SGA 4-I — Expose II: Topologies and Sheaves
Translated from the cleaned OCR transcription of the image-only SGA 4-I scan. Page markers refer to PDF pages. Mathematical notation should still be checked against the scan.
[PDF p. 234]
Expose II — Topologies and Sheaves
By J.-L. Verdier.
After the definition of topologies and pretopologies (§1) and of sheaves of sets (§2), §3 takes up the central theorem of this expose: the existence of the sheaf associated to a presheaf (3.4). In order to cover all cases encountered in practice, the existence of this functor is proved for presheaves on a -site (3.0.2).
Sections 4 and 5 draw the consequences of this theorem for the behavior of inductive and projective limits in categories of sheaves. In §6, one defines and studies sheaves with values in arbitrary categories, immediately turning attention to sheaves of rings, groups, modules, and so on.
1. Topologies. Covering Families. Pretopologies
Definition 1.1. A topology on a category is the datum, for every object of , of a set of sieves of , subject to the following axioms:
- (T1) Stability under base change. For every object of , every sieve , and every morphism , with , the sieve of belongs to .
- (T2) Local character. If and are two sieves of , if , and if, for every and every morphism , the sieve of belongs to , then belongs to .
[PDF p. 235]
- (T3) For every object of , belongs to .
1.1.1. The sieves belonging to will be called the sieves covering , or also the refinements of . From axioms (T1), (T2), and (T3), one immediately deduces that the set of sieves covering is stable under finite intersections, and that every sieve containing a covering sieve is a covering sieve. The set , ordered by inclusion, is therefore cofiltered (I 8).
1.1.2. Let be a category, and let and be two topologies on . The topology is said to be finer than the topology if, for every object of , every refinement of for the topology is a refinement of for the topology . In this way one defines an order structure on the set of topologies.
1.1.3. Let be a family of topologies on . For every object of , the sieves of that are covering for all the topologies verify axioms (T1), (T2), and (T3), and therefore define a topology: the intersection topology of the , that is, the lower bound of the . It is the finest of the topologies that is less fine than all the topologies . Consequently the family admits an upper bound: the intersection topology of the topologies finer than each of the .
[PDF p. 236]
1.1.4. In particular, the datum equal to the set of all sieves of is a topology finer than every topology on ; we shall call it the discrete topology.
There exists a topology less fine than every topology on : the topology for which for every of . This topology is called the coarse or chaotic topology.
1.1.5. A category equipped with a topology is called a site. The category is called the category underlying the site.
Definition 1.2. Let be a site and let be an object of . A family of morphisms is said to be covering if the sieve generated by the family (I 4.3.3) is a sieve covering .
1.1.6. Let be a category. Suppose we are given, for each object of , a set of families of morphisms with target . Then there exists a topology that is the least fine of the topologies for which the given families are covering, namely the intersection (1.1.3) of all the topologies in question. This topology is called the topology generated by the given sets of families of morphisms. It is difficult, in general, to describe all covering sieves of this topology. However the situation is more pleasant in the following case.
Definition 1.3. Let be a category. A pretopology on is the datum, for each object of , of a set of families of morphisms with target , subject to the following axioms:
[PDF p. 237]
- (PT0) For every object of , the morphisms in the families of morphisms of are squarable. Recall that a morphism is said to be squarable if, for every morphism , the fiber product is representable.
- (PT1) Stability under base change. For every object of , every family belonging to , and every morphism , with , the family belongs to .
- (PT2) Stability under composition. If belongs to , and if for each a family belongs to , then the composite family belongs to .
- (PT3) The family reduced to the identity morphism belongs to .
1.3.1. Definition 1.1.6 allows us, for a given pretopology on , to consider the topology on generated by this pretopology. Note that if is a category in which fiber products are representable, then every topology on can be defined by a pretopology, namely the one for which is formed by all families covering for the topology .
Proposition 1.4. Let be a category, a pretopology on , the topology defined by the pretopology (1.1.6), and an object of . Denote by the set of sieves of generated by families of the pretopology, and by the set of refinements of for
[PDF p. 238]
the topology . Then is cofinal in . In other words, for a sieve of to belong to , it is necessary and sufficient that there exist a sieve such that .
Proof. Let , for every object of , be the set of sieves of that contain a sieve of . Evidently . To show that , it suffices to show that the datum of the defines a topology on . Now the evidently verify axioms (T1) and (T3). It remains to verify (T2).
For this it suffices to prove that, if is a subsieve of such that for every the sieve of belongs to , then the sieve belongs to . But the sieve is generated by a family belonging to , and for every the sieve contains a sieve generated by a family belonging to . We deduce that the sieve contains a sieve generated by the composite family . Therefore, by axiom (PT2), contains a sieve of , and consequently belongs to . QED.
[PDF p. 238]
2. Sheaves of Sets
Definition 2.1. Let be a site whose underlying category is a -category. A presheaf with values in -Ens is said to be separated (respectively is a sheaf) if for every sieve covering , an object of , the map
is injective (respectively bijective). The full subcategory of
[PDF p. 239]
whose objects are the sheaves is called the category of sheaves of sets on , and is most often denoted (1). When no ambiguity results, we shall simply say category of sheaves on ; by contrast, we shall sometimes be more precise and say category of sheaves with values in -Ens, or category of -sheaves.
Proposition 2.2. Let be a -category, and let be a family of -presheaves on . For each object of , denote by the set of sieves such that for every morphism of with target , the sieve has the following property: the map
is bijective (respectively injective) for every . Then the sets define a topology on , which is the finest of the topologies for which each is a sheaf (respectively a separated presheaf).
Proof. The evidently verify axioms (T1) and (T3). It remains to show that they verify (T2). For this it suffices to show that:
-
If are two sieves of , with and such that for every (where is an object of ) the sieve belongs to , then the sieve belongs to .
-
If are two sieves of such that , then the sieve belongs to .
Note that, in case 2), for every (where is an object of ) the sieve belongs to . Now, in cases 1) and 2), is the inductive limit of the , objects of , over (I 3.4).
(1) or , depending on the mood of the typewriter.
[PDF p. 240]
Inductive limits in are universal (I 3.3). Consequently, in cases 1) and 2), is the inductive limit of the . Now, in cases 1) and 2), belongs to . Passing to the inductive limit over the objects of over , one deduces that the map
is a bijection (respectively an injection) for every . Consequently the maps
are, in cases 1) and 2), bijections (respectively injections).
Moreover, hypotheses 1) and 2) are visibly stable under arbitrary base change ( an object of ). It follows that for every object of over and every , the maps
are, in both cases, bijections (respectively injections); this shows that and belong to . QED.
Corollary 2.3. Let be a -category, and for every of let be a set of sieves of . Suppose that the verify axiom (T1) of (1.1). For a presheaf to be a sheaf (respectively a separated presheaf) for the topology generated (1.1.6) by the , it is necessary and sufficient that for every object of and every sieve , the map
be a bijection (respectively an injection).
[PDF p. 241]
Corollary 2.4. In particular, let be a -category equipped with a pretopology. For a presheaf to be a sheaf (respectively a separated presheaf), it is necessary and sufficient that for every object of and every family belonging to , the diagram of sets
be exact (respectively the map
be injective).
Proof. Apply 2.3, then (I 3.5) and (I 2.12).
With Corollary 2.4 one recovers the definition given in [1].
Definition 2.5. Let be a -category. The canonical topology of is the finest topology for which the representable functors are sheaves (2.2). A sieve covering for the canonical topology will be called a universally strict epimorphic sieve. A family covering for the canonical topology will be called a universally strict epimorphic family. When, moreover, the morphisms of the covering family are squarable, the family will be said to be universally effective epimorphic [2].
Remark 2.5.1. For almost all sites that have had to be used up to now, the topology is less fine than the canonical topology; in other words, the representable functors on are sheaves, i.e. the covering families of are universally strict epimorphic.
[PDF p. 242]
The only exception to this rule is the site studied in no. 5 (whose topology is finer, and most often strictly finer, than the canonical topology).
Proposition 2.6. For a sieve to be universally strict epimorphic, it is necessary and sufficient that, for every object of over and every object of , the map
be a bijection.
Proof. Immediate by applying 2.2 and then (I 5.3).
Remarks 2.7.
-
Proposition 2.6 gives a characterization of the universally strict epimorphic sieves of a category , independent of the universe in which the presheaves take their values, with the sole condition that the morphism sets of belong to this universe. It thus allows one to define the canonical topology for every category , without needing to specify universes.
-
Let be a site whose underlying category is a -category, let be a -presheaf, and let be a universe containing . The category underlying is a -category, and may be considered as a -presheaf. The -presheaf is a sheaf (respectively a separated presheaf) if and only if the -presheaf is a sheaf (respectively a separated presheaf). In other words, the condition that be a sheaf (respectively a separated presheaf) does not depend (in the sense just
[PDF p. 243]
specified) on the universe in which the presheaf takes its values. In particular, let be a site and a -presheaf; one says that is a sheaf (respectively a separated presheaf) if there exists a universe containing such that the category is a -category and such that is a -sheaf (respectively a separated -presheaf). This property does not depend on the universe .
3. Sheaf Associated to a Presheaf
Definition 3.0.1. Let be a site. A topologically generating family (or, when no confusion results, generating family) of is a set of objects of such that every object of is the target of a covering family of morphisms of (1.2) whose sources are elements of .
Definition 3.0.2. Let be a universe. A -site is a site whose underlying category is a -category (I 1.1), which possesses a small topologically generating family. Let be a -category; a -topology on is a topology on making into a -site. A site is said to be -small, or by abuse of language small, if the category underlying is small (I 1.0).
3.0.3. It follows immediately from the definitions that every topology finer than a -topology is a -topology, and that a small site is a -site.
[PDF p. 244]
Proposition 3.0.4. Let be a -site, and let be a small topologically generating family of . For every , denote by the set of sieves covering generated by a family of morphisms
- The set is small.
- The set is cofinal in the set of all sieves covering , ordered by inclusion.
- For every , there exists a small epimorphic family (I 10.2)
Proof. 1) Put . The set is small (I 0) and . Consequently is small.
-
Let . Put and let be the sieve of generated by the family , . We have , and it suffices to show that is covering. By axiom (T2) of 1.1, it suffices to show that for every morphism , , the sieve of is covering. But the sieve contains the sieve generated by the family of all morphisms with , a family which is covering by hypothesis. The sieve of is therefore covering (axiom T2 of 1.1).
-
Let . The family , , is epimorphic by hypothesis. Now, for every , is contained in , which is small. Consequently is small.
3.0.5. Let be a -site, let be a universe such that and , and let be a -small topologically generating family of . The category of presheaves of -sets on is a -category (I 1.1.1). Let
[PDF p. 245]
be an object of . The set of sieves covering is -small and, ordered by inclusion, is cofiltered (1.1.1). For every -presheaf , the inductive limit
is therefore representable by an element of (I 2.4.1). Moreover, it follows from (3.0.4 3)) that, for every , is -small; and since is a -small set cofinal in (loc. cit.), it follows from (I 2.4.2) that is -small. Choose then, for every and every , an element of representing this inductive limit and put
Let be a morphism of . The base-change functor defines a map
which makes into a presheaf on .
Since the morphism is an element of , for every object of we have a map
which visibly defines a morphism of functors
It is moreover clear that is a functor in and that the morphisms define a morphism
Finally let be a refinement of . The definition of and (I 1.4) provide a map
[PDF p. 246]
and for every morphism of , , the definition of the functor shows that the diagram below is commutative:
Z_R: Hom_{C^}(R, F) -> Hom_{C^}(X, LF)
| |
v v
Z_{R x_X Y}: Hom_{C^}(R x_X Y, F) -> Hom_{C^}(Y, LF)
where the vertical arrows are the evident arrows. We now copy [2].
Lemma 3.1.
-
For every refinement and every , the diagram
F --ell(F)--> LF ^ ^ | u | Z_R(u) R --i_R----> Xis commutative.
-
For every morphism , there exists a refinement of and a morphism such that .
-
Let be an object of and let be two morphisms such that . Then the kernel of the pair is a refinement of .
-
Let and be two refinements of , and let and be two morphisms. For , it is necessary and sufficient that and coincide on a refinement .
Proof. The only nontrivial assertion is assertion 1). We must show that . For this it suffices to show (I 3.4) that the composites of these morphisms with every morphism ( an object
[PDF p. 247]
of ) are equal. Now consider and the fiber product :
F --ell(F)--> LF
^ ^
| u | Z_R(u)
R --i_R----> X
^ ^
| g' | f
R x_X Y --i'-> Y
In the above diagram, is a monomorphism that admits a section. Consequently (I 3.3) is an isomorphism. Now by definition of the morphism , the morphism is equal to . On the other hand, the commutativity of diagram (*) gives the equality , and consequently we indeed have the equality .
Proposition 3.2.
- The functor is left exact (I 2.3.2).
- For every presheaf , is a separated presheaf.
- The presheaf is separated if and only if the morphism is a monomorphism. The presheaf is then a sheaf.
- The following properties are equivalent:
i) is an isomorphism.
ii) is a sheaf.
Proof. 1) It suffices to show (I 3.1) that, for every object of , the functor commutes with finite projective limits. But, by definition of
[PDF p. 248]
the projective limit, the functor
commutes with projective limits, and the inductive limit commutes with finite projective limits because is a cofiltered set (I 2.7).
-
Let be an object of , and let be two morphisms that coincide on a refinement of . By virtue of 3.1 2), there exists a covering sieve , which one may always suppose contained in , and two morphisms such that and . By virtue of 3.1 1), we then have . Consequently (3.1 4)) and coincide on a refinement . Let be the restriction of to . We have , and consequently . The presheaf is therefore separated.
-
If is separated, the morphism is a monomorphism because a filtered inductive limit of monomorphisms is a monomorphism. If is a monomorphism, the presheaf is a subpresheaf of a separated presheaf, hence is separated. We show that is then a sheaf. Let be a sieve covering an object of , and let be a morphism. It suffices to show that factors through . Put and :
F --ell(F)--> LF
^ ^ \ v
| pr_1 | u
R' --pr_2---> R --i--> X
^ ^
| p_1 | m
R'' --p_2---> Y
[PDF p. 249]
To show that , it suffices to show (I 3.4) that for every morphism ( an object of ), . Put and let and be the projections. The projection is a monomorphism and makes into a sieve covering (3.1 2)). We have (3.1 1)), and therefore . We deduce , i.e. . Since the presheaf is separated, we have , QED.
- Clear.
Remark 3.3. Let be a cofinal subset of . We have
In particular, if the topology of is defined by a pretopology (1.1.3), the functor can be described using the covering families of (I 2.12 and I 3.5). By spelling out the formulas one recovers the construction of [2].
Theorem 3.4. Let be a -site. The inclusion functor from sheaves into presheaves admits a left adjoint , left exact (I 2.3.2):
C^ --a--> C~
C~ --i--> C^
The functor is canonically isomorphic to the functor (cf. 3.0.5). For every presheaf , the adjunction morphism is obtained, by the preceding isomorphism, from the morphism .
[PDF p. 250 duplicates PDF p. 249 in the source scan.]
[PDF p. 251]
Definition 3.5. The sheaf is called the sheaf associated to the presheaf .
Theorem 3.4 follows immediately from Proposition 3.2.
Proposition 3.6. Let be a -site and a universe. Denote by and (respectively and ) the categories of -presheaves and -sheaves (respectively of -presheaves and -sheaves), and by (respectively ) the corresponding “associated sheaf” functors. The diagram
C^_U --a_U--> C~_U
| |
v v
C^_V --a_V--> C~_V
where the vertical functors are the canonical inclusions, is commutative up to canonical isomorphism.
Proof. This follows from the construction of the functors and (3.4 and 3.0.5).
4. Exactness Properties of the Category of Sheaves
The exactness properties of the category of sheaves are deduced from the exactness properties of the category of presheaves via Theorem 3.4. The present section makes this philosophy explicit in typical statements among the most useful ones.
Theorem 4.1. Let be a -site, the category of sheaves, the associated sheaf functor, and the inclusion functor.
- The functor commutes with inductive limits and is exact.
[PDF p. 252]
-
-inductive limits in are representable. For every small category and every functor , the canonical morphism
is an isomorphism.
-
-projective limits in are representable. For every object of , the functor on
commutes with projective limits, i.e. the inclusion functor commutes with projective limits.
Proof. These properties result essentially from Theorem 3.4 and from (I 2.11).
Thus, in the category of sheaves, products indexed by an element of , fiber products, sums indexed by an element of , amalgamated sums, kernels, cokernels, images, and coimages are representable.
Corollary 4.1.1. Let be a -site and a sheaf of sets on . The canonical homomorphism
is an isomorphism.
Proof. This follows from (I 3.4) and from the fact that commutes with inductive limits.
Proposition 4.2. Every morphism of that is both an epimorphism and a monomorphism is an isomorphism.
Proof. Let be a morphism of that is an epimorphism and a monomorphism. First note that the morphism is a monomorphism
[PDF p. 253]
of presheaves (4.1 1)). Construct the amalgamated sum and the two canonical morphisms in the category of presheaves. Since is a monomorphism of presheaves, one verifies immediately that the following diagram is cartesian and cocartesian (I 10.1):
G --u--> H
| |
u i_2
v v
H --i_1-> K
Applying the “associated sheaf” functor, we therefore obtain a cartesian and cocartesian diagram of the category of sheaves (4.1 1)):
G --u--> H
| |
u a i_2
v v
H --a i_1-> aK
Since is an epimorphism of sheaves, the morphism is an isomorphism; and since the diagram (*) is cartesian, the morphism is an isomorphism.
Proposition 4.3.
- Inductive limits in that are representable are universal (I 2.5).
- Every epimorphic family (I 10.2) of morphisms is universally effective epimorphic (2.6).
- Every equivalence relation is universally effective (I 10.6).
- Filtered -inductive limits commute with finite projective limits (I 2.6).
[PDF p. 254]
Proof. Assertions 1) to 4) are true in the category of sets, hence in the category of presheaves (I 3.1). Assertions
- and 4) then follow immediately from the corresponding assertions for presheaves and from 4.1. Let us prove 3). Let be an equivalence relation of . The diagram
is then an equivalence relation of presheaves (4.1 1)). There therefore exists a morphism of presheaves such that the diagram
R --p_2--> X
| |
p_1 u
v v
X --u----> Y
is cartesian and cocartesian in the category of presheaves. Applying the “associated sheaf” functor, one obtains a cartesian and cocartesian diagram in the category of sheaves (4.1 1)). Consequently is an effective equivalence relation. Since every equivalence relation is effective by what precedes, every equivalence relation is universally effective.
Let us prove 2). Let be an epimorphic family of . By (II 2.6) and (I 2.12), it suffices to prove the following assertions:
a) For every morphism of sheaves , the family of morphisms , , is an epimorphic family of .
b) For every sheaf , the diagram of sets
is exact.
[PDF p. 255]
Let be the union of the images of the morphisms in the sense of presheaves, and let be the canonical injection. For every , the morphism factors through a morphism and the morphism , and the family , , is an epimorphic family of . Since is a monomorphism, the morphism is a monomorphism of sheaves (4.1). Since, for every , we have , is an epimorphism of sheaves. Consequently is an isomorphism (4.2). Let be a morphism of sheaves. We deduce by base change a morphism and morphisms . Since commutes with fiber products, is an isomorphism. Since epimorphic families of remain epimorphic after base change, the family , , is epimorphic in . Since commutes with inductive limits, the family , , is epimorphic in , which proves a).
Let us prove b). Let be a sheaf. Since is an isomorphism, the map
is a bijection. Since the , , form an epimorphic family of , the diagram of sets
is exact. Finally, since is a subpresheaf of , the presheaf , , is canonically isomorphic to , which proves b).
[PDF p. 256]
Remarks 4.3.1.
-
Proposition 4.3 2) formally gives a second proof of 4.2: it is clear that, in any category, a morphism that is both a universally effective epimorphism and a monomorphism is in fact an isomorphism.
-
One deduces from the preceding propositions that every morphism in the category of sheaves factors uniquely as an effective epimorphism followed by an effective monomorphism. This property naturally generalizes, for nonadditive categories, axiom (AB2) of abelian categories () [Tohoku].
4.4.0. Let be a -site. The functor (I 1.3.1), composed with the associated sheaf functor, provides a functor
called the canonical functor from to , which will be used constantly in what follows. The functor commutes with finite projective limits. When the topology of is less fine than the canonical topology, is fully faithful and commutes with projective limits; it is then, moreover, defined when does not necessarily possess a small topologically generating family. We shall not study in detail the behavior of with respect to inductive limits (cf. [SGA 3 IV]). We shall however need the propositions below.
Theorem 4.4. Let be a -site and let be a family of morphisms of with target . The following conditions are equivalent (*):
i) The family is an epimorphic family of (I 10.2).
(*) When the site does not necessarily possess a small topologically generating family and when the topology of is less fine than the canonical topology, one has ii) i) (cf. proof of 4.4).
[PDF p. 257]
ii) The family is a covering family of (1.2).
Proof. ii) i). Let be the sieve generated by the , (I 4.3.3), and let be the morphisms induced by the . The family of the , , is an epimorphic family of , and is a sieve covering . Consequently, for every sheaf , the map
is bijective, and the map
is injective. Therefore the map
is injective, and consequently the map
is injective, whence i).
i) ii). With the notation introduced above, let be the canonical injection into of the sieve generated by the , . It follows from i) that the morphism of sheaves is both an epimorphism of sheaves and a monomorphism of sheaves. It is therefore an isomorphism of sheaves (4.2). With the notation of no. 3, we have a commutative diagram:
X --ell(X)--> LX --ell(LX)--> aX
^ ^ ^
J_R LJ_R aJ_R
R --ell(R)--> LR --ell(LR)--> aR
[PDF p. 258]
First, by 3.1 2), there exists a sieve covering and a morphism such that
Put and . The morphisms and are morphisms from to . Let be the kernel of the pair of arrows . For every object of and every morphism , we have, by virtue of (4.4.1), . It follows from (3.1 3)) that the kernel of the pair of arrows is a sieve covering . Consequently, by axiom (T2) for topologies, is a sieve covering . We therefore have a morphism and a commutative diagram:
(4.4.2)
S_2 --J_{S_2}--> X
| \ |
u_2 \ | ell(X)
v \ J_R v
LR <--ell(R)-- R --LJ_R--> LX
Let be an object of and a morphism. By 3.1 1), there exists a sieve covering and a morphism such that . Put and . Let be the kernel of the pair and let be the canonical injection. For every object of and every morphism , we have, by the commutativity of (4.4.2):
[PDF p. 259]
It follows from 3.1 2) that the kernel of the pair is a sieve covering . By axiom (T2) for topologies, is a sieve covering . For every morphism , there therefore exists a sieve covering and a morphism such that the following diagram is commutative:
Y --beta--> S_2 --J_{S_2}--> X
^ /
| J_{Q_2} / J_R
Q_2 --v_2-------------> R
Now denote by the fiber product of and over . The sieve of contains the sieve , and consequently is a sieve covering . It then follows from axiom (T2) for topologies that is a sieve covering , and consequently the sieve , which contains , is a sieve covering , QED.
Corollary 4.4.4. Let be the topology of the -site , and let (respectively ) be the finest topology on among those for which the representable functors are sheaves (respectively separated presheaves) (2.2). Then .
Indeed, we trivially have , and 4.4 and 2.2 imply that , QED.
Definition 4.5. An initial object of a category is an object representing the empty inductive limit, i.e. such that, for every , there exists one and only one arrow . An object is said to be strictly initial if it is initial and if every morphism with target is an isomorphism. Let be a family of objects of a category . Suppose that the sum is representable. The sum is said to be disjoint if the structural morphisms are squarable, if they are monomorphisms, and if, for every pair , , the product is an initial object of . The sum is said to be disjoint
[PDF p. 260]
universal if it is disjoint and if it remains a disjoint sum after every base change . It follows that for every pair , , the objects are strictly initial objects of .
Example 4.5.1. In the category of sets, direct sums are disjoint and universal. The same is therefore true in every category of presheaves of sets (I 3.1), and consequently in every category of sheaves of sets on a -site (4.1); in particular, the initial object of is strict.
Proposition 4.6. Let be a -category and let be a family of morphisms of . For every -topology on (3.0.2), denote by the corresponding category of sheaves and by the corresponding functor.
-
Let be a -topology such that
is an isomorphism. Then for every topology finer than , the morphism
is an isomorphism.
-
Let be a -topology on . The following properties (i) and (ii) are equivalent:
i) a) The family is covering for .
b) For every , the diagonal morphism of presheaves
[PDF p. 261]
is transformed by the “associated sheaf” functor (for ) into an isomorphism, which is the case if the are monomorphisms.
c) For every pair , , of elements of , the presheaf is transformed by the “associated sheaf” functor (for ) into the initial object of .
ii) is the sum of the .
Proof. 1) A sheaf for is a sheaf for . Consequently, for every sheaf for , the morphism
is an isomorphism, which implies the assertion.
- By definition, property ii) holds if and only if the morphism of presheaves
is transformed by the “associated sheaf” functor into an isomorphism, i.e. (4.2) if and only if is an epimorphism and a monomorphism of sheaves. By (4.4), the morphism is an epimorphism if and only if property a) holds. The diagonal morphism
is the direct sum of a family , , of morphisms defined as follows:
) When , is the diagonal morphism .
) When , is the morphism , where denotes the initial object of .
[PDF p. 262]
The morphism is a monomorphism if and only if is an isomorphism, and by (4.1) is an isomorphism if and only if the are isomorphisms, i.e. if and only if properties b) and c) hold.
Corollary 4.6.1. Let be a -category and let be an object of .
-
Let be a -topology on . The object of is transformed by the “associated sheaf” functor (for the topology ) into the initial object of if and only if the empty sieve covers .
-
When is a strictly initial object of , the sheaf associated to , for every -topology finer than the canonical topology, is an initial object.
Proof. 1) Take the empty set as the set in 4.6 2).
- By 1) and 4.6 1), it suffices to show that the empty sieve covers for the canonical topology, i.e. (2.6) it suffices to show that for every object over , is an initial object of , which follows from the definition (4.5).
Corollary 4.6.2. Let be a -site and let be a family of squarable morphisms of with the same target and having the following properties:
) The family of the , , is covering.
) For every , is a monomorphism.
) For every pair , , of elements of , is covered by the empty sieve, i.e. for every sheaf on , is a set reduced to one element.
Then is the sum of the , .
[PDF p. 263]
Proof. This follows immediately from 4.6 2) and from 4.6.1 1).
Corollary 4.6.3. Let be a -category and let be a family of squarable morphisms with the same target. Suppose that the canonical topology of is a -topology. The following conditions are equivalent:
i) There exists a -topology on , less fine than the canonical topology, such that is the sum of the , .
ii) The object is the disjoint and universal sum (4.5) in of the .
Proof. ii) i). Since is the universal sum of the , , the family of the , , is covering for the canonical topology of (2.6). Condition ) of 4.6.2 is therefore satisfied when is equipped with the canonical topology. Condition ) is evidently satisfied, and property ) follows from 4.6.1 2), whence i).
i) ii). Let be a topology on less fine than the canonical topology such that is the sum of the . For every sheaf for , the canonical map is a bijection. Moreover, every representable presheaf is a sheaf. It follows at once that is the sum in of the , . Applying this to the case where the set is empty, one sees that if an object of is transformed into the initial object of , that object is an initial object of . The functor is fully faithful and commutes with finite projective limits. Consequently condition b) of 4.6 2) implies that the , , are monomorphisms of , and condition c) implies, by what precedes, that the , , are initial objects of . Consequently is the disjoint sum
[PDF p. 264]
of the . Since commutes with fiber products, this latter property is stable under every base change , and consequently is the disjoint and universal sum of the , .
Corollary 4.6.4. Let be a -category and let be an object of . Suppose that the canonical topology on is a -topology. The following properties are equivalent:
i) There exists on a topology less fine than the canonical topology such that is an initial object of .
ii) is a strictly initial object of .
Proof. Take the empty set as the set in 4.6.3.
Proposition 4.7. Let be a -category, let be the category of sheaves on for the canonical topology, and let be the canonical functor (4.4.0). Let be an equivalence relation in admitting a cokernel in . Let be the canonical morphism and suppose it is squarable. Consider the following properties:
i) is the quotient of the equivalence relation .
ii) The equivalence relation is universally effective (cf. no. 7).
One has ii) i). When possesses a -topology less fine than the canonical topology, one has i) ii).
Proof. ii) i). The canonical morphism is an isomorphism. Let be the sieve generated by . For every presheaf , one has an exact diagram (I 2.12):
[PDF p. 265]
It therefore suffices to show that, for every sheaf for the canonical topology, the map
is a bijection, i.e. it suffices to show that is covering for the canonical topology, or again that the sieve generated by is covering, which follows from Definition 2.5.
i) ii). The functor commutes with finite projective limits and is fully faithful. Now the canonical morphism
is an isomorphism (4.3). Consequently the canonical morphism is an isomorphism. The morphism is an epimorphism, which implies (4.4) that is a covering morphism of for the canonical topology, QED.
Proposition 4.8. Let be a -site. The category of sheaves on has the following properties:
a) Finite projective limits are representable.
b) Direct sums indexed by an element of are representable. They are disjoint and universal (4.5).
c) Equivalence relations are universally effective.
This was seen in 4.1 2) and 3), 4.3 3), and 4.5.1.
These properties have been singled out because they will later allow one to characterize the -categories equivalent to categories of sheaves on categories belonging to (IV 1.2).
Remark. Recall that property a) is equivalent to the property: there exists a final object, and fiber products are representable (I 2.3.1).
[PDF p. 266]
4.9. Let be a site whose underlying category is a -category. Then the category of -sheaves on satisfies the conditions considered in I 7.3 ((i) or (ii), as desired); hence, by loc. cit., the various variants considered in I 7.1 for the notion of generating family in coincide. This being said:
Proposition 4.10. With the preceding notation, consider the canonical functor (4.4.0), and let be a topologically generating family in (3.0.1). Then the family of objects of is a generating family. In particular the family of objects of is generating.
One must prove that every morphism in such that
is a bijection for every , is an isomorphism. But the preceding map identifies with the map
and one must show that if this map is bijective for every , then the same holds for every . Now let be a covering family of by objects of , and for every pair of indices of , consider the set of all morphisms in , with , varying in an index set . One then obtains a homomorphism of exact diagrams of sets
F(X) -> prod_i F(X_i) => prod_{ijk} F(X_{ijk})
| | |
v v v
F'(X) -> prod_i F'(X_i) => prod_{ijk} F'(X_{ijk}),
[PDF p. 267]
where, by hypothesis, the last two vertical arrows are bijections. The first vertical arrow is therefore also a bijection, QED.
Corollary 4.11. Suppose that is a -site (3.0.2). Then the category is a -category and admits a -small generating family.
Indeed, by hypothesis one can take a small generating family for in 4.10, which proves the existence of a small generating family. On the other hand, if , a homomorphism from to is known once one knows the homomorphism (4.10.1), i.e. (4.10.2), for every (I 7.1.1). It follows that the map
is injective. Since the second member is -small, the same holds for the first, which proves that is a -category.
Corollary 4.12. Let be a -site. Then for every object of , the set of subobjects of and the set of quotient objects of are -small.
This follows from I 7.4 and I 7.5 respectively, which apply thanks to 4.11.
5. Extension of a Topology from to
Proposition 5.1. Let be a -site and let be a morphism of . The following conditions are equivalent:
i) For every , with , the corresponding morphism has as image a covering sieve of .
ii) The morphism on associated sheaves is an
[PDF p. 268]
epimorphism of .
ii bis) For every sheaf on , the map deduced from is injective.
Proof. It is clear that ii) is equivalent to ii bis).
Let us prove that ii) i). By virtue of 4.4, i) means that is an epimorphism. But, since , the functor commuting with finite limits (4.1), the morphism under consideration is obtained from by base change. Since epimorphisms of are universal, this proves the implication ii) i). Conversely, since the family of the is epimorphic, the same is true of the family of the in (4.1). Thus, to verify that is epimorphic, it suffices to see it after every base change of the preceding type , which proves i) ii), QED.
Definition 5.2.
-
A morphism satisfying the three equivalent conditions of 5.1 is called a covering morphism. A family of morphisms , , with the same target is said to be covering if the corresponding morphism is covering.
-
A morphism is said to be bicovering if it is covering and if the diagonal morphism is covering. A family , , with the same target is said to be bicovering if the corresponding morphism is bicovering.
By condition i) of 5.1, to say that a family , , is covering means that, for every morphism , with , the family of , , has as image a sieve covering
[PDF p. 269]
; or again, by ii), that the family of morphisms in is epimorphic, taking into account that the functor commutes with direct sums (4.1). (*)
Proposition 5.3. Let be a -site and let be a morphism of . The following conditions are equivalent:
i) The morphism is bicovering (5.2 2)).
i bis) The morphism is covering and, for every object of and every pair of morphisms such that , the kernel of is a sieve covering .
ii) The morphism is an isomorphism of .
ii bis) For every sheaf on , the map is a bijection.
Proof. The equivalence i) i bis) follows from condition i) of 5.1, applied to the diagonal morphism ; the equivalence ii) ii bis) is trivial.
i) ii). The morphism is an epimorphism (5.1). Since the functor commutes with the formation of fiber products (4.1), the diagonal morphism is an epimorphism (5.1). Since the diagonal morphism is always a monomorphism, it is an isomorphism (4.2). Consequently the morphism is a monomorphism. It is therefore an isomorphism (4.2).
ii) i). Since the functor commutes with the formation of fiber products, the morphism and the diagonal morphism are transformed by into isomorphisms. In particular, they are transformed by into epimorphisms. They are therefore covering, QED.
(*) Note that the property, for a morphism or a family of morphisms of , of being covering (respectively bicovering) is stable under base change.
[PDF p. 270]
5.3.1. It follows from 5.3, and from the fact that commutes with direct sums, that a family of morphisms of is bicovering if and only if the induce on associated sheaves an isomorphism from the direct sum onto ; or again if and only if, for every sheaf , the map
is bijective.
Proposition 5.4. Let be a -site. There exists on a topology, evidently unique, such that a family of arrows of with the same target is covering for this topology if and only if it is covering in the sense of 5.2. This is also the least fine topology on among the topologies having the following properties:
- a) is finer than the canonical topology of , that is, every epimorphic family of is covering for .
- b) Every covering family in is covering in .
Proof. We shall confine ourselves to giving indications. We leave to the reader the task of showing that the covering families in the sense of 5.2 are the covering families of a topology on ; one uses 4.1. The covering families of the canonical topology on are the epimorphic families on (2.6 and I 3.1). Since commutes with inductive limits (4.1), the topology is finer than the canonical topology of .
Moreover the covering families of are covering families of (5.1). Thus if denotes the least fine of the topologies on possessing properties a) and b), is finer than . Let be a covering family of . We show that is a covering family of .
[PDF p. 271]
Let be the canonical monomorphisms, and let be the morphism defined by the . The family of the is covering for . To show that is a covering family of , it therefore suffices, by virtue of axiom (T2) for topologies, to show that the morphism is covering for .
There exists an epimorphic family of , , , with (I 3.4). To show that is covering for , it therefore suffices, again by axiom (T2), to show that, for every , the morphism
is covering for . Let , , , be an epimorphic family of . The family is covering for . It is therefore a covering family of (5.1), and then a covering family of . Consequently the sieve generated by contains a sieve covering for . It is therefore covering for , and consequently is covering, QED.
Remark 5.4.1. The proof of 5.4 shows in fact that every topology on , finer than the canonical topology of , is the least fine of the topologies on that possess the following properties:
- a) is finer than the canonical topology of .
- b) Every covering family for of the form , where and the are objects of , is covering for .
In particular, every topology on , finer than the canonical topology, is uniquely determined by the families , with and the objects of , that are covering for .
[PDF p. 272]
Remark 5.4.2. One can easily show that, for every topology on finer than the canonical topology, the set of families of morphisms with the same target in that are covering for is the set of covering families of a topology on . Thus 5.4 and 5.4.1 make it possible to establish a one-to-one correspondence between topologies on and topologies on finer than the canonical topology.
5.5.0. Let be a small category. Denote by the set of strictly full subcategories of (that is, such that every object isomorphic to an object of the subcategory is again an object of the subcategory) whose injection functor admits a left adjoint that commutes with finite projective limits. Also denote by the set of topologies on .
Theorem 3.4 defines a map
We shall define a map in the opposite direction. For this, one must associate to every element
a topology on . For every object of , we define to be the set of subobjects of whose injection morphism is transformed by into an isomorphism. Using the hypotheses made on , one verifies immediately that this defines a topology on . We have therefore defined a map
We then have the following result, due to J. Giraud:
[PDF p. 273]
Theorem 5.5. The map is a bijection, and is its inverse.
Proof. The map is the identity. Indeed, this follows immediately from 5.1.
The map is the identity. Indeed, let be an element of , let be the topology corresponding to it by , and let be the category of sheaves for . Using the definition of the topology and the definition of bicovering morphisms (5.2), one then proves the equivalence of the following assertions:
- i) The morphism of is bicovering for the topology .
- ii) The morphism of is transformed by into an isomorphism.
It is clear that is a full subcategory of . It therefore suffices to show that every sheaf for the topology is an object of . But, by the equivalence above, the morphism
is bicovering; since its source and target are sheaves, 5.3 (ii) implies that is an isomorphism, QED.
6. Sheaves with Values in a Category
6.0. Let and be two categories. A contravariant functor from to ,
is called a presheaf on with values in .
Definition 6.1. Let be a site and a category. A presheaf is called a sheaf on with values in , or more briefly a sheaf with values in , if for every object of , the presheaf of sets
is a sheaf. The full subcategory of formed by sheaves on with values in will be denoted .
[PDF p. 274]
Remarks 6.2.
-
Condition 6.1 means that, for every object of , every covering sieve of , and every object of , one has an isomorphism (I 3.5 and 2.1)
Thus, when is the category of -sets, one recovers the definition of sheaves of sets (loc. cit.).
-
Let be a -category and let be a functor commuting with projective limits. The functor transforms, by composition, sheaves with values in into sheaves with values in .
6.3.0. Let be a species of algebraic structure defined by finite projective limits (I 2.9), and let - be the category of -sets belonging to . Denote by the underlying set functor. The functor commutes with projective limits and, consequently, by the preceding remark, defines by composition the functors
The functor is called the “underlying sheaf of sets” functor. In fact, it factors canonically through the category - of -objects of , and, denoting the resulting functor again by ,
one can state:
Proposition 6.3.1. The functor establishes an equivalence of categories between the category of sheaves on with values in -, the category of -objects of the category of sheaves of sets, and the category of -objects of the category of presheaves of sets whose underlying presheaf is a sheaf.
[PDF p. 275]
Proof. The proof essentially uses (I 3.2) and the fact that a finite projective limit of sheaves in the category of presheaves is a sheaf (4.1).
6.3.2. The preceding proposition justifies the abuse of language that consists in identifying a sheaf with values in - and the corresponding -object in . We shall use this abuse of language systematically from now on.
We shall use the classical terminology: sheaves of groups, sheaves of commutative groups, which we shall most often call abelian sheaves, sheaves of rings, sheaves of -modules, and so on.
6.3.3. We denote by , respectively , the category of -objects of , respectively of . When is the species of structure “-module”, one also writes , or simply when .
Suppose that is a -site. The “associated sheaf” functor is left exact (4.1), and consequently:
Proposition 6.4. The inclusion functor admits a left adjoint , left exact; that is, one has an isomorphism
Let be an object of . The sheaf of sets underlying is canonically isomorphic to the sheaf of sets associated to the presheaf of sets underlying . The morphism of presheaves of sets underlying the adjunction morphism identifies with the adjunction morphism applied to the underlying presheaf of sets.
[PDF p. 276]
Proposition 6.5. Suppose that the functor admits a left adjoint (*)
The functor , respectively , admits a left adjoint , respectively , and one has a canonical isomorphism
Proof. The proof is formal and is left to the reader.
Corollary 6.6. Let be a generating family of (4.9). Then the family is a generating family of .
Proof. The proof is formal once one has observed that the functor commutes with projective limits and is conservative: is an isomorphism if and only if is an isomorphism.
(*) One proves that such an adjoint always exists [C.F.]: free group, free abelian group, free -module, etc.
Proposition 6.7. Let be either a -sheaf of rings or a small ring, and let , respectively , be the category of sheaves, respectively presheaves, of unitary -modules (6.3.3) on a -site . Then is an abelian -category verifying axioms (AB 5), “existence of left exact filtered inductive limits”, and (AB 3)*, “existence of infinite products”, of [TOHOKU]. It has a family of generators indexed by an element of .
Proof. It is clear that the category is an abelian category verifying axioms (AB 3), (AB 4), and (AB 5) (I 2.8 and I 3.3).
[PDF p. 277]
It then follows from 6.4 that the category is an additive category in which kernels and cokernels are representable. More precisely, let and be the inclusion functor into presheaves and the associated sheaf functor, and let be a morphism of . The functor is left exact and commutes with inductive limits. The functor commutes with projective limits. Consequently the functor is an additive left exact functor. One deduces canonical isomorphisms:
Consequently one has a canonical isomorphism:
We show that the canonical morphism is an isomorphism. For this, it suffices to show, by (*), that the sequence
is exact. Using the isomorphisms () and (*) and observing that is isomorphic to , one sees that this sequence is isomorphic to the sequence
which is none other than the transform by the functor of the sequence
But this sequence is exact and the functor is left exact. The category is therefore an abelian category. The functor
[PDF p. 278]
is left exact and commutes with arbitrary inductive limits. Consequently it is additive, exact, and commutes with inductive limits. Since the category verifies axiom (AB 5), the same is true of the category .
Finally it is clear that, in the category , products indexed by an element of are representable, and consequently that axiom (AB 3)* is verified, which completes the proof of the first assertion.
For every ring that is an element of , the functor
admits a left adjoint , the free -module generated by . It therefore suffices to apply 4.10 and 6.6 to complete the proof.
Notation 6.8. Let be a sheaf of sets. The sheaf of -modules associated to the presheaf (cf. notation 6.5)
is denoted . When , the sheaf associated to the presheaf represented by , one sometimes simply writes , by abuse of notation. The proof of 6.7 shows that the family of sheaves of -modules , , is a generating family, indexed by an element of , of the category of sheaves of -modules.
Remark 6.9. By [TOHOKU], 6.7 shows that the category , when is an element of , has enough injectives. It is known, moreover, that infinite products are not necessarily exact in , and consequently that the category does not in general have enough projectives [Roos].
[PDF p. 279]
Bibliography
[1] M. Artin, Grothendieck’s topologies.
[2] M. Demazure, Seminaire de Geometrie Algebrique III, Expose IV, Lecture Notes, Springer-Verlag.
[3] A. Grothendieck, Sur quelques points d’Algebre Homologique, Tohoku Math. Journal.
[4] J. E. Roos, CR.