Exposé IV. Topologies and sheaves

by M. Demazure1

2

This Exposé is intended to acquaint the reader with the essentials of the language of topologies and sheaves (without cohomology), particularly convenient in questions of passage to the quotient (among others).

The first three sections develop the language of passage to the quotient. The fourth, which is the central part, is the exposition of the theory of sheaves, oriented principally towards the application to questions of quotients; the fifth is an application to passage to the quotient in groups and to principal homogeneous fibered objects. The last section concerns more specifically the category of schemes, and defines various useful topologies on this category.

The reader will profitably refer to [AS], [MA], [D], and SGA 4; [D] in particular as regards the applications of topologies to the theory of descent, and SGA 4 for questions of universes (particularly mistreated in this Exposé).

1. Universal effective epimorphisms

In what follows in this Exposé, we suppose fixed a category .

Definition 1.1. A morphism is called an epimorphism if, for every object , the corresponding map

X(S) = Hom(S, X) ⟶ X(T) = Hom(T, X)

is injective.3 One says that is a universal epimorphism if for every morphism , the fiber product exists, and is an epimorphism.

Definition 1.2. A diagram

A —u→ B ⇉_{v₁,v₂} C

of maps of sets is said to be exact if is injective and if its image is formed by the elements of such that . A diagram of the same type in is said to be exact if for every object of , the corresponding diagram of sets

A(X) → B(X) ⇉ C(X)

is exact; one then also says that makes into a kernel of the pair of arrows . Dually, a diagram

C ⇉_{v₁,v₂} B —u→ A

in is said to be exact if it is exact as a diagram in the opposite category , i.e. if for every object of , the corresponding diagram of sets

X(A) → X(B) ⇉ X(C)

is exact.4 One also says that makes into a cokernel of the pair of arrows .

Definition 1.3. A morphism is called an effective epimorphism if the fiber square exists, and if the diagram

T ×_S T ⇉_{pr₁,pr₂} T —u→ S

is exact, i.e. if makes into a cokernel of . One says that is a universal effective epimorphism if for every morphism , the fiber product exists, and the morphism is an effective epimorphism.

One evidently has the implications:

universal effective epimorphism  ⟹  effective epimorphism
            ⇓                                ⇓
universal epimorphism            ⟹  epimorphism,

but in general no other implication holds.5

Definition 1.4.0.6 We "recall" that a morphism is said to be squarable if for every morphism , the fiber product exists.

Lemma 1.4. Consider morphisms . Then

a) u, v epimorphisms epimorphism epimorphism,

b) u, v universal epimorphisms universal epimorphism, and squarable universal epimorphism.

Lemma 1.4 is trivial from the definitions. From it we conclude:

Corollary 1.5. Let and be universal epimorphisms, such that exists; then exists and is a universal epimorphism.

Let us also note:

Definition 1.6.0.7 One says that an object of is squarable if its product with every object of exists. (If has a final object , this is equivalent to saying that the morphism is squarable, cf. 1.4.0.)

Lemma 1.6. Let be a morphism in ; for it to be an epimorphism (resp. universal epimorphism, resp. effective epimorphism, resp. universal effective epimorphism), it suffices that the corresponding morphism in be such, and this is also necessary if one supposes that is a squarable object of .

Immediate proof left to the reader. One uses the hypothesis " squarable" in order to interpret the -morphisms from an object of into an object of as the -morphisms from into .

Lemma 1.7. With the notation of 1.4: u, v effective epimorphisms and universal epimorphism effective epimorphism.

To see this, one considers the diagram

S ⇇ T ⇇ T ×_S T
  ↑   ↑v   ↑v×_S v
    U ⇇ U ×_S U
        ↖
       U ×_T U

One notes that by hypothesis, row 1 and column 1 are exact, and that by virtue of 1.5 and 1.6, is an epimorphism ( being a universal epimorphism). The conclusion follows by an evident diagram-chase: if an element of has the same images in , it has a fortiori the same images in , hence comes from an element of since column 1 is exact. As row 1 is exact, it suffices to verify that the element under consideration has the same images in , and since is an epimorphism, it suffices to verify that the images in are the same, which is indeed the case.

Proposition 1.8. Consider morphisms . Then u, v universal effective epimorphisms universal effective epimorphism, and squarable universal effective epimorphism.

The first implication follows at once from 1.7. For the second, one looks at the diagram (of "bisimplicial" type):

S ⇇ T ⇇ T ×_S T
↑uv  ↑   ↑
U ⇇ U ×_S T ⇇ U ×_S T ×_S T
↑    ↑           ↑
U ×_S U ⇇ U ×_S U ×_S T ⇇ U ×_S U ×_S T ×_S T.

Columns 1, 2, 3 are exact by virtue of the hypothesis "uv universal effective epimorphism", row 2 is exact, since is an effective epimorphism (because it has a section over ), and the same holds for row 3 (same reason). An evident diagram-chase then shows that row 1 is exact, i.e. is an effective epimorphism. As the hypotheses made are invariant under any change of base , it follows that is in fact a universal effective epimorphism.

Corollary 1.9. Let and be universal effective epimorphisms, such that exists; then exists and is a universal effective epimorphism.

Proof as for 1.5 by the diagram

Y′ ⇇ X′
↑    ↑u′
X ⇇ X × Y′ ⇇ X × X′
↑u             ↘ u×u′
Y ⇇ Y × Y′.

Corollary 1.10. Consider a squarable morphism , and a change-of-base morphism which is a universal effective epimorphism. For to be a universal effective epimorphism, it is necessary and sufficient that be one:

T ⇇v′ T′
↓u    ↓u′
S ⇇v  S′.

Only the "it suffices" requires a proof. Now if is a universal effective epimorphism, so is by 1.8, and since , one concludes by 1.8 that is a universal effective epimorphism.

Remark 1.11. The same reasoning shows that in 1.10 one can replace "universal effective epimorphism" by "universal epimorphism" or "universal and effective epimorphism", or simply by "epimorphism" (and in this last case, the hypothesis " squarable" is evidently unnecessary).

In the proof of 1.8 we used the following result, which deserves to be made explicit:

Proposition 1.12. Let be a morphism that admits a section. Then is an epimorphism, and if exists, it is an effective epimorphism, and a universal effective epimorphism if moreover is squarable.

The first assertion is contained in 1.4 a), and the third will follow at once from the second, which it will therefore suffice to establish. In fact we have a stronger conclusion: for every functor (not necessarily representable), the diagram of sets

F(S) → F(T) ⇉ F(T ×_S T)

is exact. This may be considered as a particular case of the formalism of Čech cohomology (in dimension 0!), which we content ourselves with recalling here. Suppose simply that exists; one then sets

Ȟ⁰(T/S, F) = Ker(F(T) ⇉ F(T ×_S T)).

One may evidently regard as a contravariant functor in the argument varying in , every -morphism defining a map

(+)    Ȟ⁰(T/S, F) → Ȟ⁰(T′/S, F).

Fix and in . A well-known calculation shows that if there exists an -morphism from into , the corresponding map (+) is in fact independent of the choice of this morphism,8 so that may be regarded as a functor on the category associated to the set preordered by the relation of "domination" ( dominates if there exists an -morphism from into ). In particular, if and are isomorphic in this latter category, i.e. if each dominates the other, then (+) is an isomorphism of sets. This applies in particular to the case where is the final object of , i.e. essentially itself; in any case dominates , and the converse is true precisely if has a section. This establishes 1.12 in the strengthened form announced.

Remark 1.13. For various applications, the notions introduced in the present Exposé, and the results stated, must be developed more generally relative to a family of morphisms with the same target (instead of a single morphism ). Thus, such a family will be said to be epimorphic if for every object of , the corresponding map

is injective, and one introduces in the same way the notion of an effective epimorphic family and the "universal" variants of these notions. We shall admit, if need be, in what follows, that the results of the present Exposé extend to this more general situation.

Remark 1.14. Every morphism that is at once a monomorphism and an effective epimorphism is an isomorphism.

Indeed, in the notation of 1.3, the fact that be a monomorphism entails that the two morphisms

pr₁, pr₂ : T ×_S T ⇉ T

are equal (and are isomorphisms). Now a diagram

C ⇉_{v,v} B —u→ A

is exact only if is an isomorphism, as follows immediately from the definition.9

2. Descent morphisms

Let us recall the following definitions:

Definition 2.1. Let be a morphism such that exists, and let be an object over . One calls gluing datum on , relative to , an -isomorphism

where () denotes the inverse image (supposed to exist) of under the projection . One says that the gluing datum is a descent datum if it satisfies the "cocycle condition"

where () are the canonical projections from into (N.B. one now supposes that also exists), where is the inverse image of , considered as an -morphism from into , and where for every integer between 1 and 3, denotes the inverse image (supposed to exist) of under the projection of index , .

In the second part of the definition, we have therefore used identifications and abuses of writing in current use,10 which experience proves to be harmless, but which it is evidently fitting to avoid in a rigorous exposition of the theory of descent (which must precisely justify to a certain extent these common abuses of language). Such a formal exposition ([D]) has been written by Giraud (with the aim of justifying and making precise SGA 1, VII, which was never written). For a detailed exposition of the results of faithfully flat descent of which constant use will be made in the present Séminaire, one may consult SGA 1, VIII.

Let still be a morphism such that exists, and let be an object over such that and exist; then the inverse images of under prᵢ () exist and are canonically isomorphic, and consequently is endowed with a canonical gluing datum relative to . When and exist, this is even a descent datum. If is another object over , satisfying the same conditions as , then for every -morphism , the corresponding -morphism is "compatible with the canonical gluing data" on . If in particular is a squarable morphism, then

X ⟼ X′ = X ×_S S′

is a functor from the category into the "category of objects over equipped with a descent datum relative to " — a category whose definition is left to the reader, and which is a full subcategory of the "category of objects over equipped with a gluing datum relative to ".

This being so:

Definition 2.2. One says that a morphism is a descent morphism (resp. an effective descent morphism*) if is squarable (i.e. for every , the fiber product exists), and if the preceding functor X ⟼ X′ = X ×_S S′ from the category of objects over , into the category of objects over equipped with a descent datum relative to , is fully faithful (resp. an equivalence of categories).*

One notes that the first of these two notions can be expressed using only the notion of gluing datum (hence without involving the triple fiber product ), being a descent morphism if is squarable and X ⟼ X′ is a fully faithful functor from the category into the category of objects over equipped with a gluing datum relative to . When one makes this definition explicit, one finds that it means that for two objects X, Y over , the following diagram of sets

(×)    Hom_S(X, Y) → Hom_{S′}(X′, Y′) ⇉_{p₁, p₂} Hom_{S″}(X″, Y″)

is exact, where denotes the inverse image of under the projection , for ; indeed, the kernel of the pair is by definition none other than the set of -morphisms compatible with the canonical gluing data.

Note that, by definition of the inverse images , one has canonical bijections

Hom_{S′}(X′, Y′) ≃ Hom_S(X′, Y)    and    Hom_{S″}(X″, Y″) ≃ Hom_S(X″, Y),

so that the exactness of the diagram (×) is equivalent to that of

(××)    Hom_S(X, Y) → Hom_S(X′, Y) ⇉ Hom_S(X″, Y),

obtained by applying to the diagram in :

(×××)   X″ ⇉ X′ → X

which is deduced from

S″ ⇉ S′ → S

by the base change . This proves, taking account of 1.6, the first part of the following proposition:

Proposition 2.3. Let be a morphism. If it is a universal effective epimorphism, it is a descent morphism, and the converse is true if is a squarable object of (i.e. its product with every object of exists).

It remains to prove that if is a descent morphism, it is a universal effective epimorphism, that is, that for every morphism , the diagram (×××) is exact, i.e. for every object of , the transform of this diagram by is an exact diagram of sets. Now by hypothesis exists; let be the object of that it defines; then the transform of (×××) by is isomorphic to the transform by , and this last is exact by the hypothesis on .

One may therefore apply to universal effective epimorphisms the results on descent morphisms, such as the following:

Proposition 2.4. Let be a descent morphism (for example a universal effective epimorphism). Then:

a) For every -morphism , is an isomorphism (resp. a monomorphism) if and only if is.

b) Let X, Y be two subobjects of , and the subobjects of inverse images of the preceding ones. For to be contained in (resp. to be equal to ), it is necessary and sufficient that the same hold for .

For (a), it follows from the definition that if is an isomorphism in the category of objects with gluing datum, then is an isomorphism; now one notes at once that every isomorphism of objects over compatible with gluing data is an isomorphism of objects with gluing datum, i.e. its inverse is also compatible with the gluing data. For b), one is reduced to proving that if is contained in , i.e. if there exists an -morphism , then the same holds for X, Y over . Now since , and hence also , is a monomorphism, one sees that is automatically compatible with the gluing data, hence comes from an -morphism . Note that the proof holds more generally when one has two objects X, Y over , with a monomorphism, and one asks whether the morphism factors through : it suffices that factor through .

Corollary 2.5. Let be a universal effective epimorphism and a morphism such that exists. Suppose that is also a fiber product of with itself over , i.e. . Then is a monomorphism (and conversely, of course).

Indeed, consider the cartesian diagram

S′ ×_S S′ ⥲ S′ ×_T S′
    ↓             ↓ f
    S    →    S ×_T S,

where the second horizontal arrow is the diagonal morphism. By virtue of 1.9 the second vertical arrow is a universal effective epimorphism, by hypothesis the first horizontal arrow is an isomorphism, hence by virtue of 2.4 a) or b) at one's choice,11 the same holds for , which means precisely that is a monomorphism.

Remark 2.6. The notions introduced in 2.1, in terms of morphisms between certain inverse limits, become explicit in an obvious way in terms of the contravariant functors defined by the objects under consideration: subject to the existence of the fiber products envisaged in 2.1, there is a one-to-one correspondence between gluing data (resp. descent data) on relative to , and gluing data (resp. descent data) for the corresponding objects in . This allows one, if one wishes, to extend these notions to the case where one makes no hypothesis of existence of fiber products in .

Remark 2.7. The notions introduced in this number generalize to the case of families of morphisms. They can on the other hand be presented in a more intrinsic manner using the notion of sieve (4.1). For these questions, the reader is referred to [D].

3. Universal effective equivalence relations

3.1. Equivalence relations: definitions

Definition 3.1.1. One calls a -equivalence relation in a representable subfunctor of the functor , such that for every , is the graph of an equivalence relation in .

This definition applies in particular to the category Ĉ. If one considers as an object of Ĉ, one then sees that a Ĉ-equivalence relation in is nothing other than a subfunctor of such that is the graph of an equivalence relation in for every object of .12

If is a -equivalence relation in , one denotes by pᵢ the morphism from into induced by the projection . One thus has a diagram

p₁, p₂ : R ⇉ X.

Definition 3.1.2. A morphism is said to be compatible with if . An object of that is a cokernel of the pair is also called a quotient object of by and denoted .

One thus has an exact diagram

R ⇉_{p₁, p₂} X —p→ X/R

and represents the covariant functor

Hom_C(X/R, Z) = {morphisms from X into Z compatible with R}.

Since quotient objects have been chosen in , the quotient is unique (when it exists).

These definitions generalize at once to the case of a Ĉ-equivalence relation in , but one will notice that the identification of objects of with their images in Ĉ does not commute with the formation of quotients, that is, the quotient of by in is not a priori a quotient of by in Ĉ. One should therefore guard against rashly identifying with its image in Ĉ in questions involving passages to the quotient.13

In what follows, we shall say simply equivalence relation for Ĉ-equivalence relation; we shall specify, when appropriate, whether we are dealing with -equivalence relations.14

Definition 3.1.3. If is an object of over , one calls an equivalence relation in over an equivalence relation in such that the structural morphism is compatible with .

The canonical monomorphism then factors through the monomorphism

X ×_S X ⟶ X × X

and defines an equivalence relation in the object of . When the quotient exists, it is endowed with a canonical morphism into and the corresponding object of is a quotient of by the preceding equivalence relation.

Conversely, if is a squarable object of and if is a quotient of by this equivalence relation (in ), then is a quotient of by in . In any case, we shall never have to consider quotients in that are not already quotients in .

Definition 3.1.4. If (resp. ) is an object of equipped with an equivalence relation (resp. ), a morphism

is said to be compatible with and if the following equivalent conditions are satisfied:

(i) for every , two points of congruent modulo are transformed by into two points of congruent modulo ;

(ii) there exists a morphism (necessarily unique) making commutative the diagram

R   →   R′
↓        ↓
X × X → X′ × X′.
        (u × u)

By the universal property of , there then exists (when the quotients and exist) a unique morphism making commutative the diagram

Definition 3.1.5. A subobject (= a representable subfunctor) of is said to be stable under the equivalence relation if the following equivalent conditions are satisfied:

(i) For every , the subset of is stable under .

(ii) The two subobjects of inverse images of under and are identical.

An important particular case of stable subobject of is the following: the quotient exists and is the inverse image on of a subobject of .

Definition 3.1.6. Let be an equivalence relation in and a morphism. The equivalence relation in obtained by the cartesian diagram

R′      →    R
↓             ↓
X′ × X′ →  X × X

is said to be the equivalence relation in inverse image of the equivalence relation in . In particular, if is a subobject of , one says it is the equivalence relation induced in by , and one denotes it .

The morphism is compatible with and ; one thus has, when the quotients exist, a morphism (3.1.4). If is a subobject of , we shall see later that in certain cases one can prove that is a monomorphism, hence identifies with a subobject of . When this is so, the inverse image of this subobject in will be a subobject of containing and stable under : the saturation of for the equivalence relation .

Proposition 3.1.7. If the subobject of is stable under , one has two cartesian squares, for :

Immediate proof.

3.2. Equivalence relation defined by a group acting freely

Definition 3.2.1. Let be an object of and a -group acting on (that is, equipped with a morphism of Ĉ-groups

H ⟶ Aut(X)).

One says that acts freely on if the following equivalent conditions are satisfied:

(i) For every , the group acts freely on ;

(ii) The morphism of functors

H × X ⟶ X × X

defined set-theoretically by (h, x) ⟼ (hx, x) is a monomorphism.

(In the preceding definition, one supposed that the group acted "on the left" on . One evidently has an analogous notion in the case where the group acts "on the right", that is, when one is given a morphism of groups from the group opposite to into ).

If acts freely on , the image of under the morphism of (ii) is an equivalence relation in called the equivalence relation defined by the action of on . When the quotient of by this equivalence relation exists, it is denoted

( when acts on the right). It represents the following covariant functor: if is an object of , one has

Hom(H\backslash X, Z) = {morphisms from X into Z invariant under H}

where the morphism is said to be invariant under if for every , the corresponding morphism is invariant under the group .

Lemma 3.2.2. Under the conditions of 3.2.1, let be a subobject of . The following conditions are equivalent:

(i) is stable under the equivalence relation defined by (3.1.5);

(ii) For every , the subset of is stable under ;

(iii) There exists a morphism , necessarily unique, making commutative the diagram

H × Y  —f→  Y
↓             ↓
H × X   →   X.

Under these conditions, defines a morphism of Ĉ-groups

and the equivalence relation defined in by this action of is none other than the equivalence relation induced in by the equivalence relation defined in by the action of .

Immediate proof. One evidently has an analogous statement for a "right action". The action of on defined above will be called the action induced in by the given action of on .

Let us now consider the following situation: and are two -groups and one is given a morphism of groups

Then acts on by translations (one sets set-theoretically ) and acts freely there if and only if is a monomorphism. The quotient of by this action of is denoted, when it exists, . One defines similarly a right action of on and a quotient . These quotients are functorial with respect to the groups in question; more precisely, one has the following lemma, stated for right quotients:

Lemma 3.2.3. Let and be two monomorphisms of -groups. Suppose given a morphism of -groups

The following conditions are equivalent:

(i) is compatible with the equivalence relations defined in and by and .

(ii) For every , one has .

(iii) There exists a morphism , necessarily unique and multiplicative, such that the following diagram is commutative

Under these conditions, if the quotients and exist, there exists a unique morphism making commutative the diagram

The first part is proved by reduction to the set-theoretic case. The second follows at once from (i).

One could translate to the present situation the notions introduced above for general equivalence relations. Let us simply signal the following lemma, whose proof is immediate by reduction to the set-theoretic case:

Lemma 3.2.4. Let be a monomorphism of -groups and a sub--group of . For the subobject of to be stable under the equivalence relation defined by , it is necessary and sufficient that factor through the canonical monomorphism , and under this condition the action of on induced by the action of on defined by is none other than the action deduced from the monomorphism that factors .

3.3. Universal effective equivalence relations

Definition 3.3.1. Let be a morphism. One calls the equivalence relation defined by in , and denotes by , the Ĉ-equivalence relation in image of the canonical monomorphism

X ×_Y X → X × X.

Definition 3.3.2. Let be an equivalence relation in . One says that is effective if

(i) is representable (i.e. is a -equivalence relation);

(ii) the quotient exists in ;15

(iii) the diagram

R ⇉_{p₁, p₂} X —p→ Y

makes the fiber square of over , that is, is the equivalence relation defined by .

Scholie 3.3.2.1.16 If is an effective equivalence relation in , then is an effective epimorphism (1.3). If is an effective epimorphism, then is an effective equivalence relation in a quotient of which is . There is therefore a "Galois" one-to-one correspondence between effective equivalence relations in and effective quotients of (i.e. equivalence classes of effective epimorphisms with source ).

Definition 3.3.3. One says that the equivalence relation in is universally effective if the quotient exists, and if, for every , the fiber products and exist and is a fiber square of over . It amounts to the same to say that is effective and that is a universal effective epimorphism.

Scholie 3.3.3.1.16 There is therefore, as above, a one-to-one correspondence between universal effective equivalence relations in and universal effective quotients of .

Remark 3.3.3.2.16 Suppose that is the category of -schemes and let denote the affine space of dimension 1 over . Let be a universal effective equivalence relation and the quotient. Then, for every open of , is the set of elements of such that . In particular, if is given by the action of a group acting freely on the right on (cf. 3.2.1), then is the set of such that , for every and , .

Proposition 3.3.4. Let be a universal effective equivalence relation in . Let be a morphism compatible with , hence factoring through . The following conditions are equivalent:

(i) is a monomorphism;

(ii) is the equivalence relation defined by .

Indeed, (i) trivially entails (ii); the converse is none other than 2.5.

Definition 3.3.5. Let be a -group acting freely on . One says that acts effectively, or that the action of on is effective (resp. universally effective*), if the equivalence relation defined in by the action of is effective (resp. universally effective).*

3.4. (M)-effectivity

In practice, it is most often difficult to characterize universal effective epimorphisms. One often disposes, nevertheless, of a certain number of morphisms of this type, for example in scheme theory, of faithfully flat quasi-compact morphisms. This leads to the developments below.

3.4.1. Let us first state a certain number of conditions bearing on a family (M) of morphisms of :

(a) (M) is stable under extension of the base, i.e., every element of (M) is squarable (cf. 1.4.0) and for every , is an element of (M).

(b) The composite of two elements of (M) is in (M).

(c) Every isomorphism is an element of (M).

(d) Every element of (M) is an effective epimorphism.

Note that (a) and (b) entail:

(a′) The cartesian product of two elements of (M) is in (M): let and be two -morphisms, elements of (M). If exists, then exists and is an element of (M).

This follows from the diagram

Y′    ⇇    X′
↑           ↑u′
X  ⇇  X ×_S Y′  ⇇  X ×_S X′
↑u                   ↘ u×_S u′
Y  ⇇  Y ×_S Y′.

Likewise (a) and (d) entail:

(d′) Every element of (M) is a universal effective epimorphism.

3.4.2. The family of universal effective epimorphisms satisfies the conditions (a) through (d) of 3.4.1. Indeed, (a), (c) and (d) are satisfied by definition, (b) follows from 1.8. In what follows, we shall suppose given a family (M) of morphisms of satisfying these conditions: our results will therefore apply in particular to the family .

Definition 3.4.3. One says that the equivalence relation in is of type (M) if it is representable and if (which by (b) and (c) entails that ).

One says that is (M)-effective if it is effective and if the canonical morphism is an element of (M).

One says that the quotient of is (M)-effective if the canonical morphism is an element of (M).

The following consequences result from this definition:17

Proposition 3.4.3.1. (i) An (M)-effective equivalence relation is of type (M) and universally effective.

(ii) An (M)-effective quotient is universally effective (cf. 3.3.3).

(iii) The maps R ⟼ X/R and p ⟼ R(p) realize a one-to-one correspondence between (M)-effective equivalence relations in and (M)-effective quotients of .

(iv) -effective is equivalent to universally effective.

Let us prove point (i). Since is (M)-effective, one has a cartesian square

and . Then, by 3.4.1 (a), and belong to (M), hence is of type (M).

Set and let be an arbitrary morphism. By 3.4.1 (a), the fiber products and exist and the morphisms and belong to (M). Finally, since , one obtains, by associativity of the fiber product:

R′ = X ×_Y X ×_Y Y′ = X′ ×_{Y′} X′.

This shows that is (M)-effective; hence in particular, is universally effective. This proves (i), and also (iv). Points (ii) and (iii) follow from this, taking account of Definition 3.3.2.

3.4.4. Let be an -group whose structural morphism is an element of (M). Then if acts freely on the -object , it defines an equivalence relation of type (M). Indeed, by (a) the fiber product exists and is an element of (M). One says that the action of is (M)-effective if the equivalence relation defined in by this action is (M)-effective.

Proposition 3.4.5 ((M)-effectivity and base change). Let be an (M)-effective equivalence relation in over . Set . Let be a change of base such that exists. Then exists, is an (M)-effective equivalence relation in over and .

Indeed, the canonical morphisms and are elements of (M), hence by (a′) and are representable. By associativity of the product, is the equivalence relation defined in by the canonical morphism which is an element of (M), whence the conclusion.

Proposition 3.4.6 ((M)-effectivity and cartesian products). Let (resp. ) be an (M)-effective equivalence relation in (resp. ) over . If exists, then exists, is an (M)-effective equivalence relation in over and

(X ×_S X′)/(R ×_S R′) ≃ (X/R) ×_S (X′/R′).

Set , . By (a′), the fiber product exists and the canonical morphism

q : X ×_S X′ → Y ×_S Y′

is an element of (M). Now the formula

(X ×_S X′) ×_{(Y ×_S Y′)} (X ×_S X′) ≃ (X ×_Y X) ×_S (X′ ×_{Y′} X′)

(all products without subscript are taken over ) shows that is the equivalence relation defined by in , which completes the proof.

3.4.7.18 Suppose that has a final object and let be a morphism of -groups, such that . Then, by 3.4.1 (a), the kernel is representable by , and the morphism belongs to (M).

On the other hand, the equivalence relation defined by is the same as that defined by the action of (say, on the right) on , i.e., it is the image of the morphism , defined set-theoretically by (g, h) ⟼ (g, gh). Consequently, one deduces from 3.3.2.1 the following corollary:

Corollary 3.4.7.1. Suppose that has a final object and let be a morphism of -groups, such that . Then the action of on is (M)-effective and is the quotient .

3.5. Construction of quotients by descent

It frequently happens that one does not know how to construct a quotient directly, but that one does know how to do so after a suitable change of base. The present number gives a criterion useful in this situation.

We have seen in §2.1 the definition of a descent datum on an object over relative to a morphism .

Definition 3.5.1. One says that a descent datum on relative to is effective if equipped with this descent datum is isomorphic to the inverse image on of an object over , equipped with its canonical descent datum.

If is a descent morphism (2.2), then the of the definition is unique up to unique isomorphism. To say that is an effective descent morphism (2.2), is to say that it is a descent morphism and that every descent datum relative to this morphism is effective.

Consider now an equivalence relation in an object over . Let (resp. , resp. ) be the inverse images of on , and and let be the equivalence relations deduced from by inverse image. Suppose that the equivalence relation in is (M)-effective, and consider the quotient , which is an object over . Its two inverse images on are isomorphic to by 3.4.5. The -object is therefore endowed with a canonical gluing datum. Using likewise the uniqueness of , one sees that this is even a descent datum. (Remark: it has been implicitly supposed in this proof that all the fiber products written existed, which is the case in particular if is squarable, for example a descent morphism).

Proposition 3.5.2. Let be an equivalence relation in the object over . Let be a universal effective epimorphism. Suppose that every -morphism whose inverse image on is in (M) is itself in (M). The following conditions are equivalent:

(i) is (M)-effective in ;

(ii) is (M)-effective in and the canonical descent datum on is effective.

If this is so, the "descended" object of is canonically isomorphic to .

The fact that (i) entails (ii) is none other than the translation, in the language of descent, of 3.4.4. If one shows the converse, the last assertion of the proposition will be a consequence of the fact that a universal effective epimorphism is a descent morphism (2.3), hence that the "descended object" is unique (up to unique isomorphism).

So let us prove (ii) ⇒ (i). Let be the quotient and the descended object. As the canonical morphism is by construction compatible with the descent data (its inverse images on coincide with the canonical morphism ), it comes by inverse image on from an -morphism . Since is an element of (M), it follows from the hypothesis made on the morphism that is also an element of (M). As is compatible with the equivalence relation , is compatible with , again because a universal effective epimorphism is a descent morphism. One thus has a morphism

R ⟶ X ×_Y X.

To prove that is (M)-effective and that is isomorphic to , it suffices to prove that this morphism is an isomorphism. Now it becomes one by extension of the base from to , since is effective; it is therefore an isomorphism for the same reason as before (2.4).

Note that the hypothesis of the text is satisfied if one takes for (M) the family of universal effective epimorphisms and if has fiber products (1.10). One deduces the

Corollary 3.5.3. Suppose that has fiber products (over would suffice). Let be an equivalence relation in over and a universal effective epimorphism. The following conditions are equivalent:

(i) is universally effective in ,

(ii) is universally effective in and the canonical descent datum on is effective.

If this is so, the "descended" object of is canonically isomorphic to .

4. Topologies and sheaves

The notion of sieve, and the presentation of the notion of topology (4.2.1) adopted here (more intrinsic and in many respects more convenient than the one by covering families of [MA]), are due to J. Giraud [AS].

4.1. Sieves

Definition 4.1.1. One calls a sieve of the category a subfunctor of the final functor .

To every sieve of one associates the set of objects of such that , that is, such that the structural morphism factors through . One thus has the equivalences

(+)   X ∈ E(C) ⟺ C(X) = e(X) = {∅}.
      X ∉ E(C) ⟺ C(X) = ∅.

The set enjoys the following property:

(++)   If X ∈ E and if Hom(Y, X) ≠ ∅, then Y ∈ E.

(Note that if one equips the set Ob C with its natural preorder structure ( dominating if there exists an arrow from into ), the sets satisfying (++) are the subsets of Ob C that contain every dominator19 of one of their elements.)

Conversely, if is a subset of Ob C enjoying property (++), then is written in a unique way in the form and is defined by the formulas (+). There is therefore a one-to-one correspondence between sieves of and subsets of Ob C satisfying condition (++). By abuse of language, we shall sometimes say that the set is a sieve of .

20 Let and be two sieves of ; since they are two subfunctors of the final functor , it amounts to the same to say that dominates (for the relation of domination in ), or that is a subfunctor of , or yet that ; one then says that is finer than . One sees that this gives an order structure on the set of sieves of . Furthermore, one has and therefore the set of sieves of is filtered for the order relation "being finer".

Every subset of , for example a subset of Ob C, defines a sieve : the set of such that for at least one satisfies condition (++) and defines the sought sieve.

This sieve can also be defined as the image of the family of morphisms in the sense of the following definition:

Definition 4.1.2. Let be a family of morphisms of Ĉ with the same target . One calls image of this family the subfunctor of defined by

S ⟼ ⋃ᵢ Im Fᵢ(S) ⊂ F(S).

Proposition 4.1.3. The formation of the image commutes with base extension: in the preceding notation, denote by the image of the family ; for every

morphism of Ĉ, the image of the family of morphisms is the subfunctor of .

Definition 4.1.4.0.21 Let be a sieve of . If is a subset of Ob C such that , one says that is a base of . Every sieve has a base, for example the set .

We propose to describe the set , where is a sieve of and an object of Ĉ, using a base of . For each pair , one has a diagram in Ĉ:

whence a diagram of sets

Γ(F) = Hom(e, F) —σ→ ∏ᵢ Hom(Sᵢ, F) ⇉_{τ₁, τ₂} ∏_{i,j} Hom(Sᵢ × Sⱼ, F)

such that . One thus has a morphism

Hom(e, F) ⟶ Ker(∏ᵢ Hom(Sᵢ, F) ⇉ ∏_{i,j} Hom(Sᵢ × Sⱼ, F)).

One verifies immediately:

Proposition 4.1.4. One has an isomorphism, functorial in ,

Hom(C, F) ⥲ Ker(∏ᵢ Hom(Sᵢ, F) ⇉ ∏_{i,j} Hom(Sᵢ × Sⱼ, F)),

such that the diagram

Hom(e, F) → Ker(∏ᵢ Hom(Sᵢ, F) ⇉ ∏_{i,j} Hom(Sᵢ × Sⱼ, F))
                        ↑ ≀
Hom(e, F) →            Hom(C, F),

where the last line is induced by the canonical morphism , is commutative.

Corollary 4.1.5. Suppose that the fiber products are representable, for example

that the Sᵢ are squarable. One then has, for every , an isomorphism

Hom(C, F) ⥲ Ker(∏ᵢ F(Sᵢ) ⇉ ∏_{i,j} F(Sᵢ × Sⱼ)).

Remark 4.1.6. Let be a sieve of ; denote by the full subcategory of whose set of objects is and by

the inclusion functor. One has an isomorphism, functorial in ,

Hom(R, F) ⥲ Γ(F ∘ i_R)

such that the diagram

Hom(e, F)  →  Hom(R, F)
   ↕ ≀          ↕ ≀
   Γ(F)    →  Γ(F ∘ i_R),

where the second line is induced by the functor , is commutative.

Definition 4.1.7. Let be a category. One calls a sieve of the object of a sieve of the category .

A sieve of is therefore a sub-Ĉ-object of . To it corresponds canonically a subset of containing the source of every arrow whose target it contains. By abuse of language, such a set will also be called a sieve of .

4.2. Topologies: definitions

Definition 4.2.1. Let be a category. One calls a topology on the datum, for each of , of a set of sieves of , called covering sieves or refinements of ,

the datum satisfying the following axioms:

(T 1) For every refinement of and every morphism , the sieve of is covering ("stability under base change").

(T 2) If R, C are two sieves of , if is covering, and if for every and every morphism , the sieve of is covering, then is a refinement of .22

(T 3) If are two sieves of and if is covering, then is covering.

(T 4) For every , is a refinement of .

One can reformulate these axioms in the following way. Suppose given a topology S ⟼ J(S) on and, for every object of Ĉ, denote by the set of subfunctors of such that for every morphism of Ĉ where is representable, , which is a sieve of , is covering. By virtue of (T 1), this notation is compatible with the preceding one. One will also say that is a refinement of . One verifies immediately that the preceding axioms entail the following properties:

(T′ 0) If are two objects of Ĉ, and if for every and every morphism , , then .

(T′ 1) If , and if is a morphism of Ĉ, then .

(T′ 2) If are three objects of Ĉ, if and , then .

(T′ 3) If are three objects of Ĉ and if , then .

(T′ 4) For every , .

Conversely, if one gives oneself, for every , a set of subobjects of satisfying the properties (T′ 0) through (T′ 4), the map S ⟼ J(S) defines a topology on and the two preceding constructions are inverse to each other.

From (T′ 1), (T′ 2) and (T′ 3)23 results the following property:

(T′ 5) If and are two subobjects of and if , then .

The set , ordered by the relation , is therefore filtered; this remark will be useful later.

4.2.2. One says that the topology defined by is finer than the topology defined by if for every , (it amounts to the same to say that for every , ).

Every set of topologies on has a greatest lower bound: let be an indexing set, and for each , let S ⟼ Jᵢ(S) be a topology on . Set ; it is immediate that one has thus defined a topology on , and that this is indeed the greatest lower bound of the given set.

In particular, let us give ourselves, for each , a set of sieves of . One calls the topology generated by these sets the least fine topology for which the elements of are refinements of for every .

Definition 4.2.3. Let be a family of morphisms of Ĉ. Let be the image (4.1.2) of this family. The family is said to be covering if . A morphism is said to be covering if the family reduced to this morphism is covering.

This definition applies in particular to an inclusion: a sieve of is covering if and only if the canonical morphism is covering.

The axioms (T′ 0) through (T′ 5) entail for covering families the following properties:

(C 0) Let be a family of morphisms of Ĉ. If for every representable base change , the family is covering, then so is the initial family.

(C 1) For every covering family and every morphism , the family is covering ("stability under base change").

(C 2) If is a covering family and if, for each , is a covering family, then the composite family is covering ("stability under composition").

(C 3) If is a covering family, and if is a family of morphisms with target such that for each there exists an such that factors through , then is covering ("saturation").

(C 4) Every family reduced to an isomorphism is covering.

Note that (C 2) and (C 3) also entail:

(C 5) If is a family of morphisms with target such that there exists a covering family such that for every the family is covering, then the family is covering ("a locally covering family is covering").

4.2.4. Conversely, let be a category having fiber products and let us give ourselves, for each , a set of families of morphisms of with target , said to be covering families, the datum satisfying axioms (C 1) through (C 4) (hence also (C 5) which is a consequence). For every , let be the set

of sieves of having a covering base (or, what amounts to the same by (C 3), all of whose bases are covering). Then S ⟼ J(S) defines a topology on . The two preceding constructions are inverse to each other.

In fact, in applications, it is impractical to consider all covering families, since one sometimes has fairly simple descriptions of a "sufficient" number of such families. This leads to posing the following definitions.

Definition 4.2.5.0.24 Let be a category. Suppose given, for each , a set of families of morphisms of with target . One calls the topology generated by the least fine topology for which the given families are covering.

Definition 4.2.5. Let be a category. One calls a pretopology on the datum for each of a set of families of morphisms with target , said to be covering for the pretopology under consideration, satisfying the following axioms:

(P 1) For every family and every morphism , the fiber products exist and .

(P 2) If and if for each , , then the composite family belongs to .

(P 3) Every family reduced to an isomorphism is covering.

Proposition 4.2.6. For every , let be the set of sieves of covering for the topology generated by the pretopology . Let be the part of formed by the sieves defined by the families of . Then is cofinal in : every refinement of contains a sieve defined by a family of .

For every , let be the set of sieves of containing a sieve of . One evidently has . To show that , it suffices to show that the make a topology on , that is, that they satisfy axioms (T 1) through (T 4). Now (T 1), (T 3), (T 4) are evidently satisfied. It remains to verify (T 2).25

So let be an element of and a sieve of ; one supposes that for every , the sieve is in and one must prove that . By definition of , contains a refinement defined by a family . Since one has verified (T 3), it suffices to prove that , so one may suppose that . By hypothesis, for every , ; there therefore exists, for each , a covering family such that factors through . The morphism therefore factors through , which shows that contains the sieve defined by the composite family , and one has finished by (P 2).

The axioms (P 1) through (P 3) are those of [MA]. Given the practical interest of pretopologies, we shall interpret each important result with the aid of a pretopology defining the given topology.

Remark 4.2.7. One can introduce a somewhat more general notion: one gives, for each , a set of covering families satisfying (P 1), (P 3) and Proposition 4.2.6. This presents itself in particular when the given families satisfy (P 1), (P 3) and (C 5). The reader may consult [D].

Definition 4.2.8. Let be equipped with a topology, and let be an object of . Let be a relation involving an argument . Suppose that entails . One says that is true locally on for the topology under consideration, if the following equivalent conditions are satisfied:

(i) The set of such that is true is a refinement of .

(ii) There exists a refinement of such that is true for every of this refinement.

(iii) (If the given topology is defined by a pretopology). There exists a covering family for this pretopology such that is true for every of this family.

Example 4.2.9. Let be an -morphism. One will say that is locally an isomorphism if there exists a covering family such that for every ,

is an isomorphism. It amounts to the same to require that there exist a refinement of such that for every , is an isomorphism.

One will see in the sequel many other examples of "local" language.

4.3. Presheaves, sheaves, sheaf associated to a presheaf

Definition 4.3.1. Let be a category. One calls a presheaf of sets on any contravariant functor from into the category of sets. The category is called the category of presheaves on . If is equipped with a topology, one says that the presheaf is separated (resp. is a sheaf*) if for every and every , the canonical map*

(+)    P(S) = Hom(S, P) ⟶ Hom(R, P)

is injective (resp. bijective). One calls category of sheaves, and denotes by , the full subcategory of Ĉ whose objects are the sheaves.26

Proposition 4.3.2. Let be a separated presheaf (resp. a sheaf). For every functor and every , the canonical map

(+)    Hom(H, P) ⟶ Hom(R, P)

is injective (resp. bijective).

Indeed, let be a separated presheaf, a presheaf, , and such that . For every , , is a refinement of and :

R ×_H S —j_S→ S
   ↓f_R         ↓f
   R    —j→   H   ⇉_{u, v} P.

Since is separated, one deduces . This being true for every representable , one has .

Suppose now that is a sheaf. Let ; let us show that it factors through . For every , , factors

in a unique way through , hence defines a morphism , which is evidently functorial in , by uniqueness:

R ×_H S —f_R→ R —g→ P
   ↓j_S          ↓j   ↗h
   S      —f→   H.

One has thus defined for every a map from into functorial in , hence a morphism from into that answers the required conditions.

Corollary 4.3.2.1.27 Let R, F be two sheaves. If is a refinement of , then .

Indeed, suppose that is a refinement of and denote by the inclusion . By 4.3.2, one has , hence there exists such that . One has likewise , and the equality entails , hence is an isomorphism.

Proposition 4.3.3 ([AS], 1.3). Let be a category. Let be a presheaf on ; for every , denote by the set of sieves of such that for every , the map

(+)    Hom(T, P) ⟶ Hom(R ×_S T, P)

is injective (resp. bijective). Then the define a topology on , i.e. satisfy axioms (T 1) through (T 4).

Corollary 4.3.4. Let, for every , be a family of sieves satisfying (T 1). Let be a presheaf on . For it to be separated (resp. a sheaf) for the topology generated by the , it is necessary and sufficient that for every and every , the canonical map

(+)    Hom(S, P) ⟶ Hom(R, P)

be injective (resp. bijective).

Corollary 4.3.5. Let, for each , be a set of families of morphisms of with target , satisfying (P 1) (for example defining a pretopology). Let be a presheaf on . For to be separated (resp. a sheaf) for the topology generated by , it is necessary and sufficient that for every and every family , the map

be injective, (resp. the diagram

P(S) ⟶ ∏ᵢ P(Sᵢ) ⇉ ∏_{i,j} P(Sᵢ ×_S Sⱼ)

be exact).

Definition 4.3.6. Let be a category. One calls the canonical topology on the finest topology for which all representable functors are sheaves.

Corollary 4.3.7. For a sieve of to be a refinement for the canonical topology, it is necessary and sufficient that for every morphism of and every , the canonical map

Hom(T, X) ⟶ Hom(R ×_S T, X)

be bijective.

Definition 4.3.8. A sieve covering for the canonical topology will be called a universal effective epimorphic sieve.

Corollary 4.3.9. A universal effective epimorphic family defines a universal effective epimorphic sieve. Conversely, every squarable family defining a universal effective epimorphic sieve is universal effective epimorphic.

Let us return to the case where is equipped with an arbitrary topology and pass to the construction of the sheaf associated to a presheaf . Let be an object of . If are

two refinements of , one has a diagram

Hom(S, P) ⟶ Hom(R, P)
                 ↘
              Hom(R′, P).

The ordered set is filtered, as has already been remarked. Since is an element of , one has an evident morphism

Hom(S, P) ⟶ lim→_{R ∈ J(S)} Hom(R, P).

Definition 4.3.10.0.28 One sets . One verifies that depends functorially on , hence defines a functor LP by

(++)   Hom(S, LP) = Ȟ⁰(S, P) = lim→_{R ∈ J(S)} Hom(R, P).

One has by construction morphisms

ℓ_P : P ⟶ LP
z_R : Hom(R, P) → Hom(S, LP).

Lemma 4.3.10. (i) For every refinement of and every , the diagram

is commutative.

(ii) For every morphism , there exists a refinement of and a morphism with .

(iii) Let be a functor and such that . Then the kernel of the pair is a refinement of .

(iv) Let and ; for , it is necessary and sufficient that there exist a refinement of such that and coincide on .

Proof. (i): One must verify that . For this, it suffices to verify that the composites of these two morphisms with every morphism , where is representable, are equal. Now consider and the fiber product :29

P  —ℓ_P→  LP
↑u           ↑z_R(u)
R  —i_R→  S
↑p           ↑f
R′  —∼→   T (via i_{R′}, g)

By definition of , (this is the particular case of what one is trying to prove in which is an isomorphism), and now .

(ii) and (iv) merely translate the definition of as a direct limit.

(iii): If denotes the kernel of the pair , then for each morphism where is representable, is a subfunctor of the kernel of the pair of arrows . One is therefore reduced, by (T′ 0), to proving the assertion in the case where is representable. But in this case, it follows from (ii) and (iv) that contains a refinement of hence is a refinement of .

One verifies finally that P ⟼ LP defines a functor

and P ⟼ ℓ_P a morphism of functors

Let us now state the essential result:

Proposition 4.3.11. (i) If is any presheaf, LP is separated and is covering (4.2.3).

(ii) If is a sheaf, is an isomorphism.

(iii) For every presheaf and every separated presheaf (resp. sheaf) , the map

Hom(ℓ_P, F) : Hom(LP, F) → Hom(P, F)

is injective (resp. bijective).

(iv) If is separated, is a covering monomorphism (hence is a refinement of LP), and LP is a sheaf.

Proof. (i) First, is covering; indeed, for every morphism , there exists, by 4.3.10 (i) and (ii), a refinement of such that one has a commutative diagram

P  —ℓ_P→  LP
↑v′          ↑v
P ×_{LP} S  ←  S
↑                  ↑i_R
R

hence is covering. It follows, by (C 0), that is covering.

If on the other hand two morphisms induce the same morphism on a refinement of , let us show that they are equal. There exist refinements Rᵢ, , and morphisms such that . Taking small enough, one may suppose . It then follows from the commutative diagram of 4.3.10

(i) that . By loc. cit. (iii), and therefore coincide on a refinement of , hence a refinement of , which entails that , by loc. cit. (iv).

(ii) is clear, for if is a sheaf, is already an isomorphism for every refinement of .

(iii) Let and be two morphisms such that . To show that , it suffices to see that for every where is representable. Now there exists a refinement of and a morphism with . Then uf and vf coincide on with , hence coincide on . If is separated, one therefore has . Suppose now that is a sheaf; one then has the commutative diagram

P  —ℓ_P→  LP
↘             ↓
   F   →    LF

which shows that is surjective.

(iv) Let us show that if is separated, is a monomorphism. For this, it suffices to see that for every pair of morphisms (where is representable) such that one has . Now loc. cit. (iii) shows that and coincide on a refinement of , hence coincide because is separated. This shows that is a monomorphism; as it is covering by (i), one obtains that is a refinement of LP.

Let us finally show that LP is a sheaf. As we already know by (i) that it is a separated presheaf, it suffices to see that for every , every refinement of and every morphism , there exists a morphism with . Now is a refinement of , since is a refinement of LP, hence is

a refinement of . Set :

P  —ℓ_P→  LP
↑h′           ↑u (via h)
R′  —j→   R   —i_R→  S.

One has , whence . Since is a refinement of and since LP is separated, 4.3.2 shows that .

Corollary 4.3.12.30 Let be a sheaf and a sub-Ĉ-object of . Then is a separated presheaf, is a covering monomorphism, and one has a commutative diagram

R  —i→  F
↘         ↗ j
   LR.
↑ℓ_R

Consequently, is a refinement of if and only if is an isomorphism.

We have already noted that is separated and that is a monomorphism, cf. N.D.E. (24) and (26). By 4.3.11 (iv), is a covering monomorphism. Therefore, if is an isomorphism, is a refinement of . Conversely, if is covering, so is , hence it is an isomorphism by 4.3.2.1.

Remark 4.3.13. If is a cofinal subset of , one has

Hom(S, LP) = lim→_{R ∈ J′(S)} Hom(R, P).

In particular, let S ⟼ R(S) be a pretopology generating the given topology. The functor can be described using the covering families that are elements of . Making the formula above explicit, one recovers the construction of [MA].

Denote by the inclusion functor . From Proposition 4.3.11 results the following theorem:

Theorem 4.3.14. There exists a unique functor such that the following diagram is commutative

i.e., for every presheaf , is a sheaf. The functors and are adjoint to one another: for every presheaf and every sheaf one has an isomorphism, functorial in and ,

Hom_Ĉ(P, i(F)) ≃ Hom_C̃(a(P), F),

that is,

Hom(P, F) ≃ Hom(a(P), F).

Definition 4.3.15. The sheaf is said to be associated to the presheaf .

Remark 4.3.16. As the functor is constructed using inverse limits and filtered direct limits, it commutes with finite inverse limits.31

Moreover, if one identifies with , the morphism is identified with . It follows for example that if is a presheaf of groups, LP is also canonically equipped with a structure of presheaf of groups and the canonical morphism is a morphism of groups. The same holds for the functor , which shows that if is a presheaf of groups and a sheaf of groups, one has an isomorphism

Hom_{Ĉ-gr.}(P, i(F)) ≃ Hom_{C̃-gr.}(a(P), F).

See [D] for more details.

4.3.17. If is any category, one calls a presheaf on with values in a contravariant functor from into . To define sheaves with values in , we must first recall the definition of the inverse limit of a functor. If and are two categories, and

a contravariant functor from to , one denotes by the object of defined as follows:

lim← F(X) = Hom_V̂(X, lim← F) = lim←_{U ∈ Ob R} Hom_V(F(U), X) = Hom(c_X, F),

where is a variable object of , where denotes the contravariant functor from to that sends each object of to and each arrow of to , and where the last Hom is taken in the category . If has a final object , one has . If is the category of sets, the functor is identified with the functor .

If is an object of and a sieve of , denote by the full subcategory of whose set of objects is and the canonical functor. If is a presheaf on with values in , it defines by restriction a functor . The functor induces a morphism of :

P(S) = P_S(S) = lim← P_S ⟶ lim←(P_S ∘ i_R).

One denotes by the object of . By virtue of 4.1.6, Definition 4.3.1 generalizes to the

Definition 4.3.18. The presheaf on with values in is said to be separated (resp. a sheaf*), if for every and every , the canonical morphism of *

is a monomorphism (resp. an isomorphism).

In the case where is the category (Gr.) of groups (or any other category of sets equipped with algebraic structures defined by finite inverse limits), one can see (cf. [D]) that there is equivalence between the following notions: a presheaf on with values in (Gr.) whose underlying presheaf of sets is a sheaf, and a group in the category of sheaves of sets. Taking account of these identifications, we shall always consider sheaves with values in a category of sets equipped with algebraic structures defined by finite inverse limits

as sheaves of sets, equipped in the category with the corresponding algebraic structure.

4.4. Exactness properties of the category of sheaves

Theorem 4.4.1. (i) Arbitrary inverse limits exist in ; "they are computed in Ĉ", i.e. the inclusion functor commutes with inverse limits: if is an inverse system of sheaves, the presheaf

lim← i(X_α) : S ⟼ lim← X_α(S)

is a sheaf and one has i(lim← X_α) = lim← i(X_α).

(ii) Arbitrary direct limits exist in : if is a direct system of sheaves, one has

lim→ X_α = a(lim→ i(X_α))

where is the presheaf direct limit of the :

lim→ i(X_α) : S ⟼ lim→ X_α(S).

(iii) The functor commutes with arbitrary direct limits and with finite inverse limits.

Assertions (i) and (ii) follow formally from the adjunction formula (4.3.14), and32 the first assertion of (iii) follows from (ii). Finally, the second assertion of (iii) has already been pointed out in 4.3.16.

Scholie 4.4.2. This theorem allows one to use the following method to prove in an assertion bearing simultaneously on arbitrary direct limits and finite inverse limits (for example: "every epimorphism is universally effective", cf. below). One begins by proving the corresponding assertion in the category of sets, then one extends it "argument by argument" to the category

of presheaves. Next, one uses the preceding theorem to pass from the category of presheaves to the category of sheaves. One will see in the sequel many examples of this method (4.4.3, 4.4.6, 4.4.9, etc.).

Let us finally remark that the assertions relative to the category of presheaves are formally corollaries of the assertions relative to the category of sheaves. It suffices in fact to take as topology the least fine topology ("chaotic"), that is, the topology defined by for every ; every functor is indeed a sheaf for this topology.

Proposition 4.4.3. Let be a family of morphisms of sheaves. The following conditions are equivalent:

(i) is an epimorphic family.

(ii) is a universal effective epimorphic family (1.13).

(iii) is covering (4.2.3).

(iv) The sheaf image of (that is, the sheaf associated to the presheaf image of (4.1.2)) is .

The equivalence of (iii) and (iv) follows from 4.3.12. The other equivalences will follow from the following lemmas.

Lemma 4.4.4. Let be a monomorphism of sheaves which is an epimorphism. Then is an isomorphism.

The lemma is first clear in the category of sets. Let us prove it next in the category of presheaves. Consider the presheaf

V : S ⟼ Y(S) ∐_{X(S)} Y(S);

it is the amalgamated sum of and under in the category of presheaves.33 Denote by and the two "coordinate" morphisms . If is an epimorphism in the category of presheaves, then . In this case, for each , the map is surjective; as it is also injective, it is a bijection, and therefore is an isomorphism.

Let us place ourselves finally in the category of sheaves. By 4.4.1 (ii), the amalgamated sum in of the sheaves and under is the sheaf associated to the presheaf . Consider the diagram of morphisms:

X —f→ Y ⇉_{i₁, i₂} V —τ→ Z = a(V).

One has , whence , and therefore , since is an epimorphism in . By point (iii) of the lemma below, the presheaf is separated, hence is a monomorphism (4.3.11 (iv)). Therefore , and we have seen above that this

entails that is an isomorphism. This proves Lemma 4.4.4, once one has verified that is separated.

34 Let be the presheaf that to every associates the empty set; it is an initial object of Ĉ.

Lemma 4.4.5. (i) Suppose that , for every . If is a family of separated presheaves, the direct sum presheaf is separated.35

(ii) Consider an equivalence relation in the category of presheaves:

X ⇉_{u, v} Y

and let be the quotient. If is a sheaf and separated, then is separated.

(iii) Consider an amalgamated sum in the category of presheaves, where and are monomorphisms:

X  —u→   Y
↓u′         ↘
Y′    →    V.

If and are separated, and a sheaf, then is separated.

(i) Set . Let and a refinement of , and let be two elements of such that ; there exist indices i, j such that and . Since , there exists a morphism , with . Then, , and since is the disjoint union of the , this entails . Then, since Xᵢ is separated and a subobject of , the map is injective, and therefore . This proves that is separated.

Let us prove (iii). Consider the morphisms and , and let be the kernel of:

Y × Y′ ⇉_{p, q} V,

where and .

Let . Since and are monomorphisms, one can identify with its image in (resp. ); denote by (resp. ) the complement. Then

is identified with the disjoint union of , and , and one easily sees that the maps and are injective, and that the map

(u × u′)(S) : X(S) ⟶ K(S)

is bijective. Consequently, and are monomorphisms, and is an isomorphism.

Set , and let be a refinement of ; one has a commutative diagram:

Let whose images in coincide. By the definition of , lift to elements of ; denote by their images in . Then have the same image in .

Since and are monomorphisms, the maps and are injective. Therefore, since and are separated, if and both belong to or to , then . Otherwise, one may suppose that and , whence and . But then, since , the morphism factors through . Moreover, since (because is a sheaf), there exists such that and , whence, since and are separated, and , and therefore . This proves that is separated.

Let us prove (ii). Let us first remark that the morphism of presheaves

where denotes the kernel of the pair of morphisms , is an isomorphism. Indeed, for every , is an equivalence relation in , so that the diagram

X(T) —u⊠v→ Y(T) × Y(T) ⇉_{w∘pr₁, w∘pr₂} Z(T)

is exact in the category of sets.

Let now and be two morphisms that coincide on a refinement of . Since , one has ,

by construction of , and therefore there exist morphisms such that .

Then, and therefore, by what precedes, there exists a morphism such that and . Since is a sheaf, there exists such that , and therefore one has in the equalities:

u ψ τ = f₁ τ ,    v ψ τ = f₂ τ.

Since is injective ( being separated), this entails and , whence it follows that . This proves that is separated.

Lemma 4.4.6.36 (i) Let be a family of morphisms of presheaves, and let be the presheaf image. Then the following diagram in Ĉ is exact:

∐ᵢ Fᵢ ×_G ∐ⱼ Fⱼ ⇉_{pr₁, pr₂} ∐ᵢ Fᵢ ⟶ G

i.e., for every presheaf , the following diagram of sets is exact:

(∗)    Hom(G, H) ⟶ ∏ᵢ Hom(Fᵢ, H) ⇉_{p, q} ∏_{i,j} Hom(Fᵢ ×_G Fⱼ, H).

(ii) Every covering family of morphisms of sheaves is universally effective epimorphic.

(i) Let be a presheaf. The map which to a morphism associates the family of morphisms is injective, since for every , is determined by the , since the family is surjective. It is clear that the image of is contained in . Conversely, let be a family of morphisms such that, for every i, j, the diagram below is commutative:

Fᵢ ×_G Fⱼ → Fⱼ
↓             ↓ φⱼ
Fᵢ   —φᵢ→  H.

Then, for every , the map factors uniquely as a map , and this defines a morphism such that . This proves the exactness of the sequence (∗), and point (i).

Let us prove (ii). Since the notion of covering family is stable under base extension, it suffices to show that every covering family is effective epimorphic. So let be a covering family of morphisms of sheaves, and let be the presheaf image of this family. Since the family is covering, so is the monomorphism , hence, by 4.3.12, one has .

On the other hand, since is a monomorphism, the fiber products and are the same. Hence, by (i), the following diagram of sets is exact, for every presheaf :

(∗∗)   Hom(G, H) ⟶ ∏ᵢ Hom(Fᵢ, H) ⇉_{p, q} ∏_{i,j} Hom(Fᵢ ×_F Fⱼ, H).

If moreover is a sheaf, one has

Hom(G, H) = Hom(a(G), H) = Hom(F, H),

and then (∗∗) shows that is indeed an effective epimorphic family in the category of sheaves.

Lemma 4.4.7. Every family of morphisms of sheaves factors as a covering family and a monomorphism .

It suffices in fact to take for the sheaf image of the given family.

Proof of Proposition 4.4.3: one has seen in 4.4.6 that (iii) ⇒ (ii), and one evidently has (ii) ⇒ (i). Let finally be an epimorphic family; by Lemma 4.4.7, it factors as a covering family followed by a monomorphism. But the latter, being dominated37 by an epimorphic family, is an epimorphism, hence an isomorphism by 4.4.4.

Remark 4.4.8.38 As the presheaf image of the family is separated, the construction of the associated sheaf shows that the conditions of Proposition 4.4.3 are also equivalent to the following:

(v) For every , every is locally in the image of , that is:

(vi) For every and every , the set of such that the image of in is in the image of one of the is a refinement of .

(vii) (If the topology is defined by a pretopology ). For every and every , there exists a family such that for every the image fⱼ of in is in the image of one of the .

Remark 4.4.8.bis. If the sheaf is representable, the preceding conditions are also equivalent to:

(viii) The set of (), such that there exist an and a commutative diagram

Fᵢ → F
↑   ↗
T

is a refinement of .

Indeed, if (viii) is satisfied, the presheaf image of the Fᵢ is dominated by39 a refinement of , which entails that the family is covering. Conversely, one applies (vi) to .

This condition is expressed in pictorial language in the following way: locally on , there exists an such that has a section. In particular, a morphism

, where is a sheaf and a representable sheaf, will be covering if and only if it has locally (on ) a section.

40 The following lemmas will be useful in VI_B, 8.1 and 8.2.

Lemma 4.4.8.1. Let be presheaves of groups on , being normal in . Suppose that is separated, and that is a sheaf. Then the presheaf of groups is separated. Consequently, one has

a(P/Q) = L(P/Q) = lim→_{R ∈ J(S)} (P/Q)(R).

It suffices to show the first assertion, since the second follows from it, by 4.3.11 (iv) and (ii). Let and a sieve of , and let whose image in is the identity. One must show that . Now, by hypothesis, the morphism (resp. ) is injective (resp. an isomorphism), and in the commutative diagram below, the top row is exact:

The result will follow from this, if one shows that the bottom row is exact. Let whose image in is the identity, and let with . Then is the identity of , i.e. . Therefore, φ ⟼ f ∘ φ is a functorial map , hence defines a morphism of functors such that , whence . This proves the exactness of the bottom row, and the lemma is proved.

Lemma 4.4.8.2. Let be presheaves of groups on .

(i) If is normal in , then is normal in .

(ii) If is central in , then is central in .

Let ; it must be shown (cf. I 2.3.6) that is normal (resp. central) in . Let and ; there exist a sieve of and elements , , such that and (notation of 4.3.10). Since is a morphism of groups, one has .

In case (i), one has , whence ; in case (ii), and therefore .

Proposition 4.4.9. Every equivalence relation in is universally effective (3.3.3): let be a -equivalence relation in the sheaf ; then the sheaf associated to the separated presheaf

i(X)/i(R) : S ⟼ X(S)/R(S)

is a universal effective quotient of by .

Let be the sheaf quotient of by , which exists by 4.4.1 (ii): . We must show that is a universal effective epimorphism, and that the morphism is an isomorphism. The first assertion has already been proved (4.4.3). As for , it comes by application of the functor from the morphism or, since is separated (4.4.5 (ii)) so that is a monomorphism, from the canonical morphism .

One is therefore reduced to proving the same assertion in the category of presheaves. But is the presheaf S ⟼ X(S)/R(S) and one is reduced to proving the analogous assertion in the category of sets, where it is immediate.

Proposition 4.4.10. Under the conditions of 4.4.9, let be a subsheaf of . Denote by R_Y the equivalence relation induced in by . Then the canonical morphism (3.1.6)

is a monomorphism: it identifies with a subsheaf of , which is the sheaf image of the composite morphism

The morphism of presheaves

is a monomorphism. Since the functor is left exact (4.3.16), it transforms monomorphisms into monomorphisms, and therefore

is a monomorphism. The last assertion follows from the commutative diagram

and the fact that is covering.

By virtue of this proposition, we shall always identify with a subsheaf of .

Proposition 4.4.11. Let be a -equivalence relation in the sheaf . For every subsheaf of stable under , denote by the quotient considered as a subsheaf of . Then , and the maps Y ⟼ Y/R_Y and Y′ ⟼ Y′ ×_{X′} X realize a bijective correspondence between the set of subsheaves of stable under and the set of subsheaves of .

If is a subsheaf of , then is a subsheaf of stable

under .41 If is obtained by passage to the quotient from a subsheaf of , then is a subobject of . It therefore suffices to show that if one has two subsheaves and of , stable under , containing , and if the quotients and are identical, then . One is evidently reduced to proving the same assertion in the case where . Denoting then by (resp. ) the presheaf (resp. ), the diagram

is cartesian. Since one has a commutative diagram

and since is covering (4.3.11), the monomorphism is covering, hence is a refinement of . By base change, is a refinement of . Since and are sheaves, this entails (4.3.12) .

4.4.12. In particular, if is a subsheaf of , and if , then the preceding correspondence defines a subsheaf Ȳ of , stable under , containing and minimum for these properties, which one calls the saturation of for the equivalence relation .

Footnotes (chunk-A)

4.5. The case of a topology coarser than the canonical topology

By 4.3.6 and 4.3.8, the following conditions are equivalent for a topology on :

(i) is coarser than the canonical topology of .

(ii) Every representable presheaf is a sheaf for .

(iii) Every covering sieve for is universal effective epimorphic.

If is defined by a pretopology , these conditions are further equivalent to

(iv) Every family belonging to is universal effective epimorphic.

In the case where these conditions are satisfied, the canonical functor factors through a functor (we shall also write 42).

Proposition 4.5.1. The functor is fully faithful and commutes with arbitrary inverse limits. It is in particular left exact and therefore preserves the algebraic structures defined by finite inverse limits.

This follows at once from consideration of the commutative diagram

        C
       ╱ ╲
    j ╱   ╲ (canonical)
     ↙     ↘
    C̃  ─── i ───→  Ĉ

and from 4.4.1 (i).

Before exhibiting other properties of the functor , we need to define the topology induced on a category . No longer assuming the given topology necessarily coarser than the canonical one, this is done as follows: if is a sieve of in and one has a morphism , then defines naturally a sieve of in , denoted (since the definition of a sieve of depends only on the category ). If, for example, is defined by the family , then its image in is defined by the same family considered as a family of morphisms of . This being said, the map defines a topology on called the topology induced by the given topology. With the definitions of [AS], 2.3, it is the coarsest of the topologies on for which the canonical functor

is a comorphism43. One will note that the identifications

(C/S)/T = C/T

respect the topologies by definition.

Proposition 4.5.2. Let be a representable sheaf on and a morphism of Ĉ. In order for to be a separated presheaf (resp. a sheaf) on , it is necessary and sufficient that be a separated presheaf (resp. a sheaf) on .

For every functor , one has (I 1.4.1)

Hom(P, F) = ⨆_{f ∈ Hom(P, S)} Hom_f(P, F).

44

Let and a covering sieve. As is a sheaf, the map establishes a bijection . Consequently, the map

Hom_Ĉ(S′, F) ⟶ Hom_Ĉ(C′, F)

decomposes as a "disjoint union" of the maps

Hom_f(S′, F) ⟶ Hom_{f ∘ η}(C′, F).

The proposition follows.

Corollary 4.5.3. The topology induced on by a topology on coarser than the canonical topology of , is coarser than the canonical topology of .

Corollary 4.5.4. Suppose the given topology on is coarser than the canonical topology. For every , one has an equivalence of categories

The following diagrams are commutative up to isomorphism (all unlabelled arrows being equivalences):

              (j_C)/S                (i_C)/S̃                (a_C)/Ŝ
    C/S ─────────────→  C̃/S̃ ───────────────→  Ĉ/Ŝ ────────────────→  C̃/S̃
      ║                  ≀                     ≀                       ≀
      ║       j_{C/S}            i_{C/S}                 a_{C/S}
    C/S ──────────────→  C̃/S ───────────────→  Ĉ/S ────────────────→  C̃/S
        T̃
       ╱ ╲
      ╱   ╲
   ╱        ↘
  C̃/S̃     C̃/S / T̃
   ↘ ≃   ↗
      ╲ ╱
      C̃/T̃ ─→ C̃/T

The commutativity of the first two squares results from the definition of the equivalence . To prove the commutativity of the last, one must see that the following square is commutative:

       L′
Ĉ/Ŝ ────→  Ĉ/Ŝ
  │           │
  │           │
  ↓    L″     ↓
Ĉ/S ────→  Ĉ/S ,

where is the restriction of the functor L_C to and the functor ; this is easily seen by going back to the definition of the functors (given after 4.3.9). As for the second diagram, this is none other than the restriction to the sheaf categories of the corresponding diagram on the categories of presheaves (Exposé I, n° 1), which is commutative.

Scholie 4.5.5. The various assertions of this number show that, in the case where the given topology is coarser than the canonical topology, one may identify with a full subcategory of , itself a full subcategory of Ĉ, and that under this identification one may indulge in the usual abuses of language regarding , justified by the commutativities above. Let us explicitly remark that the first diagram of 4.5.4 shows that one may use the functor without special precaution.

We shall see in the next number that the identification of with a full subcategory of (contrary to what was the case for Ĉ) commutes with the formation of certain direct limits, and we shall then say how to use this fact.

From now on, and unless explicitly stated otherwise, we shall assume the given topology coarser than the canonical topology and we shall systematically make the identifications described above.

Proposition 4.5.6. Let and be two sheaves over and an -morphism. The following conditions on are equivalent:

(i) is a monomorphism (resp. an epimorphism, resp. an isomorphism) in .

(ii) is a monomorphism (resp. an epimorphism, resp. an isomorphism) in .

For monomorphism and isomorphism, this is evident (it is a question of presheaves). For epimorphism, it follows from the description of epimorphisms as covering morphisms (4.4.3) and from the fact that, by definition of the induced topology, these are the same in and .

Proposition 4.5.7. Let be a morphism of sheaves. The following conditions are equivalent:

(i) is a monomorphism (resp. an epimorphism, resp. an isomorphism).

(ii) For each , is locally a monomorphism (resp. an epimorphism, resp. an isomorphism), that is:

(iii) For each , the set of such that is a monomorphism (resp. an epimorphism, resp. an isomorphism) is a refinement of .

If the given topology is defined by a pretopology , these conditions are further equivalent to the following:

(iv) For each , there exists a covering family such that for every , is a monomorphism (resp. an epimorphism, resp. an isomorphism).

If the category has a final object , one may content oneself with taking in conditions (ii), (iii), and (iv).

One obviously has (ii) ⇔ (iii) ⇔ (iv). To prove the equivalence of (i) and (ii) as well as the supplement concerning the final object, one must show that (ii) ⇒ (i) and that the notions in question are stable under base extension. Let us first prove this last point. For monomorphism and isomorphism, this is evident (it is a question of presheaves). For epimorphism, it follows from the fact that every epimorphism of sheaves is universal (4.4.3).

Let us finally show that (ii) entails (i). Suppose first that is locally a monomorphism (resp. an isomorphism). There then exists a covering sieve of such that for every , is a monomorphism (resp. an isomorphism). As an inverse limit of monomorphisms (resp. isomorphisms) is one, will be a monomorphism (resp. an isomorphism) (cf. 4.1.4). The commutative diagram

            Hom(C, f)
Hom(C, F) ────────────→ Hom(C, G)
   ↑                       ↑
   ≀                       ≀
            f(S)
   F(S) ─────────────→ G(S)

then shows that is injective (resp. bijective). Suppose finally that is locally an epimorphism and let be its image. For each , is the image of . To show that is an epimorphism, one must show that is a refinement of , that is, that is a refinement of for each . But since this is so after every base change of a refinement of (since is locally covering), is indeed a refinement of (Axiom (T 2)).

Corollary 4.5.8. Let and be two sheaves over and an -morphism. In order for to be a monomorphism, resp. an epimorphism, resp. an isomorphism, it is necessary and sufficient that it be so locally on .

Remark 4.5.9. The proof of the proposition shows that it remains valid, for the part concerning monomorphisms (resp. isomorphisms), when one only assumes that is a separated presheaf (resp. a sheaf) and an arbitrary presheaf (resp. a separated presheaf).

Let us provisionally return to the case of an arbitrary topology and lay down a definition.

Definition 4.5.10. Let be a morphism of Ĉ. We say that is a relative sheaf above if for every -functor and every refinement of , the canonical map

(+)    Hom_F(H, G) ⥲ Hom_F(R, G)

is bijective.

Proposition 4.5.2 immediately generalizes:

Proposition 4.5.11. If is a sheaf, is a relative sheaf above if and only if it is a sheaf.

Lemma 4.5.12. In the situation (where X, T, S are three objects of Ĉ), if is a relative sheaf above , then is a relative sheaf above .

Indeed, one has for every -functor

Hom_S(Y, U) = Hom_T(T ×_S Y, X).

If is a sieve of , then is a sieve of ; one concludes at once.

Corollary 4.5.13. The presheaves , , etc., are sheaves when the arguments appearing in them are also.

Indeed, all these presheaves are constructed by means of fibered products and presheaves (I 1.7 and II 1). It therefore suffices to verify the result for a presheaf ; in this case, the assertion follows from 4.5.11 and 4.5.12.

4.6. Description of the quotient of a sheaf by an equivalence relation

Recall that we are assuming the topology given to be coarser than the canonical topology.

Proposition 4.6.1. Let (with morphisms ) be a -equivalence relation in the sheaf . Let be defined as follows: for each of ,

F(S) = { sub-S-sheaves Z of X_S stable under R × S[^N.D.E-IV-43], whose quotient by R_Z is S,
         i.e. such that the diagram R_Z ⇒ Z → S is exact }.

Then for every sheaf , is identified with the set:

{ sub-Y-sheaves of X × Y stable under R × Y and whose quotient is Y }.

In particular the sub-sheaf of corresponds to an element of and the diagram

R ⇒ X →^p F

is exact, hence identifies with the sheaf-quotient .

Indeed, set . For every sheaf and every morphism corresponding to a section , consider the diagram

R × Y ⇒ X × Y ──→ Q × Y
                    ↑     ↑
(∗)                 │     │ s
                    │     │
                    Z ──→ Y

where the square is cartesian. It is immediate by 4.4.11 that is a sub--sheaf of , stable under , whose quotient is , and that, conversely, every of this type comes from a unique section of over . Taking first representable, one extracts an isomorphism . Taking next arbitrary, one extracts the announced form of . Considering finally the canonical morphism , one sees at once that it corresponds to the sub--sheaf of , which completes the proof.

Corollary 4.6.2. Let be an arbitrary subfunctor of such that contains . Then the canonical morphism factors through .

Since is covering (4.4.9 and 4.4.3), it follows that is a refinement of . In particular, every sub-sheaf of satisfying the preceding condition is equal to (4.3.12).

4.6.3. We are now going to interest ourselves in the case where and are representable. Let us first introduce some terminology. Besides the conditions (a) to (d) introduced in 3.4.1, we shall use other conditions on a family (M) of morphisms of , which we state below, recalling the conditions (a) to (c) already given, for completeness.

(a) (M) is stable under base extension.

(b) The composite of two elements of (M) is in (M).

(c) Every isomorphism is an element of (M).

(d_T) Every element of (M) is covering.45

(e_T) Let be a morphism of . If there exists a refinement of such that for every , is an element of (M), then is an element of (M).

Recall that (a) and (b) entail

(a′) The cartesian product of two elements of (M) is an element of (M).

On the other hand, (a) and (d_T) entail by 4.3.9:

(d′) Every element of (M) is a universal effective epimorphism.

4.6.4. The preceding conditions are verified by the family of covering morphisms, denoted (M_T), when has fibered products. Indeed (cf. 4.2.3), (a) follows from (C 1), (b) from (C 2), (c) from (C 4), (d_T) from the definition, (e_T) from (C 5). The results we shall establish for a family satisfying these conditions will apply in particular to the family (M_T). In particular, one may take for the canonical topology and for (M) the family of universal effective epimorphisms.

Lemma 4.6.5. Let (M) be a family of morphisms satisfying the properties (a) to (e_T) above. Let be a -equivalence relation in , of type (M). Let be the sheaf defined by , the -equivalence relation in defined by and the sheaf-quotient. In order for to be (M)-effective, it is necessary and sufficient that be representable. If this is so, is represented by the quotient .

Suppose first that is (M)-effective and denote . The canonical morphism is an element of (M), hence covering by (d_T). The corresponding morphism

is therefore a universal effective epimorphism of (4.4.3), hence identifies with the quotient of by the equivalence relation defined in by . As the canonical functor commutes with fibered products, is none other than , since is the equivalence relation defined by .

Conversely, suppose representable by an object of . Let be the morphism deduced from the canonical morphism ; it is a covering morphism by 4.4.3. It is clear as before that is the equivalence relation defined by . It only remains to show that . Now the cartesian square

R ⥲ X ×_Y X ──→ X
           │      │
        pr_2     p
           ↓      ↓
           X ───→ Y

shows that becomes , which is an element of (M), after base change by the covering morphism . One concludes by (e_T).

Corollary 4.6.5.1.46 Let (M) satisfy the properties (a) to (e_T) above and let be a morphism of -groups, such that . Suppose representable (which is the case if has a final object ). Then the equivalence relation in defined by is (M)-effective and represents the sheaf-quotient for the topology .

This follows from 4.6.5 and 3.4.7.1.

We are now in a position to state the main result of this number.

Theorem 4.6.6. Let (M) be a family of morphisms verifying the axioms (a) to (e_T) of 4.6.3. Let be a -equivalence relation of type (M) (cf. 3.4.3) in the object of . Consider the functor defined as follows:

F(S) = { sub-S-sheaves Z of X_S stable under R × S whose quotient by R_Z is S }.

Let F_0 be the subfunctor of defined as follows: consists of the that are representable, that is:

F_0(S) = { sub-C/S-objects Z of X_S stable under R × S, such that
           R_Z is (M)-effective and has quotient S (i.e. such that
           Z → S belongs to (M) and R_Z ≃ Z ×_S Z) }.

Then:

(i) The morphism defined by the sub-object of identifies with the sheaf-quotient of by .

(ii) The following conditions are equivalent:

  • a) is representable.
  • b) F_0 is representable.
  • c) is (M)-effective.

Under these conditions, .

(iii) Let (N) be a family of morphisms stable under base change, such that for every covering family and every family of morphisms of (N), every descent datum on the relative to is effective. Suppose squarable (cf. 1.6.0) and the morphism an element of (N). Then .

Proof. (i) has already been proved (4.6.1).

(ii) We have seen the equivalence of a) and c) as well as the equality . It remains to prove that b) or c) implies . Let us first remark, as is moreover affirmed in the statement, that F_0 is indeed a subfunctor of ; for every and every , the morphism is squarable, hence is an element of for every . As belongs to , 4.6.2 shows that b) implies .

Now suppose c) verified and let be an object of representing . Then the morphism is an element of (M) and, for every and every , the diagram (∗) of 4.6.1 shows that is representable, and belongs to (M), hence .

(iii) Let correspond to . We must show that factors through F_0, that is, that is representable. This is clear first if factors through , by virtue of:

Lemma 4.6.7. Let . The image of in corresponds to the sub-sheaf of X_S defined by the two cartesian squares

              id_{X_S} × x_0
   X_S ─────────────────→ X_S ×_S X_S ──→ X × X
    ↑                          ↑              ↑
    │                          │              │
    Z ──────────────────→     R_S ─────────→ R

This lemma follows at once from the description of the morphism .

Let us return to the proof of the theorem. If factors through , then is representable and, as is an element of (N), the same holds for .

In general, does not necessarily factor through ; but since is covering (4.4.3), there exists by 4.4.8 (vii) a covering family and for each a morphism making commutative the diagram

X ────→ F
↑       ↑
│       │ f
S_i ──→ S .

By what precedes, the morphism defined by the preceding diagram belongs to and corresponds to the sub-sheaf of . The morphism is an element of (N) and the family covering. It therefore only remains to establish:

Proposition 4.6.8. Let be a covering family and a sheaf above . Suppose that for each , the -functor is representable by an object . Then the family of is equipped with a canonical descent datum relative to . In order for to be representable, it is necessary and sufficient that this datum be effective; if this is so, the "descended" object represents .

Let us first remark that by 4.4.3, is a universal effective epimorphic family in , hence a descent family in (2.3). If is representable by the object , then (considered as a sheaf) is isomorphic to , hence the descent datum on the is effective and the (unique) descended object is isomorphic to . Conversely, suppose that the canonical descent datum on the is effective and let be the descended object. Since the family is a descent family in , there exists an -morphism which by base extension to each recovers the canonical morphism . This morphism is locally an isomorphism; as and are sheaves, it follows from 4.5.8 that it is an isomorphism.

Corollary 4.6.9. Let be an (M)-effective equivalence relation in . For every sheaf , the map

Hom(X/R, F) ⟶ Hom(X, F)

identifies the first set with the part of the second consisting of the morphisms compatible with .

Corollary 4.6.10. Let be a topology coarser than , for which the morphisms of (M) are covering. Under the conditions of 4.6.6 (iii), is also the sheaf-quotient of by in every intermediate topology between and the canonical topology.

Remark 4.6.11. If in the statement of 4.6.6 (iii), one furthermore assumes that, under the hypotheses of the text, if one denotes the descended object, the morphism is an element of (N), then the inclusion morphisms are also elements of (N), as follows at once from the construction of by descent.

Remark 4.6.12. The implications c) ⇒ b) ⇒ a) and c) ⇒ [F_0 = F = X/R] have been established without recourse to the "if" part of Lemma 4.6.5, which is the only place where condition (e_T) is used. They therefore remain valid if (M) satisfies only conditions (a) to (d_T). An example of such a family (M) is that of squarable covering morphisms (compare with 4.6.4). In the case of the canonical topology, these are none other than the universal effective epimorphisms. One therefore has:

Corollary 4.6.13. Let be a universal effective equivalence relation in . Then the object of is the sheaf-quotient of by for the canonical topology. It represents the following functor: is the set of sub--objects of X_S stable under and such that the induced equivalence relation is universal effective and has as quotient.

Similarly, for an arbitrary topology:

Corollary 4.6.14. Let (M) be the family of squarable covering morphisms. If is an (M)-effective equivalence relation in , then the object of is the sheaf-quotient of by and represents the functor F_0 of 4.6.6.

Scholie 4.6.15. We can now bring the following precisions to 4.5.5. Whereas in questions involving exclusively inverse limits (fibered products, algebraic structures, etc.), one may, by the results of Exposé I and 4.5.5, identify indifferently with a full subcategory of or of Ĉ, it is not the same in those that mix inverse and direct limits. In all questions involving both inverse and direct limits, in particular passages to the quotient (example: group structure on the quotient of a group by an invariant subgroup), we shall consider the given category as embedded in the category of sheaves; thus if is a -equivalence relation in the object of , will denote the sheaf-quotient of by (previously denoted ), hence in the case where this sheaf is representable, the object it represents. The preceding results show that in the most important cases, a quotient in will also be a quotient in the category of sheaves; in any case, we forbid ourselves the use of the notation for a quotient in that does not coincide with the quotient in (for example one that is not universal), thus modifying the definitions of n° 3.

To study a problem of the type above, one therefore places oneself first in the category of sheaves, where all the usual results are valid (cf. n° 4.4), then one specializes the results obtained to the original category, using the results of the present number and, when one has them, descent effectivity criteria. We shall see examples of this method in the following numbers.

4.7. Use of effectivity criteria: isomorphism theorem

In this number, we give an example of the use of effectivity criteria. The data of departure are a topology on (always coarser than the canonical topology), a family (M) of morphisms of verifying the axioms (a) to (e_T) of 4.6.3, and a family (N) of morphisms of liable to verify the following axioms:

(a) (N) is stable under base extension.

(f_T) "the morphisms of (N) descend by the given topology"; that is: for every , every covering family and every family of morphisms of (N), every descent datum on the relative to is effective, and if one denotes the descended object, the morphism is an element of (N).

Since every element of (M) is covering (condition 4.6.3 (d_T)), (f_T) entails the following axiom:47

(f_M) If is an element of (N) and an element of (M), every descent datum on relative to is effective; if one denotes Y_1 the descended object, is an element of (N).

Let us at once signal an example of this situation, which will be treated later: is the category of schemes, the faithfully flat quasi-compact topology; (M) the family of faithfully flat quasi-compact morphisms, (N) the family of closed immersions, or that of quasi-compact immersions.48

Let us recall the principal result of 4.6.6 (taking into account 4.6.11):

Proposition 4.7.1. If is a squarable object of , an equivalence relation of type (M) in , such that is an element of (N), with (N) verifying (a) and (f_T), then the sheaf-quotient is defined by

(X/R)(S) = { sub-S-objects Z of X_S, stable under R × S, such that Z → X_S
             belongs to (N), Z → S is covering (or an element of (M)),
             and R_Z ≃ Z ×_S Z }.

Moreover, one has:

Proposition 4.7.2. Let and an (M)-effective equivalence relation in . Let (N) be a family of morphisms verifying (a) and (f_M).

For every sub-object of , stable under and such that belongs to (N), the equivalence relation induced in by is (M)-effective and the quotient is a sub-object of such that belongs to (N).

The map is a bijection between the set of sub-objects of , stable under , such that belongs to (N), and the set of sub-objects of such that belongs to (N). The inverse map is .

Proof. As is (M)-effective, the morphism belongs to (M). Let be a sub-object of such that the canonical morphism belongs to (N). Then, the sub-object of is stable under , and the morphism (resp. ) belongs to (N) (resp. (M)) since (N) and (M) are stable under base change. Let R_Y denote the equivalence relation induced in by . By 4.4.11, the sheaf quotient is represented by and therefore, by 4.6.5, R_Y is (M)-effective.

Conversely, let us show that every sub-object of , stable under , such that the structural morphism belongs to (N), is obtained in this way. Indeed, if is stable under , its two images in are identical and is equipped with a descent datum relative to ; the desired result follows, since the family (N) verifies the axiom (f_M).

Corollary 4.7.3. Let and an (M)-effective equivalence relation in ; assume moreover that belongs to (N), where (N) verifies (a) and (f_T). Then, for every as in 4.7.2, also belongs to (N) and therefore, by 4.7.1, one has:

(Y/R_Y)(S) = { sub-S-objects Z of Y_S, stable under R_Y × S, such that Z → Y_S
               belongs to (N) (then Z → X_S also belongs to it), Z → S is covering,
               and R_Z ≃ Z ×_S Z }.

5. Passage to the quotient and algebraic structures

5.1. Principal homogeneous bundles

Definition 5.1.0.49 We recall (III 0.1) that an object with (right) operator group is said to be formally principal homogeneous50 under if the canonical morphism (of functors)

X × H ⟶ X × X

defined by is an isomorphism. It amounts to the same thing to say (cf. loc. cit.) that for every , is formally principal homogeneous under , that is, empty or principal homogeneous under . In particular, if is made to operate on itself by (right) translations, becomes formally principal homogeneous under itself.

Definition 5.1.1. The object with operator group is said to be trivial if it is isomorphic (as object with operator group ) to on which operates by translations.

Proposition 5.1.2. Let be formally principal homogeneous under . One has an isomorphism

of principal homogeneous sets under .

To every section of one associates the morphism from to defined setwise by . The stated assertion is immediate, by reduction to the set-theoretic case.

Corollary 5.1.3. One has an isomorphism of objects with operators

Corollary 5.1.4. In order for an object with operator group to be trivial, it is necessary and sufficient that it be formally principal homogeneous and possess a section.

Definition 5.1.5. Let be a category equipped with a topology. We say that the -object with -operator group is a principal homogeneous bundle under if it is locally trivial, that is, if the following equivalent conditions are satisfied:51

(i) The set of such that (the functor) is trivial under is a refinement of .

(ii) There exists a covering family (for a pretopology defining the given topology) such that for each , the -functor with -functor-group of operators is trivial (= has a section over ).

Proposition 5.1.6. Let be a category equipped with a topology . Let (M) be a family of morphisms of verifying the axioms (a) to (e_T) of 4.6.3. Let be an -group such that the structural morphism is an element of (M) and an -object with -operator group . The following conditions are equivalent:

(i) is a principal homogeneous bundle under (Definition 5.1.5).

(ii) is formally principal homogeneous under and the structural morphism is an element of (M).

(iii) There exists a morphism element of (M) such that by base extension from to , becomes trivial, that is, is trivial under .

(iv) operates freely on , in an (M)-effective manner, and the quotient is isomorphic to .

Let us first remark that (ii) and (iv) are equivalent, taking into account that, in either case, is an element of (M), hence squarable, which ensures the representability of the fibered products and . It is clear that (ii) entails (iii), as one can take itself as , the hypothesis that is formally principal homogeneous entailing that is trivial under (5.1.4), since it has a section (the diagonal section). It is clear that (iii) entails (i), since is a covering family, by axiom (d_T). It therefore remains to show that (i) entails (ii). The morphism of sheaves is locally an isomorphism, hence an isomorphism (4.5.8); is therefore formally principal homogeneous. The structural morphism is locally isomorphic to the structural morphism which is an element of (M). It is therefore itself an element of (M) by (e_T).

The equivalence between (i) and (iv) generalizes:

Proposition 5.1.7. Under the same hypotheses on and (M), let be an -group and an -object on which operates (on the right). Suppose the structural morphism is an element of (M). The following conditions are equivalent:

(i) operates freely on and in an (M)-effective manner.

(ii) There exists an -morphism compatible with the equivalence relation defined in by the action of and such that the operation of on above thus deduced makes a principal homogeneous bundle under H_Y above .

Under these conditions identifies with the quotient .

If is a morphism compatible with the action of , then the operation of on above thus deduced defines in the same equivalence relation as the action of , by virtue of the formula

H_Y ×_Y X ⥲ H ×_S X.

The proposition follows from this remark and from the equivalence (iv) ⇔ (i) above.

Corollary 5.1.7.1.52 Let be a category having a final object, stable under fibered products, and equipped with a topology coarser than the canonical topology. Let be a morphism of -groups, and . Assume covering for the topology .

Then represents the sheaf quotient , and is a K_H-torsor. (N.B. One will also say that: " is a -torsor above ".)

Indeed, as is covering, it is a universal effective epimorphism (4.4.3), hence by 3.3.3.1, is the quotient of by the equivalence relation . On the other hand, the morphism , is an isomorphism of objects with operator group (its inverse being given by ). Hence, on the one hand, is the equivalence relation defined by ; on the other hand, since the morphism is covering, is a K_H-torsor, by 5.1.6 (ii) (or directly by definition 5.1.5 (ii)).

We can now make Theorem 4.6.6 more precise in the case of passage to the quotient by an operator group:

Proposition 5.1.8. Under the hypotheses of 5.1.7, let F_0 denote the functor over defined as follows: for each , is the set of representable sub--functors of , stable under and being principal homogeneous bundles under this -group for the induced action (3.2.2).

(i) The following conditions are equivalent:

  • a) The operation of on is (M)-effective and free.53
  • b) F_0 is representable.

Under these conditions, one has .

(ii) Let (N) be a family of morphisms, stable under base change, such that for every covering family and every family of morphisms of (N), every descent datum on the relative to is effective. Assume the morphism is an element of (N) and squarable. Then the element of corresponding to the sub-object of identifies F_0 with the sheaf-quotient .

5.2. Group structures and passage to the quotient

In this number we interest ourselves in the algebraic structures one can place on the quotient of a group by a subgroup. We shall first place ourselves in the category of sheaves on for an arbitrary topology. By taking the canonical topology and using 4.5.12, we shall obtain results for the universal effective passage to the quotient in .

Proposition 5.2.1. Let be a monomorphism of sheaves of groups. There exists on the sheaf-quotient a unique structure of object with operator group such that the canonical morphism

is a morphism of objects with operator group . This structure is functorial with respect to the pair : if one has a commutative diagram

H ──→ G
│      │
│      │ f
↓      ↓
H′ ──→ G′ ,

the morphism (3.2.3) is compatible with the morphism on the operator groups.

Indeed, the sheaf is the sheaf associated with the presheaf

as the functor is left exact, it transforms objects with operator groups into objects with operator group. As the presheaf is equipped with a structure of object with operator groups , then is equipped with a structure of object with operators . This structure obviously enjoys all the stated properties.

Corollary 5.2.2. Let be a monomorphism of -groups. Suppose that the operation of on is universal effective. There exists on the object-quotient a unique structure of object with operator group such that is a morphism of objects with operators. This structure is functorial in the pair (with operating in a universal effective manner in ), in the preceding sense.

Proposition 5.2.3. Let be a monomorphism of sheaves of groups identifying with an invariant sub-sheaf of groups of . There exists on the sheaf-quotient a unique structure of sheaf of groups such that the canonical morphism is a morphism of groups. This structure is functorial in the pair (with invariant).

The proof is similar to that of 5.2.1.

Corollary 5.2.4. Let be a monomorphism of -groups identifying with an invariant subgroup of . Suppose that the action of on is universal effective. There exists on the object-quotient a unique group structure

such that the canonical morphism is a morphism of groups. This structure is functorial with respect to the pair ( invariant, operating in a universal effective manner).

One can characterize the group structure of in a more telling manner:

Proposition 5.2.5. Under the conditions of 5.2.4, let be a -group and a morphism. The following conditions are equivalent:

(i) is a morphism of groups compatible with the equivalence relation defined by .

(ii) is a morphism of groups inducing the trivial morphism .

(iii) factors as a morphism of groups .

In particular, one has an isomorphism, functorial in the group

Hom_{C-gr.}(G/H, K) ⥲ {f ∈ Hom_{C-gr.}(G, K) | f ∘ u = e}.

The equivalence of (i) and (ii) is proved set-theoretically. One obviously has (iii) ⇒ (ii). The equivalence of (iii) and (ii) follows from the formula

Hom(G/H, K) ≃ Hom(i(G)/i(H), K)

and from the definition of the group structure of .

Remark 5.2.6. In the preceding situation, if the kernel of is exactly , the morphism that factors is a monomorphism. This follows at once from 3.3.4.

In the case of sheaves of groups, one can make 4.4.11 more precise by means of the

Proposition 5.2.7. Let be a sheaf of groups, an invariant sub-sheaf of groups. For every sub-sheaf of groups of containing , let be the quotient group considered as a subgroup of .

One has , and the maps and realize a bijection between the set of sub-sheaves of groups of containing and the set of sub-sheaves of groups of . In this correspondence, the invariant sub-sheaves of groups of and of correspond.

The first part follows easily from 4.4.11 and 3.2.4. It remains to see that is invariant in if and only if is invariant in . If is invariant in , then the presheaf is invariant in . The same holds for the associated sheaves, by virtue of the usual argument. If conversely is invariant in , then the fibered product is invariant in , as one sees immediately.

If now is an arbitrary sub-sheaf of groups of , let be the saturation of for the equivalence relation defined by ; we shall also write .

Proposition 5.2.8. Under the preceding conditions, is a sub-sheaf of groups of containing and the image of in is identified with

(L · H)/H ≃ L/(H ∩ L).

Indeed, let denote the sheaf image of in . It is a sub-sheaf of groups of corresponding to in the correspondence of the preceding proposition. As the morphism is covering, hence a universal effective epimorphism of sheaves, it follows from 4.4.9 that is identified with the quotient of by the kernel of which is obviously none other than .

Let us finally consider the following situation: we have a sheaf of groups , a sub-sheaf of groups and a sub-sheaf of groups of , invariant in . Let us first define a (right) operation of the sheaf of groups (= ) on . The group operates by right translations on . As is invariant in , this operation is compatible with the equivalence relation defined by the action of and therefore defines an operation of on , that is, a morphism from the opposite group to into . As the latter is a sheaf (4.5.13) and this morphism is trivial on , it factors through and defines the sought operation. As the right and left operations of on itself commute, the operations of and on commute.

Proposition 5.2.9. Under the preceding conditions, operates freely (on the right) on and one has a canonical isomorphism of sheaves with operator group

When is invariant in , in which case is invariant in (5.2.7), this isomorphism respects the group structures of the two sides.

One has an isomorphism of presheaves

which respects the structures of objects with operator group . The announced result is obtained by applying the functor to this relation.

Corollary 5.2.10. Let be a -group, a sub--group of , an invariant sub--group of . Let (M) be a family of morphisms of verifying the axioms (a) to (e_T). Suppose the operation of on (resp. ) on the right is (M)-effective. Then operates in a natural manner freely on the right on ; this operation commutes with that of . The following conditions are equivalent:

(i) The operation of on is (M)-effective.

(ii) The operation of on is (M)-effective.

Under these conditions, one has an isomorphism of objects54 with operator group :

5.3. Use of effectivity criteria: Noether's theorem55

Let , and (M) be as usual. Let (N) be a family of morphisms verifying the axioms (a) and (f_M) of 4.7. Putting together 5.2.7 and 4.7.2, one obtains:

Proposition 5.3.1. Let be a -group. Let be a sub--group of , invariant and operating in an (M)-effective manner in .

For every sub--group of containing and such that the morphism belongs to (N), operates in in an (M)-effective manner and the quotient is a sub--group of such that the morphism belongs to (N).

The map is a bijection between the set of sub--groups of , containing and such that belongs to (N), and the set of sub--groups of such that belongs to (N). The inverse map is . In this correspondence, the invariant subgroups of and correspond.

Corollary 5.3.2. If is an element of (N), then has a final object and the unit section is an element of (N).56

6. Topologies in the category of schemes

6.1. The Zariski topology

This is the topology generated by the following pretopology: a family of morphisms is covering if each morphism is an open immersion and the union of the images of the is all of . It is denoted (Zar).

Definition 6.1.1. A sheaf for the Zariski topology is also called a functor of local nature*: it is a contravariant functor from (Sch) to (Ens) such that for every scheme and every covering of by opens , one has an exact diagram:*

F(S) → ∏_i F(S_i) ⇒ ∏_{i,j} F(S_i ∩ S_j).

In particular, a functor of local nature transforms direct sums into products. As every representable functor is a sheaf, this topology is coarser than the canonical topology.

From the terminological point of view, whenever we say "local", "locally", without precision, it will be with reference to the Zariski topology, hence in the usual sense.

6.2. A procedure for constructing topologies

Proposition 6.2.1. Let be a category, a full subcategory, a set of families of morphisms of with the same target, stable under base change and under composition (i.e. verifying the axioms (P 1) and (P 2) of 4.2.5), a set of families of morphisms of containing the families reduced to an identity isomorphism. Equip with the topology generated by and (cf. 4.2.5.0) and suppose the three conditions below are verified:

(a) If (hence ) and if is a morphism of , then the fibered products (in ) exist and the family belongs to (hence ). (Remark: this condition entails that is stable under base change in , but is not equivalent to it, since it further supposes that the inclusion functor from to commutes with certain fibered products).

(b) For every , there exists with for each .

(c) In the following situation:

              S_{ijk}
                │ (P′)
                ↓
        (P)  S_{ij}
   S_i ←──── │
        (P′) │
             ↓
             S ,

where ; ; for each ; for each ij, there exists a family and for each a multi-index (ijk) and a commutative diagram

S_{ijk} ←──── T_n
       ╲     ╱
        ╲   ╱
         ╲ ╱
          S .

Then, in order for a sieve of to be covering, it is necessary and sufficient that there exist a composite family

R ←┄┄┄ S_{ij}
        │ (P′)
        ↓
   S ←─ S_i
        (P)

where , , for each , and the morphisms so obtained factor through (in other words, the sieve generated by this composite family is contained in ).

Proof. As the families elements of and of are covering, a family composed of such families will be also (C 2), hence a sieve of the indicated form will be covering, since it contains a covering sieve.

Conversely, it suffices to see that the sieves of this form do indeed form a topology, i.e. it suffices to verify the axioms (T 1) to (T 4) of 4.2.1.

Axiom (T 4). Let . There exists by (b) a family with . The families are elements of by hypothesis. The sieve of is therefore of the desired form:

S_i ───→ S
 │  (P′) id    id_S
 ↓  (P)    
S_i ───→ S .

Axiom (T 3). Evident.

Axiom (T 2). Let be a sieve of of the desired form and let be a sieve57 of such that, for every and every morphism factoring through , the sieve of is of the desired form. Then, as factors through , the sieve of :

        S_{ij} ←─── C_{ij}
       ╱  (P′)
      ╱
    S_i
     │ (P)
     ↓
     S ←──── C
     ↑
     R

is of the desired form; hence, for each ij, one has a diagram of the form:

S_{ijkl}
   │ (P′)
   ↓
S_{ijk}
   │
   │ (P)
   ↓
S_{ij} ←──── C_{ij} .

One has therefore proved that there exists a composite family

S_{ijkl} ──(P′)─→ S_{ijk} ──(P)─→ S_{ij} ──(P′)─→ S_i ──(P)─→ S

belonging to , factoring through , and where all objects other than are in . Applying condition (c) to each family , one deduces that for each there exists a family , such that factors through one of the , hence through :

T_{in} ──→ S_{ijkl}
   │ (P′)
   ↓
   S_i
   │ (P)
   ↓
   S ←──── C .

The sieve of is therefore of the desired form, which completes the verification.

Axiom (T 1). Let be a sieve of of the given form and let be a morphism of . Let us show that the sieve of is of the desired form.

       S_{ij} ←──────── U_{ikj}
          │ (P′)              (P′)
          ↓                   ↓
   S_i ←─────────── T_i ←─(P)── U_{ik}
          │                       ╱
          │ (P)            (P)   ╱ (P)
          ↓                     ╱
   R ──→ S ←──────── T ←───────

Let . The family belongs to (by (P 1)). Applying (b), one constructs , with the . By hypothesis (condition (P 2) on ), one has . By (a), is an object of and for each ik, . Then, the commutative diagram below

R ←──────── U_{ikj}
              │ (P′)
              ↓
              U_{ik}
              │ (P)
              ↓
S ←──── T ←── 

shows that the morphisms factor through the sieve of , which is therefore of the desired form, which completes the proof.

Corollary 6.2.2. If and if is a sieve of , is covering if and only if there exists a family , factoring through .

Indeed, such a sieve is covering. On the other hand, it suffices to apply (c) by taking the family reduced to the identity isomorphism of to deduce from the proposition that a covering sieve is of the indicated form.

Corollary 6.2.3. In order for a presheaf on to be separated (resp. a sheaf), it is necessary and sufficient that the morphism

be injective (resp. that the diagram

F(S) ⟶ ∏_i F(S_i) ⇒ ∏_{i,j} F(S_i ×_S S_j)

be exact) in the two following cases:

(i) ,

(ii) ; .

Indeed, the conditions are necessary, since the families in question are covering. If is the sieve of image of a family of morphisms , a straightforward diagram chase shows that the conditions of the corollary entail that is injective (resp. bijective). But every refinement of contains a sieve of the above type and one has a commutative diagram

                f
Hom(S, F) ───────────→ Hom(R, F)
       ╲                  ╱
        ╲ g           h ╱
         ╲             ╱
          ↘           ↙
           Hom(C, F) .

One knows that is injective, hence so is . Therefore is separated. But is a refinement of , hence is also injective. If is bijective, then is also, hence is a sheaf.

Remark 6.2.4. The preceding corollary does not follow from 4.3.5, because is not stable under base extension.

Remark 6.2.5. Condition (c) is verified in particular in the case where

(i) is stable under composition.

(ii) If is a family of morphisms of , element of , there exists a subfamily element of .

6.3. Application to the category of schemes

One takes for the category of schemes, for the full subcategory formed by affine schemes, for the set of surjective families of open immersions. One will consider several sets :

: finite surjective families, composed of flat morphisms.

: finite surjective families, composed of flat morphisms of finite presentation and quasi-finite.58

: finite surjective families, composed of étale morphisms.

: finite surjective families, composed of étale and finite morphisms.

For each of these sets , except , the conditions of Proposition 6.2.1 are verified ((c) thanks to 6.2.5, since an affine scheme being quasi-compact, every family of morphisms of , element of , contains a finite subfamily that is also in , hence in for ). The topology generated by and is denoted and called in the following way:

T_1 = (fpqc) = faithfully flat quasi-compact topology.
T_2 = (fppf) = faithfully flat (locally) of finite presentation topology.
T_3 = (ét)  = étale topology.
T_4 = (étf) = étale finite topology.

As , one has

(fpqc) ≥ (fppf) ≥ (ét) ≥ (étf) ≥ (Zar).

Proposition 6.3.1. (i) In order for the sieve of to be covering for , , it is necessary and sufficient that there exist a covering of by affine opens and for each a family element of , the being affine, such that each morphism factors through .59

(ii) In order for a presheaf on (Sch) to be a sheaf for (fpqc) (resp. (fppf), (ét), (étf)), it is necessary and sufficient that

  • a) be a sheaf for (Zar), i.e. a functor of local nature.
  • b) For every faithfully flat (resp. faithfully flat of finite presentation and quasi-finite, resp. surjective étale, resp. surjective étale finite) morphism , where and are affine, one has an exact diagram:
F(S) ⟶ F(T) ⇒ F(T ×_S T).

(iii) The topologies , , are coarser than the canonical topology.

(iv) Every surjective family formed of flat and open morphisms (resp. flat and locally of finite presentation, resp. étale, resp. étale and finite) is covering for (fpqc) (resp. (fppf), resp. (ét), resp. (étf)).

(v) Every finite surjective family formed of flat and quasi-compact morphisms is covering for (fpqc). In particular, every faithfully flat and quasi-compact morphism is covering for (fpqc).

Proof. (i) follows from 6.2.1, (ii) from 6.2.3, taking into account the fact that a sheaf for the Zariski topology transforms direct sums into products. Every representable functor being a sheaf for (Zar) and verifying condition (b) of (ii) by SGA 1, VIII 5.3, T_1 is coarser than the canonical topology, which proves (iii).

Let us prove (iv). Let be a family of morphisms as in the statement. Considering a covering of by affine opens, one reduces immediately to the case where is affine.60

Let us first treat the case where the morphisms are flat and open (resp. étale). Let be a covering of by affine opens. As the morphisms in question are open, the images of the form an open covering of . As is affine, hence quasi-compact, it is covered by a finite number of opens , for running through a finite set . Then is affine, and the morphism belongs to , resp. , hence is covering. As it factors through the given family, the latter is covering.

In the case (étf), each is finite over hence is affine; in the preceding argument, one can then take the covering of , and one obtains that belongs to .

Let us now consider the case where the morphisms are flat and locally of finite presentation. For every , there exists, by (the proof of) EGA IV_4, 17.16.2, an affine subscheme of a certain , such that and that the morphism , restriction of , is flat, of finite presentation, and quasi-finite. Then, is an open neighborhood of (EGA IV_2, 2.4.6), and, being affine, it is covered by a finite number of such opens , j = 1, ..., n. Consequently, is affine, and the morphism is surjective, flat, of finite presentation, and quasi-finite, hence belongs to .61 This completes the proof of (iv).

Let us prove (v). Let be a finite faithfully flat and quasi-compact family. Let be a covering of by affine opens. The are quasi-compact and therefore have finite affine open coverings . Each morphism is flat, and the family is finite and surjective, hence covering for T_1. The family is therefore also, by composition. It factors through the given family which is therefore also:

S_i ←──── S_{ij} ←──── T_{ijk}
                          ╱
                         ╱
                        ╱
S ←──── T_j ←─────────

Corollary 6.3.2. Let be the following family of morphisms:

  • (M_1) : faithfully flat and quasi-compact morphisms.
  • (M_2) : faithfully flat morphisms locally of finite presentation.
  • (M_3) : surjective étale morphisms.
  • (M_4) : surjective étale finite morphisms.62

The family verifies the axioms (a), (b), (c), (d_{T_i}) and (e_{T_i}) of 4.6.3.

Indeed, for (a), (b), (c), it is classical (EGA and SGA 1, passim.).63 By 6.3.1, (iv) and (v), verifies (d_{T_i}). It remains to see that verifies (e_{T_i}); for this, it suffices to see that verifies (e_{T_1}), which entails the others. This follows from SGA 1, VIII (nos 4 and 5).

Corollary 6.3.3. If is a scheme and an equivalence relation in of type , is -effective if and only if the sheaf-quotient of by for is representable and in this case it is represented by the quotient .

Indeed, this is 4.6.5.

6.4. Effectivity conditions

We now seek families (N) of morphisms verifying axiom (f_T) of 4.7. Let us first remark that (f_{T_1}) entails (f_{T_i}), so that we may restrict ourselves to the case of the topology (fpqc).

Lemma 6.4.1. The following families of morphisms verify axiom (f_{T_1}) of 4.7, that is, "descend by (fpqc)":

  • (N) : open immersions.
  • (N′) : closed immersions.
  • (N″) : quasi-compact immersions.

By virtue of 6.3.1 (ii), it suffices to verify that the given families descend by the Zariski topology and by a faithfully flat quasi-compact morphism. The first assertion is clear; let us verify the second. For (N), this is SGA 1, VIII 4.4; for (N′), this is loc. cit., 1.9. For (N″) one argues as in loc. cit., 5.5, using the two preceding results.

Corollary 6.4.2. The same result holds for quasi-compact open immersions.

These results allow one to apply to the present situation the general results of 4.7.1, 4.7.2, 5.1.8, 5.3.1, etc. Let us state one as an example, the first.

Corollary 6.4.3. (= 4.7.1 + 4.6.10). Let be a scheme and an equivalence relation in . Suppose that is faithfully flat and quasi-compact and that is a closed immersion (resp. open, resp. quasi-compact, resp. quasi-compact open). Then the sheaf-quotient is the same for the topology (fpqc) and for the canonical topology, and for each scheme , one has

(X/R)(S) = { closed (resp. open, resp. retrocompact[^N.D.E-IV-63], resp. open retrocompact)
             subschemes Z of X_S, stable under R × S, such that Z → S is
             faithfully flat quasi-compact and the diagram R_Z ⇒ Z → S is exact }.

6.5. Principal homogeneous bundles

Let us simply indicate the terminology:

   topology     principal homogeneous bundles
   (fpqc)       "         "         "             (tout court)
   (ét)         "         "         "             quasi-isotrivial
   (étf)        "         "         "             locally isotrivial
   (Zar)        "         "         "             locally trivial.

6.6. Other topologies

One sometimes uses other topologies on the category of schemes. Let us indicate one: the global étale finite topology (étfg), generated by the pretopology whose covering families are the surjective families formed of étale finite morphisms. It is not finer than the Zariski topology. The corresponding principal homogeneous bundles are called "isotrivial".

                     (canonical)
                         │
                       (fpqc)
                         │
                       (fppf)
                         │
                        (ét)
                         │
                       (étf)
                       ╱    ╲
                      ╱      ╲
                  (Zar)    (étfg)
                      ╲      ╱
                       ╲    ╱
                    (chaotic)[^N.D.E-IV-64]

6.7. Homogeneous spaces64

Let be an -group scheme, an -scheme with (left) operator group , and

Φ : G ×_S X ⟶ X ×_S X

the morphism of -schemes defined setwise by . Let us recall (cf. 5.1.0 and III.0.1) that one says that is a formally principal homogeneous space under if the following equivalent conditions are satisfied:

(i) for every , the set is empty or principal homogeneous under ,

(ii) is an isomorphism of -functors,

(iii) is an isomorphism of -schemes.

(The equivalence (i) ⇔ (ii) is clear, and one has (ii) ⇔ (iii) since is a full subcategory of Ĉ.)

The definition of formally homogeneous space (not necessarily principal homogeneous) is obtained by requiring that be an epimorphism in the category of sheaves for an appropriate topology . Indeed, the condition that be an epimorphism of -functors amounts to the requirement that, for every , the set be empty or homogeneous (not necessarily principal homogeneous) under , but this condition is too restrictive, as the following simple example shows. Let , and on which acts via . Then the morphism is étale, finite, and surjective, hence an epimorphism in the category of sheaves for the topology (étf) (a fortiori, an epimorphism of -schemes); on the other hand, the points 1 and of are not conjugate by an element of , so that the morphism is not surjective.65 One is therefore led to lay down the following definition:

Definition 6.7.1. Let be an -group, an -scheme with operator group , and a topology on , coarser than the canonical topology. We say that is a formally homogeneous space under (relative to the topology ) if the following equivalent conditions are satisfied:

(i) the morphism is an epimorphism in the category of sheaves for the topology ,

(ii) for every , and , there exists a morphism covering for the topology , and , such that .

Remark 6.7.2. Condition (i) implies, in particular, that is a universal effective epimorphism in (cf. 4.4.3). This entails, as one easily sees, that is surjective (cf. 1.3, N.D.E. (3)).

Proposition and Definition 6.7.3.66 Let be an -group, an -scheme with operator group , and a topology on , coarser than the canonical topology. The following conditions are equivalent:

(i) verifies the two hypotheses below:

  • (1) the morphism is covering, i.e. is a -formally homogeneous space,
  • (2) the morphism is also covering, i.e. locally for the topology , it has a section (cf. RefIV.4.4.8bis.bis).

(ii) "Locally on for the topology ", is isomorphic, as a scheme with operator group , to the sheaf quotient (for ) of by a sub-group-scheme , i.e. there exists a covering family such that each represents the sheaf quotient of by a certain sub-group-scheme .

Under these conditions, one says that is a -homogeneous space (relative to the topology ).

Proof. Suppose (ii) is verified. Set and . Then, has a section over , namely the composite of the unit section and the projection . Hence is covering.

On the other hand, is covering, hence is also (cf. 4.2.3 (C 1) and (C 2)), and one has a commutative diagram:

                 Φ_i
G_i ×_{S_i} X_i ────→ G_i ×_{S_i} X_i
       ↑                       ↑
  id × π_i                   π_i × π_i
       │                       │
                 Ψ_i
G_i ×_{S_i} G_i ────→ G_i ×_{S_i} G_i
                  ∼

where is deduced from by the base change and is the isomorphism defined setwise by . Then is covering, hence is also (4.2.3 (C 3)). This shows that is "locally covering", hence is covering (4.2.3 (C 5)). This proves that (ii) ⇒ (i).

Conversely, suppose (i) is verified, and suppose moreover that the structural morphism has a section . By EGA I, 5.3.13, is an immersion. Let us define by the diagram below, in which the two squares are cartesian:

H ──→ G ──id_G ⊠ σ──→ G ×_S X
│      │                 │
│      π                 Φ
↓ σ    ↓    id_X ⊠ σ     ↓
S ──→ X ──────────────→ X ×_S X

where , and denote the morphisms defined setwise, for and , , by:

π(g) = g · σ_T,    (id_G ⊠ σ)(g) = (g, σ_T),    (id_X ⊠ σ)(x) = (x, σ_T).

Then, is covering, and is a sub-group-scheme of , representing the stabilizer of (cf. I, 2.3.3), i.e., for every , one has:

H(T) = { g ∈ G(T) | g · σ_T = σ_T }.

Let us denote the presheaf , and the associated sheaf, for the topology . By what precedes, one obtains a commutative diagram of morphisms of presheaves with operator group :

G ──π──→ X
 ╲       ↗
  ╲     ╱ π̄
   ↘   ╱
   G/H

where is a monomorphism. As is covering, is also and therefore, by 4.3.12, induces an isomorphism . One has therefore proved that: if is a -homogeneous space such that admits a section , then represents the sheaf quotient , where is the stabilizer of .

In the general case, there exists by hypothesis a covering family such that each morphism has a section . Set ; then the morphism deduced from by the base change is again covering. Hence, by what precedes, , where is the stabilizer in of . This completes the proof of the implication (i) ⇒ (ii).

Bibliography

  • [AS] Analysis Situs, by J. Giraud, Sém. Bourbaki, Exp. 256, May 1963.
  • [D] Méthode de la descente, by J. Giraud, Mém. Soc. Math. France, t. 2 (1964), p. iii–viii + 1–150.
  • [MA] Grothendieck Topologies, by M. Artin, mimeographed notes, Harvard, 1962.
  • [SGA 1] Séminaire de Géométrie Algébrique du Bois-Marie 1960–61, Revêtements étales et groupe fondamental, Lecture Notes in Maths. 224 (1971), revised and annotated edition, Documents Math. 3, Soc. Math. France, 2003.
  • [SGA 4] Séminaire de Géométrie Algébrique du Bois-Marie 1963–1964, Théorie des topos et cohomologie étale des schémas, t. I, II, III, Lecture Notes in Maths. 269, 270 (1972), 305 (1973).
  • [TDTE I] Techniques de descente et théorèmes d'existence en géométrie algébrique I. Généralités. Descente par morphismes fidèlement plats, by A. Grothendieck, Sém. Bourbaki, Exp. 190, Dec. 1959.
  • [Ray70]67 M. Raynaud, Faisceaux amples sur les schémas en groupes et les espaces homogènes, Lect. Notes Math. 119, Springer-Verlag, 1970.

Footnotes

1

This text develops the substance of two oral expositions of A. Grothendieck, completing the latter on several important points which had been passed over in silence or scarcely touched on.

2

N.D.E.: Version of 13/10/2024.

3

N.D.E.: that is, if is right-cancellable.

4

N.D.E.: This implies, in particular, that be an epimorphism.

5

N.D.E.: For example, if is the category of schemes, one sees easily that every universal epimorphism is surjective. Let and ; then the morphism is an epimorphism that is not universal. On the other hand, one sees that is identified with , so that the two projections T ×_S T ⇉ T coincide; since does not descend to a morphism , this shows that is not an effective epimorphism.

6

N.D.E.: The numbering 1.4.0 has been added for later references.

7

N.D.E.: The numbering 1.6.0 has been added for later references.

8

N.D.E.: This is the following argument, communicated by M. Demazure. Let , and let be the morphism with components and , whence and . Then satisfies and . Now, for every , one has . Hence, applying to both sides, one obtains , which shows that and induce the same morphism.

9

N.D.E.: One will note that a monomorphism which is an epimorphism is not necessarily an isomorphism. For example, in , the morphism Spec(F_p) ∐ Spec(Z[1/p]) → Spec(Z) is a monomorphism and a surjective epimorphism, but is not an isomorphism.

10

N.D.E.: for example, one has identified, on the one hand, and, on the other hand, .

11

N.D.E.: applied to .

12

N.D.E.: The condition is evidently necessary. Conversely, if for every , is the graph of an equivalence relation, then this equivalence relation extends to for every , by declaring that two morphisms are equivalent if, for every and , and are equivalent in .

13

N.D.E.: Let us illustrate this by giving an outline of the sequel of this Exposé. Let be a -group, a sub--group, the equivalence relation in defined by , (g, h) ⟼ (g, gh) (cf. 3.2). The functor defined by is a quotient in Ĉ (according to 4.4.9 applied to the least fine topology, cf. 4.4.2), but it is not in general the quotient that one wishes. For example, for , one has an exact sequence of (affine) group schemes:

1 ⟶ μ₂ ⟶ G_m —p→ G_m ⟶ 1

which identifies with the quotient . Moreover, since is a finite and locally free morphism, then is the sheaf-quotient of by in the larger category of sheaves for the (fppf) topology, cf. 4.6.6 (ii) and 6.3.2. By contrast, the quotient in Ĉ is not isomorphic to since, for example, while . Hence is not an (fppf) sheaf, and a fortiori is not representable.

14

N.D.E.: Note that, even for a Ĉ-equivalence relation in , one is interested in the existence of a quotient in .

15

N.D.E.: "in " has been added.

16

N.D.E.: The numbering 3.3.2.1 (resp. 3.3.3.1) has been added for later references. On the other hand, Remark 3.3.3.2 has been added.

17

N.D.E.: The numbering 3.4.3.1 has been added for later references, and the proof of point (i) has been detailed.

18

N.D.E.: This paragraph has been added.

19

N.D.E.: Here, "dominator" is taken in the sense of the preorder relation mentioned above, i.e., dominates if there exists an arrow . On the other hand, if X, Y are two subobjects of an object , one says (cf. 2.4) that contains if . To avoid any ambiguity between these two terminologies, "majorant" has been replaced in the sequel by "dominating" in the first case, and by "containing", in the second.

20

N.D.E.: The sentence that follows has been detailed.

21

N.D.E.: The numbering 4.1.4.0 has been added in order to highlight this definition.

22

N.D.E.: that is: if is covering "locally with respect to the covering sieve ", then is covering.

23

N.D.E.: (T′ 1) and (T′ 2) suffice: belongs to , by (T′ 1), hence to , by (T′ 2).

24

N.D.E.: This definition has been placed here (placed in the original after 4.2.5), since it will be used in 6.2.1 in a slightly more general setting than that of 4.2.5.

25

N.D.E.: in what follows, typographical errors of the original have been corrected.

26

N.D.E.: One will note that if is a subpresheaf of a separated presheaf , then is separated. Indeed, for every sieve of , the composite map is injective and factors through .

27

N.D.E.: This corollary has been added.

28

N.D.E.: The numbering 4.3.10.0 has been added for later references. On the other hand, it follows from the definition that if is a monomorphism, the same holds for ; hence "preserves monomorphisms" (see also 4.3.16 for a more general result: "commutes with finite inverse limits").

29

N.D.E.: Denote by the projection . Since is a monomorphism, one has . Let be the section of defined by ; then and therefore ; this entails and therefore is an isomorphism, with inverse .

30

N.D.E.: The statement of the corollary and its proof have been detailed.

31

N.D.E.: In particular, if is the kernel of a pair of morphisms of presheaves u, v : Q ⇉ P, then LK is the kernel of Lu, Lv : LQ ⇉ LP (this will be used in 4.4.5).

32

N.D.E.: The original has been modified here.

33

N.D.E.: The continuation of the proof has been slightly modified.

34

N.D.E.: Point (i) of Lemma 4.4.5 has been corrected, and the proof of the three points detailed.

35

N.D.E.: In general, the direct sum of two sheaves F, G is not a sheaf. Indeed, let ; suppose that the direct sum exists in and that the fiber product is an initial object of (cf. I, 1.8). Let be the sieve of with base ; then is the disjoint union of and of for , hence is not a sheaf in general. On the other hand, if is the category with a single object and as sole morphism, equipped with the topology defined by , then the only separated presheaves are and (which is a sheaf), and is not separated.

36

N.D.E.: The statement of the lemma and its proof have been detailed.

37

N.D.E.: Recall (cf. N.D.E. (17)) that one says that a morphism is dominated by a family of morphisms if this family factors through .

38

N.D.E.: In order to be in accord with later references, the numbering of the original, which contained two nos. 4.4.7, has been corrected.

39

N.D.E.: "contains" has been replaced by: "is dominated by", cf. N.D.E. (17).

40

N.D.E.: Lemmas 4.4.8.1 and 4.4.8.2 have been added.

41

N.D.E.: and one has .

42

N.D.E.: and we shall also write , cf. the first commutative diagram of 4.5.4.

43

N.D.E.: i.e., such that for every object of , every covering sieve of , considered as a sieve of , is covering.

44

N.D.E.: We have expanded the original in what follows.

45

N.D.E.: Recall that denotes the given topology on , coarser than the canonical topology.

46

N.D.E.: We have added this corollary.

47

N.D.E.: We have placed the axiom (f_M) here (which figured before Proposition 4.7.2).

48

N.D.E.: cf. § 6.4, see also VI_A, 5.3.1.

49

N.D.E.: We have introduced the numbering 5.1.0, to refer to it later.

50

N.D.E.: One also says "pseudo-torsor", cf. EGA IV_4, 16.5.15. On the other hand, the more general notion of formally homogeneous object (not necessarily principal homogeneous), is defined in the addendum 6.7.1 at the end of this Exposé.

51

N.D.E.: In this case, one also says that is an -torsor.

52

N.D.E.: We have added this particular case as a corollary, which will be used several times in the following Exposés.

53

N.D.E.: We have added "and free".

54

N.D.E.: of .

55

N.D.E.: In fact, the isomorphisms established in 5.2.8 to 5.2.10 would deserve to be called "Noether isomorphism theorems".

56

N.D.E.: This follows from the proposition applied to .

57

N.D.E.: We have corrected the original here.

58

N.D.E.: Let denote the set of finite surjective families, composed of flat morphisms of of finite presentation. According to Proposition 6.3.1 below, the topology T_2 generated by and coincides with the topology generated by and . This follows from the results of EGA IV_4, § 17.16 on quasi-sections; see the proof of 6.3.1. Let us cite here the following particular case of EGA IV_4, 17.16.2: let be affine and a surjective flat morphism locally of finite presentation; then there exists a morphism faithfully flat, of finite presentation, quasi-finite, with affine, and an -morphism .

59

N.D.E.: By hypothesis, each family is finite, hence is affine and the family can therefore be replaced by the morphism , which still belongs to .

60

N.D.E.: We have simplified what follows, taking advantage of the fact that is henceforth assumed affine.

61

N.D.E.: This shows that, if one denotes the set of finite surjective families of morphisms of flat of finite presentation, the topology generated by and equals T_2. On the other hand, with the notations at the start of the proof of (iv), if one takes a covering of by affine opens, of finite presentation over , one obtains that belongs to .

62

N.D.E.: We have corrected the original by adding the surjectivity hypothesis for (M_3) and (M_4), which is automatically satisfied in the other cases.

63

N.D.E.: cf. EGA I, 6.6.4 for "quasi-compact", EGA II, 6.1.5 for "finite", EGA IV_1, 1.6.2 for "locally of finite presentation", EGA IV_2, 2.2.13 for "faithfully flat", and EGA IV_4, 17.3.3 for "étale".

64

N.D.E.: We have added the numbers that follow.

65

N.D.E.: Evidently, this difficulty arises from the fact that if is a full subcategory of Ĉ containing , for example, the category of sheaves on for a topology coarser than the canonical topology, and if is a morphism in , then the implications:

f epimorphism of Ĉ ⇒ f epimorphism of C ′ ⇒ f epimorphism of C

are in general strict.

66

N.D.E.: cf. [Ray70], Def. VI.1.1.

67

N.D.E.: We have added this reference.

68

N.D.E.: denotes the equivalence relation in defined by , and R_Z is the equivalence relation it induces in (cf. 3.1.6).

69

N.D.E.: Recall that a subscheme of a scheme is said to be retrocompact if the immersion is quasi-compact, cf. EGA 0_III, 9.1.1.

70

N.D.E.: Recall (cf. 4.4.2) that the chaotic topology is the coarsest topology, defined by for every .