Exposé IV. Flat Morphisms

Here we give above all the flatness properties that were used in the preceding exposés. A more detailed study will be found in Chapter IV of the Éléments de Géométrie Algébrique in preparation,1 where the following situation is studied systematically: locally of finite type over locally noetherian , and coherent on and -flat; one seeks relations among the properties of , those of , and those of the coherent sheaves induced by on the fibers of , especially from the viewpoints of dimension, cohomological dimension, depth, etc. There is in particular a systematic way of obtaining theorems of Seidenberg or Bertini type, for hyperplane sections. The essential result for applying flatness methods in this context is the following, proved below: if is integral, of finite type over , and coherent on , then there exists a nonempty open of such that is -flat at the points of lying over . A second, no doubt still more important, way in which flatness enters algebraic geometry is descent theory: see, for example, Grothendieck's two exposés on the subject in the Bourbaki seminar,2 and Exposés VIII and IX below. Flatness thus seems to be one of the central technical notions in algebraic geometry.

Recall that the notion of flatness and faithful flatness was introduced by Serre in GAGA. An exposition of the following numbers IV.1 and IV.2 is also found in Bourbaki's Algèbre Commutative, which of course, as the title of the book indicates, is not restricted to commutative base rings.3

Contrary to the preceding exposés, we do not suppose that the rings under consideration are necessarily noetherian.

1. Sorites on Flat Modules

A module over a ring is said to be flat, or -flat if one wants to specify , if the functor

T_M: N ↦ M ⊗_A N

which is in any case right exact, is exact, i.e. transforms monomorphisms into monomorphisms. Equivalently, the first right-derived functor, or all the right-derived functors, vanish; that is, one has

Tor^A_1(M,N) = 0     for all N,

respectively

Tor^A_i(M,N) = 0     for i > 0 and all N.

Since the commute with inductive limits, it is enough to verify these conditions for of finite type; indeed, taking then a composition series of with monogenic quotients, it is enough to have

for monogenic, i.e. of the form , where is an ideal of . Note moreover that

Tor^A_1(M,A/I) = 0  ⇔  I ⊗_A M → M = A ⊗_A M is injective,

as one sees from the exact sequence of Tor, taking into account that . Thus flat is equivalent to saying that for every ideal , the natural homomorphism

I ⊗_A M → IM

is an isomorphism. It is enough to verify this for of finite type; a fortiori it is enough to verify that the functor is exact on modules of finite type.

As always when one has an exact functor , if for a subobject of one identifies with a subobject of , then for two subobjects , of one has

T(N′ ∩ N″) = T(N′) ∩ T(N″),
T(N′ + N″) = T(N′) + T(N″).

A direct sum of flat modules, and a direct factor of a flat module, is flat. In particular, since is flat, a free module, hence also a projective module, is flat. The tensor product of two flat modules is flat; and if is flat over , then is flat over for every base change , by associativity of the tensor product and the fact that a composite of exact functors is exact. If is flat over , and flat over , then is flat over , for the same reason.

The exact sequence of Tor, plus the “commutativity” of Tor, gives:

Proposition.

Let

be an exact sequence of -modules, with flat. Then:

  1. This sequence remains exact after tensoring by any -module .
  2. is flat if and only if is flat.

Thus one may say that, from the point of view of behavior under tensor products, flat modules are “as good” as free or projective modules; in particular, the exact sequence of IV.1.1 is “as good” as if it split.

Let be a multiplicatively stable subset of . Then is flat over , because is an exact functor in . If is -flat, then is -flat; the converse is true if is an isomorphism, i.e. if the are bijective on , by transitivity of flatness, since is flat over . More generally, the case of a morphism of preschemes and a quasi-coherent sheaf on whose flatness relative to one wants to study leads to the situation with two rings.

Proposition.

Let be a homomorphism of rings, let be a -module, and let be a multiplicatively stable subset of .

  1. If is -flat, then is -flat, hence also -flat for every multiplicatively stable subset of mapping into .
  2. Conversely, if is flat over for every maximal ideal of , equivalently over where is the prime ideal of inverse image of , then is -flat.

Indeed, there is the formula, functorial in the -module :

T⁻¹M ⊗_A N = T⁻¹(M ⊗_A N),

for the two sides are functorially isomorphic to , with , by the associativity formulas for . It follows at once that if is exact in , then the same is true of , as a composite of two exact functors; this gives (i). And (ii) follows in the same way, since to verify exactness of a sequence of -modules it is enough to verify exactness of the localizations at all maximal ideals of .

Proposition.

  1. Let be a flat -module. If is not a zero-divisor in , then it is not a zero-divisor in . In particular, if is integral, is torsion-free.
  2. Suppose is integral and that for every maximal ideal of , is principal, for example is a Dedekind ring, or even a principal ideal domain. In order that the -module be flat, it is necessary and sufficient that it be torsion-free.

For (i), note that homothety by on is obtained by tensoring homothety by on with . For (ii), by IV.1.2 (ii) one may already suppose principal. One must show that if is torsion-free, then for every ideal of , the injection , tensored by , is an injection. This means that the generator of is not a zero-divisor in , as required.

2. Faithfully Flat Modules

A functor from one category to another is said to be faithful if, for all and , the map is injective. If is an additive functor between additive categories, this is equivalent to saying that implies , and this implies that implies . For to be faithful and exact, it is necessary and sufficient that the following condition hold: for every sequence of morphisms in , the transformed sequence is exact if and only if the original one is exact. Or again: is exact, and implies . To speak of exactness, of course, the categories involved must be abelian.

Suppose one has a family of nonzero objects of such that every nonzero object of has a subobject admitting a quotient isomorphic to some . Then is faithful and exact if and only if is exact and for all . If is the category of modules over a ring , one may take for , for example, the family of , with running through the maximal ideals of . Indeed, every nonzero module admits a nonzero monogenic submodule, hence one isomorphic to , with an ideal , which by Krull admits a quotient . From these sorites one deduces in particular:

Proposition.

Let be an -module. The following conditions are equivalent:

  1. The functor is faithful and exact.
  2. is flat, and implies .
  3. is flat, and for every maximal ideal of .
  4. For every sequence of homomorphisms , the sequence tensored by is exact if and only if the initial sequence is exact.

One then says that is a faithfully flat -module. In particular, if is faithfully flat, then is a monomorphism, epimorphism, or isomorphism if and only if the homomorphism obtained by tensoring by is one. A faithfully flat module is faithful, since homothety by on is obtained by tensoring homothety by on with .

As in IV.1, one sees the usual transitivity properties: the tensor product of two faithfully flat modules is faithfully flat; if is faithfully flat over , then is faithfully flat over for every extension of the base ; if is an -algebra faithfully flat over and is a faithfully flat -module, then is a faithfully flat -module.

Corollary.

Let be a local homomorphism of local rings, and let be a -module of finite type. In order that be faithfully flat over , it is necessary and sufficient that it be flat over and nonzero.

This follows from criterion (i ter) and Nakayama. In particular, for to be -flat, it is necessary and sufficient that it be faithfully -flat.

Proposition.

Let be a homomorphism of rings, and let be a -module faithfully flat over . For every prime ideal of , there exists a prime ideal of inducing it.

Dividing by , we are reduced to the case . Localizing at the prime ideal 0, we are reduced to the case where is a field. But , being faithfully flat over , is nonzero; a fortiori , hence has a prime ideal, which can only induce the unique prime ideal of . Geometrically, one may say that the existence of a quasi-coherent sheaf on that is “faithfully flat” relative to implies that is surjective.

Corollary.

Suppose is flat over , of finite type over , and , i.e. for every prime ideal of . Then the prime ideals of containing and minimal among such ideals induce .

We are again reduced to the case , since all the hypotheses are preserved by dividing, hence is integral. We are reduced to the following statement.

Corollary.

With as above, every minimal prime ideal of induces a prime ideal of that is minimal.

Indeed, localizing at and , we are reduced to proving that if and are local, is local, is a nonzero -module flat over , and has dimension 0, then has dimension 0. By IV.2.2 and IV.2.3, every prime ideal of is induced by a prime ideal of , hence by the maximal ideal of , and therefore is the maximal ideal, as required. Geometrically, IV.2.5 means that every irreducible component of dominates some irreducible component of , provided there exists a quasi-coherent sheaf of finite type on , with support , and flat relative to .

Note that in IV.2.4 we did not have to suppose faithfully flat over , but then nothing guarantees the existence of a prime ideal containing , and hence of a minimal one among such ideals.

Proposition.

Let be a homomorphism of rings. The following conditions are equivalent:

  1. is a faithfully flat -module.
  2. is flat over , and is surjective.
  3. is flat over , and every maximal ideal is induced by an ideal of .
  4. is injective and Coker i is a flat -module.
  5. The functor in the -module is exact, and the canonical functorial homomorphism is injective.
  6. For every ideal of , is an isomorphism, and the inverse image of IB in is equal to .

We have (i) ⇒ (ii) by IV.2.3; (ii) ⇒ (ii bis) is trivial; (ii bis) ⇒ (i) by criterion (i ter) of IV.2.1. We have (iii) ⇒ (iv) by IV.1.1; (iv) ⇒ (iv bis) trivially, taking in the second condition (iv bis); and (iv bis) ⇒ (i) by the flatness criterion by ideals seen at the beginning of IV.1 and by criterion IV.2.1 (i ter). Finally, (iv) ⇒ (iii) by an easy converse of IV.1.1, and (i) ⇒ (iv), because if is the kernel of , then, since is exact, is zero; hence , whence .

3. Relations with Completion

Let be a noetherian ring, let be an ideal in , let  be the separated completion of for the -preadic topology, and for every -module , let be its completion for the -preadic topology. This is an Â-module, whence a canonical homomorphism

M ⊗_A Â → M̂.

When ranges over modules of finite type, the functor is exact, as follows easily from Krull's theorem: if , then the topology of is the one induced by the topology of . Since is right exact, one easily concludes, by resolving by with and free of finite type, that the functorial homomorphism above is an isomorphism, since is also right exact, and consequently that is also an exact functor in . Therefore:

Proposition.

Let be a noetherian ring and an ideal of . Then the separated completion  of , for the -preadic topology, is flat over .

Corollary.

In order that  be faithfully flat over , it is necessary and sufficient that be contained in the radical of .

Indeed, it suffices to apply criterion IV.2.1 (i ter).

These results summarize all that can be said, from the viewpoint of linear algebra, about the relations between and Â. Corollary IV.3.2 is used especially when is a noetherian local ring and is contained in the maximal ideal , and most often is equal to it.

4. Relations with Free Modules

Proposition.

Let be a ring, let be an ideal of , and let be an -module. Suppose one is under one or the other of the following hypotheses:

  1. is nilpotent.
  2. is noetherian, lies in the radical of , and is of finite type.

In order that be free over , it is necessary and sufficient that be free over and that .

This is necessary. We prove the sufficiency. Let be a family of elements of whose image in defines a basis there over ; it is a finite family in case (b). Let be the free -module constructed on the same index set. Thus there is a homomorphism such that tensoring by induces an isomorphism . If is the cokernel of , then , whence by Nakayama, valid under either condition (a) or (b). Thus is surjective. Let be its kernel. We then have an exact sequence

whence, since , an exact sequence , whence , and hence again by Nakayama, taking into account that in case (b), is of finite type because was assumed noetherian.

Corollary.

One may replace the condition by: the canonical surjective homomorphism

gr_I^0(M) ⊗_{A/I} gr_I(A) → gr_I(M)

is an isomorphism.

Indeed, if is free, this is certainly verified. Thus one must prove that if is free over and the condition on the associated graded objects is verified, then is free. Resume the proof above by constructing . The hypothesis implies that this homomorphism induces an isomorphism on associated graded objects; hence its kernel is contained in the intersection of the , and so is zero, trivially in (a), and by a well-known fact in (b). This proves the assertion.

Corollary.

Suppose is a field. Then the following conditions on are equivalent:

  1. is free.
  2. is projective.
  3. is flat.
  4. .
  5. The canonical homomorphism IV.4.2 is bijective.

Indeed, in the case considered, is automatically free.

The preceding result is valid in the following two cases:

  1. is an arbitrary module over a local ring whose maximal ideal is nilpotent, for example an artinian local ring.
  2. is a module of finite type over a noetherian local ring.

Recall, for reference:

Corollary.

Suppose is a noetherian local integral ring with maximal ideal , residue field , and field of fractions . Let be a module of finite type over . Then the preceding equivalent conditions (i) to (v) are also equivalent to:

  1. and are vector spaces of the same dimension, i.e. the rank of over is equal to the minimum number of generators of the -module .

The proof is immediate. We leave it to the reader to generalize to the case where is only assumed to have no nilpotent elements; one must then require that the ranks of at the minimal prime ideals of be equal to the dimension of the vector space .

5. Local Flatness Criteria

Proposition.

Let be a ring equipped with an ideal , and let be an -module. Suppose

Then the canonical surjective homomorphism

gr_I^0(M) ⊗_{A/I} gr_I(A) → gr_I(M)

is an isomorphism. The converse is true if is nilpotent.

The hypothesis means that the homomorphisms

I^n ⊗_A M → I^nM

are isomorphisms, whence at once the fact that the homomorphisms

I^n/I^{n+1} ⊗_A M → I^nM/I^{n+1}M

are isomorphisms. Conversely, suppose this condition holds and is nilpotent. We prove for every . This is true for large , so proceed by descending induction on , supposing it proved for . We have a commutative diagram

        M ⊗ I^{n+1}  →  M ⊗ I^n  →  M ⊗ (I^n/I^{n+1})  →  0
              ↓              ↓                 ↓
0  →       MI^{n+1}  →      MI^n  →       MI^n/MI^{n+1} →  0

whose rows are exact. By hypothesis, the last vertical arrow is an isomorphism, and the induction hypothesis also means that the first vertical arrow is one. The same is therefore true of the middle vertical arrow, which completes the proof.

The following proposition was isolated by Serre at the time of the Seminar; it allows substantial simplifications in the present number.

Proposition.

Let be a homomorphism of rings, and let be an -module. The following conditions are equivalent:

  1. For every -module , one has .
  2. , and is -flat.

There is a functorial isomorphism

M ⊗_A N = (M ⊗_A B) ⊗_B N,

which expresses the left-hand side, regarded as a functor in , as a composite of two functors and . Since the first sends free -modules to free -modules, hence projectives to projectives, one has the spectral sequence for composite functors

Tor^A_n(M,N) ⇐ Tor^B_p(Tor^A_q(M,B),N),

whence an exact sequence in low degrees

0 ← Tor^B_1(M ⊗_A B,N) ← Tor^A_1(M,N) ← Tor^A_1(M,B) ⊗_A N.

If (i) holds, then from this exact sequence one concludes for every , i.e. is -flat, hence (ii). Conversely, if (ii) holds, then in the exact sequence the terms surrounding are zero, hence (i) holds.

Corollary.

Suppose . Then the preceding conditions are equivalent to:

  1. for every -module annihilated by a power of .

Indeed, (i) means that this holds if is annihilated by . One deduces (iii) by applying the hypothesis to the .

Corollary.

Under the conditions of IV.5.3, the conditions under consideration imply that the functorial homomorphism

gr_I^0(M) ⊗_{A/I} gr_I(A) → gr_I(M)

is an isomorphism, and that is flat over .

It suffices to apply (iii) and IV.5.1. Using the converse of IV.5.1 when is nilpotent, one finds:

Corollary.

Let be a ring equipped with a nilpotent ideal , and let be an -module. The following conditions are equivalent:

  1. is -flat.
  2. is -flat, and .
  3. is -flat, and the canonical homomorphism on associated graded objects is an isomorphism.

Indeed, these are respectively the preceding conditions (iii) and (ii), and those of Corollary IV.5.4.

No longer suppose nilpotent. Then in IV.5.5 one will only have a priori the implications (i) ⇒ (ii) ⇒ (iii). On the other hand, since condition (iii) remains stable after dividing by a power of , one sees by IV.5.5 that it implies:

  1. For every integer , is flat over .

We propose to give conditions under which one can conclude (i), i.e. that is -flat. I say that it suffices for this that be noetherian and that satisfy the following finiteness condition: for every module of finite type over , is separated for the -preadic topology. It would suffice to verify this when is an ideal of finite type in . Indeed, let us prove that under these conditions, if is a monomorphism of finite-type modules, then is a monomorphism. It is enough to show that the kernel is contained in the

I^n(M ⊗_A N′) = Im(M ⊗_A I^nN′ → M ⊗_A N′),

or again in the

Im(M ⊗_A V′_n → M ⊗_A N′) = Ker(M ⊗_A N′ → M ⊗_A (N′/V′_n)),

where runs through a countable fundamental system of neighborhoods of 0 in , endowed with its -adic topology. By Krull's theorem, the -adic topology of is induced by that of , so one may take . Consider then the commutative diagram

M ⊗_A N′          →  M ⊗_A (N′/V′_n)
↓                    ↓
M ⊗_A N           →  M ⊗_A (N/I^nN).

Since and are annihilated by , the second vertical homomorphism identifies with the one obtained from the injective homomorphism by tensoring over with the flat -module ; it is therefore injective. Consequently, the kernel of is contained in the kernel of , which is what was required.

The “finiteness” condition considered for is satisfied in particular if is a module of finite type over a noetherian -algebra such that IB is contained in the radical of : indeed, then is a module of finite type over for every module of finite type over , hence is separated by Krull for the -adic topology, which is its IB-adic topology. Thus one obtains:

Theorem.

Let be a homomorphism of noetherian rings, let be an ideal of such that IB is contained in the radical of , and let be a -module of finite type. The following conditions are equivalent:

  1. is -flat.
  2. is -flat, and .
  3. is -flat, and the canonical homomorphism
gr_I^0(M) ⊗_{A/I} gr_I(A) → gr_I(M)

is an isomorphism. 4. For every integer , is flat over .

This result applies especially when and are local noetherian rings, a local homomorphism, and an ideal of contained in its maximal ideal; and one can immediately reduce IV.5.6 to this case. An interesting case is that where is a field, i.e. is maximal; then the condition that be flat over becomes superfluous. Moreover, since the are artinian local rings, condition (iv) means that the are free over the .

Corollary.

Let be a local homomorphism of noetherian local rings, and let be a homomorphism of -modules of finite type. Suppose is flat over . Then the following conditions are equivalent:

  1. is injective, and Coker u is flat over .
  2. is injective,

where denotes the residue field of .

(i) ⇒ (ii) by IV.1.1. We prove the converse. First, is injective, for it suffices to verify this on associated graded objects, where it follows from a commutative square that the reader will write. Let be its cokernel. We then have an exact sequence

By the exact sequence of Tor, taking into account hypothesis (ii) and , we get ; hence is flat over by Theorem IV.5.6.

Corollary.

Under the conditions of IV.5.6, let be an ideal of containing IB and contained in the radical. Let  be the -adic completion of , and let and be the -adic completions of and . In order that be -flat, it is necessary and sufficient that be Â-flat.

The sufficiency would already follow easily from IV.3.2. One uses criterion (iii) of IV.5.6 in the situation and in the situation . One observes that the conditions obtained in the two cases are equivalent by IV.3.2.

Corollary.

Let be local homomorphisms of noetherian local rings, and let be a -module of finite type. Here occurs only so that a finiteness condition can be placed on . Suppose is flat over . Let be the residue field of . The following conditions are equivalent:

  1. is flat over .
  2. is flat over , and is flat over .

The implication (i) ⇒ (ii) is trivial. We prove (ii) ⇒ (i). Apply criterion (iii) of IV.5.6 to . Since

M ⊗_B (B/I) = M ⊗_B (B ⊗_A k) = M ⊗_A k,

the first condition of this criterion says precisely that is flat over . The second condition of the criterion is satisfied because is flat over and is flat over , by an associativity formula for the tensor product. Of course, referring to IV.5.5 instead of IV.5.6, one obtains an analogous statement without noetherian and finiteness assumptions, when one supposes instead that the ideal of is nilpotent. The fact that was taken maximal did not enter; but in a sense the case “ maximal” is “the best possible”.

6. Flat Morphisms and Open Sets

Recall first some results on constructible sets, which are proved in circulating notes from the Dieudonné-Rosenlicht Seminar on Schemes.4

Let be a topological space. Following Chevalley, a subset of is called constructible if it is a finite union of locally closed subsets.

Lemma.

Let be a noetherian topological space, and let be a subset of . In order that be constructible, it is necessary and sufficient that for every irreducible closed subset of , is nondense in or contains a nonempty open subset of the space .

One deduces from this, using a well-known lemma of commutative algebra:

Lemma (Chevalley).

Let be a morphism of finite type of preschemes, with noetherian. Then is constructible.

Lemma.

Let be a noetherian topological space in which every irreducible closed subset admits a generic point, let be a constructible subset of , and let . In order that be a neighborhood of , it is necessary and sufficient that every generization of , i.e. every such that , belongs to .

In particular:

Corollary.

Let be a noetherian topological space in which every irreducible closed subset admits a generic point, and let be a subset of . In order that be open, it is necessary and sufficient that it satisfy the following two conditions:

  1. contains every generization of each of its points.
  2. If , then contains a nonempty open subset of the space .

Indeed, is necessarily constructible by IV.6.1, and one applies criterion IV.6.2, which proves that is a neighborhood of each of its points.

Corollary.

Let be a morphism of finite type of preschemes, with locally noetherian, let be a point of , and let . In order that transform every neighborhood of into a neighborhood of , it is necessary and sufficient that for every generization of , there exist a generization of such that .

We may evidently suppose and affine, hence noetherian. The condition is sufficient, for it is enough to prove that is a neighborhood of ; but is constructible by IV.6.1, and it suffices to apply criterion IV.6.3. The condition is necessary: let , and let be the union of the irreducible components of that do not contain . Then is an open neighborhood of , so its image is a neighborhood of , and a fortiori contains . Thus there exists such that . Consider an irreducible component of containing ; it necessarily contains

otherwise it would be contained in . Let be its generic point. This is a generization of , and is a generization of contained in , hence is equal to , as required.

Theorem.

Let be a morphism locally of finite type, with locally noetherian, and let be a coherent sheaf on with support , flat relative to . Then is an open morphism, i.e. transforms open sets into open sets.

It suffices to prove criterion IV.6.5 for every point . The generizations of correspond to the prime ideals of , those of correspond to the prime ideals of , and therefore one must verify that every prime ideal of is induced by a prime ideal of . But is a nonzero -module, flat over , hence faithfully flat over by IV.2.2. We may therefore apply IV.2.3, which completes the proof.

Remarks. Since flatness is preserved under extension of the base, one sees that under the conditions of IV.6.5, is even universally open. I do not know, however, when is integral and is of finite type over , whether induces on every component of an open morphism, or even only an equidimensional one,5 i.e. one whose fiber components all have the same dimension; one only knows that dominates . The question is related to the following one: let be a local homomorphism of noetherian local rings, such that is flat over and is an ideal of definition of , which implies moreover . Is it true that for every minimal prime ideal of one has ? Let us only point out that the answer to the first question is negative if one replaces the flatness hypothesis of IV.6.5 by the sole hypothesis that is universally open.

Lemma.

Let be a noetherian integral ring, let be an -algebra of finite type, and let be a -module of finite type. Then there exists a nonzero element of such that is a free, a fortiori flat, module over .

Let be the field of fractions of . Then is an algebra of finite type over , and is a module of finite type over it. Let be the dimension of the support of this module; we argue by induction on . If , i.e. if , then taking a finite set of generators of over , one sees that there exists annihilating these generators, hence annihilating ; thus , and we are done. Suppose . We know that the -module admits a composition series whose successive quotients are isomorphic to modules , with prime ideals of . Since an extension of free modules is free, we are reduced to the case where itself is of the form , or again identical with , with an integral -algebra. Applying Noether's normalization lemma to the -algebra , one sees easily that there exists a nonzero element of such that is integral over the subring , where the are indeterminates. Thus we may already suppose is integral over ; it is then a finite torsion-free -module. Let be its rank. There is therefore an exact sequence of -modules

where is a torsion -module. It follows that the Krull dimension of the -module is strictly less than the dimension of . By the induction hypothesis, after localizing with respect to a suitable nonzero of , we may suppose that is a free -module. On the other hand, is a free -module. Hence is then a free -module, and we are done.

Lemma.

Let be a noetherian ring, an algebra of finite type over , a -module of finite type, a prime ideal of , and the prime ideal it induces on . Suppose is flat over , equivalently over . Then there exists such that:

  1. is flat over .
  2. .

Indeed, applying IV.6.7 to , one first sees that there exists in such that is flat over . On the other hand, since is flat over , we have

Since is a -module of finite type, there exists such that (b) holds. Replacing by gf, we may then suppose that (a) also holds, which proves the corollary.

Corollary.

With the notation of IV.6.8, for every prime ideal of containing and not containing , is flat over , equivalently over , where is the prime ideal of induced by .

It suffices to apply criterion IV.5.6 (ii) to the system , using localization of Tor.

Theorem.

Let be a morphism of finite type, with locally noetherian, and let be a coherent sheaf on . Let be the set of points such that is flat over . Then is an open set.

Proof. We may suppose and affine, with rings and , so is defined by a -module of finite type. We apply criterion IV.6.4. Condition (a) is trivially verified by IV.1.2 (i); it remains to verify condition (b) of IV.6.4. This is what was done in Lemma IV.6.8 and Corollary IV.6.9.

In many questions, the following weaker form of Theorem IV.6.10 is sufficient; it already follows from Lemma IV.6.7, and therefore requires neither the technique of constructible sets nor Theorem IV.5.6.

Corollary.

Under the conditions of IV.6.10, if one supposes integral, then there exists a nonempty open in such that is flat relative to at all points of .

Indeed, the open set contains the fiber of the generic point of , since the local ring of this point is a field; hence it contains an open set of the form , since is of finite type over . From IV.6.11 one also easily concludes the following result, where is supposed noetherian but not necessarily integral: there exists a partition of into locally closed subsets such that, giving the induced reduced structure, induces on each a sheaf flat relative to .

1

Cf. EGA IV 11 and 12.

2

And, for a more detailed exposition, Exposés VIII and IX below.

3

N. Bourbaki, Algèbre Commutative, Chap. I, Modules plats, Act. Sci. Ind. 1290, Paris, Hermann, 1961.

4

Cf. EGA 0_III 9, EGA IV 1.8 and 1.10.

5

The answer to the second question is affirmative, and to the first negative even if is étale; cf. EGA IV 12.1.1.5 and EGA Err_IV 33.