§19. Regular and transversally regular immersions

The present section is devoted, on the one hand, to a study of regular immersions (16.9.2) of preschemes, notably in the case where and are flat over the same prescheme ; on the other hand, it gives complements on -regular sequences and normal flatness, generalizing in particular flatness results established by H. Hironaka in his theory of resolution of singularities [35].

19.1. Properties of regular immersions

Proposition (19.1.1).

Let be a locally Noetherian prescheme, an immersion, a regular point of . For to be a regular immersion at the point , it is necessary and sufficient that be regular at the point .

Taking (16.9.10) into account, one is reduced to proving the

Corollary (19.1.2).

Let be a Noetherian local ring, an ideal of ; suppose that the quotient ring is regular. For to be regular, it is necessary and sufficient that the ideal be regular.

Indeed, if is regular, one knows (0, 17.1.9) that is generated by a part of a regular system of parameters of , hence is regular. Conversely, if is regular, and if is an -regular sequence generating , the form a part of a system of parameters of (0, 16.4.1), so one deduces from (0, 17.1.7) that is regular.

Definition (19.1.3).

Let be a prescheme, a sub-prescheme of , an open set of containing and such that is defined by an Ideal of ; suppose that is a locally free -Module (which is the case if is quasi-regularly immersed in ). For every , we call transversal codimension of in at the point , and denote , the rank of the free -module . If is an immersion of image , we likewise call transversal codimension of at the point the transversal codimension in at the point of the sub-prescheme .

Proposition (19.1.4).

Let be a locally Noetherian prescheme, a sub-prescheme of regularly immersed; then for every , one has

  (19.1.4.1)    codim_y^t(Y, X) = codim_y(Y, X).

By virtue of (5.1.3.2), is equal to the smallest of the dimensions of the local rings , where is the generic point of an irreducible component of containing

; as such a point is contained in every neighbourhood of in , is a free -module of rank ; it suffices to see that . Now, since is a maximal point of the sub-prescheme defined by , one has , so is an ideal of definition of the Noetherian local ring ; furthermore, the hypothesis that is quasi-regular entails that is a free -module of rank ; if is the length of the Artinian ring , the length of is therefore ; the Hilbert-Samuel polynomial of for the -preadic filtration is therefore of degree , which proves the proposition (0, 16.2.3).

By virtue of (19.1.4), we shall henceforth say "codimension" instead of "transversal codimension" when dealing with a sub-prescheme regularly immersed in (even when is not locally Noetherian).

Proposition (19.1.5).

(i) For an immersion to be open, it is necessary and sufficient that it be regular and of codimension 0 at every point.

(ii) Let be a regular (resp. quasi-regular) immersion, a flat morphism, ; then the morphism is a regular (resp. quasi-regular) immersion, and for every , if is the projection of in , the codimension of at the point is equal to the codimension of at the point . Conversely, if is an immersion and a faithfully flat and quasi-compact morphism, then, if is a quasi-regular immersion, so is .

(iii) If and are regular immersions, then is a regular immersion, and for every , the codimension of at the point is the sum of the codimension of at the point and of the codimension of at the point . Moreover, the sequence of canonical homomorphisms (16.2.7.1)

is exact, and for every , there exists a neighbourhood of in in which the restrictions of the homomorphisms of this sequence form a split sequence.

(iv) Let be a locally Noetherian prescheme, and two immersions, a point of . The following conditions are equivalent:

a) is a regular immersion at the point , and is a regular immersion at the point .

b) and are regular immersions at the point .

c) is a regular immersion at the point , is a regular immersion at the point , and the sequence of canonical homomorphisms (19.1.5.1), restricted to a sufficiently small open neighbourhood of in , is exact and split.

(v) Let be a locally Noetherian prescheme, , two immersions, a point of ; suppose that is a regular point of and a regular point of (which entails, by (19.1.1), that is a regular immersion at the point ); then,

for to be a regular immersion at the point , it is necessary and sufficient that be a regular immersion at the point .

Assertion (i) is another way of expressing (16.1.10). To prove the first assertion of (ii), one may restrict to the case where is a closed sub-prescheme of defined by a regular (resp. quasi-regular) Ideal ; then is the closed sub-prescheme of defined by the Ideal (I, 4.4.5), which is here identified with since is flat; by replacing by a sufficiently small affine open set, one may suppose that admits a regular (resp. quasi-regular) sequence of generators (sections of over ). If the sequence is regular, so is the sequence (which generates ) by virtue of (0, 15.2.5); the analogous result for quasi-regular sequences follows at once from the flatness of and the criterion (16.9.4). Likewise, if is faithfully flat and quasi-compact, and if is quasi-regular, is of finite type by virtue of (2.5.2); by flatness, one has , hence, since is locally free by hypothesis (16.9.4), so is (2.5.2). Finally, condition (iii) of (16.9.4) relative to entails the same condition for by virtue of the faithful flatness of (2.2.7). One therefore concludes by (16.9.4) that is quasi-regular.

To prove (iii), one may likewise suppose that and are closed sub-preschemes of , defined respectively by Ideals , of with ; furthermore, one may suppose that there exist sections of over generating and such that, for , the endomorphism of defined by is injective, and sections of over generating and such that, for , the endomorphism of defined by is injective. For every , let be an open neighbourhood of in such that there exists a section of over whose class in is for ; then is isomorphic to , and one therefore sees that the sequence is regular in ; as it generates , this proves the first assertion of (iii). To prove the last assertion of (iii), one may restrict (with the same notations) to the case where the images of the in (for ) form a basis of this -Module (16.9.5); for the same reason, one may suppose that the for are the canonical images of elements of a basis of the -Module , and that the canonical images of the in for form a basis of this -Module. This completes the proof of (iii).

Let us pass to the proof of (iv). The fact that a) implies c) follows from (iii). Let us prove that c) implies a). Set again , so that ,

with . By hypothesis, and are regular ideals of and one has a split exact sequence

(cf. (16.2.7)). This entails that there exist elements of whose images in form a basis of this -module, and such that moreover the of index belong to and are such that their images in form a basis of this -module. The for therefore generate , and the for generate since is Noetherian (Bourbaki, Alg. comm., chap. II, §3, n° 2, prop. 5); as is a regular ideal, it follows from (16.9.5) that the sequence is regular. By definition, the images of in therefore form a regular sequence in this -module, which completes the proof that c) entails a) in (iv).

Finally, taking (16.9.10) into account, it is immediate that (v), as well as the equivalence of a) and b) in (iv), result from the

Corollary (19.1.6).

Let be a Noetherian local ring, , two ideals of contained in its maximal ideal and such that , , . Then:

(i) If and are regular ideals, so is .

(ii) If and are regular ideals, so is .

(iii) Suppose that the rings and are regular. For the ideal to be regular, it is necessary and sufficient that be regular.

Assertion (i) is a particular case of (19.1.5, (iii)). To prove (ii), consider the canonical surjective homomorphism . As by hypothesis and are free -modules, whose respective ranks we denote and , the kernel of is a projective -module (Bourbaki, Alg., chap. II, 3rd ed., §2, n° 2, prop. 4), hence free of rank since is a Noetherian local ring ; in other words, there exists a basis of whose first elements form a basis of . This means again (by virtue of Nakayama's lemma) that there exists a system of generators of , forming a regular sequence in , such that the sequence is formed of elements of ; we must show that this latter sequence generates . To do so, denote by the ideal it generates; by considering the ring , one sees that one is reduced to proving the following lemma:

Lemma (19.1.6.1).

Let be a Noetherian local ring of maximal ideal , a regular sequence in of elements of , the ideal it generates, an ideal such that the canonical images of the in again form a regular sequence in this ring. Then one has .

The lemma being trivial for , let us reason by induction on . The induction hypothesis, applied to the ring and to the canonical images of the

and of in this quotient ring, shows that . Let be the ideal of formed by such that ; one then has ; but since by hypothesis is regular in , one necessarily has , hence . As is of finite type and is contained in the radical of , Nakayama's lemma shows that .

Let us finally prove assertion (iii) of (19.1.6). By virtue of (i) and of (19.1.2), it suffices to see that if is regular, so is . Note that is regular (19.1.2), and let us reason by induction on the rank of the free -module . If , one has by Nakayama's lemma and the assertion is trivial. As and are regular, one knows (0, 17.1.9) that is generated by a part with elements of a regular system of parameters of ; so, if is one of the elements of this system of generators of , is regular and (0, 17.1.8). Let , be the canonical images of , in A_1; one has , , so is regular and consequently is a regular ideal (19.1.2). Let us show that is a regular ideal in A_1; as is a regular ideal in , it suffices to show that is part of a regular system of generators of , and for this (16.9.5) it suffices to show that the image of in is part of a basis of this -vector space, in other words that , which indeed follows from . Then, as is a free -module of rank , the induction hypothesis shows that is a regular ideal of , and as is identified with in the canonical isomorphism between and , is regular. Q.E.D.

Remarks (19.1.7).

(i) We do not know whether the composite of two quasi-regular immersions is quasi-regular.

(ii) In a ringed space in local rings , for every Ideal of , is of finite type and so has closed support in . One may consequently define, as in (I, 4.1.3 and 4.2.1), the notions of (locally closed) ringed sub-space of defined by an Ideal of a sheaf , where is an open set of , and the notion of immersion. The definitions of quasi-regular and regular immersions, of transversal codimension then apply without modification, and the results of (19.1.5, (i) to (iv)) are still valid, supposing in the second assertion of (i) and in (iv) that the sheaf is coherent and the local rings Noetherian; as for (ii), one must define as the ringed sub-space defined by the -Ideal ; the proofs are unchanged.

(19.1.8)

We have already pointed out (notably in (16.9.6, (ii))) that in non-Noetherian rings, the notion of "regular sequence" of elements of the ring does not possess the properties that would make it usable. Recent research seems to show that the notion that should in this case replace that of regular sequence is that of a sequence having the property that the exterior-algebra complex (III, 1.1.3) has its homology zero in degrees > 0 (cf. (III, 1.1.4)); see in particular A. Grothendieck, Séminaire de géométrie algébrique, 1966, to appear shortly.

19.2. Transversally regular immersions

Definition (19.2.1).

Let be a morphism of preschemes, a quasi-coherent -Module. We say that a sequence of sections of over is transversally -regular relatively to (or relatively to ) if the sequence is -regular and such that the -Modules are -flat for . We say that a quasi-coherent Ideal of is transversally regular relatively to (or relatively to ) at a point if there exists an open neighbourhood of and a finite sequence of sections of over which is transversally -regular relatively to and which generates .

In an -algebra , we say that an ideal is transversally regular relatively to if is transversally regular relatively to at every point of in .

Definition (19.2.2).

Let be a prescheme, , two -preschemes, an -morphism which is an immersion. We say that is a transversally regular immersion relatively to at a point if there exists a neighbourhood of in such that the sub-prescheme induced on by the sub-prescheme of associated to is defined by an Ideal of transversally regular relatively to at the point .

It is clear that the set of points of where an immersion is transversally regular relatively to is open in ; if it equals , one says simply that is a transversally regular immersion relatively to .

If is a flat morphism, then if is an Ideal of transversally regular relatively to at , it is so also relatively to . An -immersion transversally regular at a point relatively to is therefore also a -immersion transversally regular relatively to at this point.

Proposition (19.2.3).

Let be a prescheme, , two -preschemes, an -morphism which is a transversally regular immersion relatively to at a point . Then, for every base change , if one sets , , the immersion is transversally regular relatively to at every point above .

This follows at once from the definition and from (0, 15.1.15).

Proposition (19.2.4).

Let be a prescheme, , two -preschemes, an -morphism which is an immersion. Suppose either that and are locally Noetherian, or that the structure morphisms , are locally of finite presentation. Let be a point of , its image in . The following conditions are equivalent:

a) is transversally regular relatively to at the point .

b) The morphisms and are flat in a neighbourhood of the points and respectively, and if one sets , , the immersion is regular at the point .

b') The morphisms and are flat in a neighbourhood of the points and respectively, and, for every base change , if one sets , , the immersion is regular at every point of above .

c) The morphism is flat in a neighbourhood of the point , and the immersion is regular in a neighbourhood of the point .

c') The morphism is flat in a neighbourhood of the point , and the immersion is quasi-regular in a neighbourhood of the point .

It was already proved in (19.2.3) that a) entails b') without finiteness hypothesis, and it is trivial that b') entails b). Let us show that b) entails c). For this (the question being local on and on ) one may suppose that is a closed sub-prescheme of defined by a quasi-coherent Ideal of , and the canonical injection. The hypothesis that is -flat entails that the sequence

  0 → 𝓘 ⊗_{𝒪_S} k(s) → 𝒪_X ⊗_{𝒪_S} k(s) → (𝒪_X/𝓘) ⊗_{𝒪_S} k(s) → 0

is exact , so is the closed sub-prescheme of defined by the Ideal of which is identified with . Consider a finite sequence of sections of over a neighbourhood of in whose images in form a basis of this -module (which is free by virtue of the hypothesis b)). Since is locally Noetherian, it follows from (16.9.11), (16.9.5) and the hypothesis b) that the images , sections of over a neighbourhood of in , form a regular sequence generating the restriction of to this neighbourhood. As is contained in the radical of , and as in both cases considered is an ideal of finite type of , it follows from Nakayama's lemma that the also generate ; as is an Ideal of finite type of in both cases considered, there is a neighbourhood of in such that the generate . The fact that b) entails c) is then a consequence of (0, 15.1.16) when and are locally Noetherian, and of (11.3.8) when and are locally of finite presentation; in this latter case, (11.3.8) also proves the equivalence of c) and c') (which is trivial when and are locally Noetherian (0, 15.1.11)). Finally, if c) is verified, to prove a) one may again restrict to the case where is a closed sub-prescheme of defined by , and by hypothesis there is a regular sequence of sections of over a neighbourhood of in such that the generate and is flat in ; to see that c) entails a), it again suffices to apply (0, 15.1.6) when and are locally Noetherian, and (11.3.8) when and are locally of finite presentation.

Remarks (19.2.5).

(i) When in condition b), one suppresses the hypothesis that is flat in a neighbourhood of , the thus modified condition no longer entails a). It suffices to take for the spectrum of a local ring which is not a field, for the closed point of , and .

(ii) Suppose that is a closed sub-prescheme of defined by a quasi-coherent Ideal of . It was proved in the course of the proof of (19.2.4) that, when the equivalent conditions of this proposition are fulfilled, then, for every system of sections of over a neighbourhood of in such that the images of the in form a basis of this -module, there exists an open neighbourhood of in such that the satisfy the conditions of definition (19.2.1).

Corollary (19.2.6).

Suppose either that and are locally Noetherian, or that and are locally of finite presentation. Suppose moreover that the fibres and are regular preschemes at the points and respectively, and that the morphisms and are flat in a neighbourhood of and respectively. Then the immersion is transversally regular at the point .

Indeed, it follows from (19.1.1) that the immersion is regular at the point , and it suffices to apply (19.2.4, b)).

Proposition (19.2.7).

(i) Every -open immersion is transversally regular relatively to .

(ii) Let be an -immersion, a morphism, and set , , . If is a transversally regular immersion relatively to , is a transversally regular immersion relatively to . The converse is true if is faithfully flat, and moreover if and are locally of finite presentation over .

(iii) If and are transversally regular immersions relatively to , so is .

(iv) Let be an -prescheme, , two -immersions. Suppose either that and are locally Noetherian, or that , and are locally of finite presentation over . Then, if and are transversally regular at the point relatively to , is transversally regular at the point relatively to .

(v) Under the general conditions of (iv), suppose that and are -flat, regular at the point and regular at the point ; then, if is transversally regular at the point , so is .

Assertion (i) is immediate, and the first assertion of (ii) follows at once from (19.2.3). To prove the second assertion of (ii), let us apply (19.2.4): and being locally of finite presentation over , it follows first from this criterion that and are flat over , so and are flat over (2.5.1); on the other hand, for every , if is a point above , is by hypothesis a regular immersion, and one deduces from (19.1.5, (ii)) that the same holds for , since these preschemes are locally Noetherian and it then amounts to the same to say that the immersion is regular or quasi-regular. To prove (iii), one reduces to the case where is a closed sub-prescheme of and a closed sub-prescheme of , and one forms as in (19.1.5, (iii)) a system of generators of the Ideal of defining . To prove (iv), one notes that the hypothesis combined with (19.2.4) shows first that , and are -flat in a neighbourhood of , and it suffices moreover, if is the image of in , to prove that the immersion is regular at the point , knowing that the same holds for and for ; but this follows from (19.1.5, (iv)). Finally, for (v), one reasons in the same way as for (iv); the hypothesis on already implies by (19.2.4, b)) that is -flat in a neighbourhood of , and the same holds for and by hypothesis; one is then reduced to proving that is regular at the point , knowing that the same holds for , which follows from (19.1.5, (v)), given the hypotheses made on and .

Proposition (19.2.8).

Let be an -prescheme, a closed sub-prescheme of defined by a quasi-coherent Ideal of , the canonical injection. If is transversally regular relatively to , the -Modules and are -flat at the points of . For every base change , if , and , one has up to isomorphism, for every .

Indeed, and are by hypothesis -flat at the points of , and is a locally free -module (16.9.3), so it is -flat at the points of . The first assertion then follows from and the second from (11.2.9.2).

The following proposition generalizes and refines (17.16.1):

Proposition (19.2.9).

Let be a morphism, a point of , . Suppose that is of finite presentation at the point , and is a Cohen-Macaulay morphism at the point (6.8.1). Then there exists a sub-prescheme of containing and having the following properties:

(i) The canonical injection is transversally regular relatively to .

(ii) The morphism is Cohen-Macaulay (which amounts to saying, taking (i) into account, that all the fibres of are Cohen-Macaulay preschemes).

(iii) The point is a maximal point of .

Set ; by hypothesis, the ring is Cohen-Macaulay; let be a system of parameters of this ring, which is therefore a regular sequence (0, 16.5.7). There is an open neighbourhood of in and sections of over such that the are the canonical images of the in the quotient ring of . Let be the Ideal of generated by the and take for the closed sub-prescheme of defined by . One may suppose that is locally of finite presentation, and it is therefore the same for by definition of ; it follows then from the equivalence of c') and a) in (11.3.8) that the sequence is regular in and that is flat over ; by virtue of (11.3.1), and are therefore flat in a neighbourhood of and it follows from (19.2.4) that is a transversally regular immersion relatively to (replacing if necessary by for an open neighbourhood of in ). One has thus proved (i) and one sees that one may suppose that is a flat morphism. As , quotient of by the ideal generated by the , is Artinian by construction (0, 16.3.6), is a maximal point of . Finally, the ring being Artinian is a Cohen-Macaulay ring (0, 16.5.1), so, by replacing if necessary by its intersection with an open neighbourhood of in , one may suppose that is a Cohen-Macaulay morphism, by virtue of (12.1.7, (iii)).

In particular, one recovers the existence of "flat quasi-sections" of a morphism at a point which is closed in and such that is flat at the point and is a Cohen-Macaulay ring (17.16.1), and one sees moreover that the immersion is transversally regular relatively to . This last complement is also valid for the étale quasi-section defined in (17.16.3, (i)).

19.3. Relative complete intersections (flat case)

Definition (19.3.1).

Let be a Noetherian local ring. We say that is a complete intersection ring (or also absolute complete intersection ring, when there is danger of confusion) if the completion  is isomorphic to the quotient of a complete regular Noetherian local ring by a regular ideal (i.e. (16.9.7) an ideal generated by a regular sequence of elements of ).

Every regular local ring is a complete intersection ring.

One says that a locally Noetherian prescheme is a (absolute) complete intersection at the point if the ring is a complete intersection ring.

Proposition (19.3.2).

Let be a regular Noetherian local ring, an ideal of . For to be a complete intersection ring, it is necessary and sufficient that the ideal be regular (i.e. (16.9.7) generated by a regular sequence of elements of ).

One may write and since is a faithfully flat -module and Noetherian, is regular if and only if is (19.1.5, (ii)). This shows (by virtue of (0, 17.1.5)) on the one hand that the condition of the statement is sufficient, and on the other hand that to prove it is necessary, one may restrict to the case where and are complete. By hypothesis, there then exists a complete regular Noetherian local ring such that is isomorphic to a quotient , where is a regular ideal of . Consider then the fibre product for the surjective homomorphisms , (0, 18.1.2); we shall first prove the following lemma:

Lemma (19.3.2.1).

Let , , be three rings, , two homomorphisms, and set (0, 18.1.2).

(i) If is surjective and if and are Noetherian rings, so is .

(ii) If , and are local rings and and are local homomorphisms, is a local ring and the canonical homomorphisms , are local.

(iii) If the conditions of (i) and (ii) are verified, if is surjective, and if and are complete Noetherian local rings, so is .

(i) Taking into account (0, 18.1.3 and 18.1.5), is an -ring augmented over , whose augmentation ideal is canonically identified with the kernel of , the -module structure on being identified with that of -module on via the canonical homomorphism . If is any ideal of , is isomorphic to , hence to an ideal of , and is therefore of finite type; on the other hand, is isomorphic to an ideal of contained in , hence is also of finite type, which proves that is of finite type, so is Noetherian.

(ii) Let , be the maximal ideals of and respectively; the set is evidently an ideal of , the inverse image of by the projection and of by the projection . One deduces at once that the elements of are non-invertible in . On the other hand, if , and if for instance , does not belong to the maximal ideal of , hence, since , and since is local, one has also ; one

concludes that and are invertible, so the same holds for , which completes the proof of (ii).

(iii) The image of by the surjective homomorphism is the maximal ideal of , so the image of is , and as is closed (hence complete) for the topology induced by the -adic topology of , is complete for the topology induced by the -preadic topology of . On the other hand, , endowed with the -preadic topology, is isomorphic to endowed with the -adic topology, hence is complete for the quotient topology of the -preadic topology of . One deduces at once that is complete for the -preadic topology.

This lemma being established, it follows from Cohen's theorem (0, 19.8.8) that B'' is isomorphic to a quotient of a complete regular Noetherian local ring . It then follows from (19.1.2) that the immersion is regular; as by hypothesis the same holds for the immersion , one concludes from (19.1.5, (iii)) that the immersion is regular; but this immersion is also written as the composite Spec(A) → Spec(B) → Spec(C). Now as is regular by hypothesis, the immersion is regular (19.1.2); so the same holds for by (19.1.5, (iv)).

Corollary (19.3.3).

Let be a locally Noetherian prescheme locally immersible in a regular prescheme. Then the set of where is a complete intersection is open in and contains the set of regular points of .

One may indeed restrict to the case where , where is a regular ring and an ideal of . The set of in such that is regular in is then open (16.9.5), and it is the set of points of where is a complete intersection by virtue of (19.3.2).

Corollary (19.3.4).

Let be a field, a prescheme locally of finite type over , an extension of , . Let be a point of , its projection in . For to be a complete intersection at , it is necessary and sufficient that be a complete intersection at .

The question being local on , one may suppose that , where is a -algebra of finite type, hence a quotient of a polynomial algebra , so that is a closed sub-prescheme of the regular scheme (0, 17.3.7). If one sets , is also a regular scheme. Now, since is faithfully flat over , it follows from (19.1.5, (ii)) and from the fact that these are Noetherian preschemes that the immersion is regular at the point if and only if the immersion is regular at the point . The conclusion therefore follows from the criterion (19.3.2).

Corollary (19.3.5).

Let , be two Noetherian local rings, a local homomorphism, a quotient ring of , the residue field of . Suppose that makes a flat -module, and that the ring is regular. Then the following conditions are equivalent:

a) The ideal is transversally regular relatively to .

b) is a flat -module, and is a complete intersection ring.

By virtue of (19.2.4), condition a) amounts to saying that is a flat -module and that (which is identified with an ideal of by virtue of the flatness of over ) is regular in this ring. Since the ring is regular, it amounts to the same, by virtue of (19.3.2), to say (when is a flat -module) that is a regular ideal of , or that is a complete intersection ring; whence the corollary.

Definition (19.3.6).

Let be a flat morphism locally of finite presentation. We say that is a relative complete intersection relatively to at a point if the fibre is an (absolute) complete intersection at the point . We say that is a relative complete intersection relatively to , or that the morphism is a complete intersection morphism, if is a relative complete intersection relatively to at each of its points.

Proposition (19.3.7).

Let , be two flat morphisms locally of finite presentation, an -immersion, a point of , , . Suppose that the fibre is regular at the point . Then, for to be a transversally regular immersion at the point , it is necessary and sufficient that be a relative complete intersection relatively to at the point .

To say that is transversally regular at the point amounts, by virtue of (19.2.4), to saying that is a regular immersion at the point . But as is regular at the point , this is equivalent, by virtue of (19.3.2), to saying that is an (absolute) complete intersection at the point , whence the proposition.

Corollary (19.3.8).

Let be a flat morphism locally of finite presentation. The set of at which is a relative complete intersection relatively to is open in . If moreover is proper, the set of such that is a complete intersection at each of its points is an open part of .

As , the second assertion follows from the first and from the fact that is a closed map. The first assertion is local on , so one may suppose that is a closed sub-prescheme of a prescheme of the form (the indeterminates). This latter being smooth over (17.3.8), its fibres are regular, and it amounts to the same, by virtue of (19.3.7), to say that or that the immersion is transversally regular at the point ; one therefore concludes the corollary by (19.2.2).

Proposition (19.3.9).

(i) Every open immersion is a complete intersection morphism.

(ii) Let be a flat morphism locally of finite presentation which is a complete intersection morphism. For every base change , is a complete intersection morphism. The converse is true if is faithfully flat and quasi-compact.

(iii) If and are two flat morphisms locally of finite presentation which are complete intersection morphisms, so is .

Assertion (i) follows at once from definition (19.3.1). To prove (ii), let us note first that if is flat and locally of finite presentation, so is

(2.1.4), the converse being true if is faithfully flat ((2.5.1) and (2.7.1)); furthermore, for every , if , it is equivalent to say that is a complete intersection or that is a complete intersection (19.3.4); whence (ii), taking into account that is surjective when it is faithfully flat.

Finally, under the hypotheses of (iii), is flat and locally of finite presentation. By definition (19.3.6), one is therefore reduced to the case where is the spectrum of a field, and to proving then that the local rings are complete intersection rings, which is contained in the more general following result:

Corollary (19.3.10).

Let be a flat morphism locally of finite presentation which is a complete intersection morphism, a point of , . If is a Noetherian complete intersection ring, so is .

One may obviously restrict to the case where with . Let us show moreover that one may restrict to the case where the local ring is complete. Indeed, set , , and let be the unique closed point of , which is the unique point of above ; if , the fibre is canonically isomorphic to , since and  have the same residue field (I, 3.6.4); there is thus a single point above and above , and one consequently has . One deduces that the local rings and have isomorphic completions, for in general, if is a local ring which is an -algebra (the homomorphism being local), the separated completion of the ring endowed with the tensor product topology is isomorphic to the separated completion of , hence to the separated completion Ê of . By virtue of definition (19.3.1), it therefore amounts to the same to say that is a complete intersection ring, or that is a complete intersection ring.

Suppose then complete; one may moreover suppose that , being a quotient of a polynomial algebra . As polynomial rings over a field are regular rings (0, 17.3.7), it follows from (19.3.7) that the immersion may be supposed transversally regular relatively to ; so one may suppose that , where is generated by a regular sequence of elements of . On the other hand, since is complete, one may by hypothesis write it , where is a regular local ring and an ideal of (0, 19.8.8), and the hypothesis that is a complete intersection ring entails that is generated by a regular sequence (19.3.2). In the polynomial ring , the elements still form a regular sequence generating the ideal (0, 15.1.4); as , one sees that if, for each , is an element of of image in , the sequence formed of the and of the is regular in ; if is the ideal of it generates, one has , whence the conclusion by virtue of (19.3.2), since is a regular ring (0, 17.3.7).

19.4. Application: regularity and smoothness criteria for blow-up preschemes

(19.4.1)

Let be a prescheme, a closed sub-prescheme of defined by a quasi-coherent Ideal of . Recall (II, 8.1.3) that the -scheme obtained by blowing up is the prescheme

  X' = Proj(𝒮),    where    𝒮 = ⊕_{n ⩾ 0} 𝓘^n.

If is of finite type, the structure morphism is projective. Without hypothesis on , the closed sub-prescheme

of is defined by the quasi-coherent Ideal of , canonically isomorphic to ; more precisely, one has an exact sequence

  (19.4.1.1)    0 → 𝓘 𝒪_{X'} → 𝒪_{X'} → 𝒪_{Y'} → 0

where is the image of (II, 8.1.7 and 8.1.8). As in a neighbourhood of a point of , the -Module is free of rank 1, is generated in such a neighbourhood by an invertible section of , and is consequently a free -Module of rank 1. This proves the first assertion of the following lemma:

Lemma (19.4.2).

The canonical immersion is regular and of codimension 1, and is canonically isomorphic to the -scheme .

The second assertion follows from (II, 3.5.3) applied to the canonical injection , taking into account that by definition, since is isomorphic to and to .

Corollary (19.4.3).

If the canonical immersion is quasi-regular, the restriction of the morphism is smooth.

Indeed, is then a locally free -Module of finite rank, and is isomorphic to (16.9.8), so is -isomorphic to and consequently is smooth over (17.3.9).

We shall see later (chap. V) that is still smooth when the immersion "presents only ordinary singularities".

Proposition (19.4.4).

With the notations of (19.4.1), suppose that is locally Noetherian and that the morphism , restriction of , is smooth. Let be a point of , . For to be regular at the point , it is necessary and sufficient that be regular at the point ; when this holds, is then regular at the point .

The first assertion follows from (17.5.8) and the second from (19.1.1) and (19.4.2).

Corollary (19.4.5).

Under the general hypotheses of (19.4.4), suppose that is regular at the point . Then, for every generization of in not belonging to , is regular at the point . If is open ((6.12), for example if is excellent (7.8.6, (iii))),

and if is regular, then there exists an open neighbourhood of in such that is regular at the points of .

To prove the first assertion, one may restrict to the case where is a local scheme of closed point ; the hypothesis then entails that is regular (0, 17.3.2), so the same holds for by (19.4.4). Note now that the morphism is projective, hence proper, and therefore (II, 7.2.1), if one considers the unique point of above (II, 8.1.3), there exists a specialization of belonging to ; applying (19.4.4) and using (0, 17.3.2) one concludes that is regular at the point , hence regular at the point since is isomorphic to (II, 8.1.3). If is regular, every point of which is a generization of a point of therefore belongs to ; if is open, is therefore locally constructible and stable under generization, hence open and answers the question.

Proposition (19.4.6).

Let be a morphism, a closed sub-prescheme of defined by a quasi-coherent Ideal , the -scheme obtained by blowing up , the inverse image of in .

(i) The following conditions are equivalent:

a) is a -flat -Algebra.

b) For every , the -Module is -flat (in other words, the -th infinitesimal neighbourhood of in (16.1.2) is -flat).

c) For every base change and every integer , if one sets , the canonical homomorphism is bijective.

When this holds, is -flat, and for every base change , if Y_1 is the inverse image of in X_1, the prescheme obtained by blowing up Y_1 in X_1 is canonically isomorphic to .

(ii) Suppose that the equivalent conditions of (i) are satisfied and moreover that the morphisms and are locally of finite presentation. Then is an -Algebra of finite presentation, the morphism is of finite presentation, the canonical immersion is transversally regular of codimension 1 relatively to , and the set of points of (resp. ) where (resp. ) is -flat is a neighbourhood of (resp. ).

(i) The equivalence of a) and b) follows at once from ; to show the equivalence of b) and c), one may restrict to the case where and are affine, with , where is an ideal of ; the exact sequence of Tor then shows that condition b) implies that for every integer and every -algebra ; by taking for an algebra direct sum , where is an arbitrary -module and where multiplication is defined by , one sees that the preceding condition is equivalent to , hence to the fact that is a flat -module. If one sets , one then has

and the assertion relative to the blown-up prescheme follows at once. Finally, to prove that is -flat, one may again restrict to the case where and are affine; if one sets , every point of then admits, by virtue of (19.4.2) and (II, 2.3.6), an affine open neighbourhood whose ring is of the form , where is a homogeneous element of degree 1 of ; since by hypothesis is a flat -module, so is its ring of fractions and the degree-0 component of this graded -module, whence our assertion.

(ii) One may always restrict to the case where and are affine, being therefore an -algebra of finite presentation, an ideal of finite type of . By virtue of (8.9.1), (8.6.3) and (11.2.9), there exist a Noetherian sub-ring A_0 of , an A_0-algebra of finite type B_0, an ideal of B_0 such that , , such that is an A_0-flat module, and that . By virtue of (i), one then has ; as is a B_0-algebra generated by the ideal of finite type of B_0, hence a B_0-algebra of finite type, and consequently of finite presentation since B_0 is Noetherian, one deduces that is a -algebra of finite presentation. Likewise, the morphism is of finite type (II, 2.7.1), and consequently of finite presentation; so the morphism , which is deduced from it by base change, is of finite presentation. One already knows (19.4.2) that the canonical immersion is regular; as is -flat, as one saw in (i), one deduces from (19.2.4) that the immersion is transversally regular relatively to and that is -flat at the points of , hence (11.3.1) in an open neighbourhood of . On the other hand, the fact that is an -flat -Algebra entails, by (11.3.4), that is -flat at the points of , hence (11.3.1) in a neighbourhood of .

Corollary (19.4.7).

Let be a morphism locally of finite presentation, a closed sub-prescheme of such that the composite morphism is locally of finite presentation; suppose moreover that is -flat and that is normally flat along (11.3.4). Then (with the notations of (19.4.1)), the morphism is of finite presentation, the canonical immersion is transversally regular of codimension 1 relatively to , is -flat (hence -flat), and the set of points of (resp. ) where (resp. ) is -flat is a neighbourhood of (resp. ).

Indeed, by hypothesis is a flat -Module, hence is -flat since is supposed to be -flat. One may consequently apply the results of (19.4.6, (i) and (ii)). As is -isomorphic to (19.4.2), the same reasoning as in (19.4.6) shows that is -flat.

Proposition (19.4.8).

Suppose the hypotheses of (19.4.7) are verified and moreover that the morphism is smooth. Let be a point of , its image in . For to be smooth over at the point , it is necessary and sufficient that be smooth over at the point ; when this holds, is then smooth over at the point .

The preceding conclusions are in particular verified when the morphism is locally of finite presentation and the immersion is transversally regular relatively to .

Taking into account the compatibility of the formation of a blow-up prescheme with base changes in the case considered (19.4.6, (i)) and (17.5.1) and (17.7.1),

it suffices to consider the case where is the spectrum of an algebraically closed field. But then it amounts to the same to say that an -prescheme locally of finite type is smooth over at a point or that it is regular at this point (17.15.1), so the conclusions follow from (19.4.4).

When is locally of finite presentation and the canonical immersion is quasi-regular, this immersion is by definition of finite presentation since the Ideal which defines is of finite type, so the composite morphism is locally of finite presentation. One has already seen (19.4.3) that under these conditions the morphism is smooth; moreover, one then knows that is a locally free -Module and that is isomorphic to , hence is a flat -Module. The hypotheses of (19.4.7) are therefore all verified and one may apply the conclusions of the statement of (19.4.8).

Corollary (19.4.9).

Under the general hypotheses of (19.4.8), suppose moreover that is smooth over at the point . Then, for every generization of in not belonging to , is smooth over at the point . If is smooth over , then there exists an open neighbourhood of in such that is smooth over at the points of .

For the first assertion, one may restrict to the case where is a local scheme of closed point , and the hypothesis then entails that is smooth over , every neighbourhood of the closed point of in being necessarily itself; one therefore concludes by the same reasoning as in (19.4.5), using the fact that the set of points where is smooth over is open in . To prove the second assertion, one may reduce to the case where is Noetherian, and of finite type over , thanks to (8.9.1), (8.6.3), (11.2.9) and (17.7.6); one then concludes as in (19.4.5), using the fact that the set of points where is smooth over is open in .

Remark (19.4.10).

In (19.4.6, (ii)), replace the hypothesis that and are locally of finite presentation over by the hypothesis that and (hence also ) are locally Noetherian. Then it is clear that is an -Algebra of finite type, that the morphism is of finite type, and it still follows from (19.2.4) applied to locally Noetherian preschemes that the immersion is transversally regular relatively to and that is -flat at the points of .

Proposition (19.4.11).

Let be a prescheme, a quasi-coherent -Module, a homomorphism of -Modules, the quasi-coherent Ideal of image of , the closed sub-prescheme of defined by . Let be the -scheme obtained by blowing up . On the other hand, let be the projective bundle on defined by (II, 4.1.1), the structure morphism, the canonical homomorphism (II, 4.1.5.1) and its kernel, so that one has the exact sequence

  0 → ℋ → p*(ℰ) →^{α♯} 𝒪_P(1) → 0.

Finally, let be the quasi-coherent Ideal of image of the restriction of , and let be the closed sub-prescheme of defined by .

(i) The image of by the closed immersion corresponding to the surjective homomorphism of graded -Algebras (II, 3.6.2) is a closed sub-prescheme of .

(ii) If is a locally free -Module of rank , is a locally free -Module of rank .

(iii) Suppose that is locally Noetherian, that is locally free of finite type, that the sub-prescheme is regular, and that at every point , the canonical immersion is regular and of codimension equal to (which we shall express by saying that the homomorphism is regular at the point (cf. chap. V)). Then the image of by the closed immersion is the closed sub-prescheme of ; at all points of (with the inverse image of in ) the immersion is transversally regular relatively to ; finally at all points of , is regular, the immersion is regular and of codimension where (in other words the homomorphism is regular at ).

All questions being local on , one may restrict to the case where is affine and , where is an -module. To do the verification of (i), one may moreover restrict to examining what happens in an affine open of of the form , where (II, 2.3.14). If one sets , one then has (II, 2.4.1), and . If one refers to the definition of (II, 2.6.2.2), one sees that to every element of (where and the are in ) makes correspond the element of . It is a matter (to prove (i)) of showing that if this latter element is zero, then so is the image of by the canonical homomorphism S_(t) → (𝓘 ⊗_𝒪_X (𝒪_X/𝓘))_t (II, 3.6.2). Now, this image is none other than in the ring , that is, the product of the element and the canonical image of by the algebra homomorphism, canonical extension to of the homomorphism of -modules. This therefore proves (i).

To establish (ii), let us note that one may suppose that is a free -module of rank ; since is surjective and is a free -module of rank 1, the exact sequence

  0 → H_(t) → (E ⊗_A S(1))_(t) → (S(1))_(t) → 0

is split, and is therefore a projective -module of rank , whence (ii).

To prove (iii), let us first examine what happens above a point . Then restricted to a neighbourhood of , is surjective, so one may restrict to the case where , in which case is canonically identified with . To see then that the closed sub-prescheme of image of is identified with , one may restrict as at the beginning to the case where is affine, and moreover , so that . One may moreover, with the notations of the beginning, suppose (the values of

verifying this condition generate the symmetric algebra ); by changing if necessary the basis in , one may therefore suppose that , and for , so that is canonically identified with . The elements of are then the elements of the form of where the must be such that in . One verifies easily that when the condition is fulfilled, the values of the may be taken arbitrarily for , and one may always then determine the for so as to verify the condition . The elements of the ideal corresponding are then the polynomials without constant term, and since is identified with , these polynomials are exactly the elements of the kernel of the homomorphism . One has thus proved (19.2.1) that the immersion is transversally regular relatively to at all points of and that at these points.

It remains to examine what happens above a point . The hypothesis of "regularity" of at this point made in (iii) is equivalent (by virtue of (19.1.1) and (0, 17.1.7)) to saying that is regular at the point and that the homomorphism is injective. This being so, as the morphism is smooth (17.3.9), for every above , is regular at the point (17.5.8); furthermore (0, 17.3.3) the canonical homomorphism

  (19.4.11.1)    (𝔪_x/𝔪_x²) ⊗_{k(x)} k(z) → 𝔪_z/𝔪_z²

is injective. One deduces that the homomorphism

  (ℰ_x ⊗_{𝒪_{X, x}} 𝒪_{P, z}) ⊗_{𝒪_{P, z}} k(z) → 𝔪_z/𝔪_z²

deduced from is also injective, for it is written as the composite

  (ℰ_x ⊗_{𝒪_{X, x}} k(z)) → (𝔪_x/𝔪_x²) ⊗_{k(x)} k(z) → 𝔪_z/𝔪_z²

where the first arrow is injective by flatness and the second is the injective homomorphism (19.4.11.1). As is (in a neighbourhood of in ) a direct factor of , the homomorphism is also injective, which establishes the "regularity" of the homomorphism at the point . It remains to see that the image by of is identical to the closed sub-prescheme . It will suffice for this to show on the one hand that the image by of the points of (where is the structure morphism) are exactly those of and on the other hand that at each of these points , the surjective homomorphism is bijective. But, taking into account the hypothesis and (19.4.3), the restriction of is smooth, and since is supposed regular, the same holds for (17.5.8); furthermore, the immersion is regular (19.4.2), so is also a regular ring (19.1.1). To prove that the homomorphism is bijective, it suffices then to show that the dimensions

of these two regular rings are equal (0, 17.1.9). But by virtue of (0, 17.3.3), this is also equivalent to the equality of the dimensions of the two rings and , so that finally one may replace and by the fibres and respectively, and reduce to proving that the morphism deduced from is an isomorphism. Now, one may restrict here to the case where with and , whence ; furthermore, the hypothesis means that there exists a regular system of parameters of such that for (0, 17.1.7); let be the image of ; if is the maximal ideal of , one has and consequently (16.9.4.1), is identified with , whence it follows that the immersion is here an isomorphism; this implies a fortiori that the closed sub-prescheme is identical to the fibre , and terminates the proof of (19.4.11).

19.5. -regularity criteria

We resume here, in completed form, the criteria for a sequence of elements of a ring to be -regular or -quasi-regular ( being an -module), already studied in (0, 15.1).

Theorem (19.5.1).

Let be a ring, a sequence of elements of , an -module. Consider the following properties:

a) The sequence is -regular.

b) One has for every (III, 1.1.3).

b') One has .

c) The sequence is -quasi-regular.

Then one has the implications

  a) ⟹ b) ⟹ b')    and    a) ⟹ c).

Set . If the modules are separated for the -preadic topology (resp. if every quotient module of a sub-module of is separated for the -preadic topology), then one also has the implication c) ⟹ a) (resp. b') ⟹ a)).

The implications a) ⟹ c), and c) ⟹ a) when the are separated for the -preadic topology, were already proved (0, 15.1.9), and are given only for the record. One has also shown (III, 1.1.4 and 1.1.3.3) that a) entails b), and b) trivially implies b'). It therefore remains to see that when every quotient module of a sub-module of is separated for the -preadic topology, b') entails a). Let us reason by induction on . For , the assertion follows trivially from the definitions, since is none other than the kernel of the homothety in (III, 1.1.1 and 1.1.2). Suppose then , and set ; with the notations of (III, 1.1.2), one has . One therefore has (III, 1.1.4.1) the exact sequence

  (19.5.1.1)    0 → H_0(f_n, H_1(f', M)) → H_1(f, M) → H_1(f_n, H_0(f', M)) → 0.

The relation therefore implies and . The first of these two relations means that one has (III, 1.1.3.5). Now, by definition (III, 1.1.1), is isomorphic to a quotient of a sub-module of ; taking on the filtration of the for , on the induced filtration, and on the quotient filtration of that of , one obtains on a finite filtration whose quotients are isomorphic to quotients of sub-modules of , hence are separated for the -preadic topology by hypothesis. As the relation implies a fortiori , one deduces from the hypothesis and the preceding remarks that . The induction hypothesis then proves that the sequence is -regular. On the other hand, the relation means that is -regular, so the sequence is -regular.

Corollary (19.5.2).

Suppose that is a Noetherian ring, that the belong to the radical of , and that is an -module of finite type. Then the four conditions a), b), b'), c) of (19.5.1) are equivalent.

One knows indeed that every -module of finite type is separated for the -preadic topology.

Proposition (19.5.3).

Let

be an exact sequence of -modules, an -regular sequence, . Consider the following properties:

a) The sequence is M''-regular.

b) The -preadic filtration of is induced on by the -preadic filtration of (in other words, the canonical homomorphism is injective).

c) The canonical homomorphism is injective.

Then one has the implications a) ⟹ b) ⟹ c), and a) entails moreover that the sequence is -regular.

If every quotient of a sub-module of M'' is separated for the -preadic topology, the conditions a), b) and c) are equivalent.

It is clear that c) is a consequence of b). Let us prove that under the hypotheses of the last assertion, c) entails a): since is -regular, one has, by (19.5.1), , whence an exact sequence, portion of the exact homology sequence

  0 → H_1(f, M'') → H_0(f, M') → H_0(f, M).

Now, condition c) expresses that the homomorphism is injective (III, 1.1.3.5); it therefore amounts to , whence, by virtue of the separation hypothesis and (19.5.1), the fact that is M''-regular.

Let us next show that a) entails c) and the fact that is -regular, by induction on . For , being -regular is also -regular and property c) is none other than lemma (3.4.1.4). For , the induction hypothesis shows that

if one sets , the sequence is -regular and one has, by virtue of c) applied to , an exact sequence

  0 → M'/(∑_{i=1}^{n−1} f_i M') → M/(∑_{i=1}^{n−1} f_i M) → M''/(∑_{i=1}^{n−1} f_i M'') → 0.

By hypothesis, is -regular and -regular, so the same reasoning shows on the one hand that is -regular, and on the other hand, by virtue of (3.4.1.4), that the sequence

  0 → M'/(∑_{i=1}^n f_i M') → M/(∑_{i=1}^n f_i M) → M''/(∑_{i=1}^n f_i M'') → 0

is exact, whence c).

Let us finally show that a) entails b). As the sequence is then -regular and -regular, hence -quasi-regular and -quasi-regular, one has canonical isomorphisms

  gr_𝔍^•(M') ≃ gr_𝔍^0(M')[T_1, …, T_n],    gr_𝔍^•(M) ≃ gr_𝔍^0(M)[T_1, …, T_n]

(0, 15.1.7), and as one has seen above that is injective, so is .

One will note again that the equivalence of conditions a), b), c) of (19.5.3) is valid in particular when is Noetherian, M'' an -module of finite type and the belong to the radical of .

(19.5.4)

In the sequel of this number, we keep the notations , , , of (19.5.1), and we suppose moreover that is endowed with a decreasing filtration formed of sub--modules, such that . Recall (0, 15.1.5) that one then defines on a second decreasing filtration formed of the sub-modules

  M'_k = M_k + 𝔍 M_{k−1} + ⋯ + 𝔍^{k−1} M_1 + 𝔍^k M_0

and that, if and are the graded -modules associated with the filtrations and respectively, one defines a surjective graded homomorphism of degree 0 (0, 15.1.5.2)

  ψ_M : (gr_•(M) ⊗_A (A/𝔍))[T_1, …, T_n] → gr'_•(M).

Recall also that one has the canonical surjective homomorphism (0, 15.1.1.1)

  φ_{gr_•(M)} : (gr_•(M) ⊗_A (A/𝔍))[T_1, …, T_n] → gr_𝔍^•(gr_•(M)).

Theorem (19.5.5).

With the notations of (19.5.4), consider the following properties:

a) The sequence is -regular.

b) The canonical homomorphism is bijective.

c) The canonical homomorphism is bijective (in other words (0, 15.1.7), the sequence is -quasi-regular).

One then has the implications1

  a) ⟹ b) ⟹ c).

Moreover, if, for every integer , the quotient modules of sub-modules of are separated for the -preadic topology (which is the case when is Noetherian, the are in the radical of and the are -modules of finite type), then the conditions a), b) and c) are equivalent.

The implication a) ⟹ b) was proved in (0, 15.1.8); the implication c) ⟹ a) under the separation hypothesis is a particular case of (0, 15.1.9). It remains to prove that b) implies c), which will be done in several steps.

We shall denote by the ideal generated by the ; for every sequence of integers , we shall set and .

(19.5.5.1)

For the map to be injective (and consequently bijective), it is necessary and sufficient that for and the following condition be verified:

(I_{q, p}) For every family of elements of , the relation

  ∑_{|q| = q} f^q x_q ∈ ∑_{i=1}^n 𝔍^{q−1} M_{p+1} + 𝔍^q M_{p+1}

implies for every such that .

One proceeds as in (0, 15.1.6) by considering the sub--module (resp. ) of terms of degree in the first (resp. second) member of (0, 15.1.5.2). One endows with the filtration

  (Q_k)_• = ⊕_{|q| ⩾ •} ((gr_{k−|q|}(M)) ⊗_A (A/𝔍)) T^q

and with the image filtration formed by ; it again suffices to prove that the homomorphisms are injective. Now, one has

  gr_•(Q_k)_i = ⊕_{|q| = i} ((M_{k−i}/M_{k−i+1}) ⊗_A (A/𝔍)) T^q = ⊕_{|q| = i} (M_{k−i}/(𝔍 M_{k−i} + M_{k−i+1})) T^q;

on the other hand, is the image of in . To write that is injective therefore amounts to writing the condition ; whence our assertion.

(19.5.5.2)

For the sequence to be -quasi-regular, it suffices that, for and , the following condition be verified:

(S_{q, p}) For every family of elements of , the relation implies for every such that .

By definition, to say that is -quasi-regular means that, for every and every , a relation

  ∑_{|q| = q} f^q x_q ∈ M_{p+q+1} + 𝔍 M_{p+q}

for a family of elements of entails for every such that . Now, the hypothesis on the sequence means that there exists a family of elements of such that one has . Condition then entails for every such that , whence ; one therefore sees that the conditions entail that the sequence is -quasi-regular.

It therefore suffices to prove the following proposition:

(19.5.5.3)

For every , if the conditions are verified for , then the conditions are verified for .

Let us introduce the conditions

(A_{q, p}) For every family of elements of , the relation implies .

It is clear that the conjunction of and of implies . On the other hand, the conditions and are trivial. Consequently, (19.5.5.3) will result from the following proposition:

(19.5.5.4)

Whatever and , the condition is entailed by the conjunction of the conditions , , , …, .

Indeed, once this proposition is proved, the conditions supposed verified for will entail, for each , the condition for , by induction on .

Let us therefore prove (19.5.5.4). Note that condition is equivalent to

  (19.5.5.5)    𝔍^q M_p ∩ M_{p+q+1} ⊂ ∑_{i=1}^n 𝔍^{q−1} M_{p+1}.

Let therefore . Applying the condition , one sees that there exists, for each such that , a family of elements of such that

  m = ∑_{i=1}^q (∑_{|q| = q − i} f^q y_q^{(i)}).

Suppose that, for , one has proved that

  y_q^{(i)} ∈ 𝔍 M_{p+i} + M_{p+i+1}    (for |q| = q − i).

One deduces by definition (19.5.4)

  ∑_{|q| = q − j} f^q y_q^{(j)} ∈ M'_{p+q+1}.

Since by hypothesis is verified, one deduces

  y_q^{(j)} ∈ 𝔍 M_{p+j} + M_{p+j+1}

and, by induction on , one sees therefore that this condition is verified for every such that . One therefore has

  m ∈ ∑_{i=1}^q 𝔍^{q−i}(𝔍 M_{p+i} + M_{p+i+1}) ⊂ ∑_{i=1}^{q+1} 𝔍^{q−i+1} M_{p+i}

and one has thus proved and terminated the proof of (19.5.5).

Remarks (19.5.6).

(i) The results of this number retain a meaning and are valid when, instead of an -module and elements , one takes an object of an arbitrary abelian category, and endomorphisms of which commute pairwise; the proofs adapt without difficulty to this more general case. The separation hypothesis for an -module relative to the -preadic topology must here be replaced by the hypothesis that every sub-object of contained in all the ( an arbitrary integer) is necessarily zero. The same remark applies to the results of 19.7.

(ii) Suppose that is a graded ring with degrees . If (resp. each ) is a graded -module with degrees bounded below and if the contain no homogeneous component of degree 0, the separation hypothesis of (0, 15.1.9) is ipso facto verified for (resp. each ), and one therefore sees that in this case one has the implication c) ⟹ a) in (19.5.1) (resp. (19.5.5)).

(iii) Recall that, without separation hypothesis, the implication c) ⟹ b) is no longer valid even for (0, 15.1.12, (iii)).

19.6. Regular sequences relative to a quotient filtered module

(19.6.1).

Let be a ring, a finite sequence of elements of , the ideal it generates, and consider an exact sequence of -modules

where one supposes that is endowed with a decreasing filtration formed of sub--modules, with . One sets (filtration induced by ), (quotient filtration of ), and one denotes by , and the graded -modules associated with these three filtrations. One sets moreover, for every ,

                       N'_k = N_k + 𝔍 N_{k-1} + ⋯ + 𝔍^{k-1} N_1 + 𝔍^k N_0

so that ; one sets moreover (filtration induced by ), and one denotes by , and the graded -modules associated with these three filtrations; one thus has a commutative diagram of exact sequences

  (19.6.1.1)         0 ──→ gr_•(R) ──→ gr_•(N) ──→ gr_•(M) ──→ 0
                              │            │            │
                              ▼            ▼            ▼
                     0 ──→ gr'_•(R) ──→ gr'_•(N) ──→ gr'_•(M) ──→ 0

where the vertical arrows are the canonical homomorphisms from a graded module associated with a filtration into the graded module associated with a coarser filtration.

One will note that, if one sets

                       R''_k = R_k + 𝔍 R_{k-1} + ⋯ + 𝔍^{k-1} R_1 + 𝔍^k R_0

one evidently has , but the filtrations and are in general distinct; one will denote by the graded -module associated with the filtration .

(19.6.2).

One has defined in (0, 15.1.5.2) the graded surjective homomorphisms of degree 0

                       ψ_N : (gr_•(N) ⊗_A (A/𝔍))[T_1, …, T_n] → gr'_•(N)
                       ψ_M : (gr_•(M) ⊗_A (A/𝔍))[T_1, …, T_n] → gr'_•(M)
                       ψ_R : (gr_•(R) ⊗_A (A/𝔍))[T_1, …, T_n] → gr''_•(R)

of which the first two are deduced from the last two vertical arrows of (19.6.1.1).

As the filtration is finer than on , one has a canonical homomorphism ; we shall denote by the composite homomorphism . It then follows at once from the definitions that one has a commutative diagram

  (19.6.2.1)         (gr_•(R) ⊗_A (A/𝔍))[T_1, …, T_n]  ──ψ_R──→  gr''_•(R)
                                  │                                  │
                                  │ j'                               │ gr''_•(j)
                                  ▼                                  ▼
                     (gr_•(N) ⊗_A (A/𝔍))[T_1, …, T_n]  ──ψ_N──→  gr'_•(N)
                                  │                                  │
                                  │ p'                               │ gr'_•(p)
                                  ▼                                  ▼
                     (gr_•(M) ⊗_A (A/𝔍))[T_1, …, T_n]  ──ψ_M──→  gr'_•(M)
                                  │                                  │
                                  ▼                                  ▼
                                  0                                  0

where the columns are exact.

Proposition (19.6.3).

With the preceding notation, suppose that the sequence is -regular. Consider the following properties:

  • a) is -regular.
  • b) is bijective.
  • c) is surjective (in other words, is generated as a -module by ).
  • d) is injective.
  • d′) The homomorphism obtained by restricting to polynomials of degree 0 is injective.
  • e) is bijective.
  • f) is injective.
  • f′) The homomorphism obtained by restricting to polynomials of degree 0 is injective.
  • g) The -preadic filtration of is induced by that of .
  • h) The filtrations and on are identical.

One then has the following implications:

                              a) ⇒ e) ⇒ b) ⇔ c) ⇔ h)
                                    ⇕     ⇕
                                       g) ⇒ d) ⇔ d′) ⇔ f) ⇔ f′)

Moreover, when every quotient module of a submodule of a is separated for the -preadic topology (which will be the case when is Noetherian, the belong to the radical of , and the are -modules of finite type), then the conditions a) through h) are all equivalent.

Note that by (19.5.5), from the hypothesis that is -regular, it follows that is bijective; as is injective, the equivalence of d) and f) follows from the diagram (19.6.2.1), as does that of d′) and f′); moreover, f) and f′) are trivially equivalent, which shows the equivalence of d), d′), f) and f′).

Since we already know that is surjective and bijective, the equivalence of b) and c) also follows from the diagram (19.6.2.1) by a particular case of the five lemma.

The relation (19.6.2) and the fact that is surjective show that condition c) is equivalent to saying that is surjective, which is equivalent in turn to condition h); one has thus proved the equivalence of b), c) and h).

It is trivial that e) is equivalent to the conjunction of c) and d), hence also of b) and d). Note now that a) implies b), and that b) implies a) under the separation hypothesis (19.5.5). On the other hand, the implications a) ⇒ g) ⇒ f), and the implication f) ⇒ a) under the separation hypothesis, follow from (19.5.3). One has thus proved that a) entails e) (equivalent to the conjunction of b) and f′)), and also that all the conditions are equivalent under the separation hypothesis. Q.E.D.

Corollary (19.6.4).

Let be an ideal of , . Suppose that the filtration (hence also ) is the -preadic filtration, so that the filtrations and are the -preadic filtrations. Consider (resp. ) as a graded module over (resp. ).

  • (i) Let be a part of which generates as a -module. Then condition c) of (19.6.3) is equivalent to the following condition:
  • c′) The image of in by generates as a -module.
  • (ii) Suppose that condition e) of (19.6.3) is satisfied, and moreover that the quotients of the are separated for the -preadic topology (which will be the case when is Noetherian, the are in the radical of , and the are -modules of finite type). Then, in order for a part of formed of homogeneous elements to generate as a -module, it is necessary and sufficient that its image by generate as a -module.

(i) Consider the homomorphism of degree 0

                       ψ_A : (gr_•(A) ⊗_A (A/𝔍))[T_1, …, T_n] → gr'_•(A) = gr_•^𝔏(A)

(0, 15.1.5.2). The fact that this homomorphism is surjective entails that the sub--algebra of generated by is none other than , whence the equivalence of c) and c′).

(ii) Since e) is satisfied, the same is true of c), and by virtue of (i), it remains to prove the sufficiency of the condition in the statement. Now, as is bijective, saying that generates regarded as a -module is equivalent to saying that generates as a -module, and by virtue of the surjectivity of , one may, in this statement, replace the ring by . It evidently amounts to the same to say that generates as a module over the ring , or equivalently as a -module. Now, if is the graded sub--module of generated by , and if one sets , one has by hypothesis , that is ; but is a graded module, and each is a quotient of , hence separated for the -preadic topology; the hypothesis therefore entails for every integer , hence , which completes the proof of the corollary.

19.7. Hironaka's normal flatness criterion

Theorem (19.7.1) (Hironaka).

Let be a ring, an ideal of , a sequence of elements of which is -regular, the ideal generated by , , an -module.

One has in this case , because , the last equality following from Bourbaki, Alg., chap. II, 3rd ed., §3, n° 7, cor. 2 of prop. 6.

Consider the following conditions:

  • a) is a flat -module.
  • b) is a flat -module, and the canonical homomorphism (0, 15.1.5.2)
                       ψ_M : ((gr_𝔍^•(M)) ⊗_A (A/𝔏))[T_1, …, T_n] → gr_𝔏^•(M)

is bijective.

  • c) is a flat -module, and the sequence is -regular.
  • d) is a flat -module, and for every , is a flat -module.

One then has the implications

                              a) ⇒ c) ⇒ b)
                                       ⇑
                                       d)

When every quotient of a submodule of a is ideally separated for (Bourbaki, Alg. comm., chap. III, §5, n° 1), the conditions a), b) and c) are equivalent.

Finally, when is Noetherian, the belong to the radical of , the sequence is -regular, and is an -module of finite type, the conditions a), b), c), d) are equivalent.

The implication a) ⇒ c) is immediate (0, 15.1.13). If c) is satisfied, it follows from (19.5.5) that is bijective; moreover, is a flat -module, hence so is , and consequently so is since is an -isomorphism; thus c) entails b). The fact that a) implies b) shows at once that a) also implies d).

When every quotient of a is ideally separated for , it follows from (19.5.5) that condition b) entails that is -regular; moreover, since , isomorphic to , is then a flat -module, the same is true of , which is a direct factor of it; this proves therefore that b) then entails c). On the other hand, under the same hypothesis, c) entails a) by virtue of (0, 15.1.21) (where one may replace the Noetherian hypothesis by the hypothesis that is ideally separated for , by virtue of (III, 10.2.1)).

It remains therefore to prove that when is Noetherian, of finite type and the belong to the radical of , d) entails a).

By hypothesis there exists a free -module of finite type and an exact sequence . It suffices to prove that d) entails b), since every quotient of a submodule of a is then ideally separated for , by virtue of the fact that is an -module of finite type, that is contained in the radical of , and that is Noetherian (Bourbaki, Alg. comm., chap. III, §5, n° 1); in other words, the question is to see that is bijective.

As by hypothesis the sequence is -regular, it is also -regular; one may therefore apply (19.6.3), for every quotient module of a submodule of a is then separated for the -preadic topology, since the belong to the radical of by hypothesis. The commutative diagram (19.6.2.1) is therefore written here

                                                          0
                                                          │
                                                          ▼
  (19.7.1.1)   (gr'_•(R, N) ⊗_{A/𝔎} (A/𝔏))[T_1, …, T_n] ──ψ_R──→ gr''_•(R, N)
                              │                                       │
                              ▼                                       ▼
               (gr'_•(N) ⊗_{A/𝔎} (A/𝔏))[T_1, …, T_n]  ──ψ_N──→  gr'_•(N)
                              │                                       │
                              ▼                                       ▼
               (gr'_•(M) ⊗_{A/𝔎} (A/𝔏))[T_1, …, T_n]  ──ψ_M──→  gr'_•(M)
                              │                                       │
                              ▼                                       ▼
                              0                                       0

where (resp. ) is the graded module associated with for the filtration of the (resp. ); recall that in this diagram the columns are exact and that

is bijective. The question is, by virtue of (19.6.3), to prove that is surjective, which will be done in several steps.

(19.7.1.2).

Set , , ; since is bijective, is identified with , and with a sub--module of ; we shall first see that one has

  (19.7.1.3)                  Q = (Q ∩ P)[T_1, …, T_n].

For this, set ; this is a sub--module of . As an -module, it is therefore isomorphic to a submodule of a direct sum of -modules isomorphic to . But is an -module isomorphic to a submodule of , which is none other than by virtue of the diagram (19.7.1.1). One has therefore

But each of the is, by hypothesis, a flat -module of finite type, hence projective since is Noetherian, and one has consequently . To see that , it suffices therefore (Bourbaki, Alg. comm., chap. IV, §1, n° 1, cor. 1 of prop. 2 and n° 3, prop. 7) to see that for every , one has . But the ideals of are the ideals of the form , where , and it amounts therefore to the same to see that for every . Now, if , where is a prime ideal of , one has and , and the images of the in form a -regular sequence (0, 15.1.14); applying to , and the implication a) ⇒ b) of the statement, one sees that the hypothesis d) entails that is bijective, hence also by virtue of (19.6.3); in other words, , where one has set , which is here a sub--module of ; in particular , hence finally one has indeed , which completes the proof of (19.7.1.3).

(19.7.1.5).

By virtue of (19.7.1.3), to prove that is surjective, it suffices to show that every homogeneous element of degree of is the image of an element of the same degree of . Now, the elements of degree of are identified (by ) with the elements of belonging to ; those which are images of elements of are identified with the elements of belonging to . It will therefore suffice in the end to prove, for every integer , the inclusion

                  ((R ∩ 𝔏^m N) + 𝔏^{m+1} N) ∩ (𝔎^m N + 𝔏^{m+1} N) ⊂ (R ∩ 𝔎^m N) + 𝔏^{m+1} N.

As , one verifies at once that the first member of this relation is equal to , so that everything reduces to proving

  (19.7.1.6)              R ∩ (𝔎^m N + 𝔏^{m+1} N) ⊂ (R ∩ 𝔎^m N) + 𝔏^{m+1} N.

We shall proceed by induction on , supposing therefore (19.7.1.6) satisfied when one replaces by an integer . We shall consider on the other hand, for fixed and an integer , the relation

  (*_d)                  R ∩ (𝔎^m N + 𝔏^{m+1} N) ⊂ (R ∩ 𝔎^m N ∩ 𝔏^d N) + 𝔏^{m+1} N.

It is clear that it suffices to prove for , and on the other hand is trivially true. We shall prove by induction on ; in other words, we suppose, for a fixed (with ), that is true, and we want to prove

  (*_{d+1})              R ∩ (𝔎^m N + 𝔏^{m+1} N) ⊂ (R ∩ 𝔎^m N ∩ 𝔏^{d+1} N) + 𝔏^{m+1} N.

Consider for this, for an , the relation

  (**_{d+1, h})  R ∩ (𝔎^m N + 𝔏^{m+1} N) ⊂ (R ∩ (𝔏^{d+1} N + 𝔍^h 𝔎^d N) ∩ 𝔎^m N) + 𝔏^{m+1} N.

The hypothesis implies that is true. We shall prove by induction on that is true for every . But one has the

Lemma (19.7.1.7).

Let be a Noetherian ring, an -module of finite type, , two submodules of , an ideal of . For every there exists such that

                       E ∩ (F + 𝔏^h N) ⊂ E ∩ F + 𝔏^k N.

Indeed, let be the canonical homomorphism, and set , , so that and . One knows that there exists such that ; if is so chosen and if is such that there exists for which , one deduces from the foregoing that there exist and such that and ; hence , which proves the lemma.

It suffices then to apply this lemma with , and to take to deduce that, with the corresponding value of , entails .

We suppose therefore in all that follows that is satisfied.

(19.7.1.8). Proof of : First case: .

Start therefore from an element

  (19.7.1.9)              g ∈ R ∩ (𝔏^{d+1} N + 𝔍^h 𝔎^d N) ∩ 𝔎^m N

congruent mod. to an element ; one has therefore

  (19.7.1.10)             g ∈ 𝔎^m N + 𝔏^{m+1} N.

It will suffice to prove that is also congruent mod. to an element , for that will entail , since one will have and . In the case which we are considering, it suffices to demonstrate the relation

  (19.7.1.11)            (𝔏^{d+1} N + 𝔍^h 𝔎^d N) ∩ 𝔎^m N ⊂ 𝔍^{m-d+1} 𝔎^d N + 𝔏^{m+1} N

for the relations (19.7.1.9) and (19.7.1.10) entail that belongs to the first member of (19.7.1.11), and as , one has , which will establish . Moreover one has , and the relation

  (19.7.1.12)            𝔍^a N ∩ 𝔏^b N ⊂ 𝔏^{a+b-d} N

will entail (19.7.1.11). As is a free -module of finite type, one sees that it therefore suffices to prove

  (19.7.1.13)            𝔏^{m+1} ∩ 𝔎^d ⊂ 𝔍^{m-d+1} 𝔎^d.

It will suffice to prove that, in general, for , one has

  (19.7.1.14)            𝔏^{m+1} ∩ 𝔎^d ⊂ 𝔍^{m-d+1} 𝔎^d + 𝔎^{d+1}.

Indeed, as , hence , the preceding relation will entail

                       𝔏^{m+1} ∩ 𝔎^{d+1} ⊂ 𝔍^{m-d} 𝔎^{d+1} + 𝔎^{d+2}

and by induction on , one will conclude

                       𝔏^{m+1} ∩ 𝔎^d ⊂ 𝔍^{m-d+1} 𝔎^d + 𝔎^{m+1}

whence (19.7.1.13), since .

As , the relation (19.7.1.14) is also written

  (19.7.1.15)        (𝔍^{m+1} + 𝔍^m 𝔎 + ⋯ + 𝔍 𝔎^m + 𝔎^{m+1}) ∩ 𝔎^d ⊂
                              𝔍^{m-d+1} 𝔎^d + 𝔍^{m-d} 𝔎^{d+1} + ⋯ + 𝔍 𝔎^m + 𝔎^{m+1} = 𝔍^{m-d+1} 𝔎^d + 𝔎^{d+1}.

We shall see that this inclusion is itself a consequence of

  (19.7.1.16)            𝔍^a 𝔎^b ∩ 𝔎^{a+1} ⊂ 𝔍^{a+1} 𝔎^{b-1}

valid for , . Indeed, to prove (19.7.1.15) it suffices to show that for , one has

  (19.7.1.17)        (𝔍^{m+1} + 𝔍^m 𝔎 + ⋯ + 𝔍^{m-q+1} 𝔎^{q-1}) ∩ 𝔎^d ⊂
                              𝔍^{m+1-q} 𝔎^{q-1} 𝔎^{d-q+1} + ⋯ + 𝔎^{m+1} ⊂ 𝔍^{m+1-q} 𝔎^d

for, taking , this will entail (19.7.1.15). Now, to prove (19.7.1.17), it suffices to proceed by induction on supposing the relation true for ; an element of the first member is written, by hypothesis, as with , ; as , one deduces that , since ; taking account of (19.7.1.16), one has , hence

                       y + z ∈ 𝔍^{m-q} 𝔎^{q+1} + ⋯ + 𝔍 𝔎^m + 𝔎^{m+1}, and y + z ∈ 𝔎^d

by hypothesis, which proves (19.7.1.17) where has been replaced by .

It remains then to establish (19.7.1.16). An element of the first member is written , where the belong to ; taking account of the fact that the sequence is -regular by hypothesis, hence also -quasi-regular (0, 15.1.9), it follows from the definition (0, 15.1.7) that the relation , where the , entails necessarily , hence ; it suffices to reason by

induction on for (the case being trivial) to complete the proof of (19.7.1.16) and consequently also of for .

(19.7.1.18). Proof of : Second case: .

We shall first demonstrate that, for every , one has

  (19.7.1.19)        R ∩ (𝔏^{d+1} N + 𝔍^h 𝔎^d N) ⊂ 𝔍^h (R ∩ (𝔎^{d+1} N + 𝔍 𝔎^d N)) + 𝔏^{d+h+1} N.

Consider for this an element of the first member of (19.7.1.19). Note that one has since ; we shall consider the class of in , which, by virtue of (19.5.5) and the hypothesis on , is identified with a sub--module of . From the fact that , we shall show that one has even (with the foregoing identification)

Indeed, one has noted, in the proof of (19.7.1.3), that for every prime ideal of such that , the map is bijective, and consequently the relation (19.7.1.6), where one replaces all the modules by their localizations at , is true. As belongs to , its image in belongs to , hence the image of in belongs to , where . In other words the image of by the canonical map is zero for every . As one has seen in the proof of (19.7.1.3) that , one deduces (Bourbaki, Alg. comm., chap. IV, §1, n° 1, cor. 1 of prop. 2 and n° 3, prop. 7) that the image of by the canonical map is zero, that is, one has the relation (19.7.1.20).

This being so, one may write by definition

  (19.7.1.21)            ẑ = Σ_{|p| = h} c_p f^p mod. 𝔏^{h+1} N

where one sets as usual , , , and where . Moreover, since identifies with , the are determined mod. , and if is the class of mod. , one has, by the preceding identification,

  (19.7.1.22)            ẑ = Σ c̄_p f^p.

Using (19.7.1.3), one deduces that one has necessarily, for every such that ( being this time identified by with an element of )

But since , the hypothesis that is true implies that belongs to the image of , hence one may suppose that . The relation (19.7.1.21) then gives

                       ẑ ∈ 𝔍^h (R ∩ 𝔎^d N) + 𝔍^h 𝔎^{d+1} N + 𝔏^{h+1} N

and since , this proves (19.7.1.19).

In particular, one may therefore write

with and ; moreover, as and , one has . On the other hand, using the hypothesis , one sees that . The element therefore answers to all the conditions stated at the beginning of (19.7.1.8), which completes the proof of (19.7.1).

Remarks (19.7.2).

(i) When one supposes that is a flat -module, the hypothesis that the sequence is -regular entails that it is also -regular (0, 15.1.14). One will note that this will be the case when and are regular rings (0, 17.3.6): in fact, is then a regular ideal of (19.1.2), hence quasi-regular, and by localizing at the maximal ideals of containing and using (0, 15.1.7), one sees that is a projective -module.

(ii) It would be interesting to be able to prove the last conclusion of (19.7.1) by supposing only that there exists an -algebra which is a Noetherian ring for which is contained in the radical of , and that is a -module of finite type.

Corollary (19.7.3).

Let be a Noetherian local ring, its maximal ideal, an ideal of such that the ring is regular and of dimension , an -module of finite type. The following conditions are equivalent:

  • a) is a flat -module.
  • b) If and are the Poincaré series of the -graded modules and , one has

Let be a sequence of elements of such that the images of the in form a regular system of parameters of (0, 17.1.6), so that is an -regular sequence; moreover, if is the ideal generated by , one has , and as is a field, is a flat -module. As is Noetherian and an -module of finite type, one may apply the equivalence of a) and b) in (19.7.1). Moreover, as is a field, and is surjective, the fact that is bijective is equivalent to saying that for every , and the submodule of elements of degree of have the same rank over . But for the second of these modules, the rank in question is evidently equal to

                       Σ_{i+j = h} binom(n + j − 1, j) rank_{A/𝔪}(gr_𝔎^i(M) ⊗_A (A/𝔪)).

As , this proves that the relation (19.7.3.1) is necessary and sufficient for to be bijective.

Corollary (19.7.4).

Let be a locally Noetherian prescheme, a regular and connected closed subprescheme of , a coherent -Module normally flat (6.10.1) along . Then there exists a formal series (independent of ) such that, for every , the Poincaré series of is given by the formula

  (19.7.4.1)             P_{ℱ(x)}(T) = R(T)(1 − T)^{-n}    where n = dim(𝒪_x).

Let be the coherent Ideal of defining ; the hypothesis on signifies that is a flat -Module. One may therefore, for every , apply (19.7.3) taking , , ; the hypothesis a) being fulfilled by hypothesis, the Poincaré series is therefore equal to , where is the Poincaré series of . Now, each of the -Modules is flat and coherent by hypothesis, hence locally free (2.1.12), and since is supposed connected, each is of constant rank on , which proves that the Poincaré series is independent of .

Corollary (19.7.5).

Let be a locally Noetherian prescheme, a closed subprescheme of , a closed subprescheme of ; one supposes and regular. Let be a coherent -Module.

(i) If is normally flat along at the points of (cf. (11.3.4) for the definition of normal flatness at a point), then is normally flat along .

(ii) Suppose moreover that is connected and that is normally flat along at the points of (which will hold in particular if is regular at the points of ). If is normally flat along and if there exists a point such that is normally flat along at the point , then is normally flat along at the points of .

Let and be the coherent Ideals of defining respectively and . For every point , the canonical immersion is regular at the point , since and are regular at that point (19.1.1); in other words, there exists a sequence of elements of which is -regular and is such that .

(i) The hypothesis signifies that is a flat -module; it therefore entails that is a flat -module, by virtue of the fact that in (19.7.1), a) entails b).

(ii) The supplementary hypothesis on entails that the sequence is -regular at every point (19.7.2). On the other hand, being regular and connected, is integral, and if is -flat at a point , then, as the generic point of is a generization of , one concludes that is -flat . One deduces that if at every point of , is a flat -module, then is a flat -module, for one may apply the equivalence of a) and d) in (19.7.1).

19.8. Properties of passage to projective limit

In this number, the notation and conventions on projective limits are those of (8.5.1) and (8.8.1).

Proposition (19.8.1).

Suppose that the transition morphisms () are flat, and moreover that one of the two following hypotheses is satisfied:

  • 1° The preschemes are locally Noetherian.
  • 2° The transition morphisms are surjective (hence faithfully flat).

Under these conditions:

(i) Suppose quasi-compact. Let be a quasi-coherent -Module, which is moreover supposed of finite type when hypothesis 1° is satisfied. Let be a finite sequence of sections of above . In order for the sequence of sections of the -Module above , corresponding to the (), to be -regular, it is necessary and sufficient that there exist such that the sequence of sections of above be -regular.

(ii) Suppose quasi-compact, and let be an immersion, which one supposes locally of finite presentation when hypothesis 2° is satisfied. Then, in order for the corresponding immersion to be regular, it is necessary and sufficient that there exist such that be regular.

(i) If is the canonical projection, one has , hence one is reduced to the case , in which case one suppresses the index . The fact that the condition is sufficient follows from the fact that is flat (8.3.8) and from (0, 15.1.5). In case 2°, is faithfully flat (8.3.8) and the necessity of the condition still follows from (0, 15.2.5), with . In case 1°, denote by (resp. ) the kernel of the homomorphism (resp. ), multiplication by (resp. by ); since is coherent by hypothesis, so is , hence the hypothesis entails for a by virtue of (8.5.8, (ii)).

(ii) Since is flat, the sufficiency of the condition follows from (19.1.5, (ii)). To prove its necessity, note that, since is quasi-compact, it is contained in a quasi-compact open of and one may therefore limit oneself to the case where is also quasi-compact and a closed immersion, so that the image of is defined by a quasi-coherent Ideal of , which one may moreover suppose of finite type in cases 1° and 2° ( being locally of finite presentation in case 2°); the image of (resp. ) in (resp. ) is then defined by (resp. ) which is still of finite type. One may moreover, taking account of (8.2.11), suppose that is generated by a regular sequence of sections above , which therefore define a surjective homomorphism . Taking account of (8.5.2, (i)) and (8.5.7), there exist and a surjective homomorphism such that , hence the are the canonical images of sections of above generating that Ideal. By virtue of (i), there exists therefore such that the sequence is -regular, hence the immersion is regular.

Proposition (19.8.2).

Let be a quasi-compact -prescheme, locally of finite presentation.

(i) Let be a quasi-coherent -Module of finite presentation, a sequence of sections of above . In order for the sequence of sections of above , corresponding to the , to be -transversally regular relative to (19.2.1), it is necessary and sufficient that there exist such that the sequence of sections of above be -transversally regular relative to .

*(ii) Let be a quasi-compact -prescheme, an -immersion locally of finite

presentation. In order for the corresponding -immersion to be transversally regular relative to , it is necessary and sufficient that there exist such that be transversally regular relative to .*

(i) The sufficiency of the condition follows from (0, 15.1.15). To prove that it is necessary, it will suffice to show that every point has an open neighbourhood for which there exist an index and an open neighbourhood of the image of in such that is the inverse image of and that the sequence of the restrictions to of the is -transversally regular relative to : indeed, it will then suffice to apply (8.3.4) to the open set of , defined by the condition of being the largest open such that the restrictions to of the form a sequence -transversally regular relative to . Let be the image of in , and for every , let be the image of in . Then the fibre is the projective limit of the fibres and is locally Noetherian since the morphism is locally of finite presentation; as moreover is quasi-compact and the transition morphisms are flat (since this is so of ), one may apply to the sections of above the result of (19.8.1, (i)); whence the existence of such that the sections of above form a -regular sequence. Moreover (11.2.6), one may suppose that is -flat at the point . It then follows from (11.3.8) that there exists an open neighbourhood of in satisfying the required conditions.

(ii) The sufficiency of the condition follows from (19.2.7, (ii)). To prove its necessity, it again suffices to show that every point admits an open neighbourhood for which there exist an index and an open neighbourhood of the image of in such that is the inverse image of and that the restriction of is transversally regular relative to . Let be the image of in , and for every , let be the image of in . As (resp. ) is the projective limit of the fibres (resp. ), one may, as in (i), use (19.8.1, (ii)), and as by hypothesis the immersion is regular, there exists an index such that for the immersion be regular. Moreover, by virtue of (19.2.4) there exists a quasi-compact open neighbourhood of in such that the structure morphisms and are flat; applying (11.2.6) and (8.2.11), one may suppose that there exist and a quasi-compact open neighbourhood of in , such that is the inverse image of and that the restrictions to and to respectively of the structure morphisms and be flat. One concludes therefore from (19.2.4) that there exists an open neighbourhood of answering the question.

Remark (19.8.3).

The preceding proofs show that the statements of (19.8.1) and (19.8.2) subsist when one replaces the conditions concerning regularity or transversal regularity in all of (resp. ) by these conditions in a neighbourhood of a point (resp. in a neighbourhood of , projection of ).

19.9. -regular sequences and depth

(19.9.1).

Recall that if is a locally Noetherian prescheme, a coherent -Module, a part of , one sets (5.10.1)

  (19.9.1.1)             prof_T(ℱ) = inf_{t ∈ T} prof(ℱ_t).

One sets moreover, for every point ,

  (19.9.1.2)             prof_{T, t}(ℱ) = inf_{z ∈ T ∩ Spec(𝒪_{X, t})} prof(ℱ_z),

and one says that is the -depth of at the point ; it is clear that one has

  (19.9.1.3)             prof_T(ℱ) = inf_{t ∈ T} prof_{T, t}(ℱ).

Lemma (19.9.2).

Let be a Noetherian ring, an -module of finite type, an ideal of . In order that, for every prime ideal , one have , it is necessary and sufficient that there exist an -regular sequence of elements belonging to .

The sufficiency of the condition follows at once from (0, 16.4.5); let us prove its necessity.

We reason by induction on (there is nothing to prove if ): since for every , there exists by hypothesis an -regular sequence formed of elements of . Set ; for every , the images of the in therefore belong to the maximal ideal , and form an -regular sequence (0, 15.1.14); one concludes therefore from (0, 16.4.6) that one has , and one sees that everything reduces to proving the lemma for . Now the hypothesis then signifies that, for every , is not associated to (0, 16.4.6), hence (Bourbaki, Alg. comm., chap. IV, §1, n° 2, prop. 5), is not associated to . In other words, is not contained in any of the prime ideals of , hence it is not contained in their union (Bourbaki, Alg. comm., chap. II, §1, n° 1, prop. 2). But as this union is the set of elements which are not -regular (Bourbaki, Alg. comm., chap. IV, §1, n° 1, cor. 2 of prop. 2), this proves the lemma.

Proposition (19.9.3).

Let be a locally Noetherian prescheme, a coherent -Module, a closed part of , a point of , an integer > 0. The following conditions are equivalent:

  • a) There exists an open neighbourhood of in such that .
  • b) There exists an open neighbourhood of in and a sequence of sections of above , which is -regular, such that for at every point .
  • c) One has .

Moreover, the set of satisfying these conditions is open in .

The last assertion follows trivially from a). The equivalence of a) and b) follows from (19.9.2) applied to an affine open neighbourhood of in and to (0, 15.1.14). It is clear that a) entails c), every open neighbourhood of in containing ; let us show conversely that c) implies b). It follows from c) and from the definition (19.9.1.1) that one may

apply (19.9.2) to the ring , to the -module and to the ideal of , hence there exists an -regular sequence of elements of . There exists therefore an open neighbourhood of in and a sequence of sections of above such that the are the germs of the at ; using (0, 15.2.4), one deduces that there exists an open neighbourhood of in such that the form an -regular sequence.

Corollary (19.9.4).

With the notation of (19.9.3), the function is lower semi-continuous in .

Proposition (19.9.5).

With the notation of (19.9.3), let be a locally Noetherian prescheme, a flat morphism, , . Then, for every point above , one has .

Indeed, for every , it follows from that one has , and if z'' is a maximal point of the fibre , generization of (hence belonging also to ), one has . The proposition follows from the fact that one has since is flat (2.3.4).

Proposition (19.9.6).

Let be a morphism locally of finite presentation, a closed part of , an -Module of finite presentation and -flat, an integer > 0. For every point , the following conditions are equivalent (denoting by the fibre of at the point , by the inverse image of in , and setting ):

  • a) One has .
  • b) There exist an open neighbourhood of in and a sequence of sections of above , which is transversally -regular relative to (cf. (19.2.1) for the terminology) and such that for and for every .

The set of satisfying these conditions is open in ; if moreover is locally constructible in , is retrocompact in .

By virtue of (19.9.3), condition a) is equivalent to the existence of an open neighbourhood of in and of an -regular sequence of sections of above such that for every . If one again denotes by a closed subprescheme of having as underlying space, by the quasi-coherent Ideal of defining , one may again say that one has for . As one may replace by a smaller affine neighbourhood of , one may suppose that the are the images of a sequence of sections of above , so that the germs belong to the maximal ideal of . The equivalence of a) and b) then follows from (11.3.8), which also proves that the set is open in . It remains to prove the last assertion; as it is again a question of local properties on , one may suppose affine and of finite presentation over ; there exists then a Noetherian subring A_0 of , a prescheme X_0 of finite type over and a coherent -Module such that and (8.9.1); one may moreover suppose that, if is the structure morphism, is -flat (11.2.6). Since is constructible, one may moreover (8.3.11) suppose

that , where is the canonical projection. Then, for every , it follows from (19.9.5) and the transitivity of fibres that one has

                       prof_{Y_{f(y)}, y}(ℱ_{f(y)}) = prof_{(Y_0)_{f_0(y_0)}, y_0}((ℱ_0)_{f_0(y_0)})

setting ; if is the set of for which the second member of this relation is , one has therefore . As X_0 is Noetherian, is a retrocompact open in X_0, and consequently (cf. proof of (1.8.2)) is retrocompact in .

Corollary (19.9.7).

Under the general hypotheses of (19.9.6), the function is lower semi-continuous in . If moreover is locally constructible, this function is locally constructible.

Proposition (19.9.8).

Let be a morphism locally of finite presentation, a closed part of , an -Module of finite presentation and -flat, a quasi-coherent -Module. Suppose that, for every , one has (resp. ). Then, if is the canonical injection, and if one sets , the canonical homomorphism relative to is injective (resp. bijective).

The question is evidently local on and , and it suffices to demonstrate the proposition in a neighbourhood of a point . In other words, one may restrict to the case where and are affine. Thanks to (19.9.6), one may suppose that there exists a transversally -regular section of above relative to (resp. a transversally -regular sequence of two sections of above relative to ) such that, if is the set of for which (resp. ), one has .

This being so, let us return to the proof of (19.9.8); by virtue of (19.9.6), one still has (resp. ) for every . If the proposition is proved replacing the pair by (X, Z') and , it is immediate that it will also be so for ; one may therefore restrict to proving it for and . In other words, one may restrict to demonstrating (19.9.8) when , , where is an -algebra of finite presentation, , where is an ideal of finite type of the ring , and , where is an -module and a -module of finite presentation which is a flat -module. One may moreover reduce to the case where is an -module of finite presentation. Indeed, is the inductive limit of a filtered inductive system of -modules of finite presentation (by double inductive limit); if one sets , is the inductive limit of the ; as and commute with inductive limits and lim is an exact functor in the category of quasi-coherent Modules, if the proposition is proved for each of the , it will be so for .

It suffices moreover to prove that the canonical homomorphism

                       Γ(X, ℋ) → Γ(X − Z, ℋ)

is injective (resp. bijective). There exists then a Noetherian subring A_0 of , an A_0-algebra of finite type B_0, an ideal of B_0 and a B_0-module of finite type M_0 which is

a flat A_0-module and an A_0-module N_0 such that , , and (8.9.1, 8.5.11 and 11.2.7). Moreover, let be the family of sub-A_0-algebras of finite type of , so that ; set , , , , , , ; if is the canonical projection, one has therefore , and . As is quasi-compact and quasi-separated, one concludes (8.5.2) that if one sets and , the homomorphisms and are bijective; by virtue of the exactness of the functor lim, it will therefore suffice to prove that for every sufficiently large , the canonical homomorphism is injective (resp. bijective). But for every , if is the projection of in , one has by virtue of (4.2.7) and (6.7.1); to be reduced to demonstrating (19.9.8) in the case where is Noetherian, one must prove that, for sufficiently large, one has (resp. ) for every point . Now, let be the set of points such that, for every generization of in , one has ; if is the canonical projection for , it follows from the hypothesis and from (19.9.5) that one has and ; on the other hand, it follows from (19.9.6) that is open in , hence our assertion follows from (8.3.4).

To finish the proof when is Noetherian, note that by virtue of the flatness hypothesis and of (6.3.1), one then has (resp. ); but the conclusion of (19.9.8) follows in this case from (5.10.2) (resp. (5.10.4)).

Remark (19.9.9).

The same proof, joined to the results of Chap. III, 3rd part on depth and local cohomology, allows one to establish the following generalization of (19.9.8):

Under the general conditions of (19.9.8), suppose that, for every , one has ; then the canonical homomorphism

                       H^i(X, ℋ) → H^i(X − Z, ℋ | (X − Z))

is bijective for and injective for (which, expressed by means of the general cohomological notion of depth introduced in Chap. III, is also written ).

1

The proof, without separation hypothesis on the , is due to P. Deligne, whom we thank.