Chapter IV (continued)

§2. Base change and flatness

This section (unlike §6) appeals only exceptionally to Noetherian techniques. Nos. 1 and 2 are scarcely more than translations of elementary properties of flatness from commutative algebra (cf. Bourbaki, Alg. comm., chap. I) and are included here for the convenience of references. The following numbers are devoted above all to "descent" properties along flat or faithfully flat morphisms: if is such a morphism, the issue is to be able to assert that a part of , or an -Module, or a morphism , has a certain property, when one knows that its inverse image under has that property. We restrict ourselves here to properties that do not appeal to the general technique of "descent", which will be developed in Chapter V.

2.1. Flat modules on preschemes

(2.1.1)

Let be a morphism of preschemes and an -Module; recall that is said to be -flat (or -flat) at a point if is a flat -module, -flat (or -flat) if it is -flat at every , and finally that the morphism is said to be flat at the point (resp. flat) if is -flat at (resp. -flat). When , one says simply that an -Module is flat at the point (resp. flat) if it is -flat at this point (resp. at every point ), that is, if is a flat -module (resp. if this holds for every ). Recall that we have proved (III, 1.4.15.1) the following property:

Proposition (2.1.2).

Let , be two rings, a ring homomorphism, , , the morphism corresponding to , a -module. For to be -flat, it is necessary and sufficient that be a flat -module.

Proposition (2.1.3).

Let be a morphism of preschemes, a quasi-coherent -Module. The following conditions are equivalent:

a) For every base change , if one sets , the functor from the category of quasi-coherent -Modules to that of quasi-coherent -Modules is exact.

a') Condition a) is satisfied for all canonical morphisms (I, 2.4.1), where runs over .

b) is -flat.

The questions being local on and , one may restrict to the case where , , , where is an -module. It is clear that a) entails a'); condition a') entails that for every the functor is exact in the category of -modules, being the ideal of ; this means that is a flat -module, and it results from and from (2.1.2) that is -flat. Finally, to see that b) entails a), one may also restrict to the case where is affine and , where is an -module; the conclusion then again follows from (2.1.2) and the definition of flatness, since .

Proposition (2.1.4).

Let , be two morphisms of preschemes, a quasi-coherent -Module; set , , , and let be the canonical projection . Let be a point of , , , . If is -flat at the point , then is -flat at the point ; in particular if is -flat, is -flat; if is flat, is flat.

It suffices to prove the first assertion; applying (I, 3.6.5) three times, as well as (I, 2.4.4), one may reduce to the case where , , , , where ; the hypothesis and (2.1.2) then entail that is -flat — in other words one is reduced to proving a particular case of the second assertion, and this last follows at once from (2.1.3).

Proposition (2.1.5).

Consider a commutative diagram of morphisms of preschemes

  X  ←─g'──  X'
  │           │
  f│         │f'
  ↓           ↓
  Y  ←──g──  Y'
              │
              │h
              ↓
              Z

where and . Let be a point of , and set , , , . Let be a quasi-coherent -Module that is -flat at the point (resp. -flat), and a quasi-coherent -Module that is -flat at the point (resp. -flat); then is a quasi-coherent -Module that is -flat at the point (resp. -flat).

As in (2.1.4), one reduces to the case where , ,

and , and it then suffices to prove that is -flat. Taking (2.1.2) into account, the proposition follows from Bourbaki, Alg. comm., chap. I, §2, n° 7, prop. 8.

Corollary (2.1.6).

Let , be two morphisms of preschemes, an -Module. If is -flat at the point and is flat at the point , then is -flat at the point . In particular, if and are flat morphisms, so is .

This is the particular case of (2.1.5) with , .

Corollary (2.1.7).

If , are two flat -morphisms, then is a flat morphism.

This follows from (2.1.4) and (2.1.6) (cf. (I, 3.5.1)).

Proposition (2.1.8).

Let be a morphism of preschemes,

an exact sequence of quasi-coherent -Modules such that is -flat.

(i) For every morphism and every quasi-coherent -Module , the sequence

  0 → ℱ' ⊗_Y 𝒢' → ℱ ⊗_Y 𝒢' → ℱ'' ⊗_Y 𝒢' → 0

of -Modules (where ) is exact.

(ii) For to be -flat, it is necessary and sufficient that be so.

One may obviously suppose , , affine; the conclusion then follows from (2.1.2) and .

Corollary (2.1.9).

Let be a complex of quasi-coherent -Modules, an index such that if one denotes by the differential, and are -flat. Then, with the notations of (2.1.8), the canonical homomorphism

  ℋ^i(ℒ^•) ⊗_Y 𝒢' → ℋ^i(ℒ^• ⊗_Y 𝒢')

is bijective.

Since the tensor product is right exact, one has

  𝒵^{i+1}(ℒ^•) ⊗_Y 𝒢' = Coker(d^i ⊗ 1) = 𝒵^{i+1}(ℒ^• ⊗_Y 𝒢')

and . Moreover, in the exact sequence

is -flat, so it follows from (2.1.8, (i)) that one has the exact sequence

  0 → ℬ^{i+1}(ℒ^•) ⊗_Y 𝒢' → ℒ^{i+1} ⊗_Y 𝒢' → 𝒵^{i+1}(ℒ^• ⊗_Y 𝒢') → 0

whence . Then, in the exact sequence

is -flat, so it follows from (2.1.8, (i)) and what precedes that one has the exact sequence

  0 → ℋ^i(ℒ^•) ⊗_Y 𝒢' → 𝒵^i(ℒ^• ⊗_Y 𝒢') → ℬ^{i+1}(ℒ^• ⊗_Y 𝒢') → 0

which proves the corollary.

Corollary (2.1.10).

Let be a morphism of preschemes, a quasi-coherent and -flat -Module, a left resolution of formed of quasi-coherent and -flat -Modules. Then, for every morphism and every quasi-coherent -Module , the complex is a left resolution of .

Moreover, if , the are -flat, and one has

  𝒵_i(ℒ_•) ⊗_Y 𝒢' = 𝒵_i(ℒ_• ⊗_Y 𝒢') = Ker(ℒ_i ⊗_Y 𝒢' → ℒ_{i−1} ⊗_Y 𝒢').

Set ; one then has the exact sequences

  0 ← ℱ ← ℒ_0 ← ℛ_0 ← 0
  ⋮
  0 ← ℛ_i ← ℒ_{i+1} ← ℛ_{i+1} ← 0
  ⋮

and since and the are -flat, one deduces from (2.1.8, (ii)) by induction that all the are also -flat; using (2.1.8, (i)), one therefore has the exact sequences

  0 ← ℱ ⊗_Y 𝒢' ← ℒ_0 ⊗_Y 𝒢' ← ℛ_0 ⊗_Y 𝒢' ← 0
  0 ← ℛ_i ⊗_Y 𝒢' ← ℒ_{i+1} ⊗_Y 𝒢' ← ℛ_{i+1} ⊗_Y 𝒢' ← 0          (i ≥ 0)

which prove the corollary.

Proposition (2.1.11).

Let be a flat morphism, a quasi-coherent -Module of finite presentation. If is the Ideal of annihilator of , then is the Ideal of annihilator of .

One has by definition an exact sequence

  0 → 𝒥 → 𝒪_Y → ℋom_{𝒪_Y}(ℱ, ℱ)

whence, since is flat, an exact sequence

  0 → f*(𝒥) → 𝒪_X → f*(ℋom_{𝒪_Y}(ℱ, ℱ))

and since by hypothesis is an -Module of finite presentation, is canonically identified with , whence the conclusion.

Proposition (2.1.12).

Let be a prescheme, an -Module of finite presentation, a point of . The following conditions are equivalent:

a) is a flat -module.

b) There exists an open neighbourhood of such that is a locally free -Module.

Indeed, is an -module of finite presentation and a local ring; it therefore amounts to the same to say that is a flat -module or a free -module (Bourbaki, Alg. comm., chap. II, §3, n° 2, cor. 2 of prop. 5); whence the conclusion, taking account of . We note that the proposition is valid for an arbitrary ringed space in local rings.

Proposition (2.1.13).

*Let be a morphism of preschemes. If is flat at a point and the ring is reduced (resp. integral and integrally closed), then the ring

is reduced (resp. integral and integrally closed). If is faithfully flat and is reduced (resp. normal), then is reduced (resp. normal).*

Set , . If is a flat -module, it is also a faithfully flat -module , so is identified with a subring of ; if is reduced, so therefore is . Suppose now that is integral and integrally closed, and let be its field of fractions; then is integral; denote by its field of fractions. The hypothesis entails that (Bourbaki, Alg. comm., chap. I, §3, n° 5, prop. 10). If then is integral over , it is also integral over , hence belongs to by hypothesis, and consequently , which proves that is integrally closed.

Proposition (2.1.14).

Let be a faithfully flat morphism. If is locally integral and the space underlying is locally Noetherian, then is locally integral.

Indeed, every is of the form for some and by hypothesis is identified with a subring of ; since is integral, so is , and this proves the proposition (I, 5.1.4).

2.2. Faithfully flat modules on preschemes

Proposition (2.2.1).

Let be a morphism of preschemes, a quasi-coherent -Module. The following conditions are equivalent:

a) For every base change , if one sets , the functor from the category of quasi-coherent -Modules to that of quasi-coherent -Modules is exact and faithful.

a') Condition a) is satisfied for all canonical morphisms (I, 2.4.1), where runs over .

a'') Condition a) is satisfied for all canonical immersions , where runs over the set of affine open subschemes of .

b) is -flat and, for every , if one denotes by the fibre , the -Module is .

It is clear that a) implies a') and a''); condition a') first implies that is -flat (2.1.3); on the other hand it implies that for every the functor is faithful in the category of -modules; taking in particular , one obtains the second assertion of b). To show that b) implies a), one may restrict to the case where is affine, the question being local on . Similarly, to prove that a'') implies a), one is reduced to proving that when is affine, the fact that is an exact and faithful functor entails condition a). In other words, one is reduced to proving the following more precise proposition:

Proposition (2.2.2).

Let be an affine scheme, a morphism of preschemes, a quasi-coherent -Module. Condition a) of (2.2.1) is equivalent to each of the following:

b') is -flat and, for every closed point of , one has .

c) The functor from the category of quasi-coherent -Modules to that of quasi-coherent -Modules is exact and faithful.

If b') is satisfied, there is at least one such that ; let be an affine open neighbourhood of , and let , where is a -module. Then b') implies that , and consequently (since is a flat -module by (2.1.2)) that is a faithfully flat -module . The relation for a closed point of implies , hence . But if for every closed point of , one concludes that ; indeed, if , the annihilator of an element of is contained in no maximal ideal of , so it equals all of . The relation therefore implies ; in other words, the functor is faithful; we know moreover that this functor is exact (2.1.3), which shows that b') entails c).

Finally, to see that c) entails a), one may restrict to the case where is also affine; as is then an affine morphism, so is the projection (II, 1.5.5); in addition, the functor is then exact in the category of quasi-coherent -Modules (I, 1.6.4), and if , one has , so the preceding functor is also faithful; to see that is exact and faithful, it therefore suffices to see that the functor is. Now, if , one has ; the fact that is affine entails that one has a canonical isomorphism

  ℱ ⊗_{𝒪_X} f*(g_*(𝒢')) ⥲ g'_*(g'*(ℱ) ⊗_{𝒪_{X'}} f'*(𝒢')).            (2.2.2.1)

Indeed, one knows (II, 1.5.2) that one has a canonical isomorphism

and on the other hand one has a canonical homomorphism ; composing the homomorphism with the canonical homomorphism , one deduces the homomorphism (2.2.2.1); the verification that it is an isomorphism is immediate by reducing to the case where is affine. This being so, the functor is exact and faithful and by hypothesis so is ; their composite is therefore exact and faithful, which completes the proof of (2.2.1) and (2.2.2).

Corollary (2.2.3).

Let , be two affine schemes, a morphism, a quasi-coherent -Module. For to satisfy the equivalent conditions of (2.2.1) (or (2.2.2)), it is necessary and sufficient that the -module be faithfully flat.

Indeed, condition c) of (2.2.2) then means that the functor from the category of -modules to that of -modules is exact and faithful, and the conclusion follows from .

Definition (2.2.4).

When the equivalent conditions of (2.2.1) are satisfied, one says that the quasi-coherent -Module is faithfully flat relative to (or to ).

One notes that this notion is local on , but not on ; in particular one can have for certain , in other words is not necessarily equal to . Nevertheless, it follows at once from criterion (2.2.1, b) that for every there exists at least one such that , and a fortiori ; in other words:

Corollary (2.2.5).

If is a quasi-coherent -Module, faithfully flat relative to , one has , and a fortiori is a surjective morphism.

This result admits a partial converse:

Corollary (2.2.6).

Let be a quasi-coherent -Module of finite type. For to be faithfully flat relative to , it is necessary and sufficient that be -flat and that .

Indeed (I, 9.1.13 and 3.6.1) one has , so in this case criterion (2.2.1, b) is none other than the condition of the corollary.

In particular, the -Module is faithfully flat relative to if and only if it is -flat and is surjective, in other words if and only if the morphism is faithfully flat .

Let us make explicit the properties involved in definition (2.2.4):

Proposition (2.2.7).

Let be a morphism of preschemes, a quasi-coherent -Module faithfully flat relative to . Then, for a sequence of quasi-coherent -Modules to be exact, it is necessary and sufficient that the corresponding sequence be so. In particular, for a homomorphism of quasi-coherent -Modules to be injective (resp. surjective, bijective, zero), it is necessary and sufficient that be so. For a quasi-coherent -Module to be zero, it is necessary and sufficient that . For every quasi-coherent -Module , the map from the set of quasi-coherent sub--Modules of to the set of quasi-coherent sub--Modules of is injective.

To prove the last assertion — that is, that for two quasi-coherent sub--Modules , of , the relation entails — one may (replacing by ) reduce to the case where , and it then suffices to apply the second assertion of the statement to the injection .

Corollary (2.2.8).

Let be a faithfully flat morphism. For every quasi-coherent -Module , the canonical map

  Γ(Y, 𝒢) → Γ(X, f*(𝒢))                                       (2.2.8.1)

is injective.

Indeed, is canonically identified with and likewise with . By virtue of (2.2.1) and (2.2.4), the hypothesis entails that the functor is exact and faithful in the category of quasi-coherent -Modules, and consequently a homomorphism is zero if and only if is zero.

Remark (2.2.9).

The results of (2.2.7) and (2.2.8) still hold when the -Modules , , appearing there are arbitrary (not necessarily quasi-coherent). Indeed, for every , there exists such that is a -module

faithfully flat, and consequently the functor is faithful; since moreover for every the functor is exact, one deduces the conclusion at once.

Proposition (2.2.10).

Let , , be three morphisms of preschemes, , a quasi-coherent -Module, faithfully flat relative to , a quasi-coherent -Module. For to be flat (resp. faithfully flat) relative to , it is necessary and sufficient that be a flat (resp. faithfully flat) -Module relative to .

One knows already that if is -flat, then so is (2.1.5). Consider an arbitrary base change and set ; if is faithfully flat relative to , the functor

  ℋ'' ↦ ℋ'' ⊗_Z 𝒢' → (ℋ'' ⊗_Z 𝒢') ⊗_Y ℱ = ℋ'' ⊗_Z (𝒢' ⊗_Y ℱ)        (2.2.10.1)

from the category of quasi-coherent -Modules to that of -Modules is the composite of two exact and faithful functors, hence is exact and faithful. Conversely, if this composite functor is exact (resp. exact and faithful), so is , since the functor (from the category of quasi-coherent -Modules to that of -Modules) is exact and faithful by hypothesis.

Corollary (2.2.11).

(i) Let , be two morphisms, , a quasi-coherent -Module. If is faithfully flat relative to , then is faithfully flat relative to .

(ii) Let , be two morphisms, a quasi-coherent -Module, faithfully flat relative to . For to be a faithfully flat morphism, it is necessary and sufficient that be faithfully flat relative to .

(iii) Let , be two morphisms, a quasi-coherent -Module. Suppose the morphism is faithfully flat. For to be flat (resp. faithfully flat) relative to , it is necessary and sufficient that be flat (resp. faithfully flat) relative to .

(iv) Let , be two morphisms, a point of . Suppose is flat at the point . For to be flat at the point , it is necessary and sufficient that be flat at the point .

To prove (i), one applies (2.2.10) replacing by and by . To prove (ii), one applies (2.2.10) taking the identity for the morphism and replacing by . To prove (iii), one applies (2.2.10) again taking the identity for and replacing by and by . Finally (iv) is deduced from (ii) by replacing by , by , by , and by , taking account of .

Corollary (2.2.12).

Let be an affine scheme, a quasi-compact morphism, a quasi-coherent -Module. If is faithfully flat relative to , there exists an affine scheme and a surjective local isomorphism such that is faithfully flat relative to .

Indeed, is quasi-compact, hence a finite union of affine open sets ; it suffices to take for the affine scheme sum of the preschemes induced on the open sets , and for the canonical morphism; it is clear that is faithfully flat, and

the hypothesis entails that is faithfully flat relative to , by virtue of (2.2.11, (iii)).

Corollary (2.2.13).

(i) Let , be two morphisms, , . If is a faithfully flat morphism, so is .

(ii) If , are two faithfully flat -morphisms, then

  f ×_S g : X ×_S Y → X' ×_S Y'

is faithfully flat.

(iii) Let , be two morphisms such that is faithfully flat. For to be a flat (resp. faithfully flat) morphism, it is necessary and sufficient that be a flat (resp. faithfully flat) morphism.

Proposition (2.2.14).

Let be a faithfully flat and quasi-compact morphism. If is locally Noetherian, so is .

The question being local on , one may suppose affine; since is quasi-compact, it follows from (2.2.12) that one may also restrict to the case where is affine. Then is a Noetherian ring by hypothesis, and a faithfully flat -module (2.2.3); hence is Noetherian .

Proposition (2.2.15).

Let be a faithfully flat morphism. If and denote respectively the set of sub-preschemes of and , the map from to is injective.

Since is surjective, one has, for the underlying set of a sub-prescheme of , . On the other hand, if is an open set of containing and in which is closed, is open in , is closed in , and the restriction of is a faithfully flat morphism. One may therefore restrict to considering only closed sub-preschemes of . Now, if is a closed sub-prescheme of corresponding to a quasi-coherent Ideal of , one knows that corresponds to the quasi-coherent Ideal of (I, 4.4.5), and since is flat, is identified with . But the map from the set of quasi-coherent Ideals of to the set of quasi-coherent Ideals of is injective (2.2.7), whence the conclusion.

Corollary (2.2.16).

Let , be two -preschemes; if is a faithfully flat morphism, the map from to is injective.

One has (I, 3.3.10), so the projection morphism is faithfully flat (2.2.13). The elements of correspond bijectively to the sub-preschemes of that are the graphs of these -morphisms (I, 5.3.11), and if is the graph of , one has (I, 5.3.12). It therefore suffices to apply proposition (2.2.15) to .

Proposition (2.2.17).

Let be a ring, an -algebra such that is a faithfully flat and finitely presented -module. Then the structure homomorphism is an isomorphism of the -module onto a direct factor of the -module . If is a local ring, is a free -module and there exists a basis of this module containing the unit element of .

By virtue of Bourbaki, Alg. comm., chap. II, §3, n° 3, prop. 12, it suffices to prove the proposition when is a local ring; one then knows (loc. cit., n° 2, cor. 2 of prop. 5) that is a free -module of finite type, and the conclusion follows from loc. cit., prop. 5.

2.3. Topological properties of flat morphisms

Lemma (2.3.1).

Let be a quasi-compact and quasi-separated morphism, a flat morphism; set , . Then, for every quasi-coherent -Module , the canonical homomorphism

  g*(f_*(ℱ)) → f'_*(ℱ ⊗_{𝒪_X} 𝒪_{X'})                            (2.3.1.1)

is bijective.

This is a particular case of (III, 1.4.15) (improved in (1.7.21)).

Proposition (2.3.2).

Let be a prescheme, a quasi-compact and quasi-separated -morphism; let be the sub-prescheme of , closed image of under ((I, 9.5.3) and (1.7.8)), and let be the canonical injection, so that one has , where is a morphism (loc. cit.). Let be a flat morphism, and set ; then is identified with the canonical injection of the sub-prescheme of closed image of under .

Since the morphism is flat (2.1.4), one may restrict to the case where (I, 3.3.11); if , one knows that is the closed sub-prescheme of defined by the (quasi-coherent) Ideal of , kernel of the homomorphism (I, 9.5.2). Since is a flat morphism, the quasi-coherent Ideal of is identified with the kernel of . Now, if , one verifies easily (for example by reducing to the case where and are affine and using (I, 2.2.4)) that is the composite of the canonical homomorphism (2.3.1.1) and of ; the conclusion therefore follows from (2.3.1) and (I, 9.5.2), since is quasi-compact and quasi-separated (1.1.2 and 1.2.2).

(2.3.3)

We shall say that a morphism is quasi-flat if there exists a quasi-coherent -Module of finite type that is -flat and whose support is equal to . We shall say that is quasi-faithfully flat if it is quasi-flat and surjective. Every flat (resp. faithfully flat) morphism is quasi-flat (resp. quasi-faithfully flat), since then satisfies the preceding conditions.

It follows at once from (2.1.4) and (I, 9.1.13) that if is quasi-flat, then for every morphism the morphism is quasi-flat. Similarly (taking (I, 3.5.2) into account), if is quasi-faithfully flat, so is .

Proposition (2.3.4).

Let be a quasi-flat morphism (2.3.3). Then possesses the following properties (which are equivalent by virtue of (1.10.4)):

(i) For every and every generization of , there exists a generization of such that .

(ii) For every , the image under of is .

(iii) For every irreducible closed part of , every irreducible component of dominates .

It suffices, for example, to prove (ii). By hypothesis, there is a quasi-coherent -Module of finite type , which is -flat and such that . For every , is then an -module of finite type, not reduced to 0, and moreover is a flat -module, for the homomorphism . Since this latter is local and , Nakayama's lemma shows that , hence is a faithfully flat -module . It results that for every prime ideal of , there exists a prime ideal of such that , which proves (ii).

Corollary (2.3.5).

Let be a morphism satisfying the equivalent conditions (i), (ii), (iii) of (2.3.4) (in particular a quasi-flat morphism, or an open morphism (1.10.4)).

(i) Let , be two irreducible closed parts of such that , and let be an irreducible component of ; then there exists an irreducible component of containing (and dominating ).

(ii) For every irreducible component of , is an irreducible component of .

(iii) Suppose is irreducible, and, if is its generic point, suppose that is irreducible. Then is irreducible.

(i) It suffices to apply (2.3.4, (i)) taking for the generic point of ( then being the generic point of ) and for the generic point of .

(ii) It is clear that is irreducible, and by virtue of (i), cannot be strictly contained in an irreducible closed part of .

(iii) By virtue of (ii), every irreducible component of dominates , hence meets ; the conclusion then follows from .

Proposition (2.3.6).

Let be a prescheme whose set of irreducible components is locally finite (which is the case if the space underlying is locally Noetherian (cf. (I, 6.1.9))).

(i) Every closed part of stable under generization is open. In particular, every connected component of is open.

(ii) Let be a closed morphism satisfying moreover the equivalent conditions (i), (ii), (iii) of (2.3.4) (which will be the case if is quasi-flat or open). Then the image under of every connected component of is a connected component of .

(iii) Let be a morphism satisfying the equivalent conditions (i), (ii), (iii) of (2.3.4) and such moreover that for every the set is finite (which will be the case if is quasi-finite or radicial). Then the set of irreducible components of is locally finite.

(i) If , the generic points () of the irreducible components of containing belong by hypothesis to ; hence, since is closed, these components themselves are contained in ; since by hypothesis there is a neighbourhood of such that is the union of the , one has , so

is open. Since for every , is connected, a connected component of is stable under generization, so the second assertion follows at once from the first.

(ii) Since is closed in , is closed in by hypothesis. Furthermore, since is stable under generization, the hypothesis on entails that is stable under generization; hence, by virtue of (i), is open and closed, and since it is connected it is a connected component of .

(iii) Let ; by hypothesis, there is an open neighbourhood of meeting only a finite number of irreducible components of (); let be the generic point of (). For every irreducible component of meeting , the generic point of is necessarily in one of the sets (2.3.4). Since each of these sets is finite by hypothesis, this proves our assertion.

Proposition (2.3.7).

Let be a morphism, a quasi-flat morphism, , .

(i) If is quasi-compact and dominant, so is .

(ii) If every irreducible component of dominates an irreducible component of , then every irreducible component of dominates an irreducible component of .

Denote by the canonical projection, which is a quasi-flat morphism (2.3.3).

(i) One already knows (1.1.2) that is quasi-compact; furthermore, if is a maximal point (1.1.4) of , is a maximal point of , as results from (2.3.4, (iii)). By hypothesis (1.1.5) there exists such that ; hence (I, 3.4.7) there exists such that .

(ii) Let be a maximal point of , and let ; it follows from (2.3.4, (ii)) that is a maximal point of , and by hypothesis is a maximal point of . Set , so that ; the issue is to show that is a maximal point of . By virtue of (I, 5.1.7) and (2.3.3), one may restrict to the case where and are reduced, so and are fields; moreover, by virtue of (I, 3.6.5) applied twice and of (I, 2.4.4), one may restrict to the case where and . Then is a flat morphism since is a field (2.1.2), and the same is true of (2.1.4); hence it follows from (2.3.4, (ii)) that is a maximal point of .

Corollary (2.3.8) (Zariski).

Let , be two Noetherian local rings, , their respective maximal ideals, a local homomorphism. Suppose the following hypotheses are satisfied:

is an -algebra essentially of finite type (1.3.8).

2° The completion  of for the -adic topology is integral.

is injective.

Then the -adic topology of is induced by the -adic topology of .

Set ; by virtue of 1°, is of the form , where is an -algebra of finite type and a multiplicative subset of , so is a Noetherian ring. Since is identified with a subring of  , is integral by 2°. Hypothesis 3° then entails that there exists a prime ideal of inducing the ideal 0 of , and consequently the local homomorphism is injective. One may therefore restrict to proving the conclusion of (2.3.8) by adding the hypothesis that is an integral local ring. Apply (2.3.7, (ii)) to , , and ; since the morphism is flat and

(which is integral) dominates the integral scheme , every irreducible component of dominates the scheme (integral by hypothesis) . If , , are the closed points of , , respectively, there is a point (in fact unique) above and (I, 3.4.9) and therefore dominates ; consequently one has a commutative diagram of local homomorphisms of Noetherian local rings

  B = 𝒪_x ────────→ 𝒪_{x'}
    ↑                  ↑
   φ│                 │v
    │                  │
  A = 𝒪_y ────u────→ 𝒪_{y'} = Â

such that and are injective (I, 1.2.7); identifying and  with subrings of , and denoting by the maximal ideal of , the intersection of the ideals is therefore zero ; since  is complete and these ideals are open in Â, this entails (Bourbaki, Alg. comm., chap. III, §2, n° 7, prop. 8) that the topology of  is induced by the -preadic topology of ; a fortiori the same is true of the topology of . Moreover one has , so the -preadic topology of induces on a topology finer than the -preadic topology; but since , these two topologies are identical. Q.E.D.

Remark (2.3.9).

We shall see further on (7.8.3, (vii)) that for the Noetherian local rings most usual in algebraic geometry, the hypothesis that is integral and integrally closed implies that the same holds for Â. That is why, in algebraic geometry over a base field, one generally states (2.3.8) under the hypothesis that is integral and integrally closed.

Theorem (2.3.10).

Let be a quasi-flat morphism (2.3.3). Then, for every proconstructible part (1.9.4) of , one has .

Since is continuous, one has and everything reduces to proving that for every such that , is adherent to ; it is clear that the question is local on , so one may suppose affine. By virtue of the hypothesis, there exists an affine scheme and a morphism such that (1.9.5, (ix)). Let Y_1 be the closed image of under (I, 9.5.3), and let X_1 be the closed sub-prescheme of ; if is the restriction of to X_1, one knows that is quasi-flat (2.3.3); one may therefore replace , by X_1, Y_1 respectively, in other words suppose that is dominant. Set then , and let and be the projections of onto and respectively, so that one has the commutative diagram

  X  ←─g'──  X'
  │           │
  f│         │f'
  ↓           ↓
  Y  ←──g──  Y'

Since is quasi-flat, quasi-compact and dominant, it follows from (2.3.7) (where the roles of and are interchanged) that is a dominant morphism, which proves the theorem.

Corollary (2.3.11).

Let be a quasi-flat and quasi-compact morphism, a closed part of such that ; then one has .

Let be the reduced sub-prescheme of having as underlying space (I, 5.2.1) and let be the canonical injection; then is quasi-compact (1.1.2),

so is pro-constructible in (1.9.5, (vii)), and the corollary follows from the fact that is closed.

One may further write the result of (2.3.11) in the form , in other words, the closed sets of the subspace of are the parts such that is closed in ; this also means that the topology induced by on is the quotient of that of by the equivalence relation defined by . In particular:

Corollary (2.3.12).

Let be a quasi-faithfully flat (2.3.3) and quasi-compact morphism. Then the topology of is the quotient of that of by the equivalence relation defined by (in other words, for to be open (resp. closed) in , it is necessary and sufficient that be open (resp. closed) in ).

Indeed, one then has .

Corollary (2.3.13).

Let , be two -preschemes, a faithfully flat and quasi-compact -morphism. For to be separated over , it is necessary and sufficient that the canonical immersion (I, 5.3.10) be closed.

Let us note for this that one has the commutative diagram (I, 5.3.5)

  X ×_Y X  ──j──→  X ×_S X
    │                │
    π│              │f ×_S f
    ↓                ↓
    Y    ──Δ_Y──→  Y ×_S Y

identifying with the product of the -preschemes and . Since is surjective, so are and , so the diagonal has as inverse image under the image (I, 3.4.8). Since is faithfully flat and quasi-compact (1.1.2 and 2.2.13), it suffices to apply (2.3.12) to this morphism.

Corollary (2.3.14).

Let be a quasi-faithfully flat and quasi-compact morphism, and let be a part of . For to be a locally closed pro-constructible part of , it is necessary and sufficient that be a locally closed pro-constructible part of .

One already knows (1.9.12) that, for to be a pro-constructible part of , it is necessary and sufficient that be a pro-constructible part of . The condition is evidently necessary. To prove that it is sufficient, consider the reduced closed sub-prescheme Y_1 of having as underlying space, and let X_1 be its inverse image under , which has as underlying space by virtue of (2.3.10). Since is locally closed in , it is open in , hence in X_1; now the morphism deduced from by restriction is quasi-faithfully flat and quasi-compact (1.1.2 and 2.3.3); one therefore deduces from (2.3.12) that is open in , which shows that is locally closed in .

Remark (2.3.15).

It does not suffice, in (2.3.12), to suppose only that is faithfully flat. For example, take for a reduced irreducible algebraic curve

(II, 7.4.2), for the prescheme sum of the schemes , where runs over (I, 3.1), for the canonical morphism, which is faithfully flat (I, 2.4.2); if denotes the generic point of , is not open in (II, 7.4.3), but since is a field, and consequently is open in .

2.4. Universally open morphisms and flat morphisms

(2.4.1)

We have already defined (II, 5.4.9) the notion of universally closed morphism; in the same way, one poses the following definitions:

Definition (2.4.2).

One says that a morphism of preschemes is universally open (resp. universally bicontinuous, resp. a universal homeomorphism) if, for every morphism , the morphism is open (resp. a homeomorphism onto its image, resp. a homeomorphism onto ).

We shall see further on (14.3.2) that when is locally Noetherian, the definition of universally open morphism given here is equivalent to the definition (III, 4.3.9) for morphisms of finite type; the reader may verify that we do not use this latter definition before §14.

Proposition (2.4.3).

(i) An immersion (resp. an open immersion, resp. a closed immersion) is universally bicontinuous (resp. universally open, resp. universally closed).

(ii) The composite of two universally open (resp. universally closed, resp. universally bicontinuous, resp. two universal homeomorphisms) morphisms is also.

(iii) If is a universally open (resp. universally closed, resp. universally bicontinuous, resp. a universal homeomorphism) -morphism, so is for every base change .

(iv) If , are two universally open (resp. universally closed, resp. universally bicontinuous, resp. two universal homeomorphism) -morphisms, so is .

(v) Let , be two morphisms such that is surjective; if is universally open (resp. universally closed, resp. universally bicontinuous, resp. a universal homeomorphism), so is .

(vi) For to be a universally open (resp. universally closed, resp. universally bicontinuous, resp. a universal homeomorphism) morphism, it is necessary and sufficient that be so.

(vii) Let be an open cover of . For to be universally open (resp. universally closed, resp. universally bicontinuous, resp. a universal homeomorphism), it is necessary and sufficient that for every , its restriction be so.

Assertion (i) results from (I, 4.3.2). Assertion (ii) follows at once from the definitions, and so does (iii) on reducing to the case where , , which one may do thanks to (I, 3.3.11); one knows that (iv) follows from (ii) and (iii) (I, 3.5.1). To prove (v), note that for every morphism , is surjective (I, 3.5.2); one may therefore restrict to proving that if is open (resp. closed, resp. a

homeomorphism onto its image, resp. a bijective homeomorphism), so is , and so the matter is a purely topological question. For the case where is open (resp. closed), the fact that is then open (resp. closed) results from Bourbaki, Top. gén., chap. I, 3rd ed., §5, n° 1, prop. 1; for the two other cases, one may restrict to supposing that , in other words to the case where is a homeomorphism of onto ; since is surjective, is necessarily bijective, and since is a continuous open map by what precedes, is indeed a homeomorphism of onto .

To prove (vi), note that saying that a morphism is open (resp. closed, resp. a homeomorphism onto its image, resp. a bijective homeomorphism) amounts to saying that has the same property. On the other hand (I, 5.1.8), for every morphism , one has (X_red ×_{Y_red} Y'_red)_red = (X ×_Y Y')_red, so the preceding remark shows that if is universally open (resp. universally closed, resp. universally bicontinuous, resp. a universal homeomorphism), so is . The converse is proved similarly, noting here that for every morphism , one has (X_red ×_{Y_red} Y'')_red = (X ×_Y Y'')_red (I, 5.1.3).

Finally, the necessity of (vii) results at once from (iii). Conversely, suppose condition (vii) holds, and let be a morphism; then the form an open cover of , and if one denotes by the restriction of , and by the morphism , the restriction is none other than . One may therefore restrict to proving that is open (resp. closed, resp. a homeomorphism onto its image, resp. a homeomorphism onto ), which is immediate.

Proposition (2.4.4).

A universally bicontinuous morphism is radicial (hence separated (1.7.7.1)).

Indeed, is universally injective by hypothesis (I, 3.5.11).

Proposition (2.4.5).

(i) A morphism that is integral, surjective and radicial is a universal homeomorphism.

(ii) Conversely, suppose locally Noetherian. Then, if a morphism of finite type is a universal homeomorphism, it is finite, surjective and radicial.

(i) It suffices to observe that the three properties of are preserved by base change (I, 3.5.2, I, 3.5.7 and II, 6.1.5), and since an integral morphism is closed (II, 6.1.10), it is clear that is a homeomorphism of onto .

(ii) Since is of finite type and universally closed by hypothesis and separated by (2.4.4), it is proper (II, 5.4.1), and for every , is reduced to a single element; hence (III, 4.4.2) is finite; it is clear that is surjective, and it is radicial since it is universally injective (I, 3.5.11).

Theorem (2.4.6).

Let be a quasi-flat (2.3.3) and locally of finite presentation morphism. Then is universally open. In particular, a flat morphism locally of finite presentation is universally open.

One knows that for every base change , is quasi-flat (2.3.3) and locally of finite presentation (1.4.3, (iii)). It therefore suffices to prove that is an

open morphism. But this follows from criterion (1.10.4) for morphisms locally of finite presentation, conditions b), b'), and c) of (1.10.4) being none other than conditions (i), (ii), (iii) of (2.3.4).

Corollary (2.4.7).

For every prescheme , the structure morphism (where , also denoted ) is universally open.

Indeed, for , , and is a free -algebra of finite presentation.

Remarks (2.4.8).

(i) One notes that a faithfully flat and quasi-compact morphism is not necessarily open, even when and are Noetherian. Take for example for a reduced irreducible algebraic curve (II, 7.4) with generic point , and let be the prescheme sum , being the canonical morphism; it is clear that is flat and surjective, hence faithfully flat, and quasi-compact, but the image under of the open part of is the set , which is not open in (II, 7.4.3).

(ii) For every prescheme , the canonical morphism is a closed immersion and a universal homeomorphism (2.4.4, (vi)); but when is locally Noetherian, is flat only if is reduced, hence (2.2.17).

Proposition (2.4.9).

Let be a discrete prescheme. Then every morphism is universally open.

The question being local on (2.4.3, (vii)), one may restrict to the case where the space underlying is reduced to a point; replacing by (2.4.3, (vi)), one may furthermore suppose that is the spectrum of a field ; on the other hand, for every base change , the open sets of inverse images of the affine open sets of cover , so one may suppose affine, being a -algebra. The issue is therefore to prove that for every -algebra , if one sets , the image under of every open set of the form () is open in . Now, is the direct limit of the increasing filtered family of its sub--algebras of finite type, hence (the functor lim commuting with tensor products), is the direct limit of the -algebras ; there exists such that is the image in of an element , hence is the inverse image under the canonical morphism of the open set (I, 1.2.2.2). But since is a field, is a -algebra of finite presentation and a flat -module, hence is an -algebra of finite presentation and a flat -module; one therefore concludes from (2.4.6) that the image of under is open in . Everything therefore reduces to seeing that ; it is clear that , and it therefore remains to see that for every point , the intersection is non-empty. Now one has , where ; in other words, is a non-empty open set (by definition of ) of the prescheme and is its inverse image under the morphism . Since is a subalgebra of and is a field, the homomorphism

is injective by flatness, so the morphism is dominant (I, 1.2.7), which completes the proof.

Corollary (2.4.10).

Let be a field; , -preschemes; then the projection morphism is universally open. In particular, for every extension of , the projection morphism is universally open.

It suffices to apply (2.4.9) to the structure morphism .

Remark (2.4.11).

If is an open morphism, one knows that, for every part of , one has (Bourbaki, Top. gén., chap. I, 3rd ed., §5, n° 4, prop. 7). This remark applies for example when is a flat morphism locally of finite presentation (2.4.6), or a projection morphism where , are preschemes over a field (2.4.10), and then generalizes (2.3.10).

2.5. Permanence of properties of Modules under faithfully flat descent

Proposition (2.5.1).

Let be a morphism, a quasi-coherent -Module, a faithfully flat morphism, , , . For to be flat (resp. faithfully flat) relative to , it is necessary and sufficient that be flat (resp. faithfully flat) relative to .

It suffices to apply (2.2.10) after replacing , , , , by , , , , respectively; one concludes that for to be flat (resp. faithfully flat) relative to , it is necessary and sufficient that be flat (resp. faithfully flat) relative to . But if is the canonical projection, is faithfully flat (2.2.13) and one has ; hence for to be flat (resp. faithfully flat) relative to , it is necessary and sufficient that be so relative to (2.2.11, (iii)).

Proposition (2.5.2).

Let be a faithfully flat and quasi-compact morphism, a quasi-coherent -Module, . Consider, for a quasi-coherent Module, the property of being:

(i) of finite type;

(ii) of finite presentation;

(iii) locally free of finite type;

(iv) locally free of rank .

Then, if denotes one of the preceding properties, for to possess the property it is necessary and sufficient that possess it.

For a quasi-coherent -Module to be locally free of finite type, it is necessary and sufficient that it be flat over and of finite presentation (Bourbaki, Alg. comm., chap. II, §5, n° 2, cor. 2 of th. 1, taking (2.1.2) into account); since is flat over if and only if is flat over by virtue of (2.5.1) (applied with taken to be the identity), one sees that in order to prove the proposition in case (iii) it suffices to have proved it in cases (i) and (ii); the same holds for (iv), since , so that if and are locally free of finite type and , the rank of at equals that of at , and our assertion follows from the surjectivity of . To treat cases (i) and (ii),

note that the question is local on , and one may therefore suppose affine; one then knows (2.2.12) that there exists an affine scheme X'' and a faithfully flat morphism which is a local isomorphism. Consequently, it amounts to the same thing to say that possesses the property or that does. We are thus reduced to the case where , ; in view of (2.2.3) and (II, 6.1.4.1), it therefore suffices to prove the

Lemma (2.5.3).

Let be a ring, a faithfully flat -algebra, an -module, . For to be of finite type (resp. of finite presentation), it is necessary and sufficient that be so.

For the proof, see Bourbaki, Alg. comm., chap. I, §3, n° 6, prop. 11.

Remark (2.5.4).

The assertions of (2.5.2) for the properties (i) and (ii) are still valid if one supposes only that is quasi-faithfully flat (2.3.3) and quasi-compact. Indeed, one is reduced (Bourbaki, loc. cit.) to proving the

Lemma (2.5.4.1).

Let be a homomorphism of rings such that the corresponding morphism is surjective; suppose there exists an -module of finite type which is -flat and has as support. Then, if is a homomorphism of -modules such that is surjective, is surjective.

Indeed, one deduces first that the homomorphism is surjective. Let be a prime ideal of , and let ; the corresponding homomorphism is surjective, and it can be written . By hypothesis , and is a flat -module ; by virtue of Nakayama's lemma, , and a fortiori , so is a faithfully flat -module . It follows that is surjective , and since this holds for every , being surjective, one finally concludes that is surjective (Bourbaki, Alg. comm., chap. II, §3, n° 3, th. 1).

Proposition (2.5.5).

Let be a morphism, a quasi-coherent -Module of finite type and -flat, a quasi-coherent -Module of finite type; for every , put . For a point to be a maximal point of , it is necessary and sufficient that be a maximal point of and that be a maximal point of in . When this is so, one has

(2.5.5.1)         long((ℱ ⊗_Y 𝒢)_x) = long(𝒢_y) · long((ℱ_y)_x).

It is clear that ; the image under of every irreducible component of is therefore contained in an irreducible component of . One may restrict to the case where . Indeed , since is of finite type, there exists a closed sub-prescheme of having as underlying space and a quasi-coherent -Module of finite type such that, if is the canonical injection, one has . If one then puts , where , it is clear that is -flat and that .

Suppose therefore ; if is an irreducible component of , one has (I, 9.1.13), and it follows from (2.3.4) that every irreducible component of dominates . In other words, if is the generic point of an irreducible component of contained in , then is the generic point of ; furthermore , is the generic point of an irreducible component of (I, 9.1.13), and conversely every generic point of one of these components is the generic point of an irreducible component of .

It remains to prove (2.5.5.1); one has (I, 9.1.12) and ; one is therefore reduced to proving the

Lemma (2.5.5.2).

Let , be two local rings, a local homomorphism, the maximal ideal of . Let be an -module, a -module which is a faithfully flat -module and is such that is a -module of finite length; then one has

(2.5.5.3)         long_B(M ⊗_A N) = long_A(M) · long_B(N/𝔪 N).

If has infinite length, then so does , for every strictly increasing sequence of sub-modules () yields sub-modules of which are pairwise distinct; since (because is a faithfully flat -module), the formula (2.5.5.3) is true in this case. Suppose then that has finite length. If , both members of (2.5.5.3) are zero, so we may suppose . The are sub-modules of , hence of finite length, and if , Nakayama's lemma implies ; consequently there exists necessarily an integer such that . The then identify with sub--modules of , and is isomorphic to , hence also to . The length of the latter, regarded as a -module, is therefore the product of by the rank of the -vector space , which equals the length of the -module . Summing over , one deduces at once the formula (2.5.5.3).

Remark (2.5.5.4).

Note that when is an -module of finite type, to say that is a faithfully flat -module amounts to saying that and that is a flat -module; indeed, Nakayama's lemma then shows that .

Lemma (2.5.6).

Let be a (not necessarily commutative) ring, , two isomorphic left -modules, , and let , equipped with its canonical structure of right -module. Then is a -module isomorphic to ; furthermore, for every , the following conditions are equivalent:

a) is a basis of the -module .

b) is an isomorphism of onto .

If is an isomorphism of onto , the map from to is obviously a bijection, so b) implies a). Conversely, suppose that is a basis of the -module . By hypothesis there exists an isomorphism of onto , and {u'} is then a basis of ; hence there is an invertible element of (i.e. an automorphism of ) such that , which implies that is an isomorphism of onto .

Corollary (2.5.7).

The hypotheses on , , being those of (2.5.6), suppose furthermore that one of the following conditions holds:

(i) and are Noetherian -modules;

(ii) and are modules of finite presentation over a commutative subring of .

Then the conditions a) and b) of (2.5.6) are also equivalent to the following:

a') is a generator of the -module .

b') is an epimorphism of onto .

One knows that an epimorphism of an -module onto itself is bijective in the following two cases: 1° is a Noetherian -module (Bourbaki, Alg., chap. VIII, §2, n° 2, lemma 3); 2° is commutative and is an -module of finite presentation (8.9.3) (¹); hence b) and b') are equivalent. On the other hand, if generates


(¹) The reader may verify that (2.5.7) and (2.5.8) are not used before §9.

and {u'} is a basis of , there exists such that , which proves that is surjective; therefore a') implies b'), and as a) evidently implies a'), this finishes the proof of the corollary.

Proposition (2.5.8).

Let be a commutative semi-local ring, an -algebra (not necessarily commutative), and two -modules. Let be a commutative -algebra which is a faithfully flat -module; put , , , so that is an -algebra and , are -modules. Suppose furthermore that one of the following conditions holds:

(i) and are Noetherian, and are -modules of finite type.

(ii) is an -module of finite type, is a projective -module of finite type and an -module of finite presentation.

Then, if and are isomorphic as -modules, and are isomorphic as -modules.

We note that in case (ii), , being -isomorphic to , is an -module of finite type, from which it follows that is an -module of finite type (Bourbaki, Alg. comm., chap. I, §3, n° 6, prop. 11); hence in all cases and are -modules of finite type. Furthermore:

(2.5.8.1) Under either of the hypotheses (i), (ii), is an -module of finite type.

This is evident in case (i), for is then an -module of finite type, and is an -sub-module of . In case (ii), is a direct factor of a free -module , so is a direct factor of , and since is an -module of finite type, so is .

Put

                  C = End_B(V),     M = Hom_B(V, W),

which are -modules of finite type in cases (i) and (ii). One knows that under either of the conditions (i), (ii), the canonical homomorphism

(2.5.8.2)         Hom_A(V, W) ⊗_A A' → Hom_{A'}(V', W')

is bijective (Bourbaki, Alg. comm., chap. II, §2, n° 10, prop. 11). Since is a flat -module, is canonically identified with a sub--module of . The image of this sub-module under the homomorphism (2.5.8.2) is contained in , for if and , the image of under (2.5.8.2) is the homomorphism such that ; for every , one has , whence our assertion. This being so:

(2.5.8.3) Under either of the hypotheses (i), (ii), the homomorphism

(2.5.8.4)         Hom_B(V, W) ⊗_A A' → Hom_{B'}(V', W')

is bijective.

For every , write (resp. h'(b)) for the homothety of (resp. ), which is an -endomorphism. Let be a system of generators of the -algebra ; the map

                  u ↦ (h'(b_α) ∘ u − u ∘ h(b_α))_α

from to is -linear, and by definition its kernel is precisely ; in other words, one has an exact sequence

                  0 → Hom_B(V, W) → Hom_A(V, W) → (Hom_A(V, W))^I.

The same reasoning applies upon replacing , , , by , , , ; moreover, one has a diagram

                  0 ──→ Hom_B(V, W) ⊗_A A' ──→ Hom_A(V, W) ⊗_A A' ──→ (Hom_A(V, W))^I ⊗_A A'

(2.5.8.5)                       │ r                       │ s                          │ t
                                ↓                         ↓                            ↓

                  0 ──→ Hom_{B'}(V', W') ───→ Hom_{A'}(V', W') ───→ (Hom_{A'}(V', W'))^I

where is the homomorphism (2.5.8.4), is the homomorphism (2.5.8.2), and is the composite homomorphism

                  (Hom_A(V, W))^I ⊗_A A' →w (Hom_A(V, W) ⊗_A A')^I →s^I (Hom_{A'}(V', W'))^I,

being the canonical homomorphism (Bourbaki, Alg., chap. II, 3rd ed., §3, n° 7). One verifies at once that the diagram (2.5.8.5) is commutative, and since is a flat -module its rows are exact. Finally, we have seen

that is an isomorphism, hence so is ; in case (ii), one may take finite, and one then knows that is bijective (Bourbaki, loc. cit., prop. 7); in case (i), one notes that if (resp. B'') is the image of under (resp. ) in (resp. ), then and B'' are -modules of finite type, so one may again take finite. Thus in all cases is bijective, and one concludes that is bijective too.

It therefore follows from (2.5.8.4) that, if one puts

                  C' = C ⊗_A A',    M' = M ⊗_A A',

one has canonical bijections

(2.5.8.6)         C' ≅ End_{B'}(V'),    M' ≅ Hom_{B'}(V', W'),

the first of which is an isomorphism of -algebras, the second forming with the first a di-isomorphism of right -modules.

(2.5.8.7) Reduction to the case . The hypothesis that and are isomorphic -modules implies that and are isomorphic right -modules (2.5.6). We show that to prove (2.5.8), it suffices to find an element which is a generator of as a right -module. Indeed, will be a generator of as a right -module; now in case (i), and are -modules of finite type, hence Noetherian since is Noetherian; in case (ii), and are (isomorphic) -modules of finite presentation. One may therefore apply (2.5.7) to , , , and conclude that is a -isomorphism of onto . But since is faithfully flat over , this implies that is bijective , which is the conclusion of (2.5.8). Noting that in case (i), and are -modules of finite type, and that in case (ii), is (as seen in (2.5.8.1)) a direct factor of , hence a projective -module of finite type, one sees that one is reduced (changing notation) to proving (2.5.8) in the particular case where , and that it suffices to prove the existence of a generator of the -module . Note that in this case is an -module of finite type.

(2.5.8.8) Reduction to the case where and are fields, a simple -algebra with centre . Let be the radical of the semi-local ring ; it suffices to prove that is a monogenic -module, for if there exists a surjective homomorphism , it gives by composition a homomorphism , which itself (since is a free -module) can be written , so that the surjective homomorphism considered is . Since is an -module of finite type, Nakayama's lemma shows that is surjective (Bourbaki, Alg. comm., chap. II, §3, n° 2, cor. 1 of prop. 4). If one puts , , , , the hypotheses (i) (resp. (ii)) remain satisfied when one replaces in them , , , , by A_1, , B_1, , W_1 respectively; furthermore, and are -isomorphic (with ), and is a faithfully flat A_1-module. One may therefore suppose, for the proof of (2.5.8), that is a finite product of (commutative) fields. Since is an -module of finite type, it is an Artinian ring; let be its radical. It will now suffice to prove that is a monogenic -module, for one sees as above, using Nakayama's lemma, that this implies is a monogenic -module; on the other hand, is -isomorphic to , and one has , and likewise . We may therefore further suppose , i.e. that is a semi-simple -algebra.

Note now that since is a finite product of fields (), is a direct composite of -algebras (), each being annihilated by the with ; the hypothesis that is a faithfully flat -module implies that the are non-zero. Consequently, there exists a quotient of which is a field, and A'', the direct composite of the , is a faithfully flat -module and a quotient of . Considering then the ring , is a B''-module isomorphic to ; one may therefore restrict to proving (2.5.8) after replacing by A'', i.e. one may also suppose that is a finite product of fields.

Let be the centre of , which is a finite product of fields and an -module of finite type; note that and are -modules of finite type, projective like every -module; furthermore, one has and , where , and is a faithfully flat -module. One may therefore replace by in the hypothesis, in other words suppose that is the centre of , with semi-simple and a finite product of fields. If () are the field components of , then is a direct composite of simple rings , being the centre of , and is a direct sum of sub-modules (), being annihilated by the with ; furthermore, the reasoning made above shows that one may suppose that is a product of fields (), being an extension of and annihilated by the with . The hypothesis that and are isomorphic -modules then implies that, for

every , and are isomorphic -modules; it therefore suffices to prove (2.5.8) when , i.e. in the case where and are fields and is a simple algebra with centre .

(2.5.8.9) End of the proof. One knows (Bourbaki, Alg., chap. VIII, §5, props. 6 and 8) that every -module is a direct sum of modules isomorphic to a minimal ideal of , and two -modules of finite rank over are therefore isomorphic if and only if they have the same rank over . By hypothesis, one has . But and ; hence , which finishes the proof.

2.6. Permanence of set-theoretic and topological properties of morphisms under faithfully flat descent

Proposition (2.6.1).

Let be an -morphism of -preschemes, a surjective morphism, , , . Consider, for a morphism, the property of being:

(i) surjective;

(ii) injective;

(iii) with finite fibres (as sets);

(iv) bijective;

(v) radicial.

Then, if denotes one of the preceding properties and possesses the property , the same holds for .

Since the projection morphism is itself surjective (I, 3.5.2), one may, by virtue of (I, 3.3.11), restrict to the case where , . For every (resp. ) denote by (resp. ) the fibre prescheme (resp. ) (I, 3.6.2); one knows that for over one has a canonical isomorphism (I, 3.6.4); since the morphism is surjective, so is the projection (I, 3.5.2). Hence if is non-empty (resp. has at most one point, resp. is a finite set), so is ; since is surjective, this proves the proposition in cases (i), (ii) and (iii), and (iv) follows from (i) and (ii). Finally, to prove (v), it suffices to show that if is universally injective (I, 3.5.11), then so is ; now let be an arbitrary morphism; put and . On the other hand, put , , ; since is surjective (I, 3.5.2) and is universally injective, is injective, and it follows from (ii) that is injective, whence our assertion.

Proposition (2.6.2).

The notations being those of (2.6.1), suppose the morphism faithfully flat and quasi-compact. Consider, for a morphism, the property of being:

(i) open;

(ii) closed;

(iii) quasi-compact and a homeomorphism onto the image subspace;

(iv) a homeomorphism.

Then, if denotes one of the preceding properties and possesses the property , the same holds for .

Since the morphism is faithfully flat and quasi-compact (2.2.13 and 1.1.2), one may, by virtue of (I, 3.3.11), restrict to the case where , . If is the projection , one knows that for every subset of , one has (I, 3.4.8). If is an open (resp. closed) morphism, then, for every open (resp. closed) subset of , is open (resp. closed) in , and since is faithfully flat and quasi-compact, one concludes that is open (resp. closed) in by virtue of (2.3.12). This proves the proposition in cases (i) and (ii). Let us prove it in case (iii) (which will imply case (iv), taking (2.6.1, (iv)) into account). By virtue of (2.6.1, (ii)), is injective, and it remains to prove that , viewed as a map of onto , is a quasi-compact and open map. Since is quasi-compact, so is (1.1.4). It therefore suffices to prove that for every closed subset of , one has ; since is surjective, this relation is equivalent to , or again to . Now, since is quasi-compact, so is its composite with the canonical injection ( here being the reduced closed sub-prescheme of having as underlying space). Applying (2.3.10) to the subset of (the image of the morphism ), one obtains ; the formula to be proved therefore amounts to , where ; but this formula follows from the hypothesis that is a homeomorphism of onto f'(X').

Remark (2.6.3).

In cases (i) and (ii), the conclusions of (2.6.2) remain valid when one supposes only quasi-faithfully flat (2.3.3) and quasi-compact; indeed, by virtue of (2.1.4), (I, 3.5.2) and (I, 9.1.13.1), one may again reduce to the case where , ; the conclusion then follows from (2.3.12). In cases (iii) and (iv), the conclusions remain valid when one supposes only quasi-faithfully flat, provided one supposes additionally that is quasi-compact; indeed, one then uses only (2.3.10) and the fact that is surjective. Finally, if is faithfully flat and locally of finite presentation, or if is surjective and discrete, the conclusion of (2.6.2) is valid when is the property:

(iii bis) being a homeomorphism onto the image subspace;

this results indeed from the proof given in (2.6.2) and from Remark (2.4.11).

Corollary (2.6.4).

The notations being those of (2.6.1), suppose the morphism faithfully flat and quasi-compact. Consider for a morphism the property of being:

(i) universally open;

(ii) universally closed;

(iii) quasi-compact and universally bicontinuous;

(iv) a universal homeomorphism;

(v) quasi-compact;

(vi) quasi-compact and dominant.

Then, if denotes one of the preceding properties, for to possess the property , it is necessary and sufficient that possess it.

Properties (v) and (vi) are mentioned only for the record, being consequences of (1.1.4), (1.1.6) and (2.3.7). As for the others, the condition is necessary by virtue of (2.4.3). Conversely, suppose for instance that is universally open, and let be an arbitrary morphism; put and . On the other hand, put , , ; since is faithfully flat and quasi-compact (2.2.13 and 1.1.2) and is universally open, is open, and it follows from (2.6.2) that is open; hence is universally open. The same reasoning applies in the other cases.

One notes again here that one may replace "faithfully flat" by "quasi-faithfully flat", and, when is additionally locally of finite presentation, or when is surjective and discrete, one may replace property (iii) by:

(iii bis) universally bicontinuous.

2.7. Permanence of various properties of morphisms under faithfully flat descent

Proposition (2.7.1).

Let be an -morphism of -preschemes, a faithfully flat and quasi-compact morphism, , , . Consider, for a morphism, the property of being:

(i) separated;

(ii) quasi-separated;

(iii) locally of finite type;

(iv) locally of finite presentation;

(v) of finite type;

(vi) of finite presentation;

(vii) proper;

(viii) an isomorphism;

(ix) a monomorphism;

(x) an open immersion;

(xi) a quasi-compact immersion;

(xii) a closed immersion;

(xiii) affine;

(xiv) quasi-affine;

(xv) finite;

(xvi) quasi-finite;

(xvii) integral.

Then, if denotes one of the preceding properties, for to possess the property , it is necessary and sufficient that possess it.

It has been proved in chapters I, II, and in chapter IV §1, that if possesses one of the above properties , the same holds for (without any hypothesis on the morphism ). It therefore remains to prove the converse; since the projection is

a faithfully flat and quasi-compact morphism (2.2.13 and 1.1.2), one may restrict to the case where , by virtue of (I, 3.3.11).

(i) To say that is separated means that the diagonal morphism is closed; since (I, 5.3.4), if is closed, so is by virtue of (2.6.2), hence is separated.

(ii) has already been proved under weaker hypotheses (1.2.5).

(iii) and (iv): The question is evidently local on and , and, taking (2.2.12) into account, it therefore suffices to prove the

Lemma (2.7.1.1).

Let be a ring, an -algebra, an -algebra which is a faithfully flat -module, . For to be an -algebra of finite type (resp. of finite presentation), it is necessary and sufficient that be an -algebra of finite type (resp. of finite presentation).

One knows already that the condition is necessary without any hypothesis on (1.3.4, 1.3.6, 1.4.3, 1.4.6). Suppose that is an -algebra of finite type; let be the filtered increasing family of -sub-algebras of , so that , and therefore also , since the tensor product commutes with inductive limits; if is a finite system of generators of the -algebra , there exists an index such that all the belong to the sub-algebra of , whence , and since is faithfully flat, .

Suppose now that is an -algebra of finite presentation; one knows already from what precedes that is an -algebra of finite type, so there exists a polynomial -algebra and a surjective -homomorphism of algebras ; let be the kernel of this homomorphism, so that one has an exact sequence , and therefore also an exact sequence (since is -flat), upon putting and (identified with an ideal of ). Since is an -algebra of finite presentation, is a -module of finite type (1.4.4); but one has , and is a faithfully flat -module (2.2.13 and 2.2.3); one knows then that the hypothesis that is a -module of finite type implies that is a -module of finite type (Bourbaki, Alg. comm., chap. I, §3, n° 6, prop. 11); hence is an -algebra of finite presentation.

(v) follows from (iii) and from (2.6.2, (v)) by virtue of (1.5.2).

(vi) follows similarly from (iv), (v) and (ii) by virtue of (1.6.1).

(vii) follows from (i), (v) and (2.6.4, (ii)) (II, 5.4.1).

(viii) Note first that since is an isomorphism, it is a universal homeomorphism, so the same is true of (2.6.4); one already concludes that is quasi-compact and separated (2.4.4). Write , where is therefore a homeomorphism; it must be proved that is an isomorphism of -Modules. Now, if one writes , the homomorphism is composed of the canonical homomorphism and of (2.3.2); but the first of these two homomorphisms is bijective by virtue of the hypothesis on (2.3.1), so if is bijective,

so is , and since is faithfully flat, is bijective (2.2.7), which proves (viii).

(ix) The proposition follows from (viii), from (I, 5.3.4), and from (I, 5.3.8), which reduces monomorphisms to isomorphisms.

(x) If is an open immersion, f'(X') is open in , and one has (I, 3.4.8); it follows from (2.3.12) that is open. One may then replace (resp. ) by the sub-prescheme induced on the open set (resp. f'(X')), taking (1.1.2) and (2.2.13) into account; then becomes an isomorphism, hence the same is true of by (viii), and this establishes (x).

(xi) If is a quasi-compact immersion, is a quasi-compact and quasi-separated morphism (1.2.2), so the same holds for by (ii) and (2.6.2, (v)). Let be the sub-prescheme of closed image of under (1.7.8), and put , where is the canonical injection; one then has with , , and one knows that identifies with the canonical injection of the sub-prescheme of , the closed image of under (2.3.2). The hypothesis on then means that is an open immersion (I, 9.5.10), hence the same holds for by (x), and this shows that is an immersion.

(xii) To say that (resp. ) is a closed immersion means that (resp. ) is a quasi-compact immersion and a closed morphism; one therefore sees that (xii) follows from (xi) and from (2.6.2, (ii)).

(xiii) and (xiv) Suppose affine (resp. quasi-affine); note then that is quasi-compact and quasi-separated (II, 5.1.1), so the same holds for by (ii) and (2.6.2, (v)). Put , ; by virtue of (2.3.1), the canonical homomorphism of -Algebras is bijective; consequently, if is the structure morphism, the structure morphism identifies with (II, 1.5.2). Let then (resp. ) be the canonical -morphism (resp. -morphism) corresponding to the identity homomorphism of (resp. ) (II, 1.2.7); since one has the commutative diagram

                       X  ←—— X'
                       │       │
                       u│      │u'
                       ↓       ↓
                       Z  ←—— Z'
                       │       │
                       h│      │h'
                       ↓       ↓
                       Y  ←—— Y'
                          g

and , it follows from (II, 1.2.7) that . Moreover, is faithfully flat and quasi-compact (1.1.2 and 2.2.13). This being so, the hypothesis on means that is an isomorphism (resp. an open immersion) (II, 5.1.6); it then follows from (viii) (resp. (x)) that is an isomorphism (resp. an open immersion), whence (xiii) (resp. (xiv)).

(xv) If is finite, it is affine, hence so is by (xiii); furthermore, with the notations of the proof of (xiii), is an -Module of finite type, and is isomorphic to ; it follows from (2.5.2) that is an -Module of finite type, hence is a finite morphism.

(xvi) To say that is quasi-finite means that is a morphism of finite type and that for every , is finite (II, 6.2.2 and I, 6.4.4); the conclusion therefore follows from (v) and (xv).

(xvii) One sees as in (xv) that is affine. One may restrict to the case where , , and then , , where ; is equal to the inductive limit of its -sub-algebras of finite type , so one has , where , and is an -algebra of finite type. But by hypothesis is integral over , so is an -module of finite type, and is therefore an -module of finite type (2.5.2). Q.E.D.

Corollary (2.7.2).

The hypotheses and notations being those of (2.7.1), suppose quasi-compact; let be an invertible -Module, its inverse image. For to be ample (resp. very ample) for , it is necessary and sufficient that be ample (resp. very ample) for .

The condition is necessary without any hypothesis on (II, 4.4.10 and 4.6.13); to see that it is sufficient, one may, as in (2.7.1), restrict to the case where , . The hypothesis on implies that is quasi-compact and separated (II, 4.6.1), hence the same holds for (by (2.6.2, (v)) and (2.7.1, (i))). Put , ; it follows from (2.3.1) that the canonical homomorphism is bijective. If is very ample for , the canonical homomorphism is surjective, and the morphism is an immersion (II, 4.4.4, b)), necessarily quasi-compact (1.1.2, (v)). The fact that is bijective implies that, if , are the structure morphisms, then identifies with (II, 4.1.3). On the other hand, denoting by the projection , is faithfully flat (2.2.13), one has , and one verifies easily that the homomorphism is identical with the composite homomorphism

(for example by reducing to the case where and are affine). Since is surjective and is bijective, one sees that is surjective, hence the same is true of (2.2.7). One concludes that the morphism is everywhere defined (II, 3.7.4); furthermore, if one puts , and if g'' is the projection , then identifies with (II, 4.2.10) and g'' is faithfully flat and quasi-compact (1.1.2 and 2.2.13). One therefore concludes from (2.7.1, (xi)) that is an immersion, and consequently is very ample (II, 4.4.4, b)).

Suppose now that is ample for ; to prove that is ample for , one may restrict to the case where is affine (II, 4.6.4), and by virtue of (2.2.12) and (II, 4.6.13), one may also suppose that is affine. Then and are

quasi-compact schemes, and to prove that is -ample, one may apply the criterion of (II, 4.6.8, a)). Let then be a quasi-coherent -Module of finite type; if is the canonical homomorphism, one sees as above that is the composite homomorphism

                  f'*(g*(f_*(ℱ ⊗ ℒ^{⊗ n}))) →u f'*(f'_*(ℱ' ⊗ ℒ'^{⊗ n})) →σ' ℱ' ⊗ ℒ'^{⊗ n},

upon putting , taking into account and denoting by the canonical homomorphism . Now, one knows that is bijective for every (2.3.1); on the other hand, since is quasi-coherent and of finite type, the hypothesis that is ample for implies the existence of an such that is surjective for ; one therefore sees that is surjective for , and since is faithfully flat, is surjective for these values of (2.2.7), which completes the proof.

Remarks (2.7.3).

(i) It follows from (2.6.1), (2.6.4) and (2.5.4.1) that the conclusions of (2.7.1) are still valid in cases (i), (iii), (v), (vii) and (xvi) when one supposes only that is quasi-compact and quasi-faithfully flat (2.3.3); we have already remarked that (2.7.1) is valid in case (ii) under the sole hypothesis that is surjective and quasi-compact.

(ii) With the notations and hypotheses of (2.7.1), it may happen that is proper and projective without being quasi-projective. Indeed, Hironaka [34] has given an example of a proper, non-projective morphism , where and are two regular algebraic schemes over the same field , with projective over ; furthermore, is the union of two affine open sets () such that is projective for . Let then be the sum prescheme; it is clear that the canonical morphism , coinciding with the canonical injections on Y_1 and Y_2, is faithfully flat, and is quasi-compact by virtue of (I, 5.5.10); yet, although (coinciding with on each of the ) is projective (II, 5.5.6), the same is not true of . There therefore exists an invertible -Module which is -ample but is not of the form for an invertible -Module , by virtue of (2.7.2).

(iii) Under the hypotheses of (2.7.1), it may happen that is a local isomorphism without being a local immersion. Indeed, let be a field, an algebraic closure of , a separable extension of finite degree of , distinct from ; then the structure morphism , where and , is not a local immersion, but if one takes , the morphism is faithfully flat and quasi-compact, and is a local isomorphism, since is a sum of a finite number of schemes isomorphic to .

2.8. Preschemes over a regular base of dimension 1; closure of a closed sub-prescheme of the generic fibre

Proposition (2.8.1).

Let be a locally Noetherian, regular, irreducible prescheme of dimension 1, with generic point , a morphism, the fibre at the

generic point, the canonical morphism. Let be a quasi-coherent -Module, , an -Module quotient of , and let be the -Module image of under the composite homomorphism (0_I, 4.4.3.2)

                  ℱ →ρ i_*(i*(ℱ)) = i_*(ℱ_η) → i_*(𝒢_η).

Then is a quasi-coherent and -flat -Module quotient of , such that , and it is the unique -Module quotient of having these properties.

Since is quasi-compact and quasi-separated (1.1.2 and 1.2.2), it follows from (1.7.4) that for every quasi-coherent -Module , is a quasi-coherent -Module; furthermore, for every open of , one has by definition . If one proves the proposition when and are affine, it will follow by gluing in the general case, in view of the uniqueness assertion valid in the affine case. In other words, one is reduced to proving the

Lemma (2.8.1.1).

Let be a regular Noetherian ring (0, 17.3.6), integral and of dimension 1, its field of fractions, an -module, a -module quotient of by a sub--module , the image of under the composite homomorphism . Then is a flat -module, and it is the unique quotient module of which is a flat -module and such that the kernel of the surjective homomorphism equals .

Since for every maximal ideal of , is a regular local ring of dimension 1, hence a discrete valuation ring, it amounts to the same thing to say that an -module is flat or that it is torsion-free . Since is a -vector space, it is a torsion-free -module, so the same is true of , a sub-module of ; furthermore, it is immediate to verify that identifies with . Conversely, if is a quotient -module of having the properties of the statement, the fact that is a flat -module implies that the canonical homomorphism is injective. Since identifies with , the conclusion follows from the commutativity of the diagram

                  M     ──→  N
                  │           │
                  ↓           ↓
                  M_{(K)} ──→ N_{(K)}.

Corollary (2.8.2).

Under the conditions of (2.8.1), for to be -flat, it is necessary and sufficient that the canonical homomorphism be injective.

(2.8.3)

The formation of the -Module is functorial in and : more precisely, if , are two quasi-coherent -Modules, an -homomorphism, an -Module quotient of () and a homomorphism making the diagram

                  (ℱ_1)_η ──→ (ℱ_2)_η
                       │          │
                       ↓          ↓
                  𝒢_{η,1}  ──v──→ 𝒢_{η,2}

commutative (homomorphism uniquely determined (when it exists) by this property), then the diagram

                  ℱ_1            ──→         ℱ_2
                       │                          │
                       ↓                          ↓
                  i_*(𝒢_{η,1})  ──i_*(v)──→  i_*(𝒢_{η,2})

is commutative, and consequently there is a unique homomorphism making the diagram

                  ℱ_1   ──→  ℱ_2
                       │          │
                       ↓          ↓
                  𝒢_1  ──w──→ 𝒢_2

commutative.

Proposition (2.8.4).

The hypotheses on being those of (2.8.1), let X_1, X_2 be two -preschemes, a quasi-coherent -Module, an -Module quotient of (). Then one has

(2.8.4.1)         (𝒢_{η,1} ⊠_{k(η)} 𝒢_{η,2})^∼ = 𝒢_1 ⊠_Y 𝒢_2.

Indeed, put ; the left-hand side of (2.8.4.1) is a quasi-coherent -Module which is -flat , whose inverse image in is (I, 9.1.5), and which is a quotient of ; the conclusion therefore follows from the uniqueness property of (2.8.1).

Proposition (2.8.5).

The hypotheses on and being those of (2.8.1), let be a closed sub-prescheme of . There then exists a unique closed sub-prescheme of which is -flat and such that .

If is the quasi-coherent Ideal of defining , it suffices to apply (2.8.1) to the case where and ; if , one has indeed , so (I, 4.4.5).

One notes that the prescheme is the closed image of under the composite morphism , where the first arrow is the canonical injection (I, 9.5.3); its underlying space is the closure in of (I, 9.5.4), which justifies the notation adopted. One also says that is the closure sub-prescheme of in .

Corollary (2.8.6).

Let X_1, X_2 be two -preschemes, a closed sub-prescheme of (). Then one has

(2.8.6.1)         (Z'_1 ×_{k(η)} Z'_2)^− = Z̄_1 ×_Y Z̄_2.

This results from (2.8.4) and (2.8.5).