Coherent Algebraic Sheaves: Jean-Pierre Serre

Coherent Algebraic Sheaves
Jean-Pierre Serre
Translated by Piotr Achinger and Lukasz Krupa
Introduction
We know that the cohomological methods, in particular sheaf theory, play an in-
creasing role not only in the theory of several complex variables ([5]), but also in
classical algebraic geometry (let me recall the recent works of Kodaira-Spencer
on the Riemann-Roch theorem). The algebraic character of these methods sug-
gested that it is possible to apply them also to abstract algebraic geometry; the
aim of this paper is to demonstrate that this is indeed the case.
The content of the particular chapters is as follows:
Chapter I is dedicated to general sheaf theory. It contains proofs of the
results of this theory needed for the two other chapters. Various algebraic
operations one might perform on sheaves are described in 1; we follow quite
exactly the exposition of Cartan ([2], [5]). In 2 we study coherent sheaves
of modules; these generalize analytic coherent sheaves (cf. [3], [5]), admitting
almost the same properties. 3 contains the denition of cohomology groups
of a space X with values in a sheaf F. In subsequent applications, X is an
algebraic variety, equipped with the Zariski topology, so it is not topologically
separated
1
. and the methods used by Leray [10] and Cartan [3] (basing on
partitions of unity or ne sheaves) do not apply; so one is led to follow the
method of

Cech and dene the cohomology groups H
q
(X, F) by passing to the
limit with ner and ner open coverings. Another diculty arising from the
non-separatedness of X regards the cohomology exact sequence (cf. n
os
24
and 25): we could construct this exact sequence only for particular cases, yet
sucient for the purposes we had in mind (cf. n
os
24 and 47).
Chapter II starts with the denition of an algebraic variety, analogous to
that of Weil ([17], Chapter VII), but including the case of reducible varieties
(note that, contrary to Weils usage, we reserved the word variety only for
irreducible ones); we dene the structure of an algebraic variety using the data
consisting of the topology (Zariski topology) and a sub-sheaf of the sheaf of
germs of functions (a sheaf of local rings). An algebraic coherent sheaf on an
algebraic variety V is simply a coherent sheaf of O
V
-modules, O
V
being the
sheaf of local rings on V ; we give various examples in 2. The results obtained
are in fact similar to related facts concerning Stein manifolds (cf. [3], [5]): if
F is a coherent algebraic sheaf on an ane variety V , then H
q
(V, F) = 0 for
all q > 0 and F
x
is generated by H
0
(V, F) for all x V . Moreover (4), F
is determined by H
0
(V, F) considered as a module over the ring of coordinates
on V .
Chapter III, concerning projective varieties, contains the results which are
essential for this paper. We start with establishing a correspondence between
coherent algebraic sheaves F on a projective space X = P
r
(K) and graded
S-modules satisfying the condition (TF) of n
56 (S denotes the polynomial

1
i.e. Hausdor
1
algebra K[t
0
, . . . , t
r
]); this correspondence is bijective if one identies two S-
modules whose homogeneous components dier only in low degrees (for precise
statements, see n
os
57, 59 and 65). In consequence, every question concerning
F could be translated into a question concerning the associated S-module M.
This way we obtain a method allowing an algebraic determination of H
q
(X, F)
starting from M, which in particular lets us study the properties of H
q
(X, F(n))
for n going to + (for the denition of F(n), see n
54); the results obtained

are stated in n
os
65 and 66. In 4, we relate the groups H
q
(X, F) to the func-
tors Ext
q
S
introduced by Cartan-Eilenberg [6]; this allows us, in 5, to study the
behavior of H
q
(X, F(n)) for n tending to and give a homological character-
ization of varieties k times of the rst kind. 6 exposes certain properties of the
Euler-Poincare characteristic of a projective variety with values in a coherent
algebraic sheaf.
Moreover, we demonstrate how one can apply the general results of this
paper in diverse particular problems, and notably extend to the abstract case
the duality theorem of [15], thus a part of the results of Kodaira-Spencer
on the Riemann-Roch theorem; in these applications, the theorems of n
os
66,
75 and 76 play an essential role. We also show that, if the base eld is the
eld of complex numbers, the theory of coherent algebraic sheaves is essentially
identical to that of coherent analytic sheaves (cf. [4]).
2
Contents
Introduction 1
I Sheaves 4
1 Operations on sheaves . . . . . . . . . . . . . . . . . . . . . . . . 5
1 Denition of a sheaf . . . . . . . . . . . . . . . . . . . . . 5
2 Sections of a sheaf . . . . . . . . . . . . . . . . . . . . . . 5
3 Construction of sheaves . . . . . . . . . . . . . . . . . . . 6
4 Glueing sheaves . . . . . . . . . . . . . . . . . . . . . . . . 7
5 Extension and restriction of a sheaf . . . . . . . . . . . . 8
6 Sheaves of rings and sheaves of modules . . . . . . . . . . 9
7 Subsheaf and quotient sheaf . . . . . . . . . . . . . . . . . 9
8 Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . 10
9 The direct sum of two sheaves . . . . . . . . . . . . . . . 11
10 The tensor product of two sheaves . . . . . . . . . . . . . 11
11 The sheaf of germs of homomorphisms . . . . . . . . . . . 13
2 Coherent sheaves of modules . . . . . . . . . . . . . . . . . . . . 14
12 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . 14
13 Main properties of coherent sheaves . . . . . . . . . . . . 15
14 Operations on coherent sheaves . . . . . . . . . . . . . . . 16
15 Coherent sheaves of rings . . . . . . . . . . . . . . . . . . 17
16 Change of ring . . . . . . . . . . . . . . . . . . . . . . . . 18
17 Extension and restriction of a coherent sheaf . . . . . . . 19
3 Cohomology of a space with values in a sheaf . . . . . . . . . . . 20
3
CONTENTS CONTENTS
18 Cochains of a covering . . . . . . . . . . . . . . . . . . . . 20
19 Simplicial operations . . . . . . . . . . . . . . . . . . . . . 20
20 Complexes of cochains . . . . . . . . . . . . . . . . . . . . 21
21 Passing to a ner covering . . . . . . . . . . . . . . . . . . 22
22 Cohomology groups of X with values in a sheaf F . . . . 23
23 Homomorphisms of sheaves . . . . . . . . . . . . . . . . . 24
24 Exact sequence of sheaves: the general case . . . . . . . . 25
25 Exact sequence of sheaves: the case of X paracompact . . 26
26 Cohomology of a closed subspace . . . . . . . . . . . . . . 27
4 Comparison of cohomology groups of dierent coverings . . . . . 29
27 Double complexes . . . . . . . . . . . . . . . . . . . . . . 29
28 The double complex dened by two coverings . . . . . . . 30
29 Applications . . . . . . . . . . . . . . . . . . . . . . . . . 31
II Algebraic Varieties
Coherent Algebraic Sheaves on Ane Varieties 33
1 Algebraic varieties . . . . . . . . . . . . . . . . . . . . . . . . . . 34
30 Spaces satisfying condition (A) . . . . . . . . . . . . . . . 34
31 Locally closed subsets of an ane space . . . . . . . . . . 35
32 Regular functions . . . . . . . . . . . . . . . . . . . . . . . 36
33 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
34 Denition of the structure of an algebraic variety . . . . . 37
35 Regular mappings, induced structures, products . . . . . 39
36 The eld of rational functions on an irreducible variety . . 40
2 Coherent algebraic sheaves . . . . . . . . . . . . . . . . . . . . . . 43
37 The sheaf of local rings on an algebraic variety . . . . . . 43
38 Coherent algebraic sheaves . . . . . . . . . . . . . . . . . 44
39 Sheaf of ideals dened by a closed subvariety . . . . . . . 44
40 Sheaves of fractional ideals . . . . . . . . . . . . . . . . . 45
41 Sheaf associated to the total space of a vector bundle . . 45
3 Coherent algebraic sheaves on ane varieties . . . . . . . . . . . 47
42 Ane varieties . . . . . . . . . . . . . . . . . . . . . . . . 47
43 Some preliminary properties of irreducible varieties . . . . 48
4
CONTENTS CONTENTS
44 Vanishing of certain cohomology groups . . . . . . . . . . 49
45 Sections of a coherent algebraic sheaf on an ane variety 51
46 Cohomology groups of an ane variety with values in a
coherent algebraic sheaf . . . . . . . . . . . . . . . . . . . 52
47 Coverings of algebraic varieties by open ane subsets . . 53
4 Correspondence between modules of nite type and coherent al-
gebraic sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
48 Sheaf associated to a module . . . . . . . . . . . . . . . . 54
49 Module associated to an algebraic sheaf . . . . . . . . . . 55
50 Projective modules and vector bundles . . . . . . . . . . . 56
IIICoherent Algebraic Sheaves on Projective Varieties 58
1 Projective varieties . . . . . . . . . . . . . . . . . . . . . . . . . . 59
51 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
52 Cohomology of subvarieties of the projective space . . . . 60
53 Cohomology of irreducible algebraic curves . . . . . . . . 61
2 Graded modules and coherent algebraic sheaves on the projective
space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
54 The operation F(n) . . . . . . . . . . . . . . . . . . . . . 63
55 Sections of F(n) . . . . . . . . . . . . . . . . . . . . . . . 64
56 Graded modules . . . . . . . . . . . . . . . . . . . . . . . 65
57 The algebraic sheaf associated to a graded S-module . . . 66
58 First properties of the functor A(M) . . . . . . . . . . . 67
59 The graded Smodule associated to an algebraic sheaf . . 68
60 The case of coherent algebraic sheaves . . . . . . . . . . . 70
3 Cohomology of the projective space with values in a coherent
algebraic sheaf . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
61 The complexes C
k
(M) and C(M) . . . . . . . . . . . . . 71
62 Calculation of H
q
k
(M) for certain modules M . . . . . . . 72
63 General properties of H
q
(M) . . . . . . . . . . . . . . . . 74
64 Comparison of the groups H
q
(M) and H
q
(X, A(M)) . . 75
65 Applications . . . . . . . . . . . . . . . . . . . . . . . . . 77
66 Coherent algebraic sheaves on projective varieties . . . . . 78
67 A supplement . . . . . . . . . . . . . . . . . . . . . . . . . 79
5
CONTENTS CONTENTS
4 Relations with the functors Ext
q
S
. . . . . . . . . . . . . . . . . . 80
68 The functors Ext
q
S
. . . . . . . . . . . . . . . . . . . . . . 80
69 Interpretation of H
q
k
(M) in terms of Ext
q
S
. . . . . . . . . 81
70 Denition of the functors T
q
(M) . . . . . . . . . . . . . . 82
71 Determination of T
r
(M). . . . . . . . . . . . . . . . . . . 83
q
(M). . . . . . . . . . . . . . . . . . . 85
5 Applications to coherent algebraic sheaves . . . . . . . . . . . . . 87
73 Relations between functors Ext
q
S
and Ext
q
Ox
. . . . . . . . 87
74 Vanishing of cohomology groups H
q
(X, F(n)) for n
+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
75 Nonsingular varieties . . . . . . . . . . . . . . . . . . . . . 89
76 Normal Varieties . . . . . . . . . . . . . . . . . . . . . . . 90
77 Homological characterization of varieties k-times of rst
kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
78 Complete intersections . . . . . . . . . . . . . . . . . . . . 94
6 The characteristic function and arithmetic genus . . . . . . . . . 95
79 Euler-Poincare characteristic . . . . . . . . . . . . . . . . 95
80 Relation with characteristic function of a graded S-module 96
81 The degree of the characteristic function . . . . . . . . . . 97
Bibliography 100
6
Chapter I
Sheaves
7
1. Operations on sheaves I
1 Operations on sheaves
1 Denition of a sheaf
Let X be a topological space. A sheaf of abelian groups on X (or simply a
sheaf ) consists of:
(a) A function x F
x
, giving for all x X an abelian group F
x
,
(b) A topology on the set F, the sum of the sets F
x
.
If f is an element of F
x
, we put (f) = x; we call the mapping of the
projection of F onto X; the family in FF consisting of pairs (f, g) such that
(f) = (g) is denoted by F +F.
Having stated the above denitions, we impose two axioms on the data (a)
and (b):
(I) For all f F there exist open neighborhoods V of f and U of (f) such
that the restriction of to V is a homeomorphism of V and U.
(In other words, is a local homeomorphism).
(II) The mapping f f is a continuous mapping from F to F, and the
mapping (f, g) f +g is a continuous mapping from F +F to F.
We shall see that, even when X is separated (which we do not assume), F is
not necessarily separated, which is shown by the example of the sheaf of germs
of functions (cf. n
3).
Example of a sheaf. For G an abelian group, set F
x
= G for all x X;
the set F can be identied with the product X G and, if it is equipped with
the product topology of the topology of X by the discrete topology on G, one
obtains a sheaf, called the constant sheaf isomorphic with G, often identied
with G.
2 Sections of a sheaf
Let F be a sheaf on a space X, and let U be a subset of X. By a section of
F over U we mean a continuous mapping s : U F for which s coincides
with the identity on U. We therefore have s(x) F
x
for all x U. The set of
sections of F over U is denoted by (U, F); axiom (II) implies that (U, F) is
an abelian group. If U V , and if s is a section over V , the restriction of s to U
is a section over U; hence we have a homomorphism
V
U
: (V, F) (U, F).
If U is open in X, s(U) is open in F, and if U runs over a base of the
topology of X, then s(U) runs over a base of the topology of F; this is only
another wording of axiom (I).
Note also one more consequence of axiom (I): for all f F
x
, there exists a
section s over an open neighborhood of x for which s(x) = f, and two sections
with this property coincide on an open neighborhood of x. In other words, F
x
is an inductive limit of (U, F) for U running over the ltering order of all open
neighborhoods of x.
8
3 Construction of sheaves
Given for all open U X an abelian group F
U
and for all pairs of open
U V a homomorphism
V
U
: F
V
F
U
, satisfying the transitivity condition
V
U

W
V
=
W
U
whenever U V W.
The collection (F
U
,
V
U
) allows us to dene a sheaf F in the following way:
(a) Put F
x
= limF
U
(inductive limit of the system of open neighborhoods
of x). If x belongs to an open subset U, we have a canonical morphism
U
x
:
F
U
F
x
.
(b) Let t F
U
and denote by [t, U] the set of
U
x
(t) for x running over U ;
we have [t, U] F and we give F the topology generated by [t, U]. Moreover,
an element f F
x
has a base of neighborhoods consisting of the sets [t, U] for
x U and
U
x
(t) = f.
One veries immediately that the data (a) and (b) satisfy the axioms (I) and
(II), in other words, that F is a sheaf. We say that this is the sheaf dened by
the system (F
U
,
V
U
).
If f F
U
, the mapping x
U
x
(t) is a section of F over U ; hence we have
a canonical morphism : F
U
(U, F).
Proposition 1. : F
U
(U, F) is injective
1
if and only if the following
condition holds:
If an element t F
U
is such that there exists an open covering U
i
of U
with
U
Ui
(t) = 0 for all i, then t = 0.
If t F
U
satises the condition above, we have
U
x
(t) =
Ui
x

U
Ui
(t) = 0 if x U
i
,
which means that (t) = 0. Conversely, suppose that (t) = 0 with t F
U
;
since
U
x
(t) = 0 for x U, there exists an open neighborhood U(x) of x such
that
U
U(x)
(t) = 0, by the denition of an inductive limit. The sets U(x) form
therefore an open covering of U satisfying the condition stated above.
Proposition 2. Let U be an open subset of X, and let : F
V
(V, F)
be injective for all open V U. Then : F
U
(U, F) is surjective
1
(and
therefore bijective) if and only if the following condition is satised:
For all open coverings U
i
of U, and all systems t
i
, t
i
F
Ui
such that
Ui
UiUj
(t
i
) =
Uj
UiUj
(t
j
) for all pairs (i, j), there exists a t F
U
with
U
Ui
(t) = t
i
for all i.
The condition is necessary: every t
i
denes a section s
i
= (t
i
) over U
i
, and
we have s
i
= s
j
over U
i
U
j
; so there exists a section s over U which coincides
with s
i
over U
i
for all i; if : F
U
(U, F) is surjective, there exists t F
U
1
Recall (cf. [1]) that a function f : E E
is injective if f(e
1
) = f(e
2
) implies e
1
= e
2
,
surjective if f(E) = E
, bijective when it is both injective and surjective. An injective (resp.

surjective, bijective) mapping is called an injection (resp. a surjection, a bijection).
9
such that (t) = s. If we put t
i
=
U
Ui
(t), the section dened by t
i
over U
i
does
not dier from s
i
; since (t
i
t
i
) = 0, which implies t
i
= t
i
for was supposed
injective.
The condition is sucient: if s is a section of F over U, there exists an
open covering U
i
of U and elements t
i
F
Ui
such that (t
i
) coincides with
the restriction of s to U
i
; it follows that the elements
Ui
UiUj
(t
i
) and
Uj
UiUj
(t
j
)
dene the same section over U
i
U
j
, so, by the assumption made on , they are
equal. If t F
U
satises
U
Ui
(t) = t
i
, (t) coincides with s over each U
i
, so also
over S, q.e.d.
Proposition 3. If F is a sheaf of abelian groups on X, the sheaf dened
by the system ((U, F),
V
U
) is canonically isomorphic with F.
This is an immediate result of properties of sections stated in n
2.
Proposition 3 shows that every sheaf can be dened by an appropriate system
(F
U
,
V
U
). We will see that dierent systems can dene the same sheaf F;
however, if we impose on (F
U
,
V
U
) the conditions of Propositions 1 and 2, we
shall have only one (up to isomorphism) possible system: the one given by
((U, F),
V
U
).
Example. Let G be an abelian group and denote by F
U
the set of functions
on U with values in G; dene
V
U
: F
V
F
U
by restriction of such functions.
We thus obtain a system (F
U
,
V
U
), and hence a sheaf F, called the sheaf of
germs of functions with values in G. One checks immediately that the system
(F
U
,
V
U
) satises the conditions of Propositions 1 and 2; we thus can identify
sections of F over an open U with the elements of F
U
.
4 Glueing sheaves
Let F be a sheaf on X, and let U be a subset of X; the set
1
(U) F, with
the topology induced from F, forms a sheaf over U, called a sheaf induced by
F on U, end denoted by F(U) (or just F, when it does not cause confusion).
We see that conversely, we can dene a sheaf on X by means of sheaves on
open subsets covering X:
Proposition 4. Let U = U
i
iI
be an open covering of X and, for all
i I, let F
i
be a sheaf over U
i
; for all pairs (i, j) let
ij
be an isomorphism
from F
j
(U
i
U
j
) to F
i
(U
i
U
j
); suppose that we have
ij

jk
=
ik
at each
point of U
i
U
j
U
k
for all triples (i, j, k).
Then there exists a sheaf F and for all i an isomorphism
i
from F(U
i
) to
F
i
, such that
ij
=
i

1
j
at each point of U
i
U
j
. Moreover, F and
i
are
determined up to isomorphism by the preceding conditions.
The uniqueness of F,
i
is evident; for the proof of existence, we could
dene F as a quotient space of the sum of F
i
, but we will rather use the methods
of n
3: if U is an open subset of X, let F

U
be the group whose elements are
10
systems s
k
kI
with s
k
(U U
k
, F
k
) and s
k
=
kj
(s
j
) on U U
j
U
k
;
if U V , we dene
V
U
in an obvious way. The sheaf dened by the system
(F
U
,
V
U
) is the sheaf F we look for; moreover, if U U
i
, the mapping sending
a system s
k
F
U
to the element s
i
(U
i
, F
i
) is an isomorphism from F
U
to (U, F
i
), because of the transitivity condition; we so obtain an isomorphism
i
: F(U
i
) F
i
, which obviously satises the stated condition.
We say that the sheaf F is obtained by glueing the sheaves F
i
by means of
the isomorphisms
ij
.
5 Extension and restriction of a sheaf
Let X be a topological space, Y its closed subspace and F a sheaf on X. We
say that F is concentrated on Y , or that it is zero outside of Y if we have
F
x
= 0 for all x X Y .
Proposition 5. If a sheaf F is concentrated on Y , the homomorphism
X
Y
: (X, F) (Y, F(Y ))
is bijective.
If a section of F over X is zero over Y , it is zero everywhere since F
x
= 0 if
x / Y , which shows that
X
Y
is injective. Conversely, let s be a section of F(Y )
over Y , and extend s onto X by putting s(x) = 0 for x / Y ; the mapping
x s(x) is obviously continuous on X Y ; on the other hand, if x Y , there
exists a section s
of F over an open neighborhood U of x for which s
(x) = s(x);
since s is continuous on Y by assumption, there exists an open neighborhood V
of x, contained in U and such that s
(y) = s(y) for all y V Y ; since F

y
= 0
if y / Y , we also have that s
(y) = s(y) for y V (V Y ); hence s and s
coincide on V , which proves that s is continuous in a neighborhood of Y , so it is

continuous everywhere. This shows that
X
Y
is surjective, which ends the proof.
We shall now prove that the sheaf F(Y ) determines the sheaf F uniquely:
Proposition 6. Let Y be a closed subspace of X, and let G be a sheaf on Y .
Put F
x
= G
x
if x Y , F
x
= 0 if x / Y , and let F be the sum of the sets F
x
.
Then F admits a unique structure of a sheaf over X such that F(Y ) = G.
Let U be an open subset of X; if s is a section of G on U Y , extend s by
0 on U (U Y ); when s runs over (U Y, G), we obtain this way a group
F
U
of mappings from U to F. Proposition 5 then shows that if F is equipped
a structure of a sheaf such that F(Y ) = G, we have F
U
= (U, F), which
proves the uniqueness of the structure in question. The existence is proved
using the methods of n
3 applied to F
U
and the restriction homomorphisms
V
U
: F
U
F
V
.
We say that a sheaf F is obtained by extension of the sheaf G by 0 outside
Y ; we denote this sheaf by G
X
, or simply G if it does not cause confusion.
11
6 Sheaves of rings and sheaves of modules
The notion of a sheaf dened in n
1 is that of a sheaf of abelian groups. It

is clear that there exist analogous denitions for all algebraic structures (we
could even dene sheaves of rings, where F
x
would not admit an algebraic
structure, and we only require axiom (I)). From now on, we will encounter
mainly sheaves of rings and sheaves of modules:
A sheaf of rings A is a sheaf of abelian groups A
x
, x X, where each A
x
has a structure of a ring such that the mapping (f, g) f g is a continuous
mapping from A +A to A (the notation being that of n
1). We shall always

assume that A
x
has a unity element, varying continuously with x.
If A is a sheaf of rings satisfying the preceding condition, (U, A) is a ring
with unity, and
V
U
: (V, A) (U, A) is a homomorphism of rings preserving
unity if U V . Conversely, given rings A
U
with unity and homomorphisms
V
U
: A
V
A
U
preserving unity and satisfying
V
U

W
V
=
W
U
, the sheaf A
dened by the system (A
U
,
V
U
) is a sheaf of rings. For example, if G is a ring
with unity, the ring of germs of functions with values in G (dened in n
3) is
a sheaf of rings.
Let A be a sheaf of rings. A sheaf F is called a sheaf of A-modules if every
F
x
carries a structure of a left unitary
2
A
x
-module, varying continuously
with x, in the following sense: if A + F is the subspace of A F consisting
of the pairs (a, f) with (a) = (f), the mapping (a, f) a f is a continuous
mapping from A +F to F.
If F is a sheaf of A-modules, (U, F) is a unitary module over (U, A).
Conversely, if A is dened by the system (A
U
,
V
U
) as above, and let F be a
sheaf dened by the system (F,
V
U
), where every F
U
is a unitary A
U
-module,
with
V
U
(a f) =
V
U
(a)
V
U
(f); then F is a sheaf of A-modules.
Every sheaf of abelian groups can be considered a sheaf of Z-modules, Z
being the constant sheaf isomorphic to the ring of integers. This will allow us
to narrow our study to sheaves of modules from now on.
7 Subsheaf and quotient sheaf
Let A be a sheaf of rings, F a sheaf of A-modules. For all x X, let G
x
be a
subset of F
x
. We say that G =
G
x
is a subsheaf of F if:
(a) G
x
is a sub-A
x
-module of F
x
for all x X,
(b) G is an open subset of F.
Condition (b) can be also expressed as:
(b) If x is a point of X, and if s is a section of F over a neighborhood of
x such that s(x) G
x
, we have s(y) G
y
for all y close enough to x.
2
i.e. with the unity acting as identity
12
It is clear that, if these conditions are satised, G is a sheaf of A-modules.
Let G be a subsheaf of F and put K
x
= F
x
/G
x
for all x X. Give
K =
K
x
the quotient topology of F; we see easily that we also obtain
a sheaf of A-modules, called the quotient sheaf of F by G, and denoted by
F/G. We can give another denition, using the methods of n
3: if U is an
open subset of X, set K
U
= (U, F)/(U, G) and let
V
U
a homomorphism
obtained by passing to the quotient with
V
U
: (V, F) (U, F); the sheaf
dened by the system (K
U
,
V
U
) coincides with K .
The second denition of K shows that, if s is a section of K over a neigh-
borhood of x, there exists a section t of F over a neighborhood of x such that
the class of t(y) mod G
y
is equal to s(y) for all y close enough to x. Of course,
this does not hold globally in general: if U is an open subset of X we only have
an exact sequence
0 (U, G) (U, F) (U, K ),
the homomorphism (U, F) (U, K ) not being surjective in general (cf. n
24).
8 Homomorphisms
Let A be a sheaf of rings, F and G two sheaves of A-modules. An A-
homomorphism (or an A-linear homomorphism, or simply a homomorphism)
from F to G is given by, for all x X, an A
x
-homomorphism
x
: F
x
G
x
,
such that the mapping : F G dened by the
x
is continuous. This
condition can also be expressed by saying that, if s is a section of F over U,
x
x
(s(x)) is a section of G over U (we denote this section by (s), or s).
For example, if G is a subsheaf of F, the injection G F and the projection
F F/G both are homomorphisms.
Proposition 7. Let be a homomorphisms from F to G. For all x X,
let N
x
be the kernel of
x
and let I
x
be the image of
x
. Then N =
N
x
is
a subsheaf of F, I =
I
x
is a subsheaf of G and denes an isomorphism
of F/N and I.
Since
x
is an A
x
-homomorphism, N
x
and I
x
are submodules of F and
G respectively, and
x
denes an isomorphism of F
x
/N
x
with I
x
. If on the
other hand s is a local section of F, such that s(x) N
x
, we have s(x) = 0,
hence s(y) = 0 for y close enough to x, so s(y) N
y
, which shows that
N is a subsheaf of F. If t is a local section of G, such that t(x) I
x
, there
exists a local section s F, such that s(x) = t(x), hence s = t in the
neighborhood of x, showing that I is a subsheaf of G, isomorphic with F/N .
The sheaf N is called the kernel of and denoted by Ker(); the sheaf I
is called the image of and denoted by Im(); the sheaf G/I is called the
cokernel of and denoted by Coker(). A homomorphism is called injective,
13
or one-to-one, if each
x
is injective, or equivalently if Ker() = 0; it is called
surjective if each
x
is surjective, or equivalently if Coker() = 0; it is called
bijective if it is both injective and surjective, and Proposition 7 shows that it
is an isomorphism of F and G and that
1
is a homomorphism. All the
denitions related to homomorphisms of modules translate naturally to sheaves
of modules; for example, a sequence of homomorphisms is called exact if the
image of each homomorphisms coincides with the kernel of the homomorphism
following it. If : F G is a homomorphism, the sequences:
0 Ker() F Im() 0
0 Im() G Coker() 0
are exact.
If is a homomorphism from F to G, the mapping s s is a (U, A)-
homomorphism from (U, F) to (U, G). Conversely, if A, F, G are dened by
the systems (A
U
,
V
U
), (F
U
,
V
U
), (G
U
,
V
U
) as in n
6, and take for every open

U X an A
U
-homomorphism
U
: F
U
G
U
such that
V
U

V
=
U

V
U
;
by passing to the inductive limit, the
U
dene a homomorphism : F G.
9 The direct sum of two sheaves
Let A be a sheaf of rings, F and G two sheaves of A-modules; for all x X,
form the module F
x
G
x
, the direct sum of F
x
and G
X
; an element of F
x
G
x
is a pair (f, g) with f F
x
and g G
x
. Let K be the sum of the sets F
x
G
x
for x X ; we can identify K with the subset of F G consisting of the pairs
(f, g) with (f) = (g). We give K the topology induced from F G and
verify immediately that K is a sheaf of A-modules; we call this sheaf the direct
sum of F and G, and denote it by F G. A section of F G is of the form
x (s(x), t(x)), where s and t are sections of F and G over U; in other words,
(U, F G) is isomorphic to the direct sum (U, F) (U, G).
The denition of the direct sum extends by recurrence to a nite number of
A-modules. In particular, a direct sum of p sheaves isomorphic to one sheaf F
is denoted by F
p
.
10 The tensor product of two sheaves
Let A be a sheaf of rings, F a sheaf right of A-modules, G a sheaf of left
A-modules. For all x X we set K
x
= F
x
G
x
, the tensor product being
taken over the ring A
x
(cf. for example [6], Chapter II, 2); let K be the sum
of the sets K
x
.
Proposition 8. There exists a structure of a sheaf on K , unique with the
property that if s and t are sections of F and G over an open subset U, the
mapping x s(x) t(x) K
x
gives a section of K over U.
14
The sheaf K thus dened is called the tensor product (over A) of F and
G, and denoted by F
A
G; if the rings A
x
are commutative, it is a sheaf of
A-modules.
If K has a structure of a sheaf satisfying the above condition, and if f
i
and g
i
are sections of F and G over an open U X, the mapping x
s
i
(x) t
i
(x)
is a section of K on U. In fact, all h K
x
can be expressed in the form
h =
f
i
g
i
, f
i
F
X
, g
i
G
x
, therefore also the form
s
i
(x) t
i
(x), where
s
i
and t
i
are dened in an open neighborhood U of x; in result, every section of
K can be locally expressed in the preceding form, which shows the uniqueness
of the structure of a sheaf on K .
Now we show the existence. We might assume that A, F, G are dened by
the systems (A
U
,
V
U
), (F
U
,
V
U
), (G
U
,
V
U
) as in n
6. Now set K
U
= F
U
G
U
,
the tensor product being taken over A
U
; the homomorphisms
V
U
and
V
U
dene,
by passing to the tensor product, a homomorphism
V
U
: K
V
K
U
; besides,
we have lim
xU
K
U
= lim
xU
F
U
lim
xU
G
U
= K
x
, the tensor product being
taken over A
x
(for the commutativity of the tensor product with inductive
limits, see for example [6], Chapter VI, Exercise 18). The sheaf dened by the
system (K
U
,
V
U
) can be identied with K , and K is thus given a structure
of a sheaf obviously satisfying the imposed condition. Finally, if the A
x
are
commutative, we can suppose that the A
U
are also commutative (it suces to
take for A
U
the ring (U, A)), so K
U
is a A
U
-module, and K is a sheaf of
A modules.
Now let be an A-homomorphism from F to F
and let be an A-
homomorphism formG to G
; in that case
x
x
is a homomorphism (of abelian
groups in general of A
x
-modules, if A
x
is commutative) and the denition of
F
A
G shows that the collection of
x
x
is a homomorphism from F
A
G
to F

A
G
; this homomorphism is denoted by ; if is the identity, we

write instead of 1.
All of the usual properties of the tensor product of two modules translate to
the tensor product of two sheaves of modules. For example, all exact sequences:
F F
0
give rise to an exact sequence:
F
A
G F

A
G F

A
G 0.
We have the canonical isomorphisms:
F
A
(G
1
G
2
) F
A
G
1
F
A
G
2
, F
A
A F,
and (supposing that A
x
are commutative, to simplify the notation):
F
A
G G
A
F, F
A
(G
A
K ) (F
A
G)
A
K .
15
11 The sheaf of germs of homomorphisms from one sheaf
to another
Let A be a sheaf of rings, F and G two sheaves of A-modules. If U is an open
subset of X, let K
U
be the group of homomorphisms from F(U) to G(U) (we
also write homomorphism from F to G over U in place of homomorphism
from F(U) to G(U)). The operation of restricting a homomorphism denes
V
U
: K
V
K
U
; the sheaf dened by (K
U
,
V
U
) is called the sheaf of germs
of homomorphisms from F to G and denoted by Hom
A
(F, G). If A
x
are
commutative, Hom
A
(F, G) is a sheaf of A-modules.
An element of Hom
A
(F, G), being a germ of a homomorphism from F to
G in a neighborhood of x, denes an A
x
-homomorphism from F
x
to G
x
; hence
a canonical homomorphism
: Hom
A
(F, G)
x
Hom
Ax
(F
x
, G
x
).
But, contrary to what happened with the operations studied up to now, the
homomorphism is not a bijection in general ; we will give in n
14 a sucient
condition for that.
If : F
F and : G G
are homomorphisms, we dene in an obvious

way a homomorphism
Hom
A
(, ) : Hom
A
(F, G) Hom
A
(F
, G
).
Every exact sequence 0 G G
gives rise to an exact sequence:

0 Hom
A
(F, G) Hom
A
(F, G
) Hom
A
(F, G
).
We also have the canonical isomorphisms: Hom
A
(A, G) G,
Hom
A
(F, G
1
G
2
) Hom
A
(F, G
1
) Hom
A
(F, G
2
)
Hom
A
(F
1
F
2
, G) Hom
A
(F
1
, G) Hom
A
(F
2
, G).
16
2. Coherent sheaves of modules I
2 Coherent sheaves of modules
In this paragraph, X denotes a topological space and A a sheaf of rings on X.
We suppose that all the rings A
x
, x X are commutative and have a unity
element varying continuously with x. All sheaves considered until n
16 are
sheaves of A-modules and all homomorphisms are A-homomorphisms.
12 Denitions
Let F be a sheaf of A-modules, and let s
1
, . . . , s
p
be sections of F over an
open U X. If we assign to any family f
1
, . . . , f
p
of elements of A
x
the element
i=p
i=1
f
i
s
i
(x) of F
x
, we obtain a homomorphism : A
p
F, dened over an
open subset U (being precise, is a homomorphism from A
p
(U) to F(U), with
the notations from n
4). The kernel R(s

1
, . . . , s
p
) of the homomorphism is
a subsheaf of A
p
, called the sheaf of relations between the s
i
; the image of is
a subsheaf of F generated by s
i
. Conversely, any homomorphism : A
p
F
denes the sections s
1
, . . . , s
p
by the formulas
s
1
(x) =
x
(1, 0, . . . , 0), . . . , s
p
(x) =
x
(0, . . . , 0, 1).
Denition 1. A sheaf of A-modules F is said to be of nite type if it is
locally generated by a nite number of its sections.
In another words, for every point x X, there exists an open neighborhood
U of x and a nite number of sections s
1
, . . . , s
p
of F over U such that every
element of F
y
, y U is a linear combination, with coecients in A
y
, of s
i
(y).
According to the preceding statements, it is another way of saying that the
restriction of F to U is isomorphic to a quotient sheaf of A
p
.
Proposition 1. Let F be a sheaf of nite type. If s
1
, . . . , s
p
are sections of
F, dened over a neighborhood of a point x X and generating F
x
, then they
also generate F
y
for all y close enough to x.
Because F is of nite type, there is a nite number of sections of F in a
neighborhood of x, say t
1
, . . . , t
q
, which generate F
y
for y close enough to x.
Since s
j
(x) generate F
x
, there exist sections f
ij
of A in a neighborhood of x
such that t
i
(x) =
j=p
j=1
f
ij
(x) s
j
(x); it follows that, for y close enough to x,
we have:
t
i
(y) =
j=p
j=1
f
ij
(y) s
j
(y),
which implies that s
j
(y) generate F
y
, q.e.d.
Denition 2. A sheaf of A-modules F is said to be coherent if:
(a) F is of nite type,
(b) If s
1
, . . . , s
p
are sections of F over an open U X, the sheaf of relations
between the s
i
is of nite type (over the open set U).
17
We will observe the local character of denitions 1 and 2.
Proposition 2. Locally, every coherent sheaf is isomorphic to the cokernel
of a homomorphism : A
q
A
p
.
This is an immediate result of the denitions and the remarks preceding
denition 1.
Proposition 3. Any subsheaf of nite type of a coherent sheaf is coherent.
Indeed, if a sheaf F satises condition (b) of denition 2, then any subsheaf
of F satises it also.
13 Main properties of coherent sheaves
Theorem 1. Let 0 F

G

K 0 be an exact sequence of homomor-
phisms. If two of the sheaves F, G, K are coherent, so is the third.
Suppose that G and K are coherent. Locally, there exists a homomorphism
: A
p
G; let I the kernel of ; since K is coherent, I is a sheaf of
nite type (condition (b)); thus (I) is a sheaf of nite type, thus coherent by
Proposition 3; since is an isomorphism from F to (I), it follows that F is
also coherent.
Suppose that F and G are coherent. Because G is of nite type, K is also of
nite type, so it remains to prove that K satises the condition (b) of denition
2. Let s
1
, . . . , s
p
be a nite number of sections of K in a neighborhood of a
point x X. The question being local, we can assume that there exist sections
s
1
, . . . , s
p
of G such that s
i
= (s
i
). Let n
1
, . . . , n
q
be a nite number of sections
of F in a neighborhood of x, generating F
y
for y close enough to x. A family
f
1
, . . . , f
p
of elements of A
y
belongs to R(s
1
, . . . , s
p
)
y
if and only if one can nd
g
1
, . . . , g
q
A
y
such that
i=p
i=1
f
i
s
i
=
j=q
j=1
g
j
(n
j
) in y.
Now the sheaf of relations between the s
i
and the (n
j
) is of nite type, because
G is coherent. The sheaf R(s
1
, . . . , s
p
), the image of the preceding by the
canonical projection from A
p+q
to A
p
is thus of nite type, which shows that
K is coherent.
Suppose that F and K are coherent. The question being local, we might
assume that F (resp. K ) is generated by a nite number of sections n
1
, . . . , n
q
(resp. s
1
, . . . , s
p
); furthermore we might assume that there exist sections s
i
of G
such that s
i
= (s
i
). It is clear that the sections s
i
and (n
j
) generate G, which
proves that G is a sheaf of nite type. Now let t
1
, . . . , t
r
be a nite number of
sections of G in a neighborhood of a point x; since K is coherent, there exist
sections f
i
j
or A
r
(1 i r, 1 j s), dened in the neighborhood of x,
18
which generate the sheaf of relations between the (t
i
). Put u
j
=
i=r
i=1
f
i
j
t
i
;
since
i=r
i=1
f
i
j
(t
i
) = 0, the u
j
are contained in (F) and, since F is coherent,
the sheaf of relations between the u
j
is generated, in a neighborhood of x, by
a nite number of sections, say g
j
k
(1 j s, 1 k t). I say that the
j=s
j=1
g
j
k
f
i
j
generate the sheaf R(t
1
, . . . , t
r
) in a neighborhood of x; indeed,
if
i=r
i=1
f
i
t
i
= 0 on y, with f
i
A
y
, we have
i=r
i=1
f
i
(t
i
) = 0 and there
exist g
j
A
y
with f
i
=
j=s
j=1
g
j
f
i
j
; noting that
i=r
i=1
f
i
t
i
= 0, one obtains
j=s
j=1
g
j
u
j
= 0, thus making the system g
j
a linear combination of the systems
g
j
k
and showing our assertion. It follows that G satises condition (b), which
ends the proof.
Corollary. A direct sum of a nite family of coherent sheaves is coherent.
Theorem 2. Let be a homomorphism from a coherent sheaf F to a
coherent sheaf G. The kernel, the cokernel and the image of are also coherent
sheaves.
Because F is coherent, () is of nite type, thus coherent by Proposition
3. We apply Theorem 1 to the exact sequences
0 Ker() F Im() 0
0 Im() G Coker() 0
seeing that Ker() and Coker() are also coherent.
Corollary. Let F and G be two coherent subsheaves of a coherent sheaf K .
The sheaves F +G and F G are coherent.
For F +G, this follows from Proposition 3; and for F G, this is the kernel
of F K /G.
14 Operations on coherent sheaves
We have just seen that a direct sum of a nite number coherent sheaves is a
coherent sheaf. We will now show analogous results for the functors and Hom.
Proposition 4. If F and G are two coherent sheaves, F
A
G is a coherent
sheaf.
By Proposition 2, F is locally isomorphic to the cokernel of a homomorphism
: A
q
A
p
; thus F
A
G is locally isomorphic to the cokernel of :
A
q
A
G A
p
A
G. But A
q
A
G and A
p
A
G are isomorphic to G
q
and
G
p
respectively, which are coherent (Corollary of Theorem 1). Thus F
A
G
is coherent (Theorem 2).
Proposition 5. Let F and G be two sheaves, F being coherent. For all
x X, the module Hom
A
(F, G)
x
is isomorphic to Hom
Ax
(F
x
, G
x
).
19
Precisely, we prove that the homomorphism
: Hom
A
(F, G)
x
Hom
A
(F, G)
x
,
dened in n
11, is bijective. First of all, let : F G be a homomorphism

dened in a neighborhood of x, being zero in F
x
; since F is of nite type, we
conclude immediately that is zero in a neighborhood of x, which proves that
is injective. We will show that is surjective, or in other words, that if is a
A
x
-homomorphism from F
x
to G
x
, there exists a homomorphism : F G,
dened in a neighborhood of x and such that
x
= . Let m
1
, . . . , m
p
be a nite
number of sections of F in a neighborhood of x, generating F
y
for all y close
enough to x, and let f
i
j
(1 i p, 1 j q) be sections of A
p
generating
R(m
1
, . . . , m
p
) in a neighborhood of x. There exist local sections of G, say
n
1
, . . . , n
p
, such that n
i
(x) = (m
i
(x)). Put p
j
=
i=p
i=1
f
i
j
n
i
, 1 j q; the
p
j
are local sections of G being zero in x, so in every point of a neighborhood
U of x. It follows that for y U, the formula
f
i
m
i
(y) = 0 with f
i
A
y
,
implies
f
i
n
i
(y) = 0; for any element m =
f
i
m
i
(y) F
y
, we thus can
put:
y
(m) =
i=p
i=1
f
i
n
i
(y) G
y
.
The collection of
y
, y U constitutes a homomorphism : F G, dened
over U and such that
x
= , which ends the proof.
Proposition 6. If F and G are two coherent sheaves, then Hom
A
(F, G)
is a coherent sheaf.
The question being local, we might assume, by Proposition 2, that we have
an exact sequence A
q
A
p
F 0. From the preceding Proposition it
follows that the sequence:
0 Hom
A
(F, G) Hom
A
(A
p
, G) Hom
A
(A
q
, G)
is exact. Now the sheaf Hom
A
(A
p
, G) is isomorphic to G
p
, thus is coherent, the
same for Hom
A
(A
q
, G). Theorem 2 then shows that Hom
A
(F, G) is coherent.
15 Coherent sheaves of rings
A sheaf of rings A can be regarded as a sheaf of A-modules; if this sheaf of
A-modules is coherent, we say that A is a coherent sheaf of rings. Since A is
clearly of nite type, this means that A satises condition (b) of Proposition
2. In other words:
Denition 3. A sheaf A is a coherent sheaf of rings if the sheaf of relations
between a nite number of sections of A over an open subset U is a sheaf of
nite type on U.
20
Examples. (1) If X is a complex analytic variety, the sheaf of germs of
holomorphic functions on X is a coherent sheaf of rings, by a theorem of K. Oka
(cf. [3], statement XV, or [5], 5).
(2) If X is an algebraic variety, the sheaf of local rings of X is a coherent
sheaf of rings (cf. n
37, Proposition 1).

When A is a coherent sheaf of rings, we have the following results:
Proposition 7. For a sheaf of A-modules, being coherent is equivalent to
being locally isomorphic to the cokernel of a homomorphism : A
q
A
p
.
The necessity part is Proposition 2; the suciency follows from the coherence
of A
p
and A
q
and from Theorem 2.
Proposition 8. A subsheaf of A is coherent if and only if it is of nite
type.
This is a special case of Proposition 3.
Corollary. The sheaf of relations between a nite number of sections of a
coherent sheaf is coherent.
In fact, this sheaf is of nite type, from the denition of a coherent sheaf.
Proposition 9. Let F be a coherent sheaf of A-modules. For all x X,
let I
x
be an ideal of A
x
consisting of those a A
x
for which a f = 0 for all
f F
x
. Then the I
x
form a coherent sheaf of ideals (called the annihilator of
F).
In fact, I
x
is the kernel of the homomorphism A
x
Hom
Ax
(F
x
, G
x
); we
then apply Propositions 5 and 6 and Theorem 2.
More generally, the conductor F : G of a coherent sheaf G into its coherent
subsheaf F is a coherent sheaf of ideals (being the annihilator of G/F).
16 Change of ring
The notions of a sheaf of nite type, and of a coherent sheaf, are dependent on
the xed sheaf of rings A. When we will consider multiple sheaves of rings,
we will say of nite type over A, A-coherent to point out that we mean
sheaves of A-modules.
Theorem 3. Let A be a coherent sheaf of rings, I a coherent sheaf of
ideals of A. Let F be a sheaf of A/I-modules. Then F is A/I-coherent if
and only if it is A-coherent. In particular, A/I is a coherent sheaf of rings.
It is clear that of nite type over A is the same as of nite type over
A/I. For the other part, if F is A-coherent, and if s
1
, . . . , s
p
are sections
of F over an open U, the sheaf of relations between the s
i
with coecients in
A, is of nite type over A. It follows immediately that the sheaf of relations
between the s
i
with coecients in A/I, is of nite type over A/I, since it
is the image of the preceding by the canonical mapping A
p
(A/I)
p
. Thus
21
F is A/I-coherent. In particular, since A/I is A-coherent, it is also A/I-
coherent, in other words, A/I is a coherent sheaf of rings. Conversely, if F
is A/I-coherent, it is locally isomorphic to the cokernel of a homomorphism
: (A/I)
q
(A/I)
p
and since A/I is A-coherent, F is coherent by
Theorem 2.
17 Extension and restriction of a coherent sheaf
Let Y be a closed subspace of a space X. If G is a sheaf over Y , we denote by
G
X
the a sheaf obtained by extending G by 0 outside Y ; it is a sheaf over X
(cf. n
5). If A is a sheaf of rings over Y , A

X
is a sheaf of rings over X, and
if F is a sheaf of A-modules, then F
X
is a sheaf of A
X
-modules.
Proposition 10. F is of nite type over A if and only if F
X
is of nite
type over A
X
.
Let U be an open subset of X, and let V = U Y . Any homomorphism
: A
p
F over V denes a homomorphism
X
: (A
X
)
p
F
X
over U,
and conversely; so is surjective if and only if
X
is. The proposition follows
immediately from this.
We therefore show:
Proposition 11. F is A-coherent if and only if F
X
is A
X
-coherent.
Hence, by putting F = A:
Corollary. A is a coherent sheaf of rings if and only if A
X
is a coherent
sheaf of rings.
22
3. Cohomology of a space with values in a sheaf I
3 Cohomology of a space with values in a
sheaf
In this paragraph, X is a topological space, separated or not. By a covering of
X we will always mean an open covering.
18 Cochains of a covering
Let U = U
i
iI
be a covering of X. If s = (i
0
, . . . , i
p
) is a nite sequence of
elements of I, we put
U
s
= U
i0...ip
= U
i0
. . . U
ip
.
Let F be a sheaf of abelian groups on the space X. If p is an integer
0, we call a p-cochain of U with values in F a function f assigning to every
s = (i
0
, . . . , i
p
) of p + 1 elements of I a section f
s
= f
i0...ip
of F over U
i0...ip
.
The p-cochains form an abelian group, denoted by C
p
(U, F); it is the product
group
(U
s
, F), the product being over all sequences s of p + 1 elements of
I. The family of C
p
(U, F), p = 0, 1, . . . is denoted by C(U, F). A p-cochain is
also called a cochain of degree p.
A p-cochain is said to be alternating if:
(a) f
i0...ip
= 0 whenever any two of the indices i
0
, . . . , i
p
are equal,
(b) f
i0...ip
=
f
i0...ip
if is a permutation of the set 0, . . . , p (
denotes
the sign of ).
The alternating cochains form a subgroup C
p
(U, F) of the group C
p
(U, F);
the family of the C
p
(U, F) is denoted by C
(U, F).
19 Simplicial operations
Let S(I) be the simplex with the set I as its set of vertices; an (ordered) simplex
of S(I) is a sequence s = (i
0
, . . . , i
p
) of elements of I; p is called the dimension
of s. Let K(I) =
p=0
K
p
(I) be the complex dened by S(I); by denition,
K
p
(I) is a free group with the set of simplexes of dimension p of S(I) as its
base.
If s is a simplex of S(I), we denote by [s[ the set of vertices of s.
A mapping h : K
p
(I) K
q
(I) is called a simplicial endomorphism if
(i) h is a homomorphism,
(ii) For any simplex s of dimension p of S(I) we have
h(s) =
c
s
s
s
, with c
s
s
Z,
23
the sum being over all simplexes s
of dimension q such that [s
[ [s[.
Let h be a simplicial endomorphism, and let f C
q
(U, F) be a cochain of
degree q. For any simplex s of dimension p put:
(
t
hf)
s
=
c
s
s

s
s
(f
s
),
denoting the restriction homomorphism: (U

s
, F) (U
s
, F), which
makes sense because [s
[ [s[. The mapping s (

t
hf)
s
is a p-cochain, de-
noted by
t
hf. The mapping f
t
hf is a homomorphism
t
h : C
q
(U, F) C
p
(U, F),
and one veries immediately the formulas:
t
(h
1
+h
2
) =
t
h
1
+
t
h
2
,
t
(h
1
h
2
) =
t
h
2

t
h
1
,
t
1 = 1.
Note. In practice, we often do not write the restriction homomorphism
s
s
.
20 Complexes of cochains
We apply the above to the simplicial endomorphism
: K
p+1
(I) K
p
(I),
dened by the usual formula:
(i
0
, . . . , i
p+1
) =
j=p+1
j=0
(1)
j
(i
0
, . . . ,
i
j
, . . . , i
p+1
),
the signmeaning, as always, that the symbol below it should be omitted.
We thus obtain a homomorphism
t
: C
p
(U, F) C
p+1
(U, F), which we
denote by d; from denition, we have that
(df)
i0...ip+1
=
j=p+1
j=0
(1)
j
j
(f
i0...
ij...ip+1
),
where
j
denotes the restriction homomorphism
j
: (U
i0...
ij...ip+1
, F) (U
i0...ip+1
, F).
Since = 0, we have d d = 0. Thus we nd that C(U, F) is equipped with
a coboundary operator making it a complex. The q-th cohomology group of the
complex C(U, F) will be denoted by H
q
(U, F). We have:
Proposition 1. H
0
(U, F) = (X, F).
24
A 0-cochain is a system (f
i
)
iI
with every f
i
being a section of F over U
i
.
It is a cocycle if and only if it satises f
i
f
j
= 0 over U
i
U
j
, or in other
words, if there is a section f of F on X coinciding with f
i
on U
i
for all i I.
Hence the Proposition.
(Thus H
0
(U, F) is independent of U; of course this is not true for H
q
(U, F)
in general).
We see immediately that df is alternating if f is alternating; in other words,
d preserves C
(U, F) which forms a subcomplex of C(U, F). The cohomology

groups of C
(U, F) are denoted by H

q
(U, F).
Proposition 2. The inclusion of C
(U, F) in C(U, F) induces an isomor-

phism of H
q
(U, F) and H
q
(U, F), for every q 0.
We equip the set I with a structure of a total order, and let h be a simplicial
endomorphism of K(I) dened in the following way:
h((i
0
, . . . , i
q
)) = 0 if any two indices i
0
, . . . , i
q
are equal,
h((i
0
, . . . , i
q
)) =
(i
0
. . . i
q
) if all indices i
0
, . . . , i
q
are distinct and is a
permutation of 0, . . . , q for which i
0
< i
1
< . . . < i
q
.
We verify right away that h commutes with and that h(s) = s if dim(s) = 0;
in result (cf. [7], Chapter VI, 5) there exists a simplicial endomorphism k,
raising the dimension by one, such that 1 h = k +k . Hence, by passing
to C(U, F),
1
t
h =
t
k d +d
t
k.
But we check immediately that
t
h is a projection fromC(U, F) onto C
(U, F);
since the preceding formula shows that it is a homotopy operator, the Proposi-
tion is proved. (Compare with [7], Chapter VI, theorem 6.10).
Corollary. H
q
(U, F) = 0 for q > dim(U).
By the denition of dim(U), we have U
i0...iq
= for q > dim(U), if the indices
i
0
, . . . , i
q
are distinct; hence C
q
(U, F) = 0, which shows that
H
q
(U, F) = H
q
(U, F) = 0.
21 Passing to a ner covering
A covering U = U
i
iI
is said to be ner than the covering V = V
j
jJ
if there
exists a mapping : I J such that U
i
V
i
for all i I. If f C
q
(V, F),
put
(f)
i0,...,iq
=
V
U
(f
i0...iq
),
V
U
denoting the restriction homomorphism dened by the inclusion of U
i0...iq
in V
i0...iq
. The mapping f f is a homomorphism from C
q
(V, F) to
C
q
(U, F), dened for all q 0 and commuting with d, thus it denes homo-
morphisms
: H
q
(V, F) H
q
(U, F).
25
Proposition 3. The homomorphisms
: H
q
(V, F) H
q
(U, F) depend
only on U and V and not on the chosen mapping .
Let and
be two mappings from I to J such that U

i
V
i
and U
i
V
i
; we have to show that
.
Let f C
q
(V, F); set
(kf)
i0...iq1
=
h=q1
h=0
(1)
h
h
(f
i0...ij
i
h
...
iq1
),
where
h
denotes the restriction homomorphism dened by the inclusion of
U
i0...iq1
in V
i0...ij
i
h
...
iq1
.
We verify by direct computation (cf. [7], Chapter VI, 3) that we have
dkf +k df =
f f,
which ends the proof of the Proposition.
Thus, if U is ner than V, there exists for every integer q 0 a canonical ho-
momorphism from H
q
(V, F) to H
q
(U, F). From now on, this homomorphism
will be denoted by (U, V).
22 Cohomology groups of X with values in a sheaf F
The relation U is ner than V (which we denote henceforth by U V)
is a relation of a preorder
3
between coverings of X; moreover, this relation
is ltered
4
, since if U = U
i
iI
and V = V
j
jJ
are two coverings, W =
U
i
V
j
(i,j)IJ
is a covering ner than U and than V.
We say that two coverings U and V are equivalent if we have U V and
V U. Any covering U is equivalent to a covering U
whose set of indices is

a subset of P(X); in fact, we can take for U
the set of open subsets of X

belonging to the family U. We can thus speak of the set of classes of coverings
with respect to this equivalence relation; this is an ordered ltered set.
5
If U V, we have dened at the end of the preceding n
a well dened
homomorphism (U, V) : H
q
(V, F) H
q
(U, F), dened for every integer
q 0 and every sheaf F on X. It is clear that (U, U) is the identity and
that (U, V) (V, W) = (U, W) if U V W. It follows that, if U is
equivalent to V, then (U, V) and (V, U) are inverse isomorphisms; in other
words, H
q
(F, U) depends only on the class of the covering U.
Denition. We call the q-th cohomology group of X with values in a sheaf
F, and denote by H
q
(X, F), the inductive limit of groups H
q
(U, F), where U
3
i.e. quasiorder
4
i.e. directed
5
To the contrary, we cannot speak about the set of coverings, because a covering is a
family whose set of indices is arbitrary.
26
runs over the ltered ordering of classes of coverings of X, with respect to the
homomorphisms (U, V).
In other words, an element of H
q
(X, F) is just a pair (U, x) with x
H
q
(U, F), and we identify two such pairs (U, x) and (V, y) whenever there
exists a W with W U, W V and (W, U)(x) = (W, V)(y) in H
q
(W, F).
Any covering U in X is thus associated a canonical homomorphism (U) :
H
q
(U, F) H
q
(X, F).
We will see that H
q
(X, F) can also be dened by an inductive limit of
H
q
(U, F) where U runs over a conal family of coverings. Thus, if X is quasi-
compact (resp. quasi-paracompact), we can consider only nite (resp. locally
nite) coverings.
When q = 0, by Proposition 1 we have:
Proposition 4. H
0
(X, F) = (X, F).
23 Homomorphisms of sheaves
Let be a homomorphism from a sheaf F to a sheaf G. If U is a covering of X,
we can assign to any f C
q
(U, F) an element f C
q
(U, G) dened by the
formula (f)
s
= (f
s
). The mapping f f is a homomorphism from C(U, F)
to C(U, G) commuting with the coboundary, thus it denes homomorphisms
: H
q
(U, F) H
q
(U, G). We have
(U, V) = (U, V)
, hence, by
passing to the limit, the homomorphisms
: H
q
(X, F) H
q
(X, G).
When q = 0,
coincides with the homomorphism from (X, F) to (X, G)

induced in the natural way by .
In general, the homomorphisms
satisfy usual formal properties:

( +)
, ( )
, 1
= 1.
In other words, for all q 0, H
q
(X, F) is a covariant additive functor of
F. Hence we gather that if F is the direct sum of two sheaves G
1
and G
2
, then
H
q
(X, F) is the direct sum of H
q
(X, G
1
) and H
q
(X, G
2
).
Suppose that F is a sheaf of A-modules. Any section of A on X denes
an endomorphism of F, therefore of H
q
(X, F). It follows that H
q
(X, F) are
modules over the ring (X, A).
27
24 Exact sequence of sheaves: the general case
Let 0 A

B

C 0 be an exact sequence of sheaves. If U is a covering
of X, the sequence
0 C(U, A)

C(U, B)

C(U, C)
is obviously exact, but the homomorphism need not be surjective in general.
Denote by C
0
(U, C) the image of this homomorphism; it is a subcomplex of
C(U, C) whose cohomology groups will be denoted by H
q
0
(U, C). The exact
sequence of complexes:
0 C(U, A) C(U, B) C
0
(U, C) 0
giving rise to an exact sequence of cohomology:
. . . H
q
(U, B) H
q
0
(U, C)
d
H
q+1
(U, A) H
q+1
(U, B) . . . ,
where the coboundary operator d is dened as usual.
Now let U = U
i
iI
and V = V
j
jJ
be two coverings and let : I J
be such that U
i
V
i
; we thus have U V. The commutative diagram:
0 C(V, A) C(V, B) C(V, C)
0 C(U, A) C(U, B) C(U, C)

shows that maps C
0
(V, C) into C
0
(U, C), thus dening the homomorphisms
: H
q
0
(V, C) H
q
0
(U, C). Moreover, the homomorphisms
are independent
of the choice of the mapping : this follows from the fact that, if f C
q
0
(V, C),
we have kf C
q1
0
(U, C), with the notations of the proof of Proposition 3. We
have thus obtained canonical homomorphisms (U, V) : H
q
0
(V, C) H
q
0
(U, C);
we might then dene H
q
0
(X, C) as the inductive limit of the groups H
q
0
(U, C).
Because an inductive limit of exact sequences is an exact sequence (cf. [7],
Chapter VIII, theorem 5.4), we obtain:
Proposition 5. The sequence
. . . H
q
(X, B)

H
q
0
(X, C)
d
H
q+1
(X, A)

H
q+1
(X, B) . . .
is exact.
(d denotes the homomorphism obtained by passing to the limit with the
homomorphisms d : H
q
0
(U, C) H
q+1
(U, A)).
To apply the preceding Proposition, it is convenient to compare the groups
H
q
0
(X, C) and H
q
(X, C). The inclusion of C
0
(U, C) in C(U, C) denes the
28
homomorphisms H
q
0
(U, C) H
q
(U, C), hence, by passing to the limit with U,
the homomorphisms:
H
q
0
(X, C) H
q
(X, C).
Proposition 6. The canonical homomorphism H
q
0
(X, C) H
q
(X, C) is
bijective for q = 0 and injective for q = 1.
We will prove the following lemma:
Lemma 1. Let V = V
j
jJ
be a covering and let f = (f
j
) be an element
of C
0
(V, C). There exists a covering U = U
i
iI
and a mapping : I J
such that U
i
V
i
and f C
0
0
(U, C).
For any x X, take a x J such that x V
x
. Since f
x
is a section of
C over V
x
, there exists an open neighborhood U
x
of x, contained in V
x
and
a section b
x
of B over U
x
such that (b
x
) = f
x
on U
x
. The U
x
xX
form
a covering U of X, and the b
x
form a 0-chain b of U with values in V; since
f = (b), we have that f C
0
0
(U, C).
We will now show that H
1
0
(X, C) H
1
(X, C) is injective. An element of the
kernel of this mapping may be represented by a 1-cocycle z = (z
j0j1
) C
0
(V, C)
such that there exists an f = (f
j
) C
0
(V, C) with df = z; applying Lemma
1 to f yields a covering U such that f C
0
0
(U, C), which shows that z is
cohomologous to 0 in C
0
(U, C), thus its image in H
1
0
(X, C) is 0. This shows
that H
0
0
(X, C) H
0
(X, C) is bijective.
Corollary 1. We have an exact sequence:
0 H
0
(X, A) H
0
(X, B) H
0
(X, C) H
1
(X, A) H
1
(X, B) H
1
(X, C).
This is an immediate consequence of Propositions 5 and 6.
Corollary 2. If H
1
(X, A) = 0, then (X, B) (X, C) is surjective.
25 Exact sequence of sheaves: the case of X paracompact
Recall that a space X is said to be paracompact if it is separated and if any
covering of X admits a locally nite ner covering. On paracompact spaces,
we can extend Proposition 6 for all values of q (I do not know whether that
extension is possible for nonseparated spaces):
Proposition 7. If X is paracompact, the canonical homomorphism
H
q
0
(X, C) H
q
(X, C)
is bijective for all q 0.
This Proposition is an immediate consequence of the following lemma, anal-
ogous to Lemma 1:
29
Lemma 2. Let V = V
j
jJ
be a covering, and let f = (f
j0...jq
) be an
element of C
q
(V, C). There exists a covering U = U
i
iI
and a mapping
: I J such that U
i
V
i
and f C
q
0
(U, C).
Since X is paracompact, we might assume that V is locally nite. Then
there exists a covering W
j
jJ
such that W
j
V
j
. For every x X, choose
an open neighborhood U
x
of x such that
(a) If x V
j
(resp. x W
j
), then U
x
V
j
(resp. U
x
W
j
),
(b) If U
x
W
j
,= , then U
x
W
j
,
(c) If x V
j0...jq
, there exists a section b of B over U
x
such that (b) = f
j0...jq
over U
x
.
The condition (c) can be satised due to the denition of the quotient sheaf
and to the fact that x belongs to a nite number of sets V
j0...jq
. Having (c)
satised, it suces to restrict U
x
appropriately to satisfy (a) and (b).
The family U
x
xX
forms a covering U; for any x X, choose x J such
that x W
x
. We now check that f belongs to C
q
0
(U, C), in other words, that
f
x0...xq
is the image by of a section of B over U
x0
. . .U
xq
. If U
x0
. . .U
xq
is empty, this is obvious; if not, we have U
x0
U
x
k
,= for 0 k q, and since
U
x
k
U
x
k
, we have U
x0
W
x
k
,= , which implies by (b) that U
x0
V
x
k
,
hence x
0
V
x0...xq
; we then apply (c), seeing that there exists a section b of
B over U
x0
such that (b)
x
= f
x0...xq
on U
x0
, so also on U
x0
. . . U
xq
, which
ends the proof.
Corollary. If X is paracompact, we have an exact sequence:
. . . H
q
(X, B)

H
q
(X, C)
d
H
q+1
(X, A)

H
q+1
(X, B) . . .
(the map d being dened as the composition of the inverse of the isomorphism
H
q
0
(X, C) H
q
(X, C) with d : H
q
0
(X, C) H
q+1
(X, A)).
The exact sequence mentioned above is called the exact sequence of coho-
mology dened by a given exact sequence of sheaves 0 A B C 0.
More generally, it exists whenever we can show that H
q
0
(X, C) H
q
(X, C) is
bijective (we will see in n
47 that this is the case when X is an algebraic variety

and when A is an algebraic coherent sheaf).
26 Cohomology of a closed subspace
Let F be a sheaf over a space X, and let Y be a subspace of Y . Let F(Y ) be
the sheaf induced by F on Y , in the sense of n
4. If U = U
i
iI
is a covering
of X, the sets U
i
= Y U
i
form a covering U
of Y ; if f
i0...i
Q
is a section of F
over U
i0...iq
, the restriction of f
i0...iq
to U
i0...iq
= Y U
i0...iq
is a section of F(Y ).
The operation of restriction is a homomorphism : C(U, F) C(U
, F(Y )),
commuting with d, thus dening
: H
q
(U, F) H
q
(U
, F(Y )). If U V,
we have U
, and
(U, V) = (U
, V
; thus the homomorphisms

30
dene, by passing to the limit with U, homomorphisms
: H
q
(X, F)
H
q
(Y, F(Y )).
Proposition 8. Assume that Y is closed in X and that F is zero outside
Y . Then
: H
q
(X, F) H
q
(Y, F(Y )) is bijective for all q 0.
The Proposition is implied by the following facts:
(a) Any covering W= W
i
iI
of Y is of the form U
for some covering U of

X.
Indeed, it suces to put U
i
= W
i
(X Y ), since Y is closed in X.
(b) For any covering U of X, : C(U, F) C(U
, F(Y )) is bijective.
Indeed, the result follows from Proposition 5 of n
5, applied to U
i0...iq
and the
sheaf F.
We can also express Proposition 8 in the following manner: If G is a sheaf
on Y , and if G
X
is the sheaf obtained by extending G by 0 outside Y , we have
H
q
(Y, G) = H
q
(X, G
X
) for all q 0; in other words, the identication of G
with G
X
is compatible with passing to cohomology groups.
31
4. Comparison of cohomology groups of dierent coverings I
4 Comparison of cohomology groups of
different coverings
In this paragraph, X denotes a topological space and F is a sheaf on X. We pose
conditions on a covering U of X, under which we have H
n
(U, F) = H
n
(X, F)
for all n 0.
27 Double complexes
A double complex (cf. [6], Chapter VI, 4) is a bigraded abelian group
K =
p,q
K
p,q
, p 0, q 0,
equipped with two endomorphisms d
and d
satisfying the following properties:

d
maps K
p,q
to K
p+1,q
and d
maps K
p,q
to K
p,q+1
,
d
= 0, d
+d
= 0, d
= 0.
An element of K
p,q
is said to be bihomogenous of bidegree (p, q), and of
total degree p + q. The endomorphism d = d
+ d
satises d d = 0, and the

cohomology groups of K with respect to this coboundary operator are denoted
by H
n
(K), where n means the total degree.
We can treat d
as a coboundary operator on K; since d
is compatible with
the bigrading of K, we also obtain cohomology groups, denoted by H
p,q
I
(K); for
d
, we have the groups H

p,q
II
(K).
We denote by K
q
II
the subgroup of K
0,q
consisting of elements x such that
d
(x) = 0, and by K
II
the direct sum of K
q
II
(q = 0, 1, . . .). We have an
analogous denition of K
I
=
p=0
K
p
I
. We note that
K
q
II
= H
0,q
I
(K) and K
p
I
= H
p,0
II
(K).
K
II
is a subcomplex of K, and the operator d coincides on K
II
with the
operator d
.
Proposition 1. If H
p,q
I
(K) = 0 for p > 0 and q 0, the inclusion K
II
K
denes a bijection from H
n
(K
II
) to H
n
(K), for all n 0.
(Cf. [4], statement XVII-6, whose proof we shall repeat here).
By replacing K by K/K
II
, we are led to prove that if H
p,q
I
(K) = 0 for p 0
and q 0, then H
n
(K) = 0 for all n 0. Put
K
h
=
qh
K
p,q
.
The groups K
h
(h = 0, 1, . . .) are subcomplexes embedded in K, and K
h
/K
h+1
is
isomorphic to
p=0
K
p,h
, equipped with the coboundary operator d
. We thus
32
have H
n
(K
h
/K
h+1
) = H
h,nh
I
(K) = 0 for any n and h, therefore H
n
(K
h
) =
H
n
(K
h+1
). Since H
n
(K
h
) = 0 if h > n, we deduce, by descending recursion on
h, that H
n
(K
h
) = 0 for all n and h, and since K
0
is equal to K, the Proposition
follows.
28 The double complex dened by two coverings
Let U = U
i
iI
and V = V
j
jJ
be two coverings of X. If s is a p-simplex of
S(I) and s
a q-simplex of S(J), we denote by U

s
the intersection of U
i
, i s
(cf. n
18), the intersection of V

j
, j s
, by V
s
the covering of U
s
formed by
U
s
V
j
jJ
and by U
s
the covering of V
s
formed by V
s
U
i
iI
.
We dene a double complex C(U, V; F) =
p,q
C
p,q
(U, V; F) as follows:
C
p,q
(U, V; F) =
(U
s
V
s
, F), the product taken over all pairs (s, s
)
where s is a simplex of dimension p of S(I) and s
is a simplex of dimension q
of S(J).
An element f C
p,q
(U, V; F) is thus a system (f
s,s
) of sections of F on
U
s
V
s
or, with the notations of n
18, it is a system
f
i0...ip,j0...jq
(U
i0...ip
V
j0...jq
, F).
We can also identify C
p,q
(U, V; F) with
s
C
p
(U
s
, F); thus, for all s
, we
have a coboundary operator d : C
p
(U
s
, F C
p+1
(U
s
, F), giving a homomor-
phism
d
U
: C
p,q
(U, V; F) C
p+1,q
(U, V; F).
Making the denition of d
U
explicit, we obtain:
(d
U
f)
i0...ip+1,j0...jq
=
k=p+1
k=0
(1)
k
k
(f
i0...
i
k
...ip+1,j0...jq
),
k
being the restriction homomorphism dened by the inclusion of
U
i0...ip
V
j0...jq
in U
i0...i
k
...ip+1
V
j0...jq
.
We dene d
V
: C
p,q
(U, V; F) C
p,q+1
(U, V; F) the same way and we have
(d
V
f)
i0...ip,j0...jq+1
=
h=q+1
h=0
(1)
h
h
(f
i0...ip,j0...
j
h
...jq+1
).
It is clear that d
U
d
U
= 0, d
U
dV = d
V
d
U
, d
V
d
V
= 0. We thus
put d
= d
U
, d
= (1)
p
d
V
, equipping C(U, V; F) with a structure of a double
complex. We now apply to K = C(U, V; F) the denitions from the preceding
n
; the groups or complexes denoted in the general case by H

n
(K), H
p,q
I
(K),
33
H
p,q
I
(K), H
p,q
II
(K), K
I
, K
II
will be denoted by H
n
(U, V; F), H
p,q
I
(U, V; F),
H
p,q
II
(U, V; F), C
I
(U, V; F) and C
II
(U, V; F), respectively.
From the denitions of d
and d
, we immediately obtain:
Proposition 2. H
p,q
I
(U, V; F) is isomorphic to
s
H
p
(U
s
, F), the prod-
uct being taken over all simplexes of dimension q of S(J). In particular,
C
q
II
(U, V; F) = H
0,q
I
(U, V; F)
is isomorphic to
s
H
0
(U
s
, F) = C
q
(V, F).
We denote by
the canonical isomorphism: C(V, F) C

II
(U, V; F). If
(f
j0...jq
) is an element of C
q
(V, F), we thus have
(
f)
i0,j0...jq
=
i0
(f
j0...jq
),
where
i0
denotes the restriction homomorphism dened by the inclusion of
U
i0
V
j0...jq
in V
j0...jq
.
Obviously, a statement analogous to Proposition 2 holds for H
p,q
I
I(U, V; F),
and we have an isomorphism
: C(U, F) C
I
(U, V; F).
29 Applications
Proposition 3. Assume that H
p
(U
s
, F) = 0 for every s
and all p > 0. Then

the homomorphism H
n
(V, F) H
n
(U, V; F), dened by
, is bijective for all

n 0.
This is an immediate consequence of Propositions 1 and 2.
Before stating Proposition 4, we prove a lemma:
Lemma 1. Let W = W
i
iI
be a covering of a space Y and let F be a
sheaf on Y . If there exists an i I such that W
i
= Y , then H
p
(W, F) = 0 for
all p > 0.
Let W
be a covering of Y consisting of a single open set Y ; we obviously

have W W
, and the assumption made on W means that W
W. In result
(n
22) we have H
p
(W, F) = H
p
(W
, F) = 0 if p > 0.
Proposition 4. Suppose that the covering V is ner than the covering U.
Then
: H
n
(V, F) H
n
(U, V; F) is bijective for all n 0. Moreover, the
homomorphism

1
: H
n
(U, F) H
n
(V, F) coincides with the homomor-
phism (V, U) dened in n
21.
We apply Lemma 1 to W = U
s
and Y = V
s
, seeing that H
p
(U
s
, F) = 0
for all p > 0, and then Proposition 3 shows that
: H
n
(V, F) H
n
(U, V; F)
34
is bijective for all n 0.
Let : J I be a mapping such that V
j
U
j
; for the proof of the second
part of the Proposition, we need to observe that if f is an n-cocycle of C(U, F),
the cocycles
(f) and
(f) are cohomologous in C(U, V; F).

For any integer p, 0 p n 1, dene g
p
C
p,np1
(U, V; F) by the
following formula
g
p
i0...ip,j0...jnp1
=
p
(f
i0...ipj0...jnp
),
p
denoting the restriction dened by the inclusion of
U
i0...ip
V
j0...jnp1
in U
i0...ip,j0...jnp1
.
We verify by a direct calculation (keeping in mind that f is a cocycle) that
we have
d
(g
0
) =
(f), . . . , d
(g
p
) = d
(g
p1
), . . . d
(g
n1
) = (1)
n
(f)
hence d(g
0
g
1
+. . . +(1)
n1
g
n1
) =
(f)
(f), which shows that
(f)
and
(f) are cohomologous.

Proposition 5. Suppose that V is ner than U and that H
q
(V
s
, F) = 0 for
all s and all q > 0. Then the homomorphism (V, U) : H
n
(U, F) H
n
(V, F)
is bijective for all n 0.
If we apply Proposition 3, switching the roles of U and V, we see that
:
H
n
(V, F) H
n
(U, V; F) is bijective. The Proposition then follows directly
from Proposition 4.
Theorem 1. Let X be a topological space, U = U
i
iI
a covering of X, F
a sheaf on X. Assume that there exists a family V
, A of coverings of X
satisfying the following properties:
(a) For any covering W of X, there exists an A with V
W,
(b) H
q
(V
s
, F) = 0 for all A, all simplexes s S(I) and every q > 0,
Then (U) : H
n
(U, F) H
n
(X, F) is bijective for all n 0.
Since V
are arbitrarily ne, we can assume that they are ner than U. In
this case, the homomorphism
(V
, U) : H
n
(U, F) H
n
(V
, F)
is bijective for all n 0, by Proposition 5. Because V
are arbitrarily ne,

H
n
(X, F) is the inductive limit of H
n
(V
, F), and the theorem follows.

Remarks. (1) It is probable that Theorem 1 remains valid when we replace
the condition (b) with the following weaker condition:
(b) lim
H
q
(V
s
, F) = 0 for any simplex s of S(I) and any q > 0.
(2) Theorem 1 is analogous to a theorem of Leray on acyclic coverings. Cf.
[10] and also [4], statement XVII-7.
35
Chapter II
Algebraic Varieties
Coherent Algebraic
Sheaves on Ane Varieties
36
1. Algebraic varieties II
From now on, K denotes a commutative algebraically closed eld of arbitrary
characteristic.
1 Algebraic varieties
30 Spaces satisfying condition (A)
Let X be a topological space. The condition (A) is the following:
(A) Any decreasing sequence of closed subsets of X is stationary.
In other words, if we have F
1
F
2
F
3
. . ., F
i
being closed in X, there
exists an integer n such that F
m
= F
n
for m n. Or:
(A) The set of closed subsets of X, ordered by inclusion, satises the
minimality condition
Examples. Equip a set X with the topology where the closed subsets are
the nite subsets of X and the whole X; the condition (A) is then satised.
More generally, any algebraic variety, equipped with Zariski topology, satises
(A) (cf. n
34).
Proposition 1. (a) If X satises the condition (A), then X is quasi-
compact,
(b) If X satises (A), any subspace of X satises it also.
(c) If X is a nite union of Y
i
, the Y
i
satisfying (A), then X also satises
(A).
If F
i
is a ltering decreasing set of closed subsets of X, and if X satises (A),
then there exists an F
i
contained in all others; if
F
i
= , there is therefore an
i such that F
i
= , which shows (a).
Let G
1
G
2
G
3
. . . be a decreasing sequence of closed subsets of a
subspace Y of X; if X satises (A), there exists an n for which

G
m
=

G
n
for
m n, hence G
m
= Y

G
m
= Y

G
n
= G
n
, which shows (b).
Let F
1
F
2
F
3
. . . be a decreasing sequence of closed subsets of a space
X satisfying (c); since all Y
i
satisfy (A), there exists for all i an n
i
such that
F
m
Y
i
= F
ni
Y
i
for m n
i
; if n = Sup(n
i
), we then have F
m
= F
n
for
m n, which shows (c).
A space X is said to be irreducible if it is not a union of two closed subspaces,
distinct from X itself; or equivalently, if any two non-empty open subsets have
a non-empty intersection. Any nite family of non-empty open subsets of X
then has a non-empty intersection, and any open subset of X is also irreducible.
Proposition 2. Any space X satisfying the condition (A) is a union of a
nite number of irreducible closed subsets Y
i
. If we suppose that that Y
i
is not
contained in Y
j
for any pair (i, j), i ,= j, the set of Y
i
is uniquely determined by
X; the Y
i
are then called the irreducible components of X.
37
The existence of a decomposition X =
Y
i
follows immediately from (A).
If Z
k
is another such decomposition of X, we have Y
i
=
Y
i
Z
k
, and, since
Y
i
is irreducible, this implies of an index k such that Z
k
Y
i
; interchanging
the roles of Y
i
and Z
k
, we conclude analogously that there exists an index i
for
which Y
i
Z
k
; thus Y
i
Z
k
Y
i
, which by the assumption made on Y
i
leads
to i = i
and Y
i
= Z
k
, hence the uniqueness of the decomposition.
Proposition 3. Let X be a topological space that is a nite union of non-
empty open subsets V
i
. Then X is irreducible if and only if all V
i
are irreducible
and V
i
V
j
,= 0 for all pairs (i, j).
The necessity of these conditions was noted above; we show that they are
sucient. If X = Y Z, where Y and Z are closed, we have V
i
= (V
i
Y )(V
i
Z), which shows that each V

i
is contained either in Y or in Z. Suppose that Y
and Z are distinct from X; we can then nd two indices i, j such that V
i
is not
contained in Y and V
j
is not contained in Z; according to our assumptions on
Y
i
, we then have V
i
Z and V
j
Y . Set T = V
j
V
i
V
j
; T is closed in V
j
and
we have V
j
= T (Z V
j
); as V
j
is irreducible, it follows that either T = V
j
,
which means that V
i
V
j
= , or Z V
j
= V
j
, which means that V
j
Z, and
in both cases this leads to a contradiction, q.e.d.
31 Locally closed subsets of an ane space
Let r be an integer 0 and let X = K
r
be the ane space of dimension r
over the eld K. We equip X with the Zariski topology; recall that a subset
of X is closed in this topology if it is the zero set of a family of polynomials
P
K[X
1
, . . . , X
r
]. Since the ring of polynomials is Noetherian, X satises
the condition (A) from the preceding n
. Moreover, one easily shows that X is

an irreducible space.
If x = (x
1
, . . . , x
r
) is a point of X, we denote by O
x
the local ring of x; recall
that this is the subring of the eld K(X
1
, . . . , X
r
) consisting of those fractions
which can be put in the form:
R = P/Q, where P and Q are polynomials and Q(x) ,= 0.
Such a fraction is said to be regular in x; for all points x X for which
Q(x) ,= 0, the function x P(x)/Q(x) is a continuous function with values in
K (K being given the Zariski topology) which can be identied with R, the eld
K being innite. The O
x
, x X thus form a subsheaf O of the sheaf F(X) of
germs of functions on X with values in K (cf. n
3); the sheaf O is a sheaf of

rings.
We will extend the above to locally closed subspaces of X (we call a subset
of a space X locally closed in X if it is an intersection of a open subset with a
closed subset of X). Let Y be such a subspace and let F(Y ) be the sheaf of
germs of functions on Y with values in K; if x is a point of Y , the operation of
restriction denes a canonical homomorphism
x
: F(X)
x
F(Y )
x
.
38
The image of O
x
under
x
is a subring of F(Y )
x
which we denote by O
x,Y
;
the O
x,Y
form a subsheaf O
Y
of F(Y ), which we call the sheaf of local rings
of Y . A section of O
Y
over an open subset V of Y is thus, by denition, a
function f : V K which is equal, in the neighborhood of any point x V ,
to a restriction to V of a rational function regular at x; such a function is said
to be regular on V ; it is a continuous function if we equip V with the induced
topology and K with the Zariski topology. The set of regular functions at all
points of V is a ring, the ring (V, O
Y
); observe also that, if f (V, O
x
) and
if f(x) ,= 0 for all x V , then 1/f also belongs to (V, O
Y
).
We can characterize the sheaf O
Y
in another way:
Proposition 4. Let U (resp. F) be a open (resp. closed) subspace of X and
let Y = U F. Let I(F) be the ideal K[X
1
, . . . , X
r
] consisting of polynomials
vanishing on F. If x is a point of Y , the kernel of the surjection
x
: O
x
O
x,Y
coincides with the ideal I(F) O
x
of O
x
.
It is clear that each element of I(F) O
x
belongs to the kernel of
x
. Con-
versely, let R = P/Q be an element of the kernel, P and Q being two polynomials
with Q(x) ,= 0. By assumption, there exists an open neighborhood W of x such
that P(y) = 0 for all y W F; let F
be the complement of W, which is

closed in X; since x F
, there exists, by the denition of the Zariski topology,

a polynomial P
1
vanishing on F
and nonzero at x; the polynomial P P

1
belongs
to I(F) and we can write R = P P
1
/Q P
1
, which shows that R I(F) O
x
.
Corollary. The ring O
x,Y
is isomorphic to the localization of K[X
1
, . . . , X
r
]/I(F)
in the maximal ideal dened by the point x.
This follows immediately from the construction of localization a quotient
ring (cf. for example [8], Chap. XV, 5, th. XI).
32 Regular functions
Let U (resp. V ) be a locally closed subspace of K
r
(resp. K
s
). A function
: U V is said to be regular on U (or simply regular) if:
is continuous,
If x U and f O
(x),V
then f O
x,U
.
Denote the coordinates of the point (x) by
i
(x), 1 i s. We then have:
Proposition 5. A map : U V is regular on U if and only if
i
: U K
are regular on U for all i, 1 i s.
As the coordinate functions are regular on V , the condition is necessary.
Conversely, suppose that we have
i
(U, O
U
) for each i; if P(X
1
, . . . , X
s
)
is a polynomial, the function P(
1
, . . . ,
s
) belongs to (U, O
U
) since (U, O
U
)
as a ring; it follows that it is a continuous function on U, thus its zero set is
closed, which shows the continuity of . If we have x U and f O
(x),V
, we
can write f locally in the form f = P/Q, where P and Q are polynomials and
39
Q((x)) ,= 0. The function f is then equal to P /Q in a neighborhood
of x; from what we gave seen, P and Q are regular in a neighborhood of
x. As Q (x) ,= 0, it follows that f is regular in a neighborhood of x, q.e.d.
A composition of two regular maps is regular. A bijection : U V is
called a biregular isomorphism (or simply an isomorphism) if and
1
are
regular; or equivalently, if is a homeomorphism of U to V which transforms
the sheaf O
U
into the sheaf O
V
.
33 Products
If r and r
are two nonnegative integers, we identify the ane space K

r+r
with
the product K
r
K
r
. The Zariski topology on K

r+r
is ner than the product

of the Zariski topologies on K
r
and K
r
; it is even strictly ner if r and r
are positive. In result, if U and U
are locally closed subspaces of K

r
and K
r
,
U U
is a locally closed subspace of K

r+r
and the sheaf O

UU
is well dened.
On the other hand, let W be a locally closed subspace of K
t
, t 0 and
let : W U and
: W U
be two maps. As an immediate result of

Proposition 5 we have:
Proposition 6. A map x ((x),
(x)) is regular from W to U U
if
and only if and
are regular.
As any constant function is regular, the preceding Proposition shows that
any section x (x, x
0
), x
0
U
is a regular function from U to U U
; on
the other hand, the projections U U
U and U U
are obviously
regular.
Let V and V
be locally closed subspaces of K

s
and K
s
and let : U V
and
: U
be two mappings. The preceding remarks, together with

Proposition 6, show that we then have (cf. [1], Chap. IV):
Proposition 7. A map
: U U
V V
is regular if and only if

and
are regular.
Hence:
Corollary. A map
is a biregular isomorphism if and only if and
are biregular isomorphisms.

34 Denition of the structure of an algebraic variety
Denition. We call an algebraic variety over K (or simply an algebraic variety)
a set X equipped with:
1
a topology,
2
a subsheaf O
x
of the sheaf F(X) of germs of functions on X with values
in K,
this data being subject to axioms (V A
I
) and (V A
II
) stated below.
40
First note that if X and Y are equipped with two structures of the above
type, we have a notion of isomorphism of X and Y : it is a homeomorphism
of X to Y which transforms O
X
to O
X
. On the other hand, if X
is an open
subset of X, we can equip X
with the induced topology and the induced sheaf:

we have a notion of an induced structure on an open subset. That being said,
we can state the axiom (V A
I
):
(V A
I
) There exists a nite open covering V = V
i
iI
of the space X
such that each V
i
, equipped with the structure induced from X, is isomorphic
to a locally closed subspace U
i
of an ane space, equipped with the sheaf O
Ui
dened in n
31.
To simplify the language, we call an prealgebraic variety a topological space
X together with a sheaf O
X
satisfying the axiom (V A
I
). An isomorphism
i
: V
i
U
i
is called a chart of the open subset V
i
; the condition (V A
I
)
means that it is possible to cover X with nitely many open subsets possessing
charts. Proposition 1 from n
30 shows that X satises condition (A), thus it

is quasi-compact and so are its subspaces.
The topology on X is called the ,,Zariski topology and the sheaf O
X
is
called the sheaf of local rings of X.
Proposition 8. Let X be a set covered by a nite family of subsets X
j
,
j J. Suppose that each X
j
is equipped with a structure of a prealgebraic
variety and that the following conditions are satised:
(a) X
i
X
j
is open in X
i
for all i, j J,
(b) the structures induced by X
i
and X
j
on X
i
X
j
coincide for all i, j J.
Then there exists a unique structure of a prealgebraic variety on X such that
X
j
are open in X and such that the structure induced on each X
i
is the given
structure.
The existence and uniqueness of the topology on X and the sheaf O
X
are
immediate; it remains to check that this topology and this sheaf satisfy (V A
I
),
which follows from the fact that X
j
form a nite family and satisfy (V A
I
).
Corollary. Let X and X
be two prealgebraic varieties. There exists a

structure of a prealgebraic variety on X X
satisfying the following condition:

If : V U and
: V
are charts (V being open in X and V
being
open in X
), then V V
is open in X X
and
: V V
U U
is a
chart.
Cover X by a nite number of open V
i
having charts
i
: V
i
U
i
and
let (V
j
, U
j
,
j
) be an analogous system for X
. The set X X
is covered by
V
i
V
j
; equip each V
i
V
j
with the structure of a prealgebraic variety induced
from U
i
U
j
by
1
i

1
j
; the assumptions (a) and (b) of Proposition 8
are satised for this covering of X X
, by the corollary of Proposition 7. We

obtain a structure of a prealgebraic variety on XX
which satises appropriate

conditions.
41
We can apply the preceding corollary to the particular case X
= X; so
X X has a structure of a prealgebraic variety, and in particular a topology.
We can now state the axiom (V A
II
):
(V A
II
) The diagonal of X X is closed in X X.
Suppose that X is a prealgebraic variety obtained by the ,,gluing procedure
of Proposition 8; then the condition (V A
II
) is satised if and only if X
ij
=
X
i
X
j
is closed in X
i
X
j
. Or X
ij
is the set of (x, x) for x X
i
X
j
.
Suppose that there exist charts : X
i
U
i
and let T
ij
=
j
(X
ij
); T
ij
is
the set of (
i
(x),
j
(x)) for x running over X
i
X
j
. The axiom (V A
II
) takes
therefore the following form:
(V A
II
) For each pair (i, j), T
ij
is closed in U
i
U
j
.
In this form we recognize Weils axiom (A) (cf. [16], p. 167), except that
Weil considered only irreducible varieties.
Examples of algebraic varieties: Any locally closed subspace U of an ane
space, equipped with the induced topology and the sheaf O
U
dened in n
31 is
an algebraic variety. Any projective variety is an algebraic variety (cf. n
51).
Any algebraic ber space (cf. [17]) whose base and ber are algebraic varieties
is an algebraic variety.
Remarks. (1) We observe an analogy between condition (V A
II
) and the
condition of separatedness imposed on topological, dierential and analytic va-
rieties.
(2) Simple examples show that condition (V A
II
) is not a consequence of
condition (V A
I
).
35 Regular mappings, induced structures, products
Let X and Y be two algebraic varieties and let be a function from X to Y .
We say that is regular if:
(a) is continuous.
(b) If x X and f O
(x),Y
then f O
x,X
.
As in n
32, the composition of two regular functions is regular and a bijec-

tion : X Y is an isomorphism if and only if and
1
are regular functions.
Regular functions form a family of morphisms for the structure of an algebraic
variety in the sense of [1], Chap. IV.
Let X be an algebraic variety and let X
be a locally closed subspace of X.

We equip X
with the topology induced from X and the sheaf O

X
induced by
O
X
(to be precise, for all x X
we dene O
x,X
as the image of O
x,X
under the
canonical homomorphism F(X)
x
F(X
)
x
). The axiom (V A
I
) is satised:
if
i
: V
i
U
i
is a system of charts such that X =
V
i
, we set V
i
= X
V
i
,
U
i
=
i
(V
i
) and
i
: V
i
U
i
is a system of charts such that X
i
. The
axiom (V A
II
) is satised as well since the topology of X
is induced from
42
X X (we could also use (V A
II
)). We dene the structure of an algebraic
variety on X
which is induced by that of X; we also say that X
is a subvariety
of X (in Weil [16], the term ,,subvariety is reserved for what we call here an
irreducible closed subvariety). If denotes the inclusion of X
in X, is a regular
mapping; moreover, if is a function from an algebraic variety Y to X
then
: Y X
is regular if and only if : Y X is regular (which justies the

term ,,induced structure, cf. [1], loc. cit.).
If X and X
are two algebraic varieties, XX
is an algebraic variety, called

the product variety; it suces to check that the axiom (V A
II
) is satised, in
other words, that if
i
: V
i
U
i
and
i
: V
i
U
i
are systems of charts such that
X =
V
i
and X
i
, then the set T
ij
T
j
is closed in U
i
U
j
U
i
V
(with the notations of n
34); this follows immediately from the fact that T

ij
and T
j
are closed in U
i
U
j
and U
i
U
j
respectively.
Propositions 6 and 7 are valid without change for arbitrary algebraic vari-
eties.
If : X Y is a regular mapping, the graph of is closed in X Y ,
because it is the inverse image of the diagonal Y Y by 1 : X Y
Y Y ; moreover, the mapping : X dened by (x) = (x, (x)) is
an isomorphism: indeed, is a regular mapping, and so is
1
(since it is a
restriction of the projection X Y X).
36 The eld of rational functions on an irreducible variety
We rst show two lemmas of purely topological nature:
Lemma 1. Let X be a connected space, G an abelian group and G a constant
sheaf on X isomorphic to G. The canonical mapping G (X, G) is bijective.
An element of (X, G) is just a continuous mapping from X to G equipped
with the discrete topology. Since X is connected, any such a mapping is con-
stant, hence the Lemma.
We call a sheaf F on a space X locally constant if any point x has an open
neighborhood U such that F(U) is constant on U.
Lemma 2. Any locally constant sheaf on an irreducible space is constant.
Let F be a sheaf, X a space and set F = (X, F); it suces to demonstrate
that the canonical homomorphism
x
: F F
x
is bijective for all x X,
because we would thus obtain an isomorphism of the constant sheaf isomorphic
to F with the given sheaf F.
If f F, the set of points x X such that f(x) = 0 is open (by the
general properties of sheaves) and closed (because F is locally constant); since
an irreducible space is connected, this set is either or X, which shows that
x
is injective.
43
Now take m F
x
and let s be a section of F over a neighborhood U of x
such that s(x) = m; cover X by nonempty open subsets U
i
such that F(U
i
)
is constant on U
i
; since X is irreducible, we have U U
i
,= ; choose a point
x
i
UU
i
; obviously there exists a section s
i
of F over U
i
such that s
i
(x
i
) = s(x
i
),
and since the sections s and s
i
coincide in x
i
, they coincide on whole U U
i
,
since U U
i
is irreducible, hence connected; analogously s
i
and s
j
coincide on
U
i
U
j
, since they coincide on U U
i
U
j
,= ; thus the sections s
i
dene a
unique section s of F over X and we have
x
(s) = m, which ends the proof.
Now let X be an irreducible algebraic variety. If U is a nonempty open
subset of X, set A
U
= (U, O
X
); A
U
is an integral domain: indeed, suppose
that we have f g = 0, f and g being regular functions from U to K; if F (resp.
G) denotes the set of x U such that f(x) = 0 (resp. g(x) = 0), we have
U = F G and F and G are closed in U, because f and g are continuous; since
U is irreducible, it follows that F = U or G = U, which means exactly that f
or g is zero on U. We can therefore form the eld of fractions of A
U
, which
we denote by K
U
; if U V , the homomorphism
V
U
: A
V
A
U
is injective,
because U is dense in V , and we have a well dened isomorphism
V
U
of K
V
to
K
U
; the system of K
U
,
V
U
denes a sheaf of elds K ; then K
x
is canonically
isomorphic with the eld of fractions of O
x,X
.
Proposition 9. For any irreducible algebraic variety X, the sheaf K dened
above is a constant sheaf.
By Lemma 2, it suces to show the Proposition when X is a locally closed
subvariety of the ane space K
r
; let F be the closure of X in K
r
and let I(F) be
the ideal in K[X
1
, . . . , X
r
] of polynomials vanishing on F (or equivalently on X).
If we set A = K[X
1
, . . . , X
r
]/I(F), the ring A is an integral domain because X
is irreducible; let K(A) be the ring of fractions of A. By corollary of Proposition
4, we can identify O
x,X
with the localization of A in the maximal ideal dened
by x; we thus obtain an isomorphism of the eld K(A) with the eld of fractions
of O
x,X
and it is easy to check that it denes an isomorphism of the constant
sheaf equal to K(A) with the sheaf K , which shows the Proposition.
By Lemma 1, the sections of the sheaf K form a eld, isomorphic with K
x
for all x X, which we denote by K(X). We call it the eld of rational functions
on X; it is an extension of nite type
1
of the eld K, whose transcendence degree
over K is the dimension of X (we extend this denition to reducible varieties
by imposing dimX = Sup dimY
i
if X is a union of closed irreducible varieties
Y
i
). In general, we identify the eld K(X) with the eld K
x
; since we have
O
x,X
K
x
, we see that we can view O
x,X
as a subring of K(X) (it is the ring
of specialization of the point x in K(X) in the sense of Weil, [16], p. 77). If U
is an open subset of X, (U, O
X
) is the intersection in K(X) of the rings O
x,X
for x running over U.
If Y is a subvariety of X, we have dimY dimX; if furthermore Y is closed
and does not contain any irreducible component of X, we have dimY < dimX,
1
i.e. nitely generated
44
as shown by reducing to the case of subvarieties of K
r
(cf. for example [8],
Chap. X, 5, th. II).
45
2. Coherent algebraic sheaves II
2 Coherent algebraic sheaves
37 The sheaf of local rings on an algebraic variety
Return to the notations of n
31: let X = K
r
and let O be the sheaf of local
rings of X. We have:
Lemma 1. The sheaf O is a coherent sheaf of rings, in the sense of n
15.
Let x X, let U be an open neighborhood of x and let f
1
, . . . , f
p
be sections
of O over U, i.e. rational functions regular at each point of U; we must show
that the sheaf of relations between f
1
, . . . , f
p
is a sheaf of nite type over O.
Possibly replacing U by a smaller neighborhood, we can assume that f
i
can be
written in the form f
i
= P
i
/Q where P
i
and Q are polynomials and Q does
not vanish on U. Let now y U and g
i
O
y
such that
i=p
i=1
g
i
f
i
is zero in a
neighborhood of y; we can again write g
i
in the form g
i
= R
i
/S where R
i
and
S are polynomials and S does not vanish in y. The relationship ,,
i=p
i=1
g
i
f
i
= 0
in a neighborhood of y is equivalent to the relationship ,,
i=p
i=1
R
i
P
i
= 0 in
a neighborhood of y, i.e. equivalent to
i=p
i=1
R
i
P
i
= 0. As the module of
relations between the polynomials P
i
is a module of nite type (because the
ring of polynomials is Noetherian), it follows that the sheaf of relations between
f
i
is of nite type.
Let now V be a closed subvariety of X = K
r
; for any x X let J
x
(V ) be
the ideal of O
x
consisting of elements f O
x
whose restriction to V is zero in
a neighborhood of x (we thus have J
x
(V ) = O
x
if x / V ). The J
x
(V ) form a
subsheaf J(V ) of the sheaf O.
Lemma 2. The sheaf J(V ) is a coherent sheaf of O-modules.
Let I(V ) be the ideal of K[X
1
, . . . , X
r
] consisting of polynomials P vanishing
on V . By Proposition 4 from n
31, J
x
(V ) is equal to I(V ) O
x
for all x V
and this formula remains valid for x / V as shown immediately. The ideal I(V )
being generated by a nite number of elements, it follows that the sheaf J(V )
is of nite type, thus coherent by Lemma 1 and Proposition 8 from n
15.
We shall now extend Lemma 1 to arbitrary algebraic varieties:
Proposition 1. If V is an algebraic variety, the sheaf O
V
is a coherent
sheaf of rings on V .
The question being local, we can suppose that V is a closed subvariety of
the ane space K
r
. By Lemma 2, the sheaf J(V ) is a coherent sheaf of ideals,
thus the sheaf O/J(V ) is a coherent sheaf of rings on X, by Theorem 3 from
n
16. This sheaf of rings is zero outside V and its restriction to V is just O
V
(n
31); thus the sheaf O

V
is a coherent sheaf of rings on V (n
17, corollary of
Proposition 11).
Remark. It is clear that Proposition is valid more generally for any preal-
gebraic variety.
46
38 Coherent algebraic sheaves
If V is an algebraic variety whose sheaf of local rings is O
V
, we call an algebraic
sheaf on V any sheaf of O
V
modules, in the sense of n
6; if F and G are
two algebraic sheaves, we say that : F G is an algebraic homomorphism
(or simply a homomorphism) if it is a O
V
homomorphism; recall that this is
equivalent to saying that
x
: F
x
G
x
is O
x,V
linear and that transforms
local sections of F into local sections of G.
If F is an algebraic sheaf on V , the cohomology groups H
q
(V, F) are mod-
ules over (V, O
V
), cf. n
23; in particular, they are vector spaces over K.

An algebraic sheaf F over V is said to be coherent if it is a coherent sheaf
of O
V
modules, in the sense of n
12; by Proposition 7 of n
15 and Proposi-
tion 1 above, these sheaves are characterized by the property of being locally
isomorphic to the cokernel of an algebraic homomorphism : O
q
V
O
p
V
.
We shall give some examples of coherent algebraic sheaves (we will see more
of them later, cf. in particular n
os
48, 57).
39 Sheaf of ideals dened by a closed subvariety
Let W be a closed subvariety of an algebraic variety V . For any x V , let
J
x
(W) be the ideal of O
x,V
consisting of elements f whose restriction to W
is zero in a neighborhood of x; let J(W) be the subsheaf of O
V
formed by
J
x
(W). We have the following Proposition, generalizing Lemma 2:
Proposition 2. The sheaf J(W) is a coherent algebraic sheaf.
The question being local, we can suppose that V (thus also W) is a closed
subvariety of the ane space K
r
. It follows from Lemma 2, applied to W, that
the sheaf of ideals dened by W in K
r
is of nite type; this shows that J(W),
which is its image under the canonical homomorphism O O
V
, is also of nite
type, thus is coherent by Proposition 8 of n
15 and Proposition 1 of n
37.
Let O
W
be the sheaf of local rings of W and let O
V
W
be the sheaf on V
obtained by extending O
W
by 0 outside W (cf. n
5); this sheaf is canonically

isomorphic to O
V
/J(W), in other words, we have an exact sequence:
0 J(W) O
V
O
V
W
0.
Let then F be an algebraic sheaf on W and let F
V
be the sheaf obtained by
extending F by 0 outside W; we can consider F
V
as a sheaf of O
V
W
modules,
thus also as a sheaf of O
V
modules whose annihilator contains J(W). We
have:
Proposition 3. If F is a coherent algebraic sheaf on W, F
V
is a coherent
algebraic sheaf on V . Conversely, if G is an coherent algebraic sheaf on V whose
47
annihilator contains J(W), the restriction of G to W is a coherent algebraic
sheaf on W.
If F is a coherent algebraic sheaf on W, F
V
is a coherent sheaf of O
V
W
modules (n
17, Proposition 11), thus a coherent sheaf of O

V
modules (n
16, Theorem 3). Conversely, if G is a coherent algebraic sheaf on V whose

annihilator contains J(W), G can be considered as a sheaf of O
V
/J(W)
modules, and is a coherent sheaf (n
16, Theorem 3); the restriction of G to W

is then a coherent sheaf of O
W
modules (n
17, Proposition 11).

So, any coherent algebraic sheaf on W can by identied with an algebraic
coherent sheaf on V (and this identication does not change cohomology groups,
by Proposition 8 of n
26). In particular, any coherent algebraic sheaf on an

ane (resp. projective) variety can be considered as a coherent algebraic sheaf
on an ane (resp. projective) space; we will frequently use this possibility later.
Remark. Let G be a coherent algebraic sheaf on V which is zero outside
W; the annihilator of G does not necessarily contain J(W) (in other words, G
not always can be considered as an coherent algebraic sheaf on W); all we can
say is that it contains a power of J(W).
40 Sheaves of fractional ideals
Let V be an irreducible algebraic variety and let K(V ) denote the constant
sheaf of rational functions on V (cf. n
36); K(V ) is an algebraic sheaf which

is not coherent if dimV > 0. An algebraic subsheaf F of K(V ) can be called a
,,sheaf of fractional ideals since each F
x
is a fractional ideal of O
x,V
.
Proposition 4. An algebraic subsheaf F of K(V ) is coherent if and only
if it is of nite type.
The necessity is trivial. To prove the suciency, it suces to prove that
K(V ) satises condition (b) of denition 2 from n
12, in other that if f

1
, . . . , f
p
are rational functions, the sheaf R(f
1
, . . . , f
p
) is of nite type. If x is a point of
V , we can nd functions g
i
and h such that f
i
= g
i
/h, g
i
and h being regular
in a neighborhood U of x and h being nonzero on U; the sheaf R(f
1
, . . . , f
p
)
is then equal to the sheaf R(g
1
, . . . , g
p
), which is of nite type, since O
V
is a
coherent sheaf of rings.
41 Sheaf associated to the total space of a vector bundle
Let E be an algebraic ber space with a vector space of dimension r as a ber
and an algebraic variety V as a base; by denition, the typical ber of E is a
vector space K
r
and the structure group is the linear group GL(r, K) acting on
K
r
in the usual way (for the denition of an algebraic ber space, cf. [17]; see
also [15], n
4 for analytic vector bundles).

48
If U is an open subset of V , let S(E)
U
denote the set of regular sections of E
on U; if V U, we have the restriction homomorphism
V
U
: S(E)
V
S(E)
U
; thus a sheaf S(E), called the sheaf of germs of sections of E. Since E is a
vector bundle, each S(E)
U
is a (U, O
V
)module and it follows that S(E) is
an algebraic sheaf on V . If we identify locally E with V K
r
, we have:
Proposition 5. The sheaf S(E) is locally isomorphic to O
r
V
; in particular,
it is a coherent algebraic sheaf.
Conversely, it is easily seen that any algebraic sheaf F on V , locally iso-
morphic to O
r
V
, is isomorphic to a sheaf S(E) where E is determined up to
isomorphism (cf. [15] for the analytic case).
If V is a variety without singularities, we can take for E the vector bundle
of p-covectors tangent to V (p being a nonnegative integer); let
p
be the sheaf
corresponding to S(E); an element of
p
x
, x V is just a dierential form of
degree p on V , regular in x. If we set h
p,q
= dim
K
H
q
(V,
p
), we know that
in the classical case (and if V is projective), h
p,q
is equal to the dimension of
harmonic forms of type (p, q) (theorem of Dolbeault
2
and, if B
n
denotes the n-th
Betti number of V , we have B
n
=
p+q=n
h
p,q
. In the general case, we could
take the above formula for the denition of the Betti numbers of a nonsingular
projective variety (we will see in n
66 that h
p,q
are nite). It is convenient to
study their properties, in particular to see if they coincide with those involved
in the Weil conjectures for varieties over nite elds
3
. We only mention that
they satisfy the ,,Poincare duality B
n
= B
2mn
when V is an irreducible of
dimension m.
The cohomology groups H
q
(V, S(E)) are also involved in other issues, in-
cluding the Riemann-Roch, as well as in the classication of algebraic ber
spaces with base V and the ane group x ax +b as the structural group (cf.
[17], 4, where the case when dimV = 1 is studied).
2
P. Dolbeault. Sur la cohomologie des varietes analytiques complexes. C. R. Paris, 246,
1953, p. 175-177.
3
Bulletin Amer. Math. Soc., 55, 1949, p.507
49
3. Coherent algebraic sheaves on ane varieties II
3 Coherent algebraic sheaves on affine
varieties
42 Ane varieties
An algebraic variety V is said to be ane if it is isomorphic to a closed subvariety
of an ane space. The product of two ane varieties is an ane variety; any
closed subvariety of an ane variety is an ane variety.
An open subset U of an algebraic variety V is said to be ane if, equipped
with the structure of an algebraic variety induced from X, it is an ane variety.
Proposition 1. Let U and V be two open subsets of an algebraic variety
X. If U and V are ane, U V is ane.
Let be the diagonal of X X; by n
35, the mapping x (x, x) is a

biregular isomorphism from X onto ; thus the restriction of this map to U V
is a biregular isomorphism of U V onto U V . Since U and V are ane
varieties, U V is also an ane variety; on the other hand, is closed in XX
by the axiom (V A
II
), thus U V is closed in U V , hence ane, q.e.d.
(It is easily seen that this Proposition is false for prealgebraic varieties; the
axiom (V A
II
) plays an essential role).
Let us now introduce a notation which will be used thorough the rest of
this paragraph: if V is an algebraic variety and f is a regular function on V ,
we denote by V
f
the open subset of V consisting of all points x V for which
f(x) ,= 0.
Proposition 2. If V is an ane algebraic variety and f is a regular function
on V , the open subset V
f
is ane.
Let W be the subset of V K consisting of pairs (x, ) such that f(x) = 1;
it is clear that W is closed in V K, thus it is an ane variety. For all (x, ) W
set (x, ) = x; the mapping is a regular mapping from W to V
f
. Conversely,
for all x V
f
, set (x) = (x, 1/f(x)); the mapping : V
f
W is regular and
we have = 1, = 1, thus V
f
and W are isomorphic, q.e.d.
Proposition 3. Let V be a closed subvariety of K
r
, F be a closed subset of
V and let U = V F. The open subsets V
P
form a base for the topology of U
when P runs over the set of polynomials vanishing on F.
Let U
= V F
be an open subset of U and let x inU
; we must show that

there exists a P for which V
P
U
and x V
P
; in other words, P has to be
zero on F
and nonzero in x; the existence of such a polynomial follows simply

from the denition of the topology of K
r
.
Theorem 1. The open ane subsets of an algebraic variety X form an
open base for the topology of X.
50
The question being local, we can assume that X is a locally closed sub-
space of an ane space K
r
; in this case, the theorem follows immediately from
Propositions 2 and 3.
Corollary. The coverings of X consisting of open ane subsets are arbi-
trarily ne.
We note that if U = U
i
iI
is such a covering, the U
i0...ip
are also open
ane subsets, by Proposition 1.
43 Some preliminary properties of irreducible varieties
Let V be a closed subvariety of K
r
and let I(V ) be the ideal of K[X
1
, . . . , X
r
]
consisting of polynomials vanishing on V ; let Abe the quotient ring K[X
1
, . . . , X
r
]/I(V );
we have a canonical homomorphism
: A (V, O
V
)
that is injective by the denition of I(V ).
Proposition 4. If V is irreducible, : A (V, O
V
) is bijective.
(In fact, this holds for any closed subvariety of K
r
, as will be shown in the
next n
).
Let K(V ) be the eld of fractions of A; by n
36, we can identify O

x,V
with
the localization of A in the maximal ideal m
x
consisting of polynomials vanishing
in x, and we have (V, O
V
) = A =
xV
O
x,V
(all O
x,V
being considered as
subrings of K(V )). But all maximal ideals of A are m
x
, since K is algebraically
closed (Hilberts theorem of zeros); it follows immediately (cf. [8], Chap. XV,
5, th. X) that A =
xV
O
x,V
= (V, O
V
), q.e.d.
Proposition 5. Let X be an irreducible algebraic variety, Q a regular func-
tion on X and P a regular function on X
Q
. Then, for n suciently large, the
rational function Q
n
P is regular on the whole of X.
By quasi-compactness of X, the question is local; by Theorem 1, we can thus
suppose that X is a closed subvariety of K
r
. The above Proposition shows that
then Q is an element of A = K[X
1
, . . . , X
r
]/(I(X)). The assumption made on
P means that for any point x X
Q
we can write P = P
x
/Q
x
with P
x
and Q
x
in A and Q
x
(x) ,= 0; if a denotes the ideal of A generated by all Q
x
, the variety
of zeros of a is contained in the variety of zeros of Q; by Hilberts theorem of
zeros, this leads to Q
n
a for n suciently large, hence Q
n
=
R
x
Q
x
and
Q
n
P =
R
x
P
x
with R
x
A, which shows that Q
n
P is regular on X.
(We could also use the fact that X
Q
is ane if X is and apply Proposition
4 to X
Q
).
Proposition 6. Let X be an irreducible algebraic variety, Q a regular func-
tion on X, F a coherent algebraic sheaf on X and s a section of F over X
51
whose restriction to X
Q
is zero. Then for n suciently large the section Q
n
s
is zero on the whole of X.
The question being again local, we can assume:
(a) that X is a closed subvariety of K
r
,
(b) that F is isomorphic to a cokernel of a homomorphism : O
p
X
O
q
X
,
(c) that s is the image of a section of O
q
X
.
(Indeed, all the above conditions are satised locally).
Set A = (X, O
X
) = K[X
1
, . . . , X
r
]/I(X). The section can be identied
with a system of q elements of A. Let on the other hand
t
1
= (1, 0, . . . , 0), . . . , t
p
= (0, . . . , 0, 1);
the t
i
, 1 i p are sections of O
q
X
over X, thus can be identied with systems
of q elements of A. The assumption made on s means that for all x X
Q
we
have (x) (O
p
x,X
), that is, can be written in the form =
i=p
i=1
f
i
t
i
with
f
i
O
x,X
; or, by clearing denominators, that there exist Q
x
A, Q
x
(x) ,= 0 for
which Q
x
=
i=p
i=1
R
i
t
i
with R
i
A. The reasoning used above shows then
that, for n suciently large, Q
n
belongs to the ideal generated by Q
x
, hence
Q
n
(x) (O
p
x,X
) for all x X, which means that Q
n
s is zero on the whole of
X.
44 Vanishing of certain cohomology groups
Proposition 7. Let X be an irreducible algebraic variety, Q
i
a nite family
of regular functions on X that do not vanish simultaneously and U the open
covering of X consisting of X
Qi
= U
i
. If F is a coherent algebraic subsheaf of
O
p
X
, we have H
q
(U, F) = 0 for all q > 0.
Possibly replacing U by an equivalent covering, we can assume that none of
the functions Q
i
vanishes identically, in other words that we have U
i
,= for all
i.
Let f = (f
i0...iq
) be a q-cocycle of U with values in F. Each f
i0...iq
is
a section of F over U
i0...iq
, thus can be identied with a system of p regular
functions on U
i0...iq
; applying Proposition 5 to Q = Q
i0
. . . Q
iq
we see that, for n
suciently large, g
i0...iq
= (Q
i0
. . . Q
iq
)
n
f
i0...iq
is a system of p regular functions
on X. Choose an integer n for which this holds for all systems i
0
, . . . , i
q
, which
is possible because there is a nite number of such systems. Consider the image
of g
i0...iq
in the coherent sheaf O
p
X
/F; this is a section vanishing on U
i0...iq
;
then applying Proposition 6 we see that for m suciently large, the product of
this section with (Q
i0
. . . Q
iq
)
m
is zero on the whole of X. Setting N = m+n,
we see that we have constructed sections h
i0...iq
of F over X which coincide
with (Q
i0
. . . Q
iq
)
N
f
i0...iq
on U
i0...iq
.
As the Q
N
i
do not vanish simultaneously, there exist functions
R
i
(X, O
X
)
52
such that
R
i
Q
N
i
= 1. Then for any system i
0
, . . . , i
q1
set
k
i0...iq1
=
i
R
i
h
ii0...iq1
/(Q
i0
. . . Q
iq1
)
N
,
which makes sense because Q
i0
. . . Q
iq1
is nonzero on U
i0...iq1
.
We have thus dened a cochain k C
q1
(U, F). I claim that f = dk, which
will show the Proposition.
We must check that (dk)
i0...iq
= f
i0...iq
; it suces to show that these two
sections coincide on U =
U
i
, since they will coincide everywhere, because they
are systems of p rational functions on X and U ,= 0. Now over U, we can write
k
i0...iq1
=
i
R
i
Q
N
i
f
ii0...iq
,
hence
(dk)
i0...iq
=
j=q
j=0
(1)
q
i
R
i
Q
N
i
f
ii0...
ij...iq
and taking into account that f is a cocycle,
(dk)
i0...iq
=
i
R
i
Q
N
i
f
i0...iq
= f
i0...iq
, q.e.d.
Corollary 1. H
q
(X, F) = 0 for q > 0.
Indeed, Proposition 3 shows that coverings of the type used in Proposition
7 are arbitrarily ne.
Corollary 2. The homomorphism (X, O
p
X
) (X, O
p
X
/F) is surjective.
This follows from Corollary 1 above and from Corollary 2 to Proposition 6
from n
24.
Corollary 3. Let V be a closed subvariety of K
r
and let
A = K[X
1
, . . . , X
r
]/I(V ).
Then the homomorphism : A (V, O
V
) is bijective.
We apply Corollary 2 above to X = K
r
, p = 1, F = J(V ), the sheaf of
ideals dened by V ; we obtain that every element of (V, O
V
) is the restriction
of a section of O on X, that is, a polynomial, by Proposition 4 applied to X.
53
45 Sections of a coherent algebraic sheaf on an ane va-
riety
Theorem 2. Let F be a coherent algebraic sheaf on an ane variety X. For
every x X, the O
x,X
module F
x
is generated by elements of (X, F).
Since X is ane, it can be embedded as a closed subvariety of an ane space
K
r
; by extending the sheaf F by 0 outside X, we obtain a coherent algebraic
sheaf on K
r
(cf. n
39) and we are led to prove the theorem for the new sheaf.
In other words, we can suppose that X = K
r
.
By the denition of a coherent sheaf, there exists a covering of X consisting
of open subsets on which F is isomorphic with a quotient of the sheaf O
p
.
Applying Proposition 3, we see that there exists a nite number of polynomials
Q
i
that do not vanish simultaneously and such that on every U
i
= X
Qi
there
exists a surjective homomorphism
i
: O
pi
F; we can furthermore assume
that none of the polynomials is identically zero.
The point x belongs to one U
i
, say U
0
; it is clear that F
x
is generated by
sections of F over U
0
; as Q
0
is invertible in O
x
, it suces to prove the following
lemma:
Lemma 1. If s
0
is a section of F over U
0
, there exists an integer N and a
section s of F over X such that s = Q
N
0
s
0
over U
0
.
By Proposition 2, U
i
U
0
is an ane variety, obviously irreducible; by ap-
plying Corollary 2 of Proposition 7 to this variety and to
i
: O
pi
F, we
see that there exists a section
0i
of O
pi
on U
i
U
0
such that
i
(
0i
) = s
0
on
U
i
U
0
; as U
i
U
0
is the set of points of U
i
in which Q
0
does not vanish, we
can apply Proposition 5 to X = U
i
, Q = Q
0
and we see that there exists, for n
suciently large, a section
i
of O
pi
over U
i
which coincides with Q
n
0

0i
over
U
i
U
0
; by setting s
i
=
i
(
i
), we obtain a section of F over U
i
that coincides
with Q
n
0
s
0
over U
i
U
0
. The sections s
i
and s
j
coincide on U
i
U
j
U
0
;
applying Proposition 6 to s
i
s
j
, we see that for m suciently large we have
Q
m
0
(s
i
s
j
) = 0 on the whole of U
i
U
j
. The Q
m
0
s
i
then dene a unique
section s of F over X, and we have s = Q
n+m
0
s
0
on U
0
, which shows the lemma
and completes the proof of Theorem 2.
Corollary 1. The sheaf F is isomorphic to a quotient sheaf of the sheaf
O
p
X
.
Because F
x
is an O
x,X
module of nite type, it follows from the above
theorem that there exists a nite number of sections of F generating F
x
; by
Proposition 1 of n
12, these sections generate F

y
for y suciently close to x.
The space X being quasi-compact, we conclude that there exists a nite number
of sections s
1
, . . . , s
p
of F generating F
x
for all x X, which means that F is
isomorphic to a quotient sheaf of the sheaf O
p
X
.
54
Corollary 2. Let A

B

C be an exact sequence of coherent algebraic
sheaf on an ane variety X. The sequence (X, A)

(X, B)

(X, C) is
also exact.
We can suppose, as in the proof of Theorem 2, that X is an ane space
K
r
, thus is irreducible. Set J = () = Ker(); everything reduces to seeing
that : (X, A) (X, J) is surjective. Now, by Corollary 1, we can nd
a surjective homomorphism : O
p
X
A and, by Corollary 2 to Proposition
7, : (X, O
p
X
) (X, J) is surjective; this is a fortiori the same for
: (X, A) (X, J), q.e.d.
46 Cohomology groups of an ane variety with values in
a coherent algebraic sheaf
Theorem 3. Let X be an ane variety, Q
i
a nite family of regular functions
on X that do not vanish simultaneously and let U be the open covering of X
consisting of X
Qi
= U
i
. If F is a coherent algebraic sheaf on X, we have
H
q
(U, F) = 0 for all q > 0.
Assume rst that X is irreducible. By Corollary 1 to Theorem 2, we can
nd an exact sequence
0 R O
p
X
F 0.
The sequence of complexes: 0 C(U, R) C(U, O
p
X
) C(U, F) 0 is exact;
indeed, this reduces to saying that every section of F over U
i0...iq
is the image
of a section of O
p
X
over U
i0...iq
, which follows from Corollary 2 to Proposition 7
applied to the irreducible variety U
i0...iq
. This exact sequence gives birth to an
exact sequence of cohomology:
. . . H
q
(U, O
p
X
) H
q
(U, F) H
q+1
(U, R) . . . ,
and as H
q
(U, O
p
X
) = H
q+1
(U, R) = 0 for q > 0 by Proposition 7, we conclude
that H
q
(U, F) = 0.
We proceed now to the general case. We can embed X as a closed subvariety
of an ane space K
r
; by Corollary 3 to Proposition 7, the functions Q
i
are
induced by polynomials P
i
; let on the other hand R
j
be a nite system of
generators of the ideal I(X). The functions P
i
, R
j
do not vanish simultaneously
on K
r
, thus dene an open covering U
of K
r
; let F
be the sheaf obtained by

extending F by 0 outside X; applying what we have proven to the space K
r
,
the functions P
i
, R
j
and the sheaf F
, we see that H
q
(U
, F
) = 0 for q > 0.
As we can immediately verify that the complex C(U
, F
) is isomorphic to the
complex C(U, F), it follows that H
q
(U, F) = 0, q.e.d.
Corollary 1. If X is an ane variety and F a coherent algebraic sheaf on
X, we have H
q
(X, F) = 0 for all q > 0.
55
Indeed, the coverings used in the above theorem are arbitrarily ne.
Corollary 2. Let 0 A B C 0 be an exact sequence of sheaves
on an ane variety X. If the sheaf A is coherent algebraic, the homomorphism
(X, V) (X, C) is surjective.
This follows from Corollary 1, by setting q = 1.
47 Coverings of algebraic varieties by open ane subsets
Proposition 8. Let X be an ane variety and let U = U
i
iI
be a nite
covering of X by open ane subsets. If F is a coherent algebraic sheaf on X,
we have H
q
(U, F) = 0 for all q > 0.
By Proposition 3, there exist regular functions P
j
on X such that the cov-
ering V = X
Pj
is ner than U. For every (i
0
, . . . , i
p
), the covering V
i0,...,ip
induced by V on U
i0...ip
is dened by restrictions of P
j
to U
i0...ip
; as U
i0...ip
is
an ane variety by Proposition 1, we can apply Theorem 3 to it and conclude
that H
q
(V
i0...ip
, F) = 0 for all q > 0. Applying then Proposition 5 of n
29,
we see that
H
q
(U, F) = H
q
(V, F),
and, as H
q
(V, F) = 0 for q > 0 by Theorem 3, the Proposition is proven.
Theorem 4. Let X be an algebraic variety, F a coherent algebraic sheaf
on X and U = U
i
iI
a nite covering of X by open ane subsets. The
homomorphism (U) : H
n
(U, F) H
n
(X, F) is bijective for all n 0.
Consider the family V
of all nite coverings of X by open ane subsets.

By the corollary of Theorem 1, these coverings are arbitrarily ne. On the other
hand, for every system (i
0
, . . . , i
p
) the covering V
i0...ip
induced by V
on U
i0...ip
is a covering by open ane subsets, by Proposition 1; by Proposition 8, we thus
have H
q
(V
i0...ip
, F) = 0 for q > 0. The conditions (a) and (b) of Theorem 1,
n
29 are satised and the theorem follows.

Theorem 5. Let X be an algebraic variety and U = U
i
iI
a nite cov-
ering of X by open ane subsets. Let 0 A B C 0 be an exact
sequence of sheaves on X, the sheaf A being coherent algebraic. The canonical
homomorphism H
q
0
(U, C) H
q
(U, C) (cf. n
24) is bijective for all q 0.

It obviously suces to prove that C
0
(U, C) = C(U, C), that is, that every
section of C over U
i0...iq
is the image of a section of B over U
i0...iq
, which follows
from Corollary 2 of Theorem 3.
Corollary 1. Let X be an algebraic variety and let 0 A B C 0
be an exact sequence of sheaves on X, the sheaf A being coherent algebraic. The
canonical homomorphism H
q
0
(X, C) H
q
(X, C) is bijective for all q 0.
This is an immediate consequence of Theorems 1 and 5.
Corollary 2. We have an exact sequence:
. . . H
q
(X, B) H
q
(X, C) H
q+1
(X, A) H
q+1
(X, B) . . .
56
4. Correspondence between modules of nite type and coherent algebraic sheaves II
4 Correspondence between modules of
finite type and coherent algebraic
sheaves
48 Sheaf associated to a module
Let V be an ane variety, O the sheaf of local rings of V ; the ring A = (V, O),
which will be called the ring of coordinates of V , is an algebra over K which
has no nilpotent elements but 0. If V is embedded as a closed subvariety of
an ane space K
r
, we know (cf. n
44) that A is identied with the quotient

algebra of K[X
1
, . . . , X
r
] by the ideal of polynomials vanishing on V ; it follows
that the algebra A is generated by a nite number of elements.
Conversely, we verify easily that if A is a commutative Kalgebra without
nilpotent elements (other that 0) and is generated by a nite number of elements,
there exists an ane variety V such that A is isomorphic to (V, O); moreover,
V is determined up to isomorphism by this property (we can identify V with
the set of characters of A equipped with the usual topology).
Let M be an Amodule; M denes a constant sheaf on V which we denote
again by M; the same way A denes a constant sheaf, and the sheaf M can
be considered as a sheaf of Amodules. Dene A(M) = O
A
M, the sheaf
O being also considered as a sheaf of Amodules; it is clear that A(M) is an
algebraic sheaf on V . Moreover, if : M M
is an Ahomomorphism, we
have a homomorphism A() = 1 : A(M) A(M
); in other words, A(M)

is a covariant functor of the module M.
Proposition 1. The functor A(M) is exact.
Let M M
be an exact sequence of Amodules. We must observe

that the sequence A(M) A(M
) A(M
) is exact, in other words, that

for all x V the sequence:
O
x

A
M O
x

A
M
O
x

A
M
is exact.
Now O
x
is nothing else that the localization A
S
of A, S being the set of
those f A for which f(x) ,= 0 (for the denition of localization, cf. [8], [12] or
[13]). Proposition 1 is thus a particular case of the following result:
Lemma 1. Let A be a ring, S a multiplicative system in A not containing
0, A
S
the localization of A in S. If M M
is an exact sequence of
Amodules, the sequence A
S

A
M A
S

A
M
A
S

A
M
is exact.
Denote by M
S
the set of fractions m/s with m M, s S, two fractions m/s
and m
/s
being identied if there exists an s
S such that s
(s
msm
) = 0;
it is easily seen that M
S
is an A
S
module and that the mapping
a/s m a m/s
57
is an isomorphism from A
S

A
A onto M
S
; we are thus led to prove that the
sequence
M
S
M
S
M
S
is exact, which is obvious.
Proposition 2. A(M) = 0 implies M = 0.
Let m be an element of M; if A(M) = 0, we have 1 m = 0 in O
x

A
M
for all x V . By the discussion above, 1 m = 0 is equivalent to existence of
an element s A, s(x) ,= 0 such that s m = 0; the annihilator of m in M is
not contained in any maximal ideal of A, which implies that it is equal to A, so
m = 0.
Proposition 3. If M is an Amodule of nite type, A(M) is a coherent
algebraic sheaf on V .
Because M is of nite type and since A is Noetherian, M is isomorphic to
the cokernel of a homomorphism : A
q
A
p
and A(M) is isomorphic to the
cokernel of A() : A(A
q
) A(A
p
). As A(A
p
) = O
p
and A(A
q
) = O
q
, it
follows that A(M) is coherent.
49 Module associated to an algebraic sheaf
Let F be an algebraic sheaf on V and let (F) = (V, F); since F is a sheaf
of Omodules, (F) is equipped with a natural structure of an Amodule.
Any algebraic homomorphism : F G denes an Ahomomorphism () :
(F) (G). If we have an exact sequence of algebraic sheaves F G K ,
the sequence
(F) (G) (K )
is exact (n
45); applying this to an exact sequence O

p
F 0 we see that
(F) is an Amodule of nite type if F is coherent.
The functors A(M) and (F) are ,,inverse to each other:
Theorem 1. (a) If M is an A-module of nite type, (A(M)) is canoni-
cally isomorphic to M.
(b) If F is a coherent algebraic sheaf on V , A((F)) is canonically iso-
morphic to F.
First let us show (a). Every element m M denes a section (m) of
A(M) by the formula: (m)(x) = 1 m O
x

A
M; hence a homomorphism
: M (A(M)). When M is a free module of nite type, is bijective (it
suces to see this when M = A, in which case it is obvious); if M is an arbitrary
module of nite type, there exists an exact sequence L
1
L
0
M 0 where
L
0
and L
1
are free of nite type; the sequence A(L
1
) A(L
0
) A(M) 0
58
is exact, thus also the sequence (A(L
1
)) (A(L
0
)) (A(M)) 0. The
commutative diagram:
L
1
L
0
M 0
(A(L
1
)) (A(L
0
)) (A(M)) 0

shows then that : M (A(M)) is bijective, which shows (a).
Let now F be an algebraic coherent sheaf on V . If we associate to every
s (F) an element s(x) F(X), we obtain an Ahomomorphism: (F)
F
x
which extends to an O
x
homomorphism
x
: O
x
A
(F) F
x
; we easily
verify that the
x
form a homomorphism of sheaves : A((F)) F. When
F = O
p
, the homomorphism is bijective; it follows by the same reasoning as
above that is bijective for every coherent algebraic sheaf F, which shows (b).
Remarks. (1) We could also deduce (b) from (a); cf. n
65, proof of
Proposition 6.
(2) We will see in Chapter III how the above correspondence should be
modied when one studies coherent sheaves on the projective space.
50 Projective modules and vector bundles
Recall ([6], Chap. I, th. 2.2) that an Amodule is called projective if it is a
direct summand of a free Amodule.
Proposition 4. Let M be an Amodule of nite type. Then M is projective
if and only if the O
x
module O
x

A
M is free for every x V .
If M is projective, O
x

A
M is O
x
projective, thus O
x
free since O
x
is a
local ring (cf. [6], Chap. VIII, th. 6.1).
Conversely, if all O
x

A
M are free, we have
dim(M) = Sup dim
xV
(O
x

A
M) = 0 (cf.[6], Chap.V II, Exer.11),
from which it follows that M is projective ([6], Chap. VI, 2).
Note that if F is a coherent algebraic sheaf on V and if F
x
is isomorphic to
O
p
x
, F is isomorphic to O
p
in a neighborhood of x; if this property is satised
in every x V , the sheaf F is thus locally isomorphic to the sheaf O
p
, the
integer p being constant on every connected component of V . Applying this to
the sheaf A(M), we obtain:
Corollary. Let F be a coherent algebraic sheaf on a connected ane variety
V . The three following properties are equivalent:
(i) (F) is a projective Amodule,
59
(ii) F is locally isomorphic to O
p
,
(iii) F is isomorphic to the sheaf of germs of sections of a vector bundle
with base V .
In other words, the mapping E (S(E) (E denoting a vector bundle)
gives a bijective correspondence between classes of vector bundles and classes
of projective Amodules of nite type; in this correspondence, a trivial bundle
corresponds to a free module and conversely.
Note that when V = K
r
(in which case A = K[X
1
, . . . , X
r
]), we do not
know if there exist projective Amodules that are not free, or equivalently, if
there exist algebraic vector bundles with base K
r
that are not trivial.
60
Chapter III
Coherent Algebraic Sheaves
on Projective Varieties
61
1. Projective varieties III
1 Projective varieties
51 Notations
(The notations introduced below will be used without reference during the whole
chapter).
Let r be an integer 0 and let Y = K
r+1
0; the multiplicative group
K
of nonzero elements of K acts on Y by the formula

(
0
, . . . ,
r
) = (
0
, . . . ,
r
).
Two points y and y
will be called equivalent if there exists K
such that
y
= y; the quotient space of Y by this equivalence relation will be denoted

by P
r
(K) or simply X; it is the projective space of dimension r over K; the
canonical projection of Y onto X will be denoted .
Let I = 0, 1, . . . , r; for every i I, we denote by t
i
the i-th coordinate
function on K
r+1
, dened by the formula:
t
i
(
0
, . . . ,
r
) =
i
.
We denote by V
i
the open subset of K
r+1
consisting of points whose t
i
is
,= 0 and by U
i
the image of V
i
by ; the U
i
form a covering U of X. If i I
and j I, the function t
j
/t
i
is regular on V
i
and invariant for K
, thus denes
a function on U
i
which we denote also by t
j
/t
i
; for xed i, the functions t
j
/t
i
,
j ,= i dene a bijection
i
: U
i
K
r
.
We equip K
r+1
with the structure of an algebraic variety and Y the induced
structure. Likewise, we equip X with the quotient topology from Y : a closed
subset of X is thus the image by of a closed cone in K
r+1
. If U is open
in X, we dene A
U
= (
1
(U), O
Y
); this is the sheaf of regular functions
on
1
(U). Let A
0
U
be the subring of A
U
consisting of elements invariant for
K
(that is, homogeneous functions of degree 0). When V U, we have a

restriction homomorphism
V
U
: A
0
V
A
0
U
and the system (A
0
U
,
V
U
) denes a
sheaf O
X
which can be considered as a subsheaf of the sheaf F(X) of germs of
functions on X. Such a function f, dened in a neighborhood of x belongs to
O
x,X
if and only if it coincides locally with a function of the form P/Q where
P and Q are homogeneous polynomials of the same degree in t
0
, . . . , t
r
with
Q(y) ,= 0 for y
1
(x) (which we write for brevity as Q(x) ,= 0).
Proposition 1. The projective space X = P
r
(K), equipped with the topology
and sheaf above, is an algebraic variety.
The U
i
, i I are open in X and we verify immediately that the bijections
i
: U
i
K
r
dened above are biregular isomorphisms, which shows that the
axiom (V A
I
) is satised. To show that (V A
II
) is also satised, we must observe
62
that the subset of K
r
K
r
consisting of all pairs (
i
(x),
j
(x)) where x U
i
U
j
is closed, which does not pose diculties.
In what follows, X will be always equipped with the structure of an algebraic
variety just dened; the sheaf O
X
will be simply denoted O. An algebraic variety
V is called projective if it is isomorphic to a closed subvariety of a projective
space. The study of coherent algebraic sheaves on projective varieties can be
reduced to the study of coherent algebraic sheaves on P
r
(K), cf. n
39.
52 Cohomology of subvarieties of the projective space
Let us apply Theorem 4 from n
47 to the covering U = U
i
iI
dened in the
preceding n
: it is possible since each U

i
is isomorphic to K
r
. We thus obtain:
Proposition 2. If F is a coherent algebraic sheaf on X = P
r
(K), the
homomorphism (U) : H
n
(U, F) H
n
(X, F) is bijective for all n ,= 0.
Since U consists of r + 1 open subsets, we have (cf. n
20, corollary to
Proposition 2):
Corollary. H
n
(X, F) = 0 for n > r.
This result can be generalized in the following way:
Proposition 3. Let V be an algebraic variety, isomorphic to a locally closed
subvariety of the projective space X. Let F be an algebraic coherent sheaf on
V and let W be the subvariety of V such that F is zero outside W. We then
have H
n
(V, F) = 0 for n > dimW.
In particular, taking W = V , we see that we have:
Corollary. H
n
(V, F) = 0 for n > dimV .
Identify V with a locally closed subvariety of X = P
r
(K); there exists an
open subset U of X such that V is closed in U. We can clearly assume that
W is closed in V , so that W is closed in U. Let F = X U. Before proving
Proposition 3, we establish two lemmas:
Lemma 1. Let k = dimW; there exists k + 1 homogeneous polynomials
P
i
(t
0
, . . . , t
r
) of degrees > 0, vanishing on F and not vanishing simultaneously
on W.
(By abuse of language, we say that a homogeneous polynomial P vanishes
in a point x of P
r
(K) if it vanishes on
1
(x)).
We proceed by induction on k, the case when k = 1 being trivial. Choose
a point on each irreducible component of W and let P
1
be a homogeneous poly-
nomial vanishing on F, of degree > 0 and nonvanishing in each of these points
(the existence of P
1
follows from the fact that F is closed, given the denition of
the topology of P
r
(K)). Let W
be a subvariety of W consisting of points x W

such that P
1
(x) = 0; by the construction of P
1
, no irreducible component of W
is contained in W
and it follows (cf. n
36) that dimW
< k. Applying the

63
induction assumption to W
, we see that there exist k homogeneous polynomials

P
2
, . . . , P
k+1
vanishing on F and nonvanishing simultaneously on W
; it is clear
that the polynomials P
1
, . . . , P
k+1
satisfy appropriate conditions.
Lemma 2. Let P(t
0
, . . . , t
r
) be a homogeneous polynomial of degree n > 0.
The set X
P
of all points x X such that P(x) ,= 0 is an open ane subset of
X.
If we assign to every point y = (
0
, . . . ,
r
) Y the point of the space K
N
having for coordinates all monomials
m0
0
. . .
mr
r
, m
0
+. . . +m
r
= n, we obtain,
by passing to quotient, a mapping
n
: X P
N1
(K). It is classical, and also
easy to verify, that
n
is a biregular isomorphism of X onto a closed subvariety
of P
N1
(K) (,,Veronese variety); now
n
transforms the open subset X
P
onto
the locus of points of
n
(X) not lying on a certain hyperplane of P
N1
(X); as
the complement of any hyperplane is isomorphic to an ane space, we conclude
that X
P
is isomorphic to a closed subvariety of an ane space.
We shall now prove Proposition 3. Extend the sheaf F by 0 on U V ;
we obtain a coherent algebraic sheaf on U which we also denote by F, and
we know (cf. n
26) that H
n
(U, F) = H
n
(V, F). Let on the other hand
P
1
, . . . , P
k+1
be homogeneous polynomials satisfying the conditions of Lemma
1; let P
k+2
, . . . , P
h
be homogeneous polynomials of degrees > 0, vanishing on
W F and not vanishing simultaneously in any point of U W (to obtain such
polynomials, it suces to take a system of homogeneous coordinates of the ideal
dened by W F in K[t
0
, . . . , t
r
]). For every i, 1 i h, let V
i
be the set of
points x X such that P
i
(x) ,= 0; we have V
i
U and the assumptions made
above show that V = V
i
is an open covering of U; moreover, Lemma 2 shows
that V
i
are open ane subsets, so H
n
(V, F) = H
n
(U, F) = H
n
(V, F) for all
n 0. On the other hand, if n > k and if the indices i
0
, . . . , i
n
are distinct,
one of the indices is > k + 1 and V
i0...in
does not meet W; we conclude that
the group of alternating cochains C
n
(V, F) is zero if n > k, which shows that
H
n
(V, F) = 0, by Proposition 2 of n
20.
53 Cohomology of irreducible algebraic curves
If V is an irreducible algebraic variety of dimension 1, the closed subsets of V
distinct from V are nite subsets. If F is a nite subset of V and x a point of
F, we set V
p
x
= (V F) x; the V
F
x
, x F form a nite open covering V
F
of V .
Lemma 3. The coverings V
F
of the above type are arbitrarily ne.
Let U = U
i
iI
be an open covering of V , which we may assume to be nite
since V is quasi-compact. We can likewise assume that U
i
,= for all i I.
If we set F
i
= V U
i
, F
i
is also nite, and so is F =
iI
F
i
. We will show
that V
F
U, which proves the lemma. Let x F; there exists an i I such
that x / F
i
, since the U
i
cover V ; we have then F x F
i
, because F F
i
,
which means that V
F
x
U
i
and shows that V
F
U.
64
Lemma 4. Let F be a sheaf on V and F a nite subset of V . We have
H
n
(V
F
, F) = 0
for n 2.
Set W = V F; it is clear that V
F
x0
. . . V
F
xn
= W if x
0
, . . . , x
n
are distinct
and if n 1. If we put G = (W, F), it follows that the alternating complex
C
(V
F
, F) is isomorphic, in dimensions 1, to C
(S(F), G), S(F) denoting

the simplex with F for the set of vertices. It follows that
H
n
(V
F
, F) = H
n
(S(F), G) = 0 for n 2,
the cohomology of a simplex being trivial.
Lemmas 3 and 4 obviously imply:
Proposition 4. If V is an irreducible algebraic curve and F is an arbitrary
sheaf in V , we have H
n
(V, F) = 0 for n 2.
Remark. I do not know whether an analogous result is true for varieties of
arbitrary dimension.
65
2. Graded modules and coherent algebraic sheaves on the projective space III
2 Graded modules and coherent algebraic
sheaves on the projective space
54 The operation F(n)
Let F be an algebraic sheaf on X = P
r
(K). Let F
i
= F(U
i
) be the restriction
of F to U
i
(cf. n
51); if n is an arbitrary integer, let

ij
(n) be the isomorphism
of F
j
(U
i
U
j
) with F
i
(U
i
U
j
) dened by multiplication by the function t
n
j
/t
n
i
;
this makes sense, since t
j
/t
i
is a regular function on U
i
U
j
with values in K
.
We have
ij
(n)
jk
(n) =
ik
(n) at every point of U
i
U
j
U
k
; we can thus
apply Proposition 4 of n
4 and obtain an algebraic sheaf denoted by F(n),

dened by gluing the sheaves F
i
= F(U
i
) using the isomorphisms
ij
(n).
We have the canonical isomorphisms: F(0) F, F(n)(m) F(n + m).
Moreover, F(n) is locally isomorphic to F, thus coherent if F is; it also follows
that every exact sequence F F
of algebraic sheaves gives birth to

exact sequences F(n) F
(n) F
(n) for all n Z.

We can apply the above procedure to the sheaf F = O and so obtain the
sheaves O(n), n Z. We will give another description of these sheaves: if U is
open in X, let A
n
U
be the subset of A
U
= (
1
(U), O
Y
) consisting of regular
functions of degree n (that is, satisfying the identity f(y) =
n
f(y) for K
and y
1
(U)); the A
n
U
are A
0
U
modules, thus give birth to an algebraic sheaf,
which we denote by O
(n). An element of O
(n)
x
, x X can be this identied
with a rational function P/Q, P and Q being homogeneous polynomials such
that Q(x) ,= 0 and deg P deg Q = n.
Proposition 1. The sheaves O(n) and O
(n) are canonically isomorphic.

By denition, a section of O(n) over an open U X is a system (f
i
) of
sections of O over U U
i
with f
i
= (t
n
j
/t
n
i
) f
j
on U U
i
U
j
; the f
j
can
be identied with regular functions, homogeneous of degree 0 over
1
(U)
1
(U
i
); set g
i
= t
n
i
f
i
; we then have g
i
= g
j
at every point of
1
(U)
1
(U
i
)
1
(U
j
), thus the g
i
are the restrictions of a unique regular function
on
1
(U), homogeneous of degree n. Conversely, such a function g denes a
system (f
i
) by setting f
i
= g/t
n
i
. The mapping (f
i
) g is thus an isomorphism
of O(n) with O
(n).
Henceforth, we will often identify O(n) with O
(n) by means of the above

isomorphism. We observe that a section of O
(n) over X is just a regular

function on Y , homogeneous of degree n. If we assume that r 1, such a
function is identically zero for n < 0 and it is a homogeneous polynomial of
degree n for n 0.
Proposition 2. For every algebraic sheaf F, the sheaves F(n) and F
O
O(n) are canonically isomorphic.
66
Since O(n) is obtained from the O
i
by gluing with respect to
ij
(n), FO(n)
is obtained from F
i
O
i
by gluing with respect to the isomorphisms 1
ij
(n);
identifying F
i
O
i
with F
i
we recover the denition of F(n).
Henceforth, we will also identify F(n) with F O(n).
55 Sections of F(n)
Let us rst show a lemma on algebraic varieties, that is quite analogous to
Lemma 1 of n
45:
Lemma 1. Let V be an ane variety, Q a regular function on V and V
Q
the set of all points x V such that Q(x) ,= 0. Let F be a coherent algebraic
sheaf on V and let s be a section of F over V
Q
. Then, for n suciently large,
there exists a section s
of F over the whole V such that s
= Q
n
s over V
Q
.
Embedding V in an ane space and extending F by 0 outside V , we are
brought to the case where V is an ane space, thus is irreducible. By Corollary 1
to Theorem 2 from n
45, there exists a surjective homomorphism : O

p
V
F;
by Proposition 2 of n
42, V
Q
is an open ane subset and thus there exists (n
44,
Corollary 2 to Proposition 7) a section of O
p
V
over V
Q
such that () = s. We
can identify with a system of p regular functions on V
Q
; applying Proposition
5 of n
43 to each of these functions, we see that there exists a section
of O
p
V
over V such that
= Q
n
on V
Q
, provided that n is suciently large. Setting
s
= (
), we obtain a section of F over V such that s
= Q
n
s on V
Q
.
Theorem 1. Let F be a coherent algebraic sheaf on X = P
r
(K). There
exists an integer n(F) such that for all n n(F) and all x X, the O
x
module
F(n)
x
is generated by elements of (X, F(n)).
By the denition of F(n), a section s of F(n) over X is a system (s
i
) of
sections of F over U
i
satisfying the compatibility conditions:
s
i
= (t
n
j
/t
n
i
) s
j
on U
i
U
j
;
we say that s
i
is the i-th component of s.
On the other hand, since U
i
is isomorphic to K
r
, there exists a nite number
of sections s
i
of F over U
i
which generate F
x
for all x U
i
(n
45, Corollary 1
to Theorem 2); if for a certain integer n we can nd sections s
of F(n) whose
i-th component is s
i
, it is clear that (X, F(n)) generates F(n)
x
for all x U
i
.
Theorem 1 is thus proven if we prove the following Lemma:
Lemma 2. Let s
i
be a section of F over U
i
. For all n suciently large,
there exists a section s of F(n) whose i-th component is equal to s
i
.
Apply Lemma 1 to the ane variety V = U
j
, the function Q = t
i
/t
j
and the
section s
i
restricted to U
i
U
j
; this is legal, because t
i
/t
j
is a regular function
on U
j
whose zero set is equal to U
j
U
i
U
j
. We conclude that there exists an
integer p and a section s
j
of F over U
j
such that s
j
= (t
p
i
/t
p
j
) s
i
on U
i
U
j
;
67
for j = i, we have s
i
= s
i
, which allows us to write the preceding formula in the
form s
j
= (t
p
i
/t
p
j
) s
i
.
The s
j
being dened for every index j (with the same exponent p), consider
s
j
(t
p
k
/t
p
j
)s
k
; it is a section of F over U
j
U
k
whose restriction to U
i
U
j
U
k
is
zero; by applying Proposition 6 of n
43 we see that for every suciently large

integer q we have (t
q
i
/t
q
j
)(s
j
(t
p
k
/t
p
j
) s
k
) = 0 on U
j
U
k
; if we then put
s
j
= (t
q
i
/t
q
j
) s
j
and n = p + q, the above formula is written s
j
= (t
n
k
/t
n
j
) s
k
and the system s = (s
j
) is a section of F(n) whose i-th component is equal to
s
i
, q.e.d.
Corollary. Every coherent algebraic sheaf F on X = P
r
(K) is isomorphic
to a quotient sheaf of a sheaf O(n)
p
, n and p being suitable integers.
By the above theorem, there exists an integer n such that F(n)
x
is gen-
erated by (X, F(n)) for every x X; by the quasi-compactness of X, this
is equivalent to saying that F(n) is isomorphic to a quotient sheaf of a sheaf
O
p
, p being an appropriate integer 0. It follows then that F F(n)(n) is
isomorphic to a quotient sheaf of O(n)
p
O
p
(n).
56 Graded modules
Let S = K[t
0
, . . . , t
r
] be the algebra of polynomials in t
0
, . . . , t
r
; for every integer
n 0, let S
n
be the linear subspace of S consisting by homogeneous polynomials
of degree n; for n < 0, we set S
n
= 0. The algebra S is a direct sum of S
n
,
n Z and we have S
p
S
q
S
p+q
; in other words, S is a graded algebra.
Recall that an Smodule M is said to be graded if there is given a decompo-
sition of M into a direct sum: M =
nZ
M
n
, M
n
being subgroups of M such
that S
p
M
q
M
p+q
for every couple (p, q) of integers. An element of M
n
is said
to be homogeneous of degree n; a submodule N of M is said to be homogeneous
if it is a direct sum of N M
n
, in which case it is a graded Smodule. If M
and M
are two graded Smodules, an Shomomorphism

: M M
is said to be homogeneous of degree s if (M

n
) M
n+s
for every n Z. A
homogeneous Shomomorphism of degree 0 is simply called a homomorphism.
If M is a graded Smodule and n an integer, we denote by M(n) the graded
Smodule:
M(n) =
pZ
M(n)
p
with M(n)
p
= M
n+p
.
We thus have M(n) = M as Smodules, but a homogeneous element of degree
p of M(n) is homogeneous of degree n +p in M; in other words, M(n) is made
from M by lowering degrees by n units.
We denote by C the class of graded Smodules M such that M
n
= 0 for n
suciently large. If A B C is an exact sequence of homomorphisms of
68
graded Smodules, the relations A C, C C clearly imply B C; in other
words, C is a class in the sense of [14], Chap. I. Generally, we use the terminol-
ogy introduced in the aforementioned article; in particular, a homomorphism
: A B is called C-injective (resp. C-surjective) if Ker() C (resp. if
Coker() C) and C-bijective if it is both C-injective and C-surjective.
A graded Smodule M is said to be of nite type if it is generated by a nite
number of elements; we say that M satises the condition (TF) if there exists
an integer p such that the submodule
np
M
n
of M is of nite type; it is the
same to say that M is C-isomorphic to a module of nite type. The modules
satisfying (TF) form a class containing C.
A graded Smodule L is called free (resp. free of nite type) if it admits a
base (resp. a nite base) consisting of homogeneous elements, in other words
if it is isomorphic to a direct sum (resp. to a nite direct sum) of the modules
S(n
i
).
57 The algebraic sheaf associated to a graded S-module
If U is a nonempty subset of X, we denote by S(U) the subset of S = K[t
0
, . . . , t
r
]
consisting of homogeneous polynomials Qsuch that Q(x) ,= 0 for all x U; S(U)
is a multiplicatively closed subset of S, not containing 0. For U = X, we write
S(x) instead of S(x).
Let M be a graded Smodule. We denote by M
U
the set of fractions m/Q
with m M, Q S(U), m and Q being homogeneous of the same degree; we
identify two fractions m/Q and m
/Q
if there exists Q
S(U) such that

Q
(Q
mQ m
) = 0;
it is clear that we have dened an equivalence relation between the pairs (m, Q).
For U = x, we write M
x
instead of M
{x}
.
Applying this to M = S, we see that S
U
is the ring of rational functions P/Q,
P and Q being homogeneous polynomials of the same degree and Q S(U); if
M is an arbitrary graded Smodule, we can equip M
U
with a structure of an
S
U
module by imposing:
m/Q+m
/Q
= (Q
m+Qm
)/QQ
(P/Q) (m/Q
) = Pm/QQ
.
If U V , we have S(V ) S(U), hence the canonical homomorphisms
V
U
: M
V
M
U
;
the system (M
U
,
V
U
), where U and V run over nonempty open subsets of X,
dene thus a sheaf which we denote by A(M); we verify immediately that
lim
xU
M
U
= M
x
,
69
that is, that A(M)
x
= M
x
. In particular, we have A(S) = O and as the M
U
are S
U
modules, it follows that A(M) is a sheaf of A(S)modules, that is, an
algebraic sheaf on X. Any homomorphism : M M
denes in a natural
way the S
U
-linear homomorphisms
U
: M
U
M
U
, thus a homomorphism of
sheaves A() : A(M) A(M
), which we frequently denote . We clearly

have
A( +) = A() +A(), A(1) = 1, A( ) = A() A().
The operation A(M) is thus a covariant additive functor dened on the category
of graded Smodules and with values in the category of algebraic sheaves on X.
(The above denitions are quite analogous to these of 4, from Chap. II; it
should be noted however that S
U
is not the localization of S in S(U), but only
its homogeneous component of degree 0.)
58 First properties of the functor A(M)
Proposition 3. The functor A(M) is an exact functor.
Let M

M
be an exact sequence of graded Smodules and show

that the sequence M
x
x
is also exact. Let m
/Q M
x
be an
element of the kernel of ; by the denition of M
x
, there exist R S(x) such
that R(m
) = 0; but then there exists m M such that (m) = Rm
and we
have (m/RQ) = m
/Q, q.e.d.
(Compare with n
48, Lemma 1.)

Proposition 4. If M is a graded Smodule and if n is an integer, A(M(n))
is canonically isomorphic to A(M)(n).
Let i I, x U
i
and m/Q M(n)
x
, with m M(n)
p
, Q S(x), deg Q = p.
Put:
i,x
(m/Q) = m/t
n
i
Q M
x
,
which is valid because m M
n+p
and t
n
i
Q S(x). We immediately see that
i,x
: M(n)
x
M
x
is bijective for all x U
i
and denes an isomorphism
i
of
A(M(n)) to A(M) over U
i
. Moreover, we have
i

1
j
=
ij
(n) over U
i
U
j
.
By the denition of the operation F(n) and Proposition 4 of n
4, this shows
that A(M(n)) is isomorphic to A(M)(n).
Corollary. A(S(n)) is canonically isomorphic to O(n).
Indeed, it has been said that A(S) was isomorphic to O.
(It is also clear that A(S(n)) is isomorphic to O
(n), because O
(n)
x
consists
precisely of the rational functions P/Q such that deg P deg Q = n and Q
S(x).)
Proposition 5. Let M be a graded Smodule satisfying the condition (TF).
The algebraic sheaf A(M) is also a coherent sheaf. Moreover A(M) = 0 if and
only if M C.
70
If M C, for all m M and x X there exists Q S(x) such that
Qm = 0; it suces to take Q of a suciently large degree; we thus have M
x
= 0,
hence A(M) = 0. Let now M be a graded Smodule satisfying the condition
(TF); there exists a homogeneous submodule N of M, of nite type and such
that M/N C; applying the above together with Proposition 3, we see that
A(N) A(M) is bijective and it thus suces to prove that A(N) is coherent.
Since N is of nite type, there exists an exact sequence L
1
L
0
N 0
where L
0
and L
1
are free modules of nite type. By Proposition 3, the sequence
A(L
1
) A(L
0
) A(N) 0 is exact. But, by the corollary to Proposition
4, A(L
0
) and A(L
1
) are isomorphic to nite direct sums of the sheaves O(n
i
)
and are thus coherent. It follows that A(N) is coherent.
Let nally M be a graded Smodule satisfying (TF) and such that A(M) =
0; by the above considerations, we can suppose that M is of nite type. If m is
a homogeneous element of M, let a
m
be the annihilator of m, that is, the set of
all polynomials Q S such that Q m = 0; it is clear that a
m
is a homogeneous
ideal. Moreover, the assumption M
x
= 0 for all x X implies that the variety
of zeros of a
m
in K
r+1
is reduced to 0; Hilberts theorem of zeros shows
that every homogeneous polynomial of suciently large degree belongs to a
m
.
Applying this to the nite system of generators of M, we conclude immediately
M
p
= 0 for p suciently large, which completes the proof.
By combining Propositions 3 and 5 we obtain:
Proposition 6. Let M and M
be two graded Smodules satisfying the

condition (TF) and let : M M
be a homomorphism of M to M
. Then
A() : A(M) A(M
)
is injective (resp. surjective, bijective) if and only if is C-injective (resp.
C-surjective, C-bijective).
59 The graded Smodule associated to an algebraic sheaf
Let F be an algebraic sheaf on X and set:
(F) =
nZ
(F)
n
, with (F)
n
= (X, F(n)).
The group (F) is a graded group; we shall equip it with a structure of an
Smodule. Let s (X, F(q)) and let P S
p
; we can identify P with a section
of O(p) (cf. n
54), thus P s is a section of O(p)F(q) = F(q)(p) = F(p+q),

using the homomorphisms from n
54; we have then dened a section of F(p+q)

which we denote by P s instead of P s. The mapping (P, s) P s equips
(F) with a structure of an Smodule that is compatible with the grading.
71
In order to compare the functors A(M) and (F) we dene two canonical
homomorphisms:
: M (A(M)) and : A((F)) F.
Denition of . Let M be a graded Smodule and let m M
0
be a
homogeneous element of M of degree 0. The element m/1 is a well-dened
element of M
x
that varies continuously with x X; thus m denes a section
(m) of A(M). If now m is homogeneous of degree n, m is homogeneous of
degree 0 in M(n), thus denes a section (m) of A(M(n)) = A(M)(n) (cf.
Proposition 4). This is the denition of : M (A(M)) and it is immediate
that it is a homomorphism.
Denition of . Let F be an algebraic sheaf on X and let s/Q be an
element of (F)
x
with s (X, F(n)), Q S
n
and Q(x) ,= 0. The function
1/Q is homogeneous of degree n and regular in x, hence a section of O(n) in
a neighborhood of x; it follows that 1/Qs is a section of O(n) F(n) = F
in a neighborhood of x, thus denes an element of F
x
which we denote by
x
(s/Q), because it depends only on s/Q. We can also dene
x
by using the
components s
i
of s: if x U
i
,
x
(s/Q) = (t
n
i
/Q) s
i
(x). The collection of the
homomorphisms
x
denes a homomorphism : A((F)) F.
The homomorphisms and are related by the following Propositions,
which are shown by direct computation:
Proposition 7. Let M be a graded Smodule. The composition of the
homomorphisms A(M) A((A(M))) A(M) is the identity.
(The rst homomorphism is dened by : M (A(M)) and the second
is , applied to F = A(M).)
Proposition 8. Let F be an algebraic sheaf on X. The composition of the
homomorphisms (F) (A((F))) (F) is the identity.
(The rst homomorphism is , applied to M = (F), while the second one
is dened by : A((F)) F.)
We will show in n
65 that : A((F)) F is bijective if F is coherent

and that : M (A(M)) is C-bijective if M satises the condition (TF).
60 The case of coherent algebraic sheaves
Let us show a preliminary result:
Proposition 9. Let L be an algebraic sheaf on X, a direct sum of a nite
number of the sheaves O(n
i
). Then (F) satises (TF) and : A((L)) L
is bijective.
It comes down immediately L = O(n), then to L = O. In this case, we
know that (O(p)) = S
p
for p 0, thus we have S (O), the quotient
72
belonging to C. It follows rst that (O) satises (TF), then that A((O)) =
A(S) = O, q.e.d.
(We observe that we have (O) = S if r 1; on the other hand, if r = 0,
(O) is not even an Smodule of nite type.)
Theorem 2. For every coherent algebraic sheaf F on X there exists a
graded Smodule M, satisfying (TF), such that A(M) is isomorphic to F.
By the corollary to Theorem 1, there exists an exact sequence of algebraic
sheaves:
L
1

L
0
F 0,
where L
1
and L
0
satisfy the assumptions of the above Proposition. Let M be
the cokernel of the homomorphism () : (L
1
) (L
0
); by Proposition 9,
M satises the condition (TF). Applying the functor A to the exact sequence:
(L
1
) (L
0
) M 0,
we obtain an exact sequence:
A((L
1
)) A((L
0
)) A(M) 0.
Consider the following commutative diagram:
A((L
1
)) A((L
0
)) A(M) 0
L
1
L
0
F 0

By Proposition 9, the two vertical homomorphisms are bijective. It follows
that A(M) is isomorphic to F, q.e.d.
73
3. Cohomology of the projective space with values in a coherent algebraic sheafIII
3 Cohomology of the projective space with
values in a coherent algebraic sheaf
61 The complexes C
k
(M) and C(M)
We preserve the notations of n
os
51 and 56. In particular, I will denote the
interval 0, 1, . . . , r and S will denote the graded algebra K[t
0
, . . . , t
r
].
Let M be a graded Smodule, k and q two integers 0; we shall dene a
group C
q
k
(M): an element of C
q
k
(M) is a mapping
(i
0
, . . . , i
q
) mi
0
. . . i
q
which associates to every sequence (i

0
, . . . , i
q
) of q + 1 elements of I a homo-
geneous element of degree k(q + 1) of M, depending in an alternating way on
i
0
, . . . , i
q
. In particular, we have mi
0
. . . i
q
= 0 if two of the indices i
0
, . . . , i
q
are equal. We dene addition in C
q
k
(M) in the obvious way. the same with
multiplication by an element K, and C
q
k
(M) is a vector space over K.
If m is an element of C
q
k
(M), we dene dm C
q+1
k
(M) by the formula:
(dm)i
0
. . . i
q+1
=
j=q+1
j=0
(1)
j
t
k
ij
mi
0
. . .
i
j
. . . i
q+1
.
We verify by a direct calculation that dd = 0; thus, the direct sum C
k
(M) =
q=r
q=0
C
q
k
(M), equipped with the coboundary operator d, is a complex, whose
q-th cohomology group is denoted by H
q
k
(M).
(We note, after [11], another interpretation of the elements of C
q
k
(M): intro-
duce r + 1 dierential symbols dx
0
, . . . , dx
r
and associate to every m C
q
k
(M)
a ,,dierential form of degree q + 1:
m
=
i0<...<iq
mi
0
. . . i
q
dx
i0
. . . dx
iq
.
If we put
k
=
i=r
i=0
t
k
i
dx
i
, we see that we have:
dm
=
k

m
,
in other words, the coboundary operation is transformed into the exterior mul-
tiplication by the form
k
).
If h is an integer k, let
h
k
: C
q
k
(M) C
q
h
(M) be the homomorphism
dened by the formula:
h
k
(m)i
0
. . . i
q
= (t
i0
. . . t
iq
)
hk
mi
0
. . . i
q
.
74
We have
h
k
d = d
h
k
and
l
h

j
k
=
l
k
if k h l. We can thus dene a
complex C(M), the inductive limit of the system (C
k
(M),
h
k
) for k +. The
cohomology groups of this complex are denoted H
q
(M). Because cohomology
commutes with inductive limits (cf. [6], Chap. V, Prop. 9.3*), we have:
H
q
(M) = lim
k
H
q
k
(M).
Every homomorphism : M M
denes a homomorphism
: C
k
(M) C
k
(M
)
by the formula: (m)i
0
. . . i
q
= (mi
0
. . . i
q
), hence, by passing to the limit,
: C(M) C(M
); moreover, these homomorphisms commute with boundary

and thus dene the homomorphisms
: Hq
k
(M) Hq
k
(M
) and : H
q
(M) H
q
(M
).
If we have an exact sequence 0 M M
0, we have an exact
sequence of complexes 0 C
k
(M) C
k
(M
) C
k
(M
) 0, hence an exact
sequence of cohomology:
. . . H
q
k
(M
) H
q
k
(M
) H
q+1
k
(M) H
q+1
k
(M
) . . .
The same results for C(M) and H
q
(M).
Remark. We shall see later (cf. n
69) that we can express H

q
k
(M) in terms
of Ext
q
S
.
62 Calculation of H
q
k
(M) for certain modules M
Let M be a graded Smodule and m M a homogeneous element of degree
0. The system of (t
k
i
m) is a 0-cocycle of C
k
(M), which we denote by
k
(m)
and identify with its cohomology class. We so obtain a K-linear homomorphism
k
: M
0
H
0
k
(M); as
h
=
h
k

k
if h k, the
k
dene by passing to the
limit a homomorphism : M
0
H
0
(M).
Let us introduce two more notations:
If (P
0
, . . . , P
h
) are elements of S, we denote by (P
0
, . . . , P
h
)M the submodule
of M consisting of the elements
i=h
i=0
P
i
m
i
with m
i
M; if the P
i
are
homogeneous, this submodule is homogeneous.
If P is an element of S and N a submodule of M, we denote by N : P the
submodule of M consisting of the elements m M such that P m N; we
clearly have N : P N; if N and P are homogeneous, so is N : P.
Having specied these notations, we have:
75
Proposition 1. Let M be a graded Smodule and k an integer 0. Assume
that for all i I we have:
(t
k
0
, . . . , t
k
i1
)M : t
k
i
= (t
k
0
, . . . , t
k
i1
)M.
Then:
(a)
k
: M
0
H
0
k
(M) is bijective (if r 1),
(b) H
q
k
(M) = 0 for 0 < q < r.
(For i = 0, the assumption means that t
k
0
m = 0 implies m = 0.)
This Proposition is a special case of a result of de Rham [11] (the de Rhams
result being also valid even if we do not assume that the mi
0
. . . i
q
are homo-
geneous). See also [6], Chap. VIII, 4 for a particular case, sucient for our
purposes.
We now apply Proposition 1 to the graded Smodule S(n):
Proposition 2. Let k be an integer 0, n an arbitrary integer. Then:
(a)
k
: S
n
H
0
k
(S(n)) is bijective (if r 1),
(b) H
q
k
(S(n)) = 0 for 0 < q < r,
(c) H
r
k
(S(n)) admits a base (over K) consisting of the cohomology classes
of the monomials t
0
0
. . . t
r
r
with 0
i
< k and
i=r
i=0
i
= k(r + 1) +n.
It is clear that the Smodule S(n) satises the assumptions of Proposition
1, which shows (a) and (b). On the other hand, for every graded Smodule M,
we have H
r
k
(M) = M
k(r+1)
/(t
k
0
, . . . , t
k
r
)M
kr
; now the monomials
t
0
0
. . . t
r
r
,
i
0,
i=r
i=0
i
= k(r + 1) +n,
form a basis of S(n)
k(r+1
) and those for which at least
i
is k form a basis
of (t
k
0
, . . . , t
k
r
)S(n)
kr
; hence (c).
It is convenient to write the exponents
i
in the form
i
= k
i
. The
conditions of (c) are then written:
0 <
i
k and
i=r
i=0
i
= n.
The second condition, together with
i
> 0, implies
i
n r; if thus
k nr, the condition
i
k is a consequence of the preceding two. Hence:
Corollary 1. For k nr, H
r
k
(S(n)) admits a basis formed of the coho-
mology classes of monomials (t
0
. . . t
r
)
k
/t
0
0
. . . t
r
r
with
i
> 0 and
i=r
i=0
i
=
n.
We also have:
76
Corollary 2. If h k n r, the homomorphism
h
k
: H
q
k
(S(n)) H
q
k
(S(n))
is bijective for all q 0.
For q ,= r, this follows from the assertions (a) and (b) of Proposition 2. For
q = r, this follows from Corollary 1, given that
h
k
transforms
(t
0
. . . t
r
)
k
/t
0
0
. . . t
r
r
into (t
0
. . . t
r
)
h
/t
0
0
. . . t
r
r
.
Corollary 3. The homomorphism : S
n
H
0
(S(n)) is bijective if r 1
or if n 0. We have H
q
(S(n)) = 0 for 0 < q < r and H
r
(S(n)) is a vector
space of dimension
_
n1
r
_
over K.
The assertion pertaining to follows from Proposition 2, (a), in the case
when r 1; it is clear if r = 0 and n 0. The rest of the Corollary is an
obvious consequence of Corollaries 1 and 2 (seeing that the binomial coecient
_
a
r
_
is zero for a < r).
63 General properties of H
q
(M)
Proposition 3. Let M be a graded Smodule satisfying the condition (TF).
Then:
(a) There exists an integer k(M) such that
h
k
: H
q
k
(M) H
q
h
(M) is bijec-
tive for h k k(M) and every q.
(b) H
q
(M) is a vector space of nite dimension over K for all q 0.
(c) There exists an integer n(M) such that for n n(M), : M
n

H
0
(M(n)) is bijective and that H
q
(M(n)) is zero for all q > 0.
This is immediately reduced to the case when M is of nite type. We say
that M is of dimension s (s being an integer 0) if there exists an exact
sequence:
0 L
s
L
s1
. . . L
0
M 0,
where L
i
are free graded Smodules of nite type. By the Hilbert syzygy
theorem (cf. [6], Chap. VIII, th. 6.5), this dimension is always r + 1.
We prove the Proposition by induction on the dimension of M. If it is 0, M is
free of nite type, i.e. a direct sum of modules S(n
i
) and the Proposition follows
from Corollaries 2 and 3 and Proposition 2. Assume that M is of dimension s
and let N be the kernel of L
0
M. The graded Smodule N is of dimension
s 1 and we have an exact sequence:
0 N L
0
M 0.
77
By the induction assumption, the Proposition is true for N and L
0
. Applying
the ve lemma ([7], Chap. I, Lemme 4.3) to the commutative diagram:
H
q
k
(N) H
q
k
(L
0
) H
q
k
(M) H
q+1
k
(N) H
q+1
k
(L
0
)
H
q
h
(N) H
q
h
(L
0
) H
q
h
(M) H
q+1
h
(N) H
q+1
h
(L
0
),
where h k Sup(k(N), k(L
0
), we show (a), thus also (b), because the H
q
k
(M)
are of nite dimension over K. On the other hand, the exact sequence
H
q
(L
0
(n)) H
q
(M(n)) H
q+1
(N(n))
shows that H
q
(M(n)) = 0 for n Sup(n(L
0
), n(N)). Finally, consider the
0 N
n
L
n
M
n
0
0 H
0
(N(n)) H
0
(L
0
(n)) H
0
(M(n)) H
1
(N(n));

for n n(N), we have H
1
(N(n)) = 0; we deduce that : M
n
H
0
(M(n)) is
bijective for n Sup(n(L
0
), n(N)), which completes the proof of the Proposi-
tion.
64 Comparison of the groups H
q
(M) and H
q
(X, A(M))
Let M be a graded Smodule and let A(M) be the algebraic sheaf on X =
P
r
(K) dened by M by the procedure of n
57. We will now compare C(M) with

C
(U, A(M)), the complex of alternating cochains of the covering U = U

i
iI
with values in the sheaf A(M).
Let m C
q
k
(M) and let (i
0
, . . . , i
q
) be a sequence of q +1 elements of I. The
polynomial (t
i0
. . . t
iq
)
k
belongs obviously to S(U
i0...iq
), with the notations of n
57. It follows that mi

0
. . . i
q
/(t
i0
. . . t
iq
)
k
belongs to M
U
, where U = U
i0...iq
,
thus denes a section of A(M) over U
i0...iq
. When (i
0
, . . . , i
q
) varies, the system
consisting of this sections is an alternating cochain of U with values in A(M),
which we denote by
k
(m). We immediately see that
k
commutes with d and
that
k
=
h

h
k
if h k. By passing to the inductive limit, the
k
thus dene
a homomorphism : C(M) C
(U, A(M)), commuting with d.

Proposition 4. If M satises the condition (TF), : C(M) C
(U, A(M))
is bijective.
78
If M C, we have M
n
= 0 for n n
0
, so C
k
(M) = 0 for k n
0
and
C(M) = 0. As every Smodule satisfying (TF) is C-isomorphic to a module of
nite type, this shows that we can restrict ourselves to the case when M is of
nite type. We can then nd an exact sequence L
1
L
0
M 0, where L
1
and L
0
are free of nite type. By Propositions 3 and 5 from n
58, the sequence

A(L
1
) A(L
0
) A(M) 0
is an exact sequence of coherent algebraic sheaves; as the U
i0...iq
are ane open
subsets, the sequence
C
(U, A(L
1
)) C
(U, A(L
0
) C
(U, A(M)) 0
is exact (cf. n
45, Corollary 2 to Theorem 2). The commutative diagram

C(L
1
) C(L
0
) C(M) 0
C
(U, A(L
1
)) C
(U, A(L
0
)) C
(U, A(M)) 0

then shows that if the Proposition is true for the module L
1
and L
0
, so it is
for M. We are thus reduced to the special case of a free module of nite type,
then, by the decomposition into direct summands, to the case when M = S(n).
In this case, we have A(S(n)) = O(n); a section f
i0...iq
of O(n) over U
i0...iq
is, by the sole denition of this sheaf, a regular function on V
i0
. . . V
iq
and
homogeneous of degree n. As V
i0
. . . V
iq
as the set of points of K
r+1
where
the function t
i0
. . . t
iq
is ,= 0, there exists an integer k such that
f
i0...iq
= Pi
0
. . . i
q
/(t
i0
. . . t
iq
)
k
,
Pi
0
. . . i
q
being a homogeneous polynomial of degree n + k(q + 1), that is, of
degree k(q+1) in S(n). Thus, every alternating cochain f C
(U, O(n)) denes

a system Pi
0
. . . i
q
that is an element of C
k
(S(n)); hence a homomorphism
: C
(U, O(n)) C(S(n)).

As we verify immediately that = 1 and = 1, it follows that is
bijective, which completes the proof.
Corollary. denes an isomorphism of H
q
(M) with H
q
(X, A(M)) for all
q 0.
Indeed, we know that H
q
(U, A(M)) = H
q
(U, A(M)) (n
20, Proposition 2)
and that H
q
(U, A(M)) = H
q
(X, A(M)) (n
52, Proposition 2, which applies

because A(M) is coherent).
Remark. It is easy to see that : C(M) C
(U, A(M)) is injective even

if M does not satisfy the condition (TF).
79
65 Applications
Proposition 5. If M is a graded Smodule satisfying the condition (TF), the
homomorphism : M (A(M)), dened in n
59, is C-bijective.
We must observe that : M
n
(X, A(M(n))) is bijective for n suciently
large. Then, by Proposition 4, (X, A(M(n))) is identied with H
0
(M(n));
the Proposition follows thus from Proposition 3, (c), given the fact that the
homomorphism is transformed by the above identication to a homomorphism
dened at the beginning of n
62, also denoted by .

Proposition 6. Let F be a coherent algebraic sheaf on X. The graded S
module (F) satises the condition (TF) and the homomorphism : A((F))
F dened in n
59 is bijective.
By Theorem 2 of n
60, we can assume that F = A(M), where M is a

module satisfying (TF). By the above Proposition, : M (A(M)) is C-
bijective; as M satises (TF), it follows that (A(M)) satises it also. Applying
Proposition 6 from n
58, we see that : A(M) A((A(M))) is bijective.

Since the composition A(M)

A((A(M)))

A(M) is the identity (n
59, Proposition 7), it follows that is bijective, q.e.d.

Proposition 7. Let F be a coherent algebraic sheaf on X. The groups
H
q
(X, F) are vector spaces of nite dimension over K for all q 0 and we
have H
q
(X, F(n)) = 0 for q > 0 and n suciently large.
We can assume, as above, that F = A(M) where M is a module satisfying
(TF). The Proposition then follows from Proposition 3 and the corollary to
Proposition 4.
Proposition 8. We have H
q
(X, O(n)) = 0 for 0 < q < r and H
r
(X, O(n))
is a vector space of dimension
_
n1
r
_
over K, admitting a base consisting of
the cohomology classes of the alternating cocycles of U
f
01...r
= 1/t
0
0
. . . t
r
r
with
i
> 0 and
i=r
i=0
i
= n.
We have O(n) = A(S(n)), hence H
q
(X, O(n)) = H
q
(S(n)), by the corollary
to Proposition 4; the Proposition follows immediately from this and from the
corollaries of Proposition 2.
We note that in particular H
r
(X, O(r 1)) is a vector space of dimension
1 over K, with a base consisting of the cohomology class of the cocycle f
01...r
=
1/t
0
. . . t
r
.
80
66 Coherent algebraic sheaves on projective varieties
Let V be a closed subvariety of the projective space X = P
r
(K) and let F
be a coherent algebraic sheaf on V . By extending F by 0 outside V , we ob-
tain a coherent algebraic sheaf on X (cf. n
39) denoted F
X
; we know that
H
q
(X, F
X
) = H
q
(V, F). The results of the preceding n
thus apply to the

groups H
q
(V, F). We obtain immediately (given n
52):
Theorem 1. The groups H
q
(V, F) are vector spaces of nite dimension
over K, zero for q > dimV .
In particular, for q = 0 we have:
Corollary. (V, F) is a vector space of nite dimension over K.
(It is natural to conjecture whether the above theorem holds for all complete
varieties, in the sense of Weil [16].)
Let U
i
= U
i
V ; the U
i
form an open covering U
of V . If F is an algebraic
sheaf on V , let F
i
= F(U
i
) and let
ij
(n) be the isomorphism of F
j
(U
i
U
j
)
to F
i
(U
i
U
j
) dened by multiplication by (t
j
/t
i
)
n
. We denote by F(n) the
sheaf obtained by gluing the F
i
with respect to
ij
(n). The operation F(n)
has the same properties as the operation dened in n
54 and generalizes it; in

particular, F(n) is canonically isomorphic to F O
V
(n).
We have F
X
(n) = F(n)
X
. Applying then Theorem 1 of n
55, together
with Proposition 7 from n
65, we obtain:
Theorem 2. Let F be a coherent algebraic sheaf on V . There exists an
integer m(F) such that we have, for all n m(F):
(a) For all x V , the O
x,V
module F(n)
x
is generated by the elements of
(V, F(n)),
(b) H
q
(V, F(n)) = 0 for all q > 0.
Remark. It is essential to observe that the sheaf F(n) does not depend
solely on F and n, but also on the embedding of V into the projective space X.
More precisely, let P be the principal bundle
1
(V ) with the structural group
K
; with n an integer, we make K
act on K by the formula:

(, )
n
if K
and K.
Let E
n
= P
K
K be the ber space associated to P and the ber K, equipped
with the above action; let S(E
n
) be the sheaf of germs of sections of E
n
(cf. n
41). Taking into account the fact that t

i
/t
j
form a system of transition maps
of P, we verify immediately that S(E
n
) is canonically isomorphic to O
V
(n)).
The formula F(n) = F O
V
(n) = F S(E
n
) shows then that the operation
F F(n) depends only on the class of the principal bundle P dened by the
embedding V X. In particular, if V is normal, F(n) depends only on the class
of linear equivalence of hyperplane sections of V in the considered embedding
(cf. [17]).
81
67 A supplement
If M is a graded Smodule satisfying (TF), we denote by M
the graded S
module (A(M)). We have seen in n
65 that : M M
is C-bijective. We
shall now give conditions for to be bijective.
Proposition 9. : M M
is bijective if and only if the following

conditions are satised:
(i) If m M is such that t
i
m = 0 for all i I, then m = 0,
(ii) If elements m
i
M, homogeneous of the same degree, satisfy t
j
m
i
=
t
i
m
j
= 0 for every couple (i, j), there exists an m M such that m
i
= t
i
m.
Let us show that the conditions (i) and (ii) are satised by M
, which will
prove the necessity. For (i), we can assume that m is homogeneous, that is, it is
a section of A(M(n)); in this case, the condition t
i
m = 0 implies that m is zero
on U
i
, and since this occurs for all i I, we have m = 0. For (ii), let n be the
degree of m
i
; we thus have m
i
(A(M(n))); as 1/t
i
is a section of O(1) over
U
i
, m
i
/t
i
is a section of A(M(n1)) over U
i
and the condition t
j
m
i
t
i
m
j
shows that these various sections are the restrictions of a unique section m of
A(M(n1)) over X; it remains to compare the sections t
i
m and m
i
; to show
that they coincide on U
j
, it suces to observe that t
j
(t
i
m m
i
) = 0 on U
j
,
which follows from the formula t
j
m
i
= t
i
m
j
and the denition of m.
We will now show that (i) implies that is injective. For n suciently large,
we know that : M
n
M
n
is bijective and we can thus proceed by descending
induction on n. If (m) = 0 with m M
n
, we have t
i
(m) = (t
i
m) = 0 and
the induction assumption, applicable since t
i
m M
n+1
, shows that m = 0.
Finally, let us show that (i) and (ii) imply that is surjective. We can, as
before, proceed by descending induction on n. If m
n
, the induction
assumption shows that there exist m
i
M
n+1
such that (m
i
) = t
i
m
; we
have (t
j
m
i
t
i
m
j
) = 0, hence t
j
m
i
t
i
m
j
= 0, because is injective. The
condition (ii) then implies that there exists an m M
n
such that t
i
m = m
i
;
we have t
i
(m
(m)) = 0, which shows that m
= (m) and completes the

proof.
Remarks. (1) The proof shows that the condition (i) is sucient and
necessary for to be injective.
(2) We can express (i) and (ii) as: the homomorphism
1
: M
n
H
0
q
(M(n))
is bijective for all n Z. Besides, Proposition 4 shows that we can identify M
with the Smodule
nZ
H
0
(M(n)) and it would be easy to provide a purely
algebraic proof of Proposition 9 (without using the sheaf A(M)).
82
4. Relations with the functors Ext
q
S
III
4 Relations with the functors Ext
q
S
68 The functors Ext
q
S
We keep the notations of n
56. If M and N are two graded S-modules, we de-

note by Hom
S
(M, N)
n
the group of homogeneous S-homomorphisms of degree
n from M to N, and by Hom
S
(M, N) the graded group
nZ
Hom
S
(M, N)
n
;
it is a graded S-module; when M is of nite type it coincides with the S-module
of all S-homomorphisms from M to N.
The derived functors (cf. [6], Chapter V) of the functor Hom
S
(M, N) are
the functors Ext
q
S
(M, N), q = 0, 1, . . .. Let us briey recall their denition:
1
One chooses a ,,resolution of M, that is, an exact sequence:
. . . L
q+1
L
q
. . . L
0
M 0,
where the L
q
are free graded S-modules and the maps are homomorphisms
(that is, as usual, homogeneous S-homomorphisms of degree 0). If we set
C
q
= Hom
S
(L
q
, N), the homomorphism L
q+1
L
q
denes by transposition
a homomorphism d : C
q
C
q+1
satisfying d d = 0; therefore C =
q0
C
q
is endowed with a structure of a complex, and the q-th cohomology group of
C is just, by denition, equal to Ext
q
S
(M, N); one shows that it does not de-
pend on the chosen resolution. As the C
q
are graded S-modules and since
d : C
q
C
q+1
is homogeneous of degree 0, the Ext
q
S
(M, N) are S-modules
graded by the subspaces Ext
q
S
(M, N)
n
); the Ext
q
S
(M, N) are the cohomology
groups of the complex formed by the Hom
S
(L
q
, N)
n
), i.e., are the derived func-
tors of the functor Hom
S
(M, N)
n
).
Recall the main properties of Ext
q
S
:
Ext
0
S
(M, N) = Hom
S
(M, N); Ext
q
S
(M, N) = 0 for q > r +1 if M is of nite
type (due to the Hilbert syzygy theorem, cf. [6], Chapter VIII, theorem 6.5);
Ext
q
S
(M, N) is an S-module of nite type if M and N are both of nite type
(because we can choose a resolution with the L
q
of nite type); for all n Z we
have the canonical isomorphisms:
Ext
q
S
(M(n), N) Ext
q
S
(M, N(n)) Ext
q
S
(M, N)(n).
The exact sequences:
0 N N
0 and 0 M M
0
give rise to exact sequences:
. . . Ext
q
S
(M, N) Ext
q
S
(M, N
) Ext
q
S
(M, N
) Ext
q+1
S
(M, N) . . .
. . . Ext
q
S
(M
, N) Ext
q
S
(M
, M) Ext
q
S
(M, N) Ext
q1
S
(M
, N) . . .
1
When M is not of nite type, the Ext
q
S
(M, N) dened above can dier from the
Ext
q
S
(M, N) dened in [6]: it is due to the fact that Hom
S
(M, N) does not have the same
meaning in both cases. However, all the proofs of [6] are valid without change in the case
considered here: this is seen either directly or by applying Appendix of [6].
83
q
S
III
69 Interpretation of H
q
k
(M) in terms of Ext
q
S
Let M be a graded S-module and let k be an integer 0. Set:
B
q
k
(M) =
nZ
H
q
k
(M(n)),
with the notations of n
61.
We thus obtain a graded group, isomorphic with the q-th cohomology group
of the complex
nZ
C
k
(M(n)); this complex can be given a structure of an
S-module, compatible with the grading by setting
(P m)i
0
i
q
= P mi
0
i
q
, if P S
p
and mi
0
i
q
C
q
k
(M(n));
as the coboundary operator is a homogeneous S-homomorphism of degree 0, it
follows that the B
q
k
(M) are themselves graded S-modules.
We put
B
q
(M) = lim
k
B
q
k
(M) =
nZ
H
q
(M(n)).
The B
q
(M) are graded S-modules. For q = 0 we have
B
0
(M) =
nZ
H
0
(M(n)),
and we recognize the module denoted by M
in n
67 (when M satises the

condition (TF)). For each n Z, we have dened in n
62 a linear map :
M
n
H
0
(M(n)); we verify immediately that the sum of these maps denes a
homomorphism, which we denote also by , from M to B
0
(M).
Proposition 1. Let k be an integer 0 and let J
k
be the ideal (t
k
0
, . . . , t
k
r
) of
S. For every graded S-module M, the graded S-modules B
q
k
(M) and Ext
q
S
(J
k
, M)
are isomorphic.
Let L
q
k
, q = 0, . . . , r be the free graded S-module with a base consisting of
the elements ei
0
i
q
, 0 i
0
< i
1
< . . . < i
q
r of degree k(q +1); we dene
an operator d : L
q+1
k
L
q
k
and an operator : L
0
k
J
k
by the formulas:
d(ei
0
i
q+1
) =
j=q+1
j=0
(1)
j
t
k
ij
ei
0

i
j
i
q+1
,
(ei) = t
k
i
.
Lemma 1. The sequence of homomorphisms:
0 L
r
k
d
L
r1
k
. . . L
0
k
J
k
0
is an exact sequence.
84
q
S
III
For k = 1, this result is well known (cf. [6], Chapter VIII, 4); the general
case is shown in the same way (or reduced to it); we can also use the theorem
shown in [11].
Proposition 1 follows immediately from the Lemma, if we observe that the
complex formed by the Hom
S
(L
q
k
, M) and the transposition of d is just the
complex
nZ
C
k
(M(n)).
Corollary 1. H
q
k
(M) is isomorphic to Ext
q
S
(J
k
, M)
0
.
Indeed, these groups are the degree 0 components of the graded groups
B
q
k
(M) and Ext
q
S
(J
k
, M).
Corollary 2. H
q
(M) is isomorphic to lim
k
Ext
q
S
(J
k
, M)
0
.
We easily see that the homomorphism
h
k
: H
q
k
(M) H
q
h
(M) from n
61 is
transformed by the isomorphism from Corollary 1 to a homomorphism from
Ext
q
S
(J
k
, M)
0
to Ext
q
S
(J
h
, M)
0
induced by the inclusion J
h
J
k
; hence the Corollary 2.
Remark. Let M be a graded S-module of nite type; M denes (cf. n
48)
a coherent algebraic sheaf F
on K
r+1
, thus on Y = K
r+1
0 and we can
verify that H
q
(Y, F
) is isomorphic to B
q
(M).
70 Denition of the functors T
q
(M)
Let us rst dene the notion of a dual module to a graded S-module. Let M
be a graded S-module; for all n Z, M
n
is a vector space over K, whose dual
vector space we denote by (M
n
)
. Set
M
nZ
M
n
, with M
n
= (M
n
)
.
We give M
the structure of an S-module compatible with the grading; for all

P S
p
, the mapping m P m is a K-linear map from M
np
to M
n
, so
denes by transposition a K-linear map from (M
n
)
= M
n
to (M
np
)
=
M
n+p
; this denes the structure of an S-module on M
. We could also dene

M
as Hom
S
(M, K), denoting by K the S-graded module S/(t
0
, . . . , t
r
).
The graded S-module M
is called the module dual to M; we have M
= M
if every M
n
is of nite dimension over K, which holds if M = (F), F being a
coherent algebraic sheaf on X, or if M is of nite type. Every homomorphism
: M N denes by transposition a homomorphism from N
to M
. If the
sequence M N P is exact, so is the sequence P
; in other
words, M
is a contravariant and exact functor of the module M. When I is a

homogeneous ideal of S, the dual of S/I is exactly the ,,inverse system of I, in
the sense of Macaulay (cf. [9], n
25).
85
q
S
III
Let now M be a graded S-module and q an integer 0. In the preceding
n
, we have dened the graded S-module B

q
(M); the module dual to B
q
(M)
will be denoted by T
q
(M). We thus have, by denition:
T
q
(M) =
nZ
T
q
(M)
n
, with T
q
(M)
n
= (H
q
(M(n)))
.
Every homomorphism : M N denes a homomorphism from B
q
(M)
to B
q
(N), thus a homomorphism from T
q
(N) to T
q
(M); thus the T
q
(M) are
contravariant functors of M (we shall see in n
72 that they can expressed very

simply in terms of Ext
S
). Every exact sequence:
0 M N P 0
gives rise to an exact sequence:
. . . B
q
(M) B
q
(N) B
q
(P) B
q+1
(M) . . . ,
thus, by transposition, an exact sequence:
. . . T
q+1
(M) T
q
(P) T
q
(N) T
q
(M) . . . .
The homomorphism : M B
0
(M) denes by transposition a homomor-
phism
: T
0
(M) M
.
Since B
q
(M) = 0 for q > r, we have T
q
(M) = 0 for q > r.
r
(M).
(In this n
, and in the following, we assume that we have r 1; the case r = 0

leads to somehow dierent, and trivial, statements).
We denote by the graded S-module S(r 1); this is a free module, with
a base consisting of an element of degree r + 1. We have seen in n
62 that
H
r
() = H
r
k
() for k suciently large, and that H
r
k
() admits a base over K
consisting of a single element (t
0
. . . t
r
)
k
/t
0
. . . t
r
; the image in H
r
() of this
element will be denoted by ; is thus a basis of H
r
().
We will now dene a scalar product h, between elements h B
r
(M)
n
and Hom
S
(M, )
n
, M being an arbitrary graded S-module. The element
can be identied with an element of Hom
S
(M(n), )
0
, that is, with a homo-
morphism from M(n) to ; it thus denes, by passing to cohomology groups,
a homomorphism from H
r
(M(n)) = B
r
(M)
n
to H
r
(), which we also de-
note by . The image of h under this homomorphism is thus a scalar multiple
of , and we dene h, by the formula:
(h) = h, .
86
q
S
III
For every Hom
S
(M, )
n
, the function h h, is a linear form on
B
r
(M)
n
, thus can be identied with an element () of the dual of B
r
(M)
n
,
which is T
r
(M)
n
. We have thus dened a homogeneous mapping of degree 0
: Hom
S
(M, ) T
r
(M),
and the formula P h, = h, P shows that is an S-homomorphism.
Proposition 2. The homomorphism : Hom
S
(M, ) T
r
(M) is bijective.
We shall rst prove the Proposition when M is a free module. If M is a
direct sum of homogeneous submodules M
, we have:
Hom
S
(M, )
n
=
Hom
S
(M
, )
n
and T
r
(M)
n
=
T
r
(M
)
n
.
So, if the proposition holds for the M
, it holds for M, and this reduces the case

of free modules to the particular case of a free module with a single generator,
that is, to the case when M = S(m). We can identify Hom
S
(M, )
n
with
Hom
S
(S, S(n m r 1))
0
, that is, with the vector space of homogeneous
polynomials of degree n m r 1. Thus Hom
S
(M, )
n
has for a base the
family of monomials t
0
0
. . . t
r
r
with
i
0 and
i=r
i=0
i
= n m r 1. On
the other hand, we have seen in n
62 that H
r
k
(S(m n)) has for a base (if
k is large enough) the family of monomials (t
0
. . . t
r
)
k
/t
0
0
. . . t
r
r
with
i
> 0
and
i=r
i=0
i
= n m. By setting
i
=
i
+ 1, we can write these monomials
in the form (t
0
. . . t
r
)
k1
/t
0
0
. . . t
r
r
, with
i
0 and
i=r
i=0
i
= n m r 1.
Comparing the denition of h, , we observe that the scalar product
(t
0
. . . t
r
)
k1
/t
0
0
. . . t
r
r
, t
0
0
. . . t
r
r

is always zero, unless
i
=
i
for all i, in which case it is equal to 1. This means
that transforms the basis of t
0
0
. . . t
r
r
to the dual basis of (t
0
. . . t
r
)
k1
/t
0
0
. . . t
r
r
,
thus is bijective, which shows the Proposition in the case when M is free.
Let us now pass to the general case. We choose an exact sequence
L
1
L
0
M 0
where L
0
and L
1
are free. Consider the following commutative diagram
0 Hom
S
(M, ) Hom
S
(L
0
, ) Hom
S
(L
1
, )
0 T
r
(M) T
r
(L
0
) T
r
(L
1
).

The rst row of this diagram is an exact sequence, by the general properties of
the functor Hom
S
; the second is also exact, because it is dual to the sequence
B
r
(L
1
) B
r
(L
0
) B
r
(M) 0,
87
q
S
III
which is exact by the cohomology exact sequence of B
q
and the fact that
B
r+1
(M) = 0 for any M. On the other hand, the two vertical homomorphisms
: Hom
S
(L
0
, ) T
r
(L
0
) and : Hom
S
(L
1
, ) T
r
(L
1
)
are bijective, as we have just seen. It follows that
: Hom
S
(M, ) T
r
(M)
is also bijective, which completes the proof.
q
(M).
We shall now prove the following theorem, which generalizes Proposition 2:
Theorem 1. Let M be a graded S-module. For q ,= r, the graded S-
modules T
rq
(M) and Ext
q
S
(M, ) are isomorphic. Moreover, we have an exact
sequence:
0 Ext
r
S
(M, ) T
0
(M)

Ext
r+1
S
(M, ) 0.
We will use the axiomatic characterization of derived functors given in [6],
Chap. III, 5. For this, we rst dene new functors E
q
(M) in the following
manner:
For q ,= r, r + 1, E
q
(M) = T
rq
(M),
For q = r, E
r
(M) = Ker(
),
For q = r + 1, E
r+1
(M) = Coker(
).
The E
q
(M) are additive functors of M, enjoying the following properties:
(i) E
0
(M) is isomorphic to Hom
S
(M, ).
This follows from Proposition 2.
(ii) If L is free, E
q
(L) = 0 for q > 0.
It suces to verify this for L = S(n), in which case it follows from n
62.
(iii) To every exact sequence 0 M N P 0 there is associated a
sequence of coboundary operators d
q
: E
q
(M) E
q+1
(P) and the sequence:
. . . E
q
(P) E
q
(N) E
q
(M)
d
q
E
q+1
(P) . . .
is exact.
88
q
S
III
The denition of d
q
is obvious if q ,= r 1, r: this is the homomorphism from
T
rq
(M) to T
rq1
(P) dened in n
70. For q = r1 or r, we use the following

T
1
(M) T
0
(P) T
0
(N) T
0
(M) 0
0 P
0.
This diagram shows immediately that the image of T

1
(M) is contained in
the kernel of
: T
0
(P) P
, which is just E
r
(P). This denes d
r1
:
E
r1
(M) E
r
(P).
To dene d
r
: Ker(T
0
(M) M
) Coker(T
0
(P) P
), we use the
process from [6], Chap. III, Lemma 3.3: if x Ker(T
0
(M) M
), there exists
y P
and z T
0
(N) such that x is the image of z and that y and z have the
same image in N
; we then set d
r
(x) = y.
The exactness of the sequence
. . . E
q
(P) E
q
(N) E
q
(M)
d
q
E
q+1
(P) . . .
follows from the exactness of the sequence
. . . T
rq
(P) T
rq
(N) T
rq
(M) T
rq1
(P) . . .
and from [6], loc. cit.
(iv) The isomorphism from (i) and the operators d
q
from (iii) are ,,natural
This follows immediately from the denitions.
As the properties (i) to (iv) characterize the derived functors of the functor
Hom
S
(M, ), we have E
q
(M) Ext
q
S
(M, ), which proves the Theorem.
Corollary 1. If M satises (TF), H
q
(M) is isomorphic to the vector space
dual to Ext
rq
S
(M, )
0
for all q 1.
In fact, we know that H
q
(M) is a vector space of nite dimension, whose
dual is isomorphic to Ext
rq
S
(M, )
0
.
Corollary 2. If M satises (TF), the T
q
(M) are graded S-modules of nite
type for q 1, and T
0
(M) satises (TF).
We can replace M by a module of nite type without changing the B
q
(M),
thus T
q
(M). The Ext
rq
S
(M, ) are then S-modules of nite type, and we have
M
C, hence the Corollary.

89
5. Applications to coherent algebraic sheaves III
5 Applications to coherent algebraic
sheaves
73 Relations between functors Ext
q
S
and Ext
q
Ox
Let M and N be two graded S-modules. If x is a point of X = P
r
(K), we
have dened in n
57 the O
x
-modules M
x
and N
x
; we will nd relation between
Ext
q
Ox
(M
x
, N
x
) and graded S-module Ext
q
S
(M, N).
Proposition 1. Suppose that M is of nite type. Then:
(a) The sheaf A(Hom
S
(M, N)) is isomorphic to the sheaf Hom
O
(A(M), A(N)).
(b) For all x X, the O
x
-module Ext
q
S
(M, N)
x
is isomorphic to the O
x
-module
Ext
q
Ox
(M
x
, N
x
).
First dene a homomorphism
x
: Hom
S
(M, N)
x
Hom
Ox
(M
x
, N
x
). An
element of rst module is a fraction /P, with Hom
S
(M, N)
n
, P S(x),
P is homogeneous of degree n; if m/P
is an element of M
x
, (m)/PP
is an
element of N
x
which does not depend on /P and m/P
, and the function

m/P
(m)/PP
is a homomorphism
x
(/P) : M
x
N
x
; this denes
x
.
After Proposition 5 of n
14, Hom
Ox
(M
x
, N
x
) can be identied with:
Hom
O
(A(M), A(N))
x
;
this identication transforms
x
into:
x
: A(Hom
S
(M, N))
x
Hom
O
(A(M), A(N))
x
,
and we easily verify that the family of
x
is a homomorphism
: A(Hom
S
(M, N)) Hom
O
(A(M), A(N)).
When M is a free module of nite type,
x
is a bijection. Indeed, it suces
to regard M = S(n), for which it is obvious.
If now M is any graded S-module of nite type, choose a resolution of M:
... L
q+1
L
q
... L
0
M 0
where L
q
are free of nite type, and consider a complex C formed by Hom
S
(L
q
, N).
The cohomology groups of C are Ext
q
S
(M, N); or else if we denote by B
q
and
Z
q
the submodules of C
q
formed respectively by the coboundaries and cocycles,
we have the exact sequences:
0 Z
q
C
q
B
q+1
0
90
and
0 B
q
Z
q
Ext
q
S
(M, N) 0.
As the functor A(M) is exact, the sequences
0 Z
q
x
C
q
x
B
q+1
x
0
and
0 B
q
x
Z
q
x
Ext
q
S
(M, N)
S
0
are also exact.
But after preceding consideration C
q
x
is isomorphic to Hom
Ox
(L
q
x
, N
x
); the
Ext
q
S
(M, N)
x
are isomorphic to cohomology groups of a complex formed by the
Hom
Ox
(L
q
x
, N
x
) and, because the L
q
x
are clearly O
x
-free, we get back the de-
nition of Ext
q
Ox
(M
x
, N
x
), which shows (b). For q = 0 preceding considerations
show that
x
is bijection, so is an isomorphism, so (a) holds.
74 Vanishing of cohomology groups H
q
(X, F(n)) for n
+
Theorem 1. Let F be a coherent algebraic sheaf on X and let q be an integer
0. The following conditions are equivalent:
(a) H
q
(X, F(n)) = 0 for n large enough.
(b) Ext
rq
Ox
(F
x
, O
x
) = 0 for all x X.
After Theorem 2 of n
60, we can suppose that F = A(M), where M is a

graded S-module of nite type, and by the n
64 H
q
(X, F(n)) is isomorphic
to H
q
(M(n)) = B
q
(m)
n
, so condition (a) is equivalent to
T
q
(M)
n
= 0
for n large enough, that is to say T
q
(M) C. After Theorem 1 of n
72 and
the fact that M
C as M is of nite type, this last condition is equivalent to

Ext
rq
S
(M, ) C; as Ext
rq
S
(M, ) is a S-module of nite type,
Ext
rq
S
(M, ) C
is equivalent to Ext
rq
S
(M, )
x
= 0 for all x X, by Proposition 5 of n
58.
Finally the Proposition 1 shows that Ext
rq
S
(M, )
x
= Ext
rq
Ox
(M
x
,
x
) and as
M
x
is isomorphism to F
x
and
x
is isomorphic to O(r 1)
x
, so to O
x
, this
completes the proof.
For announcing Theorem 2, we will need the notion of dimension of an O
x
-
module. Recall ([6], Chap VI) that O
x
-module of nite type P is of dimension
p if there is an exact sequence of O-modules:
0 L
p
L
p1
... L
0
P 0,
91
where each L
p
is free (this denition is equivalent to [6], because all projective
O
x
-modules of nite type are free (cf [6], Chap VIII, Th. 6.1.).
All O
x
-modules of nite type are of dimension r, by Hilberts syzygy
theorem. (cf. [6], Chap VIII, Th. 6.2).
Lemma 1. Let P be an O
x
-module of nite type and let p be an integer 0.
The following two conditions are equivalent:
(i) P is of dimension p.
(ii) Ext
m
Ox
(P, O
x
) = 0 for all m > p.
It is clear that (i) implies (ii). We will show that (ii) implies (i) by induction
decreasing on p. For p r the lemma is trivial, because (i) is always true. Now
pass from p + 1 to p; let N be any O
x
-module of nite type. We can nd an
exact sequence 0 R L N 0, where L is free of nite type (because
O
x
is Noetherian). The exact sequence:
Ext
p+1
Ox
(P, L) Ext
p+1
Ox
(P, N) Ext
p+2
Ox
(P, R)
shows that Ext
p+1
Ox
(P, N) = 0, so we have Ext
p+2
Ox
(P, L) = 0 by condition (ii),
and Ext
p+2
Ox
(P, R) = 0 as dimP p + 1 by the induction hypothesis. As this
property characterizes the modules of nite dimension p, the lemma is proved.
By combining Lemma with Theorem 1 we obtain:
Theorem 2. Let F be a coherent algebraic sheaf on X, and let p be an
integer 0. The following two conditions are equivalent:
(i) H
q
(X, F(n)) = 0 for all n large enough and 0 q < p.
(ii) For all x X the O
x
-module F
x
is of dimension r p.
75 Nonsingular varieties
The following results play essential role in extension of the duality theorem
[15] to an arbitrary case.
Theorem 3. Let V be a nonsingular subvariety of projective space P
r
(K).
Suppose that all irreducible components of V have the same dimension p. Let
F be a coherent algebraic sheaf on V , such that for all x V , F
x
is a free
module over O
x,V
. Then we have H
q
(V, F(n)) = 0 for all n large enough and
0 q < p.
After Theorem 2, it remains to show that O
x,V
considered as O
x
-module is of
dimension r p. Denote by g
x
(V ) the kernel of the canonical homomorphism
: O
x
O
x,V
; since the point x is simple over V , we know (cf. [18], th 1)
92
that this ideal is generated by r p elements f
1
, ..., f
rp
, and the theorem of
Cohen-Macaulay (cf. [13], p. 53, prop 2) shows that we have
(f
1
, ..., f
i1
) : f
i
= (f
1
, ..., f
i1
) for 1 i r p.
Denote by L
q
a free O
x
-module which admits a base of elements e < i
1
...i
q
>
corresponding to sequence (i
1
, ..., i
q
) such that
1 i
1
< l
2
< ... < i
q
r p;
for q = 0, take L
0
= O
x
and dene:
d(ei
1
...i
q
) =
q
j=1
(1)
j
f
i,j
ei
1
, ...
i
j
...i
q
d(ei) = f
i
After [6], Chap. VIII, prop 4.3, the sequence
0 L
rp
d
L
rp1
d
...
d
L
0
x
O
x,V
0
is exact, which shows that dim
Ox
(O
x,V
) r p, QED.
Corollary. We have H
q
(V, O
V
(n)) = 0 for n large enough and 0 q < p.
Remark. The above proof applies more generally whenever the ideal g
x
(V )
admits a system of r p generators, that is, if the variety V is a local complete
intersection at all points.
76 Normal Varieties
Lemma 2. Let M be a O
x
module of nite type and let f be a noninvertible
element of O
x
, such that the relation fm = 0 implies m = 0 if m M. Then the
dimension of the O
x
-module M/fM is equal to the dimension of M increased
by one.
By assumption, we have an exact sequence 0 M

M M/fM 0,
where is multiplication by f. If N is a O
x
-module of nite type, we have an
exact sequence:
... Ext
q
Ox
(M, N)

Ext
q
Ox
(M, N) Ext
q+1
Ox
(M/fM, N) Ext
q+1
Ox
(M, N) ...
Denote by p the dimension of M. By taking q = p + 1 in the preceding
exact sequence, we see that Ext
p+2
Ox
(M/fM, N) = 0, which (by [6], Chap. VI,
2) implies that dim(M/fM) p + 1. On the other hand, since dimM = p we
can choose N such that Ext
p
Ox
(M, N) ,= 0; by taking q = p in the above exact
sequence, we see that
Ext
p+1
Ox
(M/fM, N) can be identied with cokernel of
Ext
p
Ox
(M, N)

Ext
p
Ox
(M, N)
93
as the last homomorphism is nothing else that multiplication by f and that f
isnt invertible in the local ring O
x
. If follows from [6], Chap. VIII, prop. 5.1
that this cokernel is ,= 0, which shows that dimM/fM p +1 and nishes the
proof.
We will now show a result, that is related with the Enriques-Severi lemma
of Zariski [19]:
Theorem 4. Let V be an irreducible, normal subvariety of dimension 2,
of projective space P
r
(K). Let F be a coherent algebraic sheaf on V , such that
for all x V , F
x
is a free module over O
x,V
. Then we have H
1
(V, F(n)) = 0
for n large enough.
After Theorem 2, it remains to show that O
x,V
, considered as O
x
-module is
of dimension r 2. First choose an element f O
x
such that f(x) = 0 and
that the image of f in O
x,V
is not zero; this is possible because dimV > 0. As
V is irreducible, O
x,V
is an integral ring (domain), and we can apply Lemma 2
to the pair (O, f); we then have:
dimO
x,V
= dimO
x,V
/(f) 1, with (f) = fO
x,V
.
As O
x,V
is an integrally closed ring, all prime ideals p
of the principal ideal

(f) are minimal (cf. [12] p.136, or [9], n
37), and none of them is equal to the

maximal ideal m of O
x,V
(if not we would have dimV 1). So we can nd an
element g m, not belonging to any of p
; this element g is not divisible by 0

in the quotient ring O
x,V
/(f); we denote by g, a representation of g in O
x
. We
see that we can apply Lemma to the pair O
x,V
/(f), g); we then have:
dimO
x,V
/(f) = dimO
x,V
/(f, g) 1.
But by Hilberts syzygy theorem, we have dimO
x,V
/(f, g) r, so dimO
x,V

r 1 and dimO
x,V
r 2 QED.
Corollary. We have H
1
(V, O
V
(n)) = 0 for n large enough.
Remarks.
(1) The reasoning made before is classic in theory of syzygies. Cf. W. Grobner,
Moderne Algebraische Geometrie, 152.6 and 153.1.
(2) If the dimension of V is > 2, we can have dimO
x,V
= r 2. This is
in particular the case when V is a cone which hyperplane section W is a
normal and irregular projective variety (i.e., H
1
(W, O
W
) ,= 0).
94
77 Homological characterization of varieties k-times of rst
kind
Let M be a graded S-module of nite type. We show by a reasoning identical
to that of Lemma 1:
Lemma 3 dim k if and only if Ext
q
S
(M, S) = 0 for q > k.
As M is graded. we have Ext
q
S
(M, ) = Ext
q
S
(M, S)(r1), so the previous
condition is equivalent to Ext
q
S
(M, ) = 0 for q > k. Given Theorem 1 of n
72, we conclude:
Proposition 2.
(a) For dimM r it is necessary and sucient that M
n
H
0
(M(n)) is
injective for all n Z.
(b) If k is an integer 1, for dimM r k it is necessary and sucient that
: M
n
H
0
(M(n)) is bijective for all n Z, and that H
q
(M(n)) = 0 for
0 < q < k and all n Z.
Let V be a closed subvariety of P
r
(K), and let I(V ) be an ideal if homoge-
neous polynomials, which are zero on V .
Denote S(V ) = S/I(V ), this is a graded S-module whose associated sheaf is
O
V
. We say
2
that V is a variety k-times of rst kind of P
r
(K) if the dimension
of S-module S(V ) is r k. It is obvious that : S(V )
n
H
0
(V, O
V
(n)) is
injective for all n Z, so all varieties are 0-times of rst kind. Using preceding
proposition to M = S(V ), we obtain:
Proposition 3. Let k be an integer 1. For a subvariety V to be a k-
times of rst kind, it is necessary and sucient that the following conditions
are satised for all n Z:
(i) : S(V )
n
H
0
(V, O
V
(n)) is bijective.
(ii) H
q
(V, O
V
(n)) = 0 for 0 < q < k.
(The condition (i) can also be expressed by saying that linear series cut on
V by forms of degree n is complete, which is well known.)
By comparing with Theorem 2 (or by direct reasoning), we obtain:
Corollary. If V is k-times of rst kind, we have H
q
(V, O
V
) = 0 for 0 <
q < k and, for all x V , the dimension of O
x
-module O
x,V
is r k.
2
Cf. P. Dubreil, Sur la dimension des ideaux de polyn omes, J . Math. Pures App., 15,
1936, p. 271-283. See also W . Grobner, hloderne Algebraische Geometrie, 5.
95
If mis an integer 1, denote by
m
the embedding of P
r
(K) into a projective
space of convenient dimension, given by the monomials of degree m (cf. [8],
Chap. XVI, 6, or n
52, proof of Lemma 2). So the preceding corollary admits

following converse:
Proposition 4. Let k be an integer 1, and let V be a connected and closed
subvariety of P
r
(K). Suppose that H
q
(V, O
V
) = 0 for 0 < q < k, and that for
all x V the dimension of O
x
-module O
x,V
is r k.
Then for all m large enough,
m
(V ) is a subvariety k-times of rst kind.
Because V is connected, we have H
0
(V, O
V
) = K. So, if V is irreducible,
its evident (if not, H
0
(V, O
V
) contains a polynomial algebra and is not of nite
dimension over K); if V is reducible, all elements f H
0
(V, O
V
) induce a
constant on each of irreducible components of V , and this constants are the
same, because of connectivity of V .
By the fact that dimO
x,V
r 1, the algebraic dimension of each of irre-
ducible components of V is at least equal to 1. So it follows that
H
0
(V, O
V
(n)) = 0
for all n > 0 (because if f H
0
(V, O
V
(n)) and f ,= 0, the f
k
g with g S(V )
nk
form a vector subspace of H
0
(V, O
V
) of dimension > 1).
That being said, denote by V
m
the subvariety
m
(V ); we obviously have:
O
Vm
(n) = O
V
(nm).
For m large enough the following conditions are satised:
(a) : S(V )
nm
H
0
(V, O
V
(nm)) is bijective for all n 1.
This follows from Proposition 5 of n
65.
(b) H
q
(V, O
V
(mn)) = 0 for 0 < q < k and for all n 1.
This follows from Proposition of n
65.
(c) H
q
(V, O
V
(nm)) = 0 for 0 < q < k and for all n 1.
This follows from Theorem 2 of n
74, and hypothesis made on O

x,V
.
On the other hand, we have H
0
(V, O
V
) = K, H
0
(V, O
V
(nm)) = 0 for all
n 1, and H
q
(V, O
V
) = 0 for 0 < q < k, by the hypothesis. It follows that
V
m
satises all the hypothesis of Proposition 3, QED.
Corollary. Let k be an integer 1, and let V be a projective variety with-
out singularities, of dimension k. For V being birationally isomorphic to a
subvariety k-times of rst kind of a convenient projective space, it is necessary
and sucient that V is connected and that H
q
(V, O
V
) = 0 for 0 < q < k.
The necessity is evident, by Proposition 3. To show suciency, it suces to
remark that O
x,V
is of dimension r k (cf. n
75) and to apply the previous

proposition.
96
78 Complete intersections
A subvariety V of dimension p of projective space P
r
(K) is a complete inter-
section if the ideal I(V ) of polynomials zero at V admits a system of r p
generators P
1
, ..., P
rp
; in this case, all irreducible components of V have the
dimension p, by the theorem of Macaulay (cf. [9], n
17). It is known, that

this variety is p-times of rst kind, which implies that H
q
(V, O
V
(n)) = 0 for
0 < q < p, as we have just seen. We will determine H
p
(V, O
V
(n)) as a function
of degree m
1
, ..., m
rp
of homogeneous polynomials P
1
, ..., P
rp
.
Let S(V ) = S/I(V ) be a ring of projective coordinates of V . By theorem 1
of n
72 all it is left, is to determine the S-module Ext

rp
S
(S(V ), ). We have a
resolution, analogous to that of n
75: we take L
q
the graded free S-module, ad-
mitting for a base the elements ei
1
, ..., i
q
, corresponding to sequences (i
1
, ..., i
q
)
such that 1 i
1
< i
2
< ... < i
q
r p and of degree
q
j=1
m
j
; for L
0
we take
S. We set:
d(ei
1
, ..., i
q
) =
q
j=1
(1)
i
P
ij
ei
1
...
i
j
...i
q
d(ei) = P
i
.
The sequence 0 L
rp
d
...
d
L
0
S(V ) 0 is exact ([6], Chap.
VIII, Prop. 4.3). It follows that the Ext
q
S
(S(V ), ) are the cohomology groups
of the complex formed by the Hom
S
(L
q
, ); but we can identify an element of
Hom
S
(L
q
, )
n
with a system fi
1
, ...i
q
, where the fi
1
, ..., i
q
are homogeneous
polynomials of degree m
i1
+... +m
iq
+nr1; after this identication is made,
the operator of coboundary is given by usual formula:
(df)i
1
...i
q+1
=
q
j=1
(1)
j
P
ij
fi
1
...
i
j
...i
q+1
.
The theorem of Macaulay implies that we are in conditions of [11], and we ob-
tain that Ext
q
S
(S(V ), ) = 0 for q ,= rp. On the other hand, Ext
rp
S
(S(V ), )
n
is isomorphic to a vector subspace of S(V ) formed by homogeneous elements
of degree N + n, where N =
rp
i=1
m
i
r 1. Using Theorem 1 of n
72 we
obtain:
Proposition 5. Let V be a complete intersection, dened by the homoge-
neous polynomials P
1
, ..., P
rp
of degrees m
1
, ..., m
rp
.
(a) The function : S(V )
n
H
0
(V, O
V
(n)) is bijective for all n Z.
(b) H
q
(V, O
V
(n)) = 0 for 0 < q < p and all n Z.
(c) H
q
(V, O
V
(n)) is isomorphic to a dual vector space to H
0
(V, O
V
(N n)),
with N =
rp
i=1
m
i
r 1.
We see that in particular H
p
(V, O
V
) is zero if N < 0.
97
6. The characteristic function and arithmetic genus III
6 The characteristic function and
arithmetic genus
79 Euler-Poincare characteristic
Let V be a projective variety and F a coherent algebraic sheaf on V . Let
h
q
(V, F) = dim
K
H
q
(V, F).
We have seen (n
66, Theorem 1) that h

q
(V, F) are nite for all integer q
and zero for q > dimV . So we can dene an integer (V, F) by:
(V, F) =
q=0
(1)
q
h
q
(V, F).
This is the Euler-Poincare characteristic of V with coecient in F.
Lemma 1. Let 0 L
1
... L
p
0 be an exact sequence, with L
i
being
nite dimensional vector spaces over K, and homomorphisms L
i
L
i+1
being
K-linear. Then we have:
p
q=1
(1)
q
dim
K
L
q
= 0.
We proceed by induction on p. The lemma is evident if p 3. If L
p1
is the
kernel of L
p1
L
p
, we have two exact sequences:
0 L
1
... L
p1
0
0 L
p1
L
p1
L
p
0.
Applying induction hypothesis to each sequence, we see that
p2
q=1
(1)
q
dimL
q
+
(1)
p1
dimL
p1
= 0, and
dimL
p1
dimL
p1
+dimL
p
= 0,
which proves the lemma.
Proposition 1. Let 0 A B C 0 be an exact sequence of coherent
algebraic sheaves on a projective variety V , with homomorphisms A B and
B C being K-linear. Then we have:
(V, B) = (V, A) +(V, C).
By Corollary 2 of Theorem 5 of n
47, we have an exact sequence of coho-

mology:
... H
q
(V, B) H
q
(V, C) H
q+1
(V, A) H
q+1
(V, B) ...
98
Applying Lemma to this exact sequence of vector spaces we obtain the
Proposition.
Proposition 2. Let 0 F
1
... F
p
0 be an exact sequence
of coherent algebraic sheaves on a projective variety V , with homomorphisms
F
i
F
i+1
being algebraic. Then we have:
p
q=1
(1)
q
(V, F
q
) = 0.
We proceed by induction on p. The proposition is a particular case of Propo-
sition 1 if p 3. If we dene F
p1
to be the kernel of F
p1
F
p
, the sheaf
F
p1
is coherent algebraic because F
p1
F
p
is an algebraic homomorphism.
So we can applicate the induction hypothesis to two exact sequences
0 F
1
... F
p1
0
0 F
p1
F
p1
F
p
,
and the Proposition follows.
80 Relation with characteristic function of a graded S-
module
Let F be a coherent algebraic sheaf on the space P
r
(K). We write (F) instead
of (P
r
(K), F). We have:
Proposition 3. (F(n)) is a polynomial of n of degree r.
By Theorem 2 of n
60, there exists a graded S-module M of nite type,

such that A(M) is isomorphic to F. Applying the Hilberts syzygy theorem to
M we obtain an exact sequence of graded S-modules:
0 L
r+1
... L
0
M 0,
where L
q
are free of nite type. Applying the functor A to this sequence, we
obtain an exact sequence of sheaves:
0 L
r+1
... L
0
F 0,
where each L
q
is isomorphic to a nite direct sum of shaves O(n
i
). The propo-
sition 2 implies that (F(n)) is equal to an alternating sum of (L
0
(n)), which
brings us to case of the sheaf O(n
i
). Now it follows from n
62 that we have
(O(n)) =
_
n+r
r
_
, which is a polynomial on n of the degree r. This implies
the Proposition.
Proposition 4. Let M be a graded S-module satisfying condition (TF), and
let F = A(M). For all n large enough, we have (F(n)) = dim
K
M
n
.
99
We know (by n
65) that for n large enough, the homomorphism : M

n

H
0
(X, F(n)) is bijective, and H
q
(X, F(n)) = 0 for q > 0. So we have:
(F(n)) = h
0
(X, F(n)) = dim
K
M
n
.
We use a well known fact, that dim
K
M
n
is a polynomial of n for n large
enough. This polynomial, which we denote by P
M
is called the characteristic
function of M. For all n Z we have P
M
(n) = (F(n)), and in particular for
n = 0, we see that the constant term of P
M
is equal to (F).
Apply this to M = S/I(V ), I(V ) being a homogeneous ideal of S of poly-
nomials which are zero on a closed subvariety V of P
r
(K). The constant term
of P
M
is called in this case the arithmetic genus of V (cf. [19]). Since on the
other hand we have A(M) = O
V
, we obtain:
Proposition 5. The arithmetic genus of a projective variety V is equal to
(V, O
V
) =
q=0
(1)
q
dim
K
H
q
(V, O
V
).
Remarks.
(1) The preceding Proposition makes evident the fact, that the arithmetic genus
is independent of an embedding of V into a projective space, since its true
for H
q
(V, O).
(2) The virtual arithmetic genus (dened by Zariski in [19]) can also be re-
duced to Euler-Poincare characteristic. We return to this question later, by
Riemann-Roch theorem.
(3) For the reason of convenience, we have adopted the denition of arithmetic
genus dierent from the classical one (cf. [19]). If all irreducible compo-
nents of V have the same dimension p, two denitions are related by the
following formula: (V, O
V
) = 1 + (1)
p
p
a
(V ).
81 The degree of the characteristic function
If F is a coherent algebraic sheaf on an algebraic variety V , we call the support
of F, and denote by Supp(F), the set of points x V such that F
x
,= 0. By
the fact that F is a sheaf of nite type, this set is closed. If we have F
x
= 0,
the zero section generates F
x
, then also F
y
for y in neighborhood of x (n
12,
Proposition 1), which means that the complement of Supp(F) is open.
Let M be a graded S-module of nite type, and let F = A(M) be a sheaf
dened by M on P
r
(K) = X. We can determine Supp(F) from M in the
following manner:
100
Let 0 =
be a decomposition of 0 as an intersection of homogeneous

primary submodules M
of M. M
correspond to homogeneous primary ideals

p
(cf. [12], Chap. IV). We suppose that this decomposition is the shortest
possible, i.e. that non of M
is contained in an intersection of others. For

all x X, each p denes a primary ideal p
x
of a local ring O
x
, and we have
p
x
= O if and only if x is not an element of a variety V
dened by an ideal p
.
We have also 0 =
x
in M
x
, and we verify easily that we thereby obtain a
primary decomposition of 0 in M
x
. The M
x
correspond to primary ideals p
x
;
if x / V
, we have M
x
= M
x
, and if we restrict ourself to consider M
x
such
that x V
, we obtain the shortest possible decomposition (cf. [12], Chap

IV, th 4.). We conclude that M
x
,= 0 if and only if x is an element of V
, thus
Supp(F) =
.
Proposition 6. If F is a coherent algebraic sheaf on P
r
(K), the degree of
(F(n)) is equal to the dimension of Supp(F).
We proceed by induction on r. The case r = 0 is trivial. We can suppose
that F = A(M), where M is a graded S-module of nite type. Using notation
introduced below, we have to show that (F(n)) is an polynomial of degree
q = Sup dim V
.
Let t be a linear homogeneous form, which do not appear in any of proper
prime ideals p
. Such a form exists because the eld K is innite. Let E be a

hyperplane of X with equation t = 0. Consider the exact sequence:
0 O(1) O O
K
0,
where O O
E
is a restriction homomorphism, while O(1) O is a ho-
momorphism f tf. Applying tensor product with F, we obtain an exact
sequence:
F(1) F F
E
0, with F
E
= F
O
O
E
.
On U
i
, we can identify F(1) with F, and this identication transforms
the homomorphism F(1) F dened above to the multiplication by t/t
i
.
Because t was chosen outside p
, t/t
i
dont belong to any prime ideal of M
x
=
F
x
if x U
i
, and the preceding homomorphism is injective (cf. [12], p. 122, th.
7, b)). So we have an exact sequence:
0 F(1) F F
E
0,
from which, for all n Z the exact sequence:
0 F(n 1) F(n) F
K
(n) 0.
Applying Proposition 1, we see that:
(F(n)) (F(n 1)) = (F
E
(n)).
101
But the sheaf F
E
is a coherent sheaf of O
E
-modules, which means that it
is a coherent algebraic sheaf on E, which is a projective space of dimension
r 1. Moreover F
x,E
= 0 means that the endomorphism of F
x
dened by
multiplication by t/t
i
is surjective, which leads to F
x
= 0 (cf. [6], Chap VIII,
prop 5.1). It follows that Supp(F
K
) = E Supp(F), and because E does
not contain any of varieties V
, if follows by a known fact, that the dimension

of Supp(F
E
) is equal to q 1. By the induction hypothesis (F
E
(n)) is a
polynomial of degree q 1. As this dierence is prime to the function (F(n)),
the latter is a polynomial of degree q.
Remarks.
(1) Proposition 6 was well known for F = O/I, I being a coherent sheaf of
ideals. Cf. [9] n
24.
(2) The above proof does not use Proposition 3 and shows it once again.
102
Bibliography
[1] N. Bourbaki, Theorie des Ensembles. Paris (Hermann).
[2] H. Cartan, Seminaire E.N.S., 1950-1951.
[5] H. Cartan, Varietes analytiques complexes et cohomologie. Colloque de
Bruxelles, (1953), p. 41-55.
[6] H. Cartan and S. Eilenberg, Homological Algebra. Princeton Math.
Ser., n
19.
[7] S. Eilenberg and N. E. Steenrod, Foundations of Algebraic Topology,
Princeton Math. Ser., n
15.
[8] W. V. D. Hodge and D. Pedoe, Methods of Algebraic Geometry, Cam-
bridge Univ. Press.
[9] W. Krull, Idealtheorie, Ergenbisse IV-3. Berlin (Springer).
[10] J. Leray, Lanneau spectral et lanneau ltre dhomologie dun espace lo-
calement compact et dune application continue, J. Math. Pures App., 29,
(1950), p. 1-139.
[11] G. de Rham, Sur la division de formes et de courants par une forme
lineaire, Comment. Math. Helv., 28, (1954), p. 346-352.
[12] P. Samuel, Commutative Algebra (Notes by D. Hertzig), Cornell Univ.,
1953.
[13] P. Samuel, Alg`ebre locale, Mem. Sci. Math., CXXIII, Paris, 1953.
[14] J-P. Serre, Groupes dhomotopie et classes de groupes abeliens, Ann. of
Math., 58, (1953), p. 258-294.
[15] J-P. Serre, Un theor`eme de dualite, Comment. Math. Helv., 29, (1955),
p. 9-26.
103
BIBLIOGRAPHY BIBLIOGRAPHY
[16] A. Weil, Foundations of Algebraic Geometry, Colloq. XXIX.
[17] A. Weil, Fibre-spaces in Algebraic Geometry (Notes by A. Wallace),
Chicago Univ., 1952.
[18] O. Zariski, The concept of a simple point of an abstract algebraic variety,
Trans. Amer. Math. Soc., 62, (1947), p. 1-52.
[19] O. Zariski, Complete linear systems on normal varieties and a generaliza-
tion of a lemma of Enriques-Severi, Ann. of Math., 55, (1952), p. 552-592.
104

Coherent Algebraic Sheaves: Jean-Pierre Serre

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Coherent Algebraic Sheaves: Jean-Pierre Serre

Загружено:

Авторское право:

Доступные форматы

Coherent Algebraic Sheaves

56 (S denotes the polynomial

54); the results obtained

, bijective when it is both injective and surjective. An injective (resp.

3: if U is an open subset of X, let F

of F over an open neighborhood U of x for which s

(y) = s(y) for all y V Y ; since F

(y) = s(y) for y V (V Y ); hence s and s

coincide on V , which proves that s is continuous in a neighborhood of Y , so it is

1 is that of a sheaf of abelian groups. It

1). We shall always

6, and take for every open

; this homomorphism is denoted by ; if is the identity, we

are homomorphisms, we dene in an obvious

gives rise to an exact sequence:

4). The kernel R(s

11, is bijective. First of all, let : F G be a homomorphism

37, Proposition 1).

5). If A is a sheaf of rings over Y , A

of dimension q such that [s

denoting the restriction homomorphism: (U

[ [s[. The mapping s (

(U, F) which forms a subcomplex of C(U, F). The cohomology

(U, F) are denoted by H

(U, F) in C(U, F) induces an isomor-

be two mappings from I to J such that U

whose set of indices is

the set of open subsets of X

coincides with the homomorphism from (X, F) to (X, G)

satisfy usual formal properties:

47 that this is the case when X is an algebraic variety

; thus the homomorphisms

dene, by passing to the limit with U, homomorphisms

for some covering U of

satisfying the following properties:

satises d d = 0, and the

as a coboundary operator on K; since d

, we have the groups H

a q-simplex of S(J), we denote by U

18), the intersection of V

; the groups or complexes denoted in the general case by H

the canonical isomorphism: C(V, F) C

and all p > 0. Then

, is bijective for all

be a covering of Y consisting of a single open set Y ; we obviously

, and the assumption made on W means that W

(f) are cohomologous in C(U, V; F).

(f), which shows that

(f) are cohomologous.

are arbitrarily ne,

, F), and the theorem follows.

Z), which shows that each V

. Moreover, one easily shows that X is

3); the sheaf O is a sheaf of

be the complement of W, which is

, there exists, by the denition of the Zariski topology,

and nonzero at x; the polynomial P P

are two nonnegative integers, we identify the ane space K

. The Zariski topology on K

is ner than the product

; it is even strictly ner if r and r

are positive. In result, if U and U

are locally closed subspaces of K