Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
Course at Paris VI University, 2005/2006
1
Pierre Schapira
June 22, 2006
1
To the students: the material covered by these Notes goes beyond the con
tents of the actual course. All along the semester, the students will be informed
of what is required for the exam.
2
Contents
1 Linear algebra over a ring 7
1.1 Modules and linear maps . . . . . . . . . . . . . . . . . . . . . 7
1.2 Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 The functor Hom . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Koszul complexes . . . . . . . . . . . . . . . . . . . . . . . . . 31
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2 The language of categories 37
2.1 Categories and functors . . . . . . . . . . . . . . . . . . . . . . 37
2.2 The Yoneda Lemma . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 Adjoint functors . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6 Exact functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.7 Filtrant inductive limits . . . . . . . . . . . . . . . . . . . . . 53
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3 Additive categories 59
3.1 Additive categories . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Complexes in additive categories . . . . . . . . . . . . . . . . . 61
3.3 Simplicial constructions . . . . . . . . . . . . . . . . . . . . . 64
3.4 Double complexes . . . . . . . . . . . . . . . . . . . . . . . . . 66
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Abelian categories 71
4.1 Abelian categories . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 Complexes in abelian categories . . . . . . . . . . . . . . . . . 75
4.3 Application to Koszul complexes . . . . . . . . . . . . . . . . 79
4.4 Injective objects . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3
4 CONTENTS
4.5 Resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.6 Derived functors . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.7 Bifunctors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5 Abelian sheaves 95
5.1 Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.2 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3 Sheaf associated with a presheaf . . . . . . . . . . . . . . . . . 100
5.4 Internal operations . . . . . . . . . . . . . . . . . . . . . . . . 104
5.5 Direct and inverse images . . . . . . . . . . . . . . . . . . . . 107
5.6 Sheaves associated with a locally closed subset . . . . . . . . . 112
5.7 Locally constant and locally free sheaves . . . . . . . . . . . . 115
5.8 Gluing sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6 Cohomology of sheaves 123
6.1 Cohomology of sheaves . . . . . . . . . . . . . . . . . . . . . . 123
6.2
ˇ
Cech complexes for closed coverings . . . . . . . . . . . . . . . 126
6.3 Invariance by homotopy . . . . . . . . . . . . . . . . . . . . . 127
6.4 Cohomology of some classical manifolds . . . . . . . . . . . . . 134
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7 Homotopy and fundamental groupoid 139
7.1 Fundamental groupoid . . . . . . . . . . . . . . . . . . . . . . 139
7.2 Monodromy of locally constant sheaves . . . . . . . . . . . . . 142
7.3 The Van Kampen theorem . . . . . . . . . . . . . . . . . . . . 146
7.4 Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
CONTENTS 5
Introduction
This course is an ﬁrst introduction to Algebraic Topology from the point of
view of Sheaf Theory. An expanded version of these Notes may be found in
[23], [24].
Algebraic Topology is usually approached via the study of of homology
deﬁned using chain complexes and the fundamental group, whereas, here,
the accent is put on the language of categories and sheaves, with particular
attention to locally constant sheaves.
Sheaves on topological spaces were invented by Jean Leray as a tool to
deduce global properties from local ones. This tool turned out to be ex
tremely powerful, and applies to many areas of Mathematics, from Algebraic
Geometry to Quantum Field Theory.
The functor associating to a sheaf F on a topological space X the space
F(X) of its global sections is left exact, but not right exact in general. The
derived functors H
j
(X; F) encode the ‘ ‘obstructions” to pass from local to
global. Given a ring k, the cohomology groups H
j
(X; k
X
) of the sheaf k
X
of kvalued locally constant functions is therefore a topological invariant of
the space X. Indeed, it is a homotopy invariant, and we shall explain how
to calculate H
j
(X; k
X
) in various situations.
We also introduce the fundamental group π
1
(X) of a topological space
(with suitable assumptions on the space) and prove an equivalence of cate
gories between that of ﬁnite dimensional representations of this group and
that of local systems on X. As a byproduct, we deduce the Van Kampen
theorem from the theorem on the glueing of sheaves deﬁned on a covering.
Lectures will be organized as follows.
Chapter 1 is a brief survey of linear algebra over a ring. It serves as a
guide for the theory of additive and abelian categories which is exposed in
the subsequent chapters.
In Chapter 2 we expose the basic language of categories and functors.
A key point is the Yoneda lemma, which asserts that a category ( may be
embedded in the category (
∧
of contravariant functors on ( with values in
the category Set of sets. This naturally leads to the concept of representable
functor. Next, we study inductive and projective limits in some detail and
with many examples.
Chapters 3 and 4 are devoted to additive and abelian categories. The
aim is the construction and the study of the derived functors of a left (or
right) exact functor F of abelian categories. Hence, we start by studying
complexes (and double complexes) in additive and abelian categories. Then
we brieﬂy explain the construction of the right derived functor by using
injective resolutions and later, by using Finjective resolutions. We apply
6 CONTENTS
these results to the case of the functors Ext and Tor.
In Chapter 5, we study abelian sheaves on topological spaces (with a
brief look at Grothendieck topologies). We construct the sheaf associated
with a presheaf and the usual internal operations (Hom and ⊗) and external
operations (direct and inverse images). We also explain how to obtain locally
constant or locally free sheaves when glueing sheaves.
In Chapter 6 we prove that the category of abelian sheaves has enough
injectives and we deﬁne the cohomology of sheaves. We construct resolutions
of sheaves using open or closed
ˇ
Cech coverings and, using the fact that the
cohomology of locally constant sheaves is a homotopy invariant, we show how
to compute the cohomology of spaces by using cellular decomposition. We
apply this technique to deduce the cohomology of some classical manifolds.
In Chapter 7, we deﬁne the fundamental groupoid π
1
(X) of a locally
arcwise connected space X as well as the monodromy of a locally constant
sheaf and prove that under suitable assumptions, the monodromy functor is
an equivalence. We also show that the Van Kampen theorem may be deduced
from the theorem on the glueing of sheaves and apply it in some particular
sitations.
Conventions. In these Notes, all rings are unital and associative but not
necessarily commutative. The operations, the zero element, and the unit are
denoted by +, , 0, 1, respectively. However, we shall often write for short ab
instead of a b.
All along these Notes, k will denote a commutative ring. (Sometimes, k
will be a ﬁeld.)
We denote by ∅ the empty set and by ¦pt¦ a set with one element.
We denote by N the set of nonnegative integers, N = ¦0, 1, . . . ¦.
Chapter 1
Linear algebra over a ring
This chapter is a short review of basic and classical notions of commutative
algebra.
Many notions introduced in this chapter will be repeated later in a more
general setting.
Some references: [1], [4].
1.1 Modules and linear maps
All along these Notes, k is a commutative ring.
Let A be a kalgebra, that is, a ring endowed with a morphism of rings
ϕ: k − → A such that the image of k is contained in the center of A. Notice
that a ring A is always a Zalgebra. If A is commutative, then A is an
Aalgebra.
Since we do not assume A is commutative, we have to distinguish between
left and right structures. Unless otherwise speciﬁed, a module M over A
means a left Amodule.
Recall that an Amodule M is an additive group (whose operations and
zero element are denoted +, 0) endowed with an external law A M → M
satisfying:
(ab)m = a(bm)
(a +b)m = am +bm
a(m+m
t
) = am +am
t
1 m = m
where a, b ∈ A and m, m
t
∈ M.
Note that M inherits a structure of a kmodule via ϕ. In the sequel, if
there is no risk of confusion, we shall not write ϕ.
7
8 CHAPTER 1. LINEAR ALGEBRA OVER A RING
We denote by A
op
the ring A with the opposite structure. Hence the
product ab in A
op
is the product ba in A and an A
op
module is a right A
module.
Note that if the ring A is a ﬁeld (here, a ﬁeld is always commutative),
then an Amodule is nothing but a vector space.
Examples 1.1.1. (i) The ﬁrst example of a ring is Z, the ring of integers.
Since a ﬁeld is a ring, Q, R, C are rings. If A is a commutative ring, then
A[x
1
, . . . , x
n
], the ring of polynomials in n variables with coeﬃcients in A, is
also a commutative ring. It is a subring of A[[x
1
, . . . , x
n
]], the ring of formal
powers series with coeﬃcients in A.
(ii) Let k be a ﬁeld. Then for n > 1, the ring M
n
(k) of square matrices of
rank n with entries in k is non commutative.
(iii) Let k be a ﬁeld. The Weyl algebra in n variables, denoted W
n
(k), is the
non commutative ring of polynomials in the variables x
i
, ∂
j
(1 ≤ i, j ≤ n)
with coeﬃcients in k, and relations :
[x
i
, x
j
] = 0, [∂
i
, ∂
j
] = 0, [∂
j
, x
i
] = δ
i
j
where [p, q] = pq −qp and δ
i
j
is the Kronecker symbol.
The Weyl algebra W
n
(k) may be regarded as the ring of diﬀerential op
erators with coeﬃcients in k[x
1
, . . . , x
n
], and k[x
1
, . . . , x
n
] becomes a left
W
n
(k)module: x
i
acts by multiplication and ∂
i
is the derivation with re
spect to x
i
.
A morphism f : M − → N of Amodules is an Alinear map, i.e., f satisﬁes:
f(m+m
t
) = f(m) +f(m
t
) m, m
t
∈ M
f(am) = af(m) m ∈ M, a ∈ A.
A morphism f is an isomorphism if there exists a morphism g : N − → M
with f ◦ g = id
N
, g ◦ f = id
M
.
If f is bijective, it is easily checked that the inverse map f
−1
: N → M
is itself Alinear. Hence f is an isomorphism if and only if f is Alinear and
bijective.
A submodule N of M is a subset N of M such that n, n
t
∈ N implies
n + n
t
∈ N and n ∈ N, a ∈ A implies an ∈ N. A submodule of the
Amodule A is called an ideal of A. Note that if A is a ﬁeld, it has no
non trivial ideal, i.e., its only ideals are ¦0¦ and A. If A = C[x], then
I = ¦P ∈ C[x]; P(0) = 0¦ is a non trivial ideal.
If N is a submodule of M, the quotient module M/N is characterized
by the following “universal property”: for any module L, any morphism
1.1. MODULES AND LINEAR MAPS 9
h: M − → L which induces 0 on N factorizes uniquely through M/N. This is
visualized by the diagram
N
0
M
g
h
M/N
h
.
L
Let I be a set, and let ¦M
i
¦
i∈I
be a family of Amodules indexed by I. The
product
¸
i
M
i
is the set of families ¦(x
i
)
i∈I
¦ with x
i
∈ M
i
, and this set
naturally inherits a structure of an Amodule. There are natural surjective
morphisms:
π
k
:
¸
i
M
i
− → M
k
.
Note that given a module L and a family of morphisms f
i
: L − → M
i
, this
family factorizes uniquely through
¸
i
M
i
. This is visualized by the diagram
M
i
L
f
i
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
f
j
¸
k
M
k
π
i
π
j
M
j
.
The direct sum
¸
i
M
i
is the submodule of
¸
i
M
i
consisting of families
¦(x
i
)
i∈I
¦ with x
i
= 0 for all but a ﬁnite number of i ∈ I. In particular,
if the set I is ﬁnite, the natural injection
¸
i
M
i
− →
¸
i
M
i
is an isomorphism.
There are natural injective morphisms:
ε
k
: M
k
− →
i
M
i
.
We shall sometimes identify M
k
to its image in
¸
i
M
i
by ε
k
. Note that given
a module L and a family of morphisms f
i
: M
i
− → L, this family factorizes
uniquely through
¸
i
M
i
. This is visualized by the diagram
M
i
f
i
ε
i
¸
k
M
k
L.
M
j
ε
j
f
j
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
10 CHAPTER 1. LINEAR ALGEBRA OVER A RING
If M
i
= M for all i ∈ I, one writes:
M
(I)
:=
i
M
i
, M
I
:=
¸
i
M
i
.
An Amodule M is free of rank one if it is isomorphic to A, and M is free
if it is isomorphic to a direct sum
¸
i∈I
L
i
, each L
i
being free of rank one.
If card (I) is ﬁnite, say r, then r is uniquely determined and one says M is
free of rank r.
Let f : M → N be a morphism of Amodules. One sets :
Ker f = ¦m ∈ M; f(m) = 0¦
Imf = ¦n ∈ N; there exists m ∈ M, f(m) = n¦.
These are submodules of M and N respectively, called the kernel and the
image of f, respectively. One also introduces the cokernel and the coimage
of f:
Coker f = N/ Im f, Coimf = M/ Ker f.
Since the natural morphism Coimf − → Imf is an isomorphism, one shall not
use Coim when dealing with Amodules.
If (M
i
)
i∈I
is a family of submodules of an Amodule M, one denotes by
¸
i
M
i
the submodule of M obtained as the image of the natural morphism
¸
i
M
i
− → M. This is also the module generated in M by the set
¸
i
M
i
. One
calls this module the sum of the M
i
’s in M.
Example 1.1.2. Let W
n
(k) denote as above the Weyl algebra. Consider
the left W
n
(k)linear map W
n
(k) − → k[x
1
, . . . , x
n
], W
n
(k) ÷ P → P(1) ∈
k[x
1
, . . . , x
n
]. This map is clearly surjective and its kernel is the left ideal
generated by (∂
1
, , ∂
n
). Hence, one has the isomorphism of left W
n
(k)
modules:
W
n
(k)/
¸
j
W
n
(k)∂
j
∼
−→k[x
1
, . . . , x
n
]. (1.1)
1.2 Complexes
Deﬁnition 1.2.1. A complex M
•
of Amodules is a sequence of modules
M
j
, j ∈ Z and Alinear maps d
j
M
: M
j
− → M
j+1
such that d
j
M
◦ d
j−1
M
= 0 for
all j.
1.2. COMPLEXES 11
One writes a complex as:
M
•
: → M
j
d
j
M
−−→ M
j+1
→
If there is no risk of confusion, one writes M instead of M
•
. One also often
write d
j
instead of d
j
M
.
A morphism of complexes f : M − → N is a commutative diagram:
M
k−1
f
k−1
d
k−1
M
M
k
f
k
N
k−1
d
k−1
N
N
k
Remark 1.2.2. One also encounters ﬁnite sequences of morphisms
M
j
d
j
−→ M
j+1
− → − → M
j+k
such that d
n
◦ d
n−1
= 0 when it is deﬁned. In such a case we also call such a
sequence a complex by identifying it to the complex
− → 0 − → M
j
d
j
−→ M
j+1
− → − → M
j+k
− → 0 − → .
In particular, M
t
f
− → M
g
− → M
tt
is a complex if g ◦ f = 0.
Consider a sequence
(1.2) M
t
f
− → M
g
− → M
tt
, with g◦f = 0. (Hence, this sequence is a complex.)
Deﬁnition 1.2.3. (i) The sequence (1.2) is exact if Imf
∼
−→Ker g.
(ii) More generally, a complex M
j
− → − → M
j+k
is exact if any sequence
M
n−1
− → M
n
− → M
n+1
extracted from this complex is exact.
(iii) An exact complex 0 − → M
t
− → M − → M
tt
− → 0 is called a short exact
sequence.
Example 1.2.4. Let A = k[x
1
, x
2
] and consider the sequence:
0 − → A
d
0
−→ A
2
d
1
−→ A − → 0
where d
0
(P) = (x
1
P, x
2
P) and d
1
(Q, R) = x
2
Q− x
1
R. One checks immedi
ately that d
1
◦ d
0
= 0: the sequence above is a complex.
12 CHAPTER 1. LINEAR ALGEBRA OVER A RING
One deﬁnes the kth cohomology object of a complex M
•
as:
H
k
(M
•
) = Ker d
k
/ Imd
k−1
.
Hence, a complex M
•
is exact if all its cohomology objects are zero, that is,
Imd
k−1
= Ker d
k
for all k.
If f
•
: M
•
− → N
•
is a morphism of complexes, then for each j, f
j
sends
Ker d
j
M
• to Ker d
j
N
• and sends Imd
j−1
M
• to Imd
j−1
N
• . Hence it deﬁnes the mor
phism
H
j
(f
•
) : H
j
(M
•
) − → H
j
(N
•
).
One says that f is a quasiisomorphism (a qis, for short) if H
j
(f) is an
isomorphism for all j.
As a particular case, consider a complex M
•
of the type:
0 − → M
0
f
− → M
1
− → 0.
Then H
0
(M
•
) = Ker f and H
1
(M
•
) = Coker f.
To a morphism f : M − → N one then associates the two short exact
sequences :
0 − → Ker f − → M − → Imf − → 0,
0 − → Imf − → N − → Coker f − → 0,
and f is an isomorphism if and only if Ker f = Coker f = 0. In this case one
writes :
f : M
∼
−→N.
One says f is a monomorphism (resp. epimorphism) if Ker f (resp. Coker f)
= 0.
Proposition 1.2.5. Consider an exact sequence
(1.3) 0 − → M
t
f
− → M
g
− → M
tt
− → 0.
Then the following conditions are equivalent:
(a) there exists h : M
tt
− → M such that g ◦ h = id
M
,
(b) there exists k : M − → M
t
such that k ◦ f = id
M
(c) there exists h : M
tt
− → M and k : M − → M
t
such that such that id
M
=
f ◦ k +h ◦ g,
1.3. THE FUNCTOR HOM 13
(d) there exists ϕ = (k, g) : M − → M
t
⊕M
tt
and ψ = (f +h) : M
t
⊕M
tt
− → M,
such that ϕ and ψ are isomorphisms inverse to each other. In other
words, the exact sequence (1.3) is isomorphic to the exact sequence 0 − →
M
t
− → M
t
⊕M
tt
− → M
tt
− → 0.
Proof. (a) ⇒ (c). Since g = g ◦ h ◦ g, we get g ◦ (id
M
−h ◦ g) = 0, which
implies that id
M
−h ◦ g factors through Ker g, that is, through M
t
. Hence,
there exists k : M − → M
t
such that id
M
−h ◦ g = f ◦ k.
(b) ⇒ (c). The proof is similar and left to the reader.
(c) ⇒ (a). Since g ◦ f = 0, we ﬁnd g = g ◦ h◦ g, that is (g ◦ h−id
M
) ◦ g = 0.
Since g is onto, this implies g ◦ h −id
M
= 0.
(c) ⇒ (b). The proof is similar and left to the reader.
(d) ⇔ (a)&(b)&(c) is obvious. q.e.d.
Deﬁnition 1.2.6. In the above situation, one says that the exact sequence
(1.3) splits.
If A is a ﬁeld, all exact sequences split, but this is not the case in general.
For example, the exact sequence of Zmodules
0 − → Z
2
−→ Z − → Z/2Z − → 0
does not split.
1.3 The functor Hom
In this section, A denotes a kalgebra and the notation M ∈ Mod(A) means
that M is an Amodule. One calls Mod(A) the category of Amodules. (A
precise deﬁnition will be given in Chapter 2.)
Let M and N be two Amodules. One denotes by Hom
A
(M, N) the
set of Alinear maps f : M − → N. This is clearly a kmodule. In fact one
deﬁnes the action of k on Hom
A
(M, N) by setting: (λf)(m) = λ(f(m)).
Hence (λf)(am) = λf(am) = λaf(m) = aλf(m) = a(λf(m)), and λf ∈
Hom
A
(M, N).
We shall often set for short
Hom(M, N) = Hom
k
(M, N).
Notice that if K is a kmodule, then Hom(K, M) is an Amodule.
There is a natural isomorphism Hom
A
(A, M) · M: to u ∈ Hom
A
(A, M)
one associates u(1) and to m ∈ M one associates the linear map A − →
M, a → am. More generally, if I is an ideal of A then Hom
A
(A/I, M) ·
¦m ∈ M; Im = 0¦.
14 CHAPTER 1. LINEAR ALGEBRA OVER A RING
Functors
Although the general deﬁnition of a functor will be given in Chapter 2, we
give it in the particular case of categories of modules.
Let A and B be two kalgebras. A functor F : Mod(A) − → Mod(B) as
sociates a Bmodule F(M) to each Amodule M and associates a Blinear
map F(f): F(M) − → F(N) to each Alinear map f : M − → N such that:
F(id
M
) = id
F(M)
for any Amodule M,
F(g ◦ f) = F(f) ◦ F(g) for any morphisms M
f
− → N
g
− → L.
A functor F : Mod(A) − → Mod(B) is kadditive if it commutes to ﬁnite direct
sums, i.e., F(M ⊕N) · F(M) ⊕F(N) and the map
F : Hom
A
(M, N) − → Hom
B
(F(M), F(N)
is klinear.
A contravariant functor is almost the same as a functor, with the dif
ference that it reverses the direction of the arrows. Hence, to a morphism
f : M − → N, a contravariant functor G: Mod(A) − → Mod(B) associates a
morphism G(f): G(N) − → G(M) and satisﬁes
G(g ◦ f) = G(f) ◦ G(g).
The functors Hom
A
(M,
•
) and Hom
A
(
•
, N)
Let M ∈ Mod(A). The functor
Hom
A
(M,
•
): Mod(A) − → Mod(k)
associates Hom
A
(M, K) to the Amodule K and to an Alinear map g : K − →
L it associates
Hom
A
(M, g) : Hom
A
(M, K)
g◦
−→ Hom
A
(M, L)
(M
h
− → K) → (M
h
− → K
g
− → L).
Clearly, Hom
A
(M,
•
) is a functor from the category Mod(A) of Amodules
to the category Mod(k) of kmodules.
Similarly, for N ∈ Mod(A), the contravariant functor
Hom
A
(
•
, N): Mod(A) − → Mod(k)
1.3. THE FUNCTOR HOM 15
associates Hom
A
(K, N) to the Amodule K and to an Alinear map g : K − →
L it associates
Hom
A
(g, N) : Hom
A
(L, N)
◦g
−→ Hom
A
(K, N)
(L
h
− → N) → (K
g
− → L
h
− → N).
Clearly, the two functors Hom
A
(M,
•
) and Hom
A
(
•
, N) commute to ﬁnite
direct sums or ﬁnite products, i.e.,
Hom
A
(K ⊕L, N) · Hom
A
(K, N) Hom
A
(L, N)
Hom
A
(M, K L) · Hom
A
(M, K) Hom
A
(M, L).
Hence, these functors are additive.
Exactness
Proposition 1.3.1. (a) Let 0 − → M
t
f
− → M
g
− → M
tt
be a complex of A
modules. The assertions below are equivalent.
(i) the sequence is exact,
(ii) M
t
is isomorphic by f to Ker g,
(iii) any morphism h : L − → M such that g ◦ h = 0, factorizes uniquely
through M
t
(i.e., h = f ◦ h
t
, with h
t
: L − → M
t
). This is visualized
by
L
0
h
h
0
M
t
f
M
g
M
tt
(iv) for any module L, the sequence of kmodules
(1.4) 0 − → Hom
A
(L, M
t
) − → Hom
A
(L, M) − → Hom
A
(L, M
tt
)
is exact.
(b) Let M
t
f
− → M
g
− → M
tt
− → 0 be a complex of Amodules. The assertions
below are equivalent.
(i) the sequence is exact,
(ii) M
tt
is isomorphic by g to Coker f,
16 CHAPTER 1. LINEAR ALGEBRA OVER A RING
(iii) any morphism h : M − → L such that h ◦ f = 0, factorizes uniquely
through M
tt
(i.e., h = h
tt
◦ g, with h
tt
: M
tt
− → L). This is visualized
by
M
t
f
0
M
g
h
M
tt
h
0
L
(iv) for any module L, the sequence of kmodules
(1.5) 0 − → Hom
A
(M
tt
, L) − → Hom
A
(M, L) − → Hom
A
(M
t
, L)
is exact.
Proof. (a) (i) ⇒ (ii) is obvious, as well as (ii) ⇒ (iii), since any linear map
h : L − → M such that g ◦ h = 0 factorizes uniquely through Ker g, and this
characterizes Ker g. Finally, (iii) ⇔ (iv) is tautological.
(b) The proof is similar. q.e.d.
Deﬁnition 1.3.2. (i) An additive functor F : Mod(A) − → Mod(B) is left
exact if for any exact sequence 0 − → M
t
− → M − → M
tt
in Mod(A), the
sequence 0 − → F(M
t
) − → F(M) − → F(M
tt
) in Mod(B) is exact.
(ii) An additive functor F : Mod(A) − → Mod(B) is right exact if for any ex
act sequence M
t
− → M − → M
tt
− → 0 in Mod(A), the sequence F(M
t
) − →
F(M) − → F(M
tt
) − → 0 in Mod(B) is exact.
(iii) An additive contravariant functor G: Mod(A) − → Mod(B) is left exact
if for any exact sequence M
t
− → M − → M
tt
− → 0 in Mod(A), the sequence
0 − → G(M
tt
) − → G(M) − → G(M
t
) in Mod(B) is exact.
(iv) An additive contravariant functor G: Mod(A) − → Mod(B) is right exact
if for any exact sequence 0 − → M
t
− → M − → M
tt
in Mod(A), the sequence
G(M
tt
) − → G(M) − → G(M
t
) − → 0 in Mod(B) is exact.
(v) An additive functor is exact if it is both right and left exact.
Hence, the fact that (a)–(i) ⇔ (a)–(iv) and (b)–(i) ⇔ (b)–(iv)) is formu
lated by saying that Hom
A
(
•
, L) and Hom
A
(L,
•
)) are left exact functors.
Note that if A = k is a ﬁeld, then Hom
k
(M, k) is the algebraic dual of
M, the vector space of linear functional on M, usually denoted by M
∗
. If M
is ﬁnite dimensional, then M · M
∗∗
. If u : L − → M is a linear map, the map
Hom
k
(u, k) : M
∗
− → L
∗
is usually denoted by
t
u and called the transpose of
u.
1.3. THE FUNCTOR HOM 17
Lemma 1.3.3. Consider an additive functor F : Mod(A) − → Mod(B) and
assume that for each exact sequence 0 − → M
t
− → M − → M
tt
− → 0 in Mod(A),
the sequence 0 − → F(M
t
) − → F(M) − → F(M
tt
) is exact in Mod(B). Then F
is left exact.
Proof. Consider an exact sequence 0 − → M
t
f
− → M − → L and denote by M
tt
the
cokernel of f. We get an exact sequence 0 − → M
t
− → M − → M
tt
− → 0, hence, by
the hypothesis, an exact sequence 0 − → F(M
t
) − → F(M) − → F(M
tt
). On the
other hand, the map M − → L factorizes through a map h: M
tt
− → L which is
clearly injective. Consider the exact sequence 0 − → M
tt
− → L − → Coker h − → 0.
Applying the functor F, we obtain that F(M
tt
) − → F(L) is injective. Since
the map F(M) − → F(L) factorizes through F(M) − → F(M
tt
), the kernel of
F(M) − → F(L) is isomorphic to the kernel of F(M) − → F(M
tt
). It follows
that the sequence 0 − → F(M
t
) − → F(M) − → F(L) is exact. q.e.d.
There is a similar result for right exact functors and for contravariant func
tors. Moreover:
Lemma 1.3.4. Consider an additive functor F : Mod(A) − → Mod(B). The
conditions below are equivalent:
(i) F is exact,
(ii) for any exact sequence M
t
− → M − → M
tt
in Mod(A), the sequence
F(M
t
) − → F(M) − → F(M
tt
) is exact in Mod(B),
(iii) for any exact sequence 0 − → M
t
− → M − → M
tt
− → 0 in Mod(A), the
sequence 0 − → F(M
t
) − → F(M) − → F(M
tt
) − → 0 is exact in Mod(B).
The proof is left as an exercise.
Example 1.3.5. The functors Hom
A
(
•
, L) and Hom
A
(M,
•
) are not right
exact in general. In fact choose A = k[x], with k a ﬁeld, and consider the
exact sequence of Amodules:
0 − → A
x
−→ A − → A/Ax − → 0 (1.6)
(where x means multiplication by x). Apply Hom
A
(, A) to this sequence.
We get the sequence:
0 − → Hom
A
(A/Ax, A) − → A
x
−→ A − → 0
which is not exact since x is not surjective. On the other hand, since x is
injective and Hom
A
(
•
, A) is left exact, we ﬁnd that Hom
A
(A/Ax, A) = 0.
18 CHAPTER 1. LINEAR ALGEBRA OVER A RING
Similarly, apply Hom
A
(A/Ax,
•
) to the exact sequence (1.6). We get the
sequence:
0 − → Hom
A
(A/Ax, A) − → Hom
A
(A/Ax, A) − → Hom
A
(A/Ax, A/Ax) − → 0.
Since Hom
A
(A/Ax, A) = 0 and Hom
A
(A/Ax, A/Ax) = 0, this sequence is
not exact.
Notice moreover that the functor Hom
A
(
•
,
•
) being additive, it sends
split exact sequences to split exact sequences. This shows again that (1.6)
does not split.
The next result is of constant use.
Proposition 1.3.6. Let f : M − → N be a morphism of Amodules. The
conditions below are equivalent:
(i) f is an isomorphism,
(ii) for any Amodule L, the map Hom
A
(L, M)
f◦
−→ Hom
A
(L, N) is an iso
morphism,
(iii) for any Amodule L, the map Hom
A
(N, L)
◦f
−→ Hom
A
(M, L) is an iso
morphism.
Proof. (i) ⇒ (ii) and (i) ⇒ (iii) are obvious.
(ii) ⇒ (i). Choose L = A.
(iii) ⇒ (i). By choosing L = M and id
M
∈ Hom
A
(M, M) we ﬁnd that there
exists g : N − → M such that g ◦ f = id
M
. Hence, f is injective and moreover,
by Proposition 1.2.5 there exists an isomorphism N · M ⊕ P. Therefore,
Hom
A
(P, L) · 0 for all module L, hence Hom
A
(P, P) · 0, and this implies
P · 0. q.e.d.
Injective and projective modules
Deﬁnition 1.3.7. (i) An Amodule I is injective if for any exact sequence
0 − → M
t
− → M − → M
tt
in Mod(A), the sequence Hom
A
(M
tt
, I) − →
Hom
A
(M, I) − → Hom
A
(M
t
, I) − → 0 is exact in Mod(k) or, equivalently,
if the functor Hom
A
(
•
, I) is exact.
(ii) An Amodule P is projective if for any exact sequence M
t
− → M − →
M
tt
− → 0 in Mod(A), the sequence Hom
A
(P, M
t
) − → Hom
A
(P, M) − →
Hom
A
(P, M
tt
) − → 0 is exact in Mod(k) or, equivalently, if the functor
Hom
A
(P,
•
) is exact.
1.3. THE FUNCTOR HOM 19
Proposition 1.3.8. An Amodule I is injective if and only if for any solid
diagram in which the row is exact:
0
M
t
f
k
M
h
I
the dotted arrow may be completed, making the diagram commutative.
Proof. (i) Assume that I is injective and let M
tt
denote the cokernel of the
map M
t
− → M. Applying Hom
A
(
•
, I) to the sequence 0 − → M
t
− → M − → M
tt
,
one gets the exact sequence:
Hom
A
(M
tt
, I) − → Hom
A
(M, I)
◦f
−→ Hom
A
(M
t
, I) − → 0.
Thus there exists h : M − → I such that h ◦ f = k.
(ii) Conversely, consider an exact sequence 0 − → M
t
f
− → M
g
− → M
tt
− → 0. Then
the sequence 0 − → Hom
A
(M
tt
, I)
◦h
−→ Hom
A
(M, I)
◦f
−→ Hom
A
(M
t
, I) − → 0 is
exact by Proposition 1.3.1 and the hypothesis.
To conclude, apply Lemma 1.3.3. q.e.d.
By reversing the arrows, we get a similar result assuming P is projective. In
other words, P is projective if and only if for any solid diagram in which the
row is exact:
P
k
h
M
f
M
tt
0
the dotted arrow may be completed, making the diagram commutative.
A free module is projective and if A = k is a ﬁeld, all modules are both
injective and projective.
Generators and relations
Suppose one is interested in studying a system of linear equations
(1.7)
N
0
¸
j=1
p
ij
u
j
= v
i
, (i = 1, . . . , N
1
)
where the p
ij
’s belong to the ring A and u
j
, v
i
belong to some left Amodule
L. Using matrix notations, one can write equations (1.7) as
(1.8) Pu = v
20 CHAPTER 1. LINEAR ALGEBRA OVER A RING
where P is the matrix (p
ij
) with N
1
rows and N
0
columns, deﬁning the
Alinear map P : L
N
0
− → L
N
1
. Now consider the right Alinear map
(1.9) P : A
N
1
− → A
N
0
,
where P operates on the right and the elements of A
N
0
and A
N
1
are written
as rows. Let (e
1
, . . . , e
N
0
) and (f
1
, . . . , f
N
1
) denote the canonical basis of A
N
0
and A
N
1
, respectively. One gets:
(1.10) f
i
P =
N
0
¸
j=1
p
ij
e
j
, (i = 1, . . . , N
1
).
Hence ImP is generated by the elements
¸
N
0
j=1
p
ij
e
j
for i = 1, . . . , N
1
. Denote
by M the quotient module A
N
0
/A
N
1
P and by ψ : A
N
0
− → M the natural A
linear map. Let (u
1
, . . . , u
N
0
) denote the images by ψ of (e
1
, . . . , e
N
0
). Then
M is a left Amodule with generators (u
1
, . . . , u
N
0
) and relations
¸
N
0
j=1
p
ij
u
j
=
0 for i = 1, . . . , N
1
. By construction, we have an exact sequence of left A
modules:
(1.11) A
N
1
P
−→ A
N
0
ψ
−→ M − → 0.
Applying the left exact functor Hom
A
(
•
, L) to this sequence, we ﬁnd the
exact sequence of kmodules:
(1.12) 0 − → Hom
A
(M, L) − → L
N
0
P
−→ L
N
1
.
Hence, the kmodule of solutions of the homogeneous equations associated
to (1.7) is described by Hom
A
(M, L).
1.4 Tensor product
The tensor product, that we shall construct below, solves a “universal prob
lem”. Namely, consider a right Amodule N, a left Amodule M, and a
kmodule L. Let us say that a map f : N M − → L is (A, k)bilinear if
f is additive with respect to each of its arguments and satisﬁes f(na, m) =
f(n, am), f(n(λ), m) = λ(f(n, m)) for all (n, m) ∈ N M and a ∈ A, λ ∈ k.
We shall construct a kmodule denoted N ⊗
A
M such that f factors
uniquely through the bilinear map N M − → N ⊗
A
M followed by a klinear
map N ⊗
A
M − → L. This is visualized by:
N M
f
N ⊗
A
M
L
1.4. TENSOR PRODUCT 21
First, remark that one may identify a set I to a subset of k
(I)
as follows: to
i ∈ I, we associate ¦l
j
¦
j∈I
∈ k
(I)
given by
l
j
=
1 if j = i,
0 if j = i.
(1.13)
The tensor product N ⊗
A
M is the kmodule deﬁned as the quotient of
k
(NM)
by the submodule generated by the following elements (where n, n
t
∈
N, m, m
t
∈ M, a ∈ A, λ ∈ k and N M is identiﬁed to a subset of k
(NM)
):
(n +n
t
, m) −(n, m) −(n
t
, m)
(n, m+m
t
) −(n, m) −(n, m
t
)
(na, m) −(n, am)
λ(n, m) −(nλ, m).
The image of (n, m) in N⊗
A
M is denoted n⊗m. Hence an element of N⊗
A
M
may be written (not uniquely!) as a ﬁnite sum
¸
j
n
j
⊗m
j
, n
j
∈ N, m
j
∈ M
and:
(n +n
t
) ⊗m = n ⊗m+n
t
⊗m
n ⊗(m+m
t
) = n ⊗m+n ⊗m
t
na ⊗m = n ⊗am
λ(n ⊗m) = nλ ⊗m = n ⊗λm.
Consider an Alinear map f : M − → L. It deﬁnes a linear map id
N
f :
N M − → N L, hence a (A, k)bilinear map N M − → N ⊗
A
L, and ﬁnally
a klinear map
id
N
⊗f : N ⊗
A
M − → N ⊗
A
L.
One constructs similarly g ⊗id
M
associated to g : N − → L.
Tensor product commutes to direct sum, that is, there are natural iso
morphisms:
(N ⊕N
t
) ⊗
A
M · (N ⊗
A
M) ⊕(N
t
⊗
A
M),
N ⊗
A
(M ⊕M
t
) · (N ⊗
A
M) ⊕(N ⊗
A
M
t
).
Clearly, we have constructed additive functors
N ⊗
A
•
: Mod(A) − → Mod(k),
•
⊗
A
N: Mod(A
op
) − → Mod(k).
Note that if A is commutative, there is an isomorphism: N ⊗
A
M ·
M ⊗
A
N, given by n ⊗ m → m ⊗ n and moreover the tensor product is
associative, that is, if L, M, N are Amodules, there are natural isomorphisms
L ⊗
A
(M ⊗
A
N) · (L ⊗
A
M) ⊗
A
N. One simply writes L ⊗
A
M ⊗
A
N.
22 CHAPTER 1. LINEAR ALGEBRA OVER A RING
There is a natural isomorphism A ⊗
A
M · M. We shall often write for
short
M ⊗
k
N = M ⊗N.
Sometimes, one has to consider various rings. Consider two kalgebras,
A
1
and A
2
. Then A
1
⊗A
2
has a natural structure of a kalgebra, by setting
(a
1
⊗a
2
) (b
1
⊗b
2
) = a
1
b
1
⊗a
2
b
2
.
An (A
1
⊗A
op
2
)module M is also called a (A
1
, A
2
)bimodule (a left A
1
module
and right A
2
module). Note that the actions of A
1
and A
2
on M commute,
that is,
a
1
a
2
m = a
2
a
1
m, a
1
∈ A
1
, a
2
∈ A
2
, m ∈ M.
Let A
1
, A
2
, A
3
, A
4
denote four kalgebras.
Proposition 1.4.1. Let
i
M
j
be an (A
i
⊗A
op
j
)module. Then
1
M
2
⊗
A
2
2
M
3
is an (A
1
⊗A
op
3
)module,
Hom
A
1
(
1
M
2
,
1
M
3
) is an (A
2
⊗A
op
3
)module,
and there is a natural isomorphism of A
4
⊗A
op
3
modules
(1.14) Hom
A
1
(
1
M
4
, Hom
A
2
(
2
M
1
,
2
M
3
)) · Hom
A
2
(
2
M
1
⊗
A
1
1
M
4
,
2
M
3
).
In particular, if A is a kalgebra, M, N are left Amodules and L is a
kmodule, we have the isomorphisms
Hom
A
(L ⊗
k
N, M) · Hom
A
(N, Hom
k
(L, M)) (1.15)
· Hom
k
(L, Hom
A
(N, M)).
Proof. We shall only prove (1.15) in the particular case where A = k. In
this case, Hom
A
(L⊗
k
N, M) is nothing but the kmodule of kbilinear maps
from L N to M, and a kbilinear map from L N to M deﬁnes uniquely
a linear map from L to Hom
A
(N, M) and conversely. q.e.d.
Consider the functors
Φ(
•
) := L ⊗
k
•
: Mod(A) − → Mod(A),
Ψ(
•
) := Hom
k
(L,
•
): Mod(A) − → Mod(A),
Φ
t
(
•
) :=
•
⊗
k
N: Mod(k) − → Mod(A),
Ψ
t
(
•
) := Hom
A
(N,
•
): Mod(A) − → Mod(k).
1.4. TENSOR PRODUCT 23
The isomorphisms in (1.15) become:
Hom
A
(Φ(N), M) · Hom
A
(N, Ψ(M))
Hom
A
(Φ
t
(N), M) · Hom
A
(N, Ψ
t
(M))
One translates these isomorphisma (see Chapter 2 below) by saying that
(Φ, Ψ) and (Φ
t
, Ψ
t
) are pairs of adjoint functors.
Proposition 1.4.2. If M
t
− → M − → M
tt
− → 0 is an exact sequence of left A
modules, then the sequence of kmodules N⊗
A
M
t
− → N⊗
A
M − → N⊗
A
M
tt
− → 0
is exact.
Proof. By Proposition 1.3.1 (b), it is enough to check that for any kmodule
L, the sequence
0 − → Hom
k
(N ⊗
A
M
tt
, L) − → Hom
k
(N ⊗
A
M, L) − → Hom
k
(N ⊗
A
M
t
, L)
is exact. This sequence is isomorphic to the sequence
0 − → Hom
k
(M
tt
, Hom
A
(N, L)) − → Hom
k
(M, Hom
A
(N, L))
− → Hom
k
(M
t
, Hom
A
(N, L))
and it remains to apply Proposition 1.3.1 ((b), (i) ⇒ (ii)). q.e.d.
Hence,
•
⊗
A
M: Mod(A
op
) − → Mod(k) and N ⊗
A
•
: Mod(A) − → Mod(k) are
right exact functors.
Example 1.4.3.
•
⊗
A
M is not left exact in general. In fact, consider the
commutative ring A = C[x] and the exact sequence of Amodules:
0 − → A
x
−→ A − → A/xA − → 0.
Apply
•
⊗
A
A/Ax. We get the sequence:
0 − → A/Ax
x
−→ A/Ax − → A/xA⊗
A
A/Ax − → 0
Multiplication by x is 0 on A/Ax. Hence this sequence is the same as:
0 − → A/Ax
0
− → A/Ax − → A/Ax ⊗
A
A/Ax − → 0
which shows that A/Ax ⊗
A
A/Ax · A/Ax and moreover that this sequence
is not exact.
24 CHAPTER 1. LINEAR ALGEBRA OVER A RING
Deﬁnition 1.4.4. (i) An A
op
module N is ﬂat if for any exact sequence
0 − → M
t
− → M − → M
tt
in Mod(A), the sequence 0 − → N ⊗
A
M
t
− →
N ⊗
A
M − → N ⊗
A
M
tt
is exact in Mod(k). In other words, N is ﬂat if
and only if the functor N ⊗
A
•
is exact.
(ii) If N is ﬂat and moreover N ⊗
A
M = 0 implies M = 0, one says that N
is faithfully ﬂat.
(iii) Similarly, an Amodule N is ﬂat if the functor
•
⊗
A
N is exact and N
is faithfully ﬂat if moreover M ⊗
A
N = 0 implies M = 0.
One proves easily that a projective module is ﬂat (see Exercise 1.2).
1.5 Limits
Deﬁnition 1.5.1. Let I be a set.
(i) An order ≤ on I is a relation which satisﬁes: (a) i ≤ i, (b) i ≤ j &
j ≤ k implies i ≤ k, (c) i ≤ j and j ≤ i implies i = j.
(ii) The opposite order (I, ≤
op
) is deﬁned by i ≤
op
j if and only if j ≤ i.
(iii) An order is discrete if i ≤ j implies i = j.
The following deﬁnition will be of constant use.
Deﬁnition 1.5.2. Let (I, ≤) be an ordered set.
(i) One says that (I, ≤) is ﬁltrant (one also says “directed”) if for any
i, j ∈ I there exists k with i ≤ k and j ≤ k.
(ii) Assume I is ﬁltrant and let J ⊂ I be a subset. One says that J is
coﬁnal to I if for any i ∈ I there exists j ∈ J with i ≤ j.
Let (I, ≤) be a ordered set and let A be a ring. A projective system
¦N
i
, v
ij
¦ of Amodules indexed by (I, ≤) is the data for each i ∈ I of an A
module N
i
and for each pair i, j with i ≤ j of an Alinear map v
ij
: N
j
− → N
i
,
such that for all i, j, k with i ≤ j and j ≤ k:
v
ii
= id
N
i
v
ij
◦ v
jk
= v
ik
.
Consider the “universal problem”: to ﬁnd an Amodule N and linear
maps v
i
: N − → N
i
satisfying v
ij
◦ v
j
= v
i
for all i ≤ j, such that for any
1.5. LIMITS 25
Amodule L and linear maps g
i
: L − → N
i
, satisfying v
ij
◦g
j
= g
i
for all i ≤ j,
there is a unique linear map g : L − → N such that g
i
= v
i
◦ g for all i. If such
a family (N, v
i
) exists (and we shall show below that it does), it is unique up
to unique isomorphism and one calls it the projective limit of the projective
system (N
i
, v
ij
), denoted lim
←−
i
N
i
. This problem is visualized by the diagram:
N
i
L
g
i
g
j
lim
←−
k
N
k
v
i
v
j
N
j
.
v
ij
An inductive system ¦M
i
, u
ji
¦ of Amodules indexed by (I, ≤) is the data for
each i ∈ I of an Amodule M
i
and for each pair i, j with i ≤ j of an Alinear
map u
ji
: M
i
− → M
j
, such that for all i, j, k with i ≤ j and j ≤ k:
u
ii
= id
M
i
u
kj
◦ u
ji
= u
ki
.
Note that a projective system indexed by (I, ≤) is nothing but an inductive
system indexed by (I, ≤
op
).
Consider the “universal problem”: to ﬁnd an Amodule M and linear
maps u
i
: M
i
− → M satisfying u
j
◦ u
ji
= u
i
for all i ≤ j, such that for any
Amodule L and linear maps f
i
: M
i
− → L satisfying f
j
◦u
ji
= f
i
for all i ≤ j,
there is a unique linear map f : M − → L such that f
i
= f ◦u
i
for all i. If such
a family (M, u
i
)
i
exists (and we shall show below that it does), it is unique
up to unique isomorphism and one calls it the inductive limit of the inductive
system (M
i
, u
ji
), denoted lim
−→
i
M
i
. This problem is visualized by the diagram:
M
i
u
ji
f
i
u
i
lim
−→
k
M
k
L.
M
j
u
j
f
j
26 CHAPTER 1. LINEAR ALGEBRA OVER A RING
Theorem 1.5.3. (i) The projective limit of the projective system ¦N
i
, v
ij
¦
is the Amodule
lim
←−
i
N
i
= ¦¦x
i
¦
i
∈
¸
i
N
i
; v
ij
(x
j
) = x
i
for all i ≤ j¦.
The maps v
i
: lim
←−
j
N
j
− → N
i
are the natural ones.
(ii) The inductive limit of the inductive system ¦M
i
, u
ij
¦ is the Amodule
lim
−→
i
M
i
= (
i∈I
M
i
)/N
where N is the submodule of
¸
i∈I
M
i
generated by ¦x
i
− u
ji
(x
i
); x
i
∈
M
i
, i ≤ j¦. The maps u
i
: M
i
− → lim
−→
j
M
j
are the natural ones.
Note that if I is discrete, then lim
−→
i
M
i
=
¸
i
M
i
and lim
←−
i
N
i
=
¸
i
N
i
.
The proof is straightforward.
The universal properties on the projective and inductive limit are better
formulated by the isomorphisms which characterize lim
←−
i
N
i
and lim
−→
i
M
i
:
Hom
A
(L, lim
←−
i
N
i
)
∼
−→ lim
←−
i
Hom
A
(L, N
i
), (1.16)
Hom
A
(lim
−→
i
M
i
, L)
∼
−→ lim
←−
i
Hom
A
(M
i
, L). (1.17)
There are also natural morphisms
lim
−→
i
Hom
A
(L, M
i
) − → Hom
A
(L, lim
−→
i
M
i
) (1.18)
lim
−→
i
Hom
A
(N
i
, L) − → Hom
A
(lim
←−
i
N
i
, L). (1.19)
One should be aware that morphisms (1.18) and (1.19) are not isomorphisms
in general (see Example 1.5.7 below).
Proposition 1.5.4. Let M
t
i
f
i
−→ M
i
g
i
−→ M
tt
i
be a family of exact sequences of
Amodules, indexed by the set I. Then the sequences
¸
i
M
t
i
− →
¸
i
M
i
− →
¸
i
M
tt
i
and
i
M
t
i
− →
i
M
i
− →
i
M
tt
i
are exact.
1.5. LIMITS 27
The proof is left as an (easy) exercise.
Proposition 1.5.5. (i) Consider a projective system of exact sequences of
Amodules: 0 − → N
t
i
f
i
−→ N
i
g
i
−→ N
tt
i
. Then the sequence 0 − → lim
←−
i
N
t
i
f
− →
lim
←−
i
N
i
g
− → lim
←−
i
N
tt
i
is exact.
(ii) Consider an inductive system of exact sequences of Amodules: M
t
i
f
i
−→
M
i
g
i
−→ M
tt
i
− → 0. Then the sequence lim
−→
i
M
t
i
f
− → lim
−→
i
M
i
g
− → lim
−→
i
M
tt
i
− → 0 is
exact.
Proof. (i) Since lim
←−
i
N
t
i
is a submodule of
¸
i
N
t
i
, the fact that f is injective
follows from Proposition 1.5.4. Let ¦x
i
¦
i
∈ lim
←−
i
N
i
with g(¦x
i
¦
i
) = 0. Then
g
i
(x
i
) = 0 for all i, and there exists a unique x
t
i
∈ N
t
i
such that x
i
= f
i
(x
t
i
).
One checks immedialtely that the element ¦x
t
i
¦
i
belongs to lim
←−
i
N
t
i
.
(ii) Let L be an Amodule. The sequence
0 − → Hom
A
(lim
−→
i
M
tt
i
, L) − → Hom
A
(lim
−→
i
M
i
, L) − → Hom
A
(lim
−→
i
M
t
i
, L)
is isomorphic to the sequence
0 − → lim
←−
i
Hom
A
(M
tt
i
, L) − → lim
←−
i
Hom
A
(M
i
, L) − → lim
←−
i
Hom
A
(M
t
i
, L)
and this sequence is exact by (i) and Proposition 1.3.1. Then the result
follows, again by Proposition 1.3.1. q.e.d.
One says that “the functor lim
−→
is right exact”, and “the functor lim
←−
is
left exact”. We shall give a precise meaning to these sentences in Chapter 2.
Remark 1.5.6. (i) If all M
i
’s are submodules of a module M, and if the
maps u
ji
: M
i
− → M
j
, (i ≤ j) are the natural injective morphisms, then
lim
−→
i
M
i
·
¸
i
M
i
.
(ii) If all M
i
’s are submodules of a module M, and if the maps v
ij
: M
j
− →
M
i
, (i ≤ j) are the natural injective morphisms, then lim
←−
i
M
i
·
¸
i
M
i
.
Example 1.5.7. Let k be a commutative ring and consider the kalgebra
A := k[x]. Denote by I = A x the ideal generated by x. Notice that
28 CHAPTER 1. LINEAR ALGEBRA OVER A RING
A/I
n+1
· k[x]
≤n
, where k[x]
≤n
denotes the kmodule consisting of polyno
mials of degree less than or equal to n.
(i) For p ≤ n there are monomorphisms u
pn
: k[x]
≤p
k[x]
≤n
which deﬁne an
inductive system of kmodules. One has the isomorphism
k[x] = lim
−→
n
k[x]
≤n
.
Notice that id
k[x]
∈ lim
−→
n
Hom
k
(k[x], k[x]
≤n
). This shows that the morphism
(1.18) is not an isomorphism in general.
(ii) For p ≤ n there are epimorphisms v
pn
: A/I
n
A/I
p
which deﬁne a
projective system of Amodules whose projective limit is k[[x]], the ring of
formal series with coeﬃcients in k.
(iii) For p ≤ n there are monomorphisms I
n
I
p
which deﬁne a projective
system of Amodules whose projective limit is 0.
(iv) We thus have a projective system of complexes of Amodules
L
•
n
: 0 − → I
n
− → A − → A/I
n
− → 0.
Taking the projective limit, we get the complex 0 − → 0 − → k[x] − → k[[x]] − → 0
which is no more exact.
Tensor products and inductive limits
Let ¦M
i
, u
ji
¦ be an inductive system of Amodules, N a right Amodule.
The family of morphisms M
i
− → lim
−→
i
M
i
deﬁnes the family of morphisms
N ⊗
A
M
i
− → N ⊗
A
lim
−→
i
M
i
, hence the morphism
(1.20) lim
−→
i
(N ⊗
A
M
i
) − → N ⊗
A
lim
−→
i
M
i
.
Proposition 1.5.8. The morphism (1.20) is an isomorphism.
Proof. Let L be a kmodule. Consider the chain of isomorphisms
Hom
k
(N ⊗
A
lim
−→
i
M
i
, L) · Hom
A
(lim
−→
i
M
i
, Hom
k
(N, L))
· lim
←−
i
Hom
A
(M
i
, Hom
k
(N, L))
· lim
←−
i
Hom
k
(N ⊗
A
M
i
, L)
· Hom
k
(lim
−→
i
(N ⊗
A
M
i
), L).
Then the result follows from Proposition 1.3.6. q.e.d.
1.5. LIMITS 29
Filtrant limit
Lemma 1.5.9. Assume I is a ﬁltrant ordered set and let M = lim
−→
i
M
i
.
(i) Let x
i
∈ M
i
. Then u
i
(x
i
) = 0 ⇔ there exists k ≥ i with u
ki
(x
i
) = 0.
(ii) Let x ∈ M. Then there exists i ∈ I and x
i
∈ M
i
with u
i
(x
i
) = x.
Proof. We keep the notations of Theorem 1.5.3 (ii). Hence, N denotes the
submodule of ⊕
i
M
i
generated by the elements ¦x
i
−u
ji
(x
i
); x
i
∈ M
i
, i ≤ j¦.
Let N
t
denote the subset of ⊕
i
M
i
consisting of ﬁnite sums
¸
j∈J
x
j
, x
j
∈ M
j
such that there exists k ≥ j for all j ∈ J with
¸
j∈J
u
kj
(x
j
) = 0. Since I is
ﬁltrant, N
t
is a submodule of ⊕
i
M
i
. Let us show that N = N
t
. The inclusion
N ⊂ N
t
is obvious since u
ji
(x
i
) + u
jj
(−u
ji
(x
i
)) = 0. Conversely, let J ⊂ I
be a ﬁnite set and let x =
¸
j∈J
x
j
∈ N
t
. Then
¸
j∈J
x
j
=
¸
j∈J
x
j
−
¸
j∈J
u
kj
(x
j
)
=
¸
j∈J
(x
j
−u
kj
(x
j
)) ∈ N
t
.
(i) follows from the fact that x ∈ N
t
∩M
i
if and only if there exists k ≥ i
with u
ki
(x
i
) = 0.
(ii) Let x ∈ M. There exist a ﬁnite set J ⊂ I and x
j
∈ M
j
such that
x =
¸
j∈J
u
j
(x
j
). Choose i with i ≥ j for all j ∈ J. Then
x =
¸
j∈J
u
k
u
ij
(x
j
) = u
i
(
¸
j∈J
u
ij
(x
j
)).
Setting x
i
=
¸
j∈J
u
ij
(x
j
), the result follows. q.e.d.
Example 1.5.10. Let X be a topological space, x ∈ X and denote by I
x
the
set of open neighborhoods of x in X. We endow I
x
with the order: U ≤ V if
V ⊂ U. Given U and V in I
x
, and setting W = U ∩V , we have U ≤ W and
V ≤ W. Therefore, I
x
is ﬁltrant.
Denote by (
0
(U) the Cvector space of complex valued continuous func
tions on U. The restriction maps (
0
(U) − → (
0
(V ), V ⊂ U deﬁne an inductive
system of Cvector spaces indexed by I
x
. One sets
(
0
X,x
= lim
−→
U∈Ix
(
0
(U). (1.21)
An element ϕ of (
0
X,x
is called a germ of continuous function at 0. Such a
germ is an equivalence class (U, ϕ
U
)/ ∼ with U a neighborhood of x, ϕ
U
a
30 CHAPTER 1. LINEAR ALGEBRA OVER A RING
continuous function on U, and (U, ϕ
U
) ∼ 0 if there exists a neighborhood V
of x with V ⊂ U such that the restriction of ϕ
U
to V is the zero function.
Hence, a germ of function is zero at x if this function is identically zero in a
neighborhood of x.
Proposition 1.5.11. Consider an inductive system of exact sequences of
Amodules indexed by a ﬁltrant ordered set I: M
t
i
f
i
−→ M
i
g
i
−→ M
tt
i
. Then the
sequence
lim
−→
i
M
t
i
f
− → lim
−→
i
M
i
g
− → lim
−→
i
M
tt
i
is exact.
Proof. Let x ∈ lim
−→
i
M
i
with g(x) = 0. There exists x
i
∈ M
i
with u
i
(x
i
) =
x, and there exists j ≥ i such that u
ji
(g
i
(x
i
)) = 0. Hence g
j
(u
ji
(x
i
)) =
u
ji
(f
i
(x
i
)) = 0, which implies that there exists x
t
j
∈ M
t
j
such that u
ji
(x
i
) =
f
j
(x
t
j
). Then x
t
= u
t
j
(x
t
j
) satisﬁes f(x
t
) = f(u
t
j
(x
t
j
)) = u
j
f
j
(x
t
j
) = u
j
u
ji
(x
i
) =
x. q.e.d.
Proposition 1.5.12. Assume J ⊂ I and assume that I is ﬁltrant and J is
coﬁnal to I.
(i) Let ¦M
i
, u
ij
¦ be an inductive system of Amodules indexed by I. Then
the natural morphism lim
−→
j∈J
M
j
− → lim
−→
i∈I
M
i
is an isomorphism.
(ii) Let ¦M
i
, v
ji
¦ be a projective system of Amodules indexed by I. Then
the natural morphism lim
←−
i∈I
M
i
− → lim
←−
j∈J
M
j
is an isomorphism.
The proof is left as an exercise.
In particular, assume I = ¦0, 1¦ with 0 < 1. Then the inductive limit of
the inductive system u
10
: M
0
− → M
1
is M
1
, and the projective limit of the
projective system v
01
: M
1
− → M
0
is M
1
.
The MittagLeﬄer condition
Recall (Proposition 1.5.4) that a product of exact sequences of Amodules is
an exact sequence. Let us give another criterion in order that the projective
limit of an exact sequence remains exact. This is a particular case of the
socalled “MittagLeﬄer” condition (see [15]).
1.6. KOSZUL COMPLEXES 31
Proposition 1.5.13. Let 0 − → ¦M
t
n
¦
fn
−→ ¦M
n
¦
gn
−→ ¦M
tt
n
¦ − → 0 be an exact
sequence of projective systems of Amodules indexed by N. Assume that for
each n, the map M
t
n+1
− → M
t
n
is surjective. Then the sequence
0 − → lim
←−
n
M
t
n
f
− → lim
←−
n
M
n
g
− → lim
←−
n
M
tt
n
− → 0
is exact.
Proof. Let us denote for short by v
p
the morphisms M
p
− → M
p−1
which deﬁne
the projective system ¦M
p
¦, and similarly for v
t
p
, v
tt
p
.
Let ¦x
tt
p
¦
p
∈ lim
←−
n
M
tt
n
. Hence x
tt
p
∈ M
tt
p
, and v
tt
p
(x
tt
p
) = x
tt
p−1
.
We shall ﬁrst show that v
n
: g
−1
n
(x
tt
n
) − → g
−1
n−1
(x
tt
n−1
) is surjective. Let
x
n−1
∈ g
−1
n−1
(x
tt
n−1
). Take x
n
∈ g
−1
n
(x
tt
n
). Then g
n−1
(v
n
(x
n
) − x
n−1
)) =
0. Hence v
n
(x
n
) − x
n−1
= f
n−1
(x
t
n−1
). By the hypothesis f
n−1
(x
t
n−1
) =
f
n−1
(v
t
n
(x
t
n
)) for some x
t
n
and thus v
n
(x
n
−f
n
(x
t
n
)) = x
n−1
.
Then we can choose x
n
∈ g
−1
n
(x
tt
n
) inductively such that v
n
(x
n
) = x
n−1
.
q.e.d.
1.6 Koszul complexes
If L is a ﬁnite free kmodule of rank n, one denotes by
j
L the kmodule
consisting of jmultilinear alternate forms on the dual space L
∗
and calls it
the jth exterior power of L. (Recall that L
∗
= Hom
k
(L, k).)
Note that
1
L · L and
n
L · k. One sets
0
L = k.
If (e
1
, . . . , e
n
) is a basis of L and I = ¦i
1
< < i
j
¦ ⊂ ¦1, . . . , n¦, one
sets
e
I
= e
i
1
∧ ∧ e
i
j
.
For a subset I ⊂ ¦1, . . . , n¦, one denotes by [I[ its cardinal. The family of
e
I
’s with [I[ = j is a basis of the free module
j
L.
Let M be an Amodule and let ϕ = (ϕ
1
, . . . , ϕ
n
) be n endomorphisms of
M over A which commute with one another:
[ϕ
i
, ϕ
j
] = 0, 1 ≤ i, j ≤ n.
(Recall the notation [a, b] := ab − ba.) Set M
(j)
= M ⊗
j
k
n
. Hence
M
(0)
= M and M
(n)
· M. Denote by (e
1
, . . . , e
n
) the canonical basis of k
n
.
Hence, any element of M
(j)
may be written uniquely as a sum
m =
¸
[I[=j
m
I
⊗e
I
.
32 CHAPTER 1. LINEAR ALGEBRA OVER A RING
One deﬁnes d ∈ Hom
A
(M
(j)
, M
(j+1)
) by:
d(m⊗e
I
) =
n
¸
i=1
ϕ
i
(m) ⊗e
i
∧ e
I
and extending d by linearity. Using the commutativity of the ϕ
i
’s one checks
easily that d ◦ d = 0. Hence we get a complex, called a Koszul complex and
denoted K
•
(M, ϕ):
0 − → M
(0)
d
− → − → M
(n)
− → 0.
When n = 1, the cohomology of this complex gives the kernel and cokernel
of ϕ
1
. More generally,
H
0
(K
•
(M, ϕ)) · Ker ϕ
1
∩ . . . ∩ Ker ϕ
n
,
H
n
(K
•
(M, ϕ)) · M/(ϕ
1
(M) + +ϕ
n
(M)).
Deﬁnition 1.6.1. (i) If for each j, 1 ≤ j ≤ n, ϕ
j
is injective as an endo
morphism of M/(ϕ
1
(M) + +ϕ
j−1
(M)), one says (ϕ
1
, . . . , ϕ
n
) is a regular
sequence.
(ii) If for each j, 1 ≤ j ≤ n, ϕ
j
is surjective as an endomorphism of
Ker ϕ
1
∩ . . . ∩ Ker ϕ
j−1
, one says (ϕ
1
, . . . , ϕ
n
) is a coregular sequence.
Theorem 1.6.2. (i) Assume that (ϕ
1
, . . . , ϕ
n
) is a regular sequence. Then
H
j
(K
•
(M, ϕ)) · 0 for j = n.
(ii) Assume that (ϕ
1
, . . . , ϕ
n
) is a coregular sequence. Then H
j
(K
•
(M, ϕ))
· 0 for j = 0.
Proof. The proof will be given in Section 4.2. Here, we restrict ourselves
to the simple case n = 2 for coregular sequences. Hence we consider the
complex:
0 − → M
d
− → M M
d
− → M − → 0
where d(x) = (ϕ
1
(x), ϕ
2
(x)), d(y, z) = ϕ
2
(y) − ϕ
1
(z) and we assume ϕ
1
is
surjective on M, ϕ
2
is surjective on Ker ϕ
1
.
Let (y, z) ∈ M M with ϕ
2
(y) = ϕ
1
(z). We look for x ∈ M solution
of ϕ
1
(x) = y, ϕ
2
(x) = z. First choose x
t
∈ M with ϕ
1
(x
t
) = y. Then
ϕ
2
◦ϕ
1
(x
t
) = ϕ
2
(y) = ϕ
1
(z) = ϕ
1
◦ϕ
2
(x
t
). Thus ϕ
1
(z −ϕ
2
(x
t
)) = 0 and there
exists t ∈ M with ϕ
1
(t) = 0, ϕ
2
(t) = z−ϕ
2
(x
t
). Hence y = ϕ
1
(t+x
t
), z =
ϕ
2
(t +x
t
) and x = t +x
t
is a solution to our problem. q.e.d.
1.6. KOSZUL COMPLEXES 33
Example 1.6.3. Let k be a ﬁeld of characteristic 0 and let A = k[x
1
, . . . , x
n
].
(i) Denote by x
i
the multiplication by x
i
in A. We get the complex:
0 − → A
(0)
d
− → − → A
(n)
− → 0
where:
d(
¸
I
a
I
⊗e
I
) =
n
¸
j=1
¸
I
x
j
a
I
⊗e
j
∧ e
I
.
The sequence (x
1
, . . . , x
n
) is a regular sequence in A, considered as an A
module. Hence the Koszul complex is exact except in degree n where its
cohomology is isomorphic to k.
(ii) Denote by ∂
i
the partial derivation with respect to x
i
. This is a klinear
map on the kvector space A. Hence we get a Koszul complex
0 − → A
(0)
d
− →
d
− → A
(n)
− → 0
where:
d(
¸
I
a
I
⊗e
I
) =
n
¸
j=1
¸
I
∂
j
(a
I
) ⊗e
j
∧ e
I
.
The sequence (∂
1
, . . . , ∂
n
) is a coregular sequence, and the above complex
is exact except in degree 0 where its cohomology is isomorphic to k. Writing
dx
j
instead of e
j
, we recognize the “de Rham complex”.
Example 1.6.4. Let W = W
n
(k) be the Weyl algebra introduced in Ex
ample 1.1.2, and denote by ∂
i
the multiplication on the right by ∂
i
. Then
(∂
1
, . . . , ∂
n
) is a regular sequence on W (considered as an Wmodule) and
we get the Koszul complex:
0 − → W
(0)
δ
− → − → W
(n)
− → 0
where:
δ(
¸
I
a
I
⊗e
I
) =
n
¸
j=1
¸
I
a
I
∂
j
⊗e
j
∧ e
I
.
This complex is exact except in degree n where its cohomology is isomorphic
to k[x] (see Exercise 1.3).
Remark 1.6.5. One may also encounter coKoszul complexes. For I =
(i
1
, . . . , i
k
), introduce
e
j
e
I
=
0 if j ∈ ¦i
1
, . . . , i
k
¦
(−1)
l+1
e
I
ˆ
l
:= (−1)
l+1
e
i
1
∧ . . . ∧ ´ e
i
l
∧ . . . ∧ e
i
k
if e
i
l
= e
j
34 CHAPTER 1. LINEAR ALGEBRA OVER A RING
where e
i
1
∧. . . ∧ ´ e
i
l
∧. . . ∧e
i
k
means that e
i
l
should be omitted in e
i
1
∧. . . ∧e
i
k
.
Deﬁne δ by:
δ(m⊗e
I
) =
n
¸
j=1
ϕ
j
(m)e
j
e
I
.
Here again one checks easily that δ ◦ δ = 0, and we get the complex:
K
•
(M, ϕ) : 0 − → M
(n)
δ
− → − → M
(0)
− → 0,
This complex is in fact isomorphic to a Koszul complex. Consider the iso
morphism
∗ :
j
k
n
∼
−→
n−j
k
n
which associates ε
I
m⊗e
ˆ
I
to m⊗e
I
, where
ˆ
I = (1, . . . , n) ` I and ε
I
is the
signature of the permutation which sends (1, . . . , n) to I .
ˆ
I (any i ∈ I is
smaller than any j ∈
ˆ
I). Then, up to a sign, ∗ interchanges d and δ.
Exercises to Chapter 1
Exercise 1.1. Consider two complexes of Amodules M
t
1
− → M
1
− → M
tt
1
and
M
t
2
− → M
2
− → M
tt
2
. Prove that the two sequences are exact if and only if the
sequence M
t
1
⊕M
t
2
− → M
1
⊕M
2
− → M
tt
1
⊕M
tt
2
is exact.
Exercise 1.2. (i) Prove that a free module is projective and ﬂat.
(ii) Prove that a module P is projective if and only if it is a direct summand
of a free module (i.e., there exists a module K such that P ⊕K is free).
(iii) Deduce that projective modules are ﬂat.
Exercise 1.3. Let k be a ﬁeld of characteristic 0, W := W
n
(k) the Weyl
algebra in n variables.
(i) Denote by x
i
: W − → W the multiplication on the left by x
i
on W (hence,
the x
i
’s are morphisms of right Wmodules). Prove that ϕ = (x
1
, . . . , x
n
)
is a regular sequence and calculate H
j
(K
•
(W, ϕ)).
(ii) Denote ∂
i
the multiplication on the right by ∂
i
on W. Prove that ψ =
(∂
1
, . . . , ∂
n
) is a regular sequence and calculate H
j
(K
•
(W, ψ)).
(iii) Now consider the left W
n
(k)module O := k[x
1
, . . . , x
n
] and the klinear
map ∂
i
: O − → O (derivation with respect to x
i
). Prove that λ = (∂
1
, . . . , ∂
n
)
is a coregular sequence and calculate H
j
(K
•
(O, λ)).
Exercise 1.4. Let A = W
2
(k) be the Weyl algebra in two variables. Con
struct the Koszul complex associated to ϕ
1
= x
1
, ϕ
2
= ∂
2
and calculate its
cohomology.
Exercises to Chapter 1 35
Exercise 1.5. If M is a Zmodule, set M
∨
= Hom
Z
(M, Q/Z).
(i) Prove that Q/Z is injective in Mod(Z).
(ii) Prove that the map Hom
Z
(M, N) − → Hom
Z
(N
∨
, M
∨
) is injective for any
M, N ∈ Mod(Z).
(iii) Prove that if P is a right projective Amodule, then P
∨
is left Ainjective.
(iv) Let M be an Amodule. Prove that there exists an injective Amodule
I and a monomorphism M − → I.
(Hint: (iii) Use formula (1.15). (iv) Prove that M → M
∨∨
is an injective
map using (ii), and replace M with M
∨∨
.)
Exercise 1.6. Let k be a ﬁeld, A = k[x, y] and consider the Amodule
M =
¸
i≥1
k[x]t
i
, where the action of x ∈ A is the usual one and the action
of y ∈ A is deﬁned by y x
n
t
j+1
= x
n
t
j
for j ≥ 1, y x
n
t = 0. Deﬁne the
endomorphisms of M, ϕ
1
(m) = x m and ϕ
2
(m) = y m. Calculate the
cohomology of the Kozsul complex K
•
(M, ϕ).
Exercise 1.7. Let I be a ﬁltrant ordered set and let M
i
, i ∈ I be an inductive
sytem of kmodules indexed by I. Let M =
¸
M
i
/ ∼ where
¸
denotes the
settheoretical disjoint union and ∼ is the relation M
i
÷ x
i
∼ y
j
∈ M
j
if
there exists k ≥ i, k ≥ j such that u
ki
(x
i
) = u
kj
(y
j
).
Prove that M is naturally a kmodule and is isomorphic to lim
−→
i
M
i
.
Exercise 1.8. Let I be a ﬁltrant ordered set and let A
i
, i ∈ I be an inductive
sytem of rings indexed by I.
(i) Prove that A := lim
−→
i
A
i
is naturally endowed with a ring structure.
(ii) Deﬁne the notion of an inductive system M
i
of A
i
modules, and deﬁne
the Amodule lim
−→
i
M
i
.
(iii) Let N
i
(resp. M
i
) be an inductive system of right (resp. left) A
i
modules.
Prove the isomorphism
lim
−→
i
(N
i
⊗
A
i
M
i
)
∼
−→lim
−→
i
N
i
⊗
A
lim
−→
i
M
i
.
36 CHAPTER 1. LINEAR ALGEBRA OVER A RING
Chapter 2
The language of categories
In this chapter we introduce some basic notions of category theory which
are of constant use in various ﬁelds of Mathematics, without spending too
much time on this language. After giving the main deﬁnitions on categories
and functors, we prove the Yoneda Lemma. We also introduce the notions
of representable functors and adjoint functors.
Then we construct inductive and projective limits in categories by using
projective limits in the category Set of sets and give some examples. We
also analyze some related notions, in particular those of coﬁnal categories,
ﬁltrant categories and exact functors. Special attention will be paid to ﬁltrant
inductive limits in the category Set.
Some references: [21], [4], [20], [10], [18], [19].
2.1 Categories and functors
Deﬁnition 2.1.1. A category ( consists of:
(i) a family Ob((), the objects of (,
(ii) for each X, Y ∈ Ob((), a set Hom
(
(X, Y ), the morphisms from X to
Y ,
(iii) for any X, Y, Z ∈ Ob((), a map, called the composition, Hom
(
(X, Y )
Hom
(
(Y, Z) − → Hom
(
(X, Z), and denoted (f, g) → g ◦ f,
these data satisfying:
(a) ◦ is associative,
(b) for each X ∈ Ob((), there exists id
X
∈ Hom(X, X) such that for all
f ∈ Hom
(
(X, Y ) and g ∈ Hom
(
(Y, X), f ◦ id
X
= f, id
X
◦g = g.
37
38 CHAPTER 2. THE LANGUAGE OF CATEGORIES
Note that id
X
∈ Hom(X, X) is characterized by the condition in (b).
Remark 2.1.2. There are some settheoretical dangers, and one should men
tion in which “universe” we are working. For sake of simplicity, we shall not
enter in these considerations here.
Notation 2.1.3. One often writes X ∈ ( instead of X ∈ Ob(() and f :
X − → Y instead of f ∈ Hom
(
(X, Y ). One calls X the source and Y the
target of f.
A morphism f : X − → Y is an isomorphism if there exists g : X ←− Y
such that f ◦ g = id
Y
, g ◦ f = id
X
. In such a case, one writes f : X
∼
−→Y or
simply X · Y . Of course g is unique, and one also denotes it by f
−1
.
A morphism f : X − → Y is a monomorphism (resp. an epimorphism) if
for any morphisms g
1
and g
2
, f ◦ g
1
= f ◦ g
2
(resp. g
1
◦ f = g
2
◦ f) implies
g
1
= g
2
. One sometimes writes f : XY or else X → Y (resp. f : XY )
to denote a monomorphism (resp. an epimorphism).
Two morphisms f and g are parallel if they have the same sources and
targets, visualized by f, g : X ⇒Y .
One introduces the opposite category (
op
:
Ob((
op
) = Ob((), Hom
(
op
(X, Y ) = Hom
(
(Y, X).
A category (
t
is a subcategory of (, denoted (
t
⊂ (, if: Ob((
t
) ⊂ Ob((),
Hom
(
(X, Y ) ⊂ Hom
(
(X, Y ) for any X, Y ∈ (
t
and the composition ◦ in (
t
is induced by the composition in (. One says that (
t
is a full subcategory if
for all X, Y ∈ (
t
, Hom
(
(X, Y ) = Hom
(
(X, Y ).
A category is discrete if the only morphisms are the identity morphisms.
Note that a set is naturally identiﬁed with a discrete category.
A category ( is ﬁnite if the family of all morphisms in ( (hence, in par
ticular, the family of objects) is a ﬁnite set.
A category ( is a groupoid if all morphisms are isomorphisms.
Examples 2.1.4. (i) Set is the category of sets and maps, Set
f
is the full
subcategory consisting of ﬁnite sets.
(ii) Rel is deﬁned by: Ob(Rel) = Ob(Set) and Hom
Rel
(X, Y ) = {(XY ),
the set of subsets of X Y. The composition law is deﬁned as follows. If
f : X − → Y and g : Y − → Z, g ◦ f is the set
¦(x, z) ∈ X Z; there exists y ∈ Y with (x, y) ∈ f, (y, z) ∈ g¦.
Of course, id
X
= ∆ ⊂ X X, the diagonal of X X.
Notice that Set is a subcategory of Rel, not a full subcategory.
2.1. CATEGORIES AND FUNCTORS 39
(iii) Let A be a ring. The category of left Amodules and Alinear maps is
denoted Mod(A). In particular Mod(Z) is the category of abelian groups.
We shall often use the notations Ab instead of Mod(Z) and Hom
A
(
•
,
•
)
instead of Hom
Mod(A)
(
•
,
•
).
One denotes by Mod
f
(A) the full subcategory of Mod(A) consisting of
ﬁnitely generated Amodules.
(iv) One denotes by C(Mod(A)) the category whose objects are the complexes
of Amodules and morphisms, morphisms of such complexes.
(v) One associates to a preordered set (I, ≤) a category, still denoted by I
for short, as follows. Ob(I) = I, and the set of morphisms from i to j has a
single element if i ≤ j, and is empty otherwise. Note that I
op
is the category
associated with I endowed with the opposite order.
(vi) We denote by Top the category of topological spaces and continuous
maps.
Deﬁnition 2.1.5. (i) An object P ∈ ( is called initial if for all X ∈
(, Hom
(
(P, X) · ¦pt¦. One often denotes by ∅
(
an initial object in (.
(ii) One says that P is terminal if P is initial in (
op
, i.e., for all X ∈
(, Hom
(
(X, P) · ¦pt¦. One often denotes by pt
(
a terminal object in
(.
(iii) One says that P is a zeroobject if it is both initial and terminal. In
such a case, one often denotes it by 0. If ( has a zero object, for any
object X ∈ (, the morphism obtained as the composition X − → 0 − → X
is still denoted by 0: X − → X.
Note that initial (resp. terminal) objects are unique up to unique isomor
phisms.
Examples 2.1.6. (i) In the category Set, ∅ is initial and ¦pt¦ is terminal.
(ii) The zero module 0 is a zeroobject in Mod(A).
(iii) The category associated with the ordered set (Z, ≤) has neither initial
nor terminal object.
Deﬁnition 2.1.7. Let ( and (
t
be two categories. A functor F : ( − → (
t
consists of a map F : Ob(() − → Ob((
t
) and for all X, Y ∈ (, of a map still
denoted by F: Hom
(
(X, Y ) − → Hom
(
(F(X), F(Y )) such that
F(id
X
) = id
F(X)
, F(f ◦ g) = F(f) ◦ F(g).
A contravariant functor from ( to (
t
is a functor from (
op
to (
t
. In other
words, it satisﬁes F(g ◦ f) = F(f) ◦ F(g). If one wishes to put the emphasis
on the fact that a functor is not contravariant, one says it is covariant.
40 CHAPTER 2. THE LANGUAGE OF CATEGORIES
One denotes by op : ( − → (
op
the contravariant functor, associated with
id
(
op.
Deﬁnition 2.1.8. (i) One says that F is faithful (resp. full, resp. fully
faithful) if for X, Y in (
Hom
(
(X, Y ) − → Hom
(
(F(X), F(Y ))
is injective (resp. surjective, resp. bijective).
(ii) One says that F is essentially surjective if for each Y ∈ (
t
there exists
X ∈ ( and an isomorphism F(X) · Y .
One deﬁnes the product of two categories ( and (
t
by :
Ob(( (
t
) = Ob(() Ob((
t
)
Hom
((
((X, X
t
), (Y, Y
t
)) = Hom
(
(X, Y ) Hom
(
(X
t
, Y
t
).
A bifunctor F : ((
t
− → (
tt
is a functor on the product category. This means
that for X ∈ ( and X
t
∈ (
t
, F(X,
•
) : (
t
− → (
tt
and F(
•
, X
t
) : ( − → (
tt
are
functors, and moreover for any morphisms f : X − → Y in (, g : X
t
− → Y
t
in
(
t
, the diagram below commutes:
F(X, X
t
)
F(f,X
)
F(X,g)
F(X, Y
t
)
F(f,Y
)
F(Y, X
t
)
F(Y,g)
F(Y, Y
t
)
In fact, (f, g) = (id
Y
, g) ◦ (f, id
X
) = (f, id
Y
) ◦ (id
X
, g).
Examples 2.1.9. (i) Hom
(
(
•
,
•
) : (
op
( − → Set is a bifunctor.
(ii) If A is a kalgebra,
•
⊗
A
•
: Mod(A
op
) Mod(A) − → Mod(k) and
Hom
A
(
•
,
•
): Mod(A)
op
Mod(A) − → Mod(k) are bifunctors.
(iii) Let A be a ring. Then H
j
(
•
) : C(Mod(A)) − → Mod(A) is a functor.
(iv) The forgetful functor for : Mod(A) − → Set associates to an Amodule
M the set M, and to a linear map f the map f.
Deﬁnition 2.1.10. Let F
1
, F
2
are two functors from ( to (
t
. A morphism
of functors θ : F
1
− → F
2
is the data for all X ∈ ( of a morphism θ(X) :
F
1
(X) − → F
2
(X) such that for all f : X − → Y , the diagram below commutes:
F
1
(X)
F
1
(f)
θ(X)
F
2
(X)
F
2
(f)
F
1
(Y )
θ(Y )
F
2
(Y )
2.1. CATEGORIES AND FUNCTORS 41
A morphism of functors is visualized by a diagram:
(
F
1
F
2
θ
(
t
Hence, by considering the family of functors from ( to (
t
and the morphisms
of such functors, we get a new category.
Notation 2.1.11. (i) We denote by Fct((, (
t
) the category of functors from
( to (
t
. One may also use the shorter notation ((
t
)
(
.
Examples 2.1.12. Let k be a ﬁeld and consider the functor
∗
: Mod(k)
op
− → Mod(k),
V → V
∗
= Hom
k
(V, k).
Then there is a morphism of functors id − →
∗
◦
∗
in Fct(Mod(k), Mod(k)).
(ii) We shall encounter morphisms of functors when considering pairs of ad
joint functors (see (2.5)).
In particular we have the notion of an isomorphism of categories. If F is
an isomorphism of categories, then there exists G : (
t
− → ( such that for all
X ∈ (, G◦ F(X) = X. In practice, such a situation rarely occurs and is not
really interesting. There is an weaker notion that we introduce below.
Deﬁnition 2.1.13. A functor F : ( − → (
t
is an equivalence of categories if
there exists G : (
t
− → ( such that: G ◦ F is isomorphic to id
(
and F ◦ G is
isomorphic to id
(
.
We shall not give the proof of the following important result below.
Theorem 2.1.14. The functor F : ( − → (
t
is an equivalence of categories if
and only if F is fully faithful and essentially surjective.
If two categories are equivalent, all results and concepts in one of them
have their counterparts in the other one. This is why this notion of equiva
lence of categories plays an important role in Mathematics.
Examples 2.1.15. (i) Let k be a ﬁeld and let ( denote the category deﬁned
by Ob(() = N and Hom
(
(n, m) = M
m,n
(k), the space of matrices of type
(m, n) with entries in a ﬁeld k (the composition being the usual composition
of matrices). Deﬁne the functor F : ( − → Mod
f
(k) as follows. To n ∈ N,
F(n) associates k
n
∈ Mod
f
(k) and to a matrix of type (m, n), F associates
the induced linear map from k
n
to k
m
. Clearly F is fully faithful, and since
42 CHAPTER 2. THE LANGUAGE OF CATEGORIES
any ﬁnite dimensional vector space admits a basis, it is isomorphic to k
n
for
some n, hence F is essentially surjective. In conclusion, F is an equivalence
of categories.
(ii) let ( and (
t
be two categories. There is an equivalence
Fct((, (
t
)
op
· Fct((
op
, ((
t
)
op
). (2.1)
(iii) Let I, J and ( be categories. There are equivalences
Fct(I J, () · Fct(J, Fct(I, ()) · Fct(I, Fct(J, ()). (2.2)
2.2 The Yoneda Lemma
Deﬁnition 2.2.1. Let ( be a category. One deﬁnes the categories
(
∧
= Fct((
op
, Set), (
∨
= Fct((
op
, Set
op
),
and the functors
h
(
: ( − → (
∧
, X → Hom
(
(
•
, X)
k
(
: ( − → (
∨
, X → Hom
(
(X,
•
).
Since there is a natural equivalence of categories
(
∨
· (
op,∧,op
, (2.3)
we shall concentrate our study on (
∧
.
Proposition 2.2.2. (The Yoneda lemma.) For A ∈ (
∧
and X ∈ (, there
is an isomorphism Hom
(
∧
(h
(
(X), A) · A(X), functorial with respect to X
and A.
Proof. One constructs the morphism ϕ: Hom
(
∧
(h
(
(X), A) − → A(X) by the
chain of morphisms: Hom
(
∧
(h
(
(X), A) − → Hom
Set
(Hom
(
(X, X), A(X)) − →
A(X), where the last map is associated with id
X
.
To construct ψ : A(X) − → Hom
(
∧
(h(X)
(
, A), it is enough to associate
with s ∈ A(X) and Y ∈ ( a map from Hom
(
(Y, X) to A(Y ). It is deﬁned
by the chain of maps Hom
(
(Y, X) − → Hom
Set
(A(X), A(Y )) − → A(Y ) where
the last map is associated with s ∈ A(X).
One checks that ϕ and ψ are inverse to each other. q.e.d.
Corollary 2.2.3. The functor h
(
is fully faithful.
2.2. THE YONEDA LEMMA 43
Proof. For X and Y in (, one has Hom
(
∧
(h
(
(X), h(Y )) · h
(
(Y )(X) =
Hom
(
(X, Y ). q.e.d.
One calls h
(
the Yoneda embedding. Hence, one may consider ( as a full
subcategory of (
∧
.
Corollary 2.2.4. Let ( be a category and let f : X − → Y be a morphism in
(.
(i) Assume that for any Z ∈ (, the map Hom
(
(Z, X)
f◦
−→ Hom
(
(Z, Y ) is
bijective. Then f is an isomorphism.
(ii) Assume that for any Z ∈ (, the map Hom
(
(Y, Z)
◦f
−→ Hom
(
(X, Z) is
bijective. Then f is an isomorphism.
Proof. (i) By the hypothesis, h
(
(f) : h
(
(X) − → h
(
(Y ) is an isomorphism in
(
∧
. Since h
(
is fully faithful, this implies that f is an isomorphism. (See
Exercise 2.2 (ii).)
(ii) follows by replacing ( with (
op
. q.e.d.
Representable functors
Deﬁnition 2.2.5. (i) One says that a functor F from (
op
to Set is repre
sentable if there exists X ∈ ( and an isomorphism if F · h
(
(X)) in (
∧
,
or, in other words, if there exists an isomorphism F(Y ) · Hom
(
(Y, X),
functorially in Y ∈ (. Such an object X is called a representative of F.
(ii) Similarly, a functor G: ( − → Set is representable if there exists X ∈ (
such that G(Y ) · Hom
(
(X, Y ), functorially in Y ∈ (.
It is important to notice that the isomorphisms above determine X up to
unique isomorphism.
Representable functors provides a categorical language to deal with uni
versal problems. Let us illustrate this by an example.
Example 2.2.6. Let k be a commutative ring and let M, N, L be three k
modules. Denote by B(N M, L) the set of kbilinear maps from N M
to L. Then the functor F : L → B(N M, L) is representable by N ⊗
k
M,
since F(L) = B(N M, L) · Hom
k
(N ⊗M, L).
44 CHAPTER 2. THE LANGUAGE OF CATEGORIES
2.3 Adjoint functors
Deﬁnition 2.3.1. Let F : ( − → (
t
and G: (
t
− → ( be two functors. One says
that (F, G) is a pair of adjoint functors or that F is a left adjoint to G, or
that G is a right adjoint to F if there exists an isomorphism of bifunctors:
Hom
(
(F(
•
),
•
) · Hom
(
(
•
, G(
•
)) (2.4)
If G is an adjoint to F, then G is unique up to isomorphism. In fact,
G(Y ) is a representative of the functor X → Hom
(
(F(X), Y ).
The isomorphism (2.4) gives the isomorphisms
Hom
(
(F ◦ G(
•
),
•
) · Hom
(
(G(
•
), G(
•
)),
Hom
(
(F(
•
), F(
•
)) · Hom
(
(
•
, G◦ F(
•
)).
In particular, we have morphisms X − → G◦ F(X), functorial in X ∈ (, and
morphisms F ◦ G(Y ) − → Y , functorial in Y ∈ (
t
. In other words, we have
morphisms of functors
F ◦ G − → id
(
, id
(
− → G◦ F. (2.5)
Example 2.3.2. Let A be a kalgebra. Let K ∈ Mod(k) and let M, N ∈
Mod(A). The formula:
Hom
A
(N ⊗K, M) · Hom
A
(N, Hom(K, M)).
tells us that the functors
•
⊗K and Hom(K,
•
) from Mod(A) to Mod(A)
are adjoint.
In the preceding situation, denote by for : Mod(A) − → Mod(k) the “forget
ful functor” which, to an Amodule M associates the underlying kmodule.
Applying the above formula with N = A, we get
Hom
A
(A⊗K, M) · Hom(K, for(M)).
Hence, the functors A⊗
•
(extension of scalars) and for are adjoint.
Example 2.3.3. Let X, Y, Z ∈ Set. The bijection
Hom
Set
(X Y, Z) · Hom
Set
(X, Hom
Set
(Y, Z))
tells us that (
•
Y, Hom
Set
(Y,
•
)) is a pair of adjoint functors.
2.4. LIMITS 45
2.4 Limits
In the sequel, I will denote a category. Let ( be a category. A functor α: I − →
( (resp. β : I
op
− → () is sometimes called an inductive (resp. projective)
system in ( indexed by I.
For example, if (I, ≤) is an ordered set, I the associated category, an
inductive system indexed by I is the data of a family (X
i
)
i∈I
of objects of
( and for all i ≤ j, a morphism X
i
− → X
j
with the natural compatibility
conditions.
Assume ﬁrst that ( is the category Set and let us consider projective
systems. One sets
lim
←−
β = ¦¦x
i
¦
i
∈
¸
i
β(i); β(s)(x
j
) = x
i
for all s ∈ Hom
I
(i, j)¦. (2.6)
The next result is obvious.
Lemma 2.4.1. Let β : I
op
− → Set be a functor and let X ∈ Set. There is a
natural isomorphism
Hom
Set
(X, lim
←−
β)
∼
−→lim
←−
Hom
Set
(X, β),
where Hom
Set
(X, β) denotes the functor I
op
− → Set, i → Hom
Set
(X, β(i)).
Consider now two functors β : I
op
− → ( and α: I − → (. For X ∈ (, we get
functors from I
op
to Set:
Hom
(
(X, β): I
op
÷ i → Hom
(
(X, β(i)) ∈ Set,
Hom
(
(α, X): I
op
÷ i → Hom
(
(α, X) ∈ Set.
Deﬁnition 2.4.2. (i) Assume that the functor X → lim
←−
Hom
(
(X, β) is
representable. We denote by lim
←−
β its representative and says that the
functor β admits a projective limit in (. In particular,
Hom
(
(X, lim
←−
β) · lim
←−
Hom
(
(X, β). (2.7)
(ii) Assume that the functor X → lim
←−
Hom
(
(α, X) is representable. We
denote by lim
−→
α its representative and says that the functor α admits
an inductive limit in (. In particular,
Hom
(
(lim
−→
α, X) · lim
←−
Hom
(
(α, X), (2.8)
46 CHAPTER 2. THE LANGUAGE OF CATEGORIES
When ( = Set this deﬁnition of lim
←−
β coincides with the former one, in
view of Lemma 2.4.1.
Notice that both projective and inductive limits are deﬁned using projec
tive limits in Set.
Assume that lim
−→
α exists in (. One gets:
lim
←−
Hom
(
(α, lim
−→
α) · Hom
(
(lim
−→
α, lim
−→
α)
and the identity of lim
−→
α deﬁnes a family of morphisms
ρ
i
: α(i) − → lim
−→
α.
Consider a family of morphisms ¦f
i
: α(i) − → X¦
i∈I
in ( satisfying the com
patibility conditions
f
i
= f
j
◦ f(s) for all s ∈ Hom
I
(i, j). (2.9)
This family of morphisms is nothing but an element of lim
←−
i
Hom(α(i), X),
hence by (2.8), an element of Hom(lim
−→
α, X). Therefore, lim
−→
α is character
ized by the “universal property”:
for all X ∈ ( and all family of morphisms ¦f
i
: α(i) − → X¦
i∈I
in
( satisfying (2.9), all morphisms f
i
’s factorize uniquely through
lim
−→
α.
(2.10)
Similarly, assume that lim
←−
β exists in (. One gets:
lim
←−
Hom
(
(lim
←−
β, β) · Hom
(
(lim
←−
β, lim
←−
β)
and the identity of lim
←−
β deﬁnes a family of morphisms
ρ
i
: lim
←−
β − → β(i).
Consider a family of morphisms ¦f
i
: X − → β(i)¦
i∈I
in ( satisfying the com
patibility conditions
f
j
= f
i
◦ f(s) for all s ∈ Hom
I
(i, j). (2.11)
This family of morphisms is nothing but an element of lim
←−
i
Hom(X, β(i)),
hence by (2.7), an element of Hom(X, lim
←−
β, X). Therefore, lim
←−
β is charac
terized by the “universal property”:
for all X ∈ ( and all family of morphisms ¦f
i
: X − → β(i)¦
i∈I
in ( satisfying (2.11), all morphisms f
i
’s factorize uniquely
through lim
←−
β.
(2.12)
2.4. LIMITS 47
Inductive and projective limits are visualized by the diagrams:
α(i)
f
i
ρ
i
α(s)
lim
−→
α
X
α(j)
ρ
j
f
j
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
β(i)
X
f
i
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
f
j
lim
←−
β
ρ
i
ρ
j
β(j)
β(s)
If ϕ: J − → I, α: I − → ( and β : I
op
− → ( are functors, we have natural
morphisms:
lim
−→
(α ◦ ϕ) − → lim
−→
α, (2.13)
lim
←−
(β ◦ ϕ) ←− lim
←−
β. (2.14)
This follows immediately of (2.10) and (2.12).
Proposition 2.4.3. Let I be a category and assume that ( admits induc
tive limits indexed by I. Then for any category J, the category (
J
admits
inductive limits indexed by I. Moreover, if α: I − → (
J
is a functor, then its
inductive limit is deﬁned as follows. For Y ∈ J, denote by α(Y ): I − → ( the
functor i → α(i)(Y ). Then lim
−→
α ∈ (
J
is given by
(lim
−→
α)(Y ) = lim
−→
α(Y ), Y ∈ J.
Similarly, if β : I
op
− → (
J
is a functor, then lim
←−
β ∈ (
J
is given by
(lim
←−
β)(Y ) = lim
←−
α(Y ), Y ∈ J.
The proof is obvious.
Recall the equivalence of categories (2.2) and consider a bifunctor α: I
J − → (. It deﬁnes a functor α
J
: I − → (
J
as well as a functor α
I
: J − → (
I
.
One easily checks that
lim
−→
α · lim
−→
(lim
−→
α
J
) · lim
−→
(lim
−→
α
I
). (2.15)
Similarly, if β : I
op
J
op
− → ( is a bifunctor, then β deﬁnes a functor
β
J
: I
op
− → (
J
op
and a functor β
I
: J
op
− → (
I
op
and one has the isomorphisms
lim
←−
β · lim
←−
lim
←−
β
J
· lim
←−
lim
←−
β
I
. (2.16)
In other words:
lim
−→
i,j
α(i, j) · lim
−→
j
(lim
−→
i
(α(i, j)) · lim
−→
i
lim
−→
j
(α(i, j)), (2.17)
lim
←−
i,j
β(i, j) · lim
←−
j
lim
←−
i
(β(i, j)) · lim
←−
i
lim
←−
j
(β(i, j)). (2.18)
48 CHAPTER 2. THE LANGUAGE OF CATEGORIES
If every functor from I to ( admits an inductive limit, one says that (
admits inductive limits indexed by I. If this property holds for all categories
I (resp.
ﬁnite categories I), one says that ( admits inductive (resp.
ﬁnite inductive) limits, and similarly when replacing I with I
op
and inductive
limits with projective limits.
2.5 Examples
Empty limits.
If I is the empty category and α: I − → ( is a functor, then lim
−→
α exists in ( if
and only if ( has an initial object ∅
(
, and in this case lim
−→
α · ∅
(
. Similarly,
lim
←−
α exists in ( if and only if ( has a terminal object pt
(
, and in this case
lim
←−
α · pt
(
.
Terminal object
If I admits a terminal object, say i
o
and if α: I − → ( and β : I
op
− → ( are
functor, then
lim
−→
α · α(i
o
) lim
←−
β · β(i
o
).
This follows immediately of (2.10) and (2.12).
Sums and products
Consider a discrete category I.
Deﬁnition 2.5.1. (i) When the category I is discrete, inductive and pro
jective limits are called coproduct and products, denoted
¸
and
¸
,
respectively. Hence, writing α(i) = X
i
or β(i) = X
i
, we get for Y ∈ (:
Hom
(
(Y,
¸
i
X
i
) ·
¸
i
Hom
(
(Y, X
i
),
Hom
(
(
¸
i
X
i
, Y ) ·
¸
i
Hom
(
(X
i
, Y ).
(ii) If I is discrete with two objects, a functor I − → ( is the data of two
objects X
0
and X
1
in ( and their coproduct and product (if they exist)
are usually denoted by X
0
. X
1
and X
0
X
1
, respectively.
2.5. EXAMPLES 49
Hence, if α: I − → ( is a functor and I is discrete, one denotes by
¸
α or
¸
i∈I
α(i) its coproduct and one denotes by
¸
α or
¸
i∈I
α(i) its product.
If α(i) = X for all i ∈ I, one simply denotes this limit by X
‘
I
(resp.
X
Q
I
). One also writes X
(I)
and X
I
instead of X
‘
I
and X
Q
I
, respectively.
Example 2.5.2. In the category Set, we have for I, X, Z ∈ Set:
X
(I)
· I X,
X
I
· Hom
Set
(I, X),
Hom
Set
(I X, Z) · Hom
Set
(I, Hom
Set
(X, Z)),
· Hom
Set
(X, Z)
I
.
The coproduct and product of two objects are visualized by the diagrams:
X
0
X
0
. X
1
X
X
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
X
0
X
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
X
0
X
1
X
1
In other words, any pair of morphisms from (resp. to) X
0
and X
1
to (resp.
from) X factors uniquely through X
0
.X
1
(resp. X
0
X
1
). If ( is the category
Set, X
0
.X
1
is the disjoint union and X
0
X
1
is the product of the two sets
X
0
and X
1
.
Cokernels and kernels
Consider the category I with two objects and two parallel morphisms other
than identities, visualized by
•
•
A functor α: I − → ( is characterized by two parallel arrows in (:
(2.19) f, g : X
0
X
1
In the sequel we shall identify such a functor with the diagram (2.19).
Deﬁnition 2.5.3. Consider two parallel arrows f, g : X
0
⇒X
1
in (.
(i) A coequalizer (one also says a cokernel), if it exists, is an inductive
limit of this functor. It is denoted by Coker(f, g).
50 CHAPTER 2. THE LANGUAGE OF CATEGORIES
(ii) An equalizer (one also says a kernel), if it exists, is a projective limit of
this functor. It is denoted by Ker(f, g).
(iii) A sequence X
0
⇒ X
1
− → Z (resp. Z − → X
0
⇒ X
1
) is exact if Z is
isomorphic to the coequalizer (resp. equalizer) of X
0
⇒X
1
.
(iv) Assume that the category ( admits a zeroobject 0. Let f : X − → Y
be a morphism in (. A cokernel (resp. a kernel) of f, if it exists, is a
cokernel (resp. a kernel) of f, 0: X ⇒Y . It is denoted Coker(f) (resp.
Ker(f)).
The coequalizer L is visualized by the diagram:
X
0
f
g
X
1
h
k
L
.·
X
which means that any morphism h: X
1
− → X such that h ◦ f = h ◦ g factors
uniquely through k.
Note that
k is an epimorphism. (2.20)
Indeed, consider a pair of parallel arrows a, b: L ⇒ X such that a ◦ k =
b ◦ k = h. Then h ◦ f = a ◦ k ◦ f = a ◦ k ◦ g = b ◦ k ◦ g = h ◦ g. Hence h
factors uniquely through k, and this implies a = b.
Dually, the equalizer K is visualized by the diagram:
K
h
X
0
f
g
X
1
X
h
and
h is a monomorphism. (2.21)
We have seen that coproducts and coequalizers (resp. products and equal
izers) are particular cases of inductive (resp. projective) limits. One can show
that conversely, inductive limits (resp. ﬁnite inductive limits) can be obtained
as coequalizer of coproducts (resp. ﬁnite coproducts) and projective limits
(resp. ﬁnite projective limits) can be obtained as equalizer of products (resp.
ﬁnite products).
In particular, a category ( admits ﬁnite projective limits if and only if it
satisﬁes:
2.6. EXACT FUNCTORS 51
(i) ( admits a terminal object,
(ii) for any X, Y ∈ Ob((), the product X Y exists in (,
(iii) for any parallel arrows in (, f, g : X ⇒Y , the equalizer exists in (.
Moreover, if ( admits ﬁnite projective limits, a functor F : ( − → (
t
commutes
with such limits if and only if it commutes with the terminal object, (ﬁnite)
products and kernels.
There is a similar result for ﬁnite inductive limits, replacing a terminal
object by an initial object, products by coproducts and equalizers by co
equalizers.
Proposition 2.5.4. The category Set admits inductive limits. More pre
cisely, if I is a category and α: I − → Set is a functor, then
lim
−→
α · (
¸
i∈I
α(i))/ ∼ where ∼ is the equivalence relation generated by
α(i) ÷ x ∼ y ∈ α(j) if there exists s: i − → j with α(s)(x) = y.
In particular, the coproduct in Set is the disjoint union,
¸
=
¸
.
Proof. Let S ∈ Set. By the deﬁnition of the projective limit in Set we get:
lim
←−
Hom(α, S) · ¦¦p(i, x)¦
i∈I,x∈α(i)
; p(i, x) ∈ S, p(i, x) = p(j, y)
if there exists s: i − → j with α(s)(x) = y¦.
The result follows. q.e.d.
Notation 2.5.5. In the category Set one uses the notation
¸
rather than
¸
.
2.6 Exact functors
Let I, ( and (
t
be categories and let α: I − → (, β : I
op
− → ( and F : ( − → (
t
be functors. Recall that if ( and (
t
admit inductive (resp. projective) limits
indexed by I, there is a natural morphism lim
−→
(F ◦ α) − → F(lim
−→
α) (resp.
F(lim
←−
β) − → lim
←−
(F ◦ β)).
Deﬁnition 2.6.1. Let F : ( − → (
t
be a functor.
(i) Let I be a category and assume that ( admits inductive limits indexed
by I. One says that F commutes with such limits if for any α: I − → (,
lim
−→
(F ◦ α) exits in (
t
and is represented by F(lim
−→
α).
52 CHAPTER 2. THE LANGUAGE OF CATEGORIES
(ii) Similarly if I is a category and ( admits projective limits indexed by
I, one says that F commutes with such limits if for any β : I
op
− → (,
lim
←−
(F ◦ β) exists and is represented by F(lim
←−
β).
Example 2.6.2. Let k be a ﬁeld, ( = (
t
= Mod(k), and let X ∈ (. Then the
functor Hom
k
(X,
•
) does not commute with inductive limit if X is inﬁnite
dimensional.
Deﬁnition 2.6.3. Let F : ( − → (
t
be a functor.
(i) Assume that ( admits ﬁnite projective limits. One says that F is left
exact if it commutes with such limits.
(ii) Assume that ( admits ﬁnite inductive limits. One says that F is right
exact if it commutes with such limits.
(iii) One says that F is exact if it is both left and right exact.
Proposition 2.6.4. Let F : ( − → (
t
be a functor. Assume that
(i) F admits a left adjoint G: (
t
− → (,
(ii) ( admits projective limits indexed by a category I.
Then F commutes with projective limits indexed by I, that is, F(lim
←−
i
β(i)) ·
lim
←−
i
F(β(i)).
Proof. Let β : I
op
− → ( be a projective system indexed by I and let Y ∈ (
t
.
One has the chain of isomorphisms
Hom
(
(Y, F(lim
←−
i
β(i))) · Hom
(
(G(Y ), lim
←−
i
β(i))
· lim
←−
i
Hom
(
(G(Y ), β(i))
· lim
←−
i
Hom
(
(Y, F(β(i)))
· Hom
(
∧(Y, lim
←−
i
F(β(i))).
Then the result follows by the Yoneda lemma. q.e.d.
Of course there is a similar result for inductive limits. If ( admits inductive
limits indexed by I and F admits a right adjoint, then F commutes with
such limits.
2.7. FILTRANT INDUCTIVE LIMITS 53
Proposition 2.6.5. Let ( be a category which admits ﬁnite inductive and
ﬁnite projective limits.
(i) The functor Hom
(
: (
op
( − → Set is left exact.
(ii) Let F : ( − → (
t
be a functor. If F admits a right (resp. left) adjoint,
then F is right (resp. left) exact.
(iii) Let I be a category and assume that ( admits inductive (resp. projec
tive) limits indexed by I. Then the functor lim
−→
: Fct(I, () − → ( (resp.
lim
←−
: Fct(I
op
, () − → () is right (resp. left) exact.
(iv) Let I be a discrete category and let A be a ring. Then the functor
¸
: Mod(A)
I
− → Mod(A) is exact.
Proof. (i) follows immediately from (1.17) and (1.16).
(ii) is a particular case of Proposition 2.6.4.
(iii) Use the isomorphism (2.15) or (2.16).
(iv) is wellknown and obvious. q.e.d.
2.7 Filtrant inductive limits
We shall generalize some notions of Deﬁnition 1.5.2 as well as Lemma 1.5.9
and Proposition 1.5.11.
Deﬁnition 2.7.1. A category I is called ﬁltrant if it satisﬁes the conditions
(i)–(iii) below.
(i) I is non empty,
(ii) for any i and j in I, there exists k ∈ I and morphisms i − → k, j − → k,
(iii) for any parallel morphisms f, g : i ⇒j, there exists a morphism h: j − →
k such that h ◦ f = h ◦ g.
One says that I is coﬁltrant if I
op
is ﬁltrant.
The conditions (ii)–(iii) of being ﬁltrant are visualized by the diagrams:
i
k
j
i
j
k
54 CHAPTER 2. THE LANGUAGE OF CATEGORIES
Of course, if (I, ≤) is a nonempty directed ordered set, then the associated
category I is ﬁltrant.
We shall ﬁrst study ﬁltrant inductive limits in the category Set.
Proposition 2.7.2. Let α: I − → Set be a functor, with I ﬁltrant. Deﬁne the
relation ∼ on
¸
i
α(i) by α(i) ÷ x
i
∼ x
j
∈ α(j) if there exists s: i − → k and
t : j − → k such that α(s)(x
i
) = α(t)(x
j
). Then
(i) the relation ∼ is an equivalence relation,
(ii) lim
−→
α ·
¸
i
α(i)/ ∼.
Proof. (i) The relation ∼ is clearly symmetric and reﬂexive. Let us show it
is transitive. Let x
j
∈ α(i
j
), j = 1, 2, 3 with x
1
∼ x
2
and x
2
∼ x
3
. There
exist morphisms visualized by the diagram:
i
1
s
1
j
1
u
1
i
2
s
2
t
2
k
1
v
l
i
3
t
3
j
2
u
2
such that α(s
1
)x
1
= α(s
2
)x
2
, α(t
2
)x
2
= α(t
3
)x
3
, and v ◦ u
1
◦ s
2
= v ◦ u
2
◦ t
2
.
Set w
1
= v ◦ u
1
◦ s
1
, w
2
= v ◦ u
1
◦ s
2
= v ◦ u
2
◦ t
2
and w
3
= v ◦ u
2
◦ t
3
. Then
α(w
1
)x
1
= α(w
2
)x
2
= α(w
3
)x
3
. Hence x
1
∼ x
3
.
(ii) follows from Proposition 2.5.4. q.e.d.
Corollary 2.7.3. Let α: I − → Set be a functor, with I ﬁltrant.
(i) Let S be a ﬁnite subset in lim
−→
α. Then there exists i ∈ I such that S is
contained in the image of α(i) by the natural map α(i) − → lim
−→
α.
(ii) Let i ∈ I and let x and y be elements of α(i) with the same image in
lim
−→
α. Then there exists s: i − → j such that α(s)(x) = α(s)(y) in α(j).
The proof is left as an exercise.
Corollary 2.7.4. Let A be a ring and denote by for the forgetful functor
Mod(A) − → Set. Then the functor for commutes with ﬁltrant inductive
limits. In other words, if I is ﬁltrant and α: I − → Mod(A) is a functor, then
for ◦ (lim
−→
i
α(i)) = lim
−→
i
(for ◦ α(i)).
2.7. FILTRANT INDUCTIVE LIMITS 55
Inductive limits with values in Set indexed by ﬁltrant categories commute
with ﬁnite projective limits. More precisely:
Proposition 2.7.5. For a ﬁltrant category I, a ﬁnite category J and a func
tor α: I J
op
− → Set, one has lim
−→
i
lim
←−
j
α(i, j)
∼
−→lim
←−
j
lim
−→
i
α(i, j). In other
words, the functor
lim
−→
: Fct(I, Set) − → Set
commutes with ﬁnite projective limits.
Proof. It is enough to prove that lim
−→
commutes with equalizers and with
ﬁnite products. This veriﬁcation is left to the reader. q.e.d.
Corollary 2.7.6. Let A be a ring and let I be a ﬁltrant category. Then the
functor lim
−→
: Mod(A)
I
− → Mod(A) is exact.
Proof. Let α: I J
op
− → Mod(A) be a functor. In order to prove that
lim
−→
i
lim
←−
j
α(i, j) − → lim
←−
j
lim
−→
i
α(i, j) is an isomorphism, it is enough to check it
after applying the functor for : Mod(A) − → Set. Then the result follows from
Corollaries 2.7.4 and 2.7.4. q.e.d.
Coﬁnal functors
Deﬁnition 2.7.7. Let I be a ﬁltrant category and let ϕ: J − → I be a fully
faithful functor. One says that J is coﬁnal to I (or that ϕ : J − → I is coﬁnal)
if for any i ∈ I there exists j ∈ J and a morphism i − → ϕ(j).
Note that the hypothesis implies that J is ﬁltrant.
Proposition 2.7.8. Assume I is ﬁltrant, ϕ: J − → I is fully faithful and
J − → I is coﬁnal. Let α: I − → ( (resp. β : I
op
− → () be a functor. Assume that
lim
−→
α (resp. lim
←−
β) exists in (. Then lim
−→
(α◦ϕ) (resp. lim
←−
(β ◦ϕ
op
)) exists in
( and the natural morphism lim
−→
(α◦ ϕ) − → lim
−→
α (resp. lim
←−
β − → lim
←−
(β ◦ ϕ
op
))
is an isomorphism.
The proof is left as an exercise.
Remark 2.7.9. In these notes, we have skipped problems related to ques
tions of cardinality and universes, but we should have not. Indeed, the reader
will assume that all categories ((, (
t
etc.) belong to a given universe  and
that all limits are indexed by small categories (I, J, etc.). (We do not give
the meaning of “universe” and “small” here.)
56 CHAPTER 2. THE LANGUAGE OF CATEGORIES
Let us give an example which show that, otherwise, we may have troubles.
Let ( be a category which admits products and assume there exist X, Y ∈
( such that Hom
(
(X, Y ) has more than one element. Set M = Mor((),
where Mor(() denotes the “set” of all morphisms in (, and let π = card(M),
the cardinal of the set M. We have Hom
(
(X, Y
M
) · Hom
(
(X, Y )
M
and
therefore card(Hom
(
(X, Y
M
) ≥ 2
π
. On the other hand, Hom
(
(X, Y
M
) ⊂
Mor(() which implies card(Hom
(
(X, Y
M
) ≤ π.
The “contradiction” comes from the fact that ( does not admit products
indexed by such a big set as Mor((), or else, Mor(() is not small (in general)
in the universe to which ( belongs. (The remark was found in [9].)
Exercises to Chapter 2
Exercise 2.1. Prove that the categories Set and Set
op
are not equivalent.
(Hint: if F : Set − → Set
op
were such an equivalence, then F(∅) · ¦pt¦ and
F(¦pt¦) · ∅. Now compare Hom
Set
(¦pt¦, X) and Hom
Set
op(F(¦pt¦), F(X))
when X is a set with two elements.)
Exercise 2.2. (i) Let F : ( − → (
t
be a faithful functor and let f be a mor
phism in (. Prove that if F(f) is a monomorphism (resp. an epimorphism),
then f is a monomorphism (resp. an epimorphism).
(ii) Assume now that F is fully faithful. Prove that if F(f) is an isomorphism,
then f is an isomorphism.
Exercise 2.3. Prove that the category ( is equivalent to the opposite cate
gory (
op
in the following cases:
(i) ( denotes the category of ﬁnite abelian groups,
(ii) ( is the category Rel of relations.
Exercise 2.4. (i) Prove that in the category Set, a morphism f is a mono
morphism (resp. an epimorphism) if and only if it is injective (resp. surjec
tive).
(ii) Prove that in the category of rings, the morphism Z − → Q is an epimor
phism.
Exercise 2.5. Let ( be a category. We denote by id
(
: ( − → ( the identity
functor of ( and by End(id
(
) the set of endomorphisms of the identity functor
id
(
: ( − → (, that is,
End(id
(
) = Hom
Fct((,()
(id
(
, id
(
).
Prove that the composition law on End(id
(
) is commutative.
Exercises to Chapter 2 57
Exercise 2.6. In the category Top, give an example of a morphism which is
both a monomorphism and an epimorphism and which is not an isomorphism.
(Hint: consider a continuous injective map f : X → Y with dense image.)
Exercise 2.7. Let X, Y ∈ ( and consider the category T whose arrows are
triplets Z ∈ (, f : Z − → X, g : Z − → Y , the morphisms being the natural one.
Prove that this category admits a terminal object if and only if the product
X Y exists in (, and that in such a case this terminal object is isomorphic
to XY, XY − → X, XY − → Y . Deduce that if XY exists, it is unique
up to unique isomorphism.
Exercise 2.8. (i) Let I be a (non necessarily ﬁnite) set and (X
i
)
i∈I
a family
of sets indexed by I. Show that
¸
i
X
i
is the disjoint union of the sets X
i
.
(ii) Construct the natural map
¸
i
Hom
Set
(Y, X
i
) − → Hom
Set
(Y,
¸
i
X
i
) and
prove it is injective.
(iii) Prove that the map
¸
i
Hom
Set
(X
i
, Y ) − → Hom
Set
(
¸
i
X
i
, Y ) is not in
jective in general.
Exercise 2.9. Let I and ( be two categories and denote by ∆ the functor
from ( to (
I
which, to X ∈ (, associates the constant functor ∆(X): I ÷
i → X ∈ (, (i − → j) ∈ Mor(I) → id
X
. Assume that any functor from I to (
admits an inductive limit.
(i) Prove that lim
−→
: (
I
− → ( is a functor.
(ii) Prove the formula (for α : I − → ( and Y ∈ ():
Hom
(
(lim
−→
i
α(i), Y ) · Hom
Fct(I,()
(α, ∆(Y )).
(iii) Replacing I with the opposite category, deduce the formula (assuming
projective limits exist):
Hom
(
(X, lim
←−
i
G(i)) · Hom
Fct(I
op
,()
(∆(X), G).
Exercise 2.10. Let ( be a category which admits ﬁltrant inductive limits.
One says that an object X of ( is of ﬁnite type (resp. of ﬁnite presentation) if
for any functor α: I − → ( with I ﬁltrant, the natural map lim
−→
Hom
(
(X, α) − →
Hom
(
(X, lim
−→
α) is injective (resp. bijective).
(i) Show that this deﬁnition coincides with the classical one when ( =
Mod(A), for a ring A.
(ii) Does this deﬁnition coincide with the classical one when ( denotes the
category of commutative algebras?
58 CHAPTER 2. THE LANGUAGE OF CATEGORIES
Exercise 2.11. Let ( be a category and recall that the category (
∧
admits
inductive limits. One denotes by “lim
−→
” the inductive limit in (
∧
. Let k be a
ﬁeld and let ( = Mod(k). Prove that the Yoneda functor h
(
: ( − → (
∧
does
not commute with inductive limits.
Exercise 2.12. Consider the category I with three objects ¦a, b, c¦ and two
morphisms other than the identities, visualized by the diagram
a ←− c − → b.
Let ( be a category. A functor β : I
op
− → ( is nothing but the data of three
objects X, Y, Z and two morphisms visualized by the diagram
X
f
− → Z
g
←− Y.
The ﬁber product X
Z
Y of X and Y over Z, if it exists, is the projective
limit of β.
(i) Assume that ( admits products (of two objects) and kernels. Prove that
the sequence
X
Z
Y − → X ⇒Y
is exact. Here, the two morphisms X ⇒Y are given by f, g.
(ii) Prove that ( admits ﬁnite projective limits if and only if it admits ﬁber
products and a terminal object.
Exercise 2.13. Let I and ( be two categories and let F, G : I ⇒ ( be two
functors. Prove the isomorphism:
Hom
Fct(I,()
(F, G) ·
Ker
¸
i∈I
Hom
(
(F(i), G(i)) ⇒
¸
(j− →k)∈Mor(I)
Hom
(
(F(j), G(k))
.
Here, the double arrow is associated with the two maps:
¸
i∈I
Hom
(
(F(i), G(i)) − → Hom
(
(F(j), G(j)) − → Hom
(
(F(j), G(k)),
¸
i∈I
Hom
(
(F(i), G(i)) − → Hom
(
(F(k), G(k)) − → Hom
(
(F(j), G(k)).
Exercise 2.14. Let ( be a category which admits products of two objects
and a terminal object, denoted by pt
(
. Let A ∈ (. Construct a functor
π
A
: (Set
f
)
op
− → (, (2.22)
such that π
A
(∅) · pt
(
, π
A
(¦pt¦) · A and for S a set with 2 elements,
π
A
(S) · AA.
Chapter 3
Additive categories
Many results or constructions in the category Mod(A) of modules over a
ring A have their counterparts in other contexts, such as ﬁnitely generated
Amodules, or graded modules over a graded ring, or sheaves of Amodules,
etc. Hence, it is natural to look for a common language which avoids to
repeat the same arguments. This is the language of additive and abelian
categories.
In this chapter, we give the main properties of additive categories. We
expose some basic constructions and notions on complexes such as the shift
functor, the homotopy, the mapping cone and the simple complex associated
with a double complex. We also construct complexes associated with simpli
cial objects in an additive category and give a criterion for such a complex
to be homotopic to zero.
3.1 Additive categories
Deﬁnition 3.1.1. A category ( is additive if it satisﬁes conditions (i)(v)
below:
(i) for any X, Y ∈ (, Hom
(
(X, Y ) ∈ Ab,
(ii) the composition law ◦ is bilinear,
(iii) there exists a zero object in (,
(iv) the category ( admits ﬁnite coproducts,
(v) the category ( admits ﬁnite products.
Note that Hom
(
(X, Y ) = ∅ since it is a group. Note that Hom
(
(X, 0) =
Hom
(
(0, X) = 0 for all X ∈ (.
59
60 CHAPTER 3. ADDITIVE CATEGORIES
Notation 3.1.2. If X and Y are two objects of (, one denotes by X ⊕ Y
(instead of X.Y ) their coproduct, and calls it their direct sum. One denotes
as usual by X Y their product. This change of notations is motivated by
the fact that if A is a ring, the forgetful functor Mod(A) − → Set does not
commute with coproducts.
By the deﬁnition of a coproduct and a product in a category, for each
Z ∈ (, there is an isomorphism in Mod(Z):
Hom
(
(X, Z) Hom
(
(Y, Z) · Hom
(
(X ⊕Y, Z), (3.1)
Hom
(
(Z, X) Hom
(
(Z, Y ) · Hom
(
(Z, X Y ). (3.2)
Lemma 3.1.3. Let ( be a category satisfying conditions (i)–(iii) in Deﬁnition
3.1.1. Consider the condition
(vi) for any two objects X and Y in (, there exists Z ∈ ( and morphisms
i
1
: X − → Z, i
2
: Y − → Z, p
1
: Z − → X and p
2
: Z − → Y satisfying
p
1
◦ i
1
= id
X
, p
1
◦ i
2
= 0 (3.3)
p
2
◦ i
2
= id
Y
, p
2
◦ i
1
= 0, (3.4)
i
1
◦ p
1
+i
2
◦ p
2
= id
Z
. (3.5)
Then the conditions (iv), (v) and (vi) are equivalent and the objects X ⊕Y ,
X Y and Z are naturally isomorphic.
Proof. (a) Let us assume condition (iv). The identity of X and the zero
morphism Y − → X deﬁne the morphism p
1
: X⊕Y − → X satisfying (3.3). We
construct similarly the morphism p
2
: X ⊕Y − → Y satisfying (3.4). To check
(3.5), we use the fact that if f : X ⊕ Y − → X ⊕ Y satisﬁes f ◦ i
1
= i
1
and
f ◦ i
2
= i
2
, then f = id
X⊕Y
.
(b) Let us assume condition (vi). Let W ∈ ( and consider morphisms
f : X − → W and g : Y − → W. Set h := f ◦ p
1
⊕ g ◦ p
2
. Then h: Z − → W
satisﬁes h◦i
1
= f and h◦i
2
= g and such an h is unique. Hence Z · X⊕Y .
(c) We have proved that conditions (iv) and (vi) are equivalent and moreover
that if they are satisﬁed, then Z · X ⊕ Y . Replacing ( with (
op
, we get
that these conditions are equivalent to (v) and Z · X Y . q.e.d.
Example 3.1.4. (i) If A is a ring, Mod(A) and Mod
f
(A) are additive cate
gories.
(ii) Ban, the category of CBanach spaces and linear continuous maps is
additive.
(iii) If ( is additive, then (
op
is additive.
3.2. COMPLEXES IN ADDITIVE CATEGORIES 61
(iv) Let I be category. If ( is additive, the category (
I
of functors from I to
(, is additive.
(v) If ( and (
t
are additive, then ( (
t
are additive.
Let F : ( − → (
t
be a functor of additive categories. One says that F is
additive if for X, Y ∈ (, Hom
(
(X, Y ) − → Hom
(
(F(X), F(Y )) is a morphism
of groups. One can prove the following
Proposition 3.1.5. Let F : ( − → (
t
be a functor of additive categories. Then
F is additive if and only if it commutes with direct sum, that is, for X and
Y in (:
F(0) · 0
F(X ⊕Y ) · F(X) ⊕F(Y ).
Unless otherwise speciﬁed, functors between additive categories will be
assumed to be additive.
Example 3.1.6. Let ( be an additive category. One shall be aware that the
bifunctor (
op
( − → Mod(Z) is separately additive with respect to each of
its argument, but is not additive as a functor on the product category.
Generalization: Let k be a commutative ring. One deﬁnes the notion of a
kadditive category by assuming that for X and Y in (, Hom
(
(X, Y ) is a
kmodule and the composition is kbilinear.
3.2 Complexes in additive categories
Let ( denote an additive category.
Deﬁnition 3.2.1. (i) A diﬀerential object (X
•
, d
•
X
) in ( is a sequence of
objects X
k
and morphisms d
k
(k ∈ Z):
− → X
k−1
d
k−1
−−→ X
k
d
k
−→ X
k+1
− → . (3.6)
(ii) A complex is a diﬀerential object (X
•
, d
•
X
) such that such that d
k
◦
d
k−1
= 0 for all k ∈ Z.
A morphism of diﬀerential objects f
•
: X
•
− → Y
•
is visualized by a com
mutative diagram:
X
n
f
n
d
n
X
X
n+1
f
n+1
Y
n
d
n
Y
X
n+1
62 CHAPTER 3. ADDITIVE CATEGORIES
One deﬁnes naturally the direct sum of two diﬀerential objects. Hence, we
get a new additive category, the category Diff(() of diﬀerential objects in
(. One denotes by C(() the full additive subcategory of Diff(() consisting
of complexes.
From mow on, we shall concentrate our study on the category C(().
A complex is bounded (resp. bounded below, bounded above) if X
n
= 0
for [n[ >> 0 (resp. n << 0, n >> 0). One denotes by C
∗
(()(∗ = b, +, −)
the full additive subcategory of C(() consisting of bounded complexes (resp.
bounded below, bounded above).
One considers ( as a full subcategory of C
b
(() by identifying an object
X ∈ ( with the complex X
•
“concentrated in degree 0”:
X
•
:= − → 0 − → X − → 0 − →
where X stands in degree 0.
Shift functor
Let X ∈ C(() and k ∈ Z. One deﬁnes the shifted complex X[k] by:
(X[k])
n
= X
n+k
d
n
X[k]
= (−1)
k
d
n+k
X
If f : X − → Y is a morphism in C(() one deﬁnes f[k] : X[k] − → Y [k] by
(f[k])
n
= f
n+k
.
The shift functor X → X[1] is an automorphism (i.e. an invertible func
tor) of C(().
Homotopy
Let ( denote an additive category.
Deﬁnition 3.2.2. (i) A morphism f : X − → Y in C(() is homotopic to
zero if for all k there exists a morphism s
k
: X
k
− → Y
k−1
such that:
f
k
= s
k+1
◦ d
k
X
+d
k−1
Y
◦ s
k
.
Two morphisms f, g : X − → Y are homotopic if f − g is homotopic to
zero.
(ii) An object X in C(() is homotopic to 0 if id
X
is homotopic to zero.
3.2. COMPLEXES IN ADDITIVE CATEGORIES 63
A morphism homotopic to zero is visualized by the diagram (which is not
commutative):
X
k−1
X
k
s
k
.
f
k
d
k
X
X
k+1
s
k+1
.
Y
k−1
d
k−1
Y
Y
k
Y
k+1
Note that an additive functor sends a morphism homotopic to zero to a
morphism homotopic to zero.
Example 3.2.3. The complex 0 − → X
t
− → X
t
⊕X
tt
− → X
tt
− → 0 is homotopic
to zero.
Mapping cone
Deﬁnition 3.2.4. Let f : X − → Y be a morphism in C((). The mapping
cone of f, denoted Mc(f), is the object of C(() deﬁned by:
Mc(f)
k
= (X[1])
k
⊕Y
k
d
k
Mc(f)
=
d
k
X[1]
0
f
k+1
d
k
Y
Of course, before to state this deﬁnition, one should check that d
k+1
Mc(f)
◦
d
k
Mc(f)
= 0. Indeed:
−d
k+2
X
0
f
k+2
d
k+1
Y
◦
−d
k+1
X
0
f
k+1
d
k
Y
= 0
Notice that although Mc(f)
k
= (X[1])
k
⊕ Y
k
, Mc(f) is not isomorphic to
X[1] ⊕Y in C(() unless f is the zero morphism.
There are natural morphisms of complexes
α(f) : Y − → Mc(f), β(f) : Mc(f) − → X[1].
and β(f) ◦ α(f) = 0.
If F : ( − → (
t
is an additive functor, then F(Mc(f)) · Mc(F(f)).
The homotopy category K(()
Let ( be an additive category.
64 CHAPTER 3. ADDITIVE CATEGORIES
Starting with C((), we shall construct a new category by deciding that
a morphism of complexes homotopic to zero is isomorphic to the zero mor
phism. Set:
Ht(X, Y ) = ¦f : X − → Y ; f is homotopic to 0¦.
If f : X − → Y and g : Y − → Z are two morphisms in C(() and if f or g is
homotopic to zero, then g ◦ f is homotopic to zero. This allows us to state:
Deﬁnition 3.2.5. The homotopy category K(() is deﬁned by:
Ob(K(()) = Ob(C(())
Hom
K(()
(X, Y ) = Hom
C(()
(X, Y )/Ht(X, Y )
In other words, a morphism homotopic to zero in C(() becomes the zero
morphism in K(() and a homotopy equivalence becomes an isomorphism.
One deﬁnes similarly K
∗
((), (∗ = b, +, −). They are clearly additive
categories, endowed with an automorphism, the shift functor [1] : X → X[1].
3.3 Simplicial constructions
We
1
shall deﬁne the simplicial category and use it to construct complexes
and homotopies in additive categories.
Deﬁnition 3.3.1. (a) The simplicial category, denoted by ∆, is the cate
gory whose objects are the ﬁnite totally ordered sets and the morphisms
are the orderpreserving maps.
(b) We denote by ∆
inj
the subcategory of ∆ such that Ob(∆
inj
) = Ob(∆),
the morphisms being the injective orderpreserving maps.
For integers n, m denote by [n, m] the totally ordered set ¦k ∈ Z; n ≤
k ≤ m¦.
Proposition 3.3.2. (i) the natural functor ∆ − → Set
f
is faithful,
(ii) the full subcategory of ∆ consisting of objects ¦[0, n]¦
n≥−1
is equivalent
to ∆,
(iii) ∆ admits an initial object, namely ∅, and a terminal object, namely
¦0¦.
1
This section has not been treated in 2005/2006
3.3. SIMPLICIAL CONSTRUCTIONS 65
The proof is obvious.
Let us denote by
d
n
i
: [0, n]− →[0, n + 1] (0 ≤ i ≤ n + 1)
the injective orderpreserving map which does not take the value i. In other
words
d
n
i
(k) =
k for k < i,
k + 1 for k ≥ i.
One checks immediately that
d
n+1
j
◦ d
n
i
= d
n+1
i
◦ d
n
j−1
for 0 ≤ i < j ≤ n + 2. (3.7)
Indeed, both morphisms are the unique injective orderpreserving map which
does not take the values i and j.
The category ∆
inj
is visualized by
∅
d
−1
0
¦0¦
d
0
0
d
0
1
¦0, 1¦
d
1
0
d
1
1
d
1
2
¦0, 1, 2¦
(3.8)
Let ( be an additive category and F : ∆
inj
− → ( a functor. We set for
n ∈ Z:
F
n
=
F([0, n]) for n ≥ −1,
0 otherwise,
d
n
F
: F
n
− → F
n+1
, d
n
F
=
n+1
¸
i=0
(−)
i
F(d
n
i
).
Consider the diﬀerential object
F
•
:= − → 0 − → F
−1
d
−1
F
−−→ F
0
d
0
F
−→ F
1
− → − → F
n
d
n
F
−→ . (3.9)
Theorem 3.3.3. (i) The diﬀerential object F
•
is a complex.
(ii) Assume that there exist morphisms s
n
F
: F
n
− → F
n−1
(n ≥ 0)satisfying:
s
n+1
F
◦ F(d
n
0
) = id
F
n for n ≥ −1,
s
n+1
F
◦ F(d
n
i+1
) = F(d
n−1
i
) ◦ s
n
F
for i > 0, n ≥ 0.
Then F
•
is homotopic to zero.
66 CHAPTER 3. ADDITIVE CATEGORIES
Proof. (i) By (3.7), we have
d
n+1
F
◦ d
n
F
=
n+2
¸
j=0
n+1
¸
i=0
(−)
i+j
F(d
n+1
j
◦ d
n
i
)
=
¸
0≤j≤i≤n+1
(−)
i+j
F(d
n+1
j
◦ d
n
i
) +
¸
0≤i<j≤n+2
(−)
i+j
F(d
n+1
j
◦ d
n
i
)
=
¸
0≤j≤i≤n+1
(−)
i+j
F(d
n+1
j
◦ d
n
i
) +
¸
0≤i<j≤n+2
(−)
i+j
F(d
n+1
i
◦ d
n
j−1
)
= 0 .
Here, we have used
¸
0≤i<j≤n+2
(−)
i+j
F(d
n+1
i
◦ d
n
j−1
) =
¸
0≤i<j≤n+1
(−)
i+j+1
F(d
n+1
i
◦ d
n
j
)
=
¸
0≤j≤i≤n+1
(−)
i+j+1
F(d
n+1
j
◦ d
n
i
).
(ii) We have
s
n+1
F
◦ d
n
F
+d
n−1
F
◦ s
n
=
n+1
¸
i=0
(−1)
i
s
n+1
F
◦ F(d
n
i
) +
n
¸
i=0
(−1)
i
F(d
n−1
i
◦ s
n
F
)
= s
n+1
F
◦ F(d
n
0
) +
n
¸
i=0
(−1)
i+1
s
n+1
F
◦ F(d
n
i+1
) +
n
¸
i=0
(−1)
i
F(d
n−1
i
◦ s
n
F
)
= id
F
n +
n
¸
i=0
(−1)
i+1
F(d
n−1
i
◦ s
n
F
) +
n
¸
i=0
(−1)
i
F(d
n−1
i
◦ s
n
F
)
= id
F
n.
q.e.d.
3.4 Double complexes
Let ( be as above an additive category. A double complex (X
•,•
, d
X
) in ( is
the data of
¦X
n,m
, d
t
n,m
X
, d
tt
n,m
X
; (n, m) ∈ Z Z¦
where X
n,m
∈ ( and the “diﬀerentials” d
t
n,m
X
: X
n,m
− → X
n+1,m
, d
tt
n,m
X
:
X
n,m
− → X
n,m+1
satisfy:
(3.10) d
t
2
X
= d
tt
2
X
= 0, d
t
◦ d
tt
= d
tt
◦ d
t
.
3.4. DOUBLE COMPLEXES 67
One can represent a double complex by a commutative diagram:
X
n,m
d
n,m
d
n,m
X
n,m+1
d
n,m+1
X
n+1,m
d
n+1,m
X
n+1,m+1
One deﬁnes naturally the notion of a morphism of double complexes, and
one obtains the additive category C
2
(() of double complexes.
There are two functors F
I
, F
II
: C
2
(() − → C(C(()) which associate to
a double complex X the complex whose objects are the rows (resp. the
columns) of X. These two functors are clearly isomorphisms of categories.
Now consider the ﬁniteness condition:
(3.11) for all p ∈ Z, ¦(m, n) ∈ Z Z; X
n,m
= 0, m+n = p¦ is ﬁnite
and denote by C
2
f
(() the full subcategory of C
2
(() consisting of objects X
satisfying (3.11). To such an X one associates its “total complex” tot(X) by
setting:
tot(X)
p
= ⊕
m+n=p
X
n,m
,
d
p
tot(X)
[
X
n,m = d
t
n,m
+ (−1)
n
d
tt
n,m
.
This is visualized by the diagram:
X
n,m
(−)
n
d
d
X
n,m+1
X
n+1,m
Proposition 3.4.1. The diﬀerential object ¦tot(X)
p
, d
p
tot(X)
¦
p∈Z
is a complex
(i.e. d
p+1
tot(X)
◦ d
p
tot(X)
= 0) and tot : C
2
f
(() − → C(() is a functor of additive
categories.
Proof. For (n, m) ∈ Z Z, one has
d ◦ d(X
n,m
) = d
tt
◦ d
tt
(X
n,m
) +d
t
◦ d
t
(X
n,m
)
+(−)
n
d
tt
◦ d
t
(X
n,m
) + (−)
n+1
d
t
◦ d
tt
(X
n,m
)
= 0.
It is left to the reader to check that tot is an additive functor. q.e.d.
68 CHAPTER 3. ADDITIVE CATEGORIES
Example 3.4.2. Let f
•
: X
•
− → Y
•
be a morphism in C((). Consider the
double complex Z
•,•
such that Z
−1,•
= X
•
, Z
0,•
= Y
•
, Z
i,•
= 0 for i = −1, 0,
with diﬀerentials f
j
: Z
−1,j
− → Z
0,j
. Then
tot(Z
•,•
) · Mc(f
•
). (3.12)
Bifunctor
Let (, (
t
and (
tt
be additive categories and let F : ( (
t
− → (
tt
be an additive
bifunctor (i.e., F(
•
,
•
) is additive with respect to each argument). It deﬁnes
an additive bifunctor C
2
(F) : C(() C((
t
) − → C
2
((
tt
). In other words, if
X ∈ C(() and X
t
∈ C((
t
) are complexes, then C
2
(F)(X, X
t
) is a double
complex.
Examples 3.4.3. (i) Consider the bifunctor Hom
(
: (
op
( − → Mod(Z).
We shall write Hom
•,•
(
instead of C
2
(Hom
(
). If X and Y are two objects of
C((), one has
Hom
•,•
(
(X, Y )
n,m
= Hom
(
(X
−m
, Y
n
),
d
tn,m
= Hom
(
(X
−m
, d
n
Y
), d
ttn,m
= Hom
(
((−)
n
d
−n−1
X
, Y
m
).
Note that Hom
•,•
(
(X, Y ) is a double complex in the category Ab, which
should not be confused with the group Hom
C(()
(X, Y ).
(ii) Consider the bifunctor
•
⊗
•
: Mod(A
op
) Mod(A) − → Mod(Z). We
shall simply write ⊗ instead of C
2
(⊗). Hence, for X ∈ C
−
(Mod(A
op
)) and
Y ∈ C
−
(Mod(A)), one has
(X ⊗Y )
n,m
= X
n
⊗Y
m
,
d
t
n,m
= d
n
X
⊗Y
m
, d
t
/
n,m
= X
n
⊗d
m
Y
.
Deﬁnition 3.4.4. Let X ∈ C
−
(() and Y ∈ C
+
((). One sets
Hom
•
(
(X, Y ) = tot(Hom
•,•
(
(X, Y )). (3.13)
Exercises to Chapter 3
Exercise 3.1. Let ( be an additive category and let X ∈ C((). By using
±d
p
X
: X
p
− → X
p+1
(p ∈ Z), construct a morphism d
X
: X − → X[1] in C(()
and prove that d
X
: X − → X[1] is homotopic to zero.
Exercises to Chapter 3 69
Exercise 3.2. Let ( be an additive category, f, g : X ⇒ Y two morphisms
in C((). Prove that f and g are homotopic if and only if there exists a
commutative diagram in C(()
Y
α(f)
Mc(f)
u
β(f)
X[1]
Y
α(g)
Mc(f)
β(g)
X[1].
In such a case, prove that u is an isomorphism in C(().
Exercise 3.3. Let ( be an additive category and let f : X − → Y be a mor
phism in C(().
Prove that the following conditions are equivalent:
(a) f is homotopic to zero,
(b) f factors through α(id
X
): X − → Mc(id
X
),
(c) f factors through β(id
Y
)[−1] : Mc(id
Y
)[−1] − → Y ,
(d) f decomposes as X − → Z − → Y with Z a complex homotopic to zero.
Exercise 3.4. A category with translation (/, T) is a category / together
with an equivalence T : / − → /. A diﬀerential object (X, d
X
) in a cate
gory with translation (/, T) is an object X ∈ / together with a morphism
d
X
: X − → T(X). A morphism f : (X, d
X
) − → (Y, d
Y
) of diﬀerential objects is
a commutative diagram
X
f
d
X
TX
T(f)
Y
d
Y
TY.
One denotes by /
d
the subcategory of (/, T) consisting of diﬀerential objects
and morphisms of such objects. If / is additive, one says that a diﬀerential
object (X, d
X
) in (/, T) is a complex if the composition X
d
X
−→ T(X)
T(d
X
)
−−−→
T
2
(X) is zero. One denotes by /
c
the full subcategory of /
d
consisting of
complexes.
(i) Let ( be a category. Denote by Z
d
the set Z considered as a discrete
category and still denote by Z the ordered set (Z, ≤) considered as a category.
Prove that (
Z
:= Fct(Z
d
, () is a category with translation.
(ii) Show that the category Fct(Z, () may be identiﬁed to the category of
diﬀerential objects in (
Z
.
70 CHAPTER 3. ADDITIVE CATEGORIES
(iii) Let ( be an additive category. Show that the notions of diﬀerential
objects and complexes given above coincide with those in Deﬁnition 3.2.1
when choosing / = C(() and T = [1].
Exercise 3.5. Consider the category ∆ and for n > 0, denote by
s
n
i
: [0, n]− →[0, n −1] (0 ≤ i ≤ n −1)
the surjective orderpreserving map which takes the same value at i and i+1.
In other words
s
n
i
(k) =
k for k ≤ i,
k −1 for k > i.
Checks the relations:
s
n
j
◦ s
n+1
i
= s
n
i−1
◦ s
n+1
j
for 0 ≤ j < i ≤ n,
s
n+1
j
◦ d
n
i
= d
n−1
i
◦ s
n
j−1
for 0 ≤ i < j ≤ n,
s
n+1
j
◦ d
n
i
= id
[0,n]
for 0 ≤ i ≤ n + 1, i = j, j + 1,
s
n+1
j
◦ d
n
i
= d
n−1
i−1
◦ s
n
j
for 1 ≤ j + 1 < i ≤ n + 1.
Exercise 3.6. Let M be an Amodule and let ϕ = (ϕ
1
, . . . , ϕ
n
) be n endo
morphisms of M over A which satisfy:
ϕ
i
= 0 or ϕ
i
= id
M
.
Prove that the Koszul complex K
•
(M, ϕ) is homotopic to zero.
Chapter 4
Abelian categories
In this chapter, we give the main properties of abelian categories and expose
some basic constructions on complexes in such categories, such as the snake
Lemma. We explain the notion of injective resolutions and apply it to the
construction of derived functors, with applications to the functors Ext and
Tor.
For sake of simplicity, we shall always argue as if we were working in a full
abelian subcategory of Mod(A) for a ring A. (See Convention 4.1.1 below.)
Some important historical references are the book [5] and the paper [14].
4.1 Abelian categories
Convention 4.1.1. In these Notes, when dealing with an abelian category
( (see Deﬁnition 4.1.4 below), we shall assume that ( is a full abelian sub
category of a category Mod(A) for some ring A. This makes the proofs
much easier and moreover there exists a famous theorem (due to Freyd &
Mitchell) that asserts that this is in fact always the case (up to equivalence
of categories).
From now on, (, (
t
will denote additive categories.
Let f : X − → Y be a morphism in (. Recall that if Ker f exists, it is
unique up to unique isomorphism, and for any W ∈ (, the sequence
0 − → Hom
(
(W, Ker f) − → Hom
(
(W, X)
f
− → Hom
(
(W, Y ) (4.1)
is exact in Mod(Z).
Similarly, if Coker f exists, then for any W ∈ (, the sequence
0 − → Hom
(
(Coker f, W) − → Hom
(
(Y, W)
f
− → Hom
(
(X, W) (4.2)
is exact in Mod(Z).
71
72 CHAPTER 4. ABELIAN CATEGORIES
Example 4.1.2. Let A be a ring, I an ideal which is not ﬁnitely generated
and let M = A/I. Then the natural morphism A − → M in Mod
f
(A) has no
kernel.
Let ( be an additive category which admits kernels and cokernels. Let
f : X − → Y be a morphism in (. One deﬁnes:
Coimf = Coker h, where h : Ker f − → X
Imf = Ker k, where k : Y − → Coker f.
Consider the diagram:
Ker f
h
X
f
s
Y
k
Coker f
Coimf
˜
f
u
Imf
Since f ◦h = 0, f factors uniquely through
˜
f, and k ◦f factors through k ◦
˜
f.
Since k ◦ f = k ◦
˜
f ◦ s = 0 and s is an epimorphism, we get that k ◦
˜
f = 0.
Hence
˜
f factors through Ker k = Imf. We have thus constructed a canonical
morphism:
Coimf
u
− → Imf. (4.3)
Examples 4.1.3. (i) If A is a ring and f is a morphism in Mod(A), then
(4.3) is an isomorphism.
(ii) The category Ban admits kernels and cokernels. If f : X − → Y is a
morphism of Banach spaces, deﬁne Ker f = f
−1
(0) and Coker f = Y/Imf
where Imf denotes the closure of the space Imf. It is wellknown that there
exist continuous linear maps f : X − → Y which are injective, with dense and
non closed image. For such an f, Ker f = Coker f = 0 although f is not an
isomorphism. Thus Coimf · X and Imf · Y . Hence, the morphism (4.3)
is not an isomorphism.
Deﬁnition 4.1.4. Let ( be an additive category. One says that ( is abelian
if:
(i) any f : X − → Y admits a kernel and a cokernel,
(ii) for any morphism f in (, the natural morphism Coimf − → Imf is an
isomorphism.
In an abelian category, a morphism f is a monomorphism (resp. an
epimorphism) if and only if Ker f · 0 (resp. Coker f · 0). If f is both a
monomorphism and an epimorphism, it is an isomorphism.
4.1. ABELIAN CATEGORIES 73
Examples 4.1.5. (i) If A is a ring, Mod(A) is an abelian category.
(ii) If A is noetherian, then Mod
f
(A) is abelian.
(iii) The category Ban admits kernels and cokernels but is not abelian. (See
Examples 4.1.3 (ii).)
(iv) Let I be category. Then if ( is abelian, the category (
I
of functors
from I to (, is abelian. For example, if F, G : I − → ( are two functors
and ϕ : F − → G is a morphism of functors, deﬁne the functor Ker ϕ by
Ker ϕ(X) = Ker(F(X) − → G(X)). Then clearly, Ker ϕ is a kernel of ϕ. One
deﬁnes similarly the cokernel.
(v) If ( is abelian, then the opposite category (
op
is abelian.
Unless otherwise speciﬁed, we assume until the end of this chapter that
( is abelian.
One naturally extends Deﬁnition 1.2.1 to abelian categories. Consider a
sequence of morphisms X
t
f
− → X
g
− → X
tt
with g ◦ f = 0 (sometimes, one calls
such a sequence a complex). It deﬁnes a morphism Coimf − → Ker g, hence,
( being abelian, a morphism Imf − → Ker g.
Deﬁnition 4.1.6. (i) One says that a sequence X
t
f
− → X
g
− → X
tt
with
g ◦ f = 0 is exact if Imf
∼
−→Ker g.
(ii) More generally, a sequence of morphisms X
p
d
p
−→ − → X
n
with d
i+1
◦
d
i
= 0 for all i ∈ [p, n−1] is exact if Imd
i ∼
−→Ker d
i+1
for all i ∈ [p, n−1].
(iii) A short exact sequence is an exact sequence 0 − → X
t
− → X − → X
tt
− → 0
Any morphism f : X − → Y may be decomposed into short exact se
quences:
0 − → Ker f − → X − → Imf − → 0
0 − → Imf − → Y − → Coker f − → 0.
Proposition 4.1.7. Let 0 − → X
t
f
− → X
g
− → X
tt
− → 0 be a short exact sequence
in (. Then the conditions (i) to (iii) are equivalent.
(i) there exists h : X
tt
− → X such that g ◦ h = id
X
,
(ii) there exists k : X − → X
t
such that k ◦ f = id
X
,
(iii) there exists ϕ = (k, g) and ψ = (f + h) such that X
ϕ
−→ X
t
⊕ X
tt
and
X
t
⊕X
tt
ψ
−→ X are isomorphisms inverse to each other,
74 CHAPTER 4. ABELIAN CATEGORIES
The proof is similar to the case of Amodules and is left as an exercise.
If the conditions of the above proposition are satisﬁed, one says that the
sequence splits.
Note that an additive functor of abelian categories sends split exact se
quences into split exact sequences.
Lemma 4.1.8. (The “ﬁve lemma”.) Consider a commutative diagram:
X
0
f
0
α
0
X
1
f
1
α
1
X
2
f
2
α
2
X
3
f
3
Y
0
β
0
Y
1
β
1
Y
2
β
2
Y
3
and assume that the rows are exact sequences.
(i) If f
0
is an epimorphism and f
1
, f
3
are monomorphisms, then f
2
is a
monomorphism.
(ii) If f
3
is a monomorphism, and f
0
, f
2
are epimorphisms, then f
1
is an
epimorphism.
According to Convention 4.1.1, we shall assume that ( is a full abelian
subcategory of Mod(A) for some ring A. Hence we may choose elements in
the objects of (.
Proof. (i) Let x
2
∈ X
2
and assume that f
2
(x
2
) = 0. Then f
3
◦ α
2
(x
2
) = 0
and f
3
being a monomorphism, this implies α
2
(x
2
) = 0. Since the ﬁrst row
is exact, there exists x
1
∈ X
1
such that α
1
(x
1
) = x
2
. Set y
1
= f
1
(x
1
). Since
β
1
◦ f
1
(x
1
) = 0 and the second row is exact, there exists y
0
∈ Y
0
such that
β
0
(y
0
) = f
1
(x
1
). Since f
0
is an epimorphism, there exists x
0
∈ X
0
such
that y
0
= f
0
(x
0
). Since f
1
◦ α
0
(x
0
) = f
1
(x
1
) and f
1
is a monomorphism,
α
0
(x
0
) = x
1
. Therefore, x
2
= α
1
(x
1
) = 0.
(ii) is nothing but (i) in (
op
. q.e.d.
Let F : ( − → (
t
be an additive functor of abelian categories. Since F
is additive, F(0) · 0 and F(X ⊕ Y ) · F(X) ⊕ F(Y ). In other words, F
commutes with ﬁnite direct sums (and with ﬁnite products).
Let F : ( − → (
t
be an additive functor. Recall that F is left exact if and
only if it commutes with kernels, that is, if and only if for any exact sequence
in (, 0 − → X
t
− → X − → X
tt
the sequence 0 − → F(X
t
) − → F(X) − → F(X
tt
) is
exact in (
t
.
Similarly, F is right exact if and only if it commutes with cokernels, that
is, if and only if for any exact sequence in (, X
t
− → X − → X
tt
− → 0 the
sequence F(X
t
) − → F(X) − → F(X
tt
) − → 0 is exact.
4.2. COMPLEXES IN ABELIAN CATEGORIES 75
Lemma 4.1.9. Let F( − → (
t
be an additive functor.
(i) F is left exact if and only if for any exact sequence 0 − → X
t
− → X − →
X
tt
− → 0 in (, the sequence 0 − → F(X
t
) − → F(X) − → F(X
tt
) is exact.
(ii) F is exact if and only if for any exact sequence X
t
− → X − → X
tt
in (,
the sequence F(X
t
) − → F(X) − → F(X
tt
) is exact.
The proof is left as an exercise.
Examples 4.1.10. (i) Let ( be an abelian category. The functor Hom
(
from (
op
( to Mod(Z) is left exact.
(ii) Let A be a kalgebra. Let M and N in Mod(A). It follows from (i) that
the functors Hom
A
from Mod(A)
op
Mod(A) to Mod(k) is left exact.
The functors ⊗
A
from Mod(A
op
) Mod(A) to Mod(k) is right exact.
If A is a ﬁeld, all the above functors are exact.
(iii) Let I and ( be two categories with ( abelian. Assume that ( admits
inductive limits. Recall that the functor lim
−→
: Fct(I, () − → ( is right exact.
If ( = Mod(A) and I is ﬁltrant, then the functor lim
−→
is exact.
Similarly, if ( admits projective limits, the functor lim
←−
: Fct(I
op
, () − → (
is left exact. If ( = Mod(A) and I is discrete, the functor lim
←−
(that is, the
functor
¸
) is exact.
4.2 Complexes in abelian categories
We assume that ( is abelian. Notice ﬁrst that the categories C
∗
(() are clearly
abelian for ∗ = ∅, +, −, b. For example, if f : X − → Y is a morphism in C((),
the complex Z deﬁned by Z
n
= Ker(f
n
: X
n
− → Y
n
), with diﬀerential induced
by those of X, will be a kernel for f, and similarly for Coker f.
Let X ∈ C((). One deﬁnes the following objects of (:
Z
k
(X) := Ker d
k
X
B
k
(X) := Imd
k−1
X
H
k
(X) := Z
k
(X)/B
k
(X) (:= Coker(B
k
(X) − → Z
k
(X)))
One calls H
k
(X) the kth cohomology object of X. If f : X − → Y is a mor
phism in C((), then it induces morphisms Z
k
(X) − → Z
k
(Y ) and B
k
(X) − →
B
k
(Y ), thus a morphism H
k
(f) : H
k
(X) − → H
k
(Y ). Clearly,H
k
(X ⊕ Y ) ·
H
k
(X) ⊕H
k
(Y ). Hence we have obtained an additive functor:
H
k
(
•
) : C(() − → (.
76 CHAPTER 4. ABELIAN CATEGORIES
Notice that:
H
k
(X) = H
0
(X[k]).
Lemma 4.2.1. Let ( be an abelian category and let f : X − → Y be a mor
phism in C(() homotopic to zero. Then H
k
(f) : H
k
(X) − → H
k
(Y ) is the 0
morphism.
Proof. Let f
k
= s
k+1
◦d
k
X
+d
k−1
Y
◦s
k
. Then d
k
X
= 0 on Ker d
k
X
and d
k−1
Y
◦s
k
= 0
on Ker d
k
Y
/ Imd
k−1
Y
. Hence H
k
(f) : Ker d
k
X
/ Imd
k−1
X
− → Ker d
k
Y
/ Imd
k−1
Y
is the
zero morphism. q.e.d.
In view of Lemma 4.2.1, the functor H
0
: C(() − → ( extends as a functor
H
0
: K(() − → (.
Deﬁnition 4.2.2. One says that a morphism f : X − → Y in C(() or in K(()
is a quasiisomorphism (a qis, for short) if H
k
(f) is an isomorphism for all
k ∈ Z. In such a case, one says that X and Y are quasiisomorphic.
In particular, X ∈ C(() is qis to 0 if and only if the complex X is exact.
Remark 4.2.3. By Lemma 4.2.1, a complex homotopic to 0 is qis to 0, but
the converse is false. For example, a short exact sequence does not necessarily
split. One shall be aware that the property for a complex of being homotopic
to 0 is preserved when applying an additive functor, contrarily to the property
of being qis to 0.
Remark 4.2.4. Consider a bounded complex X
•
and denote by Y
•
the
complex given by Y
j
= H
j
(X
•
), d
j
Y
≡ 0. One has:
Y
•
= ⊕
i
H
i
(X
•
)[−i]. (4.4)
The complexes X
•
and Y
•
have the same cohomology objects. In other words,
H
j
(Y
•
) · H
j
(X
•
). However, in general these isomorphisms are neither
induced by a morphism from X
•
− → Y
•
, nor by a morphism from Y
•
− → X
•
,
and the two complexes X
•
and Y
•
are not quasiisomorphic.
There are exact sequences
X
k−1
− → Ker d
k
X
− → H
k
(X) − → 0,
0 − → H
k
(X) − → Coker d
k−1
X
− → X
k+1
,
which give rise to the exact sequence:
0 − → H
k
(X) − → Coker(d
k−1
X
)
d
k
X
−→ Ker d
k+1
X
− → H
k+1
(X) − → 0. (4.5)
4.2. COMPLEXES IN ABELIAN CATEGORIES 77
Lemma 4.2.5. (The snake lemma.) Consider the commutative diagram in
( below with exact rows:
X
f
α
Y
g
β
Z
γ
0
0
X
t
f
Y
t
g
Z
t
Then it gives rise to an exact sequence:
Ker α − → Ker β − → Ker γ
ϕ
−→ Coker α − → Coker β − → Coker γ.
The proof is similar to that of Lemma 4.1.8 and is left as an exercise.
Theorem 4.2.6. Let 0 − → X
t
f
− → X
g
− → X
tt
− → 0 be an exact sequence in C(().
(i) For each k ∈ Z, the sequence H
k
(X
t
) − → H
k
(X) − → H
k
(X
tt
) is exact.
(ii) For each k ∈ Z, there exists δ
k
: H
k
(X
tt
) − → H
k+1
(X
t
) making the
sequence:
H
k
(X) − → H
k
(X
tt
)
δ
k
−→ H
k+1
(X
t
) − → H
k+1
(X) (4.6)
exact. Moreover, one can construct δ
k
functorial with respect to short
exact sequences of C(().
Proof. The exact sequence in C(() gives rise to commutative diagrams with
exact rows:
Coker d
k−1
X
d
k
X
f
Coker d
k−1
X
d
k
X
g
Coker d
k−1
X
d
k
X
0
0
Ker d
k+1
X
f
Ker d
k+1
X g
Ker d
k+1
X
Then using the exact sequence (4.5), the result follows from Lemma 4.2.5.
q.e.d.
Remark 4.2.7. Let us denote for a while by δ
k
(f, g) the map δ
k
con
structed in Theorem 4.2.6. Then one can prove that δ
k
(−f, g) = δ
k
(f, −g) =
−δ
k
(f, g).
Corollary 4.2.8. In the situation of Theorem 4.2.6, if two of the complexes
X
t
, X, X
tt
are exact, so is the third one.
78 CHAPTER 4. ABELIAN CATEGORIES
Corollary 4.2.9. Let f : X − → Y be a morphism in C((). Then there is a
long exact sequence
− → H
k
(X)
H
k
(f)
−−−→ H
k
(Y ) − → H
k+1
(Mc(f)) − →
Proof. There are natural morphisms Y − → Mc(f) and Mc(f) − → X[1] which
give rise to an exact sequence in C(():
0 − → Y − → Mc(f) − → X[1] − → 0. (4.7)
Applying Theorem 4.2.6, one ﬁnds a long exact sequence
− → H
k
(X[1])
δ
k
−→ H
k+1
(Y ) − → H
k+1
(Mc(f)) − → .
One can prove that the morphism δ
k
: H
k+1
(X) − → H
k+1
(Y ) is H
k+1
(f) up
to a sign. q.e.d.
Double complexes
Let ( denote as above an abelian category.
Theorem 4.2.10. Let X
•,•
be a double complex such that all rows X
j,•
and
columns X
•,j
are 0 for j < 0 and are exact for j > 0.
Then H
p
(X
0,•
) · H
p
(X
•,0
) · H
p
(tot(X
•,•
)) for all p.
Proof. We shall only describe the ﬁrst isomorphism H
p
(X
0,•
) · H
p
(X
•,0
) in
the case where ( = Mod(A), by the socalled “Weil procedure”.
Let x
p,0
∈ X
p,0
, with d
t
x
p,0
= 0 which represents y ∈ H
p
(X
•,0
). Deﬁne
x
p,1
= d
tt
x
p,0
. Then d
t
x
p,1
= 0, and the ﬁrst column being exact, there exists
x
p−1,1
∈ X
p−1,1
with d
t
x
p−1,1
= x
p,1
. One can iterate this procedure until
getting x
0,p
∈ X
0,p
. Since d
t
d
tt
x
0,p
= 0, and d
t
is injective on X
0,p
for p > 0
by the hypothesis, we get d
tt
x
0,p
= 0. The class of x
0,p
in H
p
(X
0,•
) will be the
image of y by the Weil procedure. Of course, one has to check that this image
does not depend of the various choices we have made, and that it induces an
isomorphism.
4.3. APPLICATION TO KOSZUL COMPLEXES 79
This can be visualized by the diagram:
x
0,p
d
d
0
x
1,p−2
d
x
1,p−1
x
p−1,1
d
x
p,0
d
d
x
p,1
0
q.e.d.
Proposition 4.2.11. Let X
•,•
be a double complex such that all rows X
j,•
and columns X
•,j
are 0 for j < 0. Assume that all rows (resp. all columns)
of X
•,•
are exact. Then the complex tot(X
•,•
) is exact.
The proof is left as an exercise. Note that if there are only two rows let’s
say in degrees −1 and 0, then the result follows from Theorem 4.6.4
4.3 Application to Koszul complexes
Consider a Koszul complex, as in '1.6. Keeping the notations of this section,
set ϕ
t
= ¦ϕ
1
, . . . , ϕ
n−1
¦ and denote by d
t
the diﬀerential in K
•
(M, ϕ
t
). Then
ϕ
n
deﬁnes a morphism
¯ ϕ
n
: K
•
(M, ϕ
t
) − → K
•
(M, ϕ
t
) (4.8)
Proposition 4.3.1. The complex K
•
(M, ϕ)[1] is isomorphic to the mapping
cone of −¯ ϕ
n
.
Proof.
1
Consider the diagram
Mc( ¯ ϕ
n
)
p
d
p
M
λ
p
Mc( ¯ ϕ
n
)
p+1
λ
p+1
K
p+1
(M, ϕ)
d
p+1
K
K
p+2
(M, ϕ)
1
The proof has been skipped in 2005/2006
80 CHAPTER 4. ABELIAN CATEGORIES
given explicitly by:
(M ⊗
p+1
Z
n−1
) ⊕(M ⊗
p
Z
n−1
)0
@
−d
t
0
−ϕ
n
d
t
1
A
id⊕(id ⊗en∧)
M(⊗
p+2
Z
n−1
) ⊕(M ⊗
p+1
Z
n−1
)
id⊕(id ⊗en∧)
M ⊗
p+1
Z
n
−d
M ⊗
p+2
Z
n
Then
d
p
M
(a ⊗e
J
+b ⊗e
K
) = −d
t
(a ⊗e
J
) + (d
t
(b ⊗e
K
) −ϕ
n
(a) ⊗e
J
),
λ
p
(a ⊗e
J
+b ⊗e
K
) = a ⊗e
J
+b ⊗e
n
∧ e
K
.
(i) The vertical arrows are isomorphisms. Indeed, let us treat the ﬁrst one.
It is described by:
¸
J
a
J
⊗e
J
+
¸
K
b
K
⊗e
K
→
¸
J
a
J
⊗e
J
+
¸
K
b
K
⊗e
n
∧ e
K
(4.9)
with [J[ = p + 1 and [K[ = p. Any element of M ⊗
p+1
Z
n
may uniquely
be written as in the right hand side of (4.9).
(ii) The diagram commutes. Indeed,
λ
p+1
◦ d
p
M
(a ⊗e
J
+b ⊗e
K
) = −d
t
(a ⊗e
J
) +e
n
∧ d
t
(b ⊗e
K
) −ϕ
n
(a) ⊗e
n
∧ e
J
= −d
t
(a ⊗e
J
) −d
t
(b ⊗e
n
∧ e
K
) −ϕ
n
(a) ⊗e
n
∧ e
J
,
d
p+1
K
◦ λ
p
(a ⊗e
J
+b ⊗e
K
) = −d(a ⊗e
J
+b ⊗e
n
∧ e
K
)
= −d
t
(a ⊗e
J
) −ϕ
n
(a) ⊗e
n
∧ e
J
−d
t
(b ⊗e
n
∧ e
K
).
q.e.d.
Proposition 4.3.2. There exists a long exact sequence
− → H
j
(K
•
(M, ϕ
t
))
ϕn
−→ H
j
(K
•
(M, ϕ
t
)) − → H
j+1
(K
•
(M, ϕ)) − → (4.10)
Proof. Apply Proposition 4.3.1 and Corollary 4.2.9. q.e.d.
We can now give a proof to Theorem 1.6.2. Assume for example that
(ϕ
1
, . . . , ϕ
n
) is a regular sequence, and let us argue by induction on n. The
cohomology of K
•
(M, ϕ
t
) is thus concentrated in degree n−1 and is isomor
phic to M/(ϕ
1
(M) + + ϕ
n−1
(M)). By the hypothesis, ϕ
n
is injective on
this group, and Theorem 1.6.2 follows.
4.4. INJECTIVE OBJECTS 81
4.4 Injective objects
Deﬁnition 4.4.1. (i) An object I of ( is injective if Hom
(
(
•
, I) is an
exact functor.
(ii) One says that ( has enough injectives if for any X ∈ ( there exists a
monomorphism XI with I injective.
(iii) An object P is projective in ( iﬀ it is injective in (
op
, i.e. if the functor
Hom
(
(P,
•
) is exact.
(iv) One says that ( has enough projectives if for any X ∈ ( there exists
an epimorphism PX with P projective.
Example 4.4.2. Let A be a ring. An Amodule M is called injective (resp.
projective) if it is so in the category Mod(A). If M is free then it is projective.
More generally, if there exists an Amodule N such that M⊕N is free then M
is projective (see Exercise 1.2). One immediately deduces that the category
Mod(A) has enough projectives. One can prove that Mod(A) has enough
injectives (see Exercise 1.5).
If k is a ﬁeld, then any object of Mod(k) is both injective and projective.
Proposition 4.4.3. The object I ∈ ( is injective if and only if, for any
X, Y ∈ ( and any diagram in which the row is exact:
0
X
f
k
Y
h
.·
I
the dotted arrow may be completed, making the solid diagram commutative.
The proof is similar to that of Proposition 1.3.8.
Lemma 4.4.4. Let 0 − → X
t
f
− → X
g
− → X
tt
− → 0 be an exact sequence in (, and
assume that X
t
is injective. Then the sequence splits.
Proof. Applying the preceding result with k = id
X
, we ﬁnd h: X − → X
t
such
that k ◦ f = id
X
. Then apply Proposition 4.1.7. q.e.d.
It follows that if F : ( − → (
t
is an additive functor of abelian categories, and
the hypotheses of the lemma are satisﬁed, then the sequence 0 − → F(X
t
) − →
F(X) − → F(X
tt
) − → 0 splits and in particular is exact.
82 CHAPTER 4. ABELIAN CATEGORIES
Lemma 4.4.5. Let X
t
, X
tt
belong to (. Then X
t
⊕ X
tt
is injective if and
only if X
t
and X
tt
are injective.
Proof. It is enough to remark that for two additive functors of abelian cat
egories F and G, X → F(X) ⊕ G(X) is exact if and only if F and G are
exact. q.e.d.
Applying Lemmas 4.4.4 and 4.4.5, we get:
Proposition 4.4.6. Let 0 − → X
t
− → X − → X
tt
− → 0 be an exact sequence in (
and assume X
t
and X are injective. Then X
tt
is injective.
4.5 Resolutions
In this section, ( denotes an abelian category and 1
(
its full additive subcat
egory consisting of injective objects. We shall asume
the abelian category ( admits enough injectives. (4.11)
Deﬁnition 4.5.1. Let . be a full additive subcategory of (. We say that
. is cogenerating if for all X in (, there exist Y ∈ . and a monomorphism
XY .
Note that the category of injective objects is cogenerating iﬀ ( has enough
injectives.
Notations 4.5.2. Consider an exact sequence in (, 0 − → X − → J
0
− → − →
J
n
− → and denote by J
•
the complex 0 − → J
0
− → − → J
n
− → . We
shall say for short that 0 − → X − → J
•
is a resolution of X. If the J
k
’s belong
to ., we shall say that this is a .resolution of X. When . denotes the
category of injective objects one says this is an injective resolution.
Proposition 4.5.3. Assume . is cogenerating. Then for any X ∈ (, there
exists a .resolution of X.
Proof. We proceed by induction. Assume to have constructed:
0 − → X − → J
0
− → − → J
n
For n = 0 this is the hypothesis. Set B
n
= Coker(J
n−1
− → J
n
) (with J
−1
=
X). Then J
n−1
− → J
n
− → B
n
− → 0 is exact. Embed B
n
in an object of .:
0 − → B
n
− → J
n+1
. Then J
n−1
− → J
n
− → J
n+1
is exact, and the induction
proceeds. q.e.d.
4.5. RESOLUTIONS 83
Proposition 4.5.4. (i) Let f
•
: X
•
− → I
•
be a morphism in C
+
((). As
sume I
•
belongs to (
+
(1
(
) and X
•
is exact. Then f
•
is homotopic to
0.
(ii) Let I
•
∈ C
+
(1
(
) and assume I
•
is exact. Then I
•
is homotopic to 0.
Proof. (i) Consider the diagram:
X
k−2
X
k−1
f
k−1
s
k−1
.
X
k
s
k
.
f
k
X
k+1
s
k+1
.
I
k−2
I
k−1
I
k
I
k+1
We shall construct by induction morphisms s
k
satisfying:
f
k
= s
k+1
◦ d
k
X
+d
k−1
I
◦ s
k
.
For j << 0, s
j
= 0. Assume we have constructed the s
j
for j ≤ k. Deﬁne
g
k
= f
k
−d
k−1
I
◦ s
k
. One has
g
k
◦ d
k−1
X
= f
k
◦ d
k−1
X
−d
k−1
I
◦ s
k
◦ d
k−1
X
= f
k
◦ d
k−1
X
−d
k−1
I
◦ f
k−1
+d
k−1
I
◦ d
k−2
I
◦ s
k−1
= 0.
Hence, g
k
factorizes through X
k
/ Imd
k−1
X
. Since the complex X
•
is exact,
the sequence 0 − → X
k
/ Imd
k−1
X
− → X
k+1
is exact. Consider
0
X
k
/ Imd
k−1
X
g
k
X
k+1
s
k+1
.
I
k
The dotted arrow may be completed by Proposition 4.4.3.
(ii) Apply the result of (i) with X
•
= I
•
and f = id
X
. q.e.d.
Proposition 4.5.5. (i) Let f : X − → Y be a morphism in (, let 0 − → X − →
X
•
be a resolution of X and let 0 − → Y − → J
•
be a complex with the
J
k
’s injective. Then there exists a morphism f
•
: X
•
− → J
•
making the
diagram below commutative:
0
X
f
X
•
f
•
0
Y
J
•
(4.12)
84 CHAPTER 4. ABELIAN CATEGORIES
(ii) The morphism f
•
in C(() constructed in (i) is unique up to homotopy.
Proof. (i) Let us denote by d
X
(resp. d
Y
) the diﬀerential of the complex X
•
(resp. J
•
), by d
−1
X
(resp. d
−1
Y
) the morphism X − → X
0
(resp. Y − → J
0
) and
set f
−1
= f.
We shall construct the f
n
’s by induction. Morphism f
0
is obtained by
Proposition 4.4.3. Assume we have constructed f
0
, . . . , f
n
. Let g
n
= d
n
Y
◦
f
n
:X
n
− → J
n+1
. The morphism g
n
factorizes through h
n
: X
n
/ Imd
n−1
X
− →
J
n+1
. Since X
•
is exact, the sequence 0 − → X
n
/ Imd
n−1
X
− → X
n+1
is exact.
Since J
n+1
is injective, h
n
extends as f
n+1
: X
n+1
− → J
n+1
.
(ii) We may assume f = 0 and we have to prove that in this case f
•
is
homotopic to zero. Since the sequence 0 − → X − → X
•
is exact, this follows
from Proposition 4.5.4 (i), replacing the exact sequence 0 − → Y − → J
•
by the
complex 0 − → 0 − → J
•
. q.e.d.
4.6 Derived functors
In this section, ( and (
t
will denote abelian categories and F : ( − → (
t
a left
exact functor. We shall make the hypothesis
the category ( admits enough injectives. (4.13)
Recall that 1
(
denotes the additive category of injective objects in (.
Lemma 4.6.1. Assuming (4.13), there exists a functor λ: ( − → K(1
(
) and
for for each X ∈ (, a qis X − → λ(X), functorially in X ∈ (.
Proof. (i) Let X ∈ ( and let I
•
X
∈ C
+
(1
(
) be an injective resolution of
X. The image of I
•
X
in K
+
(() is unique up to unique isomorphism, by
Proposition 4.5.5.
Indeed, consider two injective resolutions I
•
X
and J
•
X
of X. By Propo
sition 4.5.5 applied to id
X
, there exists a morphism f
•
: I
•
X
− → J
•
X
making
the diagram 4.12 commutative and this morphism is unique up to homo
topy, hence is unique in K
+
((). Similarly, there exists a unique morphism
g
•
: J
•
X
− → I
•
X
in K
+
((). Hence, f
•
and g
•
are isomorphisms inverse one to
each other.
(ii) Let f : X − → Y be a morphism in (, let I
•
X
and I
•
Y
be injective resolutions
of X and Y respectively, and let f
•
: I
•
X
− → I
•
Y
be a morphism of complexes
such as in Proposition 4.5.5. Then the image of f
•
in Hom
K
+
(7
C
)
(I
•
X
, I
•
Y
)
does not depend on the choice of f
•
by Proposition 4.5.5.
In particular, we get that if g : Y − → Z is another morphism in ( and I
•
Z
is
an injective resolutions of Z, then g
•
◦f
•
= (g ◦f)
•
as morphisms in K
+
(1
(
).
q.e.d.
4.6. DERIVED FUNCTORS 85
Let F : ( − → (
t
be a left exact functor of abelian categories and assume
that ( has enough injectives. Consider the functors
(
λ
− → K
+
(1
(
)
F
−→ K
+
((
t
)
H
n
−−→ (
t
.
Deﬁnition 4.6.2. One sets
R
n
F = H
n
◦ F ◦ λ (4.14)
and calls R
n
F the nth right derived functor of F.
By its deﬁnition, the receipt to construct R
n
F(X) is as follows:
• choose an injective resolution I
•
X
of X, that is, construct an exact
sequence 0 − → X − → I
•
X
with I
•
X
∈ C
+
(1
(
),
• apply F to this resolution,
• take the nth cohomology.
In other words, R
n
F(X) · H
n
(F(I
•
X
)).
Note that R
n
F is an additive functor from ( to (
t
and
R
n
F(X) · 0 for n < 0,
R
0
F(X) · F(X),
R
n
F(X) · 0 for n = 0 if F is exact,
R
n
F(X) · 0 for n = 0 if X is injective.
The ﬁrst assertion is obvious since I
k
X
= 0 for k < 0, and the second one
follows from the fact that F being left exact, then Ker(F(I
0
X
) − → F(I
1
X
)) ·
F(Ker(I
0
X
− → I
1
X
))· F(X). The third assertion is clear since F being exact,
it commutes with H
j
(
•
). The last assertion is obvious by the construction
of R
j
F(X).
Deﬁnition 4.6.3. An object X of ( such that R
k
F(X) · 0 for all k > 0 is
called Facyclic.
Hence, injective objects are Facyclic for all left exact functors F.
Theorem 4.6.4. Let 0 − → X
t
f
− → X
g
− → X
tt
− → 0 be an exact sequence in (.
Then there exists a long exact sequence:
0 − → F(X
t
) − → F(X) − → − → R
k
F(X
t
) − → R
k
F(X) − → R
k
F(X
tt
) − →
86 CHAPTER 4. ABELIAN CATEGORIES
Sketch of the proof. One constructs an exact sequence of complexes 0 − →
X
t
•
− → X
•
− → X
tt
•
− → 0 whose objects are injective and this sequence is
quasiisomorphic to the sequence 0 − → X
t
f
− → X
g
− → X
tt
− → 0 in C((). Since
the objects X
t
j
are injectice, we get a short exact sequence in C((
t
):
0 − → F(X
t
•
) − → F(X
•
) − → F(X
tt
•
) − → 0
Then one applies Theorem 4.2.6. q.e.d.
Deﬁnition 4.6.5. Let . be a full additive subcategory of (. One says that
. is Finjective if:
(i) . is cogenerating,
(ii) for any exact sequence 0 − → X
t
− → X − → X
tt
− → 0 in ( with X
t
∈ ., X ∈
., then X
tt
∈ .,
(iii) for any exact sequence 0 − → X
t
− → X − → X
tt
− → 0 in ( with X
t
∈ ., the
sequence 0 − → F(X
t
) − → F(X) − → F(X
tt
) − → 0 is exact.
By considering (
op
, one obtains the notion of an Fprojective subcategory,
F being right exact.
Proposition 4.6.6. Let F : ( − → (
t
be a left exact functor and denote by
1
F
the full subcategory of ( consisting of Facyclic objects. Then 1
F
is F
injective.
Proof. Since injective objects are Facyclic, hypothesis (4.13) implies that 1
F
is cogenerating. The conditions (ii) and (iii) in Deﬁnition 4.6.5 are satisﬁed
by Theorem 4.6.4. q.e.d.
Examples 4.6.7. (i) If ( has enough injectives, the category 1 of injective
objects is Facyclic for all left exact functors F.
(ii) Let A be a ring and let N be a right Amodule. The full additive
subcategory of Mod(A) consisting of ﬂat Amodules is projective with respect
to the functor N ⊗
A
•
.
Lemma 4.6.8. Assume . is Finjective and let X
•
∈ C
+
(.) be a complex
qis to zero (i.e. X
•
is exact). Then F(X
•
) is qis to zero.
Proof. We decompose X
•
into short exact sequences (assuming that this
complex starts at step 0 for convenience):
0 − → X
0
− → X
1
− → Z
1
− → 0
0 − → Z
1
− → X
2
− → Z
2
− → 0
0 − → Z
n−1
− → X
n
− → Z
n
− → 0
4.6. DERIVED FUNCTORS 87
By induction we ﬁnd that all the Z
j
’s belong to ., hence all the sequences:
0 − → F(Z
n−1
) − → F(X
n
) − → F(Z
n
) − → 0
are exact. Hence the sequence
0 − → F(X
0
) − → F(X
1
) − →
is exact. q.e.d.
Theorem 4.6.9. Assume . is Finjective and contains the category 1
(
of
injective objects. Let X ∈ ( and let 0 − → X − → J
•
be a resolution of X with
J
k
∈ .. Then for each k, there is an isomorphism R
k
F(X) · H
k
(F(J
•
)).
Proof. Let 0 − → X − → J
•
be a .resolution of X and let 0 − → X − → I
•
be an injective resolution of X. Applying Proposition 4.5.5, there exists
f : J
•
− → I
•
making the diagram below commutative
0
X
id
J
0
f
0
d
0
J
J
1
f
1
d
1
J
0
X
I
0
d
0
I
I
1
d
1
I
Deﬁne the complex K
•
= Mc(f), the mapping cone of f. By the hypothesis,
K
•
belongs to C
+
(.) and this complex is qis to zero by Corollary 4.2.8. By
Lemma 4.6.8, F(K
•
) is qis to zero.
On the otherhand, F(Mc(f)) is isomorphic to Mc(F(f)), the mapping
cone of F(f) : F(J
•
) − → F(I
•
). Applying Theorem 4.2.6 to this sequence, we
ﬁnd a long exact sequence
− → H
n
(F(J
•
)) − → H
n
(F(I
•
)) − → H
n
(F(K
•
)) − →
Since F(K
•
) is qis to zero, the result follows. q.e.d.
By this result, one sees that in order to calculate the kth derived functor
of F at X, the recipe is as follows. Consider a resolution 0 − → X − → J
•
of X by objects of ., then apply F to the complex J
•
, and take the kth
cohomology object.
Proposition 4.6.10. Let F : ( − → (
t
and G : (
t
− → (
tt
be left exact functors
of abelian categories. We assume that ( and (
t
have enough injectives.
(i) If G is exact, then R
j
(G◦ F) · G◦ R
j
F.
88 CHAPTER 4. ABELIAN CATEGORIES
(ii) Assume that F is exact. There is a natural morphism R
j
(G ◦ F) − →
(R
j
G) ◦ F.
(iii) Let .
t
be a Ginjective subcategory of (
t
and assume that F sends the
injective objects of ( in .
t
. If X ∈ ( satisﬁes R
k
F(X) = 0 for k = 0,
then R
j
(G◦ F)(X) · R
j
G(F(X)).
(iv) In particular, let .
t
be a Ginjective subcategory of (
t
and assume that
F is exact and sends the injective objects of ( in .
t
. Then R
j
(G◦F) ·
R
j
G◦ F.
Proof. Let X ∈ ( and let 0 − → X − → I
•
X
be an injective resolution of X. Then
R
j
(G◦ F)(X) · H
j
(G◦ F(I
•
X
)).
(i) If G is exact, the righthand side is isomorphic to G(H
j
(F(I
•
X
)).
(ii) Consider an injective resolution 0 − → F(X) − → J
•
F(X)
of F(X). By Propo
sition 4.5.5, there exists a morphism F(I
•
X
) − → J
•
F(X)
. Applying G we get a
morphism of complexes: (G◦ F)(I
•
X
) − → G(J
•
F(X)
). Since H
j
((G◦ F)(I
•
X
)) ·
R
j
(G◦ F)(X) and H
j
(G(J
•
F(X)
)) · R
j
G(F(X)), we get the result.
(iii) By the hypothesis, F(I
•
X
) is qis to F(X) and belongs to C
+
(.
t
). Hence
R
j
G(F(X)) · H
j
(G(F(I
•
X
))).
(iv) is a particular case of (iii). q.e.d.
4.7 Bifunctors
Now consider an additive bifunctor F : ((
t
− → (
tt
of abelian categories, and
assume: F is left exact with respect of each of its arguments (i.e., F(X,
•
)
and F(
•
, Y ) are left exact).
Let 1
(
(resp. 1
(
) denote the full additive subcategory of ( (resp. (
t
)
consisting of injective objects.
Deﬁnition 4.7.1. (a) The pair (1
(
, (
t
) is Finjective if ( admits enough
injective and for all I ∈ 1
(
, F(I,
•
) is exact.
(b) If (1
(
, (
t
) is Finjective, we denote by R
k
F(X, Y ) the kth derived functor
of F(
•
, Y ) at X, i.e., R
k
F(X, Y ) = R
k
F(
•
, Y )(X).
Proposition 4.7.2. Assume that (1
(
, (
t
) is Finjective.
(i) Let 0 − → X
t
− → X − → X
tt
− → 0 be an exact sequence in ( and let Y ∈ (
t
.
Then there is a long exact sequence in (
tt
:
− → R
k−1
F(X
tt
, Y ) − → R
k
F(X
t
, Y ) − → R
k
F(X, Y ) − → R
k
F(X
tt
, Y ) − →
4.7. BIFUNCTORS 89
(ii) Let 0 − → Y
t
− → Y − → Y
tt
− → 0 be an exact sequence in (
t
and let X ∈ (.
Then there is a long exact sequence in (
tt
:
− → R
k−1
F(X, Y
tt
) − → R
k
F(X, Y
t
) − → R
k
F(X, Y ) − → R
k
F(X, Y
tt
) − →
Proof. (i) is a particular case of Theorem 4.6.4.
(ii) Let 0 − → X − → I
•
be an injective resolution of X. By the hypothesis, the
sequence in C((
tt
):
0 − → F(I
•
, Y
t
) − → F(I
•
, Y ) − → F(I
•
, Y
tt
) − → 0
is exact. By Theorem 4.2.6, it gives rise to the desired long exact sequence.
q.e.d.
Proposition 4.7.3. Assume that both (1
(
, (
t
) and ((, 1
(
) are Finjective.
Then for X ∈ ( and Y ∈ (
t
, we have the isomorphism: R
k
F(X, Y ) :=
R
k
F(
•
, Y )(X) · R
k
F(X,
•
)(Y ).
Moreover if I
•
X
is an injective resolution of X and I
•
Y
an injective resolu
tion of Y , then R
k
F(X, Y ) · totH
k
(F(I
•
X
, I
•
Y
).
Proof. Let 0 − → X − → I
•
X
and 0 − → Y − → I
•
Y
be injective resolutions of X and
Y , respectively. Consider the double complex:
0
0
0
0
0
F(I
0
X
, Y )
F(I
1
X
, Y )
0
F(X, I
0
Y
)
F(I
0
X
, I
0
Y
)
F(I
1
X
, I
0
Y
)
0
F(X, I
1
Y
)
F(I
0
X
, I
1
Y
)
F(I
1
X
, I
1
Y
)
The cohomology of the ﬁrst row (resp. column) calculates R
k
F(
•
, Y )(X)
(resp. R
k
F(X,
•
)(Y )). Since the other rows and columns are exact by the
hypotheses, the result follows from Theorem 4.2.10. q.e.d.
Example 4.7.4. Assume ( has enough injectives. Then
R
k
Hom
(
: (
op
( − → Ab
exists and is calculated as follows. Let X ∈ (, Y ∈ (. There exists a qis in
C
+
((), Y − → I
•
, the I
j
’s being injective. Then:
R
k
Hom
(
(X, Y ) · H
k
(Hom
(
(X, I
•
)).
90 CHAPTER 4. ABELIAN CATEGORIES
If ( has enough projectives, and P
•
− → X is a qis in C
−
((), the P
j
’s being
projective, one also has:
R
k
Hom
(
(X, Y ) · H
k
Hom
(
(P
•
, Y )
· H
k
tot(Hom
(
(P
•
, I
•
)).
If ( has enough injectives or enough projectives, one sets:
Ext
k
(
(
•
,
•
) = R
k
Hom
(
(
•
,
•
). (4.15)
For example, let A = k[x, y], M = k · A/xA+yA and let us calculate the
groups Ext
j
A
(M, A). Since injective resolutions are not easy to calculate, it
is much simpler to calculate a free (hence, projective) resolution of M. Since
(x, y) is a regular sequence of endomorphisms of A (viewed as an Amodule),
M is quasiisomorphic to the complex:
M
•
: 0 − → A
u
− → A
2
v
− → A − → 0,
where u(a) = (ya, −xa), v(b, c) = xb + yc and the module A on the right
stands in degree 0. Therefore, Ext
j
A
(M, N) is the jth cohomology object of
the complex Hom
A
(M
•
, N), that is:
0 − → N
v
−→ N
2
u
−→ N − → 0,
where v
t
= Hom(v, N), u
t
= Hom(u, N) and the module N on the left stands
in degree 0. Since v
t
(n) = (xn, yn) and u
t
(m, l) = ym− xl, we ﬁnd again a
Koszul complex. Choosing N = A, its cohomology is concentrated in degree
2. Hence, Ext
j
A
(M, A) · 0 for j = 2 and · k for j = 2.
Example 4.7.5. Let A be a kalgebra. Since the category Mod(A) admits
enough projective objects, the bifunctor
•
⊗
•
: Mod(A
op
) Mod(A) − → Mod(k)
admits derived functors, denoted Tor
A
−k
(
•
,
•
) or else, Tor
k
A
(
•
,
•
).
If Q
•
− → N − → 0 is a projective resolution of the A
op
module N, or
P
•
− → M − → 0 is a projective resolution of the Amodule M, then :
Tor
A
k
(N, M) · H
−k
(Q
•
⊗
A
M)
· H
−k
(N ⊗
A
P
•
)
· H
−k
(tot(Q
•
⊗
A
P
•
)).
Exercises to Chapter 4 91
Exercises to Chapter 4
Exercise 4.1. Let ( be an abelian category which admits inductive limits
and such that ﬁltrant inductive limits are exact. Let ¦X
i
¦
i∈I
be a family
of objects of ( indexed by a set I and let i
0
∈ I. Prove that the natural
morphism X
i
0
− →
¸
i∈I
X
i
is a mornomorphism.
Exercise 4.2. Let ( be an abelian category.
(i) Prove that a complex 0 − → X − → Y − → Z is exact iﬀ and only if for
any object W ∈ ( the complex of abelian groups 0 − → Hom
(
(W, X) − →
Hom
(
(W, Y ) − → Hom
(
(W, Z) is exact.
(ii) By reversing the arrows, state and prove a similar statement for a complex
X − → Y − → Z − → 0.
Exercise 4.3. Let ( be an abelian category. A square is a commutative
diagram:
V
f
g
Y
g
X
f
Z.
A square is Cartesian if moreover the sequence 0 − → V − → X Y − → Z is
exact, that is, if V · X
Z
Y (recall that X
Z
Y = Ker(f −g), where f −g :
X⊕Y − → Z). A square is coCartesian if the sequence V − → X⊕Y − → Z − → 0
is exact, that is, if Z · X⊕
V
Y (recall that X⊕
Z
Y = Coker(f
t
−g
t
), where
f
t
−g
t
: V − → X Y ).
(i) Assume the square is Cartesian and f is an epimorphism. Prove that f
t
is an epimorphism.
(ii) Assume the square is coCartesian and f
t
is a monomorphism. Prove
that f is a monomorphism.
Exercise 4.4. Let ( be an abelian category and consider two sequences of
morphisms X
t
i
f
i
−→ X
i
g
i
−→ X
tt
i
, i = 1, 2 with g
i
◦ f
i
= 0. Set X
t
= X
t
1
⊕ X
t
2
,
and deﬁne similarly X, X
tt
and f, g. Prove that the two sequences above are
exact if and only if the sequence X
t
f
− → X
g
− → X
tt
is exact.
Exercise 4.5. Let ( be an abelian category and consider a commutative
92 CHAPTER 4. ABELIAN CATEGORIES
diagram of complexes
0
0
0
0
X
t
0
X
0
X
tt
0
0
X
t
1
X
1
X
tt
1
0
X
t
2
X
2
X
tt
2
Assume that all rows are exact as well as the second and third column. Prove
that all columns are exact.
Exercise 4.6. Let ( be an abelian category. To X ∈ C
b
((), one associates
the new complex H
•
(X) =
¸
H
j
(X)[−j] with 0diﬀerential. In other words
H
•
(X) := − → H
i
(X)
0
− → H
i+1
(X)
0
− →
(i) Prove that H
•
: C
b
(() − → C
b
(() is a welldeﬁned additive functor.
(ii) Give examples which show that in general, H
•
is neither right nor left
exact.
Exercise 4.7. Let ϕ = (ϕ
1
, . . . , ϕ
n
) be n commuting endomorphisms of an
Amodule M. Let ϕ
t
= (ϕ
1
, . . . , ϕ
n−p
) and ϕ
tt
= (ϕ
n−p+1
, . . . , ϕ
n
).
Calculate the cohomology of K
•
(M, ϕ) assuming that ϕ
t
is a regular sequence
and ϕ
tt
is a coregular sequence.
Exercise 4.8. Let A = k[x
1
, x
2
]. Consider the Amodules: M
t
= A/(Ax
1
+
Ax
2
), M = A/(Ax
2
1
+Ax
1
x
2
), M
tt
= A/(Ax
1
).
(i) Show that the monomorphism Ax
1
→ A induces a monomorphism M
t
→
Mand deduce an exact sequence of Amodules 0 − → M
t
− → M − → M
tt
− → 0.
(ii) By considering the action of x
1
on these three modules, show that the
sequence above does not split.
(iii) Construct free resolutions of M
t
and M
tt
.
(iv) Calculate Ext
j
A
(M, A) for all j .
Exercise 4.9. Let ( and (
t
be two abelian categories. We assume that
(
t
admits inductive limits and ﬁltrant inductive limits are exact in (
t
. Let
¦F
i
¦
i∈I
be an inductive system of left exact functors from ( to (
t
, indexed
by a ﬁltrant categoryI.
(i) Prove that lim
−→
i
F
i
is a left exact functor.
(ii) Prove that for each k ∈ Z, ¦R
k
F
i
¦
i∈I
is an inductive system of functors
and R
k
(lim
−→
i
F
i
) · lim
−→
i
R
k
F
i
.
Exercises to Chapter 4 93
Exercise 4.10. Let F : ( − → (
t
be a left exact functor of abelian categories.
Let . be an Finjective subcategory of (, and let Y
•
be an object of C
+
(.).
Assume that H
k
(Y
•
) = 0 for all k = p for some p ∈ Z, and let X = H
p
(Y
•
).
Prove that R
k
F(X) · H
k+p
(F(Y
•
)).
Exercise 4.11. We consider the following situation: F : ( − → (
t
and G :
(
t
− → (
tt
are left exact functors of abelian categories having enough injectives,
.
t
is an Ginjective subcategory of (
t
and F sends injective objects of ( in
.
t
.
(i) Let X ∈ ( and assume that there is q ∈ N with R
k
F(X) = 0 for k = q.
Prove that R
j
(G◦ F)(X) · R
j−q
G(R
q
F(X)). (Hint: use Exercise 4.10.)
(ii) Assume now that R
j
F(X) = 0 for j = 0, 1. Prove that there is a long
exact sequence:
− → R
k−1
G(R
1
F(X)) − → R
k
(G◦ F)(X) − → R
k
G(F(X)) − →
(Hint: construct an exact sequence 0 − → X − → X
0
− → X
1
− → 0 with X
0
injective and X
1
Facyclic.)
Exercise 4.12. In the situation of Proposition 4.6.10, let X ∈ ( and assume
that R
j
F(X) · 0 for j < n. Prove that R
n
(F
t
◦ F)(X) · F
t
(R
n
F(X)).
Exercise 4.13. Here, we shall use the notation H
•
introduced in Exercise
4.6. Assume that k is a ﬁeld and consider the complexes in Mod(k):
X
•
:= X
0
f
− → X
1
,
Y
•
:= Y
0
g
− → Y
1
and the double complex
X
•
⊗Y
•
:= X
0
⊗Y
0
f⊗id
id ⊗g
X
1
⊗Y
0
id ⊗g
X
0
⊗Y
1
f⊗id
X
1
⊗Y
1
.
(i) Prove that tot(X
•
⊗Y
•
) and tot(H
•
(X
•
)⊗Y
•
) have the same cohomology
objects.
(ii) Deduce that tot(X
•
⊗ Y
•
) and tot(H
•
(X
•
) ⊗ H
•
(Y
•
)) have the same
cohomology objects.
94 CHAPTER 4. ABELIAN CATEGORIES
Exercise 4.14. Assume that k is a ﬁeld. Let X
•
and Y
•
be two objects of
C
b
(Mod(k)). Prove the isomorphism
H
p
(tot(X
•
⊗Y
•
)) ·
i+j=p
H
i
(X
•
) ⊗H
j
(Y
•
)
· H
p
(
i
H
i
(X
•
)[−i] ⊗
j
H
j
(Y
•
)[−j]).
Here, we use the convention that:
(A ⊕B) ⊗(C ⊕D) · (A⊗C) ⊕(A⊗D) ⊕(B ⊗C) ⊕(B ⊗D)
A[i] ⊗B[j] ∼ A ⊗B[i +j].
(Hint: use the result of Exercise 4.13.)
Chapter 5
Abelian sheaves
In this chapter we expose basic sheaf theory in the framework of topological
spaces. Although we restrict our study to topological spaces, it will be con
venient to consider morphisms of sites f : X − → Y which are not continuous
maps from X to Y .
Recall that all along these Notes, k denotes a commutative unitary ring.
Some references: [12], [3], [10], [17], [18].
5.1 Presheaves
Let X be a topological space. The family of open subsets of X is ordered by
inclusion and we denote by Op
X
the associated category. Hence:
Hom
Op
X
(U, V ) =
¦pt¦ if U ⊂ V,
∅ otherwise.
Note that the category Op
X
admits a terminal object, namely X, and ﬁnite
products, namely U V = U ∩ V . Indeed,
Hom
Op
X
(W, U) Hom
Op
X
(W, V ) · Hom
Op
X
(W, U ∩ V ).
Deﬁnition 5.1.1. (i) One sets PSh(X) := Fct((Op
X
)
op
, Set) calls an ob
ject of this category a presheaf of sets. In other words, a presheaf of
sets in a functor from (Op
X
)
op
to Set.
(ii) One denotes by PSh(k
X
) the subcategory of PSh(X) consisting of func
tors with values in Mod(k) and calls an object of this category a presheaf
of kmodules.
95
96 CHAPTER 5. ABELIAN SHEAVES
Hence, a presheaf F on X associates to each open subset U ⊂ X a set
F(U), and to an open inclusion V ⊂ U, a map ρ
V U
: F(U) − → F(V ), such
that for each open inclusions W ⊂ V ⊂ U, one has:
ρ
UU
= id
U
, ρ
WU
= ρ
WV
◦ ρ
V U
.
A morphism of presheaves ϕ : F − → G is thus the data for any open set U of
a map ϕ(U) : F(U) − → G(U) such that for any open inclusion V ⊂ U, the
diagram below commutes:
F(U)
ϕ(U)
G(U)
F(V )
ϕ(V )
G(V )
If F is a presheaf of kmodules, then the F(U)’s are kmodules and the maps
ρ
V U
are klinear.
The category PSh(k
X
) inherits of most of the properties of the category
Mod(k). In particular it is abelian and admits inductive and projective
limits. For example, one checks easily that if F and G are two presheaves,
the presheaf U → F(U) ⊕G(U) is the coproduct of F and G in PSh(k
X
). If
ϕ : F − → G is a morphism of presheaves, then (Ker ϕ)(U) · Ker ϕ(U) and
(Coker ϕ)(U) · Coker ϕ(U) where ϕ(U) : F(U) − → G(U).
More generally, if i → F
i
, i ∈ I is an inductive system of presheaves, one
checks that the presheaf U → lim
−→
i
F
i
(U) is the inductive limit of this system
in the category PSh(k
X
) and similarly with projective limits.
Note that for U ∈ Op
X
, the functor PSh(k
X
) − → Mod(k), F → F(U) is
exact.
Notation 5.1.2. (i) One calls the morphisms ρ
V U
, the restriction mor
phisms. If s ∈ F(U), one better writes s[
V
instead of ρ
V U
s and calls s[
V
the restriction of s to V .
(ii) One denotes by F[
U
the presheaf on U deﬁned by V → F(V ), V open in
U and calls F[
U
the restriction of F to U.
(iii) If s ∈ F(U), one says that s is a section of F on U, and if V is an open
subset of U, one writes s[
V
instead of ρ
V U
(s).
Examples 5.1.3. (i) Let M ∈ Set. The correspondence U → M is a
presheaf, called the constant presheaf on X with ﬁber M. For example, if
M = C, one gets the presheaf of Cvalued constant functions on X.
(ii) Let (
0
(U) denote the Cvector space of Cvalued continuous functions on
U. Then U → (
0
(U) (with the usual restriction morphisms) is a presheaf of
Cvector spaces, denoted (
0
X
.
5.2. SHEAVES 97
Deﬁnition 5.1.4. Let x ∈ X, and let I
x
denote the full subcategory of Op
X
consisting of open neighborhoods of x. For a presheaf F on X, one sets:
F
x
= lim
−→
U∈I
op
x
F(U). (5.1)
One calls F
x
the stalk of F at x.
Let x ∈ U and let s ∈ F(U). The image s
x
∈ F
x
of s is called the germ
of s at x. Note that any s
x
∈ F
x
is represented by a section s ∈ F(U) for
some open neighborhood U of x, and for s ∈ F(U), t ∈ F(V ), s
x
= t
x
means
that there exists an open neighborhood W of x with W ⊂ U ∩ V such that
ρ
WU
(s) = ρ
WV
(t). (See Example 1.5.10.)
Proposition 5.1.5. The functor F → F
x
from PSh(k
X
) to Mod(k) is exact.
Proof. The functor F → F
x
is the composition
PSh(k
X
) = Fct(Op
op
X
, Mod(k)) − → Fct(I
op
x
, Mod(k))
lim
− → −−→ Mod(k).
The ﬁrst functor associates to a presheaf F its restriction to the category
I
op
x
. It is clearly exact. Since U, V ∈ I
x
implies U ∩V ∈ I
x
, the category I
op
x
is ﬁltrant and it follows that the functor lim
−→
is exact (see Example 4.1.10
(iii)). q.e.d.
5.2 Sheaves
Let X be a topological space and let Op
X
denote the category of its open
subsets. Recall that if U, V ∈ Op
X
, then U ∩ V is the product of U and V
in Op
X
.
We shall have to consider families  := ¦U
i
¦
i∈I
of open subsets of U
indexed by a set I. One says that  is an open covering of U if
¸
i
U
i
= U.
Note that the empty family ¦U
i
; i ∈ I¦ with I = ∅ is an open covering of
∅ ∈ Op
X
.
Let F be a presheaf on X and consider the two conditions below.
S1 For any open subset U ⊂ X, any open covering U =
¸
i
U
i
, any s, t ∈
F(U) satisfying s
i
[
U
i
= t
i
[
U
i
for all i, one has s = t.
S2 For any open subset U ⊂ X, any open covering U =
¸
i
U
i
, any family
¦s
i
∈ F(U
i
), i ∈ I¦ satisfying s
i
[
U
ij
= s
j
[
U
ij
for all i, j, there exists
s ∈ F(U) with s[
U
i
= s
i
for all i.
98 CHAPTER 5. ABELIAN SHEAVES
Deﬁnition 5.2.1. (i) One says that F is separated if it satisﬁes S1. One
says that F is a sheaf if it satisﬁes S1 and S2.
(ii) One denotes by Sh(X) the full subcategory of PSh(X) whose objects
are sheaves.
(iii) One denotes by Mod(k
X
) the full kadditive subcategory of PSh(k
X
)
whose objects are sheaves and by ι
X
: Mod(k
X
) − → PSh(k
X
) the for
getful functor. If there is no risk of confusion, one writes ι instead of
ι
X
.
(iv) One writes Hom
k
X
(
•
,
•
) instead of Hom
Mod(k
X
)
(
•
,
•
).
Note that if F is a sheaf on X and U is open in X, then F[
U
is a sheaf on U.
Let  be an open covering of U ∈ Op
X
and let F ∈ PSh(k
X
). One sets
F() = Ker(
¸
V ∈
F(V ) ⇒
¸
V
,V
∈
F(V
t
∩ V
tt
)). (5.2)
Here the two arrows are associated with
¸
V ∈
F(V ) − → F(V
t
) − → F(V
t
∩V
tt
)
and
¸
V ∈
F(V ) − → F(V
tt
) − → F(V
t
∩ V
tt
).
In other words, a section s ∈ F() is the data of a family of sections
¦s
V
∈ F(V ); V ∈ ¦ such that for any V
t
, V
tt
∈ ,
s
V
[
V
∩V
= s
V
[
V
∩V
.
Therefore, if F is a presheaf, there is a natural map
F(U) − → F(). (5.3)
The next result is obvious.
Proposition 5.2.2. A presheaf F is separated (resp. is a sheaf) if and
only if for any U ∈ Op
X
and any open covering  of U, the natural map
F(U) − → F() is injective (resp. bijective).
One can consider  as a category. Assuming that  is stable by ﬁnite
intersection, we have
F() · lim
←−
V ∈
F(V ). (5.4)
Note that if F is a sheaf of sets, then F(∅) = ¦pt¦. If F is a sheaf of
kmodules, then F(∅) = 0. If ¦U
i
¦
i∈I
is a family of disjoint open subsets,
then F(
¸
i
U
i
) =
¸
i
F(U
i
).
If F is a sheaf on X, then its restriction F[
U
to an open subset U is a
sheaf.
In the sequel, we shall concentrate on sheaves of kmodules.
5.2. SHEAVES 99
Notation 5.2.3. (i) Let F be a sheaf on X. One can deﬁne its support,
denoted by supp F, as the complementary of the union of all open subsets U
of X such that F[
U
= 0. Note that F[
X\supp F
= 0.
(ii) Let s ∈ F(U). One can deﬁne its support, denoted by supp s, as the
complementary of the union of all open subsets U of X such that s[
U
= 0.
The next result is extremely useful. It says that to check that a morphism
of sheaves is an isomorphism, it is enough to do it at each stalk.
Proposition 5.2.4. Let ϕ : F − → G be a morphism of sheaves.
(i) ϕ is a monomorphism if and only if, for all x ∈ X, ϕ
x
: F
x
− → G
x
is
injective.
(ii) ϕ is an isomorphism if and only if, for all x ∈ X, ϕ
x
: F
x
− → G
x
is an
isomorphism.
Proof. (i) The condition is necessary by Proposition 5.1.5. Assume now ϕ
x
is injective for all x ∈ X and let us prove that ϕ : F(U) − → G(U) is injective.
Let s ∈ F(U) with ϕ(s) = 0. Then (ϕ(s))
x
= 0 = ϕ
x
(s
x
), and ϕ
x
being
injective, we ﬁnd s
x
= 0 for all x ∈ U. This implies that there exists an open
covering U = ∪
i
U
i
, with s[
U
i
= 0, and by S1, s = 0.
(ii) The condition is clearly necessary. Assume now ϕ
x
is an isomorphism
for all x ∈ X and let us prove that ϕ : F(U) − → G(U) is surjective. Let
t ∈ G(U). There exists an open covering U = ∪
i
U
i
and s
i
∈ F(U
i
) such that
t[
U
i
= ϕ(s
i
).
Then, ϕ(s
i
)[
U
i
∩U
j
= ϕ(s
j
)[
U
i
∩U
j
, hence by (i), s
i
[
U
i
∩U
j
= s
j
[
U
i
∩U
j
and by
S2, there exists s ∈ F(U) with s[
U
i
= s
i
. Since ϕ(s)[
U
i
= t[
U
i
, we have
ϕ(s) = t, by S1. q.e.d.
Examples 5.2.5. (i) The presheaf (
0
X
is a sheaf.
(ii) Let M ∈ Mod(k). The presheaf of locally constant functions on X with
values in M is a sheaf, called the constant sheaf with stalk M and denoted
M
X
. Note that the constant presheaf with stalk M is not a sheaf except if
M = 0.
(iii) Let X be a topological space in (a) below, a real manifold of class C
∞
in (b)–(d), a complex analytic manifold in (e)–(h). We have the classical
sheaves:
(iii)–(a) k
X
: kvalued locally constant functions, (iii)–(b) (
∞
X
: complex val
ued functions of class (
∞
,
(iii)–(c) Tb
X
: complex valued distributions,
(iii)–(d) (
∞,(p)
X
: pforms of class (
∞
, also denoted Ω
p
X
,
100 CHAPTER 5. ABELIAN SHEAVES
(iii)–(e) O
X
: holomorphic functions,
(iii)–(f) Ω
p
X
: holomorphic pforms (hence, Ω
0
X
= O
X
).
(iv) On a topological space X, the presheaf U → (
0,b
X
(U) of continuous
bounded functions is not a sheaf in general. To be bounded is not a local
property and axiom S2 is not satisﬁed.
(v) Let X = C, and denote by z the holomorphic coordinate. The holomor
phic derivation
∂
∂z
is a morphism from Ø
X
to Ø
X
. Consider the presheaf:
F : U → O(U)/
∂
∂z
O(U),
that is, the presheaf Coker(
∂
∂z
: O
X
− → O
X
). For U an open disk, F(U) = 0
since the equation
∂
∂z
f = g is always solvable. However, if U = C ` ¦0¦,
F(U) = 0. Hence the presheaf F does not satisfy axiom S1.
(vi) If F is a sheaf on X and U is open, then F[
U
is sheaf on U.
5.3 Sheaf associated with a presheaf
Consider the forgetful functor
ι
X
: Mod(k
X
) − → PSh(k
X
) (5.5)
which, to a sheaf F associates the underlying presheaf. In this section, we
shall rapidly construct a left adjoint to this functor.
When there is no risk of confusion, we shall often omit the symbol ι
X
. In
other words, we shall identify a sheaf and the underlying presheaf.
Theorem 5.3.1. The forgetful functor ι
X
in (5.5) admits a left adjoint
a
: Mod(k
X
) − → PSh(k
X
). (5.6)
More precisely, one has the isomorphism, functorial with respect to F ∈
PSh(k
X
) and G ∈ Mod(k
X
)
(5.7) Hom
PSh(k
X
)
(F, ι
X
G) · Hom
k
X
(F
a
, G).
Moreover (5.7) deﬁnes a morphism of presheaves θ : F − → F
a
and θ
x
: F
x
− →
F
a
x
is an isomorphism for all x ∈ X.
Note that if F is locally 0, then F
a
= 0. If F is a sheaf, then θ : F − → F
a
is an isomorphism.
If F is a presheaf on X, the sheaf F
a
is called the sheaf associated with
F.
5.3. SHEAF ASSOCIATED WITH A PRESHEAF 101
Proof. Deﬁne:
F
a
(U) =
¦s : U − →
¸
x∈U
F
x
; s(x) ∈ F
x
such that, for all x ∈ U,
there exists V ÷ x, V open in U, and there exists t ∈
F(V ) with t
y
= s(y) for all y ∈ V ¦.
Deﬁne θ : F − → F
a
as follows. To s ∈ F(U), one associates the section of F
a
:
(x → s
x
) ∈ F
a
(U).
One checks easily that F
a
is a sheaf and any morphism of presheaves ϕ :
F − → G with G a sheaf will factorize uniquely through θ. In particular,
any morphism of presheaves ϕ : F − → G extends uniquely as a morphism of
sheaves ϕ
a
: F
a
− → G
a
, and F → F
a
is functorial. q.e.d.
Example 5.3.2. Let M ∈ Mod(k). Then the sheaf associated with the
constant presheaf U → M is the sheaf M
X
of Mvalued locally constant
functions.
Theorem 5.3.3. (i) The category Mod(k
X
) admits projective limits and
such limits commute with the functor ι
X
in (5.5). More precisely, if
¦F
i
¦
i∈I
is a projective system of sheaves, its projective limit in PSh(k
X
)
is a sheaf and is a projective limit in Mod(k
X
).
(ii) The functor
a
in (5.6) commutes with kernels. More precisely, let ϕ :
F − → G be a morphism of presheaves and let ϕ
a
: F
a
− → G
a
denote the
associated morphism of sheaves. Then
(Ker ϕ)
a
· Ker ϕ
a
. (5.8)
(iii) The category Mod(k
X
) admits inductive limits. If ¦F
i
¦
i∈I
is an induc
tive system of sheaves, its inductive limit is the sheaf associated with its
inductive limit in PSh(k
X
).
(iv) The functor
a
commutes with inductive limits (in particular, with coker
nels). More precisely, if ¦F
i
¦
i∈I
is an inductive system of of presheaves,
then
lim
−→
i
(F
a
i
) · (lim
−→
i
F
i
)
a
, (5.9)
where lim
−→
on the left (resp. right) is the inductive limit in the category
of sheaves (resp. of presheaves).
(v) The category Mod(k
X
) is abelian and the functor
a
is exact.
102 CHAPTER 5. ABELIAN SHEAVES
(vi) The functor ι
X
: Mod(k
X
) − → PSh(k
X
) is fully faithful and left exact.
(vii) Filtrant inductive limits are exact in Mod(k
X
).
Let ϕ : F − → G is a morphism of sheaves and let ι
X
ϕ : ι
X
F − → ι
X
G
denote the underlying morphism of presheaves. Then Ker ι
X
ϕ is a sheaf
and coincides with ι
X
Ker ϕ. On the otherhand, one shall be aware that
Coker ι
X
ϕ is not necessarily a sheaf. The cokernel in the category of sheaves
is the sheaf associated with this presheaf. In other words, the functor ι
X
:
Mod(k
X
) − → PSh(k
X
) is left exact, but not right exact in general.
Proof. (i) Let  be an open covering of an open subset U. Since F →
F() commutes with projective limits, (lim
←−
i
F
i
)(U)
∼
−→(lim
←−
i
F
i
)(). Hence a
projective limit of sheaves in the category PSh(k
X
) is a sheaf. One has, for
G ∈ Mod(k
X
):
lim
←−
i
Hom
k
X
(G, F
i
) · lim
←−
i
Hom
PSh(k
X
)
(G, F
i
)
· Hom
PSh(k
X
)
(G, lim
←−
i
F
i
)
· Hom
k
X
(G, lim
←−
i
F
i
).
It follows that lim
←−
i
F
i
is a projective limit in the category Mod(k
X
).
(ii) The commutative diagram
0
Ker ϕ
F
G
0
Ker(ϕ
a
)
F
a
G
a
deﬁnes the morphism Ker ϕ − → Ker ϕ
a
, hence, the morphism ψ : (Ker ϕ)
a
− →
Ker ϕ
a
. Since the functor F → F
x
commutes both with Ker and with
a
, ψ
x
is an isomorphism for all x and it remains to apply Proposition 5.2.4.
(iii)–(iv) Let G ∈ Mod(k
X
) and let ¦F
i
¦
i∈I
be an inductive system of of
presheaves. We have the chain of isomorphisms
Hom
k
X
((lim
−→
i
F
i
)
a
, G) · Hom
PSh(k
X
)
(lim
−→
i
F
i
, G)
· lim
←−
i
Hom
PSh(k
X
)
(F
i
, G)
· lim
←−
i
Hom
k
X
(F
a
i
, G).
5.3. SHEAF ASSOCIATED WITH A PRESHEAF 103
(v) By (i) and (iii), the category Mod(k
X
) admits kernels and cokernels. Let
ϕ : F − → G be a morphism of sheaves and denote by ι
X
ϕ the underlying
morphism of presheaves. Using (5.9) we get that Coimϕ := Coker(Ker ϕ − →
F) is isomorphic to the sheaf associated with Coim(ι
X
ϕ). Using (5.8) we get
that Imϕ := Ker(F − → (Coker ι
X
ϕ)
a
) is isomorphic to the sheaf associated
with Im(ι
X
ϕ). The isomorphism of presheaves Coim(ι
X
ϕ)
∼
−→Im(ι
X
ϕ) yields
the isomorphism of the associated sheaves. Hence Mod(k
X
) is abelian.
The functor F → F
a
is exact since it commutes with kernels by by (5.9)
and with cokernels by (5.8).
(vi) The functor ι
X
is fully faithful by deﬁnition. Since it admits a left
adjoint, it is left exact.
(vii) Filtrant inductive limits are exact in the category Mod(k), whence in
the category PSh(k
X
). Then the result follows since
a
is exact. q.e.d.
Recall that the functor F → F
a
commutes with the functors of restriction
F → F[
U
, as well as with the functor F → F
x
.
Proposition 5.3.4. (i) Let ϕ : F − → G be a morphism of sheaves and
let x ∈ X. Then (Ker ϕ)
x
· Ker ϕ
x
and (Coker ϕ)
x
· Coker ϕ
x
. In
particular the functor F → F
x
, from Mod(k
X
) to Mod(k) is exact.
(ii) Let F
t
ϕ
−→ F
ψ
−→ F
tt
be a complex of sheaves. Then this complex is exact
if and only if for any x ∈ X, the complex F
t
x
ϕx
−→ F
x
ψx
−→ F
tt
x
is exact.
Proof. (i) The result is true in the category of presheaves. Since ι
X
Ker ϕ ·
Ker ι
X
ϕ and Coker ϕ · (Coker ι
X
ϕ)
a
, the result follows.
(ii) By Proposition 5.2.4, Imϕ · Ker ψ if and only if (Imϕ)
x
· (Ker ψ)
x
for
all x ∈ X. Hence the result follows from (i). q.e.d.
By this statement, the complex of sheaves above is exact if and only if for
each section s ∈ F(U) deﬁned in an open neighborhood U of x and satisfying
ψ(s) = 0, there exists another open neighborhood V of x with V ⊂ U and a
section t ∈ F
t
(V ) such that ϕ(t) = s[
V
.
On the other hand, a complex of sheaves 0 − → F
t
− → F − → F
tt
is exact if
and only if it is exact as a complex of presheaves, that is, if and only if, for
any U ∈ [Op
X
, the sequence 0 − → F
t
(U) − → F(U) − → F
tt
(U) is exact.
Examples 5.3.5. Let X be a real analytic manifold of dimension n. The
(augmented) de Rham complex is
(5.10) 0 − → C
X
− → (
∞,(0)
X
d
− → − → (
∞,(n)
X
− → 0
104 CHAPTER 5. ABELIAN SHEAVES
where d is the diﬀerential. This complex of sheaves is exact. The same result
holds with the sheaf (
∞
X
replaced with the sheaf (
ω
X
or the sheaf Tb
X
.
(ii) Let X be a complex manifold of dimension n. The (augmented) holo
morphic de Rham complex is
(5.11) 0 − → C
X
− → Ω
0
X
d
− → − → Ω
n
X
− → 0
where d is the holomorphic diﬀerential. This complex of sheaves is exact.
Deﬁnition 5.3.6. Let U ∈ Op
X
. We denote by Γ(U;
•
) : Mod(k
X
) − →
Mod(k) the functor F → F(U).
Proposition 5.3.7. The functor Γ(U;
•
) is left exact.
Proof. The functor Γ(U;
•
) is the composition
Mod(k
X
)
ι
X
−→ PSh(k
X
)
λ
U
−→ Mod(k),
where λ
U
is the functor F → F(U). Since ι
X
is left exact and λ
U
is exact,
the result follows. q.e.d.
The functor Γ(U;
•
) is not exact in general. Indeed, consider Example 5.2.5
(v). Recall that X = C, z is a holomorphic coordinate and U = X ` ¦0¦.
Then the sequence of sheaves 0 − → C
X
− → O
X
∂z
−→ O
X
− → 0 is exact. Applying
the functor Γ(U;
•
), the sequence one obtains is no more exact.
5.4 Internal operations
1
Internal hom
Deﬁnition 5.4.1. Let F, G ∈ PSh(k
X
). One denotes by Hom
PSh(k
X
)
(F, G)
or simply Hom(F, G) the presheaf on X, U → Hom
PSh(k
U
)
(F[
U
, G[
U
) and
calls it the “internal hom” of F and G.
Proposition 5.4.2. Let F, G ∈ Mod(k
X
). Then the presheaf Hom(F, G) is
a sheaf.
1
The proofs in this section may be skipped
5.4. INTERNAL OPERATIONS 105
Proof. Let U ∈ Op
X
and let  be an open covering of U. Let us show that
Hom(F, G)(U) · Hom(F, G)(). In other words, we shall prove that the
sequence below is exact, (in these formulas, we write Hom instead of Hom
k
W
,
W open, for short):
0 − → Hom
k
X
(F[
U
, G[
U
) − →
¸
V ∈
Hom
k
X
(F[
V
, G[
V
)
⇒
¸
V
,V
∈
Hom
k
X
(F[
V
∩V
, G[
V
∩V
).
(i) Let ϕ ∈ Hom(F[
U
, G[
U
) and assume that ϕ[
V
: F[
V
− → G[
V
is zero for all
V ∈ . Then for V ∈ , any W ∈ Op
U
and any s ∈ F(W), ϕ(s)[
W∩V
= 0.
Since ¦W ∩V ; V ∈ ¦ is a covering of W, ϕ(s) ∈ G(W) is zero. This implies
ϕ = 0.
(ii) Let ¦ϕ
V
¦ belong to
¸
V ∈
Hom(F[
V
, G[
V
). Assume that
ϕ
V
[
V
∩V
= ϕ
V
[
V
∩UV
for any V
t
, V
tt
∈ . Let W ∈ Op
U
. Then ¦ϕ
V
¦ deﬁnes a commutative
diagram
F(W)
a
W
−−→
¸
V ∈
G(W ∩ V ) ⇒
¸
V
,V
∈
G(W ∩ V
t
∩ V
tt
)
where a
W
is given by F(W) ÷ s → ϕ
V
(s[
W∩V
). Since G is a sheaf, a
W
factors
uniquely as
F(W)
ψ(W)
−−−→ G(W) − →
¸
V ∈
G(W ∩ V ).
It is easy to see that ψ : Op
U
÷ W → ψ(W) ∈ Hom(F(W), G(W)) deﬁnes
an element of Hom
k
U
(F[
U
, G[
U
). q.e.d.
The functor Hom
k
X
(
•
,
•
) being left exact, it follows that
Hom(
•
,
•
) : Mod(k
X
)
op
Mod(k
X
) − → Mod(k
X
)
is left exact. Note that
Hom
k
X
(
•
,
•
) · Γ(X;
•
) ◦ Hom(
•
,
•
).
Since a morphism: ϕ : F − → G deﬁnes a klinear map F
x
− → G
x
, we get a
natural morphism (Hom(F, G))
x
− → Hom(F
x
, G
x
). In general, this map is
neither injective nor surjective.
106 CHAPTER 5. ABELIAN SHEAVES
Tensor product
Deﬁnition 5.4.3. Let F, G ∈ Mod(k
X
).
(i) One denotes by F
psh
⊗G the presheaf on X, U → F(U) ⊗
k
G(U).
(ii) One denotes by F ⊗
k
X
G the sheaf associated with the presheaf F
psh
⊗G
(see Deﬁnition 5.4.1) and calls it the tensor product of F and G. If
there is no risk of confusion, one writes F ⊗G instead of F ⊗
k
X
G.
Proposition 5.4.4. Let F, G, K ∈ Mod(k
X
). There are natural isomor
phisms:
Hom
k
X
(G⊗F, H) · Hom
k
X
(G, Hom(F, H)),
Hom(G⊗F, H) · Hom(G, Hom(F, H)).
Proof. (i) Let us deﬁne a map λ: Hom(G, Hom(F, H)) − → Hom(F
psh
⊗G, H).
For U ∈ Op
X
, we have the chain of morphisms
Hom
PSh(k
X
)
(G, Hom(F, H)) − → Hom
k
(G(U), Hom(F, H)(U))
− → Hom
k
(G(U), Hom
k
(F(U), H(U)))
· Hom
k
(G(U) ⊗F(U), H(U)).
Since these morphisms are functorial with respect to U, they deﬁne λ.
Let us deﬁne a map µ: Hom
k
X
(F
psh
⊗G, H) − → Hom
k
X
(G, Hom(F, H)).
For V − → U in (
X
, we have the chain of morphisms
Hom
PSh(k
X
)
(F
psh
⊗G, H) − → Hom
k
(F(V )
psh
⊗G(V ), H(V ))
· Hom
k
(G(V ), Hom
k
(F(V ), H(V )))
− → Hom
k
(G(U), Hom
k
(F(V ), H(V ))).
Since these morphism are functorial with respect to V ⊂ U, they deﬁne µ.
It is easily checked that λ and µ are inverse to each other.
(ii) Applying Theorem 5.3.1 and Proposition 5.4.2, we get the chain of iso
morphisms
Hom
k
X
(G⊗F, H) · Hom
PSh(k
X
)
(G
psh
⊗F, H)
· Hom
PSh(k
X
)
(G, Hom(F, H))
· Hom
k
X
(G, Hom(F, H)).
(iii) The second isomorphism follows from the ﬁrst one. q.e.d.
5.5. DIRECT AND INVERSE IMAGES 107
Example 5.4.5. Let (
∞
X
denote as above the sheaf of real valued (
∞

functions on a real manifold X. If V is a ﬁnite Rdimensional vector space
(e.g., V = C), then the sheaf of V valued (
∞
functions is nothing but
(
∞
X
⊗
R
X
V
X
.
The functor
•
⊗
•
: Mod(k
X
) Mod(k
X
) − → Mod(k
X
)
is the composition of the right exact functor
psh
⊗ and the exact functor
a
. This
functor is thus right exact and if k is a ﬁeld, it is exact. Note that for x ∈ X
and U ∈ Op
X
:
(i) (F ⊗G)
x
· F
x
⊗G
x
,
(ii) Hom(F, G)[
U
· Hom(F[
U
, G[
U
),
(iii) Hom(k
X
, F) · F,
(iv) k
X
⊗F · F.
5.5 Direct and inverse images
Let f : X − → Y be a continuous map. We denote by f
t
the inverse image of
a set by f. Hence, we set for V ⊂ Y : f
t
(V ) := f
−1
(V ) and f
t
: Op
Y
− → Op
X
is a functor. Note that f
t
commutes with products and coverings, that is, it
satisﬁes
for any U, V ∈ Op
Y
, f
t
(U ∩ V ) = f
t
(U) ∩ f
t
(V ),
for any V ∈ Op
Y
and any covering 1 of V , f
t
(1) is a covering
of f
t
(V ).
(5.12)
Deﬁnition 5.5.1. Let U → X be an open embedding. One denotes by
j
t
U
: Op
U
− → Op
X
the functor Op
V
÷ V → V ∈ Op
X
.
The functor j
t
U
satisﬁes (5.12). It is a motivation to introduce the follow
ing deﬁnition which extends the notion of a continuous map.
Deﬁnition 5.5.2. A morphism of sites f : X − → Y is a functor f
t
: Op
Y
− →
Op
X
which satisﬁes (5.12).
One shall be aware that we do not ask that f
t
commutes with ﬁnite
projective limits. In particular, we do not ask f
t
(Y ) = X.
Note that if f : X − → Y and g : Y − → Z are morphisms of sites, then
g ◦ f : X − → Z is also a morphism of sites.
108 CHAPTER 5. ABELIAN SHEAVES
Deﬁnition 5.5.3. Let f : X − → Y be a morphism of sites and let F ∈
PSh(k
X
). One deﬁnes f
∗
F ∈ PSh(k
Y
), the direct image of F by f, by
setting: f
∗
F(V ) = F(f
t
(V )).
Proposition 5.5.4. Let F ∈ Mod(k
X
). Then f
∗
F ∈ Mod(k
Y
). In other
words, the direct image of a sheaf is a sheaf.
Proof. Let V ∈ Op
Y
and let 1 be an open covering of V . Then f
t
(1) is an
open covering of f
t
(V ) and we get
f
∗
F(V ) · F(f
t
(V )) · F(f
t
(1)) · f
∗
F(1).
q.e.d.
Deﬁnition 5.5.5. Let f : X − → Y be a morphism of sites.
(i) Let G ∈ PSh(k
Y
). One deﬁnes f
†
G ∈ PSh(k
X
) by setting for U ∈ Op
X
:
f
†
G(U) = lim
−→
U⊂f
t
(V )
G(V ).
(ii) Let G ∈ Mod(k
Y
). One deﬁnes f
−1
G ∈ Mod(k
X
), the inverse image of
G by f, by setting
f
−1
G = (f
†
G)
a
.
Note that if f is a continuous map, one has for x ∈ X:
(f
−1
G)
x
· (f
†
G)
x
· G
f(x)
. (5.13)
Example 5.5.6. Let U → X be an open embedding and let F ∈ Mod(k
X
).
Then i
†
U
F · j
U∗
F is already a sheaf. Hence:
i
†
U
= i
−1
U
· j
U
∗
. (5.14)
Theorem 5.5.7. Let f : X − → Y be a morphism of sites.
(i) The functor f
−1
: Mod(k
Y
) − → Mod(k
X
) is left adjoint to the functor
f
∗
: Mod(k
X
) − → Mod(k
Y
). In other words, we have for F ∈ Mod(k
X
)
and G ∈ Mod(k
Y
):
Hom
k
X
(f
−1
G, F) · Hom
k
Y
(G, f
∗
F).
(ii) The functor f
∗
is left exact and commutes with projective limits.
5.5. DIRECT AND INVERSE IMAGES 109
(iii) The functor f
−1
is exact and commutes with inductive limits.
(iv) There are natural morphisms of functors id − → f
∗
f
−1
and f
−1
f
∗
− → id.
Proof. (i) First we shall prove that the functor f
†
: PSh(k
Y
) − → PSh(k
X
)
is left adjoint to f
∗
: PSh(k
X
) − → PSh(k
Y
). In other words, we have an
isomorphism, functorial with respect to F ∈ PSh(k
X
) and G ∈ PSh(k
Y
):
Hom
PSh(k
X
)
(f
†
G, F) · Hom
PSh(k
Y
)
(G, f
∗
F). (5.15)
A section ϕ ∈ Hom
PSh(k
Y
)
(G, f
∗
F) is a family of maps
¦ϕ
V
: G(V ) − → F(f
t
(V ))¦
V ∈Op
Y
compatible with the restriction morphisms. Equivalently, this is a family of
maps
¦ϕ
U
: G(V ) −→ F(U); U ⊂ f
t
(V )¦
V ∈Op
Y
,U∈Op
X
compatible with the restriction morphisms. Hence it gives a family of maps
¦ψ
V
: lim
−→
U⊂f
t
(V )
G(V ) − → F(U)¦
U∈Op
X
which deﬁnes ψ ∈ Hom
PSh(k
X
)
(f
†
G, F). Clearly, the correspondence ϕ → ψ
is an isomorphism.
Using the isomorphism (5.15) we get the chain of isomorphisms
Hom
k
Y
(G, f
∗
F) · Hom
PSh(k
Y
)
(G, f
∗
F) · Hom
PSh(k
X
)
(f
†
G, F)
· Hom
k
X
((f
†
G)
a
, F) = Hom
k
X
(f
−1
G, F).
(ii) With the exception of the fact that f
−1
is left exact, the other assertions
follow by the adjunction property. If f is a continuous map, f
−1
is left exact
by (5.13). Let us give a proof in the general case. Let
0 − → G
t
− → G − → G
tt
(5.16)
be an exact sequence of sheaves and let U ∈ Op
X
, V ∈ Op
Y
with U ⊂ f
t
(V ).
The sequence 0 − → G
t
(V ) − → G(V ) − → G
tt
(V ) is exact.
Consider the category (Op
Y
)
U
:= ¦V ∈ Op
Y
, U ⊂ f
t
(V )¦. The sequence
(5.16) deﬁnes an exact sequence of functors from ((Op
Y
)
U
)
op
to Mod(k). The
category ((Op
Y
)
U
)
op
is either ﬁltrant or empty. Indeed, if U ⊂ f
t
(V
1
) and
U ⊂ f
t
(V
2
), then U ⊂ f
t
(V
1
∩ V
2
). It follows that the sequence
0 − → lim
−→
U⊂f
t
(V )
G
t
(V ) − → lim
−→
U⊂f
t
(V )
G(V ) − → lim
−→
U⊂f
t
(V )
G
tt
(V )
is exact. Hence the functor G → f
†
ι
Y
G from Mod(k
Y
) to PSh(k
X
) is left
exact. Since the functor
a
is exact, the result follows. q.e.d.
110 CHAPTER 5. ABELIAN SHEAVES
Consider morphisms of sites f : X − → Y , g : Y − → Z and g ◦ f : X − → Z.
Proposition 5.5.8. One has natural isomorphisms of functors
g
∗
◦ f
∗
· (g ◦ f)
∗
,
f
−1
◦ g
−1
· (g ◦ f)
−1
.
Proof. The functoriality of direct images is clear by its deﬁnition. The func
toriality of inverse images follows by adjunction. q.e.d.
Note that inverse image commutes with the functor
a
(see Exercise 5.9) and
inverse image commutes with tensor product (see Exercise 5.10 (ii)).
Let V ∈ Op
Y
, set U = f
t
(V ) and denote by f
V
the restriction of f to V :
X
f
Y
U
f
V
i
U
V.
i
V
(5.17)
Proposition 5.5.9. Let F ∈ Mod(k
X
). Then i
−1
V
f
∗
F · f
V
∗
i
−1
U
F.
Proof. Consider the morphisms of presites:
X
f
j
U
Y
j
V
U
f
V
V.
(5.18)
By Proposition 5.5.8, one has j
V
∗
f
∗
F · f
V
∗
j
U
∗
F. q.e.d.
We denote by a
X
the canonical map a
X
= X − → ¦pt¦. (Recall that ¦pt¦
is the set with one element.) There is a natural equivalence of categories
Mod(k
pt
)
∼
−→ Mod(k),
F → Γ(pt; F).
In the sequel, we shall identify these two categories.
Examples 5.5.10. (i) Let F ∈ Mod(k
X
). Then:
Γ(X; F) · a
X∗
F.
(ii) Let M ∈ Mod(k). Recall that M
X
denotes the sheaf associated with the
presheaf U → M. Hence:
M
X
· a
−1
X
M
pt¦
.
5.5. DIRECT AND INVERSE IMAGES 111
Let f : X − → Y be a continuous map. Since a
X
= a
Y
◦ f, we get
M
X
· f
−1
M
Y
.
(iii) Let x ∈ X and denote by i
x
: ¦x¦ → X the embedding. Then
i
−1
x
F · F
x
.
(iv) Let i
U
: U → X be the inclusion of an open subset of X and let F be a
sheaf on X. Then Γ(V ; i
−1
U
F) · Γ(V ; F) for V ∈ Op
U
.
(v) Let X = Y
¸
Y , the disjoint union of two copies of Y . Let f : X − → Y
be the natural map which induces the identity one each copy of Y . Then
f
∗
f
−1
G · G⊕G. In fact, if V is open in Y , then we have the isomorphisms
Γ(V ; f
∗
f
−1
G) · Γ(V . V ; f
−1
G)
· Γ(V ; G) ⊕Γ(V ; G).
Example 5.5.11. Let f : X − → Y be a morphism of complex manifolds. To
each open subset V ⊂ Y is associated a natural “pullback” map:
Γ(V ; O
Y
) − → Γ(V ; f
∗
O
X
)
deﬁned by:
ϕ → ϕ ◦ f
We obtain a morphism O
Y
− → f
∗
O
X
, hence a morphism:
f
−1
O
Y
− → O
X
.
For example, if X is closed in Y and f is the injection, f
−1
O
Y
will be the
sheaf on X of holomorphic functions on Y deﬁned in a neighborhood of X. If
f is smooth (locally on X, f is isomorphic to a projection Y Z − → Y ), then
f
−1
O
Y
will be the subsheaf of O
X
consisting of functions locally constant
in the ﬁbers of f.
Examples 5.5.12. (i) Let i
S
: S → X be the embedding of a closed subset
S of X. Then the functor i
S∗
is exact.
(ii) Let i
U
: U → X be the embedding of an open subset U of X. Let
F ∈ Mod(k
X
) and let x ∈ X. Then
(i
U∗
i
−1
U
F)
x
· lim
−→
V ÷x
Γ(U ∩ V ; F).
Notation 5.5.13. On X Y , we denote by q
1
and q
2
the ﬁrst and second
projection, respectively. If F ∈ Mod(k
X
) and G ∈ Mod(k
Y
) we set:
F G = q
−1
1
F ⊗q
−1
2
G.
One can recover the functor ⊗ from . Denote by δ : X → X X the
diagonal embedding and let F
1
and F
2
be in Mod(k
X
). We have:
δ
−1
(F
1
F
2
) = δ
−1
(q
−1
1
F
1
⊗q
−1
2
F
2
) · F
1
⊗F
2
.
112 CHAPTER 5. ABELIAN SHEAVES
5.6 Sheaves associated with a locally closed
subset
Let Z be a subset of X. One denotes by
i
Z
: Z → X (5.19)
the inclusion morphism. One endows Z with the induced topology, and for
F ∈ Mod(k
X
), one sets:
F[
Z
= i
−1
Z
F,
Γ(Z; F) = Γ(Z; F[
Z
).
If Z := U is open, these deﬁnitions agree with the previous ones. The mor
phism F − → i
Z∗
i
−1
Z
F deﬁnes the morphism a
X∗
F − → a
X∗
i
Z∗
i
−1
Z
F · a
Z∗
i
−1
Z
F,
hence the morphism:
Γ(X; F) − → Γ(Z; F). (5.20)
One denotes by s[
Z
the image of a section s of F on X by this morphism.
Replacing X with an open set U containing Z in (5.20), we get the mor
phism Γ(U; F) − → Γ(Z; F). Denote by I
Z
the category of open subsets con
taining Z (the morphisms are the inclusions). Then (I
Z
)
op
is ﬁltrant, F
deﬁnes a functor (I
Z
)
op
− → Mod(k) and we get a morphism
lim
−→
U⊃Z
Γ(U; F) − → Γ(Z; F). (5.21)
The morphism (5.21) is injective. Indeed, if a section s ∈ Γ(U; F) is zero in
Γ(Z; F), then s
x
= 0 for all x ∈ Z, hence s = 0 on an open neighborhood
of Z. One shall be aware that the morphism (5.21) is not an isomorphism
in general. There is a classical result which asserts that if X is paracompact
(e.g., if X is locally compact and countable at inﬁnity) and Z is closed, then
(5.21) is bijective.
A subset Z of a topological space X is relatively Hausdorﬀ if two distinct
points in Z admit disjoint neighborhoods in X. If Z = X, one says that X
is Hausdorﬀ.
Proposition 5.6.1. Let Z be compact subset of X, relatively Hausdorﬀ in X
and let F ∈ Mod(k
X
). Then the natural morphism (5.21) is an isomorphism.
Proof.
2
Let s ∈ Γ(Z; F[
Z
). There exist a ﬁnite family of open subsets
¦U
i
¦
n
i=1
covering Z and sections s
i
∈ Γ(U
i
; F) such that s
i
[
Z∩U
i
= s[
Z∩U
i
.
2
The proof may be skipped
5.6. SHEAVES ASSOCIATED WITH A LOCALLY CLOSED SUBSET113
Moreover, we may ﬁnd another family of open sets ¦V
i
¦
n
i=1
covering Z such
that Z ∩
¯
V
i
⊂ U
i
. We shall glue together the sections s
i
on a neighborhood
of Z. For that purpose we may argue by induction on n and assume n = 2.
Set Z
i
= Z ∩
¯
V
i
. Then s
1
[
Z
1
∩Z
2
= s
2
[
Z
1
∩Z
2
. Let W be an open neighborhood
of Z
1
∩ Z
2
such that s
1
[
W
= s
2
[
W
and let W
i
(i = 1, 2) be an open subset of
U
i
such that W
i
⊃ Z
i
` W and W
1
∩ W
2
= ∅. Such W
i
’s exist thanks to the
hypotheses. Set U
t
i
= W
i
∪ W, (i = 1, 2). Then s
1
[
U
1
∩U
2
= s
2
[
U
1
∩U
2
. This
deﬁnes t ∈ Γ(U
t
1
∪ U
t
2
; F) with t[
Z
= s. q.e.d.
The case of open subsets
Let U be an open subset. Recall that we have the morphisms of sites
j
U
: X − → U, U ⊃ V → V ⊂ X,
i
U
: U − → X, X ⊃ V → U ∩ V ⊂ U.
Hence, we have the pairs of adjoint functors (i
−1
U
, i
U∗
) and (j
−1
U
, j
U
∗
). Clearly,
there is an isomorphism of functors j
U
∗
· i
−1
U
: Mod(k
X
) − → Mod(k
U
).
Deﬁnition 5.6.2. (i) One deﬁnes the functor i
U!
: Mod(k
U
) − → Mod(k
X
)
by setting i
U!
:=j
−1
U
.
(ii) For F ∈ Mod(k
X
), one sets F
U
:=i
U!
i
−1
U
F = j
−1
U
j
U∗
F.
(iii) For F ∈ Mod(k
X
), one sets Γ
U
F :=i
U∗
i
−1
U
F = i
U∗
j
U
∗
F.
(iv) One sets k
XU
:= (k
X
)
U
for short.
Hence, we have the functors
Mod(k
X
) i
−1
U
Mod(k
U
),
i
U∗
i
U!
and the pairs of adjoint functors
(i
−1
U
, i
U∗
) (i
U!
, i
−1
U
).
Note that i
U!
i
−1
U
F − → F deﬁnes the morphism F
U
− → F and F − → i
U∗
i
−1
U
F
deﬁnes the morphism F − → Γ
U
F.
Moreover
(F
U
)
x
· F
x
if x ∈ U,
0 otherwise.
(5.22)
It follows that the functors i
U!
: Mod(k
U
) − → Mod(k
X
) and (
•
)
U
: Mod(k
X
) − →
Mod(k
X
) are exact, and the functor Γ
U
(
•
): Mod(k
X
) − → Mod(k
X
) is left
exact.
114 CHAPTER 5. ABELIAN SHEAVES
Proposition 5.6.3. Let U ⊂ X be an open subset and let F ∈ Mod(k
X
).
(i) We have F
U
· F ⊗k
XU
.
(ii) We have Γ(U; F) · Hom
k
X
(k
XU
, F).
(iii) Let V be another open subset. Then (F
U
)
V
= F
U∩V
.
(iv) Let U
1
and U
2
be two open subsets of X. Then the sequence below is
exact:
0 − → F
U
1
∩U
2
α
−→ F
U
1
⊕F
U
2
β
− → F
U
1
∪U
2
− → 0. (5.23)
Here α = (α
1
, α
2
) and β = β
1
−β
2
are induced by the natural morphisms
α
i
: F
U
1
∩U
2
− → F
U
i
and β
i
: F
U
i
− → F
U
1
∪U
2
.
Proof. The proofs of (i), (iii), (iv) are obvious, using (5.22).
(ii) We have the isomorphisms
Hom
k
X
(k
XU
, F) = Hom
k
X
(j
−1
U
j
U∗
k
X
, F)
· Hom
k
U
(j
U
∗
k
X
, j
U
∗
F)
· Hom
k
U
(k
U
, F[
U
) · F(U).
q.e.d.
The case of closed subsets
Deﬁnition 5.6.4. Let S be a closed subset of X.
(i) For F ∈ Mod(k
X
), one sets F
S
= i
S∗
i
−1
S
F.
(ii) One sets k
XS
:= (k
X
)
S
for short.
Note that F − → i
S∗
i
−1
S
F deﬁnes the morphism F − → F
S
. Moreover
(F
S
)
x
· F
x
if x ∈ S,
0 otherwise.
(5.24)
Proposition 5.6.5. Let S ⊂ X be a closed subset and let F ∈ Mod(k
X
).
(i) Set U :=X ` S. Then the sequence 0 − → F
U
− → F − → F
S
− → 0 is exact in
Mod(k
X
).
(ii) The functor (
•
)
S
: Mod(k
X
) − → Mod(k
X
), F → F
S
is exact.
5.7. LOCALLY CONSTANT AND LOCALLY FREE SHEAVES 115
(iii) We have F
S
· F ⊗k
XS
.
(iv) Let S
t
be another closed subset. Then (F
S
)
S
= F
S∩S
.
(v) Let S
1
and S
2
be two closed subsets of X. Then the sequence below is
exact:
0 − → F
S
1
∪S
2
α
−→ F
S
1
⊕F
S
2
β
− → F
S
1
∩S
2
− → 0. (5.25)
Here α = (α
1
, α
2
) and β = β
1
−β
2
are induced by the natural morphisms
α
i
: F
S
1
∪S
2
− → F
S
i
and β
i
: F
S
i
− → F
S
1
∩S
2
.
Proof. The proof is obvious, using (5.24). q.e.d.
The case of locally closed subsets
A subset Z of X is locally closed if there exists an open neighborhood U of
Z such that Z is closed in U. Equivalently, Z = S ∩ U with U open and S
closed in X. In this case, one sets
F
Z
:= (F
U
)
S
.
One checks easily that this deﬁnition depends only on Z, not on the choice
of U and S. Moreover, (5.24) still holds with Z instead of S.
5.7 Locally constant and locally free sheaves
Locally constant sheaves
Deﬁnition 5.7.1. (i) Let M be a kmodule. Recall that the sheaf M
X
is
the sheaf of locally constant Mvalued functions on X. It is also the
sheaf associated with the constant presheaf U → M.
(ii) A sheaf F on X is constant if it is isomorphic to a sheaf M
X
, for some
M ∈ Mod(k).
(iii) A sheaf F on X is locally constant if there exists an open covering
X =
¸
i
U
i
such that F[
U
i
is a constant sheaf of U
i
.
Recall that a morphism of sheaves which is locally an isomorphism is
an isomorphism of sheaves. However, given two sheaves F and G, it may
exist an open covering ¦U
i
¦
i∈I
of X and isomorphisms F[
U
i
∼
−→G[
U
i
for all
i ∈ I, although these isomorphisms are not induced by a globally deﬁned
isomorphism F − → G.
116 CHAPTER 5. ABELIAN SHEAVES
Example 5.7.2. Consider X = R and consider the Cvalued function t →
exp(t), that we simply denote by exp(t). Consider the sheaf C
X
exp(t)
consisting of functions which are locally a constant multiple of exp(t). Clearly
C
X
exp(t) is isomorphic to the constant sheaf C
X
, hence, is a constant sheaf.
Examples 5.7.3. (i) If X is not connected it is easy to construct locally
constant sheaves which are not constant. Indeed, let X = U
1
. U
2
be a
covering by two nonempty open subsets, with U
1
∩U
2
= ∅. Let M ∈ Mod(k)
with M = 0. Then the sheaf which is 0 on U
1
and M
U
2
on U
2
is locally
constant and not constant.
(ii) Let X = Y = C ` ¦0¦, and let f : X − → Y be the map z → z
2
, where
z denotes a holomorphic coordinate on C. If D is an open disk in Y , f
−1
D
is isomorphic to the disjoint union of two copies of D. Hence, the sheaf
f
∗
k
X
[
D
is isomorphic to k
2
D
, the constant sheaf of rank two on D. However,
Γ(Y ; f
∗
k
X
) = Γ(X; k
X
) = k, which shows that the sheaf f
∗
k
X
is locally
constant but not constant.
(iii) Let X = C ` ¦0¦ with holomorphic coordinate z and consider the dif
ferential operator P = z
∂
∂z
− α, where α ∈ C ` Z. Let us denote by K
α
the
kernel of P acting on O
X
.
Let U be an open disk in X centered at z
0
, and let A(z) denote a primitive
of α/z in U. We have a commutative diagram of sheaves on U:
O
X
exp(−A(z))
z
∂
∂z
−α
O
X
1
z
exp(−A(z))
O
X
∂
∂z
O
X
Therefore, one gets an isomorphism of sheaves K
α
[
U
∼
−→C
X
[
U
, which shows
that K
α
is locally constant, of rank one.
On the other hand, f ∈ O(X) and Pf = 0 implies f = 0. Hence
Γ(X; K
α
) = 0, and K
α
is a locally constant sheaf of rank one on C ` ¦0¦
which is not constant.
Let us show that all locally constant sheaf on the interval [0, 1] are con
stant.
Recall that for M ∈ Mod(k), M
X
is the constant sheaf with stalk M.
Lemma 5.7.4. Let M, n ∈ Mod(k). Then
(i) (M ⊗N)
X
· M
X
⊗N
X
,
(ii) (Hom(M, N))
X
· Hom
k
X
(M
X
, N
X
).
5.7. LOCALLY CONSTANT AND LOCALLY FREE SHEAVES 117
The proof is left as an exercise.
Lemma 5.7.5. let X = U
1
∪U
2
be a covering of X by two open sets. Let F
be a sheaf on X and assume that:
(i) U
12
= U
1
∩ U
2
is connected and non empty,
(ii) F[
U
i
(i = 1, 2) is a constant sheaf.
Then F is a constant sheaf.
Proof. By the hypothesis, there is M
i
∈ Mod(k) and isomorphisms θ
i
:
F[
U
i
∼
−→(M
i
)
X
[
U
i
(i = 1, 2). Since U
1
∩ U
2
is non empty and connected,
M
1
· M
2
and we may assume M
1
= M
2
= M. deﬁne the isomorphism
θ
12
= θ
1
◦ θ
−1
2
: M
X
[
U
1
∩U
2
∼
−→M
X
[
U
1
∩U
2
. Since U
1
∩ U
2
is connected and
non empty, Γ(U
1
∩ U
2
; Hom(M
X
, M
X
)) · Hom(M, M) by Lemma 5.7.4.
Hence, θ
12
deﬁnes an invertible element of Hom(M, M). Using the map
Hom(M, M) − → Γ(X; Hom(M
X
, M
X
)), we ﬁnd that θ
12
extends as an iso
morphism θ : M
X
· M
X
all over X. Now deﬁne the isomorphisms: α
i
:
F[
U
i
∼
−→(M
X
)[
U
i
by α
1
= θ
1
and α
2
= θ[
U
2
◦ θ
2
. Then α
1
and α
2
will glue
together to deﬁne an isomorphism F
∼
−→M
X
. q.e.d.
Proposition 5.7.6. Let I denote the interval [0, 1].
(i) Let F be a locally constant sheaf on I. Then F is a constant sheaf.
(ii) In particular, if t ∈ I, the morphism Γ(I; F) − → F
t
is an isomorphism.
(iii) Moreover, if F = M
I
for a kmodule M, then the composition
M · F
0
∼
←−Γ(I; M
I
)
∼
−→F
1
· M
is the identity of M.
Proof. (i) We may ﬁnd a ﬁnite open covering U
i
, (i = 1, . . . , n) such that F
is constant on U
i
, U
i
∩ U
i+1
(1 ≤ i < n) is non empty and connected and
U
i
∩ U
j
= ∅ for [i −j[ > 1. By induction, we may assume that n = 2. Then
the result follows from Lemma 5.7.5.
(ii)–(iii) are obvious. q.e.d.
118 CHAPTER 5. ABELIAN SHEAVES
Locally free sheaves
A sheaf of kalgebras (or, equivalently, a k
X
algebra) / on X is a sheaf of
kmodules such that for each U ⊂ X, /(U) is endowed with a structure
of a kalgebra, and the operations (addition, multiplication) commute to the
restriction morphisms. A sheaf of Zalgebras is simply called a sheaf of rings.
If / is a sheaf of rings, one deﬁnes in an obvious way the notion of a sheaf
F of (left) /modules (or simply, an /module) as follows: for each open set
U ⊂ X, F(U) is an /(U)module and the action of /(U) on F(U) commutes
to the restriction morphisms. One also naturally deﬁnes the notion of an /
linear morphism of /modules. Hence we have deﬁned the category Mod(/)
of /modules.
Examples 5.7.7. (i) Let A be a kalgebra. The constant sheaf A
X
is a sheaf
of kalgebras.
(ii) On a topological space, the sheaf (
0
X
is a C
X
algebra. If X is open in R
n
,
the sheaf (
∞
X
is a C
X
algebra. The sheaf Tb
X
is a (
∞
X
module.
(iii) If X is open in C
n
, the sheaf O
X
is a C
X
algebra.
The category Mod(/) is clearly an additive subcategory of Mod(k
X
).
Moreover, if ϕ : F − → G is a morphism of /modules, then Ker ϕ and Coker ϕ
will be /modules. One checks easily that the category Mod(/) is abelian,
and the natural functor Mod(/) − → Mod(k
X
) is exact and faithful (but
not fully faithful). Moreover, the category Mod(/) admits inductive and
projective limits and ﬁltrant inductive limits are exact. Now consider a sheaf
of rings /.
Deﬁnition 5.7.8. (i) A sheaf L of /modules is locally free of rank k
(resp. of ﬁnite rank) if there exists an open covering X = ∪
i
U
i
such
that L[
U
i
is isomorphic to a direct sum of k copies (resp. to a ﬁnite
direct sum) of /[
U
i
.
(ii) A locally free sheaf of rank one is called an invertible sheaf.
We shall construct locally constant and locally free sheaves by gluing
sheaves in the '5.8.
5.8 Gluing sheaves
Let X be a topological space, and let X =
¸
i∈I
U
i
be an open covering of
X. One sets U
ij
= U
i
∩ U
j
, U
ijk
= U
ij
∩ U
k
. First, consider a sheaf F on X,
5.8. GLUING SHEAVES 119
set F
i
= F[
U
i
, θ
i
: F[
U
i
∼
−→F
i
, θ
ji
= θ
j
◦ θ
−1
i
. Then clearly:
(5.26)
θ
ii
= id on U
i
,
θ
ij
◦ θ
jk
= θ
ik
on U
ijk
.
The family of isomorphisms ¦θ
ij
¦ satisfying conditions (5.26) is called a 1
cocycle. Let us show that one can reconstruct F from the data of a 1cocycle.
Theorem 5.8.1. Let X =
¸
i∈I
U
i
be an open covering of X and let F
i
be
a sheaf on U
i
. Assume to be given for each pair (i, j) an isomorphism of
sheaves θ
ji
: F
i
[
U
ij
∼
−→F
j
[
U
ij
, these isomorphisms satisfying the conditions
(5.26).
Then there exists a sheaf F on X and for each i isomorphisms θ
i
:
F[
U
i
∼
−→F
i
such that θ
j
= θ
ji
◦ θ
i
. Moreover, (F, ¦θ
i
¦
i∈I
) is unique up to
unique isomorphism.
Clearly, if the F
i
’s are locally constant, then F is locally constant.
Sketch of proof. (i) Existence. For each open subset V of X, deﬁne F(V ) as
the submodule of
¸
i∈I
F
i
(V ∩ U
i
) consisting of families ¦s
i
¦
i
such that for
any (i, j) ∈ I I, θ
ji
(s
i
[
V ∩U
ji
) = s
j
[
V ∩U
ji
. One checks that the presheaf so
obtained is a sheaf, and the isomorphisms θ
i
’s are induced by the projections
¸
k∈I
F
k
(V ∩ U
k
) − → F
i
(V ∩ U
i
).
(ii) Unicity. Let θ
i
: F[
U
i
∼
−→F
i
and λ
i
: G[
U
i
∼
−→F
i
. Then the isomor
phisms λ
−1
i
◦ θ
i
: F[
U
i
− → G[
U
i
will glue as an isomorphism G
∼
−→F on X, by
Proposition 5.4.2. q.e.d.
Example 5.8.2. Assume k is a ﬁeld, and recall that k
denote the multi
plicative group k ` ¦0¦. Let X = S
1
be the 1sphere, and consider a covering
of X by two open connected intervals U
1
and U
2
. Let U
±
12
denote the two
connected components of U
1
∩ U
2
. Let α ∈ k
. One deﬁnes a locally con
stant sheaf L
α
on X of rank one over k by gluing k
U
1
and k
U
2
as follows. Let
θ
ε
: k
U
1
[
U
ε
12
− → k
U
2
[
U
ε
12
(ε = ±) be deﬁned by θ
+
= 1, θ
−
= α.
Assume that k = C. One can give a more intuitive description of the sheaf
L
α
as follows. Let us identify S
1
with [0, 2π]/ ∼, where ∼ is the relation which
identiﬁes 0 and 2π. Choose β ∈ C with exp(iβ) = α. If β / ∈ Z, the function
θ → exp(iβθ) is not well deﬁned on S
1
since it does not take the same value
at 0 and at 2π. However, the sheaf C
X
exp(iβθ) of functions which are
a constant multiple of the function exp(iβθ) is welldeﬁned on each of the
intervals U
1
and U
2
, hence is well deﬁned on S
1
, although it does not have
any global section.
120 CHAPTER 5. ABELIAN SHEAVES
Example 5.8.3. Consider an ndimensional real manifold X of class (
∞
, and
let ¦X
i
, f
i
¦ be an atlas, that is, the X
i
are open subsets of X and f
i
: X
i
∼
−→U
i
is a (
∞
isomorphism with an open subset U
i
of R
n
. Let U
i
ij
= f
i
(X
ij
) and
denote by f
ji
the map
(5.27) f
ji
= f
j
[
X
ij
◦ f
−1
i
[
U
i
ij
: U
i
ij
− → U
j
ij
.
The maps f
ji
are called the transition functions. They are isomorphisms of
class (
∞
. Denote by J
f
the Jacobian matrix of a map f : R
n
⊃ U − → V ⊂
R
n
. Using the formula J
g◦f
(x) = J
g
(f(x)) ◦ J
f
(x), one gets that the locally
constant function on X
ij
deﬁned as the sign of the Jacobian determinant
det J
f
ji
of the f
ji
’s is a 1cocycle. It deﬁnes a sheaf locally isomorphic to Z
X
called the orientation sheaf on X and denoted by or
X
.
Remark 5.8.4. In the situation of Theorem 5.8.1, if /is a sheaf of kalgebras
on X and if all F
i
’s are sheaves of /[
U
i
modules and the isomorphisms θ
ji
are /[
U
ij
linear, the sheaf F constructed in Theorem 5.8.1 will be naturally
endowed with a structure of a sheaf of /modules.
Example 5.8.5. (i) Let X = P
1
(C), the Riemann sphere. Then Ω
X
:= Ω
1
X
is locally free of rank one over O
X
. Since Γ(X; Ω
X
) = 0, this sheaf is not
globally free.
(ii) Consider the covering of X by the two open sets U
1
= C, U
2
= X ` ¦0¦.
One can glue O
X
[
U
1
and O
X
[
U
2
on U
1
∩ U
2
by using the isomorphism f →
z
p
f (p ∈ Z). One gets a locally free sheaf of rank one. For p = 0 this sheaf
is not free.
Exercises to Chapter 5
Exercise 5.1. Let S (resp. U) be a closed (resp. an open) subset of X and
let F ∈ Mod(k
X
).
(i) Prove the isomorphism Γ(X; F
S
) · Γ(S; F[
S
).
(ii) Construct the morphism Γ(X; F
U
) − → Γ(U; F) and prove that it is not
an isomorphism in general.
Exercise 5.2. Assume that X = R, let S be a nonempty closed interval
and let U = X ` S.
(i) Prove that the natural map Γ(X; k
X
) − → Γ(X; k
XS
) is surjective and
deduce that Γ(X; k
XU
) · 0.
(ii) Let x ∈ R.Prove that the morphism k
X
− → k
Xx¦
does not split.
Exercises to Chapter 5 121
Exercise 5.3. Let F ∈ Mod(k
X
). Deﬁne
¯
F ∈ Mod(k
X
) by
¯
F =
¸
x∈X
F
x¦
.
(Here, F
x¦
∈ Mod(k
X
) and the direct sum is calculated in Mod(k
X
), not in
PSh(k
X
).) Prove that F
x
and
¯
F
x
are isomorphic for all x ∈ X, although F
and
¯
F are not isomorphic in general.
Exercise 5.4. Let Z = Z
1
. Z
2
be the disjoint union of two sets Z
1
and Z
2
in X.
(i) Assume that Z
1
and Z
2
are both open (resp. closed) in X. Prove that
k
XZ
· k
XZ
1
⊕k
XZ
2
.
(ii) Give an example which shows that (i) is no more true if one only assume
that Z
1
and Z
2
are both locally closed.
Exercise 5.5. Let X = R
2
, Y = R, S = ¦(x, y) ∈ X; xy ≥ 1¦, and let
f : X − → Y be the map (x, y) → y. calculate f
∗
k
XS
.
Exercise 5.6. Let f : X − → Y be a continuous map, and let Z be a closed
subset of Y . Construct the natural isomorphism f
−1
k
Y Z
∼
−→k
X(f
−1
Z)
.
Exercise 5.7. Assume that X is a compact space and let ¦F
i
¦
i∈I
be a ﬁl
trant inductive system of sheaves on X. Prove the isomorphism lim
−→
i
Γ(X; F
i
)
∼
−→Γ(X; lim
−→
i
F
i
).
Exercise 5.8. Let S be a set endowed with the discrete topology, let p : X
S − → X denote the projection and let F ∈ Mod(k
XS
) and set F
s
= F[
Xs¦
.
Prove that p
∗
F ·
¸
s∈S
F
s
.
Exercise 5.9. Let f : X − → Y be a continuous map and let G ∈ PSh(k
Y
).
Prove the isomorphism (f
†
G)
a ∼
−→f
−1
(G
a
).
Exercise 5.10. (i) Let f : X − → Y be a morphism of sites, let F ∈ Mod(k
X
)
and let G ∈ Mod(k
Y
). Prove that there is a natural isomorphism in Mod(k
Y
)
Hom
k
Y
(G, f
∗
F)
∼
−→f
∗
Hom
k
X
(f
−1
G, F).
(ii) Let G
1
, G
2
∈ Mod(k
Y
). Prove that there is a natural isomorphism in
Mod(k
X
)
f
−1
(G
1
⊗
k
Y
G
2
)
∼
−→f
−1
G
1
⊗
k
X
f
−1
G
2
.
Exercise 5.11. Let X =
¸
i
U
i
be an open covering of X and let F ∈
PSh(k
X
). Assume that F[
U
i
is a sheaf for all i ∈ I. Prove that F is a sheaf.
(Hint: compare F with the sheaf constructed in Theorem 5.8.1.)
122 CHAPTER 5. ABELIAN SHEAVES
Exercise 5.12. Let M be a kmodule and let X be an open subset of R
n
.
Let F be a presheaf such that for any non empty convex open subsets U ⊂ X,
there exists an isomorphism F(U) · M and this isomorphism is compatible
to the restriction morphisms for V ⊂ U. Prove that the associated sheaf is
locally constant.
Exercise 5.13. Prove Lemma 5.7.4.
Exercise 5.14. Assume k is a ﬁeld, and let L be a locally constant sheaf
of rank one over k (hence, L is locally isomorphic to the sheaf k
X
). Set
L
∗
= Hom(L, k
X
).
(i) Prove the isomorphisms L
∗
⊗L
∼
−→k
X
and k
X
∼
−→Hom(L, L).
(ii) Assume that k is a ﬁeld, X is connected and Γ(X; L) = 0. Prove that
L · k
X
. (Hint: Γ(X; L) · Γ(X; Hom(k
X
, L).)
Chapter 6
Cohomology of sheaves
We ﬁrst show that the category of abelian sheaves has enough injective ob
jects. This allows us to derive all left exact functors we have constructed.
Then, using the tools on simplicial complexes of '3.3, we construct reso
lutions of sheaves using open or closed
ˇ
Cech coverings.
Next we prove an important theorem which asserts that the cohomology
of constant sheaves is a homotopy invariant. Finally, we apply this result to
calculate the cohomology of some classical manifolds.
Some references: [12], [3], [10], [17], [18].
6.1 Cohomology of sheaves
A sheaf F of kmodules is injective if it is an injective object in the category
Mod(k
X
).
Lemma 6.1.1. (i) Let X and Y be two topological spaces and let f : X − →
Y be a morphism of sites. Assume that F ∈ Mod(k
X
) is injective. Then
f
∗
F is injective in Mod(k
Y
).
(ii) Let i
U
: U → X be an open embedding and let F ∈ Mod(k
X
) be injec
tive. Then i
−1
U
F is injective in Mod(k
U
).
Proof. (i) follows immediately from the adjunction formula:
Hom
k
X
(f
−1
(), F) · Hom
k
Y
(, f
∗
F)
and the fact that the functor f
−1
is exact.
(ii) follows from (i) since i
−1
U
· j
U
∗
. q.e.d.
Theorem 6.1.2. The category Mod(k
X
) admits enough injectives.
123
124 CHAPTER 6. COHOMOLOGY OF SHEAVES
Proof. (i) When X = ¦pt¦, the result follows from Mod(k
X
) · Mod(k).
(ii) Assume X is discrete. Then for F, G ∈ Mod(k
X
), the natural morphism
Hom
k
X
(G, F) − →
¸
x∈X
Hom
k
(G
x
, F
x
)
is an isomorphism. Since products are exact in Mod(k), it follows that a
sheaf F is injective as soon as each F
x
is injective. For each x ∈ X, choose
an injective module I
x
together with a monomorphism F
x
I
x
and deﬁne the
sheaf F
0
on X by setting Γ(U, F
0
) =
¸
x∈U
I
x
. Since the topology on X is
discrete, (F
0
)
x
= I
x
. Therefore the sequence 0 − → F − → F
0
is exact and F
0
is injective.
(iii) Let
ˆ
X denote the set X endowed with the discrete topology, and let
f :
ˆ
X − → X be the identity map. Let F ∈ Mod(k
X
). There exists an
injective sheaf G
0
on
ˆ
X and a monomorphism 0 − → f
−1
F − → G
0
. Then f
∗
G
0
is injective in Mod(k
X
) and the sequence 0 − → f
∗
f
−1
F − → f
∗
G
0
is exact. To
conclude, notice that the morphism F − → f
∗
f
−1
F is a monomorphism, since
on an open subset U of X it is deﬁned by F(U) − →
¸
x∈U
F
x
. q.e.d.
It is now possible to derive all left exact functors deﬁned on the category of
sheaves, as well as the bifunctors Hom
k
X
and Hom
k
X
. The derived functors
of these two bifunctors are respectively denoted by Ext
j
k
X
and cxt
j
k
X
. Recall
that, for G, F ∈ Mod(k
X
), the kmodule Ext
j
k
X
(G, F) is calculated as follows.
Choose an injective resolution F
•
of F. Then
Ext
j
k
X
(G, F) · H
j
(Hom
k
X
(G, F
•
)). (6.1)
Let F ∈ Mod(k
X
) and let U (resp. S, resp. Z) be an open (resp. a closed,
resp. a locally closed) subset of X.
As usual, one denotes by i
Z
: Z → X the embedding of Z in X and by
a
Z
the map Z − → ¦pt¦. Note that a
Z
= a
X
◦ i
Z
. Recall that for a sheaf F on
X, we have set: Γ(Z; F) = Γ(Z; F[
Z
). One sets
H
j
(U; F) = R
j
Γ(U; )(F). (6.2)
By (6.1), we have
H
j
(U; F) · Ext
j
k
X
(k
XU
, F). (6.3)
Proposition 6.1.3. (i) If U is open in X, then H
j
(U; F) · H
j
(U; F[
U
).
(ii) If S is closed in X, then H
j
(X; F
S
) · H
j
(S; F[
S
).
6.1. COHOMOLOGY OF SHEAVES 125
(iii) If K is compact and relatively Hausdorﬀ in X, then the natural mor
phism lim
−→
U⊃K
H
j
(U; F) − → H
j
(K; F) is an isomorphism.
Proof. (i) We have the chain of isomorphisms:
H
j
(U; F) · R
j
(a
U∗
i
−1
U
)F · (R
j
a
U∗
)i
−1
U
F
· H
j
(U; F[
U
).
The second isomorphism follows from the fact that i
−1
U
is exact and sends
injective sheaves to injective sheaves (Lemma 4.6.10 (iv)).
(ii) We have the chain of isomorphisms:
H
j
(X; F
S
) · (R
j
a
X∗
)(i
S∗
i
−1
S
)F · R
j
(a
X
◦ i
S∗
)i
−1
S
F
· H
j
(S; F[
S
).
The second isomorphism follows from the fact that i
S∗
is exact and sends
injective sheaves to injective sheaves (Lemma 4.6.10 (iv)).
(iii) The result is true for j = 0 by Proposition 5.6.1. Consider an injective
resolution F − → F
•
of F. Then
H
j
(K; F) · H
j
(Γ(K; F
•
)) · H
j
( lim
−→
U⊃K
Γ(U; F
•
))
· lim
−→
U⊃K
H
j
(Γ(U; F
•
)) · lim
−→
U⊃K
H
j
(U; F).
q.e.d.
Let 0 − → F
t
− → F − → F
tt
− → 0 be an exact sequence of sheaves. Applying
a left exact functor Ψ to it, we obtain a long exact sequence
0 − → Ψ(F
t
) − → Ψ(F) − → − → R
j
Ψ(F) − → R
j
Ψ(F
tt
) − → R
j+1
Ψ(F
t
) − →
For example, applying the functor Γ(X; ), we get a long exact sequence
(6.4) H
k−1
(X; F
tt
) − → H
k
(X; F
t
) − → H
k
(X; F) − → H
k
(X; F
tt
) − →
There are similar results, replacing Γ(X; ) with other functors, such as f
∗
.
Proposition 6.1.4. Let S
1
and S
2
be two closed subsets of X and set S
12
=
S
1
∩ S
2
. Let F ∈ Mod(k
X
). There is a long exact sequence
− → H
j
(X; F
S
1
∪S
2
) − → H
j
(X; F
S
1
) ⊕H
j
(X; F
S
2
) − → H
j
(X; F
S
12
)
− → H
j+1
(X; F
S
1
∪S
2
) − →
126 CHAPTER 6. COHOMOLOGY OF SHEAVES
Proof. Apply the functor Γ(X; ) to the exact sequence of sheaves 0 − →
F
S
1
∪S
2
− → F
S
1
⊕F
S
2
− → F
S
12
− → 0. q.e.d.
Proposition 6.1.5. Let U
1
and U
2
be two open subsets of X and set U
12
=
U
1
∩ U
2
. Let F ∈ Mod(k
X
). There is a long exact sequence
− → H
j
(U
1
∪ U
2
; F) − → H
j
(U
1
; F) ⊕H
j
(U
2
; F) − → H
j
(U
12
; F)
− → H
j+1
(U
1
∪ U
2
; F) − →
Proof. Apply the functor Hom
k
X
(
•
, F) to the exact sequence of sheaves 0 − →
k
XU
12
− → k
XU
1
⊕k
XU
2
− → k
U
1
∪U
2
− → 0 and use (6.3). q.e.d.
6.2
ˇ
Cech complexes for closed coverings
Let I be a ﬁnite totally ordered set. In this section, we shall follow the same
notations as for Koszul complexes. For J ⊂ I, we denote by [J[ its cardinal
and for J = ¦i
0
< < i
p
¦, we set
e
i
0
∧ ∧ e
ip
∈
p+1
Z
[I[
.
Let o = ¦S
i
¦
i∈I
be a family indexed by I of closed subsets of X and let
F ∈ Mod(k
X
). For J ⊂ I we set
S
J
:= ∩
j∈J
S
j
, S =
¸
i∈I
S
i
F
p
S
:=
[J[=p+1
F
S
J
⊗e
J
, F
−1
S
= F
S
.
For J ⊂ I
ord
and a ∈ I, a / ∈ J, we denote by J
a
= J ∪ ¦a¦ the ordered
subset of the ordered set I and we denote by r
J,a
: F
S
J
− → F
S
Ja
the natural
restriction morphism.
We set
δ
J,a
:=r
J,a
(
•
) ⊗e
a
∧
•
: F
S
J
⊗e
J
− → F
S
Ja
⊗e
Ja
. (6.5)
The morphisms δ
J,a
(J ⊂ I, a / ∈ J) in (6.5) deﬁne the morphisms
d
p
: F
p
S
− → F
p+1
S
.
Clearly, d
p+1
◦ d
p
= 0 and we obtain a complex
F
•
S
:= 0 − → F
S
d
−1
−−→ F
0
S
d
0
−→ F
1
S
d
1
−→ . (6.6)
6.3. INVARIANCE BY HOMOTOPY 127
Proposition 6.2.1. Consider a family o = ¦S
i
¦
i∈I
of closed subsets of X
indexed by a ﬁnite totally ordered set I. Then the complex (6.6) is exact.
Proof. It is enough to check that the stalk of the complex (6.6) at each x ∈ X
is exact. Hence, we may assume that x belongs to all S
i
’s. Set M = F
x
. Then
the complex (6.6) is a Koszul complex K
•
(M, ϕ) where ϕ = ¦ϕ
i
¦
i∈I
and all
ϕ
i
are id
M
. The sequence ϕ being both regular and coregular, this complex
is exact. q.e.d.
Example 6.2.2. Assume that X = S
0
∪ S
1
∪ S
2
, where the S
i
’s are closed
subsets. We get the exact complex of sheaves
0 − → F
d
−1
−−→ F
S
0
⊕F
S
1
⊕F
S
2
d
0
−→ F
S
12
⊕F
S
02
⊕F
S
01
d
1
−→ F
S
012
− → 0.
Let us denote by
s
i
: F − → F
S
i
, s
a
ij
: F
Sa
− → F
S
ij
, s
b
k
: F
S
ij
− → F
S
012
(a, i, j, k) ∈ ¦0, 1, 2¦),
the natural morphisms. Then
d
−1
=
¸
s
0
,
s
1
,
s
2
, d
0
=
¸
0 −s
1
12
s
2
12
s
0
02
0 −s
2
02
−s
0
01
s
1
01
0
d
1
= (s
b
2
, −s
b
0
, s
b
1
).
6.3 Invariance by homotopy
In this section, we shall prove that the cohomology of locally constant sheaves
is an homotopy invariant. First, we deﬁne what it means.
In the sequel, we denote by I the closed interval I = [0, 1].
Deﬁnition 6.3.1. Let X and Y be two topological spaces.
(i) Let f
0
and f
1
be two continuous maps from X to Y . One says that f
0
and f
1
are homotopic if there exists a continuous map h : I X − → Y
such that h(0, ) = f
0
and h(1, ) = f
1
.
(ii) Let f : X − → Y be a continuous map. One says that f is a homotopy
equivalence if there exists g : Y − → X such that f ◦ g is homotopic to
id
Y
and g ◦ f is homotopic to id
X
. In such a case one says that X and
Y are homotopic.
(iii) One says that a topological space X is contractible if X is homotopic
to a point ¦x
0
¦.
128 CHAPTER 6. COHOMOLOGY OF SHEAVES
If f
0
, f
1
: X ⇒Y are homotopic, one gets the diagram
X · ¦t¦ X
it
I X
p
h
Y
X
(6.7)
where t ∈ I, i
t
: X · ¦t¦ X − → I X is the embedding, p is the projection
and f
t
= h ◦ i
t
, t = 0, 1.
One checks easily that the relation “f
0
is homotopic to f
1
” is an equiva
lence relation.
A topological space is contractible if and only if there exist g : ¦x
0
¦ − → X
and f : X − → ¦x
0
¦ such that f ◦ g is homotopic to id
X
. Replacing x
0
with
g(x
0
), this means that there exists h : I X − → X such that h(0, x) = id
X
and h(1, x) is the map x → x
0
. Note that contractible implies non empty.
Examples 6.3.2. (i) Let V be a real vector space. A non empty convex set
in V as well as a closed non empty cone are contractible sets.
(ii) Let X = S
n−1
be the unit sphere of the Euclidian space R
n
and let
Y = R
n
`¦0¦. The embedding f : S
n−1
→ R
n
`¦0¦ is a homotopy equivalence.
Indeed, denote by g : R
n
`¦0¦ − → S
n−1
the map x → x/[[x[[. Then g◦f = id
X
and f ◦ g is homotopic to id
Y
. The homotopy is given by the map h(x, t) =
(t/[[x[[ + 1 −t)x.
Statement of the main theorem
Let f : X − → Y be a continuous map and let G ∈ Mod(k
Y
). Remark that
a
X
· a
Y
◦ f. The morphism of functors id − → f
∗
◦ f
−1
deﬁnes the morphism
R
j
a
Y ∗
− → R
j
(a
Y ∗
◦ f
∗
◦ f
−1
)
· R
j
(a
X∗
◦ f
−1
).
Using Theorem 4.6.10, we get the morphism R
j
(a
X∗
◦ f
−1
) − → (R
j
a
X∗
) ◦ f
−1
,
from which we deduce the natural morphisms:
(6.8) f
j
: H
j
(Y ; G) − → H
j
(X; f
−1
G).
Lemma 6.3.3. Let f : X − → Y and g : Y − → Z be continuous maps. Then
g
j
◦ f
j
= (f ◦ g)
j
6.3. INVARIANCE BY HOMOTOPY 129
Sketch of proof. The morphisms of functors id − → f
∗
◦ f
−1
and id − → g
∗
◦ g
−1
deﬁne
a
Z∗
− → a
Z∗
◦ g
∗
◦ g
−1
· a
Y ∗
◦ g
−1
− → a
Y ∗
◦ f
∗
◦ f
−1
◦ g
−1
· a
X∗
◦ f
−1
◦ g
−1
.
One checks easily that the composition
a
Z∗
− → a
Y ∗
◦ g
−1
− → a
X∗
◦ f
−1
◦ g
−1
is the same as the morphism
a
Z∗
− → a
Z∗
◦ g ◦ f
∗
◦ g ◦ f
−1
· a
X∗
◦ (g ◦ f)
−1
.
Applying the functor R
j
(
•
) we ﬁnd the commutative diagram:
R
j
a
Z∗
R
j
(a
X∗
f
−1
g
−1
)
R
j
(a
X∗
f
−1
) ◦ g
−1
(R
j
a
X∗
) ◦ f
−1
◦ g
−1
)
R
j
a
Z∗
R
j
(a
Y ∗
g
−1
)
(R
j
a
Y ∗
) ◦ g
−1
.
f
j
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
The composition of the arrows on the top gives the morphism (g ◦ f)
j
and
the composition of the arrows on the bottom gives the morphism g
j
. q.e.d.
The aim of this section is to prove:
Theorem 6.3.4 (Invariance by homotopy Theorem).
1
Let f
0
, f
1
:
X ⇒Y be two homotopic maps, and let G be a locally constant sheaf on Y .
Consider the two morphisms f
j
t
: H
j
(Y ; G) − → H
j
(X; f
−1
t
G), for t = 0, 1.
Then there exists an isomorphism θ
j
: H
j
(X; f
−1
0
G) − → H
j
(X; f
−1
1
G) such
that θ
j
◦ f
j
0
= f
j
1
.
If G = M
Y
for some M ∈ Mod(k), then, identifying f
−1
t
M
Y
with M
X
(t = 0, 1), we have f
j
1
= f
j
0
.
Proof of the main theorem
In order to prove Theorem 6.3.4, we need several preliminary results.
For F ∈ Mod(k
I
) and a closed interval [0, t] ⊂ I, we write H
j
([0, t]; F)
instead of H
j
([0, t]; F[
[0,t]
) for short.
Lemma 6.3.5. Let F ∈ Mod(k
I
). Then:
1
The proof of this theorem may be skipped
130 CHAPTER 6. COHOMOLOGY OF SHEAVES
(i) For j > 1, one has H
j
(I; F) = 0.
(ii) If F(I) − → F
t
is an epimorphism for all t ∈ I, then H
1
(I; F) = 0.
Proof. Let j ≥ 1 and let s ∈ H
j
(I; F). For 0 ≤ t
1
≤ t
2
≤ 1, consider the
morphism:
f
t
1
,t
2
: H
j
(I; F) − → H
j
([t
1
, t
2
]; F)
and let
J = ¦t ∈ [0, 1]; f
0,t
(s) = 0¦.
Since H
j
(¦0¦; F) = 0 for j ≥ 1, we have 0 ∈ J. Since f
0,t
(s) = 0 im
plies f
0,t
(s) = 0 for 0 ≤ t
t
≤ t, J is an interval. Since H
j
([0, t
0
]; F) =
lim
−→
t>t
0
H
j
([0, t]; F), this interval is open. It remains to prove that J is closed.
For 0 ≤ t ≤ t
0
, consider the MayerVietoris sequence (Proposition 6.1.4):
− → H
j
([0, t
0
]; F) − → H
j
([0, t]; F) ⊕H
j
([t, t
0
]; F) − → H
j
(¦t¦; F) − →
For j > 1, or else for j = 1 assuming H
0
(I; F) − → H
0
(¦t¦; F) is surjective,
we obtain:
(6.9) H
j
([0, t
0
]; F) · H
j
([0, t]; F) ⊕H
j
([t, t
0
]; F).
Let t
0
= sup ¦t; t ∈ J¦. Then f
0,t
(s) = 0, for all t < t
0
. On the other hand,
lim
−→
t<t
0
H
j
([t, t
0
]; F) = 0.
Hence, there exists t < t
0
with f
t,t
0
(s) = 0. By (6.9), this implies f
0,t
0
(s) = 0.
Hence t
0
∈ J. q.e.d.
Recall that the maps p: I X − → X and i
t
: X − → I X are deﬁned in (6.7).
We also introduce the notation I
x
:=I ¦x¦.
Lemma 6.3.6. Let G ∈ Mod(k
IX
). Then (R
j
p
∗
G)
x
· H
j
(I
x
; G[
Ix
).
Proof. Let G
•
be an injective resolution of G. We have the isomorphisms
(R
j
p
∗
G)
x
· (H
j
(p
∗
G
•
))
x
· H
j
((p
∗
G
•
)
x
)
· H
j
(lim
−→
x∈U
Γ(I U; G
•
))
· lim
−→
x∈U
H
j
(I U; G) · H
j
(I
x
; G[
Ix
).
where the last isomorphism follows from Proposition 6.1.3 (iii). q.e.d.
6.3. INVARIANCE BY HOMOTOPY 131
Lemma 6.3.7. Let G ∈ Mod(k
IX
) be a locally constant sheaf. Then the
natural morphism p
−1
p
∗
G − → G is an isomorphism.
Proof. One has
(p
−1
p
∗
G)
(t,x)
· (p
∗
G)
x
· Γ(I
x
; G[
Ix
) · G
(t,x)
.
Here the last isomorphism follows from Proposition 5.7.6. q.e.d.
Lemma 6.3.8. Let F ∈ Mod(k
X
). Then F
∼
−→p
∗
p
−1
F and (R
j
p
∗
)p
−1
F = 0
for j ≥ 1.
Proof. Let x ∈ X and let t ∈ I. Using Lemma 6.3.6 one gets the isomorphism
((R
j
p
∗
)p
−1
F)
x
· H
j
(I
x
; p
−1
F[
Ix
). Then this group is 0 for j > 0 by Lemma
6.3.5 and is isomorphic to (p
−1
F)
t,x
· F
x
for j = 0. q.e.d.
Lemma 6.3.9. Let F ∈ Mod(k
X
).
(i) The morphisms p
j
: H
j
(X; F) − → H
j
(I X; p
−1
F) are isomorphisms.
(ii) The morphisms i
j
t
: H
j
(I X; p
−1
F) − → H
j
(X; F) are isomorphisms
and do not depend on t ∈ I.
Proof. (i) We know that R
j
p
∗
(p
−1
F) = 0 for j ≥ 1. Applying Proposi
tion 4.6.10 (iii) to the functors a
X∗
, p
∗
and the object p
−1
F, we get that
R
j
(a
X∗
p
∗
)(p
−1
F) · R
j
a
X∗
(p
∗
p
−1
F) · R
j
a
X∗
F. Hence p
j
is an isomor
phism.
(ii) By Lemma 6.3.3, i
j
t
◦p
j
is the identity, and p
j
is an isomorphism by (i).
Hence, i
j
t
which is the inverse of p
j
does not depend on t. q.e.d.
End of the proof of Theorem 6.3.4. (i) Since h
−1
G is locally constant, the
morphism p
−1
p
∗
h
−1
G − → h
−1
G is an isomorphism by Lemma 6.3.7. By
Lemma 6.3.9 (ii),
i
j
t
: H
j
(I X; p
−1
p
∗
h
−1
G) − → H
j
(X; i
−1
t
p
−1
p
∗
h
−1
G)
is an isomorphism. Therefore
i
j
t
: H
j
(I X; h
−1
G) − → H
j
(X; i
−1
t
h
−1
G)
is an isomorphism. By Lemma 6.3.3, f
j
t
= i
j
t
◦ h
j
. Set θ = i
j
0
i
j
1
−1
. Then
f
j
0
= i
j
0
◦ h
j
= i
j
0
i
j
1
−1
i
j
1
◦ h
j
= θ ◦ f
j
0
.
(ii) If G = M
Y
, then h
−1
G · M
IX
= p
−1
M
X
and i
j
t
does not depend on t
by Lemma 6.3.9. q.e.d.
132 CHAPTER 6. COHOMOLOGY OF SHEAVES
Applications of Theorem 6.3.4
Corollary 6.3.10. Assume f : X − → Y is a homotopy equivalence and let G
be a locally constant sheaf on Y . Then H
j
(X, f
−1
G) · H
j
(Y ; G).
In other words, the cohomology of locally constant sheaves on topological
spaces is a homotopy invariant.
Proof. Let g : Y − → X be a map such that f ◦g and g◦f are homotopic to the
identity of Y and X, respectively. Consider f
j
: H
j
(Y ; G) − → H
j
(X; f
−1
G)
and g
j
: H
j
(X; f
−1
G) − → H
j
(Y ; G). Then: (f ◦ g)
j
= g
j
◦ f
j
· id
j
X
= id
and (g ◦ f)
j
= f
j
◦ g
j
· id
j
Y
= id . q.e.d.
Corollary 6.3.11. If X is contractible and M ∈ Mod(k), then Γ(X; M
X
) ·
M and H
j
(X; M
X
) · 0 for j > 0.
We shall apply this result together with the technique of MayerVietoris
sequences to calculate the cohomology of various spaces. We shall follow the
notations of Section 5.6.
Theorem 6.3.12. Let X =
¸
i∈I
Z
i
be a ﬁnite covering of X by closed subsets
satisfying the condition
(6.10) for each non empty subset J ⊂ I, Z
J
is contractible or empty.
Let F be a locally constant sheaf on X. Then H
j
(X; F) is isomorphic to the
jth cohomology object of the complex
Γ(X; F
•
?
) := 0 − → Γ(X; F
0
?
)
d
− → Γ(X; F
1
?
) − →
Proof. Recall (Proposition 6.1.3) that if Z is closed in X, then Γ(X; F
Z
) ·
Γ(Z; F[
Z
). Therefore the sheaves F
p
?
(p ≥ 0) are acyclic with respect to the
functor Γ(X; ), by Corollary 6.3.11. Applying Proposition 6.2.1, the result
follows from Theorem 4.6.9. q.e.d.
Corollary 6.3.13. (A particular case of the universal coeﬃcients formula.)
In the situation of Theorem 6.3.12, let M be a ﬂat kmodule. Then for all j
there are natural isomorphisms H
j
(X; M
X
) · H
j
(X; k
X
) ⊗M.
Proof. The kmodule H
j
(X; M
X
) is the jth cohomology object of the com
plex Γ(X; M
•
?
). Clearly,
Γ(X; M
•
?
) · Γ(X; k
•
?
) ⊗M
6.3. INVARIANCE BY HOMOTOPY 133
Since M is ﬂat we have for any bounded complex of modules N
•
:
H
j
(N
•
⊗M) · H
j
(N
•
) ⊗M.
Hence,
H
j
(Γ(X; M
•
?
)) · H
j
(Γ(X; k
•
?
)) ⊗M
To conclude, apply Theorem 6.3.12 to both sides. q.e.d.
Proposition 6.3.14. (A particular case of the K¨ unneth formula.) Let X and
Y be two topological spaces which both admit ﬁnite closed coverings satisfying
condition (6.10). Let F (resp. G) be a locally constant sheaf on X with ﬁber
M, (resp. on Y with ﬁber N). Assume that k is a ﬁeld. Then there are
natural isomorphisms:
H
p
(X Y ; F G) ·
i+j=p
H
i
(X; F) ⊗H
j
(Y ; G).
Sketch of the proof. First, notice that F G is locally constant on X Y .
Next, denote by o = ¦S
i
¦
i∈I
(resp. Z = ¦Z
j
¦
j∈J
) a ﬁnite covering of X
(resp. Y ) satisfying condition (6.10). Since the product of two contractible
sets is clearly contractible, the covering o Z = ¦S
i
Z
j
¦
(i,j)∈IJ
is a ﬁnite
covering of X Y satisfying condition (6.10). Then H
p
(X Y ; F G) is
the pth cohomology object of the complex
Γ(X Y ; (F G)
•
S?
).
One checks that this complex is the simple complex associated with the
double complex
Γ(X Y ; F
•
S
G
•
?
)
and this double complex is isomorphic to
Γ(X; F
•
S
) ⊗Γ(Y ; G
•
?
)
It remains to apply the result of Exercise 4.14. q.e.d.
It may be convenient to reformulate the K¨ unneth formula by saying that
H
p
(XY ; FG) is isomorphic to the pth cohomology object of the complex
(
i
H
i
(X; F)[−i]) ⊗(
j
H
j
(Y ; G)[−j]).
(See Exercise 4.14.)
134 CHAPTER 6. COHOMOLOGY OF SHEAVES
6.4 Cohomology of some classical manifolds
Here, k denotes as usual a commutative unitary ring and M denotes a k
module.
Example 6.4.1. Let X be the circle S
1
and let Z
j
’s be a closed covering
by intervals such that the Z
ij
’s are single points and Z
012
= ∅. Applying
Theorem 6.3.12, we ﬁnd that if F is a locally constant sheaf on X, the
cohomology groups H
j
(X; F) are the cohomology objects of the complex:
0 − → F
Z
0
⊕F
Z
1
⊕F
Z
2
d
− → F
Z
12
⊕F
Z
20
⊕F
Z
01
− → 0.
Recall Example 5.8.2: S
1
= U
1
∪ U
2
, U
1
∩ U
2
has two connected components
U
+
12
and U
−
12
, k is a ﬁeld, α ∈ k
and L
α
denotes the locally constant sheaf of
rank one over k obtained by gluing k
U
1
and k
U
2
by the identity on U
+
12
and
by multiplication by α ∈ k
on U
−
12
.
Then for j = 0 (resp. for j = 1), H
j
(S
1
; L
α
) is the kernel (resp. the
cokernel) of the matrix
¸
0 −1 1
1 0 −α
−1 1 0
acting on k
3
. (See Example 6.2.2.)
Note that these kernel and cokernel are zero except in case of α = 1 which
corresponds to the constant sheaf k
X
.
It follows that if M is a kmodule, then H
j
(S
1
; M
S
1) · M for j = 0, 1
and 0 otherwise.
Example 6.4.2. Consider the topological nsphere S
n
. Recall that it can
be deﬁned as follows. Let E be an Rvector space of dimension n + 1 and
denote by
˙
E the set E ` ¦0¦. Then
S
n
·
˙
E/R
+
,
where R
+
denotes the multiplicative group of positive real numbers and S
n
is
endowed with the quotient topology. (See Deﬁnition 7.4.4 below.) In other
words, S
n
is the set of all halflines in E. If one chooses an Euclidian norm
on E, then one may identify S
n
with the unit sphere in E.
We have S
n
=
¯
D
+
∪
¯
D
−
, where
¯
D
+
and
¯
D
−
denote the closed hemispheres,
and
¯
D
+
∩
¯
D
−
· S
n−1
. Let us prove that for n ≥ 1:
(6.11) H
j
(S
n
; M
S
n) =
M j = 0 or j = n,
0 otherwise.
Consider the MayerVietoris long exact sequence
− → H
j
(
¯
D
+
; M¯
D
+) ⊕H
j
(
¯
D
−
; M¯
D
−) − → H
j
(S
n−1
; M
S
n−1) (6.12)
− → H
j+1
(S
n
; M
S
n) − →
6.4. COHOMOLOGY OF SOME CLASSICAL MANIFOLDS 135
Then the result follows by induction on n since the closed hemispheres being
contractible, their cohomology is concentrated in degree 0.
Let E be a real vector space of dimension n + 1, and let X = E ` ¦0¦.
Assume E is endowed with a norm [ [. The map x → x((1 − t) + t/[x[)
deﬁnes an homotopy of X with the sphere S
n
. Hence the cohomology of a
constant sheaf with stalk M on V ` ¦0¦ is the same as the cohomology of the
sheaf M
S
n
As an application, one obtains that the dimension of a ﬁnite dimensional
vector space is a topological invariant. In other words, if V and W are two
real ﬁnite dimensional vector spaces and are topologically isomorphic, they
have the same dimension. In fact, if V has dimension n, then V ` ¦0¦ is
homotopic to S
n−1
.
Notice that S
n
is not contractible, although one can prove that any locally
constant sheaf on S
n
for n ≥ 2 is constant.
Example 6.4.3. Denote by a the antipodal map on S
n
(the map deduced
from x → −x) and denote by a
n
the action of a on H
n
(S
n
; M
S
n). Using
(6.12) and Remark 4.2.7, one deduces the commutative diagram:
H
n−1
(S
n−1
; M
S
n−1)
u
a
n−1
H
n
(S
n
; M
S
n)
a
n
H
n−1
(S
n−1
; M
S
n−1)
−u
H
n
(S
n
; M
S
n)
(6.13)
For n = 1, the map a is homotopic to the identity (in fact, it is the same as
a rotation of angle π). By (6.13), we deduce:
a
n
acting on H
n
(S
n
; M
S
n) is (−)
n+1
. (6.14)
Example 6.4.4. Let T
n
denote the ndimensional torus, T
n
· (S
1
)
n
. Using
the K¨ unneth formula, one gets that (if k is a ﬁeld)
¸
j
H
j
(T
n
; k
T
n) · (k ⊕
k[−1])
⊗n
. For example, H
j
(T
2
; k
T
2) is k for j = 0, 2, is k
2
for j = 1 and is 0
otherwise.
Let us recover this result (when n = 2) by using MayerVietoris sequences.
One may represent T
2
as follows. Consider the two cylinders Z
0
= S
1
I
0
,
Z
1
= S
1
I
1
where I
0
= I
1
= [0, 1]. Then
T
2
· (Z
0
. Z
1
)/ ∼
where ∼ is the relation which identiﬁes S
1
¦0¦ ⊂ Z
0
with S
1
¦0¦ ⊂ Z
1
and S
1
¦1¦ ⊂ Z
0
with S
1
¦1¦ ⊂ Z
1
. Then
Z
01
:= Z
0
∩ Z
1
· S
1
. S
1
.
136 CHAPTER 6. COHOMOLOGY OF SHEAVES
We have a short exact sequence of sheaves
0 − → k
T
2
α
− → k
Z
0
⊕k
Z
1
β
− → k
Z
01
− → 0
Write H
j
(X) instead of H
j
(X; k
X
) for short. Applying the functor Γ(T
2
; ),
we get the long exact sequence of Theorem 6.1.4:
0 − → H
0
(T
2
)
α
0
−→ H
0
(Z
0
) ⊕H
0
(Z
1
)
β
0
−→ H
0
(Z
01
) − → H
1
(T
2
)
α
1
−→ H
1
(Z
0
) ⊕H
1
(Z
1
)
β
1
−→ H
1
(Z
01
) − → H
2
(T
2
) − → 0.
Although we know the groups H
j
(Z
0
), H
j
(Z
1
) and H
j
(Z
01
) (since Z
0
and Z
1
are homotopic to S
1
), this sequence does not allow us to conclude, unless we
know the morphisms α or β. Now we remark that identifying Z
0
and Z
1
to
S
1
, the morphism β is given by the matrix
id
S
1 −id
S
1
−id
S
1 id
S
1
. Then one
easily recovers that H
j
(T
2
; k
T
2) is k for j = 0, 2 and k
2
for j = 1.
Exercises to Chapter 6
Exercise 6.1. Let ¦F
i
¦
i∈I
be a family of sheaves on X. Assume that this
family is locally ﬁnite, that is, each x ∈ X has an open neighborhood U such
that all but a ﬁnite number of the F
i
[
U
’s are zero.
(i)Prove that (
¸
i
F
i
)
x
·
¸
i
(F
i
)
x
.
(ii) Prove that if each F
i
is injective, then
¸
i
F
i
is injective.
(iii) Let F
i
•
be an injective resolution of F
i
. Prove that
¸
i
F
i
•
is an injective
resolution of
¸
i
F
i
.
(iv) Prove that H
j
(X;
¸
i
F
i
) ·
¸
i
H
j
(X; F
i
).
Exercise 6.2. In this exercise, we shall admit the following theorem: for
any open subset U of the complex line C, one has H
j
(U; O
C
) · 0 for j > 0.
Let ω be an open subset of R, and let U
1
⊂ U
2
be two open subsets of C
containing ω as a closed subset.
(i) Prove that the natural map O(U
2
` ω)/O(U
2
) − → O(U
1
` ω)/O(U
1
) is an
isomorphism. One denote by B(ω) this quotient.
(ii) Construct the restriction morphism to get the presheaf ω − → B(ω), and
prove that this presheaf is a sheaf (the sheaf B
R
of Sato’s hyperfunctions on
R).
(iii) Prove that the restriction morphisms B(R) − → B(ω) are surjective.
(iv) Let Ω an open subset of C and let P =
¸
m
j=1
a
j
(z)
∂
∂z
j
be a holomor
phic diﬀerential operator (the coeﬃcients are holomorphic in Ω). Recall the
Cauchy theorem which asserts that if Ω is simply connected and if a
m
(z)
Exercises to Chapter 6 137
does not vanish on Ω, then P acting on O(Ω) is surjective. Prove that if ω is
an open subset of R and if P is a holomorphic diﬀerential operator deﬁned
in a open neighborhood of ω, then P acting on B(ω) is surjective
Exercise 6.3. Let X and Y be two topological spaces, S and S
t
two closed
subsets of X and Y respectively, f : S · S
t
a topological isomorphism.
Deﬁne X .
S
Y as the quotient space X . Y/ ∼ where ∼ is the relation
which identiﬁes x ∈ X and y ∈ Y if x ∈ S, y ∈ S
t
and f(x) = y. One still
denotes by X, Y, S the images of X, Y, S . S
t
in X .
S
Y .
(i) Let F be a sheaf on X.
S
Y . Write the long exact MayerVietoris sequence
associated with X, Y, S.
(ii) Application (a). Let S
n
denote the unit sphere of the Euclidian space
R
n+1
, B the intersection of S
n
with an open ball of radius ε (0 < ε << 1)
centered in some point of S
n
, Σ its boundary in S
n
. Set X = S
n
` B,
S = Σ and let Y and S
t
be a copy of X and S, respectively. Calculate
H
j
(X .
S
Y ; k
X
S
Y
).
(iii) Application (b). Same question by replacing the sphere S
n
by the torus
T
2
embedded into R
3
.
Exercise 6.4. Let X = R
4
and consider the locally closed subset Z =
¦(x, y, z, t) ∈ R
4
; t
4
= x
2
+y
2
+z
2
; t > 0¦. Denote by f : Z → X the natural
injection. Calculate (R
j
f
∗
k
Z
)
0
for j ≥ 0.
Exercise 6.5. Let p, q, n be integers ≥ 1 with n = p + q and let X = R
n
endowed with the coordinates x = (x
1
, . . . , x
n
). Set S
0
= ¦x ∈ X;
¸
n
i=1
x
2
i
=
1¦, S
1
= ¦x ∈ X;
¸
p
i=1
x
2
i
+ 2
¸
n
i=p+1
x
2
i
= 1¦, S = S
0
∪ S
1
. Calculate
H
j
(S; k
S
) for all j.
Exercise 6.6. Let γ = ¦(x, y, z, t) ∈ X = R
4
; x
2
+ y
2
+ z
2
= t
2
¦ and let
U = X ` γ.
(i) Show that γ is contractible.
(ii) Calculate H
j
(X; k
XU
) for all j. (Recall that there exists an exact se
quence 0 − → k
XU
− → k
X
− → k
Xγ
− → 0.)
Exercise 6.7. A closed subset Z of a space X is called a retract of X if
there exists a continuous map f : X − → Z which induces the identity on Z.
Show that S
1
is not a retract of the closed disk
¯
D in R
2
.
(Hint: denote by ι : S
1
→
¯
D the embedding and assume that there exists a
continuous map f :
¯
D − → S
1
such that the composition f ◦ ι is the identity.
We get that the composition
H
1
(S
1
; Z
S
1)
f
1
−−→ H
1
(
¯
D; Z¯
D
)
ι
1
−→ H
1
(S
1
; Z
S
1)
is the identity.)
138 CHAPTER 6. COHOMOLOGY OF SHEAVES
Exercise 6.8. Consider the unit ball B
n+1
= ¦x ∈ E; [x[ ≤ 1¦ and consider
a map f : B
n+1
− → B
n+1
. Prove that f has at least one ﬁxed point. (Hint:
otherwise, construct a map g : B
n+1
− → S
n
which induces the identity on S
n
and use the same argument as in Exercise 6.7.)
(Remark: the result of this exercise is known as the Brouwer’s Theorem.)
Chapter 7
Homotopy and fundamental
groupoid
In this chapter we study locally constant sheaves of sets and sheaves of k
modules and introduce the fundamental group of locally connected topolog
ical spaces. We deﬁne the monodromy of a locally constant sheaf and prove
the equivalence between the category of representations of the fundamental
group and that of locally constant sheaves.
1
Some references: [11], [7], [13], [3], [22]. In this chapter, we shall admit
some results treated with all details in [11].
7.1 Fundamental groupoid
Let us recall some classical notions of topology. We denote as usual by I the
closed interval [0, 1] and by S
1
the circle. Note that topologically S
1
· I/ ∼
where ∼ is the equivalence relation on I which identiﬁes the two points 0
and 1. We shall also consider the space
D := I I/ ∼
where ∼ is the equivalence relation which identiﬁes I ¦0¦ to a single point
(denoted a
0
) and I ¦1¦ to a single point (denoted a
1
). Note that topolog
ically, D is isomorphic to the closed unit disk, or else, to I I.
Let X denote a topological space.
Deﬁnition 7.1.1. (i) A path from x
0
to x
1
in X is a continuous map
σ : I − → X, with σ(0) = x
0
and σ(1) = x
1
. The two points x
0
and x
1
are called the ends of the path.
1
This chapter will not be treated during the course 2005/2006
139
140 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
(ii) Two paths σ
0
and σ
1
are called homotopic if there exists a continuous
function ϕ : I I − → X such that ϕ(i, t) = σ
i
(t) for i = 0, 1.
(iii) If the two paths have the same ends, x
0
and x
1
, one says they are
homotopic with ﬁxed ends if moreover ϕ(s, 0) = x
0
, ϕ(s, 1) = x
1
for all
s. This is equivalent to saying that there exists a continuous function
ψ : D − → X such that ψ(i, t) = σ
i
(t) for i = 0, 1.
(iv) A loop in X is continuous map γ : S
1
− → X. One can also consider a
loop as a path γ such that γ(0) = γ(1). A trivial loop is a constant
map γ : S
1
− → ¦x
0
¦. Two loops are homotopic if they are homotopic as
paths.
It is left to the reader to check that “homotopy” is an equivalence relation.
If σ is a path from x
0
to x
1
and τ a path from x
1
to x
2
one can deﬁne
a new path τσ (in this order) from x
0
to x
2
by setting τσ(t) = σ(2t) for
0 ≤ t ≤ 1/2 and τσ(t) = τ(2t −1) for 1/2 ≤ t ≤ 1.
If σ is a path from x
0
to x
1
, one can deﬁne the path σ
−1
from x
1
to x
0
by setting σ
−1
(t) = σ(1 −t).
Let us denote by [σ] the homotopy class of a path σ. It is easily checked
that the homotopy class of τσ depends only on the homotopy classes of σ and
τ. Hence, we can deﬁne [τ][σ] as [τσ]. The next result is left as an exercise.
Lemma 7.1.2. The product [σ][τ] is associative, and [σσ
−1
] is the homotopy
class of the trivial loop at x
0
.
By this lemma, the set of homotopy classes of loops at x
0
is a group.
Deﬁnition 7.1.3. The set of homotopy classes of loops at x
0
endowed with
the above product is called the fundamental group of X at x
0
and denoted
π
1
(X; x
0
).
Deﬁnition 7.1.4. Let X be a topological space.
(i) X is arcwise connected (or “path connected”) if given x
0
and x
1
in X,
there exists a path with ends x
0
and x
1
,
(ii) X is simply connected if any loop in X is homotopic to a trivial loop,
(iii) X is locally connected (resp. locally arcwise connected, resp. locally
simply connected) if each x ∈ X has a neighborhood system consisting
of connected (resp. arcwise connected, resp. simply connected) open
subsets.
7.1. FUNDAMENTAL GROUPOID 141
Clearly, if X is arcwise connected, it is connected. If X is locally arcwise
connected and connected, it is arcwise connected.
Example 7.1.5. In R
2
denote by X the union of the graph of the function
y = sin(1/x), x > 0, the interval ¦(x, y); x = 0, −1 ≤ y ≤ 1¦ and the interval
¦(x, y); y = 0, x ≥ 0. Then X is arcwise connected but not locally arcwise
connected.
In the sequel, we shall make the hypothesis
X is locally arcwise connected. (7.1)
Assume (7.1). If σ is a path from x
0
to x
1
in X, then the map γ → σ
−1
γσ
deﬁnes an isomorphism
π
1
(X; x
0
) · π
1
(X; x
1
).
Hence, if X is connected, all groups π
1
(X; x) are isomorphic for x ∈ X.
Deﬁnition 7.1.6. The fundamental groupoid Π
1
(X) is the category given
by
Ob(Π
1
(X)) = X,
Hom
Π
1
(X)
(x
0
, x
1
)=¦the set of homotopy classes of paths from
x
0
to x
1
¦.
Note that for x ∈ X, Hom
Π
1
(X)
(x, x) = π
1
(X, x).
Consider a continuous map f : X − → Y . If γ is a path in X, then f ◦ γ is
a path in Y , and if two paths γ
0
and γ
1
are homotopic in X, then f ◦ γ
0
and
f ◦ γ
1
are are homotopic in Y . Hence, we get a functor:
f
∗
: Π
1
(X) − → Π
1
(Y ). (7.2)
In particular, if i
U
: U → X denotes the embedding of an open subset U of
X, we get the functor
i
U∗
: Π
1
(U) − → Π
1
(X). (7.3)
Proposition 7.1.7. Let f
0
, f
1
: X − → Y be two continuous maps and assume
f
0
and f
1
are homotopic. Then the two functors f
0
∗
and f
1
∗
are isomorphic.
In particular, if f : X − → Y is a homotopy equivalence, then the two
groupoids Π
1
(Y ) and Π
1
(X) are equivalent.
Proof. Let h : I X − → Y be a continuous map such that h(i, ) = f
i
(),
i = 0, 1, and let γ : I − → X be a path. Then h ◦ γ : I I − → Y deﬁnes a
homotopy between f
0
◦ γ and f
1
◦ γ. q.e.d.
142 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
Examples 7.1.8. (i) A contractible space is simply connected. This follows
from Proposition 7.1.7.
(ii) One has π
1
(S
1
) · Z and 1 ∈ Z corresponds to the identity map, con
sidered as a loop in the space X = S
1
. We refer to [11] a for proof. As a
corollary, one gets π
1
(R
2
` ¦0¦) · Z.
Remark 7.1.9. (i) Assume (7.1) and moreover X is non empty and con
nected. Then all objects x ∈ Π
1
(X) are isomorphic and after choosing
x
0
∈ X, one sets π
1
(X) = π
1
(X; x
0
). One calls π
1
(X) the fundamental
group of X.
(ii) Remark that X being arcwise connected, it is simply connected if and
only if π
1
(X) · ¦1¦.
(iii) It is easily seen that if X is simply connected, two paths γ and τ with
the same ends are homotopic.
Let X and Y be two topological spaces satisfying (7.1). Denote by p
i
the
projection from X Y to X and Y respectively. These projections deﬁne
functors p
1
∗
: Π
1
(X Y ) − → Π
1
(X) and p
2
∗
: Π
1
(X Y ) − → Π
1
(Y ), hence a
functor
(p
1∗
p
2∗
)Π
1
(X Y ) − → Π
1
(X) Π
1
(Y ). (7.4)
Proposition 7.1.10. The functor in (7.4) is an equivalence.
Proof. (i) The functor in (7.4) is obviously essentially surjective.
(ii) Let x
0
, x
1
∈ X, y
0
, y
1
∈ Y . Let us show the isomorphism
Hom
Π
1
(XY )
((x
0
, y
0
), (x
1
, y
1
))
∼
−→Hom
Π
1
(X)
(x
0
, x
1
) Hom
Π
1
(Y )
(y
0
, y
1
) (7.5)
The map in (7.5) is surjective. Indeed, if σ is a path in X and τ is a path in
Y , the path σ τ in X Y satisﬁes
(p
1
∗
p
2
∗
)[σ τ] = [σ] [τ].
(iii) The map in (7.5) is injective. Indeed, let γ
0
and γ
1
be two paths in
X Y . Assume that p
1∗
γ
0
is homotopic to p
1∗
γ
1
and p
2∗
γ
0
is homotopic to
p
2
∗
γ
1
. The product of these two homotopies deﬁnes an homotopy from γ
0
to
γ
1
. q.e.d.
7.2 Monodromy of locally constant sheaves
Locally constant sheaves
Remark 7.2.1. We shall work here with sheaves of kmodules, but many
results remain true without any change for sheaves of sets.
7.2. MONODROMY OF LOCALLY CONSTANT SHEAVES 143
Let M ∈ Mod(k). Recall that a constant sheaf F with stalk M on X is
a sheaf isomorphic to the sheaf M
X
of locally constant functions with values
in M. We shall denote by LCSH(k
X
) (resp. CSH(k
X
)) the full additive
subcategory of Mod(k
X
) consisting of locally constant (resp. of constant)
sheaves.
Denote as usual by a
X
the map X − → pt.
Proposition 7.2.2. (i) Assume X is connected and non empty. Then the
two functors
CSH(k
X
)
a
X∗
Mod(k)
a
−1
X
are equivalences of categories, inverse one to each other. In particular,
the category CSH(k
X
) is abelian and if M and N are two kmodules,
there is an isomorphism Hom
k
X
(M
X
, N
X
) · (Hom
k
(M, N))
X
.
(ii) Assume X is locally connected. Then the category LCSH(k
X
) is abelian.
Proof. (i) If F is a constant sheaf on X, and if one sets M = Γ(X; F), then
F · M
X
. Therefore, if M ∈ Mod(k), then a
X∗
a
−1
X
M · M and a
−1
X
a
X∗
M
X
·
M
X
.
(ii) Let ϕ : F − → G is a morphism of locally constant sheaf, and let x ∈ X. If
U is suﬃciently small connected open neighborhood of x, the restriction to U
of Ker ϕ and Coker ϕ will be constant sheaves, by (i). Hence, LCSH(k
X
) is a
full additive subcategory of an abelian category (namely Mod(k
X
)) admitting
kernels and cokernels. This implies it is abelian. q.e.d.
Lemma 7.2.3. Let F be a locally constant sheaf on X = I I. Then F is
a constant sheaf.
Proof. Since I I is compact, there exists ﬁnite coverings of I by intervals
¦U
i
¦
1≤i≤N
0
and ¦V
j
¦
1≤j≤N
1
such that F[
U
i
V
j
is a constant sheaf. Since (U
i
V
j
) ∩(U
i+1
V
j
) is connected, the argument of the proof of Proposition 5.7.6
shows that F[
IV
j
is a constant sheaf for all j. Since (I V
j
) ∩ (I V
j+1
)
is connected, the argument of the proof of Proposition 5.7.6 shows that F is
constant. q.e.d.
Representations
For a group G and a ring k one deﬁnes the category Rep(G, Mod(k)) of
representations of G in Mod(k) as follows. An object is a pair (M, µ
M
) with
M ∈ Mod(k) and µ
M
∈ Hom(G, Gl(M)). A morphism µ
f
: (M, µ
M
) − →
(N, µ
N
) is a klinear map f : M − → N which satisﬁes µ
N
◦ f = f ◦ µ
M
(i.e.
144 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
µ
N
(g) ◦ f = f ◦ µ
M
(g) for any g ∈ G). Note that Rep(G, Mod(k)) contains
the full abelian subcategory Mod(k), identiﬁed to the trivial representations
of G.
We identify G with a category ( with one object c, the morphisms being
given by Hom
C
(c, c)= G. One gets that
Rep(G, Mod(k)) · Fct((, Mod(k)). (7.6)
Now we shall consider the category Fct(Π
1
(X), Mod(k)), a generalization of
the category of representations Rep(π
1
(X), Mod(k)). In fact, if X is con
nected, non empty and satisﬁes (7.7), then all x ∈ Π
1
(X) are isomorphic,
and the groupoids Π
1
(X) is equivalent to the group π
1
(X, x
0
) identiﬁed to
the category with one object x
0
and morphisms π
1
(X, x
0
). In this case, the
two categories Rep(π
1
(X), Mod(k)) and Fct(Π
1
(X), Mod(k)) are equivalent.
Monodromy
In this section, we make the hypothesis (7.1) that we recall
X is locally arcwise connected. (7.7)
Deﬁnition 7.2.4. (i) One calls an object of Fct(Π
1
(X), Mod(k)) a repre
sentation of the groupoid Π
1
(X) in Mod(k).
(ii) Let θ ∈ Fct(Π
1
(X), Mod(k)). One says that θ is a trivial representation
if θ is isomorphic to a constant functor ∆
M
which associates the module
M to any x ∈ X, and id
M
to any [γ] ∈ Hom
Π
1
(X)
(x, y).
Let Fct
0
(Π
1
(X), Mod(k)) be the full subcategory of Fct(Π
1
(X), Mod(k))
consisting of trivial representations. Then the functor M → ∆
M
from
Mod(k) to Fct
0
(Π
1
(X), Mod(k)) is an equivalence of categories.
Let F be a locally constant sheaf of kmodules on X. Let γ be a path
from x
0
to x
1
. We shall construct an isomorphism
µ(F)(γ) : F
x
0
· F
x
1
.
Since γ
−1
F is a locally constant sheaf on [0, 1], it is a constant sheaf. We get
the isomorphisms, which deﬁne µ(F)(γ):
F
x
0
· (γ
−1
F)
0
∼
←−Γ(I; γ
−1
F)
∼
−→(γ
−1
F)
1
· F
x
1
. (7.8)
Example 7.2.5. (i) Let X = I, denote by t a coordinate on I, and con
sider the constant sheaf F = C
X
exp(αt). Then µ(F)(I): F
0
∼
−→F
1
is the
multiplication by exp(α).
7.2. MONODROMY OF LOCALLY CONSTANT SHEAVES 145
(ii) Let X = S
1
· [0, 2π]/ ∼ (where ∼ identiﬁes 0 and 2π) and denote
by theta a coordinate on S
1
. Consider the constant locally constant sheaf
F = C
X
exp(iβθ). (See Example 5.8.2.) Let γ be the identity loop. Then
µ(F)(γ) : F
x
0
∼
−→F
x
0
is the multiplication by exp(2iπβ).
Lemma 7.2.6. The isomorphism µ(F)(γ) depends only on the homotopy
class of γ in X.
Proof. Let ϕ be a continuous function D − → X such that ϕ(i, t) = γ
i
(t),
i = 0, 1. The sheaf ϕ
−1
F is constant by Lemma 7.2.3. The isomorphisms
µ(F)(γ
i
) (i = 0, 1) are described by the commutative diagram:
F
x
0
· (ϕ
−1
F)
a
0
∼
Γ(D; ϕ
−1
F)
∼
∼
∼
(ϕ
−1
F)
a
1
· F
x
1
∼
F
x
0
· (γ
−1
i
F)
0
Γ(I; γ
−1
i
F)
∼
∼
(γ
−1
i
F)
1
· F
x
1
(7.9)
This shows that µ(F)(γ
0
) = µ(F)(γ
1
). q.e.d.
If τ is another path from x
1
to x
2
, then:
µ(F)(γτ) = µ(F)(γ) ◦ µ(F)(τ).
Hence we have constructed a functor of µ(F): Π
1
(X) − → Mod(k) given by
µ(F)(x) = F
x
, µ(F)([γ]) = µ(F)(γ) where γ is a representative of [γ]. This
correspondence being functorial in F, we get a functor
µ : LCSH(k
X
) − → Fct(Π
1
(X), Mod(k)). (7.10)
Deﬁnition 7.2.7. The functor µ in (7.10) is called the monodromy functor.
The functor µ is also “functorial” with respect to the space X. More
precisely, let f : X − → Y be a continuous map, and assume that both X and Y
are locally arcwise connected. We have the commutative (up to isomorphism)
diagram of categories and functors:
LCSH(k
Y
)
µ
f
−1
LCSH(k
X
)
µ
Fct(Π
1
(Y ), Mod(k))
f
−1
Fct(Π
1
(X), Mod(k))
(7.11)
Theorem 7.2.8. Assume (7.7). The functor µ in (7.10) is fully faithful.
146 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
Proof. (i) µ is faithful. Let ϕ, ψ : F − → G be morphisms of locally constant
sheaves and assume that µ(ϕ) · µ(ψ). This implies that ϕ
x
0
· ψ
x
0
: F
x
0
− →
G
x
0
for any x
0
∈ X, and the sheaf Hom(F, G) being locally constant, this
implies that ϕ = ψ on the connected component of x
0
.
(ii) µ is full. Consider a morphism u : µ(F) − → µ(G). It deﬁnes a morphism
ϕ
x
0
: F
x
0
− → G
x
0
. To each x
1
∈ X we get a well deﬁned morphism ϕ
x
1
:
F
x
1
− → G
x
1
, given by ϕ
x
1
= µ(G)(σ) ◦ ϕ
x
0
◦ µ(F)(σ
−1
), where σ is any path
from x
0
to x
1
. This isomorphism does not depend on the choice of σ by
the hypothesis. Since F and G are locally constant, each x ∈ X has an
open neighborhood U
x
such that the morphism ϕ
x
: F
x
− → G
x
extends as a
morphism ϕ
Ux
: F[
Ux
− → G[
Ux
. These morphisms will glue to each other and
deﬁne a morphism ϕ : F − → G with µ(ϕ) = u. q.e.d.
Proposition 7.2.9. Assume (7.7) and X is non empty and connected. Then
the functor µ in (7.10) induces an equivalence
µ
0
: CSH(k
X
)
∼
−→Fct
0
(Π
1
(X), Mod(k)) · Mod(k). (7.12)
Proof. (i) First, let us show that µ
0
takes its values in Fct
0
(Π
1
(X), Mod(k)).
Let F be a constant sheaf. We may assume that F = M
X
, for a module M,
and we shall show that µ
0
(F) is the constant functor x → M.
Let σ : I − → X be a path with x
0
= σ(0) = σ(1) = x
1
. It follows from
Lemma 7.2.3 (iii) that the morphism µ(F)(γ): M · F
x
0
∼
−→F
x
1
· M is the
identity.
(ii) By Theorem 7.2.8, µ
0
is fully faithful.
(iii) Since any M ∈ Mod(k) deﬁnes a constant sheaf, µ
0
is essentially surjec
tive. q.e.d.
Corollary 7.2.10. Assume that X is connected, locally arcwise connected
and simply connected. Then any locally constant sheaf F on X is a constant
sheaf.
Proof. By the hypothesis, Fct(Π
1
(X), Mod(k)) · Mod(k). Let F be a locally
constant sheaf. Then µ(F) ∈ Mod(k) and there exists G ∈ CSH(k
X
) such
that µ(F) · µ(G). Since µ is fully faithful, this implies F · G. q.e.d.
7.3 The Van Kampen theorem
In this section, we shall assume (7.7) and also
there exists an open covering stable by ﬁnite intersec
tions by connected and simply connected subsets.
(7.13)
7.3. THE VAN KAMPEN THEOREM 147
Theorem 7.3.1. Assume (7.7) and (7.13). The functor µ in (7.10) is an
equivalence of categories.
Proof. (i) By Proposition 7.2.8, it remains to show that µ is essentially sur
jective.
(ii) By Corollary 7.2.10, the theorem is true if X is connected and simply
connected.
(iii) Let  = ¦U
i
¦
i∈I
be an open covering of X as in (7.13). The func
tors i
U
i ∗
: Π
1
(U
i
) − → Π
1
(X) deﬁne functors λ
i
from Fct(Π
1
(X), Mod(k)) to
Fct(Π
1
(U
i
), Mod(k)). Let G ∈ Fct(Π
1
(X), Mod(k)) and set G
i
= λ
i
(G).
Using the result in (ii) for U
i
, we ﬁnd sheaves F
i
such that µ(F
i
) = G
i
. Set
ting λ
ji
= λ
j
◦ λ
−1
i
, and using the result in (ii) for U
ij
, we get isomorphisms
θ
ji
: F
i
[
U
ij
∼
−→F
j
[
U
ij
, and the cocycle condition (5.26) will be clearly satis
ﬁed. Applying Theorem 5.8.1, we ﬁnd a sheaf F on X, which will be locally
isomorphic to the F
i
’s, hence, will be locally constant.
It remains to show that µ(F) · G. For any U
i
as above, λ
i
(µ(F)) · G
i
and λ
i
(G) are isomorphic in Fct(Π
1
(U
i
), Mod(k)). Hence, the result follows
from Lemma 7.3.2 below. q.e.d.
Lemma 7.3.2. Let  = ¦U
i
¦
i∈I
be an open covering of X. Consider the
functor
λ =
¸
i∈I
λ
i
: Fct(Π
1
(X), Mod(k)) − →
¸
i∈I
Fct(Π
1
(U
i
), Mod(k)).
If G
1
, G
2
∈ Fct(Π
1
(X), Mod(k)) satisfy λ(G
1
) · λ(G
2
), then G
1
· G
2
.
Proof. (i) For each x ∈ X, G
1
(x) · G
2
(x), since x ∈ U
i
for some i.
(ii) Let [γ] ∈ Hom
Π
1
(X)
(x, y) and let γ be a path which represents [γ]. Assume
γ is contained in some U
i
, and denote by [γ
t
] the corresponding element in
Hom
Π
1
(U
i
)
(x, y). Then for any G ∈ Fct(Π
1
(X), Mod(k)), one has G([γ]) =
λ
i
(G)([γ
t
]). Hence, G
1
([γ]) = G
2
([γ]) in this case.
(iii) We may decompose γ as γ = γ
1
γ
n
, each γ
j
(1 ≤ j ≤ n) being
contained in some U
i
j
. By the hypothesis, for 1 ≤ j ≤ n, there exist isomor
phisms G
1
([γ
j
]) · G
2
([γ
j
]). Since G
ν
([γ]) = G
ν
([γ
n
]) G
ν
([γ
1
]) for ν = 1, 2,
the result follows. q.e.d.
Corollary 7.3.3. Assume (7.7)and (7.13) and X is connected. Then X is
simply connected if and only if any locally constant sheaf on X is constant.
Proof. By Theorem 7.3.1 and Proposition 7.2.9, any representation of π
1
(X)
is trivial if and only if any locally constant sheaf is constant. It remains to
notice that if G is a group such that any representation of G is trivial, then
G = ¦1¦. (See Exercise 7.3.) q.e.d.
148 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
Example 7.3.4. A locally constant sheaf of Cvector spaces of ﬁnite rank
on X is called a local system. Hence, it is now possible to classify all
local systems on the space X = R
2
` ¦0¦. In fact, π
1
(X) · Z, hence,
Hom(π
1
(X), Gl(C
n
)) = Gl(C
n
). A local system F of rank n is determined,
up to isomorphism, by its monodromy µ(F) ∈ Gl(C
n
). The classiﬁcation of
such sheaves is thus equivalent to that of invertible n n matrices over C
up to conjugation, a well known theory (JordanH¨ older decomposition). In
particular, when n = 1, Gl(C) = C
.
Hence, a local system of rank one is determined, up to isomorphism, by
its monodromy α ∈ C
.
We shall deduce a particular case of the Van Kampen theorem.
Theorem 7.3.5. Let X be a connected, locally arcwise connected, and locally
simply connected space. Let X = ¦U
i
¦
i∈I
be an open covering stable by
ﬁnite intersection, the U
i
’s being connected. Assume that there exists x which
belongs to all U
i
’s and identify each groupoids Π
1
(U
i
) with the group π
1
(U
i
) =
π
1
(U
i
, x). Then π
1
(X) · lim
−→
i∈I
π
1
(U
i
).
Sketch of proof. There is a natural morphism of groups
lim
−→
i
π
1
(U
i
) − → π
1
(X). (7.14)
Let M ∈ Mod(k). One has
Hom(π
1
(X), Gl(M) · lim
←−
i
Hom(π
1
(U
i
), Gl(M))
· Hom(lim
−→
i
π
1
(U
i
), Gl(M)),
where the ﬁrst isomorphism follows from Theorem 7.3.1. To conclude, remark
that if u: G
1
− → G
2
is a morphism of groups which induces an isomorphism
Hom(G
2
, Gl(M))
∼
−→Hom(G
1
, Gl(M)) for all M ∈ Mod(Z), then u is an
isomorphism (see Exercise 7.3). q.e.d.
7.4 Coverings
Let S be a set. We endow it with the discrete topology. Then X S ·
¸
s∈S
X
s
where X
s
= X ¦s¦ is a copy of X, and the coproduct is taken in
the category of topological spaces. In particular each X
s
is open.
7.4. COVERINGS 149
Deﬁnition 7.4.1. (i) A continuous map f : Z − → X is a trivial covering if
there exists a non empty set S, a topological isomorphism h : Z
∼
−→X
S where S is endowed with the discrete topology, and f = p ◦ h where
p : X S − → X is the projection.
(ii) A continuous map f : Z − → X is a covering
2
if f is surjective and any
x ∈ X has an open neighborhood U such that f[
f
−1
(U)
: f
−1
(U) − → U
is a trivial covering.
(iii) If f : Z − → X is a covering, a section u of f is a continuous map
u : X − → Z such that f ◦ u = id
X
. A local section is a section deﬁned
on an open subset U of X.
(iv) A morphism of coverings f : Z − → X to f
t
: Z
t
− → X is a continuous
map h : Z − → Z
t
such that f = f
t
◦ h.
Hence, we have deﬁned the category Cov(X) of coverings above X, and
the full subcategory of trivial coverings. Roughly speaking, a covering is
locally isomorphic to a trivial covering.
Notation 7.4.2. Let f : Z − → X be a covering. One denotes by Aut (f) the
group of automorphisms of this covering, that is, the group of isomorphisms
of the object (f : Z − → X) ∈ Cov(X).
The deﬁnition of a covering is visualized as follows.
¸
s∈S
U
s
p
f
−1
U
h
∼
f
Z
f
U
X
If X is connected and S is ﬁnite for some x, S will be ﬁnite for all x,
with the same cardinal, say n. In this case one says that f is a ﬁnite (or an
n)covering.
Example 7.4.3. Let Z = C` ¦0, ±i, ±i
√
2¦, X = C` ¦0, 1¦ and let f : Z − →
X be the map z → (z
2
+ 1)
2
. Then f is a 4covering.
Many coverings appear naturally as the quotient of a topological space
by a discrete group.
Deﬁnition 7.4.4. Let X be a locally compact topological space and let G
be a group, that we endow with the discrete topology. We denote by e the
unit in G.
2
“revˆetement” in French, not to be confused with “recouvrement”.
150 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
(i) An action µ of G on X is a map µ : GX − → X such that:
(a) for each g ∈ G, µ(g) : X − → X is continuous,
(b) µ(e) = id
X
,
(c) µ(g
1
◦ g
2
) = µ(g
1
) ◦ µ(g
2
).
In the sequel, we shall often write g x instead of µ(g)(x), for g ∈ G
and x ∈ X.
(ii) Let x ∈ X. The orbit of x in X is the subset G x of X.
(iii) One says that G acts transitively on X if for any x ∈ X, X = G x.
(iv) One says that G acts properly on X if for any compact subset K of X,
the set G
K
= ¦g ∈ G; g K ∩ K = ∅¦ is ﬁnite.
(v) One says that G acts freely if for any x ∈ X, the group G
x
= ¦g ∈
G; g x = x¦ is trivial, that is, is reduced to ¦e¦.
If a group G acts on X, it deﬁnes an equivalence relation on X, namely,
x ∼ y if and only if there exists g ∈ G with x = g y. One denotes by X/G
the quotient space, endowed with the quotient topology.
Theorem 7.4.5. Assume that a discrete group G acts properly and freely on
a locally compact space X. Then p : X − → X/G is a covering. Moreover, if
X is connected, then Aut (p) = G.
For the proof, we refer to [11].
Examples 7.4.6. (i) The map p : R − → R/Z (where Z acts on R by
translation) is a covering, and the map t → exp(2iπt) : R − → S
1
induces
an isomorphism h : R/Z
∼
−→S
1
such that exp(2iπt) = h ◦ p. Therefore,
t → exp(2iπt) : R − → S
1
is a covering.
Similarly, the map p : C − → C/2iπZ is a covering and the map z →
exp(z) : C − → C ` ¦0¦ induces an isomorphism h : C/2iπZ
∼
−→C ` ¦0¦ such
that exp(z) = h ◦ p. Therefore, z → exp(z) : C − → C ` ¦0¦ is a covering.
(ii) Consider the group H
n
of nroots of unity in C, that is, the subgroup of
C
generated by exp(2iπ/n). Then p : S
1
− → S
1
/H
n
is a covering and the
map z → z
n
: S
1
− → S
1
induces an isomorphism h : S
1
/H
n
∼
−→S
1
such that
z
n
= h ◦ p. Therefore, z → z
n
: S
1
− → S
1
is an ncovering.
(iii) The projection R
n
− → R
n
/Z
n
is a covering, and there is are isomorphisms
R
n
/Z
n
· (R/Z)
n
· (S
1
)
n
.
7.4. COVERINGS 151
Example 7.4.7. The projective space of dimension n, denoted P
n
(R), is
constructed as follows. Let E be an n +1dimensional Rvector space. Then
P
n
(R) is the set of lines in E, in other words,
P
n
(R) =
˙
E/R
, (7.15)
where
˙
E = E ` ¦0¦, and R
is the multiplicative group of nonzero elements
of R.
Identifying E with R
n+1
, a point x ∈ R
n+1
is written x = (x
0
, x
1
, , x
n
)
and x ∈ P
n
(R) may be written as x = [x
0
, x
1
, , x
n
], with the rela
tion [x
0
, x
1
, , x
n
] = [λx
0
, λx
1
, , λx
n
] for any λ ∈ R
. One says that
[x
0
, x
1
, , x
n
] are homogeneous coordinates.
The map R
n
− → P
n
(R) given by (y
1
, , y
n
) → [1, y
1
, , y
n
] allows
us to identify R
n
to the open subset of P
n
(R) consisting of the set ¦x =
[x
0
, x
1
, , x
n
]; x
0
= 0¦.
In the sequel, we shall often write for short P
n
instead of P
n
(R). Since
S
n
=
˙
E/R
+
, we get P
n
· S
n
/a where a is the “antipodal” relation on S
n
which identiﬁes x and −x. The map a deﬁnes an action of the group Z/2Z
on S
n
, and this action is clearly proper and free. Denote by
γ : S
n
− → P
n
(7.16)
the natural map. This is a 2covering and P
n
· S
n
/(Z/2Z).
Coverings and locally constant sheaves
We make hypothesis (7.1), that is, X is locally arcwise connected.
Let f : Z − → X be a covering. We associates a sheaf of sets F
f
on X as
follows. For U open in X, F
f
(U) is the set of sections of f[
U
: f
−1
(U) − → U.
Since, locally, f is isomorphic to the projection U S − → U, the sheaf F is
locally isomorphic the the constant sheaf with values in S. We have thus
constructed a functor
Φ: Cov(X) − → LCSH(X). (7.17)
Proposition 7.4.8. Assume (7.1). Then the functor Φ in (7.17) is an equiv
alence.
Sketch of proof of the proof. We shall construct a quasiinverse Ψ to Φ. Let
F ∈ LCSH(X). Consider the set Z
F
=
¸
x∈X
F
x
. We endow Z
F
with the
following topology. A basis of open subsets for this topology is given by the
sets U M such that U is open in X, F[
U
· M
U
is a constant sheaf with
stalk M (hence, F
x
· M) and M is endowed with the discrete topology.
q.e.d.
152 CHAPTER 7. HOMOTOPY AND FUNDAMENTAL GROUPOID
Note that the functor Φ induces an equivalence between trivial coverings
and constant sheaves.
Note that Theorem 7.3.1 is also true when replacing sheaves of kmodules
with sheaves of sets. Hence, assuming (7.7) and (7.13) we get the equivalences
Cov(X) · LCSH(X) · Fct(Π
1
(X), Set).
Exercises to Chapter 7
Exercise 7.1. Classify all locally constant sheaves of Cvector spaces on the
space X = S
1
S
1
.
Exercise 7.2. Let γ = ¦(x, y, z, t) ∈ X = R
4
; x
2
+y
2
+z
2
= t
2
¦ ˙ γ = γ ` ¦0¦.
Classify all locally constant sheaves of rank one of Cvector spaces on ˙ γ.
(Hint: one can use the fact that ˙ γ is homotopic to its intersection with the
unit sphere S
3
of R
4
.)
Exercise 7.3. (i) Let G be a group and assume that all representation of G
in Mod(k) are trivial. Prove that G = ¦1¦.
(ii) Let u: G
1
− → G
2
is a morphism of groups which induces an isomorphism
Hom(G
2
, Gl(M))
∼
−→Hom(G
1
, Gl(M)) for all M ∈ Mod(Z). Prove that u is
an isomorphism.
(Hint: (i) use the free kmodule k[G] generated over k by the element g ∈ G.)
Exercise 7.4. Assume X satisﬁes (7.1), X = U
1
∪ U
2
, U
1
and U
2
are con
nected and simply connected and U
1
∩ U
2
is connected.
(i) Prove that X is simply connected.
(ii) Deduce that for n > 1, the sphere S
n
as well as R
n+1
` ¦0¦ are simply
connected.
Bibliography
[1] M. Atiyah and I.G. Macdonald, Introduction to commutative algebra,
AddisonWeisley (1969)
[2] M. Berger and B. Gostiaux, Geom´etrie diﬀ´erentielle, Armand Colin Ed.
(1972)
[3] R. Bott and L.W. Tu, Diﬀerential forms in algebraic topology, Graduate
Texts in Math. 82, Springer (1982)
[4] N. Bourbaki, Elements de Mathematiques, Alg`ebre Ch 10, Masson (1980)
[5] H. Cartan and S. Eilenberg, Homological algebra, Princeton University
Press (1956)
[6] C. Chevalley, The construction and study of certain important algebras,
Publ. Soc. Math. Japan (1955)
[7] P. Deligne, Equations diﬀ´erentielles ` a points singuliers r´eguliers, Lecture
Notes in Math. 163, Springer Paris (1970)
[8] G. De Rham, Vari´et´es diﬀ´erentiables, Hermann, Paris (1960)
[9] JP. Freyd, Abelian categories, Harper & Row (1964)
[10] S.I. Gelfand and Yu.I. Manin, Methods of homological algebra, Springer
(1996)
[11] C. Godbillon, El´ements de topologie alg´ebrique, Hermann (1971)
[12] R. Godement, Topologie alg´ebrique et th´eorie des faisceaux, Hermann
(1958)
[13] M. Greenberg, Lectures on algebric topology, Benjamin (1967)
[14] A. Grothendieck, Sur quelques points d’alg`ebre homologique, Tohoku
Math. Journ. 119183 (1957)
153
154 BIBLIOGRAPHY
[15] A. Grothendieck, Elements de g´eom´etrie alg´ebrique III, Publ. IHES 11
(1961), 17 (1963)
[16] L. H¨ ormander, An introduction to complex analysis, Van Norstrand
(1966)
[17] B. Iversen, Cohomology of sheaves, Springer (1987)
[18] M. Kashiwara and P. Schapira, Sheaves on manifolds, Grundlehren der
Math. Wiss. 292 SpringerVerlag (1990)
[19] M. Kashiwara and P. Schapira, Categories and sheaves, SpringerVerlag,
to appear
[20] JP. Lafon, Les formalismes fondamentaux de l’alg`ebre commutative,
Hermann
[21] S. MacLane, Categories for the working mathematician, Graduate Texts
in Math. 5 Springer 2nd ed. (1998)
[22] J.P. May, A concise course in algebraic toplogy, Chicago Lectures in
Mathematics, The University of Chicago Press (1999)
[23] P. Schapira, Categories and Homological Algebra, unpub
lished course at Paris VI University, (Ps), (Dvi), (Pdf)
http://www.math.jussieu.fr/˜ schapira/publications/
[24] P. Schapira, An introduction to Sheaves, unpub
lished course at Paris VI University, (Ps), (Dvi), (Pdf)
http://www.math.jussieu.fr/˜ schapira/publications/
[25] C. Weibel, An introduction to homological algebra, Cambridge Studies
in Advanced Math. 38 (1994)
Institut de Math´ematiques, Analyse Alg´ebrique
Universit´e Pierre et Marie Curie, Case 82
4, place Jussieu F75252, Paris Cedex 05, France
email: schapira@math.jussieu.fr
Homepage: www.math.jussieu.fr/~schapira
2
Contents
1 Linear algebra over a ring 1.1 Modules and linear maps 1.2 Complexes . . . . . . . . 1.3 The functor Hom . . . . 1.4 Tensor product . . . . . 1.5 Limits . . . . . . . . . . 1.6 Koszul complexes . . . . Exercises . . . . . . . . . . . . 2 The language of categories 2.1 Categories and functors . 2.2 The Yoneda Lemma . . 2.3 Adjoint functors . . . . . 2.4 Limits . . . . . . . . . . 2.5 Examples . . . . . . . . 2.6 Exact functors . . . . . . 2.7 Filtrant inductive limits Exercises . . . . . . . . . . . . 7 7 10 13 20 24 31 34 37 37 42 44 45 48 51 53 56 59 59 61 64 66 68 71 71 75 79 81
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
3 Additive categories 3.1 Additive categories . . . . . . . . 3.2 Complexes in additive categories . 3.3 Simplicial constructions . . . . . 3.4 Double complexes . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . 4 Abelian categories 4.1 Abelian categories . . . . . . . . . 4.2 Complexes in abelian categories . 4.3 Application to Koszul complexes 4.4 Injective objects . . . . . . . . . . 3
5. . . . . 5. . . . . . . . . 6. .4 4. . . . . . . . . . . . . . . . . . . .5 Resolutions . . . . . .4 Internal operations . . . . . . . . . . . . . . . . . . . . . . . .6 Sheaves associated with a locally closed subset 5. . . . . . . . . . . . . . . . . . .3 Invariance by homotopy . . . . . . . . .8 Gluing sheaves . . . . . . . . . 123 . . . . . . . CONTENTS . . . 7 Homotopy and fundamental groupoid 7. . . . . . . . . . .2 Sheaves . . . . .2 Monodromy of locally constant sheaves 7. .1 Cohomology of sheaves . . . . . . . . . . Exercises .1 Presheaves . . . . . . 126 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Derived functors 4. . . . . . . . 82 84 88 91 95 95 97 100 104 107 112 115 118 120 5 Abelian sheaves 5. . . . . . . . . . Exercises . .3 The Van Kampen theorem . . . . . . . . . 6 Cohomology of sheaves 6. . . . . . . . . . . . . . . . . . . . . 6. .2 Cech complexes for closed coverings . . . . . . 5. . . . . . . .3 Sheaf associated with a presheaf . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . 4. . . . . . . . . . . . . . . . . . . . . . . . . . 134 .1 Fundamental groupoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Locally constant and locally free sheaves . . . . . . . . Exercises . . . . . . . . . . . . .4 Coverings . . . . . .4 Cohomology of some classical manifolds Exercises . . . . . . . . . . ˇ 6. . . . . . . . . . . . . . . 139 139 142 146 148 152 . . . 5. . . 123 . . . . . . . . . . . . . . 127 . . . . . . . . .7 Bifunctors . . . 5. . . . . . . . . . . . . . . . . . 136 . . . . . . . . . . . . . . . . . . . . . . .5 Direct and inverse images . . . . . . . . 7. . . . . . . . . . . . . 7. .
In Chapter 2 we expose the basic language of categories and functors. Chapters 3 and 4 are devoted to additive and abelian categories. and applies to many areas of Mathematics. here. Next. Lectures will be organized as follows. As a byproduct. The derived functors H j (X. We apply . F ) encode the ‘ ‘obstructions” to pass from local to global. The functor associating to a sheaf F on a topological space X the space F (X) of its global sections is left exact. kX ) of the sheaf kX of kvalued locally constant functions is therefore a topological invariant of the space X. it is a homotopy invariant. kX ) in various situations. [24]. and we shall explain how to calculate H j (X. An expanded version of these Notes may be found in [23].CONTENTS 5 Introduction This course is an ﬁrst introduction to Algebraic Topology from the point of view of Sheaf Theory. Algebraic Topology is usually approached via the study of of homology deﬁned using chain complexes and the fundamental group. the cohomology groups H j (X. Hence. Sheaves on topological spaces were invented by Jean Leray as a tool to deduce global properties from local ones. which asserts that a category C may be embedded in the category C ∧ of contravariant functors on C with values in the category Set of sets. by using F injective resolutions. We also introduce the fundamental group π1 (X) of a topological space (with suitable assumptions on the space) and prove an equivalence of categories between that of ﬁnite dimensional representations of this group and that of local systems on X. Given a ring k. A key point is the Yoneda lemma. This tool turned out to be extremely powerful. It serves as a guide for the theory of additive and abelian categories which is exposed in the subsequent chapters. the accent is put on the language of categories and sheaves. Chapter 1 is a brief survey of linear algebra over a ring. from Algebraic Geometry to Quantum Field Theory. we deduce the Van Kampen theorem from the theorem on the glueing of sheaves deﬁned on a covering. This naturally leads to the concept of representable functor. with particular attention to locally constant sheaves. whereas. Then we brieﬂy explain the construction of the right derived functor by using injective resolutions and later. we start by studying complexes (and double complexes) in additive and abelian categories. Indeed. but not right exact in general. The aim is the construction and the study of the derived functors of a left (or right) exact functor F of abelian categories. we study inductive and projective limits in some detail and with many examples.
In these Notes. we show how to compute the cohomology of spaces by using cellular decomposition. (Sometimes. and the unit are denoted by +. 1. However. All along these Notes. we deﬁne the fundamental groupoid π1 (X) of a locally arcwise connected space X as well as the monodromy of a locally constant sheaf and prove that under suitable assumptions. ·. We also show that the Van Kampen theorem may be deduced from the theorem on the glueing of sheaves and apply it in some particular sitations. In Chapter 6 we prove that the category of abelian sheaves has enough injectives and we deﬁne the cohomology of sheaves. all rings are unital and associative but not necessarily commutative. . 0. using the fact that the cohomology of locally constant sheaves is a homotopy invariant. . we shall often write for short ab instead of a · b. We denote by N the set of nonnegative integers.6 CONTENTS these results to the case of the functors Ext and T or. the zero element. We construct resolutions ˇ of sheaves using open or closed Cech coverings and. Conventions. we study abelian sheaves on topological spaces (with a brief look at Grothendieck topologies). In Chapter 5. k will be a ﬁeld. . k will denote a commutative ring. N = {0. . respectively. the monodromy functor is an equivalence. In Chapter 7. We apply this technique to deduce the cohomology of some classical manifolds. }. We also explain how to obtain locally constant or locally free sheaves when glueing sheaves. 1.) We denote by ∅ the empty set and by {pt} a set with one element. We construct the sheaf associated with a presheaf and the usual internal operations (Hom and ⊗) and external operations (direct and inverse images). The operations.
[4]. b ∈ A and m. then A is an Aalgebra. 7 All along these Notes. Many notions introduced in this chapter will be repeated later in a more general setting. k is a commutative ring. we shall not write ϕ. m ∈ M . 0) endowed with an external law A × M → M satisfying: (ab)m = a(bm) (a + b)m = am + bm a(m + m ) = am + am 1·m=m . In the sequel. Some references: [1]. a module M over A means a left Amodule. Let A be a kalgebra. Notice − that a ring A is always a Zalgebra. Recall that an Amodule M is an additive group (whose operations and zero element are denoted +. Since we do not assume A is commutative. a ring endowed with a morphism of rings ϕ : k → A such that the image of k is contained in the center of A.1 Modules and linear maps where a.Chapter 1 Linear algebra over a ring This chapter is a short review of basic and classical notions of commutative algebra. that is. Note that M inherits a structure of a kmodule via ϕ. If A is commutative. 1. Unless otherwise speciﬁed. we have to distinguish between left and right structures. if there is no risk of confusion.
The Weyl algebra Wn (k) may be regarded as the ring of diﬀerential operators with coeﬃcients in k[x1 . . . . . [∂i . n ∈ N implies n + n ∈ N and n ∈ N. is also a commutative ring. the ring of formal powers series with coeﬃcients in A. . xi ] = δj i where [p. If N is a submodule of M . q] = pq − qp and δj is the Kronecker symbol. Hence the product ab in Aop is the product ba in A and an Aop module is a right Amodule.8 CHAPTER 1. Note that if A is a ﬁeld..e. ∂j ] = 0. the ring Mn (k) of square matrices of rank n with entries in k is non commutative. . it has no non trivial ideal. xj ] = 0. . If A = C[x]. . .. the quotient module M/N is characterized by the following “universal property”: for any module L.e. the ring of polynomials in n variables with coeﬃcients in A. R. a ∈ A. ∂j (1 ≤ i. P (0) = 0} is a non trivial ideal. and relations : i [xi . C are rings.1. xn ] becomes a left Wn (k)module: xi acts by multiplication and ∂i is the derivation with respect to xi . f satisﬁes: − f (m + m ) = f (m) + f (m ) m. (ii) Let k be a ﬁeld. i. m ∈ M f (am) = af (m) m ∈ M. then I = {P ∈ C[x]. If A is a commutative ring. a ∈ A implies an ∈ N . Then for n > 1. A morphism f is an isomorphism if there exists a morphism g : N → M − with f ◦ g = idN . then A[x1 . . and k[x1 . the ring of integers. Since a ﬁeld is a ring. j ≤ n) with coeﬃcients in k. . Note that if the ring A is a ﬁeld (here. a ﬁeld is always commutative). it is easily checked that the inverse map f −1 : N → M is itself Alinear. A morphism f : M → N of Amodules is an Alinear map. Q. xn ]]. [∂j . (iii) Let k be a ﬁeld.1. then an Amodule is nothing but a vector space. . (i) The ﬁrst example of a ring is Z. A submodule N of M is a subset N of M such that n. its only ideals are {0} and A. Examples 1. i. Hence f is an isomorphism if and only if f is Alinear and bijective. . xn ]. LINEAR ALGEBRA OVER A RING We denote by Aop the ring A with the opposite structure. denoted Wn (k). xn ]. any morphism . It is a subring of A[[x1 . . The Weyl algebra in n variables. is the non commutative ring of polynomials in the variables xi . If f is bijective. g ◦ f = idM . . A submodule of the Amodule A is called an ideal of A. .
i We shall sometimes identify Mk to its image in i Mi by εk . vv vv k kkk vvv πi kkk v k kkk / M k kq q π qqqqj qq q# fj ) fi kkkkk L Mj . − There are natural injective morphisms: ε k : Mk → − Mi . and let {Mi }i∈I be a family of Amodules indexed by I. This is − visualized by the diagram Nd d / 0d d M L g / M/N d d d h d  h Let I be a set. this − family factorizes uniquely through i Mi . Mj . if the set I is ﬁnite. This is visualized by the diagram Mi r rr rr i f r εi rrr # /) 5 k Mk kkk L. − Note that given a module L and a family of morphisms fi : L → Mi . this family factorizes − uniquely through i Mi .1.1. There are natural surjective morphisms: πk : i Mi → M k . MODULES AND LINEAR MAPS 9 h : M → L which induces 0 on N factorizes uniquely through M/N . Note that given a module L and a family of morphisms fi : Mi → L. the natural injection i Mi → i Mi is an isomorphism. This is visualized by the diagram k5 Mi kkkv. The product i Mi is the set of families {(xi )i∈I } with xi ∈ Mi . The direct sum i Mi is the submodule of i Mi consisting of families {(xi )i∈I } with xi = 0 for all but a ﬁnite number of i ∈ I. and this set naturally inherits a structure of an Amodule. In particular. kkk ww ww kkkkkk ww kkk fj w ww kk kkk εj .
f (m) = 0} there exists m ∈ M. then r is uniquely determined and one says M is free of rank r. f (m) = n}. M I := i Mi . Example 1. If (Mi )i∈I is a family of submodules of an Amodule M . An Amodule M is free of rank one if it is isomorphic to A. . and M is free if it is isomorphic to a direct sum i∈I Li . say r.1.10 CHAPTER 1. . . . one denotes by i Mi the submodule of M obtained as the image of the natural morphism − i Mi → M . LINEAR ALGEBRA OVER A RING If Mi = M for all i ∈ I. xn ]. · · · . Let f : M → N be a morphism of Amodules.1. Let Wn (k) denote as above the Weyl algebra. ∂n ). . called the kernel and the image of f . j ∈ Z and Alinear maps dj : M j → M j+1 such that dj ◦ dj−1 = 0 for M M M all j. . Hence.1) Wn (k)/ j ∼ Wn (k)∂j − k[x1 . A complex M • of Amodules is a sequence of modules − M j . xn ]. Wn (k) − P → P (1) ∈ k[x1 .2. one has the isomorphism of left Wn (k)modules: (1. . Coim f = M/ Ker f. This is also the module generated in M by the set i Mi . . Since the natural morphism Coim f → Im f is an isomorphism. These are submodules of M and N respectively. . xn ]. Consider the left Wn (k)linear map Wn (k) → k[x1 . . One also introduces the cokernel and the coimage of f : Coker f = N/ Im f. one shall not − use Coim when dealing with Amodules. If card (I) is ﬁnite. This map is clearly surjective and its kernel is the left ideal generated by (∂1 . One calls this module the sum of the Mi ’s in M . .2.2 Complexes Deﬁnition 1. . One sets : Ker f = {m ∈ M . → 1. each Li being free of rank one. Im f = {n ∈ N . . one writes: M (I) := i Mi . respectively.
2. Let A = k[x1 . In such a case we also call such a sequence a complex by identifying it to the complex · · · → 0 → M j − M j+1 → · · · → M j+k → 0 → · · · .4. M A morphism of complexes f : M → N is a commutative diagram: − / M / k−1 dk−1 M / Mk fk / f k−1 N k−1 dk−1 N / Nk / Remark 1.2. − − → − − − − In particular. (Hence. − − (iii) An exact complex 0 → M → M → M → 0 is called a short exact − − − − sequence. ∼ (i) The sequence (1. x2 ] and consider the sequence: 0 → A − A2 − A → 0 − → → − where d0 (P ) = (x1 P. Example 1. with g◦f = 0. − − Consider a sequence (1. M → M → M is a complex if g ◦ f = 0. R) = x2 Q − x1 R. a complex M j → · · · → M j+k is exact if any sequence − − M n−1 → M n → M n+1 extracted from this complex is exact.2.1.2) is exact if Im f − Ker g. d0 d1 f g f g dj dj . One also encounters ﬁnite sequences of morphisms M j − M j+1 → · · · → M j+k → − − such that dn ◦ dn−1 = 0 when it is deﬁned.) − − Deﬁnition 1. One also often write dj instead of dj .2. One checks immediately that d1 ◦ d0 = 0: the sequence above is a complex.2. → (ii) More generally. this sequence is a complex.2) M → M → M . x2 P ) and d1 (Q. COMPLEXES One writes a complex as: M • : · · · → M j −M M j+1 → · · · − → dj 11 If there is no risk of confusion. one writes M instead of M • .3.
− − − − and f is an isomorphism if and only if Ker f = Coker f = 0.2. Hence it deﬁnes the morM N phism H j (f • ) : H j (M • ) → H j (N • ). To a morphism f : M → N one then associates the two short exact − sequences : 0 → Ker f → M → Im f → 0. consider a complex M • of the type: 0 → M 0 → M 1 → 0. Hence. then for each j. As a particular case. In this case one writes : ∼ f : M − N. f j sends − j−1 j−1 Ker dj • to Ker dj • and sends Im dM • to Im dN • . − (b) there exists k : M → M such that k ◦ f = idM − (c) there exists h : M → M and k : M → M such that such that idM = − f ◦ k + h ◦ g.5. that is. a complex M • is exact if all its cohomology objects are zero.3) − 0 → M → M → M → 0. − − − f g f Then the following conditions are equivalent: − (a) there exists h : M → M such that g ◦ h = idM . − − − − 0 → Im f → N → Coker f → 0. epimorphism) if Ker f (resp. for short) if H j (f ) is an isomorphism for all j. Im dk−1 = Ker dk for all k. If f • : M • → N • is a morphism of complexes.12 CHAPTER 1. → One says f is a monomorphism (resp. LINEAR ALGEBRA OVER A RING One deﬁnes the kth cohomology object of a complex M • as: H k (M • ) = Ker dk / Im dk−1 . Coker f ) = 0. − One says that f is a quasiisomorphism (a qis. − − − Then H 0 (M • ) = Ker f and H 1 (M • ) = Coker f . Consider an exact sequence (1. Proposition 1. .
The proof is similar and left to the reader. Since g is onto. There is a natural isomorphism Hom A (A. In other words. Hence. In the above situation. N ) by setting: (λf )(m) = λ(f (m)). through M . then Hom (K. we get g ◦ (idM −h ◦ g) = 0. M ) {m ∈ M . Hence (λf )(am) = λf (am) = λaf (m) = aλf (m) = a(λf (m)). If A is a ﬁeld. More generally. For example. − − such that ϕ and ψ are isomorphisms inverse to each other. This is clearly a kmodule.6. Deﬁnition 1.d.3. N ). which implies that idM −h ◦ g factors through Ker g. (d) ⇔ (a)&(b)&(c) is obvious. We shall often set for short Hom (M. (c) ⇒ (a). M ) is an Amodule. there exists k : M → M such that idM −h ◦ g = f ◦ k. ·2 1. Im = 0}. − (b) ⇒ (c). g) : M → M ⊕M and ψ = (f +h) : M ⊕M → M . M ) one associates u(1) and to m ∈ M one associates the linear map A → − M. we ﬁnd g = g ◦ h ◦ g. THE FUNCTOR HOM 13 (d) there exists ϕ = (k. M ) M : to u ∈ Hom A (A. all exact sequences split. A denotes a kalgebra and the notation M ∈ Mod(A) means that M is an Amodule. but this is not the case in general.3 The functor Hom In this section. the exact sequence (1. that is (g ◦ h − idM ) ◦ g = 0. if I is an ideal of A then Hom A (A/I. a → am. q. N ) the set of Alinear maps f : M → N . (c) ⇒ (b).2.) Let M and N be two Amodules.e. Notice that if K is a kmodule. One denotes by Hom A (M. Since g ◦ f = 0. this implies g ◦ h − idM = 0. the exact sequence of Zmodules 0 → Z − Z → Z/2Z → 0 − → − − does not split. (a) ⇒ (c). In fact one − deﬁnes the action of k on Hom A (M. .1. One calls Mod(A) the category of Amodules. (A precise deﬁnition will be given in Chapter 2. one says that the exact sequence (1.3) is isomorphic to the exact sequence 0 → − M → M ⊕ M → M → 0.3) splits. − − − Proof. N ). N ) = Hom k (M. and λf ∈ Hom A (M. Since g = g ◦ h ◦ g. that is. The proof is similar and left to the reader.
N ) → Hom B (F (M ). Hence. − − − Clearly. L) → (M → K) → (M → K → L).14 Functors CHAPTER 1. The functor Hom A (M. N ) Let M ∈ Mod(A). a contravariant functor G : Mod(A) → Mod(B) associates a − − morphism G(f ) : G(N ) → G(M ) and satisﬁes − G(g ◦ f ) = G(f ) ◦ G(g). g) : Hom A (M. F (N ) − is klinear. Let A and B be two kalgebras. F (M ⊕ N ) F (M ) ⊕ F (N ) and the map F : Hom A (M. Similarly. with the difference that it reverses the direction of the arrows. for N ∈ Mod(A). • ) : Mod(A) → Mod(k) − associates Hom A (M. Hom A (M. N ) : Mod(A) → Mod(k) − h h g g◦ f g . K) − Hom A (M. A contravariant functor is almost the same as a functor. A functor F : Mod(A) → Mod(B) as− sociates a Bmodule F (M ) to each Amodule M and associates a Blinear map F (f ) : F (M ) → F (N ) to each Alinear map f : M → N such that: − − F (idM ) = idF (M ) for any Amodule M . K) to the Amodule K and to an Alinear map g : K → − L it associates Hom A (M. • ) and Hom A ( • .e. the contravariant functor Hom A ( • .. LINEAR ALGEBRA OVER A RING Although the general deﬁnition of a functor will be given in Chapter 2. The functors Hom A (M. • ) is a functor from the category Mod(A) of Amodules to the category Mod(k) of kmodules. to a morphism f : M → N . we give it in the particular case of categories of modules. i. − − A functor F : Mod(A) → Mod(B) is kadditive if it commutes to ﬁnite direct − sums. F (g ◦ f ) = F (f ) ◦ F (g) for any morphisms M → N → L.
− − − Clearly. N ) : Hom A (L. (a) Let 0 → M → M → M − − − modules. N ) − Hom A (K.3. the two functors Hom A (M.3. K) × Hom A (M. N ) × Hom A (L. L). N ) Hom A (M. (i) the sequence is exact. M ) → Hom A (L. M ) → Hom A (L. i. (iii) any morphism h : L → M such that g ◦ h = 0. h = f ◦ h . The assertions − − − below are equivalent. (ii) M is isomorphic by f to Ker g. Hom A (K ⊕ L. with h : L → M ). N ) Hom A (M.1. The assertions below are equivalent. N ) commute to ﬁnite direct sums or ﬁnite products. N ) to the Amodule K and to an Alinear map g : K → − L it associates Hom A (g. be a complex of A hh 0 / M } / M g 0h h h h! /M (iv) for any module L.. Exactness Proposition 1. f g 0 → Hom A (L. (b) Let M → M → M → 0 be a complex of Amodules.e.4) is exact. factorizes uniquely − through M (i. K × L) Hence.1. (ii) M is isomorphic by g to Coker f .e. the sequence of kmodules (1.. M ) − − − . N ) → (L → N ) → (K → L → N ). (i) the sequence is exact. these functors are additive. THE FUNCTOR HOM 15 associates Hom A (K. • ) and Hom A ( • . This is visualized − by L hh h h f f g h g h ◦g Hom A (K.
• )) are left exact functors. the sequence − − − 0 → G(M ) → G(M ) → G(M ) in Mod(B) is exact. (a) (i) ⇒ (ii) is obvious.3. usually denoted by M ∗ . LINEAR ALGEBRA OVER A RING (iii) any morphism h : M → L such that h ◦ f = 0. L) → Hom A (M . (iii) ⇔ (iv) is tautological. − − − (ii) An additive functor F : Mod(A) → Mod(B) is right exact if for any ex− act sequence M → M → M → 0 in Mod(A). If M − is ﬁnite dimensional. Hence. then Hom k (M. Proof. the sequence F (M ) → − − − − F (M ) → F (M ) → 0 in Mod(B) is exact. Deﬁnition 1. the fact that (a)–(i) ⇔ (a)–(iv) and (b)–(i) ⇔ (b)–(iv)) is formulated by saying that Hom A ( • . Finally. − − (iii) An additive contravariant functor G : Mod(A) → Mod(B) is left exact − if for any exact sequence M → M → M → 0 in Mod(A). the sequence of kmodules (1. and this − characterizes Ker g. the − − − sequence 0 → F (M ) → F (M ) → F (M ) in Mod(B) is exact. This is visualized − by M g gg g f / M L g /M / 0 0g g gg h ! } h (iv) for any module L. L) − − − . Note that if A = k is a ﬁeld. then M M ∗∗ . since any linear map h : L → M such that g ◦ h = 0 factorizes uniquely through Ker g. If u : L → M is a linear map. the map ∗ ∗ Hom k (u.16 CHAPTER 1.2. (i) An additive functor F : Mod(A) → Mod(B) is left − exact if for any exact sequence 0 → M → M → M in Mod(A). q. L) → Hom A (M.d. k) : M → L is usually denoted by t u and called the transpose of − u. − − − (iv) An additive contravariant functor G : Mod(A) → Mod(B) is right exact − if for any exact sequence 0 → M → M → M in Mod(A). − − − (v) An additive functor is exact if it is both right and left exact.5) is exact. L) and Hom A (L.e. h = h ◦ g. as well as (ii) ⇒ (iii).e. with h : M → L). k) is the algebraic dual of M . the sequence − − − G(M ) → G(M ) → G(M ) → 0 in Mod(B) is exact. the vector space of linear functional on M . (b) The proof is similar. factorizes uniquely − through M (i.. 0 → Hom A (M .
− − (iii) for any exact sequence 0 → M → M → M → 0 in Mod(A).1. A) is left exact.5. Consider an additive functor F : Mod(A) → Mod(B) and − assume that for each exact sequence 0 → M → M → M → 0 in Mod(A). we ﬁnd that Hom A (A/Ax. Proof. the sequence − − F (M ) → F (M ) → F (M ) is exact in Mod(B). − − − − The proof is left as an exercise. the map M → L factorizes through a map h : M → L which is − − clearly injective. Consider an exact sequence 0 → M → M → L and denote by M the − − − cokernel of f . an exact sequence 0 → F (M ) → F (M ) → F (M ). L) and Hom A (M. On the other hand. The − conditions below are equivalent: (i) F is exact. Apply Hom A (·. − − − − Applying the functor F . Then F − − − is left exact.4. In fact choose A = k[x]. Consider the exact sequence 0 → M → L → Coker h → 0. Moreover: Lemma 1. There is a similar result for right exact functors and for contravariant functors. The functors Hom A ( • . • ) are not right exact in general. the kernel of − − F (M ) → F (L) is isomorphic to the kernel of F (M ) → F (M ). Example 1. We get an exact sequence 0 → M → M → M → 0. the − − − − sequence 0 → F (M ) → F (M ) → F (M ) → 0 is exact in Mod(B).3.3. by − − − − the hypothesis. A) = 0. − − − q. and consider the exact sequence of Amodules: (1.3. x· . We get the sequence: → − 0 → Hom A (A/Ax. with k a ﬁeld. we obtain that F (M ) → F (L) is injective.e.d. Consider an additive functor F : Mod(A) → Mod(B).3. Since − the map F (M ) → F (L) factorizes through F (M ) → F (M ).6) 0 → A − A → A/Ax → 0 − → − − ·x f (where ·x means multiplication by x). − − − − the sequence 0 → F (M ) → F (M ) → F (M ) is exact in Mod(B). A) → A − A → 0 − − which is not exact since x· is not surjective.3. (ii) for any exact sequence M → M → M in Mod(A). It follows − − that the sequence 0 → F (M ) → F (M ) → F (L) is exact. On the − − − other hand. A) to this sequence. hence. since x· is injective and Hom A ( • . THE FUNCTOR HOM 17 Lemma 1.
it sends split exact sequences to split exact sequences. A/Ax) = 0.6. the sequence Hom A (M . Injective and projective modules Deﬁnition 1. M ) → − − Hom A (P.7. • ) being additive. A) = 0 and Hom A (A/Ax. M ) → 0 is exact in Mod(k) or. (i) ⇒ (ii) and (i) ⇒ (iii) are obvious. A/Ax) → 0. − − if the functor Hom A ( • . − by Proposition 1. Proof. (iii) for any Amodule L. (i) An Amodule I is injective if for any exact sequence 0 → M → M → M in Mod(A). I) → 0 is exact in Mod(k) or. apply Hom A (A/Ax. • ) to the exact sequence (1. P ) 0. I) is exact. Hence. M ) − Hom A (L. (iii) ⇒ (i). the sequence Hom A (P. equivalently. We get the sequence: 0 → Hom A (A/Ax. the map Hom A (N. and this implies P 0. M ) → Hom A (P. hence Hom A (P.2.d. L) is an iso→ morphism. Notice moreover that the functor Hom A ( • .6) does not split. f is injective and moreover. Therefore. The − conditions below are equivalent: (i) f is an isomorphism. A) → Hom A (A/Ax. This shows again that (1. − (ii) An Amodule P is projective if for any exact sequence M → M → − − M → 0 in Mod(A). − − − − Since Hom A (A/Ax. N ) is an iso→ morphism. L) 0 for all module L.3.3. I) → Hom A (M . (ii) ⇒ (i).e.6). M ) we ﬁnd that there exists g : N → M such that g ◦ f = idM . Proposition 1. if the functor − Hom A (P. By choosing L = M and idM ∈ Hom A (M. Let f : M → N be a morphism of Amodules. q. ◦f f◦ . L) − Hom A (M. LINEAR ALGEBRA OVER A RING Similarly. Hom A (P. A) → Hom A (A/Ax. equivalently. The next result is of constant use. • ) is exact.18 CHAPTER 1. I) → − − − − Hom A (M. the map Hom A (L. this sequence is not exact. (ii) for any Amodule L.5 there exists an isomorphism N M ⊕ P . Choose L = A.
I) − Hom A (M . THE FUNCTOR HOM 19 Proposition 1. To conclude. all modules are both injective and projective. P is projective if and only if for any solid diagram in which the row is exact: P h k /0 /M M the dotted arrow may be completed. N1 ) where the pij ’s belong to the ring A and uj . .3. I) − Hom A (M. − − − − one gets the exact sequence: Hom A (M . q.3. I) → 0 is − → → − exact by Proposition 1. Then − − − − the sequence 0 → Hom A (M .8) Pu = v . . . I) to the sequence 0 → M → M → M . } f Generators and relations Suppose one is interested in studying a system of linear equations N0 (1. vi belong to some left Amodule L. − → − Thus there exists h : M → I such that h ◦ f = k.3. . In other words. apply Lemma 1.e. one can write equations (1.1. Proof. (i) Assume that I is injective and let M denote the cokernel of the map M → M .d.7) j=1 pij uj = vi . A free module is projective and if A = k is a ﬁeld. I) → Hom A (M. Applying Hom A ( • . An Amodule I is injective if and only if for any solid diagram in which the row is exact: 0 / M k h f / M I } the dotted arrow may be completed.1 and the hypothesis. we get a similar result assuming P is projective.3. (i = 1. consider an exact sequence 0 → M → M → M → 0.8. making the diagram commutative. − f g (ii) Conversely. I) − Hom A (M .7) as (1. Using matrix notations. making the diagram commutative. ◦h ◦f ◦f By reversing the arrows. I) → 0.3.
10) fi · P = pij ej . . .9) ·P : AN1 → AN0 . . L). We shall construct a kmodule denoted N ⊗A M such that f factors uniquely through the bilinear map N × M → N ⊗A M followed by a klinear − map N ⊗A M → L. consider a right Amodule N . fN1 ) denote the canonical basis of AN0 and AN1 . Let us say that a map f : N × M → L is (A. . deﬁning the Alinear map P · : LN0 → LN1 . Now consider the right Alinear map − (1. . N1 ). λ ∈ k. solves a “universal problem”. − − P· Hence. L) to this sequence. . This is visualized by: − /N⊗ M N ×M A www www www f www & L . we ﬁnd the exact sequence of kmodules: (1. that we shall construct below. . . m) = λ(f (n. eN0 ) and (f1 . uN0 ) denote the images by ψ of (e1 .4 Tensor product The tensor product. m) = f (n. . we have an exact sequence of left Amodules: (1. m)) for all (n.7) is described by Hom A (M. . . a left Amodule M . N0 Hence Im P is generated by the elements j=1 pij ej for i = 1. . . One gets: N0 (1. Denote by M the quotient module AN0 /AN1 · P and by ψ : AN0 → M the natural A− linear map. .11) → − → AN1 − AN0 − M → 0. the kmodule of solutions of the homogeneous equations associated to (1. . ·P ψ Applying the left exact functor Hom A ( • . Then M is a left Amodule with generators (u1 .12) → 0 → Hom A (M. Let (e1 . k)bilinear if − f is additive with respect to each of its arguments and satisﬁes f (na. . and a kmodule L. By construction. . eN0 ). . Namely. . . . respectively. . LINEAR ALGEBRA OVER A RING where P is the matrix (pij ) with N1 rows and N0 columns. − where ·P operates on the right and the elements of AN0 and AN1 are written as rows. . . . .20 CHAPTER 1. . . uN0 ) and relations N0 pij uj = j=1 0 for i = 1. 1. N1 . . f (n(λ). am). . . Let (u1 . j=1 (i = 1. . m) ∈ N × M and a ∈ A. L) → LN0 − LN1 . N1 .
− Tensor product commutes to direct sum. . m + m ) − (n.4. mj ∈ M and: (n + n ) ⊗ m = n ⊗ m + n ⊗ m n ⊗ (m + m ) = n ⊗ m + n ⊗ m na ⊗ m = n ⊗ am λ(n ⊗ m) = nλ ⊗ m = n ⊗ λm. a ∈ A. hence a (A. that is. The tensor product N ⊗A M is the kmodule deﬁned as the quotient of k (N ×M ) by the submodule generated by the following elements (where n. m) − (n. − One constructs similarly g ⊗ idM associated to g : N → L. m ∈ M. m. The image of (n. remark that one may identify a set I to a subset of k (I) as follows: to i ∈ I. and ﬁnally − − a klinear map idN ⊗f : N ⊗A M → N ⊗A L. M. we have constructed additive functors N ⊗A • : Mod(A) → Mod(k). if L. there are natural isomorphisms: (N ⊕ N ) ⊗A M N ⊗A (M ⊕ M ) (N ⊗A M ) ⊕ (N ⊗A M ). there is an isomorphism: N ⊗A M M ⊗A N . One simply writes L ⊗A M ⊗A N . m) (n. (N ⊗A M ) ⊕ (N ⊗A M ). TENSOR PRODUCT 21 First. m ) (na. m) − (nλ. It deﬁnes a linear map idN ×f : − N × M → N × L. that is. given by n ⊗ m → m ⊗ n and moreover the tensor product is associative. there are natural isomorphisms L ⊗A (M ⊗A N ) (L ⊗A M ) ⊗A N . m) − (n. Hence an element of N ⊗A M may be written (not uniquely!) as a ﬁnite sum j nj ⊗ mj . m). − A Note that if A is commutative.13) lj = 1 if 0 if j = i. k)bilinear map N × M → N ⊗A L. Consider an Alinear map f : M → L. − op • ⊗ N : Mod(A ) → Mod(k). we associate {lj }j∈I ∈ k (I) given by (1. Clearly. m) − (n . nj ∈ N. j = i. m) in N ⊗A M is denoted n⊗m. N are Amodules. λ ∈ k and N × M is identiﬁed to a subset of k (N ×M ) ): (n + n . m) − (n.1. am) λ(n. n ∈ N.
Sometimes.4. a2 ∈ A2 . Hom A2 (2 M1 . if A is a kalgebra. 2 M3 )) Hom A2 (2 M1 ⊗A1 1 M4 . In particular. A2 . M )) Hom k (L. Note that the actions of A1 and A2 on M commute. we have the isomorphisms (1. M ) and conversely. 1 M3 ) is an (A2 ⊗ Aop )module. Proof. − Ψ ( • ) := Hom A (N. We shall only prove (1. M. Hom A (L ⊗k N.d. • ) : Mod(A) → Mod(A). M )). a1 a2 m = a2 a1 m. LINEAR ALGEBRA OVER A RING There is a natural isomorphism A ⊗A M M .e. In this case. Let A1 . An (A1 ⊗Aop )module M is also called a (A1 . by setting (a1 ⊗ a2 ) · (b1 ⊗ b2 ) = a1 b1 ⊗ a2 b2 . − Φ ( • ) := • ⊗k N : Mod(k) → Mod(A). Hom A (N. Consider the functors Φ( • ) := L ⊗k • : Mod(A) → Mod(A). M ) Hom A (N. Let i Mj be an (Ai ⊗ Aj )module. Hom k (L.22 CHAPTER 1.1. Consider two kalgebras. A2 )bimodule (a left A1 module 2 and right A2 module). m ∈ M. Then ⊗A2 2 M3 is an (A1 ⊗ Aop )module. one has to consider various rings. 3 1 M2 and there is a natural isomorphism of A4 ⊗ Aop modules 3 (1. op Proposition 1.15) Hom A (L ⊗k N. A1 and A2 .15) in the particular case where A = k. − Ψ( • ) := Hom k (L. N are left Amodules and L is a kmodule. • ) : Mod(A) → Mod(k).14) Hom A1 (1 M4 . We shall often write for short M ⊗k N = M ⊗ N. A4 denote four kalgebras. Then A1 ⊗ A2 has a natural structure of a kalgebra. 3 Hom A1 (1 M2 . − . and a kbilinear map from L × N to M deﬁnes uniquely a linear map from L to Hom A (N. q. that is. a1 ∈ A1 . M ) is nothing but the kmodule of kbilinear maps from L × N to M . 2 M3 ). A3 .
15) become: Hom A (Φ(N ). • ⊗A M : Mod(Aop ) → Mod(k) and N ⊗A − right exact functors. Hence. Ψ (M )) 23 One translates these isomorphisma (see Chapter 2 below) by saying that (Φ.1. In fact. L) → Hom k (N ⊗A M.1 ((b). Hom A (N.3.e. it is enough to check that for any kmodule L. Proof. L) → Hom k (N ⊗A M . • ⊗A M is not left exact in general. This sequence is isomorphic to the sequence 0 → Hom k (M .3.d. (i) ⇒ (ii)).4. TENSOR PRODUCT The isomorphisms in (1. L)) − − → Hom k (M . We get the sequence: 0 → A/Ax − A/Ax → A/xA ⊗A A/Ax → 0 − → − − x· Multiplication by x is 0 on A/Ax. M ) Hom A (Φ (N ). Ψ(M )) Hom A (N. Hence this sequence is the same as: 0 → A/Ax → A/Ax → A/Ax ⊗A A/Ax → 0 − − − − which shows that A/Ax ⊗A A/Ax is not exact.4. M ) Hom A (N. L)) → Hom k (M. • q. the sequence 0 → Hom k (N ⊗A M .4.3. Hom A (N. A/Ax and moreover that this sequence 0 . L) − − − is exact. then the sequence of kmodules N ⊗A M → N ⊗A M → N ⊗A M → 0 − − − is exact. If M → M → M → 0 is an exact sequence of left A− − − modules. Hom A (N. Proposition 1. By Proposition 1. L)) − and it remains to apply Proposition 1. Ψ ) are pairs of adjoint functors. − → − − Apply • x· ⊗A A/Ax. : Mod(A) → Mod(k) are − Example 1. consider the commutative ring A = C[x] and the exact sequence of Amodules: 0 → A − A → A/xA → 0. Ψ) and (Φ .1 (b).2.
the sequence 0 → N ⊗A M → − − − − − N ⊗A M → N ⊗A M is exact in Mod(k). LINEAR ALGEBRA OVER A RING Deﬁnition 1. Deﬁnition 1. (i) One says that (I. (c) i ≤ j and j ≤ i implies i = j. Let (I. vij } of Amodules indexed by (I. ≤) is the data for each i ∈ I of an Amodule Ni and for each pair i. such that for any − . Let I be a set. ≤) be an ordered set. 1. j. N is ﬂat if − and only if the functor N ⊗A • is exact. k with i ≤ j and j ≤ k: vii = idNi vij ◦ vjk = vik . One says that J is coﬁnal to I if for any i ∈ I there exists j ∈ J with i ≤ j.5. A projective system {Ni . One proves easily that a projective module is ﬂat (see Exercise 1.4. (ii) The opposite order (I.24 CHAPTER 1. (i) An Aop module N is ﬂat if for any exact sequence 0 → M → M → M in Mod(A). one says that N is faithfully ﬂat. Consider the “universal problem”: to ﬁnd an Amodule N and linear maps vi : N → Ni satisfying vij ◦ vj = vi for all i ≤ j. (i) An order ≤ on I is a relation which satisﬁes: (a) i ≤ i.4. j with i ≤ j of an Alinear map vij : Nj → Ni .2). − such that for all i. (ii) If N is ﬂat and moreover N ⊗A M = 0 implies M = 0.5 Limits Deﬁnition 1. ≤) be a ordered set and let A be a ring. (ii) Assume I is ﬁltrant and let J ⊂ I be a subset. (iii) Similarly.1. ≤) is ﬁltrant (one also says “directed”) if for any i. In other words. The following deﬁnition will be of constant use. an Amodule N is ﬂat if the functor • ⊗A N is exact and N is faithfully ﬂat if moreover M ⊗A N = 0 implies M = 0. (iii) An order is discrete if i ≤ j implies i = j. ≤op ) is deﬁned by i ≤op j if and only if j ≤ i. (b) i ≤ j & j ≤ k implies i ≤ k. j ∈ I there exists k with i ≤ k and j ≤ k.2. Let (I.5.
1.5. LIMITS
25
Amodule L and linear maps gi : L → Ni , satisfying vij ◦ gj = gi for all i ≤ j, − there is a unique linear map g : L → N such that gi = vi ◦ g for all i. If such − a family (N, vi ) exists (and we shall show below that it does), it is unique up to unique isomorphism and one calls it the projective limit of the projective system (Ni , vij ), denoted lim Ni . This problem is visualized by the diagram: ← −
i
L
6N mmm< O i mmmzzzz gi mmm mm zzz mmm z vi mm m mmm / lim Nk vij ← − k hh v hh j gj hhh h ! (
Nj .
An inductive system {Mi , uji} of Amodules indexed by (I, ≤) is the data for each i ∈ I of an Amodule Mi and for each pair i, j with i ≤ j of an Alinear map uji : Mi → Mj , such that for all i, j, k with i ≤ j and j ≤ k: − uii = idMi ukj ◦ uji = uki . Note that a projective system indexed by (I, ≤) is nothing but an inductive system indexed by (I, ≤op ). Consider the “universal problem”: to ﬁnd an Amodule M and linear maps ui : Mi → M satisfying uj ◦ uji = ui for all i ≤ j, such that for any − Amodule L and linear maps fi : Mi → L satisfying fj ◦ uji = fi for all i ≤ j, − there is a unique linear map f : M → L such that fi = f ◦ ui for all i. If such − a family (M, ui )i exists (and we shall show below that it does), it is unique up to unique isomorphism and one calls it the inductive limit of the inductive system (Mi , uji ), denoted lim Mi . This problem is visualized by the diagram: − →
i
Mi h
uji
Mj
hh hh i f h ui hh" ( lim Mk / − → mm6 L. k mm = mm uj zz mmm zz mmmmm fj zz m zz mmm mm
26
CHAPTER 1. LINEAR ALGEBRA OVER A RING
Theorem 1.5.3. (i) The projective limit of the projective system {Ni , vij } is the Amodule lim Ni = {{xi }i ∈ ← −
i j i
Ni ; vij (xj ) = xi for all i ≤ j}.
− The maps vi : lim Nj → Ni are the natural ones. ← − (ii) The inductive limit of the inductive system {Mi , uij } is the Amodule lim Mi = ( − →
i
Mi )/N
i∈I
where N is the submodule of i∈I Mi generated by {xi − uji (xi ); xi ∈ Mi , i ≤ j}. The maps ui : Mi → lim Mj are the natural ones. − − →
j
Note that if I is discrete, then lim Mi = i Mi and lim Ni = i Ni . ← − − → i i The proof is straightforward. The universal properties on the projective and inductive limit are better formulated by the isomorphisms which characterize lim Ni and lim Mi : − → ← −
i i
(1.16) (1.17)
∼ lim Hom (L, Ni ), Hom A (L, lim Ni ) − → A ← − ← −
i i
∼ lim Hom (Mi , L). Hom A (lim Mi , L) − → A − → ← −
i i
There are also natural morphisms (1.18) (1.19) lim Hom A (L, Mi ) → Hom A (L, lim Mi ) − − → − →
i i
lim Hom A (Ni , L) → Hom A (lim Ni , L). − − → ← −
i i
One should be aware that morphisms (1.18) and (1.19) are not isomorphisms in general (see Example 1.5.7 below). Proposition 1.5.4. Let Mi − Mi − Mi be a family of exact sequences of → → Amodules, indexed by the set I. Then the sequences Mi → − Mi → − Mi and
i i fi gi
i
i
Mi → −
i
Mi → −
Mi
i
are exact.
1.5. LIMITS The proof is left as an (easy) exercise. Proposition 1.5.5.
g
27
(i) Consider a projective system of exact sequences of
fi gi f i
− Amodules: 0 → Ni − Ni − Ni . Then the sequence 0 → lim Ni → − → → − ← − lim Ni → lim Ni is exact. − ← − ← −
i i gi
(ii) Consider an inductive system of exact sequences of Amodules: M i − → Mi − Mi → 0. Then the sequence lim Mi → lim Mi → lim Mi → 0 is → − − − − − → − → − → i i i exact.
f g
fi
Proof. (i) Since lim Ni is a submodule of i Ni , the fact that f is injective ← − i follows from Proposition 1.5.4. Let {xi }i ∈ lim Ni with g({xi }i ) = 0. Then ← − i gi (xi ) = 0 for all i, and there exists a unique xi ∈ Ni such that xi = fi (xi ). One checks immedialtely that the element {xi }i belongs to lim Ni . ← − i (ii) Let L be an Amodule. The sequence 0 → Hom A (lim Mi , L) → Hom A (lim Mi , L) → Hom A (lim Mi , L) − − − − → − → − →
i i i
is isomorphic to the sequence 0 → lim Hom A (Mi , L) → lim Hom A (Mi , L) → lim Hom A (Mi , L) − − − ← − ← − ← −
i i i
and this sequence is exact by (i) and Proposition 1.3.1. Then the result follows, again by Proposition 1.3.1. q.e.d. One says that “the functor lim is right exact”, and “the functor lim is − → ← − left exact”. We shall give a precise meaning to these sentences in Chapter 2. Remark 1.5.6. (i) If all Mi ’s are submodules of a module M , and if the maps uji : Mi → Mj , (i ≤ j) are the natural injective morphisms, then − lim Mi i Mi . − → i − (ii) If all Mi ’s are submodules of a module M , and if the maps vij : Mj → Mi , (i ≤ j) are the natural injective morphisms, then lim Mi Mi . i − ←
i
Example 1.5.7. Let k be a commutative ring and consider the kalgebra A := k[x]. Denote by I = A · x the ideal generated by x. Notice that
(i) For p ≤ n there are monomorphisms upn : k[x]≤p k[x]≤n which deﬁne an inductive system of kmodules. hence the morphism − − → i (1. (ii) For p ≤ n there are epimorphisms vpn : A/I n A/I p which deﬁne a projective system of Amodules whose projective limit is k[[x]].8. L). − → n Notice that idk[x] ∈ lim Hom k (k[x].5. (iv) We thus have a projective system of complexes of Amodules L• : 0 → I n → A → A/I n → 0. Consider the chain of isomorphisms Hom k (N ⊗A lim Mi .3.20) lim(N ⊗A Mi ) → N ⊗A lim Mi . The morphism (1. L) ← − i i lim Hom A (Mi . Proof. L)) − → i lim Hom k (N ⊗A Mi . uji } be an inductive system of Amodules. . − − − − n Taking the projective limit. we get the complex 0 → 0 → k[x] → k[[x]] → 0 − − − − which is no more exact.d.28 CHAPTER 1.18) is not an isomorphism in general. k[x]≤n ). the ring of formal series with coeﬃcients in k.20) is an isomorphism. (iii) For p ≤ n there are monomorphisms I n I p which deﬁne a projective system of Amodules whose projective limit is 0. One has the isomorphism k[x] = lim k[x]≤n . Then the result follows from Proposition 1. where k[x]≤n denotes the kmodule consisting of polynomials of degree less than or equal to n.6. − − → − → i i Proposition 1. Let L be a kmodule. LINEAR ALGEBRA OVER A RING A/I n+1 k[x]≤n . L)) ← − i Hom k (lim(N ⊗A Mi ). Hom k (N. This shows that the morphism − → n (1. The family of morphisms Mi → lim Mi deﬁnes the family of morphisms − − → i N ⊗A Mi → N ⊗A lim Mi .e. Tensor products and inductive limits Let {Mi . − → q. Hom k (N. N a right Amodule. L) − → i Hom A (lim Mi .
e. q. Ix is ﬁltrant. Example 1. Hence. (i) follows from the fact that x ∈ N ∩ Mi if and only if there exists k ≥ i with uki (xi ) = 0. Setting xi = j∈J uij (xj ). xi ∈ Mi . LIMITS Filtrant limit Lemma 1. Such a germ is an equivalence class (U. Since I is ﬁltrant. V ⊂ U deﬁne an inductive − system of Cvector spaces indexed by Ix .3 (ii). (ii) Let x ∈ M . One sets (1.x = lim C 0 (U ).10.5.5. Let X be a topological space. N is a submodule of ⊕i Mi . We keep the notations of Theorem 1.d. Let N denote the subset of ⊕i Mi consisting of ﬁnite sums j∈J xj . Conversely.1. Given U and V in Ix . let J ⊂ I be a ﬁnite set and let x = j∈J xj ∈ N . We endow Ix with the order: U ≤ V if V ⊂ U . i ≤ j}. − → U ∈Ix 0 An element ϕ of CX. Therefore. N denotes the submodule of ⊕i Mi generated by the elements {xi − uji (xi ). Then there exists i ∈ I and xi ∈ Mi with ui (xi ) = x. Then ui (xi ) = 0 ⇔ there exists k ≥ i with uki (xi ) = 0.5. ϕU )/ ∼ with U a neighborhood of x.9. Then xj = j∈J j∈J xj − ukj (xj ) j∈J = j∈J (xj − ukj (xj )) ∈ N . Then x= j∈J uk uij (xj ) = ui ( j∈J uij (xj )). and setting W = U ∩ V .21) 0 CX. There exist a ﬁnite set J ⊂ I and xj ∈ Mj such that x = j∈J uj (xj ).5. xj ∈ Mj such that there exists k ≥ j for all j ∈ J with j∈J ukj (xj ) = 0. Assume I is a ﬁltrant ordered set and let M = lim Mi . x ∈ X and denote by Ix the set of open neighborhoods of x in X. The inclusion N ⊂ N is obvious since uji (xi ) + ujj (−uji (xi )) = 0. − → i 29 (i) Let xi ∈ Mi . Let us show that N = N . Choose i with i ≥ j for all j ∈ J. we have U ≤ W and V ≤ W . The restriction maps C 0 (U ) → C 0 (V ). ϕU a . the result follows.x is called a germ of continuous function at 0. (ii) Let x ∈ M . Proof. Denote by C 0 (U ) the Cvector space of complex valued continuous functions on U .
Let us give another criterion in order that the projective limit of an exact sequence remains exact.12. This is a particular case of the socalled “MittagLeﬄer” condition (see [15]). a germ of function is zero at x if this function is identically zero in a neighborhood of x. ϕU ) ∼ 0 if there exists a neighborhood V of x with V ⊂ U such that the restriction of ϕU to V is the zero function. Proposition 1. − The MittagLeﬄer condition Recall (Proposition 1. Then the natural morphism lim Mi → lim Mj is an isomorphism. . Let x ∈ lim Mi with g(x) = 0. Then − the natural morphism lim Mj → lim Mi is an isomorphism. − → − → j∈J i∈I (ii) Let {Mi . assume I = {0.5.d. (i) Let {Mi . vji } be a projective system of Amodules indexed by I. Then the inductive limit of the inductive system u10 : M0 → M1 is M1 . There exists xi ∈ Mi with ui (xi ) = − → i x. and there exists j ≥ i such that uji (gi (xi )) = 0.e. and the projective limit of the − projective system v01 : M1 → M0 is M1 . Then x = uj (xj ) satisﬁes f (x ) = f (uj (xj )) = uj fj (xj ) = uj uji (xi ) = x. − ← − ← − i∈I j∈J The proof is left as an exercise. 1} with 0 < 1. Hence. Hence gj (uji (xi )) = uji (fi (xi )) = 0. which implies that there exists xj ∈ Mj such that uji (xi ) = fj (xj ). uij } be an inductive system of Amodules indexed by I.11.30 CHAPTER 1. Proof. Then the → → sequence lim Mi → lim Mi → lim Mi − − − → − → − → i i i f g is exact. LINEAR ALGEBRA OVER A RING continuous function on U . and (U.4) that a product of exact sequences of Amodules is an exact sequence.5. Consider an inductive system of exact sequences of fi gi Amodules indexed by a ﬁltrant ordered set I: Mi − Mi − Mi . In particular. q. Proposition 1.5. Assume J ⊂ I and assume that I is ﬁltrant and J is coﬁnal to I.
. any element of M (j) may be written uniquely as a sum m= I=j mI ⊗ e I . . . n}. j ≤ n. vp . Denote by (e1 . One sets 0 L = k. For a subset I ⊂ {1. Assume that for each n. (Recall the notation [a. .) Note that 1 L L and n L k. Let − −1 −1 −1 xn−1 ∈ gn−1 (xn−1 ). n}. Let 0 → {Mn } − {Mn } − {Mn } → 0 be an exact − → → − sequence of projective systems of Amodules indexed by N. . ϕj ] = 0.) Set M (j) = M ⊗ j k n . . . . . ϕn ) be n endomorphisms of M over A which commute with one another: [ϕi . 1 ≤ i. .d. Then the sequence − 0 → lim Mn → lim Mn → lim Mn → 0 − − − − ← − ← − ← − n n n f g is exact.1. Hence. en ) is a basis of L and I = {i1 < · · · < ij } ⊂ {1. one sets e I = e i1 ∧ · · · ∧ e ij . . Let {xp }p ∈ lim Mn . Let us denote for short by vp the morphisms Mp → Mp−1 which deﬁne − the projective system {Mp }. . If (e1 . . . Let M be an Amodule and let ϕ = (ϕ1 . en ) the canonical basis of k n .6 Koszul complexes If L is a ﬁnite free kmodule of rank n. . −1 Then we can choose xn ∈ gn (xn ) inductively such that vn (xn ) = xn−1 . . Hence xp ∈ Mp . and similarly for vp .13. KOSZUL COMPLEXES fn gn 31 Proposition 1.e. Hence vn (xn ) − xn−1 = fn−1 (xn−1 ). Proof. ← − n −1 We shall ﬁrst show that vn : gn (xn ) → gn−1 (xn−1 ) is surjective. q. The family of eI ’s with I = j is a basis of the free module j L.6. . . Take xn ∈ gn (xn ). 1. one denotes by j L the kmodule consisting of jmultilinear alternate forms on the dual space L∗ and calls it the jth exterior power of L. . Hence M (0) = M and M (n) M . k). b] := ab − ba. and vp (xp ) = xp−1 . the map Mn+1 → Mn is surjective. By the hypothesis fn−1 (xn−1 ) = fn−1 (vn (xn )) for some xn and thus vn (xn − fn (xn )) = xn−1 . Then gn−1 (vn (xn ) − xn−1 )) = 0. . .5. one denotes by I its cardinal. (Recall that L∗ = Hom k (L.
ϕ)) 0 for j = n.2. Then ϕ2 ◦ ϕ1 (x ) = ϕ2 (y) = ϕ1 (z) = ϕ1 ◦ ϕ2 (x ). Using the commutativity of the ϕi ’s one checks easily that d ◦ d = 0. M/(ϕ1 (M ) + · · · + ϕn (M )). one says (ϕ1 . (ii) Assume that (ϕ1 . The proof will be given in Section 4. ϕj is surjective as an endomorphism of Ker ϕ1 ∩ . LINEAR ALGEBRA OVER A RING One deﬁnes d ∈ Hom A (M (j) . . We look for x ∈ M solution of ϕ1 (x) = y.6. ϕj is injective as an endomorphism of M/(ϕ1 (M ) + · · · + ϕj−1 (M )). ϕ2 is surjective on Ker ϕ1 . . (ii) If for each j. ϕn ) is a regular sequence. ϕ2 (t) = z−ϕ2 (x ). z = ϕ2 (t + x ) and x = t + x is a solution to our problem. . .1. d Deﬁnition 1. . 1 ≤ j ≤ n.32 CHAPTER 1. .d. ϕ2 (x)). called a Koszul complex and denoted K • (M. one says (ϕ1 . (i) Assume that (ϕ1 . the cohomology of this complex gives the kernel and cokernel of ϕ1 . Here. Proof. . . − − − − When n = 1. ∩ Ker ϕn . More generally. Hence we consider the complex: 0→M →M ×M →M →0 − − − − where d(x) = (ϕ1 (x). ϕ)) H n (K • (M. Then H j (K • (M. 1 ≤ j ≤ n. ϕ2 (x) = z. ϕ)) 0 for j = 0. ϕn ) is a coregular sequence. (i) If for each j. . . . Hence y = ϕ1 (t+x ). Then H j (K • (M. z) ∈ M × M with ϕ2 (y) = ϕ1 (z). Hence we get a complex. z) = ϕ2 (y) − ϕ1 (z) and we assume ϕ1 is surjective on M . Thus ϕ1 (z − ϕ2 (x )) = 0 and there exists t ∈ M with ϕ1 (t) = 0. . Theorem 1.e. ϕn ) is a coregular sequence. we restrict ourselves to the simple case n = 2 for coregular sequences. q. d(y. . ϕ): 0 → M (0) → · · · → M (n) → 0. . ∩ Ker ϕj−1 . . First choose x ∈ M with ϕ1 (x ) = y. .6. ϕ)) Ker ϕ1 ∩ . Let (y. . ϕn ) is a regular sequence. . M (j+1) ) by: n d(m ⊗ eI ) = i=1 ϕi (m) ⊗ ei ∧ eI and extending d by linearity.2. . H 0 (K • (M. . d d .
. ·∂n ) is a regular sequence on W (considered as an W module) and we get the Koszul complex: 0 → W (0) → · · · → W (n) → 0 − − − − where: δ( I n δ aI ⊗ e I ) = j=1 I aI · ∂j ⊗ e j ∧ eI . introduce ej e I = 0 if j ∈ {i1 . For I = (i1 . . . .5. (i) Denote by xi · the multiplication by xi in A. . . ik } (−1)l+1 eIˆ := (−1)l+1 ei1 ∧ . . . Hence the Koszul complex is exact except in degree n where its cohomology is isomorphic to k. . Then (·∂1 . . Remark 1. Let k be a ﬁeld of characteristic 0 and let A = k[x1 . we recognize the “de Rham complex”.6. . ∂n ·) is a coregular sequence. . . . . . This complex is exact except in degree n where its cohomology is isomorphic to k[x] (see Exercise 1.4. (ii) Denote by ∂i the partial derivation with respect to xi . Writing dxj instead of ej . xn ]. Example 1. Let W = Wn (k) be the Weyl algebra introduced in Example 1. . and the above complex is exact except in degree 0 where its cohomology is isomorphic to k. . This is a klinear map on the kvector space A.6. .6. and denote by ·∂i the multiplication on the right by ∂i .6.2.1. KOSZUL COMPLEXES 33 Example 1. . . One may also encounter coKoszul complexes. The sequence (x1 ·. .3). The sequence (∂1 ·. . considered as an Amodule.1. . ∧ eil ∧ .3. . . We get the complex: 0 → A(0) → · · · → A(n) → 0 − − − − where: d( I n d aI ⊗ e I ) = j=1 I xj · a I ⊗ e j ∧ e I . . Hence we get a Koszul complex 0 → A(0) → · · · → A(n) → 0 − − − − where: d( I n d d aI ⊗ e I ) = j=1 I ∂j (aI ) ⊗ ej ∧ eI . . ik ). xn ·) is a regular sequence in A. ∧ eik if eil = ej l .
∂n ) − is a coregular sequence and calculate H j (K • (O. . ϕ2 = ·∂2 and calculate its cohomology. Let k be a ﬁeld of characteristic 0.3. Prove that ψ = (·∂1 . (ii) Prove that a module P is projective if and only if it is a direct summand of a free module (i. . .1. xn ] and the klinear map ∂i : O → O (derivation with respect to xi ). . − the xi ·’s are morphisms of right W modules). Prove that λ = (∂1 . . . (iii) Deduce that projective modules are ﬂat. Prove that ϕ = (x1 ·. . Exercise 1. . .4. ϕ) : 0 → M (n) → · · · → M (0) → 0.34 CHAPTER 1. ψ)). and we get the complex: K• (M. (i) Denote by xi · : W → W the multiplication on the left by xi on W (hence. . .2.∧ eik . Consider two complexes of Amodules M1 → M1 → M1 and − − M2 → M2 → M2 . there exists a module K such that P ⊕ K is free). ϕ)). n) \ I and εI is the ˆ ˆ signature of the permutation which sends (1. λ)). . . . . (iii) Now consider the left Wn (k)module O := k[x1 . . . LINEAR ALGEBRA OVER A RING where ei1 ∧ .e. Deﬁne δ by: n δ(m ⊗ eI ) = ϕj (m)ej eI . . . . . (i) Prove that a free module is projective and ﬂat. Exercises to Chapter 1 Exercise 1.. ·∂n ) is a regular sequence and calculate H j (K • (W. . Construct the Koszul complex associated to ϕ1 = ·x1 . − − Exercise 1. . . Let A = W2 (k) be the Weyl algebra in two variables. W := Wn (k) the Weyl algebra in n variables. . . Prove that the two sequences are exact if and only if the − − sequence M1 ⊕ M2 → M1 ⊕ M2 → M1 ⊕ M2 is exact. . (ii) Denote ·∂i the multiplication on the right by ∂i on W . Consider the isomorphism j n−j δ ∗: ∼ kn − → kn ˆ which associates εI m ⊗ eI to m ⊗ eI . up to a sign. . smaller than any j ∈ I). − − − − This complex is in fact isomorphic to a Koszul complex. xn ·) is a regular sequence and calculate H j (K • (W. ∗ interchanges d and δ. where I = (1. n) to I I (any i ∈ I is ˆ Then.∧ eil ∧ . . .∧ eik means that eil should be omitted in ei1 ∧ . . Exercise 1. j=1 Here again one checks easily that δ ◦ δ = 0.
Let I be a ﬁltrant ordered set and let Ai . y · xn t = 0. Prove the isomorphism ∼ lim(Ni ⊗Ai Mi ) − lim Ni ⊗A lim Mi . then P ∨ is left Ainjective. → − → − → − → i i i . Mi ) be an inductive system of right (resp. Let k be a ﬁeld. left) Ai modules. A = k[x. where the action of x ∈ A is the usual one and the action of y ∈ A is deﬁned by y · xn tj+1 = xn tj for j ≥ 1. (iii) Prove that if P is a right projective Amodule.6. Deﬁne the endomorphisms of M .Exercises to Chapter 1 35 Exercise 1. and replace M with M ∨∨ . k ≥ j such that uki (xi ) = ukj (yj ). N ∈ Mod(Z). ϕ). − → i Exercise 1.15). Calculate the cohomology of the Kozsul complex K • (M. If M is a Zmodule. (iv) Let M be an Amodule. − → i (ii) Deﬁne the notion of an inductive system Mi of Ai modules. Let M = Mi / ∼ where denotes the settheoretical disjoint union and ∼ is the relation Mi xi ∼ yj ∈ Mj if there exists k ≥ i. i ∈ I be an inductive sytem of kmodules indexed by I. Exercise 1. − (Hint: (iii) Use formula (1.) Exercise 1. − → i (iii) Let Ni (resp. and deﬁne the Amodule lim Mi . (iv) Prove that M → M ∨∨ is an injective map using (ii). y] and consider the Amodule M = i≥1 k[x]ti .8. N ) → Hom Z (N ∨ . Prove that M is naturally a kmodule and is isomorphic to lim Mi . i ∈ I be an inductive sytem of rings indexed by I. (i) Prove that Q/Z is injective in Mod(Z).5. set M ∨ = Hom Z (M. Prove that there exists an injective Amodule I and a monomorphism M → I. M ∨ ) is injective for any − M. (i) Prove that A := lim Ai is naturally endowed with a ring structure. (ii) Prove that the map Hom Z (M. Q/Z). Let I be a ﬁltrant ordered set and let Mi . ϕ1 (m) = x · m and ϕ2 (m) = y · m.7.
LINEAR ALGEBRA OVER A RING .36 CHAPTER 1.
(iii) for any X.1 Categories and functors Deﬁnition 2. there exists idX ∈ Hom (X. [4]. [10]. ﬁltrant categories and exact functors. X). Y ) and g ∈ Hom C (Y. Y )× Hom C (Y. the morphisms from X to Y. X) such that for all f ∈ Hom C (X. Hom C (X. Z ∈ Ob(C).1. Some references: [21]. we prove the Yoneda Lemma. Y ∈ Ob(C). a set Hom C (X. f ◦ idX = f . Then we construct inductive and projective limits in categories by using projective limits in the category Set of sets and give some examples. Z) → Hom C (X. Z). − these data satisfying: (a) ◦ is associative. idX ◦g = g.Chapter 2 The language of categories In this chapter we introduce some basic notions of category theory which are of constant use in various ﬁelds of Mathematics. [19]. (b) for each X ∈ Ob(C). A category C consists of: (i) a family Ob(C). a map. After giving the main deﬁnitions on categories and functors. g) → g ◦ f . [18]. 2. and denoted (f. Y ). (ii) for each X. without spending too much time on this language. We also introduce the notions of representable functors and adjoint functors. [20].1. the objects of C. Y. called the composition. 37 . We also analyze some related notions. in particular those of coﬁnal categories. Special attention will be paid to ﬁltrant inductive limits in the category Set.
For sake of simplicity. A category is discrete if the only morphisms are the identity morphisms. There are some settheoretical dangers. we shall not enter in these considerations here. Remark 2. not a full subcategory.4. Y ) ⊂ Hom C (X. denoted C ⊂ C. Y ). visualized by f. Y ) = Hom C (X. there exists y ∈ Y with (x. f ◦ g1 = f ◦ g2 (resp. the diagonal of X × X. If f : X → Y and g : Y → Z.3. the set of subsets of X × Y. Notation 2. Setf is the full subcategory consisting of ﬁnite sets. Hom C op (X. One introduces the opposite category C op : Ob(C op ) = Ob(C). Of course. Hom C (X. Of course g is unique. A category C is a subcategory of C. Y ∈ C and the composition ◦ in C is induced by the composition in C. One says that C is a full subcategory if for all X. . Y ∈ C . One calls X the source and Y the − target of f . One often writes X ∈ C instead of X ∈ Ob(C) and f : X → Y instead of f ∈ Hom C (X. Y ) = Hom C (Y.38 CHAPTER 2. A category C is a groupoid if all morphisms are isomorphisms. The composition law is deﬁned as follows.1. Two morphisms f and g are parallel if they have the same sources and targets. Note that a set is naturally identiﬁed with a discrete category. Notice that Set is a subcategory of Rel. z) ∈ g}. and one should mention in which “universe” we are working. g ◦ f is the set − − {(x. Y ) for any X. (ii) Rel is deﬁned by: Ob(Rel) = Ob(Set) and Hom Rel (X. an epimorphism) if − for any morphisms g1 and g2 . g : X Y. X) is characterized by the condition in (b). Examples 2.2. and one also denotes it by f . In such a case. One sometimes writes f : X Y or else X → Y (resp. g1 ◦ f = g2 ◦ f ) implies g1 = g2 . THE LANGUAGE OF CATEGORIES Note that idX ∈ Hom (X. z) ∈ X × Z. Y ). A category C is ﬁnite if the family of all morphisms in C (hence. in particular. y) ∈ f. if: Ob(C ) ⊂ Ob(C). A morphism f : X → Y is an isomorphism if there exists g : X ← Y − − ∼ Y or such that f ◦ g = idY .1. f : X Y ) to denote a monomorphism (resp. A morphism f : X → Y is a monomorphism (resp. (y. (i) Set is the category of sets and maps. Hom C (X. idX = ∆ ⊂ X × X. X). an epimorphism). one writes f : X − → −1 simply X Y . g ◦ f = idX . the family of objects) is a ﬁnite set.1. Y ) = P(X × Y ).
F (f ◦ g) = F (f ) ◦ F (g). of a map still − denoted by F : Hom C (X. Examples 2.1. (iii) One says that P is a zeroobject if it is both initial and terminal. one often denotes it by 0. (iv) One denotes by C(Mod(A)) the category whose objects are the complexes of Amodules and morphisms. ≤) a category. ≤) has neither initial nor terminal object. Ob(I) = I. for all X ∈ C. as follows. Deﬁnition 2. P ) {pt}. • ). F (Y )) such that − F (idX ) = idF (X) . X) {pt}.1. • ) instead of Hom Mod(A) ( • .. A contravariant functor from C to C is a functor from C op to C . still denoted by I for short. We shall often use the notations Ab instead of Mod(Z) and Hom A ( • .5.7. In particular Mod(Z) is the category of abelian groups. CATEGORIES AND FUNCTORS 39 (iii) Let A be a ring. one says it is covariant. (ii) One says that P is terminal if P is initial in C op . A functor F : C → C − consists of a map F : Ob(C) → Ob(C ) and for all X. (i) In the category Set. (ii) The zero module 0 is a zeroobject in Mod(A). and is empty otherwise. Hom C (X. (v) One associates to a preordered set (I. In other words. The category of left Amodules and Alinear maps is denoted Mod(A). it satisﬁes F (g ◦ f ) = F (f ) ◦ F (g). the morphism obtained as the composition X → 0 → X − − is still denoted by 0 : X → X. Deﬁnition 2.2. One denotes by Modf (A) the full subcategory of Mod(A) consisting of ﬁnitely generated Amodules. for any object X ∈ C. If C has a zero object. ∅ is initial and {pt} is terminal. morphisms of such complexes. (vi) We denote by Top the category of topological spaces and continuous maps. If one wishes to put the emphasis on the fact that a functor is not contravariant. terminal) objects are unique up to unique isomorphisms. Y ) → Hom C (F (X). . and the set of morphisms from i to j has a single element if i ≤ j. Note that I op is the category associated with I endowed with the opposite order. Y ∈ C.e. (i) An object P ∈ C is called initial if for all X ∈ C. One often denotes by ptC a terminal object in C. (iii) The category associated with the ordered set (Z.1. i. − Note that initial (resp.6. Hom C (P. One often denotes by ∅C an initial object in C. In such a case.1. Let C and C be two categories.
Deﬁnition 2. full.g) / F (Y. and moreover for any morphisms f : X → Y in C. Y ) Examples 2. / F2 (X) F2 (f ) F1 (Y ) θ(Y ) / F2 (Y ) . F (X. − (iii) Let A be a ring. This means that for X ∈ C and X ∈ C . X ) F (Y. − A bifunctor F : C ×C → C is a functor on the product category.1. Y ). − (iv) The forgetful functor f or : Mod(A) → Set associates to an Amodule − M the set M .40 CHAPTER 2. One deﬁnes the product of two categories C and C by : Hom C×C Ob(C × C ) = Ob(C) × Ob(C ) ((X. Then H j ( • ) : C(Mod(A)) → Mod(A) is a functor. Let F1 . g) = (idY . • ) : C → C and F ( • . Y ) × Hom C (X . the diagram below commutes: F (X. bijective).X ) F (X. THE LANGUAGE OF CATEGORIES One denotes by op : C → C op the contravariant functor. Y )) = Hom C (X.8. surjective. (i) Hom C ( • . • ) : C op × C → Set is a bifunctor. • ): Mod(A)op × Mod(A) → Mod(k) are bifunctors. X ). A morphism of functors θ : F1 → F2 is the data for all X ∈ C of a morphism θ(X) : − F1 (X) → F2 (X) such that for all f : X → Y .1. g). g) ◦ (f. fully faithful) if for X.1. resp. X ) F (f. (Y. Mod(A ) × Mod(A) → Mod(k) and − A Hom A ( • . Y ) F (f. Y in C Hom C (X. idX ) = (f. idY ) ◦ (idX . the diagram below commutes: − − F1 (X) F1 (f ) θ(X) In fact. X ) : C → C are − − functors. − op • ⊗ •: (ii) If A is a kalgebra. Deﬁnition 2. g : X → Y in − − C . F (Y )) − is injective (resp.9. (f. (ii) One says that F is essentially surjective if for each Y ∈ C there exists X ∈ C and an isomorphism F (X) Y .g) / F (X.10. (i) One says that F is faithful (resp. and to a linear map f the map f . F2 are two functors from C to C .Y ) F (Y. resp. Y ) → Hom C (F (X). associated with − idC op .
k). Clearly F is fully faithful. CATEGORIES AND FUNCTORS A morphism of functors is visualized by a diagram: F1 41 C F2 11 11 θ ' 7 C Hence. n) with entries in a ﬁeld k (the composition being the usual composition of matrices).1. F associates the induced linear map from k n to k m .1. If F is an isomorphism of categories. Then there is a morphism of functors id → ∗ ◦ ∗ in Fct(Mod(k). − f n F (n) associates k ∈ Mod (k) and to a matrix of type (m.1. Theorem 2.1.15. Let k be a ﬁeld and consider the functor : Mod(k)op → Mod(k). If two categories are equivalent. A functor F : C → C is an equivalence of categories if − there exists G : C → C such that: G ◦ F is isomorphic to idC and F ◦ G is − isomorphic to idC . n).14. There is an weaker notion that we introduce below.2. One may also use the shorter notation (C )C .1. − (ii) We shall encounter morphisms of functors when considering pairs of adjoint functors (see (2. This is why this notion of equivalence of categories plays an important role in Mathematics.13. C ) the category of functors from C to C .5)). In particular we have the notion of an isomorphism of categories. such a situation rarely occurs and is not really interesting. we get a new category. − ∗ V → V = Hom k (V. by considering the family of functors from C to C and the morphisms of such functors. the space of matrices of type (m. To n ∈ N. and since ∗ . Notation 2. G ◦ F (X) = X. Mod(k)). (i) We denote by Fct(C. then there exists G : C → C such that for all − X ∈ C. The functor F : C → C is an equivalence of categories if − and only if F is fully faithful and essentially surjective.n (k).11.12. Examples 2. m) = Mm. Deﬁnition 2. (i) Let k be a ﬁeld and let C denote the category deﬁned by Ob(C) = N and Hom C (n. We shall not give the proof of the following important result below. Examples 2. In practice.1. all results and concepts in one of them have their counterparts in the other one. Deﬁne the functor F : C → Modf (k) as follows.
(ii) let C and C be two categories. A(X)) → − − A(X). There is an equivalence (2. Let C be a category. it is enough to associate − with s ∈ A(X) and Y ∈ C a map from Hom C (Y.2. A). Corollary 2. Set). One checks that ϕ and ψ are inverse to each other. X). Since there is a natural equivalence of categories (2. − X → Hom C ( • .2. where the last map is associated with idX . X) to A(Y ). hence F is essentially surjective. C)) Fct(I. − C → C ∨. X) → Hom Set (A(X). Proof. One constructs the morphism ϕ : Hom C ∧ (hC (X). There are equivalences (2. it is isomorphic to k n for some n. It is deﬁned by the chain of maps Hom C (Y.e. A) → A(X) by the − chain of morphisms: Hom C ∧ (hC (X). . Fct(J. (The Yoneda lemma.2. X) X → Hom C (X.3. C) Fct(J.2. One deﬁnes the categories and the functors hC kC : : C → C ∧. C)).2 The Yoneda Lemma C ∧ = Fct(C op . q. A(Y )) → A(Y ) where − − the last map is associated with s ∈ A(X). 2. J and C be categories. Setop ).3) C∨ C op. we shall concentrate our study on C ∧ .2) Fct(I × J. (C )op ). (iii) Let I. functorial with respect to X and A. Fct(I. In conclusion. C ∨ = Fct(C op . Proposition 2.1. A) A(X). Deﬁnition 2. A) → Hom Set (Hom C (X. there is an isomorphism Hom C ∧ (hC (X). THE LANGUAGE OF CATEGORIES any ﬁnite dimensional vector space admits a basis.42 CHAPTER 2. The functor hC is fully faithful.op .d. • ).1) Fct(C. C )op Fct(C op .∧. F is an equivalence of categories. To construct ψ : A(X) → Hom C ∧ (h(X)C .) For A ∈ C ∧ and X ∈ C.
(See Exercise 2. functorially in Y ∈ C.e. Then f is an isomorphism. one has Hom C ∧ (hC (X). (i) One says that a functor F from C op to Set is representable if there exists X ∈ C and an isomorphism if F hC (X)) in C ∧ . a functor G : C → Set is representable if there exists X ∈ C − such that G(Y ) Hom C (X. Hence.2.2 (ii). X) − Hom C (Z. N. Proof.4. if there exists an isomorphism F (Y ) Hom C (Y. h(Y )) Hom C (X. (ii) Assume that for any Z ∈ C. Y ) is → bijective. Such an object X is called a representative of F . Denote by B(N × M.5. Let C be a category and let f : X → Y be a morphism in − C. Then f is an isomorphism. L be three kmodules. Representable functors provides a categorical language to deal with universal problems.2.2. Y ). q.e. ◦f f◦ .2. (i) Assume that for any Z ∈ C. Z) is → bijective. Let k be a commutative ring and let M. hC (f ) : hC (X) → hC (Y ) is an isomorphism in − ∧ C . Corollary 2. Since hC is fully faithful. (ii) Similarly. one may consider C as a full subcategory of C ∧ . Let us illustrate this by an example. One calls hC the Yoneda embedding. in other words. 43 hC (Y )(X) = q. Y ). this implies that f is an isomorphism. L) the set of kbilinear maps from N × M to L. L) Hom k (N ⊗ M.2. the map Hom C (Y.) (ii) follows by replacing C with C op . functorially in Y ∈ C. For X and Y in C. since F (L) = B(N × M. L) is representable by N ⊗k M . THE YONEDA LEMMA Proof. Z) − Hom C (X. the map Hom C (Z. X). Example 2.d.d.6. L). (i) By the hypothesis. It is important to notice that the isomorphisms above determine X up to unique isomorphism. Representable functors Deﬁnition 2. or. Then the functor F : L → B(N × M.
we have morphisms X → G ◦ F (X). Z)) tells us that ( • × Y. Z) Hom Set (X. Let K ∈ Mod(k) and let M. The bijection Hom Set (X × Y. . or that G is a right adjoint to F if there exists an isomorphism of bifunctors: (2. Z ∈ Set.3 Adjoint functors Deﬁnition 2. for (M )). Example 2. Hom C ( • . functorial in X ∈ C. (extension of scalars) and for are adjoint. Let X. denote by for : Mod(A) → Mod(k) the “forget− ful functor” which.3. In fact. Let F : C → C and G : C → C be two functors. G( • )).4) gives the isomorphisms Hom C (F ◦ G( • ). functorial in Y ∈ C . Y. Hom Set (Y.1. we have − morphisms of functors (2. G ◦ F ( • )). The isomorphism (2. G(Y ) is a representative of the functor X → Hom C (F (X). Applying the above formula with N = A. In particular. and − morphisms F ◦ G(Y ) → Y . In the preceding situation. Hom Set (Y.3.2. F ( • )) Hom C (G( • ).3.3. tells us that the functors • ⊗ K and Hom (K. G) is a pair of adjoint functors or that F is a left adjoint to G. In other words. G( • )) If G is an adjoint to F . The formula: Hom A (N ⊗ K. • ) from Mod(A) to Mod(A) are adjoint. Y ). Let A be a kalgebra. • ) Hom C ( • .44 CHAPTER 2. Hom (K. M ) Hom A (N. THE LANGUAGE OF CATEGORIES 2. the functors A ⊗ • Hom (K. we get Hom A (A ⊗ K. N ∈ Mod(A). • )) is a pair of adjoint functors. then G is unique up to isomorphism.5) F ◦ G → idC . M )). One says − − that (F. M ) Hence. − Example 2.4) Hom C (F ( • ). − idC → G ◦ F. to an Amodule M associates the underlying kmodule. • ) Hom C (F ( • ).
2. One sets lim β = {{xi }i ∈ ← − β(i). X) is representable. (i) Assume that the functor X → lim Hom C (X. Lemma 2.7) Hom C (X. ← − . Let β : I op → Set be a functor and let X ∈ Set.4 Limits In the sequel. lim β) ← − lim Hom C (X. we get − − op functors from I to Set: Hom C (X. Let C be a category. β) is ← − representable. In particular. ≤) is an ordered set. lim β) − lim Hom Set (X. β(i)). (2. For example. β(s)(xj ) = xi for all s ∈ Hom I (i. X) ∈ Set.1. (2. an inductive system indexed by I is the data of a family (Xi )i∈I of objects of C and for all i ≤ j. j)}.8) Hom C (lim α. We ← − denote by lim α its representative and says that the functor α admits − → an inductive limit in C.2. Assume ﬁrst that C is the category Set and let us consider projective systems. ← − (ii) Assume that the functor X → lim Hom C (α. β(i)) ∈ Set. β).4. Deﬁnition 2. → ← − ← − where Hom Set (X. i → Hom C (α. For X ∈ C. − Consider now two functors β : I op → C and α : I → C. A functor α : I → − op C (resp. There is a − natural isomorphism ∼ Hom Set (X. X) − → lim Hom C (α. if (I. β). In particular.4. X). a morphism Xi → Xj with the natural compatibility − conditions.4. LIMITS 45 2. (2. projective) − system in C indexed by I. β) : I op Hom C (α. We denote by lim β its representative and says that the ← − functor β admits a projective limit in C. X) : I op i → Hom C (X. I will denote a category. i → Hom Set (X.6) i The next result is obvious. β) denotes the functor I op → Set. I the associated category. β : I → C) is sometimes called an inductive (resp.
7). j).9) fi = fj ◦ f (s) for all s ∈ Hom I (i. lim β) ← − ← − Consider a family of morphisms {fi : X → β(i)}i∈I in C satisfying the com− patibility conditions (2. β) ← − ← − Hom C (lim β.8). X). all morphisms fi ’s factorize uniquely through lim β.9). lim β is charac← − ← − terized by the “universal property”: − for all X ∈ C and all family of morphisms {fi : X → β(i)}i∈I (2. lim α) ← − − → Hom C (lim α. One gets: − → and the identity of lim α deﬁnes a family of morphisms − → − ρi : α(i) → lim α. lim β. Therefore.46 CHAPTER 2. This family of morphisms is nothing but an element of lim Hom (X. lim α is character− → − → ized by the “universal property”: for all X ∈ C and all family of morphisms {fi : α(i) → X}i∈I in − C satisfying (2. an element of Hom (X. ← − . − → Similarly. Therefore. j).10) and the identity of lim β deﬁnes a family of morphisms ← − ρi : lim β → β(i). THE LANGUAGE OF CATEGORIES When C = Set this deﬁnition of lim β coincides with the former one. lim α) − → − → Consider a family of morphisms {fi : α(i) → X}i∈I in C satisfying the com− patibility conditions (2.12) in C satisfying (2. − → lim Hom C (α. X). One gets: ← − (2. X).4. β(i)).11) fj = fi ◦ f (s) for all s ∈ Hom I (i. − ← − lim Hom C (lim β. ← − i hence by (2. Notice that both projective and inductive limits are deﬁned using projective limits in Set.1. all morphisms fi ’s factorize uniquely through lim α. an element of Hom (lim α. Assume that lim α exists in C.11). This family of morphisms is nothing but an element of lim Hom (α(i). in ← − view of Lemma 2. ← − i hence by (2. assume that lim β exists in C.
14) lim (α ◦ ϕ) → lim α.2.3.2) and consider a bifunctor α : I × J → C. Recall the equivalence of categories (2. the category C J admits inductive limits indexed by I. j)) − − → → j i lim β(i. Then for any category J. if α : I → C J is a functor.4. j) − → i. − → − → op J Similarly. LIMITS Inductive and projective limits are visualized by the diagrams: α(i) α(s) 47 α(j) qq qq i f q ρi qqq q# )/ lim α . denote by α(Y ) : I → C the − functor i → α(i)(Y ). Then lim α ∈ C J is given by − → (lim α)(Y ) = lim α(Y ). j) ← − i. O lllxxx fi lll lll xxxx i ρ lll l x lll / β(s) lim βp ← − ppp ρj pp pp fj # ) β(j) If ϕ : J → I.12).17) (2. if β : I → C is a functor. − llllll5 X → ww ρj ww lll w wwllllllj f ww l ll X β(i) ll5 . then β deﬁnes a functor − op op βJ : I op → C J and a functor βI : J op → C I and one has the isomorphisms − − lim β ← − lim α(i. − ← − ← − This follows immediately of (2.j (2.15) lim α lim (lim αJ ) lim (lim αI ). ← − ← − The proof is obvious. Y ∈ J.j lim lim(β(i. ← ← − − i j . Let I be a category and assume that C admits inductive limits indexed by I. if β : I × J → C is a bifunctor.18) lim lim βJ ← ← − − lim lim βI . Proposition 2. j)) ← ← − − j i lim lim(β(i. we have natural − − − morphisms: (2. − − − One easily checks that (2.13) (2.16) In other words: (2. α : I → C and β : I op → C are functors. − − → → i j lim(lim(α(i. j)). − → − − → → − − → → op op Similarly. then its − inductive limit is deﬁned as follows. j)). Y ∈ J.4.10) and (2. ← ← − − lim lim(α(i. For Y ∈ J. then lim β ∈ C J is given by − ← − (lim β)(Y ) = lim α(Y ). Moreover. − − → − → lim (β ◦ ϕ) ← lim β. It deﬁnes a functor αJ : I → C J as well as a functor αI : J → C I .
Y ). Deﬁnition 2. writing α(i) = Xi or β(i) = Xi . This follows immediately of (2. (i) When the category I is discrete. THE LANGUAGE OF CATEGORIES If every functor from I to C admits an inductive limit. Hom C (Xi . ﬁnite inductive) limits. and in this case ← − lim α ptC . we get for Y ∈ C: Hom C (Y. . one says that C admits inductive limits indexed by I. ﬁnite categories I).5 Examples Empty limits. Xi ).12). one says that C admits inductive (resp. respectively. respectively. i Xi ) i Hom C (Y. denoted and . then lim α − → α(io ) lim β ← − β(io ). ← − Terminal object If I admits a terminal object. If I is the empty category and α : I → C is a functor. Y ) (ii) If I is discrete with two objects.48 CHAPTER 2. inductive and projective limits are called coproduct and products. then lim α exists in C if − − → and only if C has an initial object ∅C . and similarly when replacing I with I op and inductive limits with projective limits.10) and (2.1. If this property holds for all categories I (resp.5. 2. − → lim α exists in C if and only if C has a terminal object ptC . and in this case lim α ∅C . say io and if α : I → C and β : I op → C are − − functor. Similarly. i Hom C ( i Xi . Sums and products Consider a discrete category I. Hence. a functor I → C is the data of two − objects X0 and X1 in C and their coproduct and product (if they exist) are usually denoted by X0 X1 and X0 × X1 .
5. One also writes X (I) and X I instead of X I and X I . g : X0 X1 in C. Z)I . Hom Set (I. from) X factors uniquely through X0 X1 (resp.3. Hom Set (I. g : X0 / / X1 In the sequel we shall identify such a functor with the diagram (2. Q Q X I ). visualized by • / /• A functor α : I → C is characterized by two parallel arrows in C: − (2. Hom Set (X. EXAMPLES 49 Hence. X). . Example 2. X0 X1 is the disjoint union and X0 × X1 is the product of the two sets X0 and X1 .2. g). X. The coproduct and product of two objects are visualized by the diagrams: X0 ss ss ss ss s$ ) X:0 X1 jjj5/ X X1 j u jjjj uu uu jjjjjjj uu uu jj jjjj 5X u jjjj: 0 jjjj uuu u jjj jjjj uuuu jjjj jj / X0 × X1 s ssss ss s s$ * X X1 In other words.19) f. Z) I × X. any pair of morphisms from (resp. is an inductive limit of this functor. Consider two parallel arrows f. Deﬁnition 2. respectively. Hom Set (X. to) X0 and X1 to (resp.2. Cokernels and kernels Consider the category I with two objects and two parallel morphisms other than identities. In the category Set. Z ∈ Set: X (I) XI Hom Set (I × X. one denotes by α or − α(i) its coproduct and one denotes by α or i∈I α(i) its product. X0 ×X1 ).19). if α : I → C is a functor and I is discrete. one simply denotes‘this limit by X I (resp. if it exists. we have for I. (i) A coequalizer (one also says a cokernel).5. i∈I ‘ If α(i) = X for all i ∈ I. Z)). If C is the category Set.5. It is denoted by Coker(f.
One can show that conversely. A cokernel (resp.20) k is an epimorphism. Dually. the equalizer K is visualized by the diagram: K` h / X0 O X / / X1 {= { { h {{ { {{ g f and (2.21) h is a monomorphism. and this implies a = b. Hence h factors uniquely through k.50 CHAPTER 2. is a cokernel (resp. We have seen that coproducts and coequalizers (resp. THE LANGUAGE OF CATEGORIES (ii) An equalizer (one also says a kernel). if it exists. (iii) A sequence X0 X1 → Z (resp. In particular. ﬁnite projective limits) can be obtained as equalizer of products (resp. products and equalizers) are particular cases of inductive (resp. Note that (2. The coequalizer L is visualized by the diagram: X0 g f g gg gg gg h g! ~ / / X1 k / L X which means that any morphism h : X1 → X such that h ◦ f = h ◦ g factors − uniquely through k. inductive limits (resp. equalizer) of X0 X1 . It is denoted by Ker(f. projective) limits. is a projective limit of this functor. a category C admits ﬁnite projective limits if and only if it satisﬁes: . a kernel) of f . Then h ◦ f = a ◦ k ◦ f = a ◦ k ◦ g = b ◦ k ◦ g = h ◦ g. Let f : X → Y − be a morphism in C. a kernel) of f. ﬁnite products). Indeed. 0 : X Y . ﬁnite coproducts) and projective limits (resp. Z → X0 − − X1 ) is exact if Z is isomorphic to the coequalizer (resp. ﬁnite inductive limits) can be obtained as coequalizer of coproducts (resp. if it exists. (iv) Assume that the category C admits a zeroobject 0. Ker(f )). b : L X such that a ◦ k = b ◦ k = h. consider a pair of parallel arrows a. It is denoted Coker(f ) (resp. g).
replacing a terminal object by an initial object.5. x) ∈ S. if I is a category and α : I → Set is a functor. More precisely. EXACT FUNCTORS (i) C admits a terminal object.5. the coproduct in Set is the disjoint union.6. (iii) for any parallel arrows in C. x)}i∈I. there is a natural morphism lim (F ◦ α) → F (lim α) (resp. Y ∈ Ob(C). The category Set admits inductive limits. One says that F commutes with such limits if for any α : I → C.d. β : I op → C and F : C → C be functors.e. Let S ∈ Set. then − lim α − → ( i∈I α(i))/ ∼ where ∼ is the equivalence relation generated by x ∼ y ∈ α(j) if there exists s : i → j with α(s)(x) = y. the product X × Y exists in C. α(i) In particular. x) = p(j. Notation 2. f. Recall that if C and C admit inductive (resp. y) if there exists s : i → j with α(s)(x) = y}. Proof. − − → − → F (lim β) → lim (F ◦ β)).x∈α(i) . − → − → . p(i. (ii) for any X.4. {{p(i. if C admits ﬁnite projective limits. − = . By the deﬁnition of the projective limit in Set we get: lim Hom (α. There is a similar result for ﬁnite inductive limits. In the category Set one uses the notation . 51 Moreover. p(i. − ← − ← − Deﬁnition 2. (ﬁnite) products and kernels.2. g : X Y . projective) limits indexed by I. C and C be categories and let α : I → C. Proposition 2. − q.6.6 Exact functors − − − Let I. − (i) Let I be a category and assume that C admits inductive limits indexed by I. S) ← − The result follows. a functor F : C → C commutes − with such limits if and only if it commutes with the terminal object.5. Let F : C → C be a functor. − lim (F ◦ α) exits in C and is represented by F (lim α). the equalizer exists in C. rather than 2. products by coproducts and equalizers by coequalizers.1.
2.d. THE LANGUAGE OF CATEGORIES Example 2.6. − (i) Assume that C admits ﬁnite projective limits. ← − i (ii) Similarly if I is a category and C admits projective limits indexed by I. • ) does not commute with inductive limit if X is inﬁnite dimensional.e. Let β : I op → C be a projective system indexed by I and let Y ∈ C . One says that F is right exact if it commutes with such limits. (iii) One says that F is exact if it is both left and right exact. Assume that − (i) F admits a left adjoint G : C → C. Let k be a ﬁeld. (ii) Assume that C admits ﬁnite inductive limits. − (ii) C admits projective limits indexed by a category I. Let F : C → C be a functor. one says that F commutes with such limits if for any β : I op → C. Then F commutes with projective limits indexed by I. Let F : C → C be a functor. C = C = Mod(k). ← − i lim Hom C (Y.52 CHAPTER 2. that is. q. β(i)) ← − i Then the result follows by the Yoneda lemma.3.6. If C admits inductive limits indexed by I and F admits a right adjoint. Then the functor Hom k (X.4. One says that F is left exact if it commutes with such limits. lim β(i)) ← − i Hom C ∧ (Y. Deﬁnition 2. F (β(i))) ← − i lim Hom C (G(Y ). . ← − ← − Proof. and let X ∈ C. F (lim β(i))) ← − i Hom C (G(Y ). − lim (F ◦ β) exists and is represented by F (lim β). Proposition 2. Of course there is a similar result for inductive limits.6. F (lim β(i)) ← − i lim F (β(i)). lim F (β(i))). then F commutes with such limits. − One has the chain of isomorphisms Hom C (Y.
g : i k such that h ◦ f = h ◦ g.6. left) adjoint.6. left) exact. (i) follows immediately from (1. − then F is right (resp. − (ii) Let F : C → C be a functor. Proof. q. − − (iii) for any parallel morphisms f. left) exact. If F admits a right (resp. projective) limits indexed by I. The conditions (ii)–(iii) of being ﬁltrant are visualized by the diagrams: i @k i / / j k j .16). Let C be a category which admits ﬁnite inductive and ﬁnite projective limits.d.11.7 Filtrant inductive limits We shall generalize some notions of Deﬁnition 1. (iv) is wellknown and obvious. A category I is called ﬁltrant if it satisﬁes the conditions (i)–(iii) below. (i) I is non empty. (iii) Let I be a category and assume that C admits inductive (resp. Deﬁnition 2. there exists a morphism h : j → − One says that I is coﬁltrant if I op is ﬁltrant.7.5. (iii) Use the isomorphism (2. C) → C (resp.5. FILTRANT INDUCTIVE LIMITS 53 Proposition 2.15) or (2. − ← − (iv) Let I be a discrete category and let A be a ring. Then the functor − : Mod(A)I → Mod(A) is exact. (ii) for any i and j in I. j.e.7.9 and Proposition 1. Then the functor lim : Fct(I.5. − − → op lim : Fct(I .16). there exists k ∈ I and morphisms i → k.17) and (1. C) → C) is right (resp. (ii) is a particular case of Proposition 2.4.2 as well as Lemma 1.5. (i) The functor Hom C : C op × C → Set is left exact. 2.2.1. j → k.
with I ﬁltrant. with I ﬁltrant. − − → (ii) Let i ∈ I and let x and y be elements of α(i) with the same image in lim α.e. Let α : I → Set be a functor.d. − − → The proof is left as an exercise. ≤) is a nonempty directed ordered set. Let α : I → Set be a functor.7. THE LANGUAGE OF CATEGORIES Of course. Corollary 2. (i) The relation ∼ is clearly symmetric and reﬂexive. In other words. q. Then − (i) the relation ∼ is an equivalence relation. Corollary 2. Deﬁne the − − relation ∼ on i α(i) by α(i) xi ∼ xj ∈ α(j) if there exists s : i → k and t : j → k such that α(s)(xi ) = α(t)(xj ).54 CHAPTER 2.4. Then α(w1 )x1 = α(w2 )x2 = α(w3 )x3 . (ii) lim α i α(i)/ ∼. Then there exists i ∈ I such that S is − → contained in the image of α(i) by the natural map α(i) → lim α. − (i) Let S be a ﬁnite subset in lim α. α(t2 )x2 = α(t3 )x3 . We shall ﬁrst study ﬁltrant inductive limits in the category Set.2. w2 = v ◦ u1 ◦ s2 = v ◦ u2 ◦ t2 and w3 = v ◦ u2 ◦ t3 . Let xj ∈ α(ij ).4. Then there exists s : i → j such that α(s)(x) = α(s)(y) in α(j). There exist morphisms visualized by the diagram: i1 /j ? 1 cc cc ÐÐ Ð c ÐÐs2 u1 c c ÐÐ bbt2 bb b t3 / j2 s1 i2 b b i3 ? k1 u2 v / l such that α(s1 )x1 = α(s2 )x2 . then the associated category I is ﬁltrant.7.3. Hence x1 ∼ x3 . Then the functor f or commutes with ﬁltrant inductive − limits. (ii) follows from Proposition 2. j = 1. Let us show it is transitive.5. if (I. if I is ﬁltrant and α : I → Mod(A) is a functor. Let A be a ring and denote by f or the forgetful functor Mod(A) → Set. − → Proof. − → − → i i .7. 2. Set w1 = v ◦ u1 ◦ s1 . Proposition 2. 3 with x1 ∼ x2 and x2 ∼ x3 . and v ◦ u1 ◦ s2 = v ◦ u2 ◦ t2 . then − f or ◦ (lim α(i)) = lim(f or ◦ α(i)).
Corollary 2. Let I be a ﬁltrant category and let ϕ : J → I be a fully − faithful functor. Then lim (α ◦ ϕ) (resp. Assume that − − − lim α (resp. q.4. lim (β ◦ ϕop )) exists in − → ← − − → ← − C and the natural morphism lim (α ◦ ϕ) → lim α (resp. a ﬁnite category J and a func∼ → tor α : I × J op → Set. j) is an isomorphism.d. Remark 2.) belong to a given universe U and that all limits are indexed by U small categories (I. − Note that the hypothesis implies that J is ﬁltrant. we have skipped problems related to questions of cardinality and universes. it is enough to check it − − ← → − ← − − → Proposition 2.e. Then the result follows from − Corollaries 2. one has lim lim α(i.4 and 2. β : I → C) be a functor. Let α : I → C (resp.2. Let A be a ring and let I be a ﬁltrant category. but we should have not. Set) → Set − − → Proof. the reader will assume that all categories (C.7. J. In other − ← − − → − ← → − i j j i words. C etc. lim : Fct(I. One says that J is coﬁnal to I (or that ϕ : J → I is coﬁnal) − if for any i ∈ I there exists j ∈ J and a morphism i → ϕ(j). Let α : I × J op → Mod(A) be a functor. The proof is left as an exercise. Indeed.5. q. In these notes.).7.7.7. j) − lim lim α(i.9. Assume I is ﬁltrant.7. Coﬁnal functors Deﬁnition 2.) . In order to prove that − lim lim α(i. It is enough to prove that lim commutes with equalizers and with − → ﬁnite products. For a ﬁltrant category I. lim β) exists in C. ϕ : J → I is fully faithful and − op J → I is coﬁnal.8. the functor commutes with ﬁnite projective limits. lim β → lim (β ◦ ϕop )) − − − → − → ← − ← − is an isomorphism.7. j). Then the functor lim : Mod(A)I → Mod(A) is exact.7. FILTRANT INDUCTIVE LIMITS 55 Inductive limits with values in Set indexed by ﬁltrant categories commute with ﬁnite projective limits.d.e. Proof. More precisely: Proposition 2.7. − − → i j j i after applying the functor for : Mod(A) → Set. j) → lim lim α(i.7. etc. This veriﬁcation is left to the reader.6. (We do not give the meaning of “universe” and “small” here.
Y M ) ≥ 2π . − End (idC ) = Hom Fct(C. Let C be a category. Y M ) ≤ π. (Hint: if F : Set → Setop were such an equivalence. Prove that the composition law on End (idC ) is commutative.) Exercise 2. an epimorphism) if and only if it is injective (resp. Prove that if F (f ) is an isomorphism. surjective). (ii) C is the category Rel of relations. X) and Hom Setop (F ({pt}). F (X)) when X is a set with two elements. Prove that the categories Set and Setop are not equivalent.5. Now compare Hom Set ({pt}. we may have troubles. Set M = Mor(C). where Mor(C) denotes the “set” of all morphisms in C. (ii) Assume now that F is fully faithful. Exercise 2. (i) Let F : C → C be a faithful functor and let f be a mor− phism in C. Y )M and therefore card(Hom C (X. We have Hom C (X. then F (∅) {pt} and − F ({pt}) ∅. then f is an isomorphism. (i) Prove that in the category Set. or else. Exercise 2. On the other hand. Hom C (X. Y ) has more than one element.4. an epimorphism). Prove that if F (f ) is a monomorphism (resp. that is. a morphism f is a monomorphism (resp. We denote by idC : C → C the identity − functor of C and by End (idC ) the set of endomorphisms of the identity functor idC : C → C.C) (idC . (The remark was found in [9]. Y ∈ C such that Hom C (X. idC ). Mor(C) is not small (in general) in the universe to which C belongs.1.3. otherwise. THE LANGUAGE OF CATEGORIES Let us give an example which show that. Let C be a category which admits products and assume there exist X. the cardinal of the set M . Y M ) Hom C (X. The “contradiction” comes from the fact that C does not admit products indexed by such a big set as Mor(C).2.56 CHAPTER 2. an epimorphism). Exercise 2.) Exercises to Chapter 2 Exercise 2. . the morphism Z → Q is an epimor− phism. and let π = card(M ). (ii) Prove that in the category of rings. then f is a monomorphism (resp. Y M ) ⊂ Mor(C) which implies card(Hom C (X. Prove that the category C is equivalent to the opposite category C op in the following cases: (i) C denotes the category of ﬁnite abelian groups.
Y ) → Hom Set ( i Xi . (i) Let I be a (non necessarily ﬁnite) set and (Xi )i∈I a family of sets indexed by I.) Exercise 2. Show that i Xi is the disjoint union of the sets Xi . (iii) Prove that the map i Hom Set (Xi . (i) Prove that lim : C I → C is a functor. Deduce that if X × Y exists.8. (i → j) ∈ Mor(I) → idX . lim G(i)) ← − i Hom Fct(I op . α) → − − − → Hom C (X. In the category Top. X × Y → X.6. One says that an object X of C is of ﬁnite type (resp. (ii) Does this deﬁnition coincide with the classical one when C denotes the category of commutative algebras? .10. − → (i) Show that this deﬁnition coincides with the classical one when C = Mod(A). deduce the formula (assuming projective limits exist): Hom C (X. (Hint: consider a continuous injective map f : X → Y with dense image. X × Y → Y . Y ∈ C and consider the category D whose arrows are triplets Z ∈ C. G).7. f : Z → X. Let C be a category which admits ﬁltrant inductive limits. Exercise 2. lim α) is injective (resp.9. give an example of a morphism which is both a monomorphism and an epimorphism and which is not an isomorphism. − − Prove that this category admits a terminal object if and only if the product X × Y exists in C. − − → (ii) Prove the formula (for α : I → C and Y ∈ C): − Hom C (lim α(i).C) (∆(X). Y ) − → i Hom Fct(I. ∆(Y )). the morphisms being the natural one. (iii) Replacing I with the opposite category.C) (α. Exercise 2. Let X. i Xi ) and − prove it is injective. bijective). (ii) Construct the natural map i Hom Set (Y. g : Z → Y . of ﬁnite presentation) if for any functor α : I → C with I ﬁltrant. Xi ) → Hom Set (Y. and that in such a case this terminal object is isomorphic to X × Y.Exercises to Chapter 2 57 Exercise 2. Y ) is not in− jective in general. it is unique − − up to unique isomorphism. for a ring A. Exercise 2. the natural map lim Hom C (X. Let I and C be two categories and denote by ∆ the functor from C to C I which. Assume that any functor from I to C − admits an inductive limit. associates the constant functor ∆(X) : I i → X ∈ C. to X ∈ C.
Prove that the sequence X ×Z Y → X − Y f g − − a ← c → b. G(j)) → Hom C (F (j). Here. Y. G : I functors. − − − − Hom C (F (i).C) (F. is the projective limit of β. Exercise 2. Hom C (F (i). i∈I Exercise 2. . Prove the isomorphism: Hom Fct(I.12. Exercise 2. One denotes by “lim” the inductive limit in C ∧ . − ptC . b. Let A ∈ C. c} and two morphisms other than the identities.13. the double arrow is associated with the two maps: i∈I (j →k)∈Mor(I) − Hom C (F (j).58 CHAPTER 2.22) such that πA (∅) πA (S) A × A. Construct a functor (2. G(k)) . G(i)) → Hom C (F (k). visualized by the diagram Let C be a category. G(i)) → Hom C (F (j). − − The ﬁber product X ×Z Y of X and Y over Z. the two morphisms X Y are given by f. Let C be a category and recall that the category C ∧ admits inductive limits. Let I and C be two categories and let F. (i) Assume that C admits products (of two objects) and kernels. Consider the category I with three objects {a. πA ({pt}) A and for S a set with 2 elements. Let k be a − → ﬁeld and let C = Mod(k). A functor β : I op → C is nothing but the data of three − objects X. Prove that the Yoneda functor hC : C → C ∧ does − not commute with inductive limits. g. if it exists. Let C be a category which admits products of two objects and a terminal object.14. THE LANGUAGE OF CATEGORIES Exercise 2. denoted by ptC . G(k)). G(k)). Z and two morphisms visualized by the diagram X → Z ← Y. is exact. (ii) Prove that C admits ﬁnite projective limits if and only if it admits ﬁber products and a terminal object. G(k)) → Hom C (F (j). πA : (Setf )op → C. G(i)) Here. G) Ker i∈I C be two Hom C (F (i).11.
This is the language of additive and abelian categories. Hom C (X. it is natural to look for a common language which avoids to repeat the same arguments. such as ﬁnitely generated Amodules. We also construct complexes associated with simplicial objects in an additive category and give a criterion for such a complex to be homotopic to zero.1. Y ) ∈ Ab. We expose some basic constructions and notions on complexes such as the shift functor. Note that Hom C (X. (v) the category C admits ﬁnite products. Hence. (ii) the composition law ◦ is bilinear.1 Additive categories Deﬁnition 3. 0) = Hom C (0. 59 . (iii) there exists a zero object in C.Chapter 3 Additive categories Many results or constructions in the category Mod(A) of modules over a ring A have their counterparts in other contexts. Note that Hom C (X. the homotopy. In this chapter. the mapping cone and the simple complex associated with a double complex. (iv) the category C admits ﬁnite coproducts. X) = 0 for all X ∈ C. A category C is additive if it satisﬁes conditions (i)(v) below: (i) for any X. Y ∈ C. we give the main properties of additive categories. or sheaves of Amodules. or graded modules over a graded ring.1. Y ) = ∅ since it is a group. 3. etc.
5).1.2. p1 : Z → X and p2 : Z → Y satisfying − − − − (3. By the deﬁnition of a coproduct and a product in a category.1. the category of CBanach spaces and linear continuous maps is additive. Z) Hom C (Z. Z) × Hom C (Y. then C op is additive. Then the conditions (iv). Consider the condition (vi) for any two objects X and Y in C.5) p1 ◦ i1 = idX . The identity of X and the zero morphism Y → X deﬁne the morphism p1 : X ⊕ Y → X satisfying (3. Replacing C with C op . Z). the forgetful functor Mod(A) → Set does not − commute with coproducts. there exists Z ∈ C and morphisms i1 : X → Z. X) × Hom C (Z. (b) Let us assume condition (vi). then Z X ⊕ Y .3. Example 3. Lemma 3. Set h := f ◦ p1 ⊕ g ◦ p2 . This change of notations is motivated by the fact that if A is a ring. we get that these conditions are equivalent to (v) and Z X × Y . Mod(A) and Modf (A) are additive categories.1.1. Let C be a category satisfying conditions (i)–(iii) in Deﬁnition 3. Then h : Z → W − − − satisﬁes h ◦ i1 = f and h ◦ i2 = g and such an h is unique.2) Hom C (X. q. One denotes as usual by X × Y their product. for each Z ∈ C. p2 ◦ i1 = 0. (i) If A is a ring.4). then f = idX⊕Y .1. Proof. (v) and (vi) are equivalent and the objects X ⊕ Y . and calls it their direct sum. If X and Y are two objects of C. (ii) Ban.4) (3.d. To check − (3. i1 ◦ p1 + i2 ◦ p2 = idZ . there is an isomorphism in Mod(Z): (3.60 CHAPTER 3. X × Y and Z are naturally isomorphic. (a) Let us assume condition (iv). Hence Z X ⊕ Y . Hom C (Z. p1 ◦ i2 = 0 p2 ◦ i2 = idY .1) (3. one denotes by X ⊕ Y (instead of X Y ) their coproduct. (c) We have proved that conditions (iv) and (vi) are equivalent and moreover that if they are satisﬁed. i2 : Y → Z. we use the fact that if f : X ⊕ Y → X ⊕ Y satisﬁes f ◦ i1 = i1 and − f ◦ i2 = i2 . ADDITIVE CATEGORIES Notation 3. Y ) Hom C (X ⊕ Y. .e.3) (3. (iii) If C is additive.4. Let W ∈ C and consider morphisms f : X → W and g : Y → W . X × Y ).3). We − − construct similarly the morphism p2 : X ⊕ Y → Y satisfying (3.
Then − F is additive if and only if it commutes with direct sum. One can prove the following Proposition 3. One shall be aware that the bifunctor C op × C → Mod(Z) is separately additive with respect to each of − its argument. − dk−1 dk Let C denote an additive category. COMPLEXES IN ADDITIVE CATEGORIES 61 (iv) Let I be category.1. 3. (v) If C and C are additive. One deﬁnes the notion of a kadditive category by assuming that for X and Y in C.6.1. the category C I of functors from I to C. Generalization: Let k be a commutative ring. Unless otherwise speciﬁed.5. that is. If C is additive. but is not additive as a functor on the product category. Hom C (X. Example 3. Hom C (X. − A morphism of diﬀerential objects f • : X • → Y • is visualized by a commutative diagram: ··· / (ii) A complex is a diﬀerential object (X • .6) → − − · · · → X k−1 − → X k − X k+1 → · · · . One says that F is − additive if for X. d• ) such that such that dk ◦ X dk−1 = 0 for all k ∈ Z. d• ) in C is a sequence of X objects X k and morphisms dk (k ∈ Z): (3. Y ) → Hom C (F (X). for X and Y in C: F (0) F (X ⊕ Y ) 0 F (X) ⊕ F (Y ).2 Complexes in additive categories Deﬁnition 3. Let F : C → C be a functor of additive categories. is additive. (i) A diﬀerential object (X • .3. Xn fn dn X / X n+1 / f n+1 ··· / ··· / Yn dn Y / X n+1 ··· . Y ∈ C. functors between additive categories will be assumed to be additive. F (Y )) is a morphism − of groups.2.2. Y ) is a kmodule and the composition is kbilinear. Let F : C → C be a functor of additive categories. then C × C are additive. Let C be an additive category.1.
−) the full additive subcategory of C(C) consisting of bounded complexes (resp. n >> 0). ADDITIVE CATEGORIES One deﬁnes naturally the direct sum of two diﬀerential objects. One denotes by C(C) the full additive subcategory of Dif f (C) consisting of complexes. (ii) An object X in C(C) is homotopic to 0 if idX is homotopic to zero. One denotes by C∗ (C)(∗ = b. bounded below. n << 0. +. Shift functor Let X ∈ C(C) and k ∈ Z.2. Deﬁnition 3. an invertible functor) of C(C). bounded below. we shall concentrate our study on the category C(C). The shift functor X → X[1] is an automorphism (i. bounded above) if X n = 0 for n >> 0 (resp.2.62 CHAPTER 3. Hence. One deﬁnes the shifted complex X[k] by: (X[k])n = X n+k n+k dn = (−1)k dX X[k] If f : X → Y is a morphism in C(C) one deﬁnes f [k] : X[k] → Y [k] by − − n n+k (f [k]) = f . (i) A morphism f : X → Y in C(C) is homotopic to − zero if for all k there exists a morphism sk : X k → Y k−1 such that: − k−1 f k = sk+1 ◦ dk + dY ◦ sk . we get a new additive category. From mow on. X Two morphisms f. A complex is bounded (resp. One considers C as a full subcategory of Cb (C) by identifying an object X ∈ C with the complex X • “concentrated in degree 0”: X • := · · · → 0 → X → 0 → · · · − − − − where X stands in degree 0.e. g : X → Y are homotopic if f − g is homotopic to − zero. . the category Dif f (C) of diﬀerential objects in C. Homotopy Let C denote an additive category. bounded above).
− β(f ) : Mc(f ) → X[1].3. Let f : X → Y be a morphism in C(C).3. before to state this deﬁnition. If F : C → C is an additive functor. Mc(f ) is not isomorphic to X[1] ⊕ Y in C(C) unless f is the zero morphism. − Mc(F (f )).2. There are natural morphisms of complexes α(f ) : Y → Mc(f ). COMPLEXES IN ADDITIVE CATEGORIES 63 A morphism homotopic to zero is visualized by the diagram (which is not commutative): X / X k+1 / Xk X x x xx x x f k xxx xx xx sk x sk+1 {x {x /Yk / Y k+1 Y k−1 k−1 dk dk−1 Y Note that an additive functor sends a morphism homotopic to zero to a morphism homotopic to zero. is the object of C(C) deﬁned by: Mc(f )k = (X[1])k ⊕ Y k dk 0 k X[1] dMc(f ) = f k+1 dk Y Of course. Indeed: Mc(f −dk+2 0 X f k+2 dk+1 Y ◦ −dk+1 0 X f k+1 dk Y =0 Notice that although Mc(f )k = (X[1])k ⊕ Y k . . The complex 0 → X → X ⊕ X → X → 0 is homotopic − − − − to zero. The mapping − cone of f . one should check that dk+1 ) ◦ Mc(f dk ) = 0.4. then F (Mc(f )) − The homotopy category K(C) Let C be an additive category. Example 3. Mapping cone Deﬁnition 3. denoted Mc(f ).2.2. and β(f ) ◦ α(f ) = 0.
Y ) In other words. (i) the natural functor ∆ → Setf is faithful. endowed with an automorphism. and a terminal object.64 CHAPTER 3. (b) We denote by ∆inj the subcategory of ∆ such that Ob(∆inj ) = Ob(∆). (∗ = b. namely ∅. n]}n≥−1 is equivalent to ∆. the morphisms being the injective orderpreserving maps. − (ii) the full subcategory of ∆ consisting of objects {[0. denoted by ∆.3 Simplicial constructions We1 shall deﬁne the simplicial category and use it to construct complexes and homotopies in additive categories. For integers n. n ≤ k ≤ m}. Y ) = Hom C(C) (X. Y ) = {f : X → Y . (iii) ∆ admits an initial object. Y )/Ht(X. This allows us to state: Deﬁnition 3. m] the totally ordered set {k ∈ Z.3. then g ◦ f is homotopic to zero. a morphism homotopic to zero in C(C) becomes the zero morphism in K(C) and a homotopy equivalence becomes an isomorphism. the shift functor [1] : X → X[1]. f is homotopic to 0}. The homotopy category K(C) is deﬁned by: Ob(K(C)) = Ob(C(C)) Hom K(C) (X. Set: Ht(X. Deﬁnition 3. −).5. One deﬁnes similarly K∗ (C).2. They are clearly additive categories. − If f : X → Y and g : Y → Z are two morphisms in C(C) and if f or g is − − homotopic to zero. namely {0}. +. 1 This section has not been treated in 2005/2006 . 3. (a) The simplicial category. Proposition 3. m denote by [n.2.3. ADDITIVE CATEGORIES Starting with C(C). we shall construct a new category by deciding that a morphism of complexes homotopic to zero is isomorphic to the zero morphism. is the category whose objects are the ﬁnite totally ordered sets and the morphisms are the orderpreserving maps.1.
n]− [0. We set for − n ∈ Z: Fn = dn F n F ([0. for i > 0.3.8) ∅ d−1 0 / {0} d0 0 d0 1 / / {0. n]) for n ≥ −1. both morphisms are the unique injective orderpreserving map which does not take the values i and j.3. 1. dn F = i=0 n (−)i F (di ). d−1 d0 dn Theorem 3. − − − → → − − → (i) The diﬀerential object F • is a complex. The category ∆inj is visualized by (3. 2} / / / Let C be an additive category and F : ∆inj → C a functor. − (ii) Assume that there exist morphisms sn : F n → F n−1 (n ≥ 0)satisfying: F n+1 sF ◦ F (dn ) = idF n 0 n+1 n−1 sF ◦ F (dn ) = F (di ) ◦ sn i+1 F for n ≥ −1. 0 otherwise.9) F • := · · · → 0 → F −1 −F F 0 −F F 1 → · · · → F n −F · · · . 1} d1 0 d1 1 1 d2 / / / / {0.3. Indeed. n + 1] → i (0 ≤ i ≤ n + 1) 65 the injective orderpreserving map which does not take the value i. k + 1 for k ≥ i. Let us denote by dn : [0. n+1 n+1 :F →F − . In other words dn (k) = i One checks immediately that (3. . SIMPLICIAL CONSTRUCTIONS The proof is obvious. Consider the diﬀerential object (3.3. j−1 i i j k for k < i. Then F • is homotopic to zero. n ≥ 0.7) dn+1 ◦ dn = dn+1 ◦ dn for 0 ≤ i < j ≤ n + 2.
3.m .10) dX =d 2 2 X = 0. dX ) in C is the data of n. d ◦ d = d ◦ d . ADDITIVE CATEGORIES dn+1 ◦ dn = F F = = j=0 i=0 (−)i+j F (dn+1 ◦ dn ) i j (−)i+j F (dn+1 ◦ dn ) + i j 0≤i<j≤n+2 0≤j≤i≤n+1 (−)i+j F (dn+1 ◦ dn ) i j (−)i+j F (dn+1 ◦ dn ) j−1 i 0≤j≤i≤n+1 (−)i+j F (dn+1 ◦ dn ) + i j 0≤i<j≤n+2 = 0.• . (i) By (3.m+1 satisfy: − n.4 Double complexes Let C be as above an additive category. d − X n. . (n.m where X n. m) ∈ Z × Z} n.7). d X .66 Proof. i=0 n−1 (−1)i+1 F (di ◦ sn ) + F i=0 n (−1)i F (dn−1 ◦ sF ) i q.d.m → X n+1.m X : (3.m → X n. we have n+2 n+1 CHAPTER 3.e.m n. A double complex (X •. i j = i=0 n+1 (−1)i sF ◦ F (dn ) i n + i=0 n−1 (−1)i F (di ◦ sn ) F n = sn+1 ◦ F (dn ) + 0 F n i=0 (−1)i+1 sn+1 ◦ F (dn ) + i+1 F n i=0 (−1)i F (dn−1 ◦ sn ) F i = idF n + = idF n .m ∈ C and the “diﬀerentials” d X : X n. Here.m .m {X n. d X . we have used (−)i+j F (dn+1 ◦ dn ) = j−1 i = (ii) We have n−1 sn+1 ◦ dn + dF ◦ sn F F n+1 n 0≤i<j≤n+2 0≤i<j≤n+1 (−)i+j+1 F (dn+1 ◦ dn ) j i 0≤j≤i≤n+1 (−)i+j+1 F (dn+1 ◦ dn ).
These two functors are clearly isomorphisms of categories. FII : C2 (C) → C(C(C)) which associate to − a double complex X the complex whose objects are the rows (resp.m .3.m+1 / n. m) ∈ Z × Z.m ) + (−)n+1 d ◦ d (X n. one has d ◦ d(X n.4. Proof. the columns) of X.m / X n+1. Now consider the ﬁniteness condition: (3.m d n. dp tot(X) }p∈Z is a complex p+1 p 2 (i.d.m n.m d (−)n d n.m .m / X n. m + n = p} is ﬁnite 2 and denote by Cf (C) the full subcategory of C2 (C) consisting of objects X satisfying (3.m+1 / One deﬁnes naturally the notion of a morphism of double complexes. DOUBLE COMPLEXES One can represent a double complex by a commutative diagram: 67 / X n.e. dtot(X) ◦ dtot(X) = 0) and tot : Cf (C) → C(C) is a functor of additive − categories. The diﬀerential object {tot(X)p . n) ∈ Z × Z. and one obtains the additive category C2 (C) of double complexes.11) for all p ∈ Z. q. There are two functors FI .m ) +(−)n d ◦ d (X n. n. .m d n+1.1. {(m.m = 0. For (n. X n.m+1 X n+1.m ) = 0.11).4.m ) = d ◦ d (X n.m+1 d / X n+1.m dp = d + (−1)n d tot(X) X This is visualized by the diagram: X n.e. / X n. To such an X one associates its “total complex” tot(X) by setting: tot(X)p = ⊕m+n=p X n. It is left to the reader to check that tot is an additive functor.m ) + d ◦ d (X n.m d n.m Proposition 3.
X ) is a double complex.• (X. Y ) is a double complex in the category Ab. One sets (3. construct a morphism dX : X → X[1] in C(C) − − X and prove that dX : X → X[1] is homotopic to zero.e.4. Let X ∈ C− (C) and Y ∈ C+ (C). Y n ). Then − (3. − •. • ) is additive with respect to each argument). X Note that Hom•. C C Exercises to Chapter 3 Exercise 3. In other words.m = Hom C (X −m . C and C be additive categories and let F : C × C → C be an additive − bifunctor (i. ADDITIVE CATEGORIES Example 3. Let f • : X • → Y • be a morphism in C(C). = Hom C ((−)n d−n−1 .• = Y • . one has d n. which C should not be confused with the group Hom C(C) (X. dn ).4. It deﬁnes an additive bifunctor C2 (F ) : C(C) × C(C ) → C2 (C ). We − 2 − shall simply write ⊗ instead of C (⊗).2. Z i.13) Hom• (X.m C d n.. Examples 3. Let C be an additive category and let X ∈ C(C). if − X ∈ C(C) and X ∈ C(C ) are complexes. d X n. with diﬀerentials f j : Z −1. n.12) Bifunctor Let C.• = X • .• = 0 for i = −1.3. By using ±dp : X p → X p+1 (p ∈ Z).• (X.1.• ) Mc(f • ). Y (X ⊗ Y )n. one has Hom•.j .j → Z 0. If X and Y are two objects of C(C). F ( • . then C2 (F )(X.m = X n ⊗ Y m . Hence.4. 0.m = X n ⊗ dm .• 2 We shall write Hom C instead of C (Hom C ).4. Consider the − double complex Z •. Y )n. Y )). Y m ). Deﬁnition 3. Y d = Hom C (X −m .m = dn ⊗ Y m . (ii) Consider the bifunctor • ⊗ • : Mod(Aop ) × Mod(A) → Mod(Z). (i) Consider the bifunctor Hom C : C op × C → Mod(Z).• (X. Y ) = tot(Hom•. − . Z 0. for X ∈ C (Mod(Aop )) and Y ∈ C− (Mod(A)). Y ).• such that Z −1.68 CHAPTER 3.m tot(Z •.
T ) is an object X ∈ A together with a morphism dX : X → T (X). − (c) f factors through β(idY )[−1] : Mc(idY )[−1] → Y . C) may be identiﬁed to the category of diﬀerential objects in C Z . A morphism f : (X. One denotes by Ac the full subcategory of Ad consisting of complexes. (i) Let C be a category. (ii) Show that the category Fct(Z. A diﬀerential object (X. prove that u is an isomorphism in C(C). Let C be an additive category and let f : X → Y be a mor− phism in C(C). Let C be an additive category.2. One denotes by Ad the subcategory of (A. Prove that the following conditions are equivalent: (a) f is homotopic to zero. one says that a diﬀerential d T (dX ) object (X. ≤) considered as a category. A category with translation (A. T ) is a complex if the composition X −X T (X) − − → −→ T 2 (X) is zero. Prove that C Z := Fct(Zd . dX ) in a cate− gory with translation (A. − − Exercise 3. T ) is a category A together with an equivalence T : A → A. Denote by Zd the set Z considered as a discrete category and still denote by Z the ordered set (Z. T ) consisting of diﬀerential objects and morphisms of such objects. (b) f factors through α(idX ) : X → Mc(idX ). If A is additive. Prove that f and g are homotopic if and only if there exists a commutative diagram in C(C) Y α(f ) / Mc(f ) u β(f ) / X[1] X[1].Exercises to Chapter 3 69 Exercise 3.3. − (d) f decomposes as X → Z → Y with Z a complex homotopic to zero.4. dX ) in (A. Exercise 3. Y α(g) / Mc(f ) β(g) / In such a case. dX ) → (Y. g : X Y two morphisms in C(C). C) is a category with translation. f. . dY ) of diﬀerential objects is − − a commutative diagram X f dX / TX T (f ) Y dY / T Y.
n − 1] → (0 ≤ i ≤ n − 1) the surjective orderpreserving map which takes the same value at i and i+1.n] i n+1 n−1 sj ◦ dn = di−1 ◦ sn i j for for for for 0 ≤ j < i ≤ n.2.70 CHAPTER 3. 0 ≤ i < j ≤ n.6. Consider the category ∆ and for n > 0. ϕ) is homotopic to zero. 0 ≤ i ≤ n + 1. i = j.5. Prove that the Koszul complex K • (M. In other words n si (k) = k for k ≤ i. k − 1 for k > i.1 when choosing A = C(C) and T = [1]. j + 1. n]− [0. ϕn ) be n endomorphisms of M over A which satisfy: ϕi = 0 or ϕi = idM . Exercise 3. Show that the notions of diﬀerential objects and complexes given above coincide with those in Deﬁnition 3. ADDITIVE CATEGORIES (iii) Let C be an additive category. . . . denote by n si : [0. Exercise 3. . . Let M be an Amodule and let ϕ = (ϕ1 . Checks the relations: n n+1 sj ◦ si = sn ◦ sn+1 i−1 j n+1 s n−1 n ◦ di = di ◦ sn j−1 j n+1 sj ◦ dn = id[0. 1 ≤ j + 1 < i ≤ n + 1.
if Coker f exists. the sequence (4. such as the snake Lemma. In these Notes.1 Abelian categories Convention 4.1 below. with applications to the functors Ext and Tor. Recall that if Ker f exists.Chapter 4 Abelian categories In this chapter. X) → Hom C (W.1. From now on. Let f : X → Y be a morphism in C.2) 0 → Hom C (Coker f.1) 0 → Hom C (W. Y ) − − − f is exact in Mod(Z). we give the main properties of abelian categories and expose some basic constructions on complexes in such categories.4 below).1. when dealing with an abelian category C (see Deﬁnition 4.1. we shall assume that C is a full abelian subcategory of a category Mod(A) for some ring A. the sequence (4. Similarly. (See Convention 4. it is − unique up to unique isomorphism.) Some important historical references are the book [5] and the paper [14]. W ) − − − f is exact in Mod(Z). Ker f ) → Hom C (W. then for any W ∈ C. We explain the notion of injective resolutions and apply it to the construction of derived functors. C will denote additive categories. we shall always argue as if we were working in a full abelian subcategory of Mod(A) for a ring A. C. For sake of simplicity. This makes the proofs much easier and moreover there exists a famous theorem (due to Freyd & Mitchell) that asserts that this is in fact always the case (up to equivalence of categories).1. W ) → Hom C (Y. and for any W ∈ C. 71 . W ) → Hom C (X. 4.
a morphism f is a monomorphism (resp. where k : Y → Coker f.1.1. For such an f . − (ii) for any morphism f in C. ˜ ˜ Since k ◦ f = k ◦ f ◦ s = 0 and s is an epimorphism. One deﬁnes: − Coim f = Coker h. Let C be an additive category. ABELIAN CATEGORIES Example 4. then (4. deﬁne Ker f = f −1 (0) and Coker f = Y /Im f where Im f denotes the closure of the space Im f . − Consider the diagram: Ker f h f ˜ f k / X s /Y : O / / Coker f Coim f u Im f ˜ ˜ Since f ◦ h = 0. If f : X → Y is a − morphism of Banach spaces. . an epimorphism) if and only if Ker f 0 (resp. In an abelian category. it is an isomorphism. (i) If A is a ring and f is a morphism in Mod(A). Let A be a ring.4. We have thus constructed a canonical Hence f morphism: (4. Ker f = Coker f = 0 although f is not an isomorphism. I an ideal which is not ﬁnitely generated and let M = A/I.3) Coim f → Im f. the natural morphism Coim f → Im f is an − isomorphism. If f is both a monomorphism and an epimorphism. we get that k ◦ f = 0.3) is an isomorphism. Let C be an additive category which admits kernels and cokernels. − u Examples 4.1. and k ◦ f factors through k ◦ f. One says that C is abelian if: (i) any f : X → Y admits a kernel and a cokernel.2. Coker f 0). where h : Ker f → X − Im f = Ker k.72 CHAPTER 4. f factors uniquely through f .3.3) is not an isomorphism. ˜ factors through Ker k = Im f . It is wellknown that there exist continuous linear maps f : X → Y which are injective. the morphism (4. Then the natural morphism A → M in Modf (A) has no − kernel. with dense and − non closed image. Let f : X → Y be a morphism in C. Deﬁnition 4. (ii) The category Ban admits kernels and cokernels. Hence. Thus Coim f X and Im f Y .
− (i) there exists h : X → X such that g ◦ h = idX .7.5.1.1. a sequence of morphisms X p − · · · → X n with di+1 ◦ ∼ → di = 0 for all i ∈ [p. (iii) The category Ban admits kernels and cokernels but is not abelian. − (iii) there exists ϕ = (k. n−1] is exact if Im di − Ker di+1 for all i ∈ [p. One − deﬁnes similarly the cokernel. Then if C is abelian.1.2.) (iv) Let I be category. Ker ϕ is a kernel of ϕ. then the opposite category C op is abelian. g) and ψ = (f + h) such that X − X ⊕ X and → X ⊕ X − X are isomorphisms inverse to each other. if F. the category C I of functors from I to C. For example. One naturally extends Deﬁnition 1.1. (iii) A short exact sequence is an exact sequence 0 → X → X → X → 0 − − − − Any morphism f : X → Y may be decomposed into short exact se− quences: 0 → Ker f → X → Im f → 0 − − − − 0 → Im f → Y → Coker f → 0. deﬁne the functor Ker ϕ by − Ker ϕ(X) = Ker(F (X) → G(X)). Then clearly. hence. (i) If A is a ring. Mod(A) is an abelian category. (ii) there exists k : X → X such that k ◦ f = idX .1 to abelian categories. It deﬁnes a morphism Coim f → Ker g. one calls − − such a sequence a complex). (i) One says that a sequence X → X → X with − − ∼ g ◦ f = 0 is exact if Im f − Ker g. − − − − Proposition 4.4. a morphism Im f → Ker g. → → − (ii) More generally.6. Consider a f g sequence of morphisms X → X → X with g ◦ f = 0 (sometimes. → ψ ϕ f g dp f g . n−1]. (See Examples 4. G : I → C are two functors − and ϕ : F → G is a morphism of functors. − C being abelian.1. then Modf (A) is abelian. − Deﬁnition 4. is abelian. Let 0 → X → X → X → 0 be a short exact sequence − − − − in C. Then the conditions (i) to (iii) are equivalent. (ii) If A is noetherian.3 (ii). we assume until the end of this chapter that C is abelian. (v) If C is abelian. Unless otherwise speciﬁed. ABELIAN CATEGORIES 73 Examples 4.
In other words. Let F : C → C be an additive functor. there exists x1 ∈ X1 such that α1 (x1 ) = x2 . Since f 1 ◦ α0 (x0 ) = f 1 (x1 ) and f 1 is a monomorphism. one says that the sequence splits. there exists y0 ∈ Y 0 such that β0 (y0 ) = f 1 (x1 ). F is right exact if and only if it commutes with cokernels. Proof. Hence we may choose elements in the objects of C. − − − . Lemma 4. α0 (x0 ) = x1 . Since the ﬁrst row is exact. (ii) is nothing but (i) in C op . F (0) 0 and F (X ⊕ Y ) F (X) ⊕ F (Y ).1.) Consider a commutative diagram: X0 f0 α0 / X1 / α1 / X2 / α2 / X3 /Y3 f1 f2 f3 Y0 β0 Y1 β1 Y2 β2 and assume that the rows are exact sequences. ABELIAN CATEGORIES The proof is similar to the case of Amodules and is left as an exercise. that is. Similarly.74 CHAPTER 4. and f 0 . Then f 3 ◦ α2 (x2 ) = 0 and f 3 being a monomorphism. F commutes with ﬁnite direct sums (and with ﬁnite products). if and only if for any exact sequence in C. Recall that F is left exact if and − only if it commutes with kernels. (i) Let x2 ∈ X2 and assume that f 2 (x2 ) = 0. Since F − is additive. that is. x2 = α1 (x1 ) = 0. 0 → X → X → X the sequence 0 → F (X ) → F (X) → F (X ) is − − − − − − exact in C . if and only if for any exact sequence in C. If the conditions of the above proposition are satisﬁed. Since β1 ◦ f 1 (x1 ) = 0 and the second row is exact. f 3 are monomorphisms.d. then f 2 is a monomorphism. we shall assume that C is a full abelian subcategory of Mod(A) for some ring A. there exists x0 ∈ X 0 such that y0 = f 0 (x0 ). then f 1 is an epimorphism. (ii) If f 3 is a monomorphism. Set y1 = f 1 (x1 ). Since f 0 is an epimorphism. Let F : C → C be an additive functor of abelian categories. According to Convention 4. (The “ﬁve lemma”. X → X → X → 0 the − − − sequence F (X ) → F (X) → F (X ) → 0 is exact. Note that an additive functor of abelian categories sends split exact sequences into split exact sequences. (i) If f 0 is an epimorphism and f 1 . f 2 are epimorphisms.8. q.e. Therefore. this implies α2 (x2 ) = 0.1.1.
If C = Mod(A) and I is discrete. Assume that C admits inductive limits. then the functor lim is exact. 4.2. C) → C − ← − is left exact. (i) Let C be an abelian category. Hence we have obtained an additive functor: H k ( • ) : C(C) → C. all the above functors are exact. Recall that the functor lim : Fct(I. the sequence 0 → F (X ) → F (X) → F (X ) is exact. − − − − (ii) F is exact if and only if for any exact sequence X → X → X in C. Clearly. the functor lim : Fct(I op . − n n n the complex Z deﬁned by Z = Ker(f : X → Y n ).H (X ⊕ Y ) − k k H (X) ⊕ H (Y ). Examples 4.4. and similarly for Coker f . − − → If C = Mod(A) and I is ﬁltrant. Notice ﬁrst that the categories C∗ (C) are clearly abelian for ∗ = ∅. The functors ⊗A from Mod(Aop ) × Mod(A) to Mod(k) is right exact. (iii) Let I and C be two categories with C abelian. COMPLEXES IN ABELIAN CATEGORIES Lemma 4. b. the functor lim (that is. − − the sequence F (X ) → F (X) → F (X ) is exact. − − The proof is left as an exercise. If f : X → Y is a mor− phism in C(C). if C admits projective limits. with diﬀerential induced − by those of X. −.2 Complexes in abelian categories We assume that C is abelian. Let X ∈ C(C). It follows from (i) that the functors Hom A from Mod(A)op × Mod(A) to Mod(k) is left exact. will be a kernel for f . C) → C is right exact. (ii) Let A be a kalgebra.1. the ← − functor ) is exact. then it induces morphisms Z k (X) → Z k (Y ) and B k (X) → − − k k k k k B (Y ). The functor Hom C from C op × C to Mod(Z) is left exact. if f : X → Y is a morphism in C(C). − . − 75 (i) F is left exact if and only if for any exact sequence 0 → X → X → − − − X → 0 in C.10. Let M and N in Mod(A). One deﬁnes the following objects of C: Z k (X) := Ker dk X k k−1 B (X) := Im dX H k (X) := Z k (X)/B k (X) (:= Coker(B k (X) → Z k (X))) − One calls H k (X) the kth cohomology object of X. − → Similarly. +. Let F C → C be an additive functor. For example.9.1. thus a morphism H (f ) : H (X) → H (Y ). If A is a ﬁeld.
Then H (f ) : H (X) → H k (Y ) is the 0 − morphism.2. In particular.1. H j (Y • ) H j (X • ). For example.2. for short) if H k (f ) is an isomorphism for all k ∈ Z.2.2. In such a case. contrarily to the property of being qis to 0. one says that X and Y are quasiisomorphic. However. dj ≡ 0. Lemma 4. By Lemma 4. Let f k = sk+1 ◦dk +dY ◦sk . the functor H 0 : C(C) → C extends as a functor − H 0 : K(C) → C. a short exact sequence does not necessarily split.1.1. There are exact sequences − − X k−1 → Ker dk → H k (X) → 0. In view of Lemma 4. k−1 k−1 Proof. q. a complex homotopic to 0 is qis to 0. X ∈ C(C) is qis to 0 if and only if the complex X is exact. The complexes X • and Y • have the same cohomology objects. − Deﬁnition 4.76 Notice that: CHAPTER 4. Let C be an abelian category and let f : X → Y be a mor− k k phism in C(C) homotopic to zero. One says that a morphism f : X → Y in C(C) or in K(C) − is a quasiisomorphism (a qis. Consider a bounded complex X • and denote by Y • the complex given by Y j = H j (X • ). − − − which give rise to the exact sequence: k−1 (4. in general these isomorphisms are neither induced by a morphism from X • → Y • . Remark 4. X − k k−1 0 → H (X) → Coker dX → X k+1 . − − → − − X dk .2.e. nor by a morphism from Y • → X • .4.d. Then dk = 0 on Ker dk and dY ◦sk = 0 X X X k−1 k−1 on Ker dk / Im dY . ABELIAN CATEGORIES H k (X) = H 0 (X[k]).2. Remark 4. In other words. One shall be aware that the property for a complex of being homotopic to 0 is preserved when applying an additive functor. Hence H k (f ) : Ker dk / Im dX → Ker dk / Im dk−1 is the − Y X Y Y zero morphism.5) 0 → H k (X) → Coker(dX ) −X Ker dk+1 → H k+1 (X) → 0.4) Y • = ⊕i H i (X • )[−i].2. One has: Y (4. but the converse is false.3. − − • • and the two complexes X and Y are not quasiisomorphic.
4.2. COMPLEXES IN ABELIAN CATEGORIES
77
Lemma 4.2.5. (The snake lemma.) Consider the commutative diagram in C below with exact rows: X
α f
/
Y Y
g
/
Z Z
/
0
β
γ g
0
/X
f
/
/
Then it gives rise to an exact sequence: Ker α → Ker β → Ker γ − Coker α → Coker β → Coker γ. − − → − − The proof is similar to that of Lemma 4.1.8 and is left as an exercise. Theorem 4.2.6. Let 0 → X → X → X → 0 be an exact sequence in C(C). − − − − (i) For each k ∈ Z, the sequence H k (X ) → H k (X) → H k (X ) is exact. − − (ii) For each k ∈ Z, there exists δ k : H k (X ) → H k+1 (X ) making the − sequence: (4.6) H k (X) → H k (X ) − H k+1 (X ) → H k+1 (X) − → −
δk f g ϕ
exact. Moreover, one can construct δ k functorial with respect to short exact sequences of C(C). Proof. The exact sequence in C(C) gives rise to commutative diagrams with exact rows:
k−1 Coker dX dk X f
/
k−1 Coker dX dk X
g
/
k−1 Coker dX
/
0
dk X
0
/
Ker dk+1 X
f
/
Ker dk+1 X
g
/
Ker dk+1 X
Then using the exact sequence (4.5), the result follows from Lemma 4.2.5. q.e.d. Remark 4.2.7. Let us denote for a while by δ k (f, g) the map δ k constructed in Theorem 4.2.6. Then one can prove that δ k (−f, g) = δ k (f, −g) = −δ k (f, g). Corollary 4.2.8. In the situation of Theorem 4.2.6, if two of the complexes X , X, X are exact, so is the third one.
78
CHAPTER 4. ABELIAN CATEGORIES
Corollary 4.2.9. Let f : X → Y be a morphism in C(C). Then there is a − long exact sequence · · · → H k (X) − − H k (Y ) → H k+1 (Mc(f )) → · · · − −→ − − Proof. There are natural morphisms Y → Mc(f ) and Mc(f ) → X[1] which − − give rise to an exact sequence in C(C): (4.7) 0 → Y → Mc(f ) → X[1] → 0. − − − −
H k (f )
Applying Theorem 4.2.6, one ﬁnds a long exact sequence · · · → H k (X[1]) − H k+1 (Y ) → H k+1 (Mc(f )) → · · · . − → − − One can prove that the morphism δ k : H k+1 (X) → H k+1 (Y ) is H k+1 (f ) up − to a sign. q.e.d.
δk
Double complexes Let C denote as above an abelian category. Theorem 4.2.10. Let X •,• be a double complex such that all rows X j,• and columns X •,j are 0 for j < 0 and are exact for j > 0. Then H p (X 0,• ) H p (X •,0 ) H p (tot(X •,• )) for all p. Proof. We shall only describe the ﬁrst isomorphism H p (X 0,• ) H p (X •,0 ) in the case where C = Mod(A), by the socalled “Weil procedure”. Let xp,0 ∈ X p,0 , with d xp,0 = 0 which represents y ∈ H p (X •,0 ). Deﬁne p,1 x = d xp,0 . Then d xp,1 = 0, and the ﬁrst column being exact, there exists xp−1,1 ∈ X p−1,1 with d xp−1,1 = xp,1 . One can iterate this procedure until getting x0,p ∈ X 0,p . Since d d x0,p = 0, and d is injective on X 0,p for p > 0 by the hypothesis, we get d x0,p = 0. The class of x0,p in H p (X 0,• ) will be the image of y by the Weil procedure. Of course, one has to check that this image does not depend of the various choices we have made, and that it induces an isomorphism.
4.3. APPLICATION TO KOSZUL COMPLEXES This can be visualized by the diagram: x0,p x
1,p−2 d d / x1,p−1 d
79
/
0
xp−1,1
/
d
x
d
p,0
d
/
xp,1
0 q.e.d. Proposition 4.2.11. Let X •,• be a double complex such that all rows X j,• and columns X •,j are 0 for j < 0. Assume that all rows (resp. all columns) of X •,• are exact. Then the complex tot(X •,• ) is exact. The proof is left as an exercise. Note that if there are only two rows let’s say in degrees −1 and 0, then the result follows from Theorem 4.6.4
4.3
Application to Koszul complexes
Consider a Koszul complex, as in §1.6. Keeping the notations of this section, set ϕ = {ϕ1 , . . . , ϕn−1 } and denote by d the diﬀerential in K • (M, ϕ ). Then ϕn deﬁnes a morphism (4.8) ϕn : K • (M, ϕ ) → K • (M, ϕ ) −
Proposition 4.3.1. The complex K • (M, ϕ)[1] is isomorphic to the mapping cone of −ϕn . Proof.
1
Consider the diagram Mc(ϕn )p
λp
p dM
/
Mc(ϕn )p+1
λp+1
K p+1 (M, ϕ)
1
dp+1 K
/ K p+2 (M, ϕ)
The proof has been skipped in 2005/2006
Assume for example that (ϕ1 .2 follows. The cohomology of K • (M. and Theorem 1.6. Indeed. . Proposition 4. p+1 λp+1 ◦ dp (a ⊗ eJ + b ⊗ eK ) = −d (a ⊗ eJ ) + en ∧ d (b ⊗ eK ) − ϕn (a) ⊗ en ∧ eJ M = −d (a ⊗ eJ ) − d (b ⊗ en ∧ eK ) − ϕn (a) ⊗ en ∧ eJ .2. ϕn is injective on this group.9) J aJ ⊗ e J + K bK ⊗ e K → J aJ ⊗ e J + K bK ⊗ e n ∧ e K Zn may uniquely with J = p + 1 and K = p. ϕn . M λp (a ⊗ eJ + b ⊗ eK ) = a ⊗ eJ + b ⊗ en ∧ eK . Indeed. ϕ ) is thus concentrated in degree n − 1 and is isomorphic to M/(ϕ1 (M ) + · · · + ϕn−1 (M )). Any element of M ⊗ be written as in the right hand side of (4.3.d. q. ϕ )) → H j+1 (K • (M. ϕ )) − H j (K • (M.1 and Corollary 4. .3. and let us argue by induction on n. .9).2. There exists a long exact sequence (4. . let us treat the ﬁrst one.6.2.e.d. Apply Proposition 4. It is described by: (4. ϕ)) → · · · → − − − Proof.9. q.e. ABELIAN CATEGORIES Zn−1 ) ⊕ (M ⊗ p 0 Zn−1 ) @ id ⊕(id ⊗en ∧) −d 0 −ϕn d /M 1 (⊗ A p+2 Zn−1 ) ⊕ (M ⊗ p+1 Zn−1 ) id ⊕(id ⊗en ∧) M⊗ Then p+1 Z n −d / M⊗ p+2 Zn dp (a ⊗ eJ + b ⊗ eK ) = −d (a ⊗ eJ ) + (d (b ⊗ eK ) − ϕn (a) ⊗ eJ ). We can now give a proof to Theorem 1. By the hypothesis. ϕn ) is a regular sequence. (ii) The diagram commutes. p+1 p dK ◦ λ (a ⊗ eJ + b ⊗ eK ) = −d(a ⊗ eJ + b ⊗ en ∧ eK ) = −d (a ⊗ eJ ) − ϕn (a) ⊗ en ∧ eJ − d (b ⊗ en ∧ eK ). (i) The vertical arrows are isomorphisms.10) · · · → H j (K • (M.80 given explicitly by: (M ⊗ p+1 CHAPTER 4.
3.4. and − the hypotheses of the lemma are satisﬁed.4. if the functor Hom C (P. making the solid diagram commutative. An Amodule M is called injective (resp. Example 4. (i) An object I of C is injective if Hom C ( • . Proposition 4. Let 0 → X → X → X → 0 be an exact sequence in C.2.e. It follows that if F : C → C is an additive functor of abelian categories. The object I ∈ C is injective if and only if.d.1.4.8. and − − − − assume that X is injective. Y ∈ C and any diagram in which the row is exact: 0 / X k h f / Y I ~ the dotted arrow may be completed. One can prove that Mod(A) has enough injectives (see Exercise 1.4. Then the sequence splits. Let A be a ring. i.4 Injective objects Deﬁnition 4. (ii) One says that C has enough injectives if for any X ∈ C there exists a monomorphism X I with I injective. if there exists an Amodule N such that M ⊕N is free then M is projective (see Exercise 1. we ﬁnd h : X → X such − that k ◦ f = idX . then the sequence 0 → F (X ) → − − F (X) → F (X ) → 0 splits and in particular is exact.1. then any object of Mod(k) is both injective and projective. More generally. If k is a ﬁeld.4.2). q. (iv) One says that C has enough projectives if for any X ∈ C there exists an epimorphism P X with P projective. Applying the preceding result with k = idX .3. I) is an exact functor. − − f g .5). The proof is similar to that of Proposition 1. If M is free then it is projective.7. Lemma 4. One immediately deduces that the category Mod(A) has enough projectives. Then apply Proposition 4. INJECTIVE OBJECTS 81 4. Proof. (iii) An object P is projective in C iﬀ it is injective in C op . for any X.e.4. projective) if it is so in the category Mod(A).4. • ) is exact.
6. there exists a J resolution of X. Then X is injective. Then X ⊕ X is injective if and only if X and X are injective.3.5.d. Embed B n in an object of J : − − − 0 → B n → J n+1 . Consider an exact sequence in C.5. Assume to have constructed: 0 → X → J0 → · · · → Jn − − − − For n = 0 this is the hypothesis.4 and 4. Set B n = Coker(J n−1 → J n ) (with J −1 = − − − − X). We proceed by induction. If the J k ’s belong − − to J . We − − − − − shall say for short that 0 → X → J • is a resolution of X. there exist Y ∈ J and a monomorphism X Y. q. We shall asume (4.11) the abelian category C admits enough injectives. When J denotes the category of injective objects one says this is an injective resolution. Notations 4. Let 0 → X → X → X → 0 be an exact sequence in C − − − − and assume X and X are injective.82 CHAPTER 4.e.4.4.4. Let X . we get: Proposition 4. .d. We say that J is cogenerating if for all X in C.e. Then J n−1 → J n → J n+1 is exact.4. X belong to C.5. It is enough to remark that for two additive functors of abelian categories F and G. ABELIAN CATEGORIES Lemma 4. Assume J is cogenerating. and the induction − proceeds. X → F (X) ⊕ G(X) is exact if and only if F and G are exact.2. Then for any X ∈ C.1. Let J be a full additive subcategory of C.5. 4. 0 → X → J 0 → · · · → − − − − n • 0 n J → · · · and denote by J the complex 0 → J → · · · → J → · · · . Proof. Proof. Note that the category of injective objects is cogenerating iﬀ C has enough injectives. we shall say that this is a J resolution of X. Proposition 4.5. C denotes an abelian category and IC its full additive subcategory consisting of injective objects.5 Resolutions In this section. Applying Lemmas 4. Then J n−1 → J n → B n → 0 is exact. Deﬁnition 4. q.
k−1 Hence.4. (i) Consider the diagram: X k−2 sk−1 / X k−1 sk / Xk sk+1 / X k+1 / f k−1 fk I k−2 z / I k−1 { / Ik { I k+1 We shall construct by induction morphisms sk satisfying: k−1 f k = sk+1 ◦ dk + dI ◦ sk . Deﬁne k−1 g k = f k − dI ◦ sk .4. As− sume I • belongs to C + (IC ) and X • is exact.e. RESOLUTIONS 83 Proposition 4. Assume we have constructed the sj for j ≤ k. q. Then there exists a morphism f • : X • → J • making the − diagram below commutative: (4. X For j << 0. Then I • is homotopic to 0. − − − Proposition 4. (ii) Let I • ∈ C+ (IC ) and assume I • is exact. Then f • is homotopic to 0. One has k−1 k−1 k−1 k−1 g k ◦ dX = f k ◦ dX − dI ◦ sk ◦ dX k−1 k−1 k−1 k−2 = f k ◦ dX − dI ◦ f k−1 + dI ◦ dI ◦ sk−1 = 0. (i) Let f • : X • → I • be a morphism in C+ (C).d.12) 0 0 / X f / X• / f• / Y J• . Proof.5. g k factorizes through X k / Im dX .5. sj = 0.5.4. let 0 → X → • • X be a resolution of X and let 0 → Y → J be a complex with the − − k J ’s injective. Consider − − 0 / k−1 X k / Im dX / X k+1 gk sk+1 Ik x The dotted arrow may be completed by Proposition 4. (ii) Apply the result of (i) with X • = I • and f = idX .5. k−1 the sequence 0 → X k / Im dX → X k+1 is exact. Since the complex X • is exact. (i) Let f : X → Y be a morphism in C.3.
Hence. (i) Let X ∈ C and let IX ∈ C+ (IC ) be an injective resolution of • + X.5. q. Y → J 0 ) and − − X Y set f −1 = f .4 (i). hn extends as f n+1 : X n+1 → J n+1 . . We shall make the hypothesis (4.13) the category C admits enough injectives. replacing the exact sequence 0 → Y → J • by the − − • complex 0 → 0 → J . We shall construct the f n ’s by induction. by d−1 (resp. we get that if g : Y → Z is another morphism in C and IZ is − • • • + an injective resolutions of Z. By Propo• • sition 4. The morphism g n factorizes through hn : X n / Im dn−1 → − − X n−1 n+1 • n n+1 J .84 CHAPTER 4. hence is unique in K+ (C).5.d. functorially in X ∈ C.5. . Let g n = dY ◦ f n :X n → J n+1 . J • ). let IX and IY be injective resolutions − • of X and Y respectively. consider two injective resolutions IX and JX of X.5. . Similarly. • • (ii) Let f : X → Y be a morphism in C. ABELIAN CATEGORIES (ii) The morphism f • in C(C) constructed in (i) is unique up to homotopy. Proof. by Proposition 4.5.5. Assuming (4. Morphism f 0 is obtained by n Proposition 4. 4. − • Proof. d−1 ) the morphism X → X 0 (resp.6.12 commutative and this morphism is unique up to homotopy. Assume we have constructed f 0 .5 applied to idX .5. Lemma 4. IY ) • does not depend on the choice of f by Proposition 4. there exists a functor λ : C → K(IC ) and − for for each X ∈ C. Since X is exact. this follows − − from Proposition 4.e. and let f • : IX → IY be a morphism of complexes − • • • such as in Proposition 4. dY ) the diﬀerential of the complex X • (resp.d. • In particular. C and C will denote abelian categories and F : C → C a left − exact functor. Then the image of f • in Hom K+ (IC ) (IX . f • and g • are isomorphisms inverse one to each other.4.13). f n .e. there exists a unique morphism • − • g • : JX → IX in K+ (C). . • • Indeed. The image of IX in K (C) is unique up to unique isomorphism. there exists a morphism f : IX → JX making − • the diagram 4.1. Recall that IC denotes the additive category of injective objects in C. Since the sequence 0 → X → X • is exact. (i) Let us denote by dX (resp. . Since J n+1 is injective. − − q. then g ◦ f = (g ◦ f ) as morphisms in K (IC ). the sequence 0 → X / Im dX → X − − is exact.6 Derived functors In this section.3.5. − (ii) We may assume f = 0 and we have to prove that in this case f • is homotopic to zero. a qis X → λ(X).
then Ker(F (IX ) → F (IX )) 0 F (Ker(IX → IX )) F (X). The last assertion is obvious by the construction of Rj F (X). Rn F (X) H n (F (IX )). − − • • apply F to this resolution. and the second one 1 0 − follows from the fact that F being left exact. construct an exact • sequence 0 → X → IX with IX ∈ C+ (IC ). injective objects are F acyclic for all left exact functors F . 0 for all k > 0 is Hence.4. One sets (4.4. • take the nth cohomology. Deﬁnition 4. − → − Deﬁnition 4. − − − Theorem 4. The third assertion is clear since F being exact. F (X). By its deﬁnition. k The ﬁrst assertion is obvious since IX = 0 for k < 0. DERIVED FUNCTORS 85 Let F : C → C be a left exact functor of abelian categories and assume − that C has enough injectives.2. Let 0 → X → X → X → 0 be an exact sequence in C.6. • In other words. 0 for n = 0 if F is exact. − 1 it commutes with H j ( • ).6.14) Rn F = H n ◦ F ◦ λ λ F Hn and calls Rn F the nth right derived functor of F . the receipt to construct Rn F (X) is as follows: • • choose an injective resolution IX of X.6.3. − Then there exists a long exact sequence: − − − − − − − 0 → F (X ) → F (X) → · · · → Rk F (X ) → Rk F (X) → Rk F (X ) → · · · f g . An object X of C such that Rk F (X) called F acyclic.6. n Note that R F is an additive functor from C to C and Rn F (X) R0 F (X) Rn F (X) Rn F (X) 0 for n < 0. that is. 0 for n = 0 if X is injective. Consider the functors C → K+ (IC ) − K+ (C ) −→ C .
2. Then I F is F injective.6. Since − − − − the objects X j are injectice.6. then X ∈ J . The conditions (ii) and (iii) in Deﬁnition 4. ABELIAN CATEGORIES Sketch of the proof. One constructs an exact sequence of complexes 0 → − X • → X • → X • → 0 whose objects are injective and this sequence is − − − f g quasiisomorphic to the sequence 0 → X → X → X → 0 in C(C). − − − − • • By considering C op . (ii) for any exact sequence 0 → X → X → X → 0 in C with X ∈ J .13) implies that IF is cogenerating.e.e.6. One says that J is F injective if: (i) J is cogenerating. Proof.5 are satisﬁed by Theorem 4. the category I of injective objects is F acyclic for all left exact functors F . the sequence 0 → F (X ) → F (X) → F (X ) → 0 is exact. Let F : C → C be a left exact functor and denote by − IF the full subcategory of C consisting of F acyclic objects.6. Deﬁnition 4.7. hypothesis (4. Lemma 4. The full additive subcategory of Mod(A) consisting of ﬂat Amodules is projective with respect to the functor N ⊗A • . one obtains the notion of an F projective subcategory. q.4. − − − − (iii) for any exact sequence 0 → X → X → X → 0 in C with X ∈ J . Assume J is F injective and let X • ∈ C+ (J ) be a complex qis to zero (i. X ∈ − − − − J .d. q.6. X • is exact). F being right exact.8.6.6.86 CHAPTER 4. Proposition 4. Proof. (i) If C has enough injectives. We decompose X • into short exact sequences (assuming that this complex starts at step 0 for convenience): 0 → X0 → X1 → Z1 → 0 − − − − 1 2 2 0→Z →X →Z →0 − − − − ··· 0 → Z n−1 → X n → Z n → 0 − − − − . we get a short exact sequence in C(C ): − − − 0 → F (X ) → F (X • ) → F (X ) → 0 − Then one applies Theorem 4.d. Since injective objects are F acyclic. Then F (X • ) is qis to zero. Let J be a full additive subcategory of C.e.5. Examples 4.6. (ii) Let A be a ring and let N be a right Amodule.
e. Applying Proposition 4. Assume J is F injective and contains the category IC of injective objects. F (K • ) is qis to zero. one sees that in order to calculate the kth derived functor of F at X.6.6. Consider a resolution 0 → X → J • − − of X by objects of J .e. and take the kth cohomology object. (i) If G is exact. the mapping cone of F (f ) : F (J • ) → F (I • ).8.4.9.2. We assume that C and C have enough injectives. By Lemma 4. there exists f : J • → I • making the diagram below commutative − 0 0 / / X id / J0 f0 d0 J / J1 f1 d1 J / ··· ··· X / I 0 d0 I / I 1 d1 I / Deﬁne the complex K • = Mc(f ).2. F (Mc(f )) is isomorphic to Mc(F (f )). K • belongs to C+ (J ) and this complex is qis to zero by Corollary 4. then apply F to the complex J • . Proposition 4. Let X ∈ C and let 0 → X → J • be a resolution of X with − − J k ∈ J .10.d. Then for each k.6.d. By this result. Applying Theorem 4. the result follows. we − ﬁnd a long exact sequence · · · → H n (F (J • )) → H n (F (I • )) → H n (F (K • )) → · · · − − − − Since F (K • ) is qis to zero. then Rj (G ◦ F ) G ◦ Rj F .6.6 to this sequence. there is an isomorphism Rk F (X) H k (F (J • )). By the hypothesis. hence all the sequences: 0 → F (Z n−1 ) → F (X n ) → F (Z n ) → 0 − − − − are exact. Proof.8. q. Hence the sequence 0 → F (X 0 ) → F (X 1 ) → · · · − − − is exact.5. On the otherhand. the recipe is as follows. DERIVED FUNCTORS 87 By induction we ﬁnd that all the Z j ’s belong to J . the mapping cone of f . . Let F : C → C and G : C → C be left exact functors − − of abelian categories.5. q. Theorem 4. Let 0 → X → J • be a J resolution of X and let 0 → X → I • − − − − be an injective resolution of X.
88 CHAPTER 4. (iv) In particular. C ) consisting of injective objects. Let X ∈ C and let 0 → X → IX be an injective resolution of X. F (X.7. • ) is exact. Rk F (X. • (iii) By the hypothesis. (iii) Let J be a Ginjective subcategory of C and assume that F sends the injective objects of C in J . Assume that (IC . Y ) are left exact). C ) is F injective. Proposition 4.5.e. there exists a morphism F (IX ) → JF (X) . • ) and F ( • . (a) The pair (IC . 4. F (IX ) is qis to F (X) and belongs to C+ (J ). Y )(X).5.d.2. Y ) → Rk F (X.. (iv) is a particular case of (iii). the righthand side is isomorphic to G(H j (F (IX )). (ii) Consider an injective resolution 0 → F (X) → JF (X) of F (X). Then − − • • Rj (G ◦ F )(X) H j (G ◦ F (IX )). Y ) → − − − − − . C ) is F injective if C admits enough injective and for all I ∈ IC . There is a natural morphism R j (G ◦ F ) → − (Rj G) ◦ F .7. i.e. − − Then there is a long exact sequence in C : · · · → Rk−1 F (X .7 Bifunctors Now consider an additive bifunctor F : C ×C → C of abelian categories. Y ) at X.. Since H j ((G ◦ F )(IX )) − • Rj (G ◦ F )(X) and H j (G(JF (X) )) Rj G(F (X)). Y ) → Rk F (X . If X ∈ C satisﬁes Rk F (X) = 0 for k = 0. Y ) the kth derived functor of F ( • . − − (i) Let 0 → X → X → X → 0 be an exact sequence in C and let Y ∈ C . Let IC (resp. let J be a Ginjective subcategory of C and assume that F is exact and sends the injective objects of C in J . IC ) denote the full additive subcategory of C (resp. Applying G we get a − • • • • morphism of complexes: (G ◦ F )(IX ) → G(JF (X) ). (b) If (IC . Y ) = Rk F ( • .1. ABELIAN CATEGORIES (ii) Assume that F is exact. we denote by Rk F (X. C ) is F injective. Then Rj (G ◦ F ) Rj G ◦ F . By Propo− − • • sition 4. F (I. Deﬁnition 4. Proof.e. q. • (i) If G is exact. we get the result. and − assume: F is left exact with respect of each of its arguments (i. Y ) → Rk F (X . then Rj (G ◦ F )(X) Rj G(F (X)). Hence j j • R G(F (X)) H (G(F (IX ))).
4.7. BIFUNCTORS
89
(ii) Let 0 → Y → Y → Y → 0 be an exact sequence in C and let X ∈ C. − − − − Then there is a long exact sequence in C : · · · → Rk−1 F (X, Y ) → Rk F (X, Y ) → Rk F (X, Y ) → Rk F (X, Y ) → − − − − − Proof. (i) is a particular case of Theorem 4.6.4. (ii) Let 0 → X → I • be an injective resolution of X. By the hypothesis, the − − sequence in C(C ): 0 → F (I • , Y ) → F (I • , Y ) → F (I • , Y ) → 0 − − − − is exact. By Theorem 4.2.6, it gives rise to the desired long exact sequence. q.e.d. Proposition 4.7.3. Assume that both (IC , C ) and (C, IC ) are F injective. Then for X ∈ C and Y ∈ C , we have the isomorphism: Rk F (X, Y ) := Rk F ( • , Y )(X) Rk F (X, • )(Y ). • • Moreover if IX is an injective resolution of X and IY an injective resolu• • tion of Y , then Rk F (X, Y ) totH k (F (IX , IY ). Proof. Let 0 → X → IX and 0 → Y → IY be injective resolutions of X and − − • − − • Y , respectively. Consider the double complex: 0 0 0 0
/ / /
0
/ / /
0 F (IX , Y )
0
/ / /
1 F (IX , Y )
0
/ / /
0 F (X, IY ) 1 F (X, IY )
0 0 F (IX , IY ) 0 1 F (IX , IY )
1 0 F (IX , IY ) 1 1 F (IX , IY )
The cohomology of the ﬁrst row (resp. column) calculates R k F ( • , Y )(X) (resp. Rk F (X, • )(Y )). Since the other rows and columns are exact by the hypotheses, the result follows from Theorem 4.2.10. q.e.d. Example 4.7.4. Assume C has enough injectives. Then Rk Hom C : C op × C → Ab − exists and is calculated as follows. Let X ∈ C, Y ∈ C. There exists a qis in C+ (C), Y → I • , the I j ’s being injective. Then: − Rk Hom C (X, Y ) H k (Hom C (X, I • )).
90
CHAPTER 4. ABELIAN CATEGORIES
If C has enough projectives, and P • → X is a qis in C− (C), the P j ’s being − projective, one also has: Rk Hom C (X, Y ) H k Hom C (P • , Y ) H k tot(Hom C (P • , I • )).
If C has enough injectives or enough projectives, one sets: (4.15) Extk ( • , • ) = Rk Hom C ( • , • ). C
For example, let A = k[x, y], M = k A/xA+yA and let us calculate the groups Extj (M, A). Since injective resolutions are not easy to calculate, it A is much simpler to calculate a free (hence, projective) resolution of M . Since (x, y) is a regular sequence of endomorphisms of A (viewed as an Amodule), M is quasiisomorphic to the complex: M • : 0 → A → A2 → A → 0, − − − − where u(a) = (ya, −xa), v(b, c) = xb + yc and the module A on the right stands in degree 0. Therefore, Extj (M, N ) is the jth cohomology object of A the complex Hom A (M • , N ), that is: 0 → N − N 2 − N → 0, − → → − where v = Hom (v, N ), u = Hom (u, N ) and the module N on the left stands in degree 0. Since v (n) = (xn, yn) and u (m, l) = ym − xl, we ﬁnd again a Koszul complex. Choosing N = A, its cohomology is concentrated in degree 2. Hence, Extj (M, A) 0 for j = 2 and k for j = 2. A Example 4.7.5. Let A be a kalgebra. Since the category Mod(A) admits enough projective objects, the bifunctor
•
u
v
v
u
⊗
•
: Mod(Aop ) × Mod(A) → Mod(k) −
admits derived functors, denoted TorA ( • , • ) or else, Tork ( • , • ). −k A − If Q• → N → 0 is a projective resolution of the Aop module N , or − P • → M → 0 is a projective resolution of the Amodule M , then : − −
A Tork (N, M )
H −k (Q• ⊗A M ) H −k (tot(Q• ⊗A P • )). H −k (N ⊗A P • )
Exercises to Chapter 4
91
Exercises to Chapter 4
Exercise 4.1. Let C be an abelian category which admits inductive limits and such that ﬁltrant inductive limits are exact. Let {Xi }i∈I be a family of objects of C indexed by a set I and let i0 ∈ I. Prove that the natural − morphism Xi0 → i∈I Xi is a mornomorphism. Exercise 4.2. Let C be an abelian category. (i) Prove that a complex 0 → X → Y → Z is exact iﬀ and only if for − − − any object W ∈ C the complex of abelian groups 0 → Hom C (W, X) → − − Hom C (W, Y ) → Hom C (W, Z) is exact. − (ii) By reversing the arrows, state and prove a similar statement for a complex X → Y → Z → 0. − − − Exercise 4.3. Let C be an abelian category. A square is a commutative diagram: V
g f
/
Y
g
X
f
/
Z.
A square is Cartesian if moreover the sequence 0 → V → X × Y → Z is − − − exact, that is, if V X ×Z Y (recall that X ×Z Y = Ker(f − g), where f − g : X ⊕ Y → Z). A square is coCartesian if the sequence V → X ⊕ Y → Z → 0 − − − − is exact, that is, if Z X ⊕V Y (recall that X ⊕Z Y = Coker(f − g ), where f − g : V → X × Y ). − (i) Assume the square is Cartesian and f is an epimorphism. Prove that f is an epimorphism. (ii) Assume the square is coCartesian and f is a monomorphism. Prove that f is a monomorphism. Exercise 4.4. Let C be an abelian category and consider two sequences of fi gi morphisms Xi − Xi − Xi , i = 1, 2 with gi ◦ fi = 0. Set X = X1 ⊕ X2 , → → and deﬁne similarly X, X and f, g. Prove that the two sequences above are f g exact if and only if the sequence X → X → X is exact. − − Exercise 4.5. Let C be an abelian category and consider a commutative
show that the sequence above does not split. {Rk Fi }i∈I is an inductive system of functors and Rk (lim Fi ) lim Rk Fi . Let A = k[x1 . . A Exercise 4. ϕn−p ) and ϕ = (ϕn−p+1 . Consider the Amodules: M = A/(Ax1 + Ax2 ). (i) Prove that lim Fi is a left exact functor. .9. one associates the new complex H • (X) = H j (X)[−j] with 0diﬀerential. . Prove that all columns are exact.92 diagram of complexes CHAPTER 4. − − − − (ii) By considering the action of x1 on these three modules. M = A/(Ax1 ). Let {Fi }i∈I be an inductive system of left exact functors from C to C . ϕn ). (iii) Construct free resolutions of M and M . . Let C and C be two abelian categories. Exercise 4. M = A/(Ax2 + Ax1 x2 ). . Calculate the cohomology of K • (M. We assume that C admits inductive limits and ﬁltrant inductive limits are exact in C . A) for all j . indexed by a ﬁltrant categoryI. . . Exercise 4.8. Let C be an abelian category. To X ∈ Cb (C). . ABELIAN CATEGORIES 0 0 0 0 / / / 0 / / / 0 / / / X0 X1 X2 X0 X1 X2 X0 X1 X2 Assume that all rows are exact as well as the second and third column. 1 (i) Show that the monomorphism Ax1 → A induces a monomorphism M → M and deduce an exact sequence of Amodules 0 → M → M → M → 0.6.7. Exercise 4. ϕn ) be n commuting endomorphisms of an Amodule M . Let ϕ = (ϕ1 . − → − → . In other words H • (X) := · · · → H i (X) → H i+1 (X) → · · · − − − 0 0 (i) Prove that H • : Cb (C) → Cb (C) is a welldeﬁned additive functor. Let ϕ = (ϕ1 . − → i i i (ii) Prove that for each k ∈ Z. − (ii) Give examples which show that in general. . . x2 ]. H • is neither right nor left exact. . ϕ) assuming that ϕ is a regular sequence and ϕ is a coregular sequence. . (iv) Calculate Extj (M.
and let Y • be an object of C+ (J ). − 0 g Y → Y1 − f / f ⊗id X1 ⊗ Y 0 id ⊗g X ⊗Y 0 1 / X ⊗ Y 1.6. We consider the following situation: F : C → C and G : − C → C are left exact functors of abelian categories having enough injectives.12. (Hint: use Exercise 4. Exercise 4.11. Prove that Rn (F ◦ F )(X) F (Rn F (X)). Here. .Exercises to Chapter 4 93 Exercise 4. − Let J be an F injective subcategory of C. Exercise 4. let X ∈ C and assume that Rj F (X) 0 for j < n. Prove that Rj (G ◦ F )(X) Rj−q G(Rq F (X)).10. Assume that H k (Y • ) = 0 for all k = p for some p ∈ Z.6. Assume that k is a ﬁeld and consider the complexes in Mod(k): X • := Y • := and the double complex X • ⊗ Y • := X0 ⊗ Y 0 id ⊗g f ⊗id X 0 → X 1. Prove that there is a long exact sequence: · · · → Rk−1 G(R1 F (X)) → Rk (G ◦ F )(X) → Rk G(F (X)) → · · · − − − − (Hint: construct an exact sequence 0 → X → X 0 → X 1 → 0 with X 0 − − − − 1 injective and X F acyclic.) (ii) Assume now that Rj F (X) = 0 for j = 0.) Exercise 4. 1.10. (ii) Deduce that tot(X • ⊗ Y • ) and tot(H • (X • ) ⊗ H • (Y • )) have the same cohomology objects. Let F : C → C be a left exact functor of abelian categories.13. we shall use the notation H • introduced in Exercise 4. 1 (i) Prove that tot(X • ⊗Y • ) and tot(H • (X • )⊗Y • ) have the same cohomology objects. In the situation of Proposition 4.10. Prove that Rk F (X) H k+p (F (Y • )). and let X = H p (Y • ). (i) Let X ∈ C and assume that there is q ∈ N with Rk F (X) = 0 for k = q. − J is an Ginjective subcategory of C and F sends injective objects of C in J .
Assume that k is a ﬁeld.14. (Hint: use the result of Exercise 4.94 CHAPTER 4. we use the convention that: H i (X • )[−i] ⊗ H j (Y • )[−j]).13. Prove the isomorphism H p (tot(X • ⊗ Y • )) H i (X • ) ⊗ H j (Y • ) i i+j=p H p( Here. j (A ⊕ B) ⊗ (C ⊕ D) (A ⊗ C) ⊕ (A ⊗ D) ⊕ (B ⊗ C) ⊕ (B ⊗ D) A[i] ⊗ B[j] ∼ A ⊗ B[i + j].) . ABELIAN CATEGORIES Exercise 4. Let X • and Y • be two objects of Cb (Mod(k)).
1. V ) Hom OpX (W. it will be con− venient to consider morphisms of sites f : X → Y which are not continuous maps from X to Y . a presheaf of sets in a functor from (OpX )op to Set. otherwise. In other words. (i) One sets PSh(X) := Fct((OpX )op .Chapter 5 Abelian sheaves In this chapter we expose basic sheaf theory in the framework of topological spaces. [17]. Although we restrict our study to topological spaces. Indeed. Hom OpX (W. [18]. Some references: [12]. Set) calls an object of this category a presheaf of sets.1. k denotes a commutative unitary ring. namely X. Note that the category OpX admits a terminal object. 95 . Recall that all along these Notes. namely U × V = U ∩ V . (ii) One denotes by PSh(kX ) the subcategory of PSh(X) consisting of functors with values in Mod(k) and calls an object of this category a presheaf of kmodules. Deﬁnition 5. [10]. V ) = {pt} ∅ if U ⊂ V.1 Presheaves Let X be a topological space. The family of open subsets of X is ordered by inclusion and we denote by OpX the associated category. 5. Hence: Hom OpX (U. and ﬁnite products. U ∩ V ). [3]. U ) × Hom OpX (W.
the − diagram below commutes: F (U ) ϕ(U ) / G(U ) F (V ) ϕ(V ) / G(V ) If F is a presheaf of kmodules.2. one says that s is a section of F on U . − More generally.3.96 CHAPTER 5. For example. V open in U and calls F U the restriction of F to U . For example. The correspondence U → M is a presheaf. (i) Let M ∈ Set. Examples 5. If − Ker ϕ(U ) and ϕ : F → G is a morphism of presheaves. ABELIAN SHEAVES Hence. i ∈ I is an inductive system of presheaves. In particular it is abelian and admits inductive and projective limits. the restriction morphisms. (iii) If s ∈ F (U ). one checks easily that if F and G are two presheaves. A morphism of presheaves ϕ : F → G is thus the data for any open set U of − a map ϕ(U ) : F (U ) → G(U ) such that for any open inclusion V ⊂ U . a map ρV U : F (U ) → F (V ). such − that for each open inclusions W ⊂ V ⊂ U . Note that for U ∈ OpX . then (Ker ϕ)(U ) (Coker ϕ)(U ) Coker ϕ(U ) where ϕ(U ) : F (U ) → G(U ). one gets the presheaf of Cvalued constant functions on X. Notation 5. and if V is an open subset of U . the presheaf U → F (U ) ⊕ G(U ) is the coproduct of F and G in PSh(kX ). (ii) One denotes by F U the presheaf on U deﬁned by V → F (V ). if i → Fi . . Then U → C 0 (U ) (with the usual restriction morphisms) is a presheaf of 0 Cvector spaces. one better writes sV instead of ρV U s and calls sV the restriction of s to V . ρW U = ρ W V ◦ ρ V U . If s ∈ F (U ). if M = C. the functor PSh(kX ) → Mod(k).1. called the constant presheaf on X with ﬁber M . The category PSh(kX ) inherits of most of the properties of the category Mod(k). one writes sV instead of ρV U (s). then the F (U )’s are kmodules and the maps ρV U are klinear. (i) One calls the morphisms ρV U . (ii) Let C 0 (U ) denote the Cvector space of Cvalued continuous functions on U . a presheaf F on X associates to each open subset U ⊂ X a set F (U ). one checks that the presheaf U → lim Fi (U ) is the inductive limit of this system − → i in the category PSh(kX ) and similarly with projective limits. and to an open inclusion V ⊂ U . denoted CX .1. one has: ρU U = idU . F → F (U ) is − exact.
then U ∩ V is the product of U and V in OpX . The functor F → Fx from PSh(kX ) to Mod(k) is exact. Since U. t ∈ F (V ). Note that the empty family {Ui . 5.1. One says that U is an open covering of U if i Ui = U . and let Ix denote the full subcategory of OpX consisting of open neighborhoods of x. The functor F → Fx is the composition op → − PSh(kX ) = Fct(Opop . (See Example 1. The image sx ∈ Fx of s is called the germ of s at x. any open covering U = i Ui . and for s ∈ F (U ). It is clearly exact. V ∈ OpX . − op → U ∈Ix One calls Fx the stalk of F at x. Note that any sx ∈ Fx is represented by a section s ∈ F (U ) for some open neighborhood U of x. i Ui .1. Mod(k)) → Fct(Ix . any open covering U = F (U ) satisfying si Ui = ti Ui for all i. Proof. any family {si ∈ F (Ui ).10 − → (iii)). For a presheaf F on X. i ∈ I} with I = ∅ is an open covering of ∅ ∈ OpX .1) Fx = lim F (U ). We shall have to consider families U := {Ui }i∈I of open subsets of U indexed by a set I. j.4.5. sx = tx means that there exists an open neighborhood W of x with W ⊂ U ∩ V such that ρW U (s) = ρW V (t).1.5. S1 For any open subset U ⊂ X. one has s = t. V ∈ Ix implies U ∩ V ∈ Ix . . Let x ∈ U and let s ∈ F (U ). Let F be a presheaf on X and consider the two conditions below.2.d. t ∈ S2 For any open subset U ⊂ X.5.) Proposition 5. any s.e. Recall that if U. Let x ∈ X.10. the category Ix is ﬁltrant and it follows that the functor lim is exact (see Example 4. one sets: (5. i ∈ I} satisfying si Uij = sj Uij for all i. there exists s ∈ F (U ) with sUi = si for all i. Mod(k)) −→ Mod(k). − − X lim The ﬁrst functor associates to a presheaf F its restriction to the category op op Ix . q. SHEAVES 97 Deﬁnition 5.2 Sheaves Let X be a topological space and let OpX denote the category of its open subsets.
One says that F is a sheaf if it satisﬁes S1 and S2. Let U be an open covering of U ∈ OpX and let F ∈ PSh(kX ). − ∩V = s V V ∩V . V ∈ U } such that for any V . bijective).2) F (U ) = Ker( V ∈U F (V ) V .V ∈U F (V ∩ V )). ← − Note that if F is a sheaf of sets. a section s ∈ F (U ) is the data of a family of sections {sV ∈ F (V ). If {Ui }i∈I is a family of disjoint open subsets. If F is a sheaf on X. then F (∅) = {pt}. In the sequel. V ∈ U . (ii) One denotes by Sh(X) the full subcategory of PSh(X) whose objects are sheaves. there is a natural map F (U ) → F (U ). (iv) One writes Hom kX ( • . − One can consider U as a category. Assuming that U is stable by ﬁnite intersection. − − In other words. (iii) One denotes by Mod(kX ) the full kadditive subcategory of PSh(kX ) whose objects are sheaves and by ιX : Mod(kX ) → PSh(kX ) the for− getful functor. One sets (5. we shall concentrate on sheaves of kmodules. ABELIAN SHEAVES Deﬁnition 5. then its restriction F U to an open subset U is a sheaf. then F ( i Ui ) = i F (Ui ). one writes ι instead of ιX . . A presheaf F is separated (resp.2.98 CHAPTER 5. we have (5.2.2. s V V (5. If F is a sheaf of kmodules. the natural map F (U ) → F (U ) is injective (resp. • ).1. Therefore. if F is a presheaf. Proposition 5. Here the two arrows are associated with V ∈U F (V ) → F (V ) → F (V ∩ V ) − − and V ∈U F (V ) → F (V ) → F (V ∩ V ). then F U is a sheaf on U . Note that if F is a sheaf on X and U is open in X.3) The next result is obvious. then F (∅) = 0. (i) One says that F is separated if it satisﬁes S1. If there is no risk of confusion. • ) instead of Hom Mod(kX ) ( • . is a sheaf) if and only if for any U ∈ OpX and any open covering U of U .4) F (U ) V ∈U lim F (V ).
5. a complex analytic manifold in (e)–(h).1. we have ϕ(s) = t. Let − t ∈ G(U ). X . and ϕx being injective. Assume now ϕx is an isomorphism for all x ∈ X and let us prove that ϕ : F (U ) → G(U ) is surjective. Proposition 5.(p) (iii)–(d) CX : pforms of class C ∞ . This implies that there exists an open covering U = ∪i Ui .2. as the complementary of the union of all open subsets U of X such that sU = 0.2.2. (ii) Let M ∈ Mod(k).d. (iii) Let X be a topological space in (a) below. One can deﬁne its support. ϕ(si )Ui ∩Uj = ϕ(sj )Ui ∩Uj . Let ϕ : F → G be a morphism of sheaves. for all x ∈ X. a real manifold of class C ∞ in (b)–(d). s = 0. denoted by supp F . there exists s ∈ F (U ) with sUi = si . q. with sUi = 0. by S1. 0 Examples 5. as the complementary of the union of all open subsets U of X such that F U = 0. (iii)–(c) DbX : complex valued distributions. − Let s ∈ F (U ) with ϕ(s) = 0. It says that to check that a morphism of sheaves is an isomorphism. we ﬁnd sx = 0 for all x ∈ U . Assume now ϕx is injective for all x ∈ X and let us prove that ϕ : F (U ) → G(U ) is injective. (ii) The condition is clearly necessary. The next result is extremely useful. (ii) ϕ is an isomorphism if and only if.5. (iii)–(b) CX : complex val∞ ued functions of class C . (i) The condition is necessary by Proposition 5. for all x ∈ X. (i) Let F be a sheaf on X. Note that the constant presheaf with stalk M is not a sheaf except if M = 0.2. also denoted Ωp . denoted by supp s. There exists an open covering U = ∪i Ui and si ∈ F (Ui ) such that tUi = ϕ(si ).5. SHEAVES 99 Notation 5. and by S1.3. (i) The presheaf CX is a sheaf. ϕx : Fx → Gx is an − isomorphism. hence by (i). The presheaf of locally constant functions on X with values in M is a sheaf. si Ui ∩Uj = sj Ui ∩Uj and by S2. called the constant sheaf with stalk M and denoted MX . Then (ϕ(s))x = 0 = ϕx (sx ). Since ϕ(s)Ui = tUi . ∞. (ii) Let s ∈ F (U ). − (i) ϕ is a monomorphism if and only if. Note that F X\supp F = 0. Proof. We have the classical sheaves: ∞ (iii)–(a) kX : kvalued locally constant functions. One can deﬁne its support. Then.4. ϕx : Fx → Gx is − injective.e. it is enough to do it at each stalk.
Note that if F is locally 0.6) a − : Mod(kX ) → PSh(kX ). functorial with respect to F ∈ PSh(kX ) and G ∈ Mod(kX ) (5. In other words. If F is a presheaf on X. ιX G) Hom kX (F a .3 (5. then F a = 0. More precisely. When there is no risk of confusion. ABELIAN SHEAVES (iii)–(e) OX : holomorphic functions. then F U is sheaf on U . we shall identify a sheaf and the underlying presheaf. the presheaf Coker( ∂z : OX → OX ). To be bounded is not a local property and axiom S2 is not satisﬁed. ∂z ∂ that is. (vi) If F is a sheaf on X and U is open. F (U ) = 0 − ∂ since the equation ∂z f = g is always solvable. Moreover (5. 0 (iii)–(f) Ωp : holomorphic pforms (hence.5) admits a left adjoint (5. F (U ) = 0.3. If F is a sheaf. if U = C \ {0}. .7) deﬁnes a morphism of presheaves θ : F → F a and θx : Fx → − − a Fx is an isomorphism for all x ∈ X. 5.b (iv) On a topological space X. However. to a sheaf F associates the underlying presheaf. ΩX = OX ). Theorem 5. the presheaf U → CX (U ) of continuous bounded functions is not a sheaf in general. we shall often omit the symbol ιX . The holomor∂ phic derivation ∂z is a morphism from ØX to ØX . G). X 0. the sheaf F a is called the sheaf associated with F. Consider the presheaf: F : U → O(U )/ ∂ O(U ).100 CHAPTER 5.7) Hom PSh(kX ) (F. and denote by z the holomorphic coordinate. one has the isomorphism.5) Sheaf associated with a presheaf ιX : Mod(kX ) → PSh(kX ) − Consider the forgetful functor which. In this section. The forgetful functor ιX in (5.1. then θ : F → F a − is an isomorphism. Hence the presheaf F does not satisfy axiom S1. (v) Let X = C. we shall rapidly construct a left adjoint to this functor. For U an open disk.
2. and there exists t ∈ F (V ) with ty = s(y) for all y ∈ V }. If {Fi }i∈I is an inductive system of sheaves.3. let ϕ : F → G be a morphism of presheaves and let ϕa : F a → Ga denote the − − associated morphism of sheaves.5.5). Then the sheaf associated with the constant presheaf U → M is the sheaf MX of M valued locally constant functions. Example 5. (ii) The functor a in (5. with cokernels).8) (Ker ϕ)a Ker ϕa .6) commutes with kernels. − any morphism of presheaves ϕ : F → G extends uniquely as a morphism of − sheaves ϕa : F a → Ga .e. (i) The category Mod(kX ) admits projective limits and such limits commute with the functor ιX in (5. Deﬁne: F (U ) = a 101 {s : U → x∈U Fx . Let M ∈ Mod(k). More precisely. Deﬁne θ : F → F a as follows. More precisely. s(x) ∈ Fx such that. To s ∈ F (U ). − → i where lim on the left (resp. if {Fi }i∈I is an inductive system of of presheaves. − there exists V x. for all x ∈ U .d.9) lim(Fia ) − → i (lim Fi )a . SHEAF ASSOCIATED WITH A PRESHEAF Proof.3. More precisely. of presheaves).3. one associates the section of F a : − One checks easily that F a is a sheaf and any morphism of presheaves ϕ : F → G with G a sheaf will factorize uniquely through θ. its inductive limit is the sheaf associated with its inductive limit in PSh(kX ). Then (5. (iv) The functor a commutes with inductive limits (in particular. (iii) The category Mod(kX ) admits inductive limits. and F → F a is functorial. Theorem 5.3. right) is the inductive limit in the category − → of sheaves (resp. (x → sx ) ∈ F a (U ). V open in U . (v) The category Mod(kX ) is abelian and the functor a is exact. − q. . if {Fi }i∈I is a projective system of sheaves. In particular. then (5. its projective limit in PSh(kX ) is a sheaf and is a projective limit in Mod(kX ).
Fi ) ← − i Hom kX (G. Since F → ∼ →(lim Fi )(U ). The cokernel in the category of sheaves is the sheaf associated with this presheaf. lim Fi ) ← − i It follows that lim Fi is a projective limit in the category Mod(kX ). G) − → i i lim Hom PSh(kX ) (Fi .102 CHAPTER 5. ← − i (ii) The commutative diagram 0 0 / / Ker ϕ / F Fa / / G Ga Ker(ϕa ) / deﬁnes the morphism Ker ϕ → Ker ϕa . − Proof. − (vii) Filtrant inductive limits are exact in Mod(kX ). Fi ) ← − i lim Hom PSh(kX ) (G. (iii)–(iv) Let G ∈ Mod(kX ) and let {Fi }i∈I be an inductive system of of presheaves. (lim Fi )(U ) − ← − ← − i i projective limit of sheaves in the category PSh(kX ) is a sheaf. for G ∈ Mod(kX ): lim Hom kX (G. Let ϕ : F → G is a morphism of sheaves and let ιX ϕ : ιX F → ιX G − − denote the underlying morphism of presheaves. In other words. ← − i . one shall be aware that Coker ιX ϕ is not necessarily a sheaf. ← − i Hom PSh(kX ) (G. (i) Let U be an open covering of an open subset U . We have the chain of isomorphisms Hom kX ((lim Fi )a .2. hence. Hence a F (U ) commutes with projective limits. but not right exact in general. Since the functor F → Fx commutes both with Ker and with . One has. G). On the otherhand. lim Fi ). ψx is an isomorphism for all x and it remains to apply Proposition 5. the functor ιX : Mod(kX ) → PSh(kX ) is left exact. ABELIAN SHEAVES (vi) The functor ιX : Mod(kX ) → PSh(kX ) is fully faithful and left exact. Then Ker ιX ϕ is a sheaf and coincides with ιX Ker ϕ. the morphism ψ : (Ker ϕ)a → − − a a Ker ϕ .4. G) − → i Hom PSh(kX ) (lim Fi . G) ← − lim Hom kX (Fia .
Then the result follows since a is exact. whence in the category PSh(kX ).8) we get that Im ϕ := Ker(F → (Coker ιX ϕ)a ) is isomorphic to the sheaf associated − ∼ with Im(ιX ϕ). − − − Examples 5.3. The (augmented) de Rham complex is (5. if and only if.9) and with cokernels by (5. (i) Let ϕ : F → G be a morphism of sheaves and − let x ∈ X. Since it admits a left adjoint. (ii) By Proposition 5.5. Then (Ker ϕ)x Ker ϕx and (Coker ϕ)x Coker ϕx .4. By this statement.e. Im ϕ Ker ψ if and only if (Im ϕ)x (Ker ψ)x for all x ∈ X. that is. (i) The result is true in the category of presheaves. Let ϕ : F → G be a morphism of sheaves and denote by ιX ϕ the underlying − morphism of presheaves. as well as with the functor F → Fx .d. the category Mod(kX ) admits kernels and cokernels. In particular the functor F → Fx .10) 0 → C X → CX − − ∞. Using (5. ϕx ψx ϕ ψ Proof. the complex of sheaves above is exact if and only if for each section s ∈ F (U ) deﬁned in an open neighborhood U of x and satisfying ψ(s) = 0. Then this complex is exact → → → → if and only if for any x ∈ X.2. SHEAF ASSOCIATED WITH A PRESHEAF 103 (v) By (i) and (iii).e. Since ιX Ker ϕ Ker ιX ϕ and Coker ϕ (Coker ιX ϕ)a . The isomorphism of presheaves Coim(ιX ϕ) − Im(ιX ϕ) yields → the isomorphism of the associated sheaves.5.(n) →0 − . On the other hand.d. q. there exists another open neighborhood V of x with V ⊂ U and a section t ∈ F (V ) such that ϕ(t) = sV . the sequence 0 → F (U ) → F (U ) → F (U ) is exact. the complex Fx − Fx − Fx is exact. The functor F → F a is exact since it commutes with kernels by by (5.8). (ii) Let F − F − F be a complex of sheaves.(0) d → · · · → CX − − ∞. from Mod(kX ) to Mod(k) is exact. Let X be a real analytic manifold of dimension n.3.4. the result follows. for any U ∈ OpX . it is left exact.3. Recall that the functor F → F a commutes with the functors of restriction F → F U . a complex of sheaves 0 → F → F → F is exact if − − − and only if it is exact as a complex of presheaves. Using (5. (vii) Filtrant inductive limits are exact in the category Mod(k). Proposition 5.9) we get that Coim ϕ := Coker(Ker ϕ → − F ) is isomorphic to the sheaf associated with Coim(ιX ϕ). Hence Mod(kX ) is abelian. (vi) The functor ιX is fully faithful by deﬁnition. q. Hence the result follows from (i).
the result follows. consider Example 5. We denote by Γ(U . Proposition 5. • ) : Mod(kX ) → − Mod(k) the functor F → F (U ). z is a holomorphic coordinate and U = X \ {0}. • ) is the composition X Mod(kX ) − PSh(kX ) −U Mod(k). GU ) and calls it the “internal hom” of F and G. Since ιX is left exact and λU is exact. One denotes by Hom PSh(kX ) (F.d. G) the presheaf on X.7. Proof. The functor Γ(U . • ). ABELIAN SHEAVES where d is the diﬀerential.4.2.4. Then the presheaf Hom (F. G) is a sheaf.11) 0 → C X → Ω0 → · · · → Ω X → 0 − − X− − n − d where d is the holomorphic diﬀerential. The functor Γ(U .4 1 Internal operations Internal hom Deﬁnition 5. The (augmented) holomorphic de Rham complex is (5. G ∈ PSh(kX ). Let F. ∂ Then the sequence of sheaves 0 → CX → OX −z OX → 0 is exact. q.2.e. G ∈ Mod(kX ). 5. Let U ∈ OpX .6.5 (v). Let F. G) or simply Hom (F. 1 The proofs in this section may be skipped . → → ι λ where λU is the functor F → F (U ).3.3. U → Hom PSh(kU ) (F U . Proposition 5. the sequence one obtains is no more exact.104 CHAPTER 5. Applying − − → − the functor Γ(U . The same result ω ∞ holds with the sheaf CX replaced with the sheaf CX or the sheaf DbX .1. This complex of sheaves is exact. (ii) Let X be a complex manifold of dimension n. • ) is not exact in general. This complex of sheaves is exact. The functor Γ(U . Recall that X = C. Indeed. Deﬁnition 5. • ) is left exact.
Let W ∈ OpU .d. this map is neither injective nor surjective. ϕ(s) ∈ G(W ) is zero. . Assume that ϕ V V for any V .5. G(W )) deﬁnes an element of Hom kU (F U . Then {ϕV } deﬁnes a commutative a F (W ) −W − → V ∈U G(W ∩ V ) V .V ∈U G(W ∩ V ∩ V ) where aW is given by F (W ) uniquely as s → ϕV (sW ∩V ).4. GV ). Then for V ∈ U . Let U ∈ OpX and let U be an open covering of U . (ii) Let {ϕV } belong to V ∈U Hom (F V . The functor Hom kX ( • . V diagram ∩V = ϕ V V ∩U V ∈ U . Gx ). INTERNAL OPERATIONS 105 Proof. Let us show that Hom (F. we shall prove that the sequence below is exact. (in these formulas. any W ∈ OpU and any s ∈ F (W ). In general. we get a − − − natural morphism (Hom (F. (i) Let ϕ ∈ Hom (F U . for short): 0 → Hom kX (F U . it follows that Hom ( • . GV ) V . • ) Γ(X. G)(U ). Since a morphism: ϕ : F → G deﬁnes a klinear map Fx → Gx . GU ) → − − Hom kX (F V . GV ∩V ). ϕ(s)W ∩V = 0.V ∈U V ∈U Hom kX (F V ∩V . • ) : Mod(kX )op × Mod(kX ) → Mod(kX ) − is left exact. • ) being left exact. we write Hom instead of Hom kW . W open. Since G is a sheaf. V ∈ U } is a covering of W .e. In other words. Since {W ∩ V . GU ). G)(U ) Hom (F. aW factors V ∈U F (W ) − − G(W ) → −→ − ψ(W ) G(W ∩ V ). GU ) and assume that ϕV : F V → GV is zero for all − V ∈ U . Note that Hom kX ( • . This implies ϕ = 0. q. • ). It is easy to see that ψ : OpU W → ψ(W ) ∈ Hom (F (W ). G))x → Hom (Fx . • ) ◦ Hom ( • .
H) Hom kX (G. Hom (F. H)).4. we get the chain of isomorphisms Hom kX (G ⊗ F. (iii) The second isomorphism follows from the ﬁrst one. psh psh psh psh Since these morphisms are functorial with respect to U . G. H) → Hom k (F (V ) ⊗ G(V ). Hom (F. we have the chain of morphisms − Hom PSh(kX ) (F ⊗ G. H)) → Hom (F ⊗ G.3. H) psh Hom kX (G. Hom PSh(kX ) (G. H(V ))). they deﬁne µ. ABELIAN SHEAVES Deﬁnition 5. K ∈ Mod(kX ). Hom k (F (U ).2. Hom (F. (i) One denotes by F ⊗ G the presheaf on X. one writes F ⊗ G instead of F ⊗kX G. H) Hom PSh(kX ) (G ⊗ F. H(U ))) − Hom k (G(U ) ⊗ F (U ). Hom (G ⊗ F. G ∈ Mod(kX ). H) → Hom kX (G. H)). H)). . H)) q. Hom (F.1 and Proposition 5. − For V → U in CX . Hom k (F (V ).d. H).3. Let F. H(V ))) → Hom k (G(U ). − For U ∈ OpX .4. (i) Let us deﬁne a map λ : Hom (G.4. we have the chain of morphisms Hom PSh(kX ) (G. There are natural isomorphisms: Hom kX (G ⊗ F. H)). Hom (F. Hom (F. U → F (U ) ⊗k G(U ). H(U )). H)) → Hom k (G(U ). Hom (F. − Since these morphism are functorial with respect to V ⊂ U . If there is no risk of confusion. Proof.e. H(V )) − psh psh Hom k (G(V ).4.106 Tensor product CHAPTER 5. (ii) Applying Theorem 5. H) Hom (G. Let F. they deﬁne λ.1) and calls it the tensor product of F and G. Proposition 5. Let us deﬁne a map µ : Hom kX (F ⊗ G.4. H)(U )) − → Hom k (G(U ). Hom (F. Hom k (F (V ). (ii) One denotes by F ⊗kX G the sheaf associated with the presheaf F ⊗ G (see Deﬁnition 5. It is easily checked that λ and µ are inverse to each other.
It is a motivation to introduce the following deﬁnition which extends the notion of a continuous map. Deﬁnition 5. V ∈ OpY . f t (U ∩ V ) = f t (U ) ∩ f t (V ). f t (V) is a covering of f t (V ). V = C).1. Hom (F U . Note that if f : X → Y and g : Y → Z are morphisms of sites. we do not ask f t (Y ) = X.5.12).5. In particular. Hence.. Let CX denote as above the sheaf of real valued C ∞ functions on a real manifold X.12) for any V ∈ OpY and any covering V of V .5 Direct and inverse images Let f : X → Y be a continuous map. 5. GU ). A morphism of sites f : X → Y is a functor f t : OpY → − − OpX which satisﬁes (5. it is exact. The functor • ⊗ • : Mod(kX ) × Mod(kX ) → Mod(kX ) − psh is the composition of the right exact functor ⊗ and the exact functor a . Note that f commutes with products and coverings. − . (ii) Hom (F.5. DIRECT AND INVERSE IMAGES 107 ∞ Example 5. we set for V ⊂ Y : f t (V ) := f −1 (V ) and f t : OpY → OpX − t is a functor. If V is a ﬁnite Rdimensional vector space (e. (5. Deﬁnition 5. G)U (iii) Hom (kX .g. that is. F.2. − t The functor jU satisﬁes (5. Let U → X be an open embedding. One denotes by t jU : OpU → OpX the functor OpV V → V ∈ OpX . then the sheaf of V valued C ∞ functions is nothing but ∞ CX ⊗RX VX . it satisﬁes for any U. then − − g ◦ f : X → Z is also a morphism of sites.12).4. This functor is thus right exact and if k is a ﬁeld. We denote by f t the inverse image of − a set by f . Note that for x ∈ X and U ∈ OpX : (i) (F ⊗ G)x Fx ⊗ G x .5. F ) (iv) kX ⊗ F F. One shall be aware that we do not ask that f t commutes with ﬁnite projective limits.5.
the direct image of a sheaf is a sheaf. − (i) Let G ∈ PSh(kY ). Then f t (V) is an open covering of f t (V ) and we get f∗ F (V ) F (f t (V )) F (f t (V)) f∗ F (V). the direct image of F by f . the inverse image of G by f . F ) Hom kY (G. Let f : X → Y be a morphism of sites.14) i† = i−1 U U jU ∗ . − (i) The functor f −1 : Mod(kY ) → Mod(kX ) is left adjoint to the functor − f∗ : Mod(kX ) → Mod(kY ). Note that if f is a continuous map. Proof. by setting: f∗ F (V ) = F (f t (V )). we have for F ∈ Mod(kX ) − and G ∈ Mod(kY ): Hom kX (f −1 G. by setting f −1 G = (f † G)a . Then i† F jU ∗ F is already a sheaf.5. One deﬁnes f −1 G ∈ Mod(kX ).5.d. ABELIAN SHEAVES Deﬁnition 5. Theorem 5. Hence: U (5. Proposition 5. q. . Let f : X → Y be a morphism of sites.5.5.108 CHAPTER 5. − → (ii) Let G ∈ Mod(kY ). Deﬁnition 5. Let f : X → Y be a morphism of sites and let F ∈ − PSh(kX ). In other words.3.4. In other words.5.13) (f −1 G)x (f † G)x Gf (x) . (ii) The functor f∗ is left exact and commutes with projective limits. Let U → X be an open embedding and let F ∈ Mod(kX ). One deﬁnes f † G ∈ PSh(kX ) by setting for U ∈ OpX : f † G(U ) = U ⊂f t (V ) lim G(V ). Let F ∈ Mod(kX ). One deﬁnes f∗ F ∈ PSh(kY ). one has for x ∈ X: (5. Example 5. f∗ F ). Then f∗ F ∈ Mod(kY ).7. Let V ∈ OpY and let V be an open covering of V .5.e.6.
If f is a continuous map. F ) Hom PSh(kY ) (G. The sequence 0 → G (V ) → G(V ) → G (V ) is exact. Indeed.15) Hom PSh(kX ) (f † G. the result follows. A section ϕ ∈ Hom PSh(kY ) (G. (ii) With the exception of the fact that f −1 is left exact.15) we get the chain of isomorphisms Hom kY (G. F ) Hom kX ((f † G)a . Using the isomorphism (5. − − − U Consider the category (OpY ) := {V ∈ OpY . F ) = Hom kX (f −1 G. The category ((OpY )U )op is either ﬁltrant or empty.d. the correspondence ϕ → ψ is an isomorphism. U ⊂ f t (V )}. F ). DIRECT AND INVERSE IMAGES (iii) The functor f −1 is exact and commutes with inductive limits. It follows that the sequence 0→ − U ⊂f t (V ) lim G (V ) → − − → U ⊂f t (V ) lim G(V ) → − − → U ⊂f t (V ) lim G (V ) − → is exact.16) 0→G →G→G − − − be an exact sequence of sheaves and let U ∈ OpX . then U ⊂ f t (V1 ∩ V2 ). U ⊂ f t (V )}V ∈OpY .5. we have an − isomorphism. if U ⊂ f t (V1 ) and U ⊂ f t (V2 ). f∗ F ) Hom PSh(kY ) (G. Hence it gives a family of maps {ψV : U ⊂f (V ) (iv) There are natural morphisms of functors id → f∗ f −1 and f −1 f∗ → id. (i) First we shall prove that the functor f † : PSh(kY ) → PSh(kX ) − is left adjoint to f∗ : PSh(kX ) → PSh(kY ).U ∈OpX → compatible with the restriction morphisms. functorial with respect to F ∈ PSh(kX ) and G ∈ PSh(kY ): (5. Clearly. f∗ F ) is a family of maps {ϕV : G(V ) → F (f t (V ))}V ∈OpY − compatible with the restriction morphisms. q. F ). f∗ F ). V ∈ OpY with U ⊂ f t (V ). f∗ F ) Hom PSh(kX ) (f † G. 109 Proof. − − − lim G(V ) → F (U )}U ∈OpX −t → which deﬁnes ψ ∈ Hom PSh(kX ) (f † G. Let (5. Since the functor a is exact. Let us give a proof in the general case. Hence the functor G → f † ιY G from Mod(kY ) to PSh(kX ) is left exact. this is a family of maps {ϕU : G(V ) − F (U ).5. Equivalently.e. In other words.16) deﬁnes an exact sequence of functors from ((OpY )U )op to Mod(k). . The sequence (5.13). f −1 is left exact by (5. the other assertions follow by the adjunction property.
d. q. Then: Γ(X. One has natural isomorphisms of functors g∗ ◦ f ∗ f ◦ g −1 −1 (g ◦ f )∗ . Hence: MX a−1 M{pt} . U U fV / Proposition 5. F ) aX ∗ F.5. ABELIAN SHEAVES Consider morphisms of sites f : X → Y .9) and inverse image commutes with tensor product (see Exercise 5. Consider the morphisms of presites: (5. fV ∗ i−1 F . (ii) Let M ∈ Mod(k). U fV / By Proposition 5. − − − Proposition 5. The functoriality of inverse images follows by adjunction. Proof. Let F ∈ Mod(kX ).) There is a natural equivalence of categories ∼ Mod(k). Then i−1 f∗ F V Proof.110 CHAPTER 5.10 (ii)).8. F ). Examples 5. Recall that MX denotes the sheaf associated with the presheaf U → M .d.9.5. (Recall that {pt} − is the set with one element. (g ◦ f )−1 .17) X O iU f / iV YO V. Note that inverse image commutes with the functor a (see Exercise 5.e. g : Y → Z and g ◦ f : X → Z. The functoriality of direct images is clear by its deﬁnition. we shall identify these two categories.5. X .5. f V ∗ jU ∗ F . (i) Let F ∈ Mod(kX ).e. Let V ∈ OpY . set U = f t (V ) and denote by fV the restriction of f to V : (5. one has jV ∗ f∗ F We denote by aX the canonical map aX = X → {pt}. Mod(kpt ) − → F → Γ(pt. In the sequel.18) X jU f / jV Y V.10.8. q.
we denote by q1 and q2 the ﬁrst and second projection. hence a morphism: − For example. f −1 G) Γ(V . OY ) → Γ(V .11. f −1 OY will be the sheaf on X of holomorphic functions on Y deﬁned in a neighborhood of X. (iii) Let x ∈ X and denote by ix : {x} → X the embedding. If f is smooth (locally on X.5. if V is open in Y . f is isomorphic to a projection Y × Z → Y ). respectively. F ).13. G). Denote by δ : X → X × X the diagonal embedding and let F1 and F2 be in Mod(kX ). On X × Y . F 1 ⊗ F2 . Fx . In fact. U (v) Let X = Y Y . If F ∈ Mod(kX ) and G ∈ Mod(kY ) we set: F One can recover the functor ⊗ from . Then i−1 F x 111 (iv) Let iU : U → X be the inclusion of an open subset of X and let F be a sheaf on X. f∗ f −1 G) Γ(V V . if X is closed in Y and f is the injection.5.5. F ) for V ∈ OpU .5. Since aX = aY ◦ f . then − f −1 OY will be the subsheaf of OX consisting of functions locally constant in the ﬁbers of f . . Then the functor iS ∗ is exact. f∗ OX ) − ϕ→ϕ◦f We obtain a morphism OY → f∗ OX . Then f∗ f −1 G G ⊕ G. Let f : X → Y − be the natural map which induces the identity one each copy of Y . Let F ∈ Mod(kX ) and let x ∈ X. i−1 F ) Γ(V . the disjoint union of two copies of Y . we get − MX f −1 MY . − lim Γ(U ∩ V . (ii) Let iU : U → X be the embedding of an open subset U of X. then we have the isomorphisms Γ(V .12. Examples 5. We have: δ −1 (F1 −1 −1 F2 ) = δ −1 (q1 F1 ⊗ q2 F2 ) −1 −1 G = q1 F ⊗ q2 G. Then (iU ∗ i−1 F )x U V x f −1 OY → OX . − → Notation 5. G) ⊕ Γ(V . Let f : X → Y be a morphism of complex manifolds. DIRECT AND INVERSE IMAGES Let f : X → Y be a continuous map.5. Example 5. (i) Let iS : S → X be the embedding of a closed subset S of X. Then Γ(V . To − each open subset V ⊂ Y is associated a natural “pullback” map: deﬁned by: Γ(V .
Z Γ(Z. F ) → Γ(Z. Let Z be compact subset of X.19) the inclusion morphism.21) is not an isomorphism in general.6 Sheaves associated with a locally closed subset iZ : Z → X Let Z be a subset of X. Indeed. and for F ∈ Mod(kX ). F ). Proposition 5. F ) is zero in Γ(Z.21) is bijective. ABELIAN SHEAVES 5. A subset Z of a topological space X is relatively Hausdorﬀ if two distinct points in Z admit disjoint neighborhoods in X. Replacing X with an open set U containing Z in (5.112 CHAPTER 5.21) U ⊃Z lim Γ(U .20) Γ(X. F ). − − → The morphism (5. One endows Z with the induced topology. Then the natural morphism (5. hence s = 0 on an open neighborhood of Z. F ). − One denotes by sZ the image of a section s of F on X by this morphism. F Z ). we get the morphism Γ(U . There is a classical result which asserts that if X is paracompact (e. Z Z hence the morphism: (5. one says that X is Hausdorﬀ. if X is locally compact and countable at inﬁnity) and Z is closed. one sets: F Z = i−1 F. these deﬁnitions agree with the previous ones.20). F Z ). F ) such that si Z∩Ui = sZ∩Ui .1. F ) → Γ(Z. Then (IZ )op is ﬁltrant. F ) = Γ(Z. then sx = 0 for all x ∈ Z. One shall be aware that the morphism (5. then (5. If Z := U is open.6. Proof. F ). F deﬁnes a functor (IZ )op → Mod(k) and we get a morphism − (5. 2 Let s ∈ Γ(Z.21) is injective. if a section s ∈ Γ(U . There exist a ﬁnite family of open subsets {Ui }n covering Z and sections si ∈ Γ(Ui . The mor−1 phism F → iZ ∗ iZ F deﬁnes the morphism aX ∗ F → aX ∗ iZ ∗ i−1 F − − aZ ∗ i−1 F . relatively Hausdorﬀ in X and let F ∈ Mod(kX ). F ) → Γ(Z.g.21) is an isomorphism. Denote by IZ the category of open subsets con− taining Z (the morphisms are the inclusions). i=1 2 The proof may be skipped . One denotes by (5. If Z = X..
X ⊃ V → U ∩ V ⊂ U. 2) be an open subset of Ui such that Wi ⊃ Zi \ W and W1 ∩ W2 = ∅. (i) One deﬁnes the functor iU ! : Mod(kU ) → Mod(kX ) − −1 by setting iU ! := jU . − (iii) For F ∈ Mod(kX ). Such Wi ’s exist thanks to the hypotheses. For that purpose we may argue by induction on n and assume n = 2. Clearly. . one sets FU := iU ! i−1 F = jU jU ∗ F . −1 (ii) For F ∈ Mod(kX ). we may ﬁnd another family of open sets {Vi }n covering Z such i=1 ¯ that Z ∩ Vi ⊂ Ui . − U ⊃ V → V ⊂ X. 2). 0 otherwise.2. Then s1 Z1 ∩Z2 = s2 Z1 ∩Z2 . jU ∗ ). iU ∗ ) U (iU ! . U Note that iU ! i−1 F → F deﬁnes the morphism FU → F and F → iU ∗ i−1 F − − − U U deﬁnes the morphism F → ΓU F . ¯ Set Zi = Z ∩ Vi . and the functor ΓU ( X − X exact.d. one sets ΓU F := iU ∗ i−1 F = iU ∗ jU ∗ F .e. Deﬁnition 5. U −1 there is an isomorphism of functors jU ∗ iU : Mod(kX ) → Mod(kU ).6.6. i−1 ). Hence. F ) with tZ = s. U −1 Hence. Then s1 U 1 ∩U 2 = s2 U 1 ∩U 2 . This deﬁnes t ∈ Γ(U 1 ∪ U 2 .22) (FU )x Fx if x ∈ U. − iU : U → X. We shall glue together the sections si on a neighborhood of Z.5. Recall that we have the morphisms of sites jU : X → U. It follows that the functors iU ! : Mod(kU ) → Mod(kX ) and ( • )U : Mod(kX ) → − − • ) : Mod(k ) → Mod(k ) is left Mod(kX ) are exact. we have the functors Mod(kX ) o o iU ! i−1 U / Mod(kU ). U (iv) One sets kXU := (kX )U for short. SHEAVES ASSOCIATED WITH A LOCALLY CLOSED SUBSET113 Moreover. iU ∗ ) and (jU . Let W be an open neighborhood of Z1 ∩ Z2 such that s1 W = s2 W and let Wi (i = 1. iU ∗ and the pairs of adjoint functors (i−1 . The case of open subsets Let U be an open subset. Set U i = Wi ∪ W . we have the pairs of adjoint functors (i−1 . (i = 1. − Moreover (5. q.
(iii).5. (i) For F ∈ Mod(kX ). F ).6. Then (FU )V = FU ∩V . → α β Here α = (α1 .6.d.114 CHAPTER 5. The proofs of (i).3. (ii) We have Γ(U . Then the sequence 0 → FU → F → FS → 0 is exact in − − Mod(kX ).23) − − − 0 → FU1 ∩U2 − FU1 ⊕ FU2 → FU1 ∪U2 → 0. Let S ⊂ X be a closed subset and let F ∈ Mod(kX ). ABELIAN SHEAVES Proposition 5. jU ∗ F ) Hom kU (kU . −1 Note that F → iS ∗ iS F deﬁnes the morphism F → FS . S (ii) One sets kXS := (kX )S for short. one sets FS = iS ∗ i−1 F . F ) = Hom kX (jU jU ∗ kX .6. (i) We have FU F ⊗ kXU . Let U ⊂ X be an open subset and let F ∈ Mod(kX ). (ii) The functor ( • )S : Mod(kX ) → Mod(kX ). 0 otherwise. (iv) are obvious. Moreover − − (5. Hom kX (kXU . F → FS is exact.e. (iv) Let U1 and U2 be two open subsets of X. The case of closed subsets Deﬁnition 5.22). (ii) We have the isomorphisms −1 Hom kX (kXU . − − (i) Set U := X \ S. q. F ) (iii) Let V be another open subset. α2 ) and β = β1 −β2 are induced by the natural morphisms αi : FU1 ∩U2 → FUi and βi : FUi → FU1 ∪U2 . Proposition 5. using (5.24) (FS )x Fx if x ∈ S. F U ) F (U ). − − Proof.4. Let S be a closed subset of X. Then the sequence below is exact: (5. F ) Hom kU (jU ∗ kX . − .
24) still holds with Z instead of S. Recall that a morphism of sheaves which is locally an isomorphism is an isomorphism of sheaves. However. The proof is obvious. (i) Let M be a kmodule. The case of locally closed subsets A subset Z of X is locally closed if there exists an open neighborhood U of Z such that Z is closed in U . − − Proof. Then the sequence below is exact: (5. (iii) A sheaf F on X is locally constant if there exists an open covering X = i Ui such that F Ui is a constant sheaf of Ui . q. given two sheaves F and G.25) → − − 0 → FS1 ∪S2 − FS1 ⊕ FS2 → FS1 ∩S2 → 0. Then (FS )S = FS∩S . it may ∼ exist an open covering {Ui }i∈I of X and isomorphisms F Ui − GUi for all → i ∈ I. (v) Let S1 and S2 be two closed subsets of X. 5. LOCALLY CONSTANT AND LOCALLY FREE SHEAVES (iii) We have FS F ⊗ kXS . 115 (iv) Let S be another closed subset. using (5. Z = S ∩ U with U open and S closed in X. for some M ∈ Mod(k). − α β Here α = (α1 . Recall that the sheaf MX is the sheaf of locally constant M valued functions on X. It is also the sheaf associated with the constant presheaf U → M . although these isomorphisms are not induced by a globally deﬁned isomorphism F → G. One checks easily that this deﬁnition depends only on Z. − . one sets FZ := (FU )S .7 Locally constant and locally free sheaves Locally constant sheaves Deﬁnition 5. Equivalently. not on the choice of U and S. (ii) A sheaf F on X is constant if it is isomorphic to a sheaf MX . In this case.e.1.7.5. Moreover.24).d. (5.7. α2 ) and β = β1 −β2 are induced by the natural morphisms αi : FS1 ∪S2 → FSi and βi : FSi → FS1 ∩S2 .
Γ(Y . one gets an isomorphism of sheaves Kα U − CX U . Let U be an open disk in X centered at z0 . and let f : X → Y be the map z → z 2 . where − z denotes a holomorphic coordinate on C.7. is a constant sheaf. Then (i) (M ⊗ N )X MX ⊗ NX . Examples 5.116 CHAPTER 5. f −1 D is isomorphic to the disjoint union of two copies of D. If D is an open disk in Y . Lemma 5. f ∈ O(X) and P f = 0 implies f = 0. (ii) Let X = Y = C \ {0}. Recall that for M ∈ Mod(k). where α ∈ C \ Z. n ∈ Mod(k). (i) If X is not connected it is easy to construct locally constant sheaves which are not constant. However.3. that we simply denote by exp(t). let X = U1 U2 be a covering by two nonempty open subsets. Hence Γ(X. MX is the constant sheaf with stalk M . (ii) (Hom (M. Kα ) = 0. the sheaf 2 f∗ kX D is isomorphic to kD . Let us denote by Kα the kernel of P acting on OX . and Kα is a locally constant sheaf of rank one on C \ {0} which is not constant. Then the sheaf which is 0 on U1 and MU2 on U2 is locally constant and not constant. the constant sheaf of rank two on D.7. f∗ kX ) = Γ(X.7. Consider the sheaf CX · exp(t) consisting of functions which are locally a constant multiple of exp(t). ABELIAN SHEAVES Example 5. NX ). with U1 ∩ U2 = ∅. Hence. hence. N ))X . which shows that the sheaf f∗ kX is locally constant but not constant.4. Let us show that all locally constant sheaf on the interval [0. Indeed. which shows → that Kα is locally constant. Consider X = R and consider the Cvalued function t → exp(t). Let M. We have a commutative diagram of sheaves on U : OX exp(−A(z)) ∂ z ∂z −α / OX / OX ∂ ∂z 1 z exp(−A(z)) OX ∼ Therefore. 1] are constant. of rank one. On the other hand. kX ) = k. Let M ∈ Mod(k) with M = 0. (iii) Let X = C \ {0} with holomorphic coordinate z and consider the dif∂ ferential operator P = z ∂z − α.2. Hom kX (MX . and let A(z) denote a primitive of α/z in U . Clearly CX ·exp(t) is isomorphic to the constant sheaf CX .
.7. Γ(U1 ∩ U2 . then the composition M is the identity of M . M ) by Lemma 5. there is Mi ∈ Mod(k) and isomorphisms θi : ∼ F Ui − →(Mi )X Ui (i = 1. − (iii) Moreover. Then α1 and α2 will glue ∼ together to deﬁne an isomorphism F − MX . ∼ ∼ F0 ← Γ(I.7. Hom (MX .5. the morphism Γ(I. LOCALLY CONSTANT AND LOCALLY FREE SHEAVES The proof is left as an exercise. 1]. MX )). we ﬁnd that θ12 extends as an iso− morphism θ : MX MX all over X. n) such that F is constant on Ui . F ) → Ft is an isomorphism. (ii) F Ui (i = 1. By induction.7.5. Now deﬁne the isomorphisms: αi : ∼ F Ui − →(MX )Ui by α1 = θ1 and α2 = θU2 ◦ θ2 . Proof. 2) is a constant sheaf. → q. 2).d. . (ii) In particular. M ) → Γ(X. Then F is a constant sheaf. θ12 deﬁnes an invertible element of Hom (M. Proof. (i) We may ﬁnd a ﬁnite open covering Ui . Let I denote the interval [0. (ii)–(iii) are obvious. By the hypothesis. Proposition 5. Using the map Hom (M. Let F be a sheaf on X and assume that: (i) U12 = U1 ∩ U2 is connected and non empty. Hom (MX .e. Then F is a constant sheaf. . 117 Lemma 5. (i) Let F be a locally constant sheaf on I. we may assume that n = 2. . Then the result follows from Lemma 5. if F = MI for a kmodule M . if t ∈ I. Ui ∩ Ui+1 (1 ≤ i < n) is non empty and connected and Ui ∩ Uj = ∅ for i − j > 1.7. M1 M2 and we may assume M1 = M2 = M . MI ) − F1 − → M .6.5. let X = U1 ∪ U2 be a covering of X by two open sets. Hence.d. q. (i = 1. deﬁne the isomorphism −1 ∼ θ12 = θ1 ◦ θ2 : MX U1 ∩U2 − MX U1 ∩U2 .7. Since U1 ∩ U2 is connected and → non empty.e. Since U1 ∩ U2 is non empty and connected. MX )) Hom (M.4. M ).
then Ker ϕ and Coker ϕ − will be Amodules. One also naturally deﬁnes the notion of an Alinear morphism of Amodules. of ﬁnite rank) if there exists an open covering X = ∪i Ui such that LUi is isomorphic to a direct sum of k copies (resp. and the operations (addition. Hence we have deﬁned the category Mod(A) of Amodules. The category Mod(A) is clearly an additive subcategory of Mod(kX ). 0 (ii) On a topological space. A(U ) is endowed with a structure of a kalgebra. F (U ) is an A(U )module and the action of A(U ) on F (U ) commutes to the restriction morphisms. Examples 5. a kX algebra) A on X is a sheaf of kmodules such that for each U ⊂ X. Deﬁnition 5. The sheaf DbX is a CX module.7.8. (ii) A locally free sheaf of rank one is called an invertible sheaf. and let X = i∈I Ui be an open covering of X. Uijk = Uij ∩ Uk . If A is a sheaf of rings. equivalently. Moreover. A sheaf of Zalgebras is simply called a sheaf of rings. Moreover. First. One checks easily that the category Mod(A) is abelian. .118 Locally free sheaves CHAPTER 5. We shall construct locally constant and locally free sheaves by gluing sheaves in the §5. to a ﬁnite direct sum) of AUi . One sets Uij = Ui ∩ Uj . n (iii) If X is open in C . an Amodule) as follows: for each open set U ⊂ X. and the natural functor Mod(A) → Mod(kX ) is exact and faithful (but − not fully faithful). the category Mod(A) admits inductive and projective limits and ﬁltrant inductive limits are exact. the sheaf CX is a CX algebra. if ϕ : F → G is a morphism of Amodules. (i) Let A be a kalgebra. consider a sheaf F on X. (i) A sheaf L of Amodules is locally free of rank k (resp. multiplication) commute to the restriction morphisms.8 Gluing sheaves Let X be a topological space. one deﬁnes in an obvious way the notion of a sheaf F of (left) Amodules (or simply. The constant sheaf AX is a sheaf of kalgebras.7. ∞ ∞ the sheaf CX is a CX algebra.8. Now consider a sheaf of rings A. 5.7. the sheaf OX is a CX algebra. If X is open in Rn . ABELIAN SHEAVES A sheaf of kalgebras (or.
Let X = S1 be the 1sphere. Then clearly: → 119 (5. q. deﬁne F (V ) as the submodule of i∈I Fi (V ∩ Ui ) consisting of families {si }i such that for any (i. Assume that k = C. GLUING SHEAVES −1 ∼ set Fi = F Ui . Then the isomor→ → −1 ∼ − → phisms λi ◦ θi : F Ui → GUi will glue as an isomorphism G − F on X. j) ∈ I × I. Clearly.2. and consider a covering ± of X by two open connected intervals U1 and U2 . if the Fi ’s are locally constant.8. the function / 1 θ → exp(iβθ) is not well deﬁned on S since it does not take the same value at 0 and at 2π. (i) Existence. j) an isomorphism of ∼ sheaves θji : Fi Uij − Fj Uij . and the isomorphisms θi ’s are induced by the projections − k∈I Fk (V ∩ Uk ) → Fi (V ∩ Ui ). θ− = α.1.e. If β ∈ Z. where ∼ is the relation which identiﬁes 0 and 2π. Moreover. Assume k is a ﬁeld. Let α ∈ k × .5. hence is well deﬁned on S1 . However. Theorem 5. One checks that the presheaf so obtained is a sheaf. Let θi : F Ui − Fi and λi : GUi − Fi . θij ◦ θjk = θik on Uijk . θi : F Ui − Fi .26) θii = id on Ui .d. . One deﬁnes a locally constant sheaf Lα on X of rank one over k by gluing kU1 and kU2 as follows. then F is locally constant. the sheaf CX · exp(iβθ) of functions which are a constant multiple of the function exp(iβθ) is welldeﬁned on each of the intervals U1 and U2 . {θi }i∈I ) is unique up to → unique isomorphism. 2π]/ ∼. ∼ ∼ (ii) Unicity. For each open subset V of X. Let us show that one can reconstruct F from the data of a 1cocycle. One can give a more intuitive description of the sheaf Lα as follows.26).8. Assume to be given for each pair (i. by Proposition 5. Example 5. these isomorphisms satisfying the conditions → (5. Let X = i∈I Ui be an open covering of X and let Fi be a sheaf on Ui . Let ε ε − θε : kU1 U12 → kU2 U12 (ε = ±) be deﬁned by θ+ = 1. and recall that k × denote the multiplicative group k \ {0}. although it does not have any global section. Then there exists a sheaf F on X and for each i isomorphisms θi : ∼ F Ui − Fi such that θj = θji ◦ θi .4. Sketch of proof. θji = θj ◦ θi . The family of isomorphisms {θij } satisfying conditions (5. Let U12 denote the two connected components of U1 ∩ U2 .8. Choose β ∈ C with exp(iβ) = α. Let us identify S1 with [0.2. θji (si V ∩Uji ) = sj V ∩Uji .26) is called a 1cocycle. (F.
120
CHAPTER 5. ABELIAN SHEAVES
Example 5.8.3. Consider an ndimensional real manifold X of class C ∞ , and ∼ let {Xi , fi } be an atlas, that is, the Xi are open subsets of X and fi : Xi − Ui → ∞ n i is a C isomorphism with an open subset Ui of R . Let Uij = fi (Xij ) and denote by fji the map (5.27)
i i fji = fj Xij ◦ fi−1 Uij : Uij → Uij . − j
The maps fji are called the transition functions. They are isomorphisms of class C ∞ . Denote by Jf the Jacobian matrix of a map f : Rn ⊃ U → V ⊂ − n R . Using the formula Jg◦f (x) = Jg (f (x)) ◦ Jf (x), one gets that the locally constant function on Xij deﬁned as the sign of the Jacobian determinant det Jfji of the fji ’s is a 1cocycle. It deﬁnes a sheaf locally isomorphic to ZX called the orientation sheaf on X and denoted by orX . Remark 5.8.4. In the situation of Theorem 5.8.1, if A is a sheaf of kalgebras on X and if all Fi ’s are sheaves of AUi modules and the isomorphisms θji are AUij linear, the sheaf F constructed in Theorem 5.8.1 will be naturally endowed with a structure of a sheaf of Amodules.
1 Example 5.8.5. (i) Let X = P1 (C), the Riemann sphere. Then ΩX := ΩX is locally free of rank one over OX . Since Γ(X; ΩX ) = 0, this sheaf is not globally free. (ii) Consider the covering of X by the two open sets U1 = C, U2 = X \ {0}. One can glue OX U1 and OX U2 on U1 ∩ U2 by using the isomorphism f → z p f (p ∈ Z). One gets a locally free sheaf of rank one. For p = 0 this sheaf is not free.
Exercises to Chapter 5
Exercise 5.1. Let S (resp. U ) be a closed (resp. an open) subset of X and let F ∈ Mod(kX ). (i) Prove the isomorphism Γ(X; FS ) Γ(S; F S ). (ii) Construct the morphism Γ(X; FU ) → Γ(U ; F ) and prove that it is not − an isomorphism in general. Exercise 5.2. Assume that X = R, let S be a nonempty closed interval and let U = X \ S. (i) Prove that the natural map Γ(X; kX ) → Γ(X; kXS ) is surjective and − deduce that Γ(X; kXU ) 0. (ii) Let x ∈ R.Prove that the morphism kX → kX{x} does not split. −
Exercises to Chapter 5
121
Exercise 5.3. Let F ∈ Mod(kX ). Deﬁne F ∈ Mod(kX ) by F = x∈X F{x} . (Here, F{x} ∈ Mod(kX ) and the direct sum is calculated in Mod(kX ), not in PSh(kX ).) Prove that Fx and Fx are isomorphic for all x ∈ X, although F and F are not isomorphic in general. Exercise 5.4. Let Z = Z1 Z2 be the disjoint union of two sets Z1 and Z2 in X. (i) Assume that Z1 and Z2 are both open (resp. closed) in X. Prove that kXZ kXZ1 ⊕ kXZ2 . (ii) Give an example which shows that (i) is no more true if one only assume that Z1 and Z2 are both locally closed. Exercise 5.5. Let X = R2 , Y = R, S = {(x, y) ∈ X; xy ≥ 1}, and let f : X → Y be the map (x, y) → y. calculate f∗ kXS . − Exercise 5.6. Let f : X → Y be a continuous map, and let Z be a closed − ∼ subset of Y . Construct the natural isomorphism f −1 kY Z − kX(f −1 Z) . → Exercise 5.7. Assume that X is a compact space and let {Fi }i∈I be a ﬁltrant inductive system of sheaves on X. Prove the isomorphism lim Γ(X; Fi ) − → i ∼ Γ(X; lim Fi ). − → − →
i
Exercise 5.8. Let S be a set endowed with the discrete topology, let p : X × S → X denote the projection and let F ∈ Mod(kX×S ) and set Fs = F X×{s} . − Prove that p∗ F s∈S Fs . Exercise 5.9. Let f : X → Y be a continuous map and let G ∈ PSh(kY ). − † ∼ Prove the isomorphism (f G)a − f −1 (Ga ). → Exercise 5.10. (i) Let f : X → Y be a morphism of sites, let F ∈ Mod(kX ) − and let G ∈ Mod(kY ). Prove that there is a natural isomorphism in Mod(kY ) ∼ Hom kY (G, f∗ F ) − f∗ Hom kX (f −1 G, F ). → (ii) Let G1 , G2 ∈ Mod(kY ). Prove that there is a natural isomorphism in Mod(kX ) ∼ f −1 (G1 ⊗kY G2 ) − f −1 G1 ⊗kX f −1 G2 . → Exercise 5.11. Let X = i Ui be an open covering of X and let F ∈ PSh(kX ). Assume that F Ui is a sheaf for all i ∈ I. Prove that F is a sheaf. (Hint: compare F with the sheaf constructed in Theorem 5.8.1.)
122
CHAPTER 5. ABELIAN SHEAVES
Exercise 5.12. Let M be a kmodule and let X be an open subset of Rn . Let F be a presheaf such that for any non empty convex open subsets U ⊂ X, there exists an isomorphism F (U ) M and this isomorphism is compatible to the restriction morphisms for V ⊂ U . Prove that the associated sheaf is locally constant. Exercise 5.13. Prove Lemma 5.7.4. Exercise 5.14. Assume k is a ﬁeld, and let L be a locally constant sheaf of rank one over k (hence, L is locally isomorphic to the sheaf kX ). Set L∗ = Hom (L, kX ). ∼ ∼ (i) Prove the isomorphisms L∗ ⊗ L − kX and kX − Hom (L, L). → → (ii) Assume that k is a ﬁeld, X is connected and Γ(X; L) = 0. Prove that L kX . (Hint: Γ(X; L) Γ(X; Hom (kX , L).)
[18]. Then f∗ F is injective in Mod(kY ). and the fact that the functor f −1 is exact.d. 123 . 6. (ii) Let iU : U → X be an open embedding and let F ∈ Mod(kX ) be injective. Then. Then i−1 F is injective in Mod(kU ). Assume that F ∈ Mod(kX ) is injective. [3]. This allows us to derive all left exact functors we have constructed. (i) follows immediately from the adjunction formula: Hom kX (f −1 (·). Finally. using the tools on simplicial complexes of §3.3. U Theorem 6. [17]. Next we prove an important theorem which asserts that the cohomology of constant sheaves is a homotopy invariant. we construct resoˇ lutions of sheaves using open or closed Cech coverings.1. (ii) follows from (i) since i−1 jU ∗ . f∗ F ) q. Lemma 6.1.1. F ) Hom kY (·. (i) Let X and Y be two topological spaces and let f : X → − Y be a morphism of sites. Some references: [12].1 Cohomology of sheaves A sheaf F of kmodules is injective if it is an injective object in the category Mod(kX ). The category Mod(kX ) admits enough injectives.2. we apply this result to calculate the cohomology of some classical manifolds. [10].e. U Proof.Chapter 6 Cohomology of sheaves We ﬁrst show that the category of abelian sheaves has enough injective objects.
F ) Extj (kXU . S. one denotes by iZ : Z → X the embedding of Z in X and by aZ the map Z → {pt}. (i) If U is open in X. Recall k that. F ) H j (S. Then for F. There exists an f : X − ˆ injective sheaf G0 on X and a monomorphism 0 → f −1 F → G0 .3) Proposition 6. − q. ·)(F ).1). a locally closed) subset of X. choose an injective module Ix together with a monomorphism Fx Ix and deﬁne the sheaf F 0 on X by setting Γ(U. a closed. F ) X X X (6. F ∈ Mod(kX ). As usual.1. It is now possible to derive all left exact functors deﬁned on the category of sheaves. G ∈ Mod(kX ). since − on an open subset U of X it is deﬁned by F (U ) → x∈U Fx . H j (U . Let F ∈ Mod(kX ) and let U (resp. F ). the natural morphism Hom kX (G. (ii) Assume X is discrete.e. COHOMOLOGY OF SHEAVES Proof. (ii) If S is closed in X.2) By (6. then H j (X. F • )). Fx ) x∈X is an isomorphism.124 CHAPTER 6. For each x ∈ X. we have (6. Z) be an open (resp. F 0 ) = x∈U Ix . FS ) . One sets (6. Recall that for a sheaf F on − X. the result follows from Mod(kX ) Mod(k). F U ). H j (U . resp. To − − conclude. F ) = Rj Γ(U . The derived functors of these two bifunctors are respectively denoted by Ext jk and Extj . resp. and let ˆ → X be the identity map. Then f∗ G0 − − is injective in Mod(kX ) and the sequence 0 → f∗ f −1 F → f∗ G0 is exact. Since products are exact in Mod(k). Therefore the sequence 0 → F → F 0 is exact and F 0 − − is injective. ˆ (iii) Let X denote the set X endowed with the discrete topology.1) H j (Hom kX (G. (i) When X = {pt}. F ) → − Hom k (Gx . then H j (U . notice that the morphism F → f∗ f −1 F is a monomorphism. k X H j (U . Then Extjk (G. we have set: Γ(Z. X Choose an injective resolution F • of F . F S ). (F 0 )x = Ix . Since the topology on X is discrete. F ) = Γ(Z. Let F ∈ Mod(kX ). as well as the bifunctors Hom kX and Hom kX .3. it follows that a sheaf F is injective as soon as each Fx is injective. F ) is calculated as follows. Note that aZ = aX ◦ iZ . for G. the kmodule Extjk (G.d. F Z ).
we get a long exact sequence (6. Consider an injective resolution F → F • of F .1. F ) → H j (K. COHOMOLOGY OF SHEAVES 125 (iii) If K is compact and relatively Hausdorﬀ in X. Then − H j (K.1. (iii) The result is true for j = 0 by Proposition 5. then the natural morphism lim H j (U . F • )) − → U ⊃K U ⊃K U ⊃K lim H (Γ(U . Applying − − − − a left exact functor Ψ to it. FS12 ) − − − j+1 → H (X. F ) → H k (X. F ).1. ·) with other functors. Let 0 → F → F → F → 0 be an exact sequence of sheaves. we obtain a long exact sequence 0 → Ψ(F ) → Ψ(F ) → · · · → Rj Ψ(F ) → Rj Ψ(F ) → Rj+1 Ψ(F ) → · · · − − − − − − − For example.4) H k−1 (X. FS ) (Rj aX ∗ )(iS ∗ i−1 )F S H j (S. FS1 ∪S2 ) → · · · − − . Let F ∈ Mod(kX ). − → q. F • )) j • H j ( lim Γ(U .10 (iv)).6. FS2 ) → H j (X. such as f∗ . Proposition 6.4.6. applying the functor Γ(X. F ) → · · · − − − − There are similar results. F ) → H k (X.6. − − → U ⊃K Proof.e. Let S1 and S2 be two closed subsets of X and set S12 = S1 ∩ S2 . F U ). F ) Rj (aU ∗ i−1 )F U j H (U . (Rj aU ∗ )i−1 F U The second isomorphism follows from the fact that i−1 is exact and sends U injective sheaves to injective sheaves (Lemma 4. FS1 ) ⊕ H j (X. Rj (aX ◦ iS ∗ )i−1 F S The second isomorphism follows from the fact that iS ∗ is exact and sends injective sheaves to injective sheaves (Lemma 4. F ) H j (Γ(K. F ) is an isomorphism. There is a long exact sequence · · · → H j (X. (ii) We have the chain of isomorphisms: H j (X. FS1 ∪S2 ) → H j (X. replacing Γ(X.10 (iv)). ·). F )) − → lim H j (U . F S ).d. F ) → H k (X.6. (i) We have the chain of isomorphisms: H j (U .
dp+1 ◦ dp = 0 and we obtain a complex (6. F ) → H j (U12 . F ) → H j (U1 .5. we denote by Ja = J ∪ {a} the ordered / subset of the ordered set I and we denote by rJ.126 CHAPTER 6.1. we denote by J its cardinal and for J = {i0 < · · · < ip }.5) δJ.a := rJ. We set (6. Let U1 and U2 be two open subsets of X and set U12 = U1 ∩ U2 . COHOMOLOGY OF SHEAVES Proof.a ( • ) ⊗ ea ∧ • : FSJ ⊗ eJ → FSJ a ⊗ eJa . p FS S= i∈I Si −1 FS = F S . − − − q. − p d p : FS → F S .a (J ⊂ I. − p+1 The morphisms δJ. Let F ∈ Mod(kX ). we set p+1 e i0 ∧ · · · ∧ e ip ∈ ZI . Proposition 6.e. − − − q. F ) − − − → H j+1 (U1 ∪ U2 . In this section. Let S = {Si }i∈I be a family indexed by I of closed subsets of X and let F ∈ Mod(kX ). 6. := J=p+1 FSJ ⊗ e J . For J ⊂ I we set SJ := ∩j∈J Sj . a ∈ J) in (6. Apply the functor Γ(X.6) • 0 FS := 0 → FS −→ FS − FS − · · · . ·) to the exact sequence of sheaves 0 → − FS1 ∪S2 → FS1 ⊕ FS2 → FS12 → 0. F ) to the exact sequence of sheaves 0 → − kXU12 → kXU1 ⊕ kXU2 → kU1 ∪U2 → 0 and use (6.d.a : FSJ → FSJ a the natural − restriction morphism.5) deﬁne the morphisms / Clearly.3).2 ˇ Cech complexes for closed coverings Let I be a ﬁnite totally ordered set. Apply the functor Hom kX ( • . For J ⊂ Iord and a ∈ I. a ∈ J.e. F ) ⊕ H j (U2 . we shall follow the same notations as for Koszul complexes. For J ⊂ I. There is a long exact sequence · · · → H j (U1 ∪ U2 . F ) → · · · − − Proof. − − → 1 → d−1 d0 d1 .d.
sij : FSa → FSij . 1]. 2}).6) is a Koszul complex K • (M. Example 6. Set M = Fx .3. −sb . . d−1 = s1 . this complex is exact. We get the exact complex of sheaves 0 → F −→ FS0 ⊕ FS1 ⊕ FS2 − FS12 ⊕ FS02 ⊕ FS01 − FS012 → 0. Proof. Hence. (i) Let f0 and f1 be two continuous maps from X to Y . First. − − → → − Let us denote by a − − si : F → FSi .6. sb : FSij → FS012 (a. One says that f is a homotopy − equivalence if there exists g : Y → X such that f ◦ g is homotopic to − idY and g ◦ f is homotopic to idX . ·) = f0 and h(1. In the sequel. where the Si ’s are closed subsets. (iii) One says that a topological space X is contractible if X is homotopic to a point {x0 }. It is enough to check that the stalk of the complex (6. Then the complex (6. i. INVARIANCE BY HOMOTOPY 127 Proposition 6.6) is exact. q. Then 0 −s1 s2 s0 . 1. − k d−1 d0 d1 the natural morphisms. In such a case one says that X and Y are homotopic.6) at each x ∈ X is exact. Let X and Y be two topological spaces.d. ϕ) where ϕ = {ϕi }i∈I and all ϕi are idM . sb).3 Invariance by homotopy In this section. k) ∈ {0. d0 = s0 02 02 2 0 1 0 1 −s01 s01 0 s2 6. we denote by I the closed interval I = [0. (ii) Let f : X → Y be a continuous map. Then the complex (6. ·) = f1 .1.e. we shall prove that the cohomology of locally constant sheaves is an homotopy invariant.3.2. 12 12 0 −s2 d1 = (sb. Assume that X = S0 ∪ S1 ∪ S2 . The sequence ϕ being both regular and coregular. j.1.2. . we deﬁne what it means. Deﬁnition 6. One says that f0 and f1 are homotopic if there exists a continuous map h : I × X → Y − such that h(0. Consider a family S = {Si }i∈I of closed subsets of X indexed by a ﬁnite totally ordered set I. we may assume that x belongs to all Si ’s.2.
Remark that − aX aY ◦ f . The embedding f : Sn−1 → Rn \{0} is a homotopy equivalence. G) → H j (X. we get the morphism Rj (aX ∗ ◦ f −1 ) → (Rj aX ∗ ) ◦ f −1 . f −1 G). − Lemma 6. x) = idX − and h(1. f1 : X (6. this means that there exists h : I × X → X such that h(0. p is the projection − and ft = h ◦ it . x) is the map x → x0 . A topological space is contractible if and only if there exist g : {x0 } → X − − and f : X → {x0 } such that f ◦ g is homotopic to idX . Statement of the main theorem Let f : X → Y be a continuous map and let G ∈ Mod(kY ). Then g ◦ f = idX − and f ◦ g is homotopic to idY .3. A non empty convex set in V as well as a closed non empty cone are contractible sets. Indeed. 1. it : X {t} × X → I × X is the embedding. Examples 6.10. Let f : X → Y and g : Y → Z be continuous maps.2. denote by g : Rn \ {0} → Sn−1 the map x → x/x. t = 0.3. Replacing x0 with g(x0 ). (i) Let V be a real vector space. one gets the diagram X {t} × X it / I ×X p h / Y X where t ∈ I. (ii) Let X = Sn−1 be the unit sphere of the Euclidian space Rn and let Y = Rn \{0}.6. The homotopy is given by the map h(x. The morphism of functors id → f∗ ◦ f −1 deﬁnes the morphism − Rj aY ∗ → Rj (aY ∗ ◦ f∗ ◦ f −1 ) − Rj (aX ∗ ◦ f −1 ).7) CHAPTER 6. Note that contractible implies non empty. t) = (t/x + 1 − t)x. COHOMOLOGY OF SHEAVES Y are homotopic. Then − − g j ◦f j = (f ◦ g) j .128 If f0 .8) f j : H j (Y . One checks easily that the relation “f0 is homotopic to f1 ” is an equivalence relation.3. − from which we deduce the natural morphisms: (6. Using Theorem 4.
f1 : X Y be two homotopic maps. Let F ∈ Mod(kI ). t].t]) for short. For F ∈ Mod(kI ) and a closed interval [0. t] ⊂ I.3.5. Proof of the main theorem In order to prove Theorem 6. The aim of this section is to prove: Theorem 6. The morphisms of functors id → f∗ ◦ f −1 and id → g∗ ◦ g −1 − − deﬁne aZ ∗ → aZ ∗ ◦ g∗ ◦ g −1 aY ∗ ◦ g −1 − → aY ∗ ◦ f∗ ◦ f −1 ◦ g −1 aX ∗ ◦ f −1 ◦ g −1 . Then: 1 The proof of this theorem may be skipped .3.3. If G = MY for some M ∈ Mod(k). ft−1 G). 1. 1). F ) instead of H j ([0. INVARIANCE BY HOMOTOPY 129 Sketch of proof.3.e. Applying the functor Rj ( • ) we ﬁnd the commutative diagram: R j aZ ∗ R j aZ ∗ / / Rj (aX ∗ f −1 g −1 ) O / Rj (aX ∗ f −1 ) ◦ g −1 O / / Rj (aY ∗ g −1 ) (Rj aY ∗ ) ◦ g −1 .6.4 (Invariance by homotopy Theorem). F [0. q.4. f1 G) such − j j that θ j ◦ f0 = f1 . f0 G) → H j (X. G) → H j (X. we need several preliminary results. we write H j ([0. for t = 0. identifying ft−1 MY with MX (t = 0. Consider the two morphisms ft j : H j (Y . 1 Let f0 . Lemma 6. − One checks easily that the composition aZ ∗ → aY ∗ ◦ g −1 → aX ∗ ◦ f −1 ◦ g −1 − − is the same as the morphism aZ ∗ → aZ ∗ ◦ g ◦ f ∗ ◦ g ◦ f −1 − aX ∗ ◦ (g ◦ f )−1 . and let G be a locally constant sheaf on Y .d. we have f1 j = f0 j . j jjjj jjjj jjj jjjj f j (Rj aX ∗ ) ◦ f −1 ◦ g −1 ) 4 The composition of the arrows on the top gives the morphism (g ◦ f ) j and the composition of the arrows on the bottom gives the morphism g j . t]. then. − −1 −1 Then there exists an isomorphism θ j : H j (X.
t0 ].7). − we obtain: (6. G) − → H j (Ix .e. F ) = 0. Let G ∈ Mod(kI×X ). F ) = lim H j ([0. . GIx ). Hence t0 ∈ J. consider the morphism: ft1 . F ) → H j ([0.t (s) = 0 implies f0. F ) − and let Since H j ({0}.1. q. Since H j ([0.t0 (s) = 0. G• )) − → x∈U x∈U lim H j (I × U . Let G• be an injective resolution of G. For 0 ≤ t ≤ t0 . Then f0. For 0 ≤ t1 ≤ t2 ≤ 1. F ) → H 0 ({t}.e. F ) ⊕ H j ([t. Lemma 6. then H 1 (I. J is an interval. F ) = 0. F ) → H j ({t}. there exists t < t0 with ft. We have the isomorphisms (H j (p∗ G• ))x H j ((p∗ G• )x ) H j (lim Γ(I × U .t (s) = 0. F ) = 0 for j ≥ 1. t]. Then (Rj p∗ G)x (Rj p∗ G)x H j (Ix . − Proof. t0 ].d.t2 : H j (I.9). F ) = 0. for all t < t0 . GIx ). t2 ]. F ). t]. F ) → H j ([t1 . COHOMOLOGY OF SHEAVES (i) For j > 1.d.1. t<t0 lim H j ([t. (ii) If F (I) → Ft is an epimorphism for all t ∈ I.3. 1]. t0 ].t0 (s) = 0. It remains to prove that J is closed. this interval is open. On the other hand.6. F ) ⊕ H j ([t. F ) H j ([0. f0. Recall that the maps p : I × X → X and it : X → I × X are deﬁned in (6. q. t0 ].3 (iii).t (s) = 0 for 0 ≤ t ≤ t. or else for j = 1 assuming H 0 (I. By (6. F ) → · · · − − − − For j > 1. Let j ≥ 1 and let s ∈ H j (I. F ) is surjective. Proof.130 CHAPTER 6. Let t0 = sup {t. consider the MayerVietoris sequence (Proposition 6.4): · · · → H j ([0. t ∈ J}. − − We also introduce the notation Ix := I × {x}. t0 ]. − → t>t0 J = {t ∈ [0. − → Hence. this implies f0. t]. Since f0. F ). we have 0 ∈ J. where the last isomorphism follows from Proposition 6. F ).9) H j ([0. t0 ]. one has H j (I.t (s) = 0}.
3. Let F ∈ Mod(kX ). − (ii) The morphisms itj : H j (I × X. we get that Rj (aX ∗ p∗ )(p−1 F ) Rj aX ∗ (p∗ p−1 F ) Rj aX ∗ F . q.3. Hence. (i) Since h−1 G is locally constant.3. End of the proof of Theorem 6. itj ◦ p j is the identity.6.e.3. Proof. ft j = itj ◦ h j . q.7. One has (p−1 p∗ G)(t.4.e. (i) The morphisms p j : H j (X.3. p−1 p∗ h−1 G) → H j (X.3.3.3. GIx ) G(t. Lemma 6.3. Let G ∈ Mod(kI×X ) be a locally constant sheaf. h−1 G) → H j (X. F ) → H j (I × X. and p j is an isomorphism by (i).e. INVARIANCE BY HOMOTOPY 131 Lemma 6. Set θ = i0j i1j f0 j = i0j ◦ h j = i0j i1j (ii) If G = MY . Then this group is 0 for j > 0 by Lemma 6. Here the last isomorphism follows from Proposition 5. Hence p j is an isomorphism. (ii) By Lemma 6. Let F ∈ Mod(kX ).9.d. Using Lemma 6. By Lemma 6. q.x) (p∗ G)x Γ(Ix . Proof.8. − itj : H j (I × X.3. −1 j i1 −1 .3.7. p−1 F Ix ). (i) We know that Rj p∗ (p−1 F ) = 0 for j ≥ 1.x) .d. Let x ∈ X and let t ∈ I. p−1 F ) are isomorphisms. ∼ Lemma 6. p−1 F ) → H j (X.5 and is isomorphic to (p−1 F )t.3. F ) are isomorphisms − and do not depend on t ∈ I. By − Lemma 6. MI×X = p−1 MX and itj does not depend on t q. Then ◦ h j = θ ◦ f0 j .6.6.x Fx for j = 0. Then F − p∗ p−1 F and (Rj p∗ )p−1 F = 0 → for j ≥ 1. i−1 p−1 p∗ h−1 G) t is an isomorphism.3.7.9. i−1 h−1 G) − t is an isomorphism. .9 (ii). p∗ and the object p−1 F . Then the natural morphism p−1 p∗ G → G is an isomorphism. the morphism p−1 p∗ h−1 G → h−1 G is an isomorphism by Lemma 6. Therefore itj : H j (I × X.e. itj which is the inverse of p j does not depend on t.10 (iii) to the functors aX ∗ .d. then h−1 G by Lemma 6. − Proof.6 one gets the isomorphism ((Rj p∗ )p−1 F )x H j (Ix .3.d. Applying Proposition 4.
(A particular case of the universal coeﬃcients formula. FZ ) → · · · − − − d Proof. G) → H j (X.3. q. Let F be a locally constant sheaf on X. Then H j (X. MX ) M and H j (X.6. MX ) H j (X. then Γ(X. the result follows from Theorem 4. Let g : Y → X be a map such that f ◦g and g ◦f are homotopic to the − identity of Y and X. Theorem 6.10.e. The kmodule H j (X. Then for all j there are natural isomorphisms H j (X.d.132 CHAPTER 6. Recall (Proposition 6.1.2. FZ ) p Γ(Z.12. kZ ) ⊗ M . G). let M be a ﬂat kmodule. q. ZJ is contractible or empty. Applying Proposition 6. Corollary 6. F ) is isomorphic to the jth cohomology object of the complex • 0 1 Γ(X. COHOMOLOGY OF SHEAVES Applications of Theorem 6.e.10) i∈I Zi be a ﬁnite covering of X by closed subsets for each non empty subset J ⊂ I.3. ·).9.1. MZ ).3.11. kX ) ⊗ M . F Z ).d. We shall follow the notations of Section 5.) In the situation of Theorem 6. FZ ) → Γ(X. If X is contractible and M ∈ Mod(k).3. Proof.3) that if Z is closed in X. • Γ(X. Corollary 6. MX ) 0 for j > 0.12. Proof. the cohomology of locally constant sheaves on topological spaces is a homotopy invariant. Let X = satisfying the condition (6. Clearly. MZ ) • Γ(X. Therefore the sheaves FZ (p ≥ 0) are acyclic with respect to the functor Γ(X. G).3. f G) → H (Y . respectively. Then: (f ◦ g) = g ◦ f j − idX = id j and (g ◦ f ) j = f j ◦ g j idY = id . f −1 G) − j j j −1 j j j and g : H (X. FZ ) := 0 → Γ(X.3.13. Then H j (X. Assume f : X → Y is a homotopy equivalence and let G − be a locally constant sheaf on Y .3. then Γ(X. by Corollary 6.6. MX ) is the jth cohomology object of the com• plex Γ(X.4 Corollary 6. We shall apply this result together with the technique of MayerVietoris sequences to calculate the cohomology of various spaces. f −1 G) H j (Y . In other words.11. Consider f j : H j (Y .
S×Z One checks that this complex is the simple complex associated with the double complex • Γ(X × Y . q. F G) is the pth cohomology object of the complex Γ(X × Y .10).14.d.) Let X and u Y be two topological spaces which both admit ﬁnite closed coverings satisfying condition (6. G• ) Z It remains to apply the result of Exercise 4. FS G• ) Z and this double complex is isomorphic to • Γ(X.14.3.10).e. Assume that k is a ﬁeld.6. Y ) satisfying condition (6. F G) i+j=p H i (X. notice that F G is locally constant on X × Y . on Y with ﬁber N ). To conclude. Proposition 6. INVARIANCE BY HOMOTOPY Since M is ﬂat we have for any bounded complex of modules N • : H j (N • ⊗ M ) Hence. Since the product of two contractible sets is clearly contractible.14. It may be convenient to reformulate the K¨ nneth formula by saying that u p H (X ×Y . First. G) be a locally constant sheaf on X with ﬁber M .10).j)∈I×J is a ﬁnite covering of X × Y satisfying condition (6. (resp. apply Theorem 6.12 to both sides. F G) is isomorphic to the pth cohomology object of the complex ( i H i (X. F ) ⊗ H j (Y . Next. Then there are natural isomorphisms: H p (X × Y . F )[−i]) ⊗ ( H j (Y .e. kZ )) ⊗ M 133 H j (N • ) ⊗ M. Let F (resp. • H j (Γ(X. Then H p (X × Y .) . G). Sketch of the proof. (F G)• ). j (See Exercise 4. the covering S × Z = {Si × Zj }(i. G)[−j]). Z = {Zj }j∈J ) a ﬁnite covering of X (resp.3.3. FS ) ⊗ Γ(Y . denote by S = {Si }i∈I (resp. (A particular case of the K¨ nneth formula.d. MZ )) • H j (Γ(X. q.
Lα ) is the kernel (resp.8.11) H j (Sn . MSn ) = M j = 0 or j = n. Example 6. the H 0 −1 1 0 −α acting on k 3 .2. + − n−1 ¯ ¯ and D ∩ D S .12. Let E be an Rvector space of dimension n + 1 and ˙ denote by E the set E \ {0}.134 CHAPTER 6.2: S1 = U1 ∪ U2 . COHOMOLOGY OF SHEAVES 6. Example 6. MD− ) → H j (Sn−1 . Recall that it can be deﬁned as follows. Then for j = 0 (resp.4. (See Deﬁnition 7. j (S1 . Consider the topological nsphere Sn . U1 ∩ U2 has two connected components + − U12 and U12 . F ) are the cohomology objects of the complex: 0 → FZ0 ⊕ FZ1 ⊕ FZ2 → FZ12 ⊕ FZ20 ⊕ FZ01 → 0. Let X be the circle S1 and let Zj ’s be a closed covering by intervals such that the Zij ’s are single points and Z012 = ∅. 1 and 0 otherwise.2. Let us prove that for n ≥ 1: (6. If one chooses an Euclidian norm on E.4 below.) cokernel) of the matrix 1 −1 1 0 Note that these kernel and cokernel are zero except in case of α = 1 which corresponds to the constant sheaf kX .2. MSn ) → · · · − .12) ¯ ¯ → H j (D + . then H j (S1 . It follows that if M is a kmodule. k denotes as usual a commutative unitary ring and M denotes a kmodule.3. for j = 1).1. MD+ ) ⊕ H j (D − .4 Cohomology of some classical manifolds Here. (See Example 6.4.) In other words. then one may identify Sn with the unit sphere in E. we ﬁnd that if F is a locally constant sheaf on X. 0 otherwise. Sn is the set of all halflines in E. the cohomology groups H j (X. ¯ ¯ ¯ ¯ We have Sn = D + ∪ D − . α ∈ k × and Lα denotes the locally constant sheaf of + rank one over k obtained by gluing kU1 and kU2 by the identity on U12 and − by multiplication by α ∈ k × on U12 . MSn−1 ) − − ¯ ¯ j+1 n − → H (S . Then Sn ˙ E/R+ . k is a ﬁeld. MS1 ) M for j = 0. d Consider the MayerVietoris long exact sequence (6. where R+ denotes the multiplicative group of positive real numbers and Sn is endowed with the quotient topology. − − − Recall Example 5. where D + and D − denote the closed hemispheres. Applying Theorem 6.4.
one obtains that the dimension of a ﬁnite dimensional vector space is a topological invariant.12) and Remark 4. MSn ) For n = 1. 2. Denote by a the antipodal map on Sn (the map deduced from x → −x) and denote by a n the action of a on H n (Sn . Notice that Sn is not contractible. Assume E is endowed with a norm  · . By (6. Then Z01 := Z0 ∩ Z1 S1 S1 . if V and W are two real ﬁnite dimensional vector spaces and are topologically isomorphic. it is the same as a rotation of angle π). In other words. MSn−1 ) −u / H n (Sn . . Hence the cohomology of a constant sheaf with stalk M on V \ {0} is the same as the cohomology of the sheaf MSn As an application. Let Tn denote the ndimensional torus. Tn (S1 )n . kT2 ) is k for j = 0.13). if V has dimension n.7. Using (6. we deduce: (6. The map x → x((1 − t) + t/x) deﬁnes an homotopy of X with the sphere Sn . Example 6. they have the same dimension.3. Example 6.2. then V \ {0} is homotopic to Sn−1 . H j (T2 .6. Let E be a real vector space of dimension n + 1. their cohomology is concentrated in degree 0. one deduces the commutative diagram: (6.4. Consider the two cylinders Z0 = S1 × I0 . MSn ) a n H n−1 (Sn−1 .4.4. Let us recover this result (when n = 2) by using MayerVietoris sequences.13) H n−1 (Sn−1 . One may represent T2 as follows. For example. MSn−1 ) a n−1 u / H n (Sn . and let X = E \ {0}.4. Then T2 (Z0 Z1 )/ ∼ where ∼ is the relation which identiﬁes S1 × {0} ⊂ Z0 with S1 × {0} ⊂ Z1 and S1 × {1} ⊂ Z0 with S1 × {1} ⊂ Z1 . 1]. Using the K¨ nneth formula. In fact. although one can prove that any locally constant sheaf on Sn for n ≥ 2 is constant. one gets that (if k is a ﬁeld) j H j (Tn . is k 2 for j = 1 and is 0 otherwise. COHOMOLOGY OF SOME CLASSICAL MANIFOLDS 135 Then the result follows by induction on n since the closed hemispheres being contractible. the map a is homotopic to the identity (in fact.14) a n acting on H n (Sn . MSn ) is (−)n+1 . kTn ) (k ⊕ u ⊗n k[−1]) . MSn ). Z1 = S1 × I1 where I0 = I1 = [0.
Applying the functor Γ(T2 . and − prove that this presheaf is a sheaf (the sheaf BR of Sato’s hyperfunctions on R). i Fi ) i H (X. and let U1 ⊂ U2 be two open subsets of C containing ω as a closed subset. we shall admit the following theorem: for any open subset U of the complex line C. In this exercise. Exercises to Chapter 6 Exercise 6. Exercise 6. j (iv) Prove that H j (X. Let ω be an open subset of R. H j (Z1 ) and H j (Z01 ) (since Z0 and Z1 are homotopic to S1 ). → → − − Although we know the groups H j (Z0 ).136 CHAPTER 6. OC ) 0 for j > 0. this sequence does not allow us to conclude. Assume that this family is locally ﬁnite. (iii) Prove that the restriction morphisms B(R) → B(ω) are surjective. COHOMOLOGY OF SHEAVES We have a short exact sequence of sheaves 0 → kT2 → kZ0 ⊕ kZ1 → kZ01 → 0 − − − − Write H j (X) instead of H j (X. (ii) Prove that if each Fi is injective. the morphism β is given by the matrix − idS1 idS1 easily recovers that H j (T2 . one has H j (U . − ∂ j (iv) Let Ω an open subset of C and let P = m aj (z) ∂z be a holomorj=1 phic diﬀerential operator (the coeﬃcients are holomorphic in Ω). we get the long exact sequence of Theorem 6. Fi ). Let {Fi }i∈I be a family of sheaves on X.2. kT2 ) is k for j = 0. One denote by B(ω) this quotient. Then one S1 . each x ∈ X has an open neighborhood U such that all but a ﬁnite number of the Fi U ’s are zero. (ii) Construct the restriction morphism to get the presheaf ω → B(ω). then i Fi is injective. Prove that i Fi • is an injective resolution of i Fi . Recall the Cauchy theorem which asserts that if Ω is simply connected and if am (z) .4: 0 → H 0 (T2 ) −0 H 0 (Z0 ) ⊕ H 0 (Z1 ) − H 0 (Z01 ) → H 1 (T2 ) − → → − α β1 α β0 α β −1 H 1 (Z0 ) ⊕ H 1 (Z1 ) − H 1 (Z01 ) → H 2 (T2 ) → 0. ·). Now we remark that identifying Z0 and Z1 to idS1 − idS1 . (i)Prove that ( i Fi )x i (Fi )x . unless we know the morphisms α or β. 2 and k 2 for j = 1. (i) Prove that the natural map O(U2 \ ω)/O(U2 ) → O(U1 \ ω)/O(U1 ) is an − isomorphism. that is. (iii) Let Fi • be an injective resolution of Fi . kX ) for short.1.1.
Calculate (Rj f∗ kZ )0 for j ≥ 0.) − − − − Exercise 6. Σ its boundary in Sn . n be integers ≥ 1 with n = p + q and let X = Rn endowed with the coordinates x = (x1 . q.) f 1 ι 1 . B the intersection of Sn with an open ball of radius ε (0 < ε << 1) centered in some point of Sn . S. Let p.5. 1 ¯ (Hint: denote by ι : S → D the embedding and assume that there exists a ¯ → S1 such that the composition f ◦ ι is the identity.4. t) ∈ R4 . Calculate i i=1 i H j (S. (ii) Calculate H j (X. kX S Y ).Exercises to Chapter 6 137 does not vanish on Ω. Calculate H j (X S Y . z. y. Let γ = {(x. kS ) for all j. (ii) Application (a). S and S two closed subsets of X and Y respectively. kXU ) for all j. S1 = {x ∈ X. then P acting on B(ω) is surjective Exercise 6. y ∈ S and f (x) = y. S the images of X. x2 + y 2 + z 2 = t2 } and let U = X \ γ. t > 0}. A closed subset Z of a space X is called a retract of X if − there exists a continuous map f : X → Z which induces the identity on Z. (i) Let F be a sheaf on X S Y . ZD ) − H 1 (S1 . Let Sn denote the unit sphere of the Euclidian space Rn+1 . One still denotes by X.3. .6. respectively. 1 ¯ Show that S is not a retract of the closed disk D in R2 . y. p x2 + 2 i=p+1 x2 = 1}. (i) Show that γ is contractible. t4 = x2 + y 2 + z 2 . .7. Let X = R4 and consider the locally closed subset Z = {(x. S S in X S Y . Exercise 6. Set S0 = {x ∈ X. Y. ZS1 ) − H 1 (D. Denote by f : Z → X the natural injection. Exercise 6. xn ). then P acting on O(Ω) is surjective. ZS1 ) is the identity. continuous map f : D − We get that the composition ¯ ¯ → − → H 1 (S1 . Write the long exact MayerVietoris sequence associated with X. Y. t) ∈ X = R4 . Deﬁne X S Y as the quotient space X Y / ∼ where ∼ is the relation which identiﬁes x ∈ X and y ∈ Y if x ∈ S. S = Σ and let Y and S be a copy of X and S. z. . Prove that if ω is an open subset of R and if P is a holomorphic diﬀerential operator deﬁned in a open neighborhood of ω. f : S S a topological isomorphism. n x2 = i=1 i n 1}. S = S0 ∪ S1 . Y. (Recall that there exists an exact sequence 0 → kXU → kX → kXγ → 0. Exercise 6. Let X and Y be two topological spaces. Set X = Sn \ B. . Same question by replacing the sphere Sn by the torus T2 embedded into R3 . (iii) Application (b).
7. Consider the unit ball Bn+1 = {x ∈ E. (Hint: − otherwise. x ≤ 1} and consider a map f : Bn+1 → Bn+1 . Prove that f has at least one ﬁxed point.) (Remark: the result of this exercise is known as the Brouwer’s Theorem.8. COHOMOLOGY OF SHEAVES Exercise 6.) . construct a map g : Bn+1 → Sn which induces the identity on Sn − and use the same argument as in Exercise 6.138 CHAPTER 6.
we shall admit some results treated with all details in [11]. We deﬁne the monodromy of a locally constant sheaf and prove the equivalence between the category of representations of the fundamental group and that of locally constant sheaves. [3]. We denote as usual by I the closed interval [0. [13]. Let X denote a topological space.Chapter 7 Homotopy and fundamental groupoid In this chapter we study locally constant sheaves of sets and sheaves of kmodules and introduce the fundamental group of locally connected topological spaces. to I × I. D is isomorphic to the closed unit disk. or else. 1 This chapter will not be treated during the course 2005/2006 139 . In this chapter. 7.1. with σ(0) = x0 and σ(1) = x1 . [22]. The two points x0 and x1 − are called the ends of the path. Note that topologically S1 I/ ∼ where ∼ is the equivalence relation on I which identiﬁes the two points 0 and 1.1 Fundamental groupoid Let us recall some classical notions of topology. 1 Some references: [11]. Note that topologically.1. 1] and by S1 the circle. We shall also consider the space D := I × I/ ∼ where ∼ is the equivalence relation which identiﬁes I × {0} to a single point (denoted a0 ) and I × {1} to a single point (denoted a1 ). [7]. Deﬁnition 7. (i) A path from x0 to x1 in X is a continuous map σ : I → X.
3.140 CHAPTER 7. arcwise connected. It is left to the reader to check that “homotopy” is an equivalence relation. one says they are homotopic with ﬁxed ends if moreover ϕ(s. This is equivalent to saying that there exists a continuous function ψ : D → X such that ψ(i. It is easily checked that the homotopy class of τ σ depends only on the homotopy classes of σ and τ . Deﬁnition 7. If σ is a path from x0 to x1 . (i) X is arcwise connected (or “path connected”) if given x0 and x1 in X. The set of homotopy classes of loops at x0 endowed with the above product is called the fundamental group of X at x0 and denoted π1 (X. resp. One can also consider a − loop as a path γ such that γ(0) = γ(1). Two loops are homotopic if they are homotopic as − paths. one can deﬁne the path σ −1 from x1 to x0 by setting σ −1 (t) = σ(1 − t). If σ is a path from x0 to x1 and τ a path from x1 to x2 one can deﬁne a new path τ σ (in this order) from x0 to x2 by setting τ σ(t) = σ(2t) for 0 ≤ t ≤ 1/2 and τ σ(t) = τ (2t − 1) for 1/2 ≤ t ≤ 1. HOMOTOPY AND FUNDAMENTAL GROUPOID (ii) Two paths σ0 and σ1 are called homotopic if there exists a continuous function ϕ : I × I → X such that ϕ(i. t) = σi (t) for i = 0. 0) = x0 . 1. Lemma 7. (iii) X is locally connected (resp. (ii) X is simply connected if any loop in X is homotopic to a trivial loop.2. we can deﬁne [τ ][σ] as [τ σ]. simply connected) open subsets. and [σσ −1 ] is the homotopy class of the trivial loop at x0 . 1. 1) = x1 for all s. Deﬁnition 7. Hence. . there exists a path with ends x0 and x1 .4. t) = σi (t) for i = 0. By this lemma. Let X be a topological space. − (iii) If the two paths have the same ends. A trivial loop is a constant map γ : S1 → {x0 }.1. ϕ(s. resp. The next result is left as an exercise. locally arcwise connected. The product [σ][τ ] is associative. x0 and x1 . Let us denote by [σ] the homotopy class of a path σ. − (iv) A loop in X is continuous map γ : S1 → X. x0 ). locally simply connected) if each x ∈ X has a neighborhood system consisting of connected (resp.1. the set of homotopy classes of loops at x0 is a group.1.
we get the functor (7. . if iU : U → X denotes the embedding of an open subset U of X. and if two paths γ0 and γ1 are homotopic in X. we shall make the hypothesis (7. Hence. − Proposition 7. x1 )={the set of homotopy classes of paths from x0 to x1 }. x ≥ 0. Hom Π1 (X) (x. the interval {(x. y = 0. Proof. Then X is arcwise connected but not locally arcwise connected. If σ is a path from x0 to x1 in X. −1 ≤ y ≤ 1} and the interval {(x. f1 : X → Y be two continuous maps and assume − f0 and f1 are homotopic.7.e. Deﬁnition 7. x > 0.7.3) iU ∗ : Π1 (U ) → Π1 (X). Example 7.6. it is connected. Then the two functors f0 ∗ and f1 ∗ are isomorphic.1). Hom Π1 (X) (x0 . and let γ : I → X be a path. x) = π1 (X. Note that for x ∈ X. 1. x).1) X is locally arcwise connected.2) (7. x) are isomorphic for x ∈ X. Consider a continuous map f : X → Y . then the two − groupoids Π1 (Y ) and Π1 (X) are equivalent. if X is connected. If γ is a path in X.d. then f ◦ γ is − a path in Y . x1 ).5.1. Let f0 . y). In R2 denote by X the union of the graph of the function y = sin(1/x).1. FUNDAMENTAL GROUPOID 141 Clearly. Let h : I × X → Y be a continuous map such that h(i.1. The fundamental groupoid Π1 (X) is the category given by Ob(Π1 (X)) = X. if X is arcwise connected. − In particular. − i = 0. q.1. if f : X → Y is a homotopy equivalence. Assume (7. all groups π1 (X. we get a functor: f∗ : Π1 (X) → Π1 (Y ). x = 0. If X is locally arcwise connected and connected. y). ·) = fi (·). then f ◦ γ0 and f ◦ γ1 are are homotopic in Y . then the map γ → σ −1 γσ deﬁnes an isomorphism π1 (X. Then h ◦ γ : I × I → Y deﬁnes a − − homotopy between f0 ◦ γ and f1 ◦ γ. Hence. In particular. x0 ) π1 (X. it is arcwise connected. In the sequel.
1. two paths γ and τ with the same ends are homotopic. This follows from Proposition 7. We refer to [11] a for proof. Indeed.10.1) and moreover X is non empty and connected. . y1 )) − Hom Π1 (X) (x0 . Assume that p1 ∗ γ0 is homotopic to p1 ∗ γ1 and p2 ∗ γ0 is homotopic to p2 ∗ γ1 . Proof. (x1 . (ii) Let x0 .4) (p1 ∗ × p2 ∗ )Π1 (X × Y ) → Π1 (X) × Π1 (Y ). As a corollary. 7. Denote by pi the projection from X × Y to X and Y respectively. Remark 7.e. x0 ). (iii) The map in (7. Indeed.9. y0 . The functor in (7. considered as a loop in the space X = S1 . hence a − − functor (7. x1 ) × Hom Π1 (Y ) (y0 . (ii) One has π1 (S1 ) Z and 1 ∈ Z corresponds to the identity map. one sets π1 (X) = π1 (X. HOMOTOPY AND FUNDAMENTAL GROUPOID Examples 7. if σ is a path in X and τ is a path in Y . the path σ × τ in X × Y satisﬁes (p1 ∗ × p2 ∗ )[σ × τ ] = [σ] × [τ ]. y0 ).5)Hom Π1 (X×Y ) ((x0 . Then all objects x ∈ Π1 (X) are isomorphic and after choosing x0 ∈ X.5) is injective. These projections deﬁne functors p1 ∗ : Π1 (X × Y ) → Π1 (X) and p2 ∗ : Π1 (X × Y ) → Π1 (Y ).142 CHAPTER 7.4) is obviously essentially surjective. One calls π1 (X) the fundamental group of X. let γ0 and γ1 be two paths in X × Y .2. (i) The functor in (7.1. Let X and Y be two topological spaces satisfying (7.1).4) is an equivalence. y1 ∈ Y .5) is surjective. (ii) Remark that X being arcwise connected. γ1 .7.1. but many results remain true without any change for sheaves of sets. x1 ∈ X. y1 ) → The map in (7. Let us show the isomorphism ∼ (7.1. The product of these two homotopies deﬁnes an homotopy from γ0 to q.d. one gets π1 (R2 \ {0}) Z. it is simply connected if and only if π1 (X) {1}. − Proposition 7. (i) A contractible space is simply connected. We shall work here with sheaves of kmodules.2 Monodromy of locally constant sheaves Locally constant sheaves Remark 7.8. (i) Assume (7.1. (iii) It is easily seen that if X is simply connected.
if M ∈ Mod(k). Denote as usual by aX the map X → pt. (i) If F is a constant sheaf on X.d. Let F be a locally constant sheaf on X = I × I.6 shows that F I×Vj is a constant sheaf for all j. inverse one to each other. Hence.e. Proof. then aX ∗ a−1 M M and a−1 aX ∗ MX X X MX .e. µN ) is a klinear map f : M → N which satisﬁes µN ◦ f = f ◦ µM (i.7. and let x ∈ X. µM ) → (N. Then the CSH(kX ) o aX ∗ / a−1 X Mod(k) are equivalences of categories.d. q.6 shows that F is q. LCSH(kX ) is a full additive subcategory of an abelian category (namely Mod(kX )) admitting kernels and cokernels. Therefore. the restriction to U of Ker ϕ and Coker ϕ will be constant sheaves. Then F is a constant sheaf. constant. MONODROMY OF LOCALLY CONSTANT SHEAVES 143 Let M ∈ Mod(k).2. A morphism µf : (M. (ii) Assume X is locally connected. and if one sets M = Γ(X. NX ) (Hom k (M. Representations For a group G and a ring k one deﬁnes the category Rep(G. by (i). − Proposition 7. An object is a pair (M. − . the argument of the proof of Proposition 5. We shall denote by LCSH(kX ) (resp. two functors (i) Assume X is connected and non empty. Lemma 7. Since I × I is compact. µM ) with − M ∈ Mod(k) and µM ∈ Hom (G.7. there is an isomorphism Hom kX (MX .2.e. of constant) sheaves.3. This implies it is abelian. the argument of the proof of Proposition 5. the category CSH(kX ) is abelian and if M and N are two kmodules. Recall that a constant sheaf F with stalk M on X is a sheaf isomorphic to the sheaf MX of locally constant functions with values in M . then F MX . there exists ﬁnite coverings of I by intervals {Ui }1≤i≤N0 and {Vj }1≤j≤N1 such that F Ui ×Vj is a constant sheaf. Since (Ui × Vj ) ∩ (Ui+1 × Vj ) is connected. Gl(M )).2. Since (I × Vj ) ∩ (I × Vj+1 ) is connected. N ))X .7. If − U is suﬃciently small connected open neighborhood of x. Proof. Then the category LCSH(kX ) is abelian. (ii) Let ϕ : F → G is a morphism of locally constant sheaf. F ). Mod(k)) of representations of G in Mod(k) as follows. CSH(kX )) the full additive subcategory of Mod(kX ) consisting of locally constant (resp. In particular.2.
identiﬁed to the trivial representations of G. Mod(k)). and con∼ sider the constant sheaf F = CX exp(αt). it is a constant sheaf. Note that Rep(G. Deﬁnition 7. and the groupoids Π1 (X) is equivalent to the group π1 (X. Mod(k)) are equivalent. and idM to any [γ] ∈ Hom Π1 (X) (x. 1]. Then the functor M → ∆M from Mod(k) to Fct0 (Π1 (X). One says that θ is a trivial representation if θ is isomorphic to a constant functor ∆M which associates the module M to any x ∈ X. One gets that (7.1) that we recall (7. (ii) Let θ ∈ Fct(Π1 (X). Let F be a locally constant sheaf of kmodules on X. which deﬁne µ(F )(γ): (7. we make the hypothesis (7. if X is connected. Example 7. a generalization of the category of representations Rep(π1 (X). Mod(k)) contains the full abelian subcategory Mod(k). c)= G. x0 ). Mod(k)) be the full subcategory of Fct(Π1 (X). (i) Let X = I. Mod(k)).6) Rep(G. Monodromy In this section. In this case. . Mod(k)) Fct(G. Mod(k)) consisting of trivial representations.8) F x0 ∼ −1 F )1 ∼ − →(γ (γ −1 F )0 ← Γ(I.2. We shall construct an isomorphism µ(F )(γ) : Fx0 F x1 . (i) One calls an object of Fct(Π1 (X).7). the two categories Rep(π1 (X).144 CHAPTER 7. γ −1 F ) − F x1 . Mod(k)). Let γ be a path from x0 to x1 . non empty and satisﬁes (7. Then µ(F )(I) : F0 − F1 is the → multiplication by exp(α). Mod(k)) is an equivalence of categories. then all x ∈ Π1 (X) are isomorphic. denote by t a coordinate on I. Now we shall consider the category Fct(Π1 (X).5.2. the morphisms being given by Hom G (c. Mod(k)) a representation of the groupoid Π1 (X) in Mod(k). Mod(k)) and Fct(Π1 (X). Mod(k)). x0 ) identiﬁed to the category with one object x0 and morphisms π1 (X. Since γ −1 F is a locally constant sheaf on [0.7) X is locally arcwise connected. y). In fact. Let Fct0 (Π1 (X).4. We identify G with a category G with one object c. We get the isomorphisms. HOMOTOPY AND FUNDAMENTAL GROUPOID µN (g) ◦ f = f ◦ µM (g) for any g ∈ G).
.2.) Let γ be the identity loop. γi−1 F ) ∼ / (γi−1 F )1 F x1 q.7). → Lemma 7. let f : X → Y be a continuous map.9) F x0 (ϕ−1 F )a0 o ∼ ∼ Γ(D.7. Hence we have constructed a functor of µ(F ) : Π1 (X) → Mod(k) given by − µ(F )(x) = Fx . Mod(k)). This correspondence being functorial in F . 2π]/ ∼ (where ∼ identiﬁes 0 and 2π) and denote by theta a coordinate on S1 .8. ϕ−1 F ) ∼ ∼ / (ϕ−1 F )a1 ∼ F x1 F x0 (γi−1 F )0 o ∼ Γ(I. We have the commutative (up to isomorphism) diagram of categories and functors: (7.10) is fully faithful. Consider the constant locally constant sheaf F = CX exp(iβθ). The functor µ in (7. If τ is another path from x1 to x2 . Then ∼ µ(F )(γ) : Fx0 − Fx0 is the multiplication by exp(2iπβ).2.e.10) is called the monodromy functor.7.6. − −1 i = 0. µ(F )([γ]) = µ(F )(γ) where γ is a representative of [γ].2.10) µ : LCSH(kX ) → Fct(Π1 (X). then: µ(F )(γτ ) = µ(F )(γ) ◦ µ(F )(τ ).2. MONODROMY OF LOCALLY CONSTANT SHEAVES 145 (ii) Let X = S1 [0.8.3.2. Mod(k)) f −1 / Fct(Π1 (X). This shows that µ(F )(γ0 ) = µ(F )(γ1 ). − Deﬁnition 7. we get a functor (7. More precisely. Assume (7. Mod(k)) Theorem 7. Let ϕ be a continuous function D → X such that ϕ(i. (See Example 5. t) = γi (t). 1) are described by the commutative diagram: (7. The isomorphism µ(F )(γ) depends only on the homotopy class of γ in X. The functor µ in (7. The sheaf ϕ F is constant by Lemma 7. 1.2. The isomorphisms µ(F )(γi ) (i = 0. The functor µ is also “functorial” with respect to the space X. and assume that both X and Y − are locally arcwise connected.d.11) LCSH(kY ) µ f −1 / LCSH(kX ) µ Fct(Π1 (Y ). Proof.
Let ϕ. Mod(k)).e.2. Corollary 7. each x ∈ X has an open neighborhood Ux such that the morphism ϕx : Fx → Gx extends as a − morphism ϕUx : F Ux → GUx . q.2. Mod(k)) → Mod(k). HOMOTOPY AND FUNDAMENTAL GROUPOID Proof.13) . Proof. Let σ : I → X be a path with x0 = σ(0) = σ(1) = x1 . Consider a morphism u : µ(F ) → µ(G). It follows from − ∼ Lemma 7. and we shall show that µ0 (F ) is the constant functor x → M . 7. let us show that µ0 takes its values in Fct0 (Π1 (X). − q.7) and X is non empty and connected. Then any locally constant sheaf F on X is a constant sheaf. where σ is any path − from x0 to x1 . q.d. given by ϕx1 = µ(G)(σ) ◦ ϕx0 ◦ µ(F )(σ −1 ). this implies that ϕ = ψ on the connected component of x0 . Then µ(F ) ∈ Mod(k) and there exists G ∈ CSH(kX ) such that µ(F ) µ(G). Mod(k)) Mod(k). Assume that X is connected. It deﬁnes a morphism − ϕx0 : Fx0 → Gx0 .2. ψ : F → G be morphisms of locally constant − sheaves and assume that µ(ϕ) µ(ψ). we shall assume (7. Fct(Π1 (X).12) ∼ µ0 : CSH(kX ) − Fct0 (Π1 (X). This isomorphism does not depend on the choice of σ by the hypothesis.8. This implies that ϕx0 ψx0 : Fx0 → − Gx0 for any x0 ∈ X.10) induces an equivalence (7. In this section. (i) µ is faithful. (iii) Since any M ∈ Mod(k) deﬁnes a constant sheaf. (i) First. µ0 is essentially surjective. for a module M . (ii) By Theorem 7. Proposition 7. this implies F G. Assume (7. Proof. Then the functor µ in (7.d. These morphisms will glue to each other and − deﬁne a morphism ϕ : F → G with µ(ϕ) = u. G) being locally constant.146 CHAPTER 7. By the hypothesis. Since µ is fully faithful. locally arcwise connected and simply connected.d. We may assume that F = MX .7) and also (7. Since F and G are locally constant.2.e.10. µ0 is fully faithful.e.3 (iii) that the morphism µ(F )(γ) : M Fx0 − Fx1 M is the → identity. and the sheaf Hom (F. To each x1 ∈ X we get a well deﬁned morphism ϕx1 : − Fx1 → Gx1 .3 The Van Kampen theorem there exists an open covering stable by ﬁnite intersections by connected and simply connected subsets. (ii) µ is full.9. Let F be a locally constant sheaf. Let F be a constant sheaf.
7) and (7. It remains to notice that if G is a group such that any representation of G is trivial. then G1 G2 . It remains to show that µ(F ) G. Mod(k)) to − Fct(Π1 (Ui ).e. Proof. Hence.7.13) and X is connected. The functor µ in (7.e. there exist isomorphisms G1 ([γj ]) G2 ([γj ]). the theorem is true if X is connected and simply connected. The functors iUi ∗ : Π1 (Ui ) → Π1 (X) deﬁne functors λi from Fct(Π1 (X).26) will be clearly satis→ ﬁed. and using the result in (ii) for Uij . G2 ∈ Fct(Π1 (X).3.3.3. q.7)and (7. we ﬁnd a sheaf F on X.3. Applying Theorem 5. one has G([γ]) = λi (G)([γ ]). G1 ([γ]) = G2 ([γ]) in this case. Assume (7. Let U = {Ui }i∈I be an open covering of X. we ﬁnd sheaves Fi such that µ(Fi ) = Gi . Corollary 7.9.) q. For any Ui as above.2.1 and Proposition 7. Assume γ is contained in some Ui . Lemma 7. each γj (1 ≤ j ≤ n) being contained in some Uij . Since Gν ([γ]) = Gν ([γn ]) · · · Gν ([γ1 ]) for ν = 1. Then for any G ∈ Fct(Π1 (X).2. Mod(k)). Let G ∈ Fct(Π1 (X). By the hypothesis. Mod(k)).2. hence. the result follows. and the cocycle condition (5.d. then G = {1}.2 below. . it remains to show that µ is essentially surjective.d. Setting λji = λj ◦ λ−1 . y) and let γ be a path which represents [γ]. THE VAN KAMPEN THEOREM 147 Theorem 7. Mod(k)) satisfy λ(G1 ) λ(G2 ). By Theorem 7. Assume (7.13). the result follows from Lemma 7.3.2. since x ∈ Ui for some i. Proof.1. any representation of π1 (X) is trivial if and only if any locally constant sheaf is constant. (i) For each x ∈ X. (See Exercise 7. i∈I If G1 .8. Using the result in (ii) for Ui .3. λi (µ(F )) Gi and λi (G) are isomorphic in Fct(Π1 (Ui ).d.1. will be locally constant.e. which will be locally isomorphic to the Fi ’s.13). Hence. we get isomorphisms i ∼ θji : Fi Uij − Fj Uij .10) is an equivalence of categories. G1 (x) G2 (x). Consider the functor λ= i∈I λi : Fct(Π1 (X). (ii) Let [γ] ∈ Hom Π1 (X) (x. Proof.10.8. (i) By Proposition 7. Mod(k)).3. (iii) We may decompose γ as γ = γ1 · · · · · γn . y). Mod(k)) and set Gi = λi (G). 2. for 1 ≤ j ≤ n. q. (ii) By Corollary 7. Then X is simply connected if and only if any locally constant sheaf on X is constant. and denote by [γ ] the corresponding element in Hom Π1 (Ui ) (x.3. (iii) Let U = {Ui }i∈I be an open covering of X as in (7. Mod(k)) → − Fct(Π1 (Ui ). Mod(k)).
3.1. when n = 1. We endow it with the discrete topology. up to isomorphism. and the coproduct is taken in the category of topological spaces. a local system of rank one is determined. We shall deduce a particular case of the Van Kampen theorem.5. HOMOTOPY AND FUNDAMENTAL GROUPOID Example 7. Hence. In particular each Xs is open. In fact.3.4 Coverings Let S be a set. In o × particular. Hence. by its monodromy µ(F ) ∈ Gl(Cn ).4. Gl(M )) ← − i i lim π (U ) → π1 (X). n n Hom (π1 (X).e. Gl(C )) = Gl(C ). A locally constant sheaf of Cvector spaces of ﬁnite rank on X is called a local system.d. and locally simply connected space. by its monodromy α ∈ C× . Assume that there exists x which belongs to all Ui ’s and identify each groupoids Π1 (Ui ) with the group π1 (Ui ) = π1 (Ui .3).148 CHAPTER 7. Theorem 7. − → where the ﬁrst isomorphism follows from Theorem 7. Then X × S s∈S Xs where Xs = X × {s} is a copy of X.14) Let M ∈ Mod(k). Gl(C) = C . then u is an → isomorphism (see Exercise 7. Then π1 (X) lim π1 (Ui ). π1 (X) Z. − − 1 i → i Hom (lim π1 (Ui ). Let X = {Ui }i∈I be an open covering stable by ﬁnite intersection. x). A local system F of rank n is determined. 7. To conclude. − → i∈I Sketch of proof. There is a natural morphism of groups (7. . Gl(M )) − Hom (G1 . remark − that if u : G1 → G2 is a morphism of groups which induces an isomorphism ∼ Hom (G2 . Gl(M )). Let X be a connected. a well known theory (JordanH¨lder decomposition). One has Hom (π1 (X). hence. Gl(M ) lim Hom (π1 (Ui ). the Ui ’s being connected. q. up to isomorphism. locally arcwise connected. it is now possible to classify all local systems on the space X = R2 \ {0}. The classiﬁcation of such sheaves is thus equivalent to that of invertible n × n matrices over C up to conjugation. Gl(M )) for all M ∈ Mod(Z).3.
(i) A continuous map f : Z → X is a trivial covering if − ∼ there exists a non empty set S. ±i 2}.2.1. − (ii) A continuous map f : Z → X is a covering 2 if f is surjective and any − x ∈ X has an open neighborhood U such that f f −1 (U ) : f −1 (U ) → U − is a trivial covering. − Hence. 1} and let f : Z → − X be the map z → (z 2 + 1)2 . with the same cardinal. ±i. X = C \ {0. Let f : Z → X be a covering. (iii) If f : Z → X is a covering. we have deﬁned the category Cov(X) of coverings above X. − The deﬁnition of a covering is visualized as follows.4. 2 “revˆtement” in French. S will be ﬁnite for all x.4. Many coverings appear naturally as the quotient of a topological space by a discrete group. and f = p ◦ h where p : X × S → X is the projection. Let Z = C \ {0. Then f is a 4covering.4. A local section is a section deﬁned − on an open subset U of X. e . (iv) A morphism of coverings f : Z → X to f : Z → X is a continuous − − map h : Z → Z such that f = f ◦ h. √ Example 7.4.3. Deﬁnition 7. say n. that we endow with the discrete topology. that is. a covering is locally isomorphic to a trivial covering. and the full subcategory of trivial coverings. a topological isomorphism h : Z − X × → S where S is endowed with the discrete topology. Notation 7.4.7. the group of isomorphisms of the object (f : Z → X) ∈ Cov(X). Roughly speaking. Let X be a locally compact topological space and let G be a group.4. One denotes by Aut (f ) the − group of automorphisms of this covering. COVERINGS 149 Deﬁnition 7. a section u of f is a continuous map − u : X → Z such that f ◦ u = idX . In this case one says that f is a ﬁnite (or an n)covering. not to be confused with “recouvrement”. We denote by e the unit in G. s∈S Us o ∼ uu h uu uu p uuuu % f −1 U / f Z f U / X If X is connected and S is ﬁnite for some x.
for g ∈ G and x ∈ X. g · x = x} is trivial. x ∼ y if and only if there exists g ∈ G with x = g · y. In the sequel. we refer to [11]. the subgroup of − C× generated by exp(2iπ/n). Assume that a discrete group G acts properly and freely on a locally compact space X. The orbit of x in X is the subset G · x of X. and there is are isomorphisms − Rn /Zn (R/Z)n (S1 )n . Therefore. the set GK = {g ∈ G. an isomorphism h : R/Z − → 1 t → exp(2iπt) : R → S is a covering. namely. the group Gx = {g ∈ G. that is. Theorem 7.150 CHAPTER 7. is reduced to {e}. µ(g) : X → X is continuous. Examples 7. (iv) One says that G acts properly on X if for any compact subset K of X. endowed with the quotient topology. (i) The map p : R → R/Z (where Z acts on R by − translation) is a covering. then Aut (p) = G. Moreover. − (ii) Consider the group Hn of nroots of unity in C. (a) for each g ∈ G. For the proof. Then p : S1 → S1 /Hn is a covering and the ∼ − map z → z n : S1 → S1 induces an isomorphism h : S1 /Hn − S1 such that → n n 1 1 z = h ◦ p. the map p : C → C/2iπZ is a covering and the map z → − ∼ exp(z) : C → C \ {0} induces an isomorphism h : C/2iπZ − C \ {0} such − → that exp(z) = h ◦ p. − (c) µ(g1 ◦ g2 ) = µ(g1 ) ◦ µ(g2 ). − n n n (iii) The projection R → R /Z is a covering. X = G · x. that is. − Similarly. Therefore. (v) One says that G acts freely if for any x ∈ X. If a group G acts on X. Then p : X → X/G is a covering.4.6. if − X is connected. we shall often write g · x instead of µ(g)(x). z → z : S → S is an ncovering. . (iii) One says that G acts transitively on X if for any x ∈ X. (ii) Let x ∈ X.5. g · K ∩ K = ∅} is ﬁnite. One denotes by X/G the quotient space. and the map t → exp(2iπt) : R → S1 induces − ∼ S1 such that exp(2iπt) = h ◦ p. z → exp(z) : C → C \ {0} is a covering. it deﬁnes an equivalence relation on X. Therefore.4. HOMOTOPY AND FUNDAMENTAL GROUPOID (i) An action µ of G on X is a map µ : G × X → X such that: − (b) µ(e) = idX .
˙ where E = E \ {0}. · · · . Let F ∈ LCSH(X). in other words. xn ) and x ∈ Pn (R) may be written as x = [x0 . Then Pn (R) is the set of lines in E. One says that [x0 .4.4. Assume (7. we shall often write for short Pn instead of Pn (R). a point x ∈ Rn+1 is written x = (x0 .8.4. · · · . X is locally arcwise connected. (7. and R× is the multiplicative group of nonzero elements of R.1). Denote by (7. the sheaf F is − locally isomorphic the the constant sheaf with values in S. λxn ] for any λ ∈ R× .e. λx1 . · · · . For U open in X. We shall construct a quasiinverse Ψ to Φ. F U MU is a constant sheaf with stalk M (hence. We have thus constructed a functor (7. · · · .7. · · · . and this action is clearly proper and free.17) Φ : Cov(X) → LCSH(X). x1 . x1 . The projective space of dimension n.7. q. denoted Pn (R). x1 . Let f : Z → X be a covering. x0 = 0}. is constructed as follows. locally. xn ] are homogeneous coordinates. This is a 2covering and Pn Coverings and locally constant sheaves We make hypothesis (7. yn ) → [1. x1 . xn ]. · · · . Let E be an n + 1dimensional Rvector space. Consider the set ZF = x∈X Fx . y1 . that is. − Proposition 7. x1 . Ff (U ) is the set of sections of f U : f −1 (U ) → U . Then the functor Φ in (7. The map Rn → Pn (R) given by (y1 . f is isomorphic to the projection U × S → U .17) is an equivalence. xn ] = [λx0 . − Since.d. . · · · . we get Pn Sn /a where a is the “antipodal” relation on Sn which identiﬁes x and −x. Fx M ) and M is endowed with the discrete topology. Since n ˙ S = E/R+ .16) γ : Sn → P n − Sn /(Z/2Z).15) ˙ Pn (R) = E/R× . We associates a sheaf of sets Ff on X as − follows. COVERINGS 151 Example 7. · · · . A basis of open subsets for this topology is given by the sets U × M such that U is open in X. yn ] allows − n us to identify R to the open subset of Pn (R) consisting of the set {x = [x0 . In the sequel. Identifying E with Rn+1 . the natural map. The map a deﬁnes an action of the group Z/2Z on Sn . Sketch of proof of the proof.1). xn ]. with the relation [x0 . We endow ZF with the following topology.
HOMOTOPY AND FUNDAMENTAL GROUPOID Note that the functor Φ induces an equivalence between trivial coverings and constant sheaves.3.1). Assume X satisﬁes (7.) Exercise 7. y. Gl(M )) − Hom (G1 . (i) Prove that X is simply connected. assuming (7. Hence.13) we get the equivalences Cov(X) LCSH(X) Fct(Π1 (X). Prove that G = {1}. Gl(M )) for all M ∈ Mod(Z).1 is also true when replacing sheaves of kmodules with sheaves of sets. ˙ Classify all locally constant sheaves of rank one of Cvector spaces on γ. (Hint: (i) use the free kmodule k[G] generated over k by the element g ∈ G.7) and (7. Prove that u is → an isomorphism. Exercise 7. the sphere Sn as well as Rn+1 \ {0} are simply connected. X = U1 ∪ U2 . z.2.4. U1 and U2 are connected and simply connected and U1 ∩ U2 is connected.152 CHAPTER 7. (ii) Deduce that for n > 1.1. Set). ˙ (Hint: one can use the fact that γ is homotopic to its intersection with the ˙ unit sphere S3 of R4 . Let γ = {(x. .) Exercise 7.3. x2 + y 2 + z 2 = t2 } γ = γ \ {0}. (i) Let G be a group and assume that all representation of G in Mod(k) are trivial. (ii) Let u : G1 → G2 is a morphism of groups which induces an isomorphism − ∼ Hom (G2 . Note that Theorem 7. Classify all locally constant sheaves of Cvector spaces on the space X = S1 × S1 . t) ∈ X = R4 . Exercises to Chapter 7 Exercise 7.
Armand Colin Ed. Elements de Mathematiques. Hermann e e (1958) [13] M. Abelian categories. Freyd. Benjamin (1967) [14] A. Vari´t´s diﬀ´rentiables. Geom´trie diﬀ´rentielle. Springer (1982) [4] N. Gelfand and Yu. Atiyah and I. Graduate Texts in Math. Tohoku e Math. Deligne. Harper & Row (1964) [10] S. e e (1972) [3] R. Soc. Bourbaki. Homological algebra. 163. Gostiaux. Hermann (1971) e e [12] R.G. Hermann. Grothendieck.I. Publ. Berger and B. Japan (1955) [7] P. Introduction to commutative algebra.Bibliography [1] M. Manin.I. AddisonWeisley (1969) [2] M. Sur quelques points d’alg`bre homologique. Tu. Lecture e ` e Notes in Math. Diﬀerential forms in algebraic topology. Bott and L. 119183 (1957) 153 . Springer Paris (1970) [8] G. Macdonald. Methods of homological algebra. Math. Princeton University Press (1956) [6] C. Eilenberg. Chevalley. Springer (1996) [11] C. Alg`bre Ch 10. The construction and study of certain important algebras.W. Topologie alg´brique et th´orie des faisceaux. Cartan and S. Lectures on algebric topology. Journ. De Rham. Godbillon. Greenberg. El´ments de topologie alg´brique. Masson (1980) e [5] H. Equations diﬀ´rentielles a points singuliers r´guliers. Paris (1960) ee e [9] JP. Godement. 82.
Kashiwara and P. Publ. 5 Springer 2nd ed. Analyse Alg´brique e e Universit´ Pierre et Marie Curie. place Jussieu F75252. The University of Chicago Press (1999) [23] P. Elements de g´om´trie alg´brique III.jussieu. An introduction to homological algebra.math. Schapira. (1998) [22] J. SpringerVerlag. (Dvi). An introduction to complex analysis. Categories and Homological Algebra.P. e Hermann [21] S. 38 (1994) Institut de Math´matiques. Weibel.jussieu. Iversen. Categories for the working mathematician. Lafon. (Ps). Cambridge Studies in Advanced Math. Kashiwara and P. to appear [20] JP. IHES 11 e e e (1961). Les formalismes fondamentaux de l’alg`bre commutative.math. MacLane. Case 82 e 4. Wiss. Schapira.fr/˜ schapira/publications/ [24] P. Schapira.fr Homepage: www. (Ps). Grundlehren der Math. (Dvi). An introduction to Sheaves. Graduate Texts in Math. 17 (1963) [16] L. H¨rmander.jussieu. unpublished course at Paris VI University. Paris Cedex 05. A concise course in algebraic toplogy. 292 SpringerVerlag (1990) [19] M.154 BIBLIOGRAPHY [15] A.jussieu.fr/~ schapira . Springer (1987) [18] M. unpublished course at Paris VI University. Cohomology of sheaves. May.fr/˜ schapira/publications/ [25] C. Categories and sheaves. Van Norstrand o (1966) [17] B. Schapira. (Pdf) http://www. Sheaves on manifolds. Grothendieck. Chicago Lectures in Mathematics. France email: schapira@math.math. (Pdf) http://www.
This action might not be possible to undo. Are you sure you want to continue?
Use one of your book credits to continue reading from where you left off, or restart the preview.