§12. Study of the fibres of flat morphisms of finite presentation

12.0. Introduction

Throughout this section we shall use the general notations of (9.4.1).

(12.0.1)

Given a morphism locally of finite presentation, we saw (9.9) that for certain local properties of preschemes over fields or of Modules on such preschemes, the set of such that the property holds for the fibre at the point of that fibre is locally constructible in . We propose to show that, for most of these properties, if one supposes moreover that the morphism is flat, then the set is even open in . Likewise ((9.2) through (9.8)) we have shown that if is of finite presentation, and if this time denotes certain global properties of preschemes over fields or of Modules on such preschemes, then the set of such that the property holds for the fibre is locally constructible in . We shall show that if one supposes moreover that the morphism is proper and flat, then is even open in .

(12.0.2)

The general method of proof of the properties in question comprises three steps. One first reduces to the case where is affine and of finite presentation; then:

A) Using (8.9.1) (and possibly other results of §8) and (11.2.6), one reduces to the case where and are Noetherian.

B) One applies the results of §9 recalled in (12.0.1) proving that (resp. ) is constructible.

C) To see that is open, it suffices, by virtue of , to show that if , then every generization of also belongs to . Using (II, 7.1.7), one sees, since is Noetherian, that there exists a spectrum of a discrete valuation ring Y_1 and a morphism such that, if (resp. ) is the closed point (resp. the generic point) of Y_1, one has , . One then makes the base change ; taking into account the results of §§4 and 6 on locally Noetherian preschemes over fields and changes of base field, one is reduced to proving the assertion in question for a point of above and for a generization of above . Since , where is a discrete valuation ring with uniformizer , one must in the end, for an -module , prove a property of , knowing that has the same property and that is -regular (which follows from the flatness hypothesis); for this one uses the results of (3.4) and (5.12). One proceeds in the same way for the set .

12.1. Local properties of the fibres of a flat morphism locally of finite presentation

Theorem (12.1.1).

Let be a morphism locally of finite presentation, an -Module that is -flat and of finite presentation, a finite subset of , an integer. The following subsets of are open:

(i) The set of such that the dimensions of the associated prime cycles of containing are elements of .

(ii) The set of such that the associated prime cycles of containing all have dimension .

(iii) The set of such that belongs to no embedded associated prime cycle of .

(iv) The set of such that is equidimensional at the point and possesses property at the point (5.7.2).

(v) The set of such that (0, 16.4.9).

(vi) The set of such that is a Cohen-Macaulay -Module at the point (5.7.1).

(vii) The set of such that is geometrically reduced at the point (4.6.22).

(viii) The set of such that is geometrically pointwise integral at the point (4.6.22).

The questions being local on , one reduces first to the case where and are affine, with a morphism of finite presentation. There then exists a Noetherian sub-ring A_0 of , an A_0-prescheme of finite type X_0, and a coherent -Module such that is isomorphic to (8.9.1); in addition, by virtue of (11.2.6), one may suppose that is Y_0-flat (with ). If is the canonical projection, the set of points where one of properties (i) to (viii) holds is equal to , where E_0 is the set of where the corresponding property for holds: this follows, for properties (i) to (iii), from (4.2.7); for properties (iv) to (vi), from (6.7.1); for properties (vii) and (viii), from (4.7.11).

One is thus reduced to the case where and are Noetherian, of finite type, and , where is a -module of finite type. By virtue of (9.9.2), one knows that the set is constructible for all the properties considered, and there remains in each case step C) of (12.0.2), where one must prove that is stable under generization.

(12.1.1.1)

Let us begin with case (iii), which is the simplest. Let be a generization of in . Set , ; there exists a spectrum of a discrete valuation ring Y_1 and a morphism such that, if and are the closed point and the generic point of Y_1, one has and (II, 7.1.7); set ; if , there exists a Y_1-section such that , where is the canonical projection. Setting , , one therefore has and , and is a generization of . Applying (4.2.7) again, one sees

that one is reduced to the case where is the spectrum of a discrete valuation ring, being the closed point and the generic point of . If is a uniformizer of , the hypothesis that is -flat entails that is -regular . By hypothesis none of the embedded associated prime cycles of contains . Then, it follows from (3.4.4) that belongs to no embedded associated prime cycle of , and the same is therefore true of every generization of . In particular, belongs to no embedded associated prime cycle of , nor a fortiori to any of those associated to ( being a sub-prescheme induced on an open set of ).

(12.1.1.2)

Let us consider next cases (iv) and (v) ((vi) is only a special case of (v)). One proceeds as above (using (6.7.1) instead of (4.2.7)) and one is reduced to the case where is the spectrum of a discrete valuation ring .

The ring is then a localization of a finitely generated -algebra, hence catenary (5.6.4), and by hypothesis the -module satisfies and is equidimensional (resp. one has ); one therefore deduces from (5.12.2) (resp. from (0, 16.4.10)) that satisfies and is equidimensional (resp. that ), since is -regular and belongs to the maximal ideal of . Whence the conclusions, since is a generization of and .

(12.1.1.3)

Let us pass to cases (vii) and (viii). One may evidently replace by the integral sub-scheme having as underlying space, and by , without changing the fibres at and ; one may therefore suppose integral, with field of fractions . One knows ((4.5.11) and (4.7.8)) that there exists a finite extension of such that, for the -Module , the associated prime cycles are geometrically irreducible and the geometric lengths of at the maximal points of its support are respectively equal to the lengths of at these points; it therefore amounts to the same to say that is geometrically reduced (resp. geometrically pointwise integral) at a point of , or to say that is reduced (resp. integral) at the points of above , taking (4.2.7) into account. Let be a finite -algebra of which is the field of fractions; it follows from (4.7.11) that at every point of above , has property (vii) (resp. (viii)). Replacing by and by , one sees that one may confine oneself to proving that is reduced (resp. integral) at every generization of . One then proceeds as in (12.1.1.1), this time using (4.7.11), and one is once again reduced to the case where is a discrete valuation ring, being the closed point and the generic point of , with the uniformizer of being an -regular element. Since by hypothesis is reduced (resp. integral) at the point , it follows from (3.4.6) (resp. (3.4.5)) that is reduced (resp. integral) at the point , hence also in a neighbourhood of , and in particular at the point , which completes the proof in cases (vii) and (viii).

(12.1.1.4)

It remains to examine cases (i) and (ii). Replacing by the integral sub-scheme having as underlying space, one may confine oneself to the case where is irreducible and its generic point. Let be a generic point of an associated prime cycle of

containing , and let be the closure of in , so that one has . To treat case (i), it will suffice to prove:

Proposition (12.1.1.5).

Let be a locally Noetherian prescheme, a morphism locally of finite type, a coherent -Module that is -flat; for every , set . Let be a point of , , and suppose that . Let (resp. ) be the reduced sub-prescheme of (resp. ) having as underlying space the closure of {z'} in (resp. of {y'} in ). Then, for every , the dimensions of all the irreducible components of are equal to (the dimension of the closure of in ) (which we shall express later (13.2.2) by saying that the restriction of is an equidimensional morphism); moreover, at every maximal point of , one has .

For this, we shall reduce to the case where is the spectrum of a discrete valuation ring. Take a spectrum of a discrete valuation ring Y_1 and a morphism such that , , where and are the closed point and the generic point of Y_1 respectively (II, 7.1.7). Let , , and let , be the canonical projections; there is a Y_1-section such that (I, 3.3.14). If , one knows (4.2.6) that the irreducible components of dominate ; let be the generic point of one of these components which contains , so that and . Since is a specialization of , it is a fortiori a specialization of ; let be a generic point of one of the irreducible components of containing . Then is a specialization of in , and is a specialization of ; but since and is a generic point of , one necessarily has .

Suppose now that one has proved that, if one sets , one has ; it will follow from (4.2.7) that ; moreover, the dimensions of and of are equal, and likewise the dimensions of and of (4.2.7); one has therefore indeed reduced, as announced, to the case where is the spectrum of a discrete valuation ring , and being its closed point and its generic point respectively.

This being so, since , one has a fortiori . If is a uniformizer of , is -regular by flatness, hence (3.4.3) one has . As for the assertion concerning dimensions, it follows from (7.1.13), applied to a closed sub-prescheme of having as underlying space.

Let us tackle finally case (ii) of (12.1.1). With the same notation, one must prove that if all the associated prime cycles of containing have dimension , then the same is true of all the associated prime cycles of containing . Now we have just seen that every associated prime cycle of containing has the same dimension as one of the associated prime cycles of containing ; this therefore proves (ii).

Remarks (12.1.2).

(i) Under the conditions of (12.1.1.5) with (so that is flat), one cannot in general assert that the restriction of is a flat morphism, nor even an open morphism. This is what the example (6.5.5, (ii)) shows,

where one takes for one of the two irreducible components of . It is immediate that this restriction morphism is not open at the points of above the "double point" of .

(ii) In the hypotheses of (12.1.1.5), with , one cannot weaken the condition " is flat" to " is universally open" (cf. (2.4.6)), as we shall see later from an example (14.4.10, (i)).

Corollary (12.1.3).

Under the hypotheses of (12.1.1), the function is upper semi-continuous in and the function (10.8.1) is lower semi-continuous in .

The first assertion is none other than an equivalent formulation of (12.1.1, (v)). To prove the second, one may first, taking (10.8.7) and (10.8.8) into account, reduce as in (12.1.1) to the case where is the spectrum of a discrete valuation ring with closed point and generic point , with affine, , where is a -module of finite type. Since is -flat, every irreducible component of that contains dominates (2.3.4), in other words its generic point belongs to ; the irreducible components of that contain a generization of belonging to are therefore exactly those that contain , and moreover, by (12.1.1.5), one has , whence, if , . In addition, for a uniformizer of , one saw in (12.1.1, (v)) that one has , and since on the other hand (6.11.5), one sees that one has ; the relation then follows from (10.8.7).

Remark (12.1.4).

One also deduces from (12.1.1, (i)) that the function is upper semi-continuous in , but we shall see later (13.1.3) that this property is true even without supposing flat.

Corollary (12.1.5).

Let be a locally Noetherian prescheme, a morphism locally of finite type, a coherent -Module that is -flat, a coherent -Module, a point of , . Suppose that has property at the point , and that has property at the point and is equidimensional at the point . Then:

(i) possesses property at the point .

(ii) If moreover is locally immersible in a regular scheme, or is an excellent prescheme (7.8.5), there exists a neighbourhood of in such that has property in this neighbourhood.

Indeed, by (12.1.1, (iv)), there exists an open neighbourhood of such that for every , satisfies at the point . On the other hand, satisfies at every point of (5.7.2). Replacing by and by , one is reduced (taking (I, 3.6.5) into account) to the case where satisfies in entirely and where, for every , possesses property at every point of ; since is -flat, one then knows (6.4.1) that possesses property in , which proves (i). To prove (ii), it suffices to observe that, under the hypotheses made, possesses

property in a neighbourhood of in ((6.11.2) and (7.8.3, (iv))); one then concludes in the same way, replacing this time by and by .

Theorem (12.1.6).

Let be a flat morphism locally of finite presentation, an integer. The following subsets of are open:

(i) The set of points such that satisfies property at the point .

(ii) The set of such that satisfies geometric property at the point , is equidimensional at the point , and moreover belongs to no embedded associated prime cycle of .

(iii) The set of such that is geometrically regular (i.e. smooth) at the point .

(iv) The set of such that is geometrically normal at the point .

Steps A) and B) of (12.0.2) are carried out here as in (12.1.1); for step A), one uses (6.7.8), as well as (4.2.7) for (ii); for step B), one uses (9.9.2) and (9.9.4). It remains to examine step C) in each case.

(i) As in (12.1.1.2), one reduces (using (6.7.8)) to the case where is the spectrum of a discrete valuation ring , , where is an -algebra of finite type, , two points of such that is the closed point and the generic point of , with moreover a generization of . Since the task is to prove that satisfies at the point , one may replace by , that is, suppose the ring local, the homomorphism local, and essentially of finite type over (1.3.8). Then, by hypothesis, if is a uniformizer of , is a -regular element by flatness, and satisfies . But since is a universally catenary ring (5.6.4), is catenary, hence, by (5.12.4), it follows that satisfies , which completes the proof in case (i).

(iii) Here step C) is unnecessary; since is flat, one knows in effect (6.8.7) that when is locally Noetherian and locally of finite type, the set of such that is geometrically regular at the point is open in .

(ii) Reasoning as in (12.1.1.3), to prove that the property considered is stable under generization, one may first, by considering an arbitrary finite extension of , and taking account of definition (6.7.6) of geometric property , as well as of the invariance under base-field change of the two other properties figuring in (ii), replace in (ii) the geometric property by property . Proceeding then as in (i), one reduces to the case where is the spectrum of a discrete valuation ring , and since is catenary, it suffices to apply (5.12.5) to conclude.

(iv) The set of points such that is geometrically normal at the point is contained in the set of such that is geometrically pointwise integral and satisfies (S_2) at the point , and the latter is open in by virtue of (12.1.1, (viii)). One may therefore already suppose that, for every , is geometrically pointwise integral and satisfies (S_2) at the point , and a fortiori it is equidimensional. On the other hand, by virtue of Serre's criterion (5.8.6), to say that is geometrically normal at the point means that satisfies (S_2) and geometric property (R_1) at ;

but by virtue of (ii) and the preceding remarks, this set is the intersection of two open sets of , hence is open in . Q.E.D.

Corollary (12.1.7).

Let be a morphism locally of finite presentation. Then the set of points where possesses one of the following properties (6.8.1):

(i) satisfies property ;

(ii) is of codepth ;

(iii) is Cohen-Macaulay;

(iv) is regular (or smooth, which amounts to the same);

(v) is normal;

(vi) is reduced;

is open in .

Indeed, it follows from (11.3.1) that the set of where is flat is open. One may therefore confine oneself to the case where is flat, and then the corollary follows from (12.1.1, (iv), (vi), and (vii)) and from (12.1.6, (i), (ii), and (iv)).

Remarks (12.1.8).

(i) In (12.1.6, (ii)), one cannot suppress the hypothesis that belongs to no embedded associated prime cycle of . This is what the example (5.12.3) shows, where one takes , being the spectrum of the local ring A_0 of k[T] corresponding to the prime ideal (T) and the morphism corresponding to the injective homomorphism , deduced by localization and passage to the quotient from the injection ; this morphism is flat since is a torsion-free A_0-module . Here the fibre at the closed point of , equal to , is irreducible, of dimension 1, and satisfies the geometric condition (R_0), since the local ring at its generic point is a field. By contrast, at the generic point of , the fibre has two irreducible components of respective dimensions 0 and 1, and the one of dimension 0 is not reduced, so does not satisfy condition (R_0).

(ii) In (12.1.6, (ii)), neither can one suppress the hypothesis that is equidimensional at the point . One sees this here on the example (5.12.6) with , , where A_0 is defined as in (i), the morphism coming again from the injection by localization and passage to the quotient, and being flat since is a torsion-free A_0-module . The fibre at the closed point of is reduced (hence satisfies (S_1)) and satisfies (R_1), but has two irreducible components of dimensions 2 and 1, while the fibre at the generic point of does not satisfy [condition (R_1)].

12.2. Local and global properties of the fibres of a proper, flat morphism of finite presentation

Theorem (12.2.1).

Let be a proper morphism of finite presentation, an -Module that is -flat and of finite presentation, a finite subset of , an integer. The following subsets of are open:

(i) The set of such that the set of dimensions of the associated prime cycles of is contained in .

(ii) The set of such that the set of dimensions of the irreducible components of contains .

(iii) The set of such that all the associated prime cycles of have the same dimension equal to .

(iv) The set of such that has no embedded associated prime cycle and that the set of dimensions of the irreducible components of is equal to .

(v) The set of such that has property and is equidimensional at every point of .

(vi) The set of such that .

(vii) The set of such that is a Cohen-Macaulay -Module.

(viii) The set of such that is geometrically reduced.

(ix) The set of such that is geometrically pointwise integral at each point of .

(x) The set of such that is geometrically integral.

(xi) The set of such that has no embedded associated prime cycle and that the sum of the total multiplicities (4.7.12) of at the maximal points of is .

With the exception of (ii), (iii), (iv), (x), and (xi), the properties considered are of the following form: "for every , has property ", where is one of properties (i) to (viii) of (12.1.1). It follows from (12.1.1) that the set of such that is true is open, and the set of such that is true is none other than the set ; in all these cases, the theorem is therefore already true on the sole hypothesis that the morphism is closed. For (iii), one applies (i) with reduced to a single element. There remain cases (ii), (iv), and (xi) to examine separately ((x) deducing at once from (xi) and (4.7.14)), always applying the method described in (12.0.2).

(12.2.1.1)

Case (ii): Steps A) and B) of the proof proceed as in the beginning of (12.1.1); for step A), one uses (8.9.1), (11.2.6), and (4.2.7); for step B), one uses (9.5.5) applied to . It remains to show that if (supposing Noetherian) the set of dimensions of the irreducible components of contains , then the same is true of the set of dimensions of the irreducible components of , for every generization of in . Let Y_1 be a spectrum of a discrete valuation ring such that, if and are the closed point and the generic point of Y_1, there is a morphism with , (II, 7.1.7). Applying (4.2.7), one sees that one may replace by Y_1 and by , in other words confine oneself to the case where is the spectrum of a discrete valuation ring, its closed point and its generic point.

Using , one may confine oneself to the case where , so that is quasi-flat (2.3.3); the irreducible components of then dominate (2.3.4), in other words their generic points belong to and the are the irreducible components of . But every irreducible component of is contained in one of the , hence is an irreducible component of ; now one knows (7.1.13) that the dimensions of all the irreducible components of are equal to , which completes the proof of (ii).

(12.2.1.2)

Case (iv): The same reasoning as at the beginning shows, using (12.1.1, (iii)), that one may already suppose (replacing by an open set of ) that for every , has no embedded associated prime cycle. The assertion of case (iv) is then an immediate consequence of the assertions of cases (i) and (ii).

(12.2.1.3)

Case (xi): For step A) of the reasoning, one uses (8.9.1), (11.2.6), (8.10.5, (xii)) (to preserve the hypothesis that is proper), as well as (4.2.7), (4.5.6), and (4.7.9). For step B), one sees, as in case (iv), that one may suppose that for every , has no embedded associated prime cycle, and one applies (9.8.8), which proves that the function is constructible. It therefore remains to see (supposing Noetherian and proper) that for every generization of a point , one has . Reasoning as in (12.1.1.3), one sees (using (4.7.8) and (4.5.11)) that one may suppose that the irreducible components of are geometrically irreducible and that is the sum of the lengths of at the maximal points of . Applying (4.2.7), (4.5.6), and (4.7.9) again, one reduces, as in (12.2.1.1), to the case where is the spectrum of a discrete valuation ring, its closed point and its generic point. The fact that will then be a consequence of:

Lemma (12.2.1.4).

Let be the spectrum of a discrete valuation ring, its closed point, its generic point, a proper morphism, a coherent -Module that is -flat, (resp. ) the maximal points of (resp. ). Suppose that has no embedded associated prime cycle. Then one has

  (12.2.1.4.1)         ∑_j long((ℱ_{y'})_{z'_j}) ≤ ∑_i long((ℱ_y)_{z_i}).

One has , where is a discrete valuation ring, of which we denote by a uniformizer, so that . Since is -flat, the are also the maximal points of (2.3.4); for every , let us denote by those of the that are generizations of ; it follows from (3.4.1.1) that one has

whence on summing

  ∑_i long((ℱ_y)_{z_i}) ≥ ∑_{i, j} long((ℱ_{y'})_{z'_{ij}}).

The lemma will therefore be proved if we establish that for every there is at least one index such that is one of the . Now, since is proper (hence closed) and , there exists which is a specialization of and is such that , in other words is non-empty. Since is -regular by flatness, one deduces from (3.4.3) that there is at least one point of of which is a generization; but since the points of are by hypothesis the , this completes the proof.

Corollary (12.2.2).

Under the hypotheses of (12.2.1):

(i) The function is continuous (hence locally constant) in .

(ii) The function (5.7.1) is upper semi-continuous in .

(iii) The function (12.2.1, (xi)) is upper semi-continuous in when the have no embedded associated prime cycle.

(iv) The function (10.8.1) is lower semi-continuous in .

(i) It follows from (12.2.1, (i)) applied to equal to the smallest interval of containing the dimensions of the associated prime cycles of , that is upper semi-continuous at the point ; it follows on the other hand from (12.2.1, (ii)) applied to equal to the set of dimensions of the irreducible components of , that this same function is lower semi-continuous at the point , whence the conclusion. Assertions (ii) and (iii) are nothing but other formulations of (12.2.1, (vi) and (xi)). Finally, by (12.1.3), the set of such that is open, and the conclusion of (iv) follows by the same reasoning as at the beginning of (12.2.1).

Remarks (12.2.3).

(i) One will observe that for (12.2.1, (ii)), one may dispense with the hypothesis that is proper.

(ii) In (12.2.1, (ii)), one cannot replace the condition that the set of dimensions of the irreducible components of contain , by the condition that be equal to . Indeed, let be a field, and consider the projective space of dimension 3, , where (with indeterminates); in , let X_0 be the closed sub-scheme "union of the line X_1 defined by and of the plane X_2 defined by ", which describes in geometric terms the fact that X_0 is equal to , where the graded ideal is equal to , with , ; consider the morphism which, in geometric terms, is the "projection of X_0 onto X_1 from the line at infinity defined by "; in algebraic form, corresponds to the homomorphism of graded algebras obtained by passage to the quotient from the canonical injection . It is clear that is a projective morphism, hence proper; so is its restriction , where and ; to see that is flat, it suffices to remark that is a principal ring and the ring a torsion-free -module . For every distinct from the point defined by , then has two irreducible components, of dimensions 0 and 1, while has only one irreducible component of dimension 1.

(iii) In (12.2.1, (i)), one cannot either replace the condition that the set of dimensions of the associated prime cycles of be contained in by the condition that it be equal to . Indeed, let be a field, the polynomial ring k[t], the sub-ring of the polynomial ring k[t, u] (with , indeterminates); is not integrally closed, the element belonging to its integral closure but not being in ; if is the maximal ideal of , generated by , , , and , is an embedded associated prime ideal of . Set , , and let be the morphism corresponding to the homomorphism which transforms into . Since this homomorphism makes a torsion-free -module, is flat . If is the point of corresponding to the maximal ideal tA, the prescheme is irreducible and of dimension 1, but admits an embedded associated prime cycle of dimension 0; on the contrary, if is the generic point of , has no embedded associated prime cycle, being the spectrum of an integral -algebra of dimension 1. In this example, is an affine morphism that is not proper; one can immediately deduce from it an analogous example where is proper and flat by considering as a dense open set of a projective scheme over ((II, 5.3.4 and 5.3.2)), or by proceeding directly as in Remark (ii).

Remarks (ii) and (iii) show the necessity of including in (12.2.1, (iv)) the condition that have no embedded associated prime cycle.

(iv) In (12.2.1, (xi)), one cannot suppress the hypothesis that is proper. Indeed, let be a field, the "affine line", spectrum of the polynomial ring k[t] ( indeterminate), the sum prescheme of and of the complementary open set of the closed point of defined by . It is clear that the morphism which is equal to the canonical injections on each of the components of is flat and that if one takes , has no embedded associated prime cycle for any . However, one sees at once that one has and for every .

(v) In (12.2.1, (xi)), neither can one suppress the hypothesis that is without embedded associated prime cycle. This is what the example of Remark (ii) shows: one sees at once in effect that one has, with , and for in .

(vi) We do not know whether in (12.2.1, (xi)) one may replace the hypothesis that is -flat by the hypothesis that and that is universally open. Taking up the proof of (12.2.1.4) again, one sees (also using (14.3.6)) that one would have to resolve the following question: let be a Noetherian local ring, an -module

of finite type, an element of the maximal ideal of that is contained in no minimal prime ideal of ; if there exists one of these minimal prime ideals such that , is it true that is an associated prime ideal of ( not being supposed -regular)?

One will note however that one cannot, in (12.2.1, (xi)), suppress purely and simply the hypothesis that is -flat. One will take here as in Remark (ii), then in the closed sub-scheme X_0 union of the three lines X_1, X_2, X_3 defined respectively by , , ; one defines the projection of X_0 onto X_1 as before, and its restriction , where is the affine line and ; is proper but not flat, and if one takes , has no embedded associated prime cycles for any . But one has and for in .

Theorem (12.2.4).

Let be a proper, flat morphism of finite presentation, an integer . The following subsets of are open:

(i) The set of such that possesses property .

(ii) The set of such that satisfies geometric property , is equidimensional at every point, and has no embedded associated prime cycle.

(iii) The set of such that is geometrically regular (i.e. smooth over ).

(iv) The set of such that is geometrically normal.

(v) The set of such that is geometrically reduced.

(vi) The set of such that is geometrically reduced and that the geometric number of connected components of is equal to .

(vii) The set of such that is geometrically pointwise integral (4.6.9).

(viii) The set of such that is geometrically integral.

(ix) The set of such that has no embedded associated prime cycle and that the total multiplicity (4.7.4) of over is .

Except for (vi), these assertions are special cases of assertions of (12.2.1), or are deduced from assertions of (12.1.6) as at the beginning of the proof of (12.2.1). For (vi), one reduces as always (taking into account the invariance under base change of the geometric number of connected components (4.5.6)) to the case where is Noetherian. The set of such that is geometrically reduced is open in by virtue of (v); it then follows from (III, 7.8.7 and 7.8.6) that for every , there is a neighbourhood of and an integer such that, for every , is isomorphic to ; but by virtue of (III, 4.3.4), is then the geometric number of connected components of , whence the conclusion. One will give another proof of (vi) in (15.5.9).

12.3. Local cohomological properties of the fibres of a flat morphism locally of finite presentation

Lemma (12.3.1).

Let be a ring, an -algebra of finite presentation that is a flat -module, a -module of finite presentation that is a flat -module. Then the following conditions are equivalent:

a) is a projective -module.

b) is a flat -module.

c) For every , is a projective -module.

The equivalence of a) and b) follows from Bourbaki, Alg. comm., chap. II, §5, n° 2, cor. 2 of th. 1. Since a) implies c) trivially, it remains to prove that c) implies b), which follows from the fibrewise flatness criterion (11.3.10), applied with , .

Proposition (12.3.2).

Let be a ring, an -algebra of finite presentation that is a flat -module, a -module of finite presentation that is a flat -module. Then:

(i) There exists a left resolution of by free -modules of finite type.

(ii) One has

  (12.3.2.1)    dim. proj_B(M) = sup_{y ∈ Spec(A)} dim. proj_{B ⊗_A k(y)}(M ⊗_A k(y)) = Tor.dim_B(M)

where is the smallest integer such that for every and every -module (and if no such integer exists).

(i) By virtue of (8.9.1), there exists a Noetherian sub-ring A_0 of , an A_0-algebra of finite type B_0 and a B_0-module of finite type M_0 such that is isomorphic to and to . Moreover (11.2.7), one may suppose that B_0 and M_0 are flat A_0-modules. There then exists a left resolution of M_0 formed of free B_0-modules of finite type, and since M_0 and the are flat A_0-modules, is a left resolution of formed of free -modules of finite type (2.1.10).

(ii) By virtue of (0, 17.2.2, (ii)), one has Tor.dim_B(M) ≤ dim. proj_B(M), and the definition of the projective dimension immediately shows that, for every , one has dim. proj_{B ⊗_A k(y)}(M ⊗_A k(y)) ≤ dim. proj_B(M). To prove the reverse inequalities, consider a left resolution of by free -modules of finite type, and suppose that (resp. for every ). Then is a -module of finite type that is also a flat -module, by virtue of the hypothesis on and and of (2.1.10). In addition, one has for every -module (M, V, 7). The hypothesis therefore entails for every -module , that is to say that is a flat -module, hence projective by virtue of (12.3.1); this establishes that . The hypothesis for every entails on the other hand, by tensorization with , that in each of the sequences (exact by virtue of the flatness over of , of the , and of (2.1.10))

  0 → R ⊗_A k(y) → L_{n−1} ⊗_A k(y) → ⋯ → L_0 ⊗_A k(y) → M ⊗_A k(y) → 0,

is a projective -module (for ). One concludes once again from (12.3.1) that is a projective -module, hence .

Proposition (12.3.3).

Let be a morphism locally of finite presentation, a complex formed of quasi-coherent -Modules of finite presentation; for every , let be the complex of -Modules of finite type. Suppose that is -flat. Then the set of such that is open in . If moreover is -flat, then one has .

One may evidently confine oneself to the case where and where the complex reduces to . One may first reduce to the case where and

are affine, then to the case where is Noetherian; indeed, by (8.9.1) and (8.5.2), one knows that there exist a Noetherian prescheme , where is a sub-ring of , a morphism of finite type , and three coherent -Modules () such that , , , as well as two homomorphisms , such that , , and . One may moreover suppose that is -flat (11.2.7), and that is -flat when is supposed -flat. By faithful flatness, the hypothesis is equivalent to , where is the projection of in ; if is the set of such that , then is the inverse image of by the projection , which reduces, for the first assertion, to the case where is Noetherian. For the second assertion, one further remarks that is an inductive limit of its sub-rings that are -algebras of finite type; set , where , , and let , , so that one has . Now, since the functor lim is exact in the category of commutative groups, one has Ker(v) = lim Ker(v_λ), , and . If one has supposed that is -flat and (by reducing to the case where ) that one has proved for every , one will indeed deduce the assertion.

I) Suppose henceforth Noetherian. One knows (without flatness hypothesis on ) that the set is constructible in (9.9.6). Using now , it remains to show that for every generization of in , one has also . The method exposed in (12.0.2) applies without change (taking into account the fact that for a prescheme over a field and an extension of , the projection is faithfully flat). One may therefore suppose that is the spectrum of a discrete valuation ring , of which one denotes by a uniformizer, with above the closed point of and above the generic point of . The hypothesis that is -flat entails that is -regular . One is then reduced to proving the following lemma (where , , , will be replaced by , , , and respectively):

Lemma (12.3.3.1).

Let be a Noetherian local ring, , , three -modules, with of finite type, two homomorphisms such that . Let be an element of the maximal ideal of such that is -regular and such that the sequence

  (12.3.3.2)        M/tM ─u⊗1→ N/tN ─v⊗1→ P/tP

is exact. Then the sequence is exact.

Let us first note that if one replaces by its image in and by the injection , the image of is the same as that of , hence one may, without changing the hypothesis nor the conclusion, suppose injective. On the other hand, if is the image of and the canonical surjection, is evidently -regular, one has , and the kernel of is contained in that of ; it is consequently equal to it if the sequence (12.3.3.2) is exact; since on the other hand one has evidently ,

one sees that one may, to prove the lemma, suppose moreover surjective. The lemma will then be a consequence of:

Lemma (12.3.3.3).

Let be a ring, , , three -modules, , two homomorphisms such that is injective, surjective, and . Let be an element of such that is -regular. Then one has

  (12.3.3.4)        Ker(v ⊗ 1)/Im(u ⊗ 1) = (Ker(v)/Im(u)) ⊗_B (B/tB)

up to a canonical isomorphism.

Indeed, the hypothesis that the sequence (12.3.3.2) is exact will then entail , and since belongs to the maximal ideal of and is a -module of finite type (since is supposed Noetherian in (12.3.3.1)), Nakayama's lemma will prove that .

It therefore remains to prove (12.3.3.3). Set , , , , so that one has the exact sequences

  0 → M → Z → H → 0
  0 → Z → N ─v→ P → 0

whence, by tensorizing with and using lemma (3.4.1.4) and the fact that is -regular, the exact sequences

  M/tM ─w'→ Z/tZ → H/tH → 0
  0 → Z/tZ → N/tN ─v'→ P/tP → 0

whence , and since , one obtains (12.3.3.4).

II) Suppose now in addition that is -flat; replacing furthermore possibly by , one may suppose that . The task is to see that for every , one has . Let , and let be the ideal , which is contained in the maximal ideal of ; since is an -module of finite type ( being locally Noetherian), it is separated for the -adic topology , hence it suffices to show that its separated completion for this topology is 0. Now, by virtue of (III, 7.4.7.2), this separated completion is equal to , and it will therefore suffice to prove that each of the terms of this projective system is null. But this is true by hypothesis for ; let us therefore reason by induction on . The conclusion will follow from the more general lemma below (which one will apply for and equal to the maximal ideal of ):

Lemma (12.3.3.5).

Let be a ring, a nilpotent ideal of , a complex of -modules. Set , , and suppose that is a flat A_0-module, and that and are flat -modules. Then the relation entails .

Set ; since there exists an integer such that , it will suffice to prove, by induction on , that the sequence is

exact (this assertion following from the hypothesis for ). Let then be an element such that . Since by the inductive hypothesis the sequence is exact, the canonical image of in belongs to , hence there exists such that with . Since one has , the relation entails , and on the other hand one evidently has ; everything therefore comes down to proving that the sequence

  𝔍^k P/𝔍^{k+1} P ─u''→ 𝔍^k Q/𝔍^{k+1} Q ─v''→ 𝔍^k R/𝔍^{k+1} R

(where u'' and v'' come from and by restriction and passage to the quotients) is exact. Now, by hypothesis, is a flat -module, hence the sequence

  (P/𝔍P) ⊗_{A/𝔍} (𝔍^k/𝔍^{k+1}) ─u''→ (Q/𝔍Q) ⊗_{A/𝔍} (𝔍^k/𝔍^{k+1}) ─v''→ (R/𝔍R) ⊗_{A/𝔍} (𝔍^k/𝔍^{k+1})

is exact. But identifies with , by virtue of the flatness of over , and likewise identifies with . Finally, the image of identifies with that of u'', and this completes the proof.

Corollary (12.3.4).

Let be a flat morphism locally of finite presentation, , two -Modules of finite presentation and -flat. Let be an integer , the set of such that (resp. ). Then is open, and one has (resp. ).

One may evidently confine oneself to the case where and are affine. Then (12.3.2), there exists a left resolution of formed of free -Modules of finite type. If one sets (resp. ), one deduces that each of the (resp. ) is isomorphic to an -Module of the form , hence the and are -Modules of finite presentation and -flat. In addition, for every base change , if one sets , , , then is still a left resolution of by free -Modules of finite type (2.1.10), and is equal to (resp. is equal to ) according to what precedes. In particular, one has, for every , (resp. ). Applying (12.3.3) to the complexes of -flat Modules and , one deduces the corollary at once.