§15. -regular sequences and -regular sequences

15.1. -regular sequences and -quasi-regular sequences

(15.1.1) Let be a ring, an -module; in what follows, we shall denote by , for short, the graded -module (the indeterminates), where is graded by the trivial grading and by total degree. Let be an -module, () elements of , the ideal of generated by the , and equip with the -preadic filtration; one then defines a surjective graded homomorphism of graded -modules, of degree 0,

(15.1.1.1)              φ : (gr_0(M))[T_1, …, T_n] → gr_•(M)

in the following way: if is the class mod of an element , then to the pair formed by and a polynomial homogeneous of degree , one associates the class mod of ; it is immediate that this class depends only on and , and if is the set of homogeneous polynomials of degree in , one has thus defined an -linear map

                        φ_k : gr_0(M) ⊗_A S_k → gr_k(M),

which is surjective by the definition of .

We shall say in what follows that an element is not a zero-divisor in an -module if the homothety of is injective.

Lemma (15.1.2).

Let be an -module, an element of , and equip with the (fA)-preadic filtration. If is not a zero-divisor in , the canonical homomorphism defined in (15.1.1):

is bijective. The converse holds if is separated for the (fA)-preadic topology.

One has here ; to say that is injective means that for every and every , the relation implies ; it is clear that this is so if is not a zero-divisor in . Conversely, observe that the homothety defines by passage to the quotients homomorphisms , hence a graded homomorphism of degree 1 of into itself; and to say that is injective is to say that is injective. If one assumes the filtration of to be separated, it follows that is injective (Bourbaki, Alg. comm., chap. III, §2, n° 8, cor. 1 of th. 1).

Corollary (15.1.3).

Let be a Noetherian ring, an -module of finite type. For to be not a zero-divisor in , it is necessary and sufficient that be bijective.

One has only to prove that the condition is sufficient; to show that is injective, it suffices to show that for every prime ideal of , the homothety of ratio in the -module is injective (Bourbaki, Alg. comm., chap. II, §3, n° 4, th. 1). Now one has , and it is immediate that the canonical homomorphism is equal to ; if is injective, then so is . One is thus reduced to the case where is a Noetherian local ring, whose maximal ideal we denote by . If , then is invertible and there is nothing to prove. If , one knows that is separated for the (fA)-preadic topology , and it therefore suffices to apply (15.1.2).

Definition (15.1.4).

Let be a ring, an -module. One says that an element is -regular (resp. -quasi-regular) if is not a zero-divisor in (resp. if the homomorphism (15.1.2.1) is bijective).

We have thus just proved that every -regular element is -quasi-regular, and that the converse holds if is Noetherian and an -module of finite type. When one no longer assumes of finite type, this converse may fail: for example, if is a principal ring which is not a field, its field of fractions and the -module , one has for every in , hence both members of (15.1.2.1) reduce to 0, but is then a zero-divisor in .

(15.1.5) Let us now assume equipped with a filtration such that , and denote by the associated graded -module. With the notations of (15.1.1), let us set, for every ,

(15.1.5.1)              M'_k = M_k + 𝔍 M_{k-1} + … + 𝔍^{k-1} M_1 + 𝔍^k M_0.

It is clear that is a filtration on . When , where is an ideal of , the filtration is none other than the -preadic filtration. We shall denote by the graded -module associated with . Let us further denote by the tensor product of graded -modules (II, 2.1.2) . One defines a surjective graded homomorphism of graded -modules, of degree 0,

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

in the following way: if is the class of an element , the class of an element and a polynomial homogeneous of degree , one associates to the triple the class mod of ; one verifies once more at once that this element depends only on , and , and this defines the graded homomorphism (15.1.5.2), which is surjective by virtue of the definition (15.1.5.1). One recovers the homomorphism (15.1.1.1) by considering the filtration on such that .

Lemma (15.1.6).

One keeps the notations of (15.1.5) and one assumes , setting . If is -regular, the homomorphism (15.1.5.2) is bijective.

Let (resp. ) be the sub--module of terms of degree in the first (resp. second) member of (15.1.5.2). Equip with the filtration

(15.1.6.1)              (Q_k)_i = ∑_{j ≤ k-i} (gr_{k-j}(M) ⊗_A (A/fA)) T^j

so that and , and for this filtration one has

                        gr_i(Q_k) = (gr_i(M) ⊗_A (A/fA)) T^{k-i}.

Equip with the filtration formed by the . Since these filtrations are finite, it suffices to prove that the homomorphisms deduced from are injective (Bourbaki, Alg. comm., chap. III, §3, n° 8, cor. 1 of th. 1). Now one has ; on the other hand, is the image of

                        M_k + fM_{k-1} + … + f^{k-i-1} M_{i+1}

in . One is thus reduced to writing that, for , the relation

(15.1.6.2)              f^{k-i} x ∈ f M_i + f^2 M_{i-1} + … + f^{k-i-1} M_{i+1} + M'_{k+1}

implies . Now, the second member of (15.1.6.2) is contained in , and by virtue of (15.1.5.1), is contained in . The hypothesis on implies that is not a zero-divisor in for every (Bourbaki, loc. cit.). The relation therefore implies (reasoning

in ) the existence of a such that , and since , one has , whence (reasoning in ), , and finally . Q.E.D.

Definition (15.1.7).

Let be a sequence of elements of . One says that it is -regular if, for , is -regular. One says that the sequence is -quasi-regular if the canonical homomorphism defined in (15.1.1.1) is bijective.

When , one simply says "regular sequence" and "quasi-regular sequence".

Proposition (15.1.8).

With the notations of (15.1.5), suppose that the sequence is -regular. Then the canonical homomorphism (15.1.5.2) is bijective.

The proposition is none other than (15.1.6) for ; we reason by induction on . Set , so that , and let

                        M''_k = M_k + 𝔍'' M_{k-1} + … + 𝔍''^{k-1} M_1 + 𝔍''^k M_0;

one can then write

                        M'_k = M''_k + f_n M''_{k-1} + … + f_n^{k-1} M''_1 + f_n^k M''_0.

Denote by the graded -module associated with the filtration of . One can write, up to canonical isomorphisms,

        (gr_•(M) ⊗ (A/𝔍)) ⊗_A A[T_1, …, T_n]
                = (gr_•(M) ⊗ (A/𝔍''))[T_1, …, T_{n-1}] ⊗ (A/f_n A) ⊗ A[T_n]

and consequently (15.1.5.2) factors as

(15.1.8.1)              (gr''_•(M) ⊗ (A/f_n A))[T_n] → gr'_•(M)

and

        ((gr_•(M) ⊗ (A/𝔍''))[T_1, …, T_{n-1}]) ⊗ (A/f_n A) ⊗ A[T_n]
                ──ψ' ⊗ 1──→ gr''_•(M) ⊗ (A/f_n A) ⊗ A[T_n]

where is the map (15.1.5.2) in which one replaces by , by and by . The induction hypothesis implies that is bijective, and it remains therefore to see that the same holds for (15.1.8.1). Applying (15.1.6), it suffices to show that is -regular. But is identified with ; by hypothesis, the homothety of ratio in is bijective; consequently the same holds for the homothety of ratio in , since is a free -module.

Proposition (15.1.9).

Let be an -module, a sequence of elements of . If the sequence is -regular, it is -quasi-regular. The converse holds if one assumes in addition that the -modules , , …, are separated for the -preadic topology (where ).

The first assertion follows from (15.1.8), where one takes , so that the homomorphism (15.1.5.2) reduces to (15.1.1.1). Conversely, suppose the homomorphism of (15.1.1.1) is bijective. Note that if one sets ,

and if one denotes by the graded -module associated with filtered by the -preadic filtration, then factors as

(15.1.9.1)              (gr''_0(M) ⊗ (A/f_n A))[T_n] → gr_•(M)

and as , where is the canonical homomorphism

Since both these homomorphisms are surjective, they are bijective if is assumed bijective; hence is necessarily injective, and consequently bijective since it is surjective. One can then reason by induction on (the case resulting from (15.1.2)), since on , , …, the -preadic topology is finer than the -preadic topology, and is consequently separated. One thus sees that the sequence is -regular. On the other hand, let us show that the hypothesis implies that if is such that , then one has . Indeed, if is a representative of , let us show that one has ; our assertion will then result from the fact that (15.1.9.1) is injective. The hypothesis implies , whence, since is injective, ; by induction, one thus obtains , whence finally . One finishes the reasoning as in (15.1.2): on , the -preadic filtration is identical to the -preadic filtration, hence is separated by hypothesis, and one concludes that the homothety of ratio in is injective, which completes the proof that the sequence is -regular.

Corollary (15.1.10).

With the notations of (15.1.9), suppose that for every -module of finite type, is separated for the -preadic topology. Then, for the sequence to be -regular, it is necessary and sufficient that it be -quasi-regular.

One will note that the hypothesis of (15.1.10) implies that, if the sequence is -regular, the same holds for the sequence for every permutation of the interval [1, n].

Corollary (15.1.11).

Let be a Noetherian ring, an -module of finite type, a finite sequence of elements contained in the radical of . Then, for the sequence to be -regular, it is necessary and sufficient that it be -quasi-regular.

This is a particular case of (15.1.10), since every -module of finite type is then separated for the -preadic topology .

Remarks (15.1.12).

(i) When is a Noetherian local ring, an -module of finite type, and when the are contained in the maximal ideal of , the notion of -regular sequence was introduced by J.-P. Serre under the name -sequence [17].

(ii) Even if is Noetherian and , a sequence of two elements of which is quasi-regular is not necessarily regular. For example, the sequence (0, 1) is quasi-regular, since then , hence ; but if it is not regular; one will note however that in this case the sequence (1, 0) is regular, since 1 is not a zero-divisor in , and since , 0 is not a zero-divisor in this module (cf. (15.1.1)).

(iii) In (15.1.6), one cannot in general replace the hypothesis that is -regular by the hypothesis that it is -quasi-regular. To see this, take up again the example (15.1.4) where is a principal ring with field of fractions , and take , the filtration of being defined by , , for , whence , , for . If in , one has seen that is not -regular; but since one has , it is immediate that the homomorphism (15.1.2.1) (where is replaced by ) is injective, hence is -quasi-regular. But with the notations of (15.1.6), one then has for every , hence , while the first member of (15.1.5.2) is not reduced to 0.

Proposition (15.1.13).

Let be a ring, , two -modules, a sequence of elements of . If the sequence is -regular and if is a flat -module, then the sequence is -regular. The converse holds if is a faithfully flat -module.

Indeed (without hypothesis on ), is isomorphic to ; it then suffices to apply and to the homothety of ratio in .

Corollary (15.1.14).

Let be a ring, an -module, a finite sequence of elements of . Let be a ring homomorphism, , for . If is a flat -module and if the sequence is -regular, then the sequence is -regular; the converse holds if is a faithfully flat -module.

Indeed, is identified with , and the homothety of ratio in the -module with the homothety of ratio in the -module ; the assertion therefore follows from (15.1.13).

Proposition (15.1.15).

Let be a ring, a commutative -algebra, a finite sequence of elements of , a -module. Suppose that the sequence is -regular and that each of the -modules () is flat. Then, for every ring homomorphism , if one sets , and (), the sequence is -regular and the -modules () are flat.

Consider the sequence (exact by hypothesis)

                        0 → M ──f_1──→ M → M/f_1 M → 0

(the arrow being the homothety of ratio in ). The hypothesis that is -flat implies that the sequence

                        0 → M ⊗_A A' ──f_1 ⊗ 1──→ M ⊗_A A' → (M/f_1 M) ⊗_A A' → 0

is exact . This shows that the homothety of ratio in is injective. Set next, for , , ; one has , , ; it suffices to replace in the preceding reasoning and by and to conclude that the homothety of ratio in is injective, by virtue of the flatness of . The last assertion follows from .

Proposition (15.1.16).

Let , be two Noetherian local rings, the residue field of , a local homomorphism, a -module of finite type. Let be a sequence of elements of the maximal ideal of , and for every , let be the image of in . The following conditions are equivalent:

a) The sequence is -regular and the quotients are flat -modules for .

b) The sequence is -regular and the -module is flat.

c) The -module is flat and the sequence is -regular.

d) The -module is flat, and for every ring homomorphism , if one sets , , (), the sequence is -regular.

It is clear that a) implies b); to show that b) implies d), it suffices, by virtue of (15.1.15), to show that b) implies that the are flat -modules for (setting ). Since , one is reduced, by descending induction, to proving that if is an element of the maximal ideal of such that the homothety of ratio in is injective and is a flat -module, then is a flat -module; but this is none other than . It is clear that c) is the particular case of d) obtained by taking . It therefore remains to prove that c) implies a). Since is a flat -module and a -module of finite type, it follows from c) and that the homothety of ratio in is injective and that its cokernel is a flat -module. We reason by induction on and suppose it proved that is a flat -module; since the homothety of ratio in is injective by hypothesis, it again follows from that the homothety of ratio in is injective and that is a flat -module, which completes the proof. One will note that the proof that c) implies a) does not use the fact that the belong to the maximal ideal of .

Corollary (15.1.17).

The hypotheses on and being those of (15.1.16), let be a -module of finite type, flat over , a sub--module of such that is a flat -module. Let be a sequence of elements of the maximal ideal of having the following properties:

(i) for .

(ii) If is the image of in , the sequence is -regular.

(iii) The sum of the sub--modules of is the canonical image of in (equal to the kernel of the canonical homomorphism ).

Then the sequence is -regular and one has .

Since is a flat -module, it follows from the exactness of the sequence that the sequence is exact ; one therefore has . Since is a -module of finite type and is Noetherian, is also a -module of finite type, whence by virtue of Nakayama's lemma, the belonging to the maximal ideal of .

Lemma (15.1.18).

Let be a ring, an -module equipped with a separated and exhaustive filtration , and let be a -regular element. Then:

(i) is -regular.

(ii) For every , one has , and the canonical image of in is therefore canonically isomorphic to .

(iii) The sequence

        0 → M_{k+1}/fM_{k+1} → M_k/fM_k → gr_k(M)/f · gr_k(M) → 0

is exact, and if one equips M'' with the quotient filtration , is therefore canonically isomorphic to .

Assertion (i) follows from the hypothesis on and from Bourbaki, Alg. comm., chap. III, §2, n° 8, cor. 1 of th. 1. Since is a direct summand of , is -regular for every , hence -regular by (i), which establishes (ii). To prove (iii), it suffices to see that the map is injective: now, if is such that , it follows from (ii) that , which completes the proof of the lemma.

Proposition (15.1.19).

Let be a ring, an -module equipped with a filtration such that , a -regular sequence of elements of . Suppose in addition that for , the quotient filtration of on is separated. Then the sequence is -regular, and if one sets and equips with the quotient filtration of , then is canonically isomorphic to .

For , the proposition follows from Lemma (15.1.18). We reason by induction on , setting and denoting by the quotient filtration of on , which is separated by hypothesis; in addition is isomorphic to

                        gr_•(M)/(∑_{i ≤ n-1} f_i gr_•(M)).

By hypothesis is therefore -regular, and it therefore follows from Lemma (15.1.18) applied to that is -regular and isomorphic to , which proves the proposition.

The proposition (15.1.19) will apply in particular for filtrations satisfying one or the other of the following hypotheses:

1° The filtration is finite and separated (since this implies for large enough);

is a Noetherian ring, an ideal contained in the radical of , an -module of finite type and the -preadic filtration .

Corollary (15.1.20).

Let be a ring, an -module, an -regular sequence of elements of . Then, whatever the integers , the sequence is -regular and has a composition series whose quotients are isomorphic to .

We reason by induction on ; for the assertions are immediate since is isomorphic to , the homothety of ratio in being injective by hypothesis. Set then ; by the induction hypothesis, admits a finite filtration for which the are isomorphic to ; by hypothesis is therefore -regular; it consequently follows from (15.1.19) that is -regular. Applying the case proved at the beginning, one concludes that is -regular and that admits a composition series whose quotients are isomorphic to ; but for the filtration of quotient of the filtration , is isomorphic to by (15.1.19) applied to , hence to , which proves the corollary.

Proposition (15.1.21).

Let , be two Noetherian local rings, a local homomorphism, a -module of finite type, an -regular sequence formed of elements of the maximal ideal of . The following properties are equivalent:

a) is a flat -module.

b) If is the ideal , is a flat -module and the sequence is -regular.

The fact that a) implies b) follows from (15.1.13) and from . Conversely, if the sequence is -regular, it is -quasi-regular (15.1.9), hence the homomorphism (15.1.1.1) is bijective. But since the sequence is also assumed to be -regular, it is -quasi-regular (15.1.9), and the corresponding canonical homomorphism (15.1.1.1) is bijective. One concludes that the canonical homomorphism

                        gr_0(M) ⊗_{A/𝔍} gr_•(A) → gr_•(M)

is bijective. The conclusion then follows from , the being in the maximal ideal of and the homomorphism being local.

15.2. -regular sequences

(15.2.1) Let be a ringed space, an -Module, a sequence of sections of over , the Ideal of generated by the (image of by the homomorphism defined canonically by the ). The sub--Modules define on a filtration, and one further sets ; if is the set of homogeneous polynomials of total degree in the ring , can be considered as a simple sheaf on , and one defines as in (15.1.1) a canonical homomorphism

(15.2.1.1)              φ_k : gr_0(ℱ) ⊗_ℤ S_k → gr_k(ℱ)

in the following way: for every open set of (or of a basis of the topology of ), one considers the homomorphism (15.1.1.1)

                        φ_{k, U} : gr_0(Γ(U, ℱ)) ⊗_ℤ S_k → gr_k(Γ(U, ℱ))

defined by the sections , the -module being filtered by the submodules . The -Module being associated with the presheaf for every , the define a homomorphism of -Modules (15.2.1.1). By virtue of the exactness properties of inductive limits of modules and of their commutation with tensor products, it follows from what precedes that for every , the homomorphism on the fibres of the two members of (15.2.1.1) at the point is identified with the homomorphism of (15.1.1)

(15.2.1.2)              gr_0(ℱ_x) ⊗_ℤ S_k → gr_k(ℱ_x)

defined by the sequence of elements of .

(15.2.2) With the notations of (15.2.1), we shall say that the sequence is -regular if, denoting by the sub--Module of image of the endomorphism of defined by , the endomorphism of defined by is injective for ; we shall say that the sequence is -quasi-regular if the homomorphisms (15.2.1.1) are injective. It follows from these definitions and from the remarks of (15.2.1) that for the sequence to be -regular (resp. -quasi-regular), it is necessary and sufficient that, for every , the sequence be -regular (resp. -quasi-regular). Every -regular sequence is therefore -quasi-regular (15.1.9); the converse holds if is an -Module of finite type, if the are Noetherian rings for every and if in addition the belong to the radical of for every (15.1.11).

If is an affine scheme, and , it follows from (I, 1.3) that for the sequence to be -regular (resp. -quasi-regular) it is necessary and sufficient that it be -regular (resp. -quasi-regular).

When , one will suppress the mention of in the preceding definitions.

Remark (15.2.3).

When and , the sequence is quasi-regular for every -module in every open set not meeting , since at every point , one has and both members of (15.2.1.2) are zero. The notion of -regular sequence is therefore of interest only in the neighbourhood of points of .

Proposition (15.2.4).

Let be a ringed space, a coherent -Module, a point of , a sequence of elements of . If the sequence is -regular, there exists an open neighbourhood of such that the sequence is -regular.

Indeed, for , the -Module is coherent , hence the same holds for the kernel of the endomorphism of this -Module defined by ; since the fibre at the point of this kernel is zero, the same holds for its restriction to a neighbourhood of .

Proposition (15.2.5).

Let be a flat morphism of ringed spaces ; let be an -Module, , a sequence of elements of , the sequence of their images . If the sequence is -regular, the sequence is -regular. The converse holds if is a faithfully flat morphism .

Let be a point of , ; one has

                        𝒢_x = ψ_x^*(ℱ_y) ⊗_{ψ_x^*(𝒪_y)} 𝒪_x      and      (g_i)_x = θ_x^♯(ψ_x^*((f_i)_y));

since by hypothesis makes a flat -module, and since the functor is exact, the first assertion follows from the corresponding assertion of (15.1.14); the same holds for the second, taking into account that the hypothesis that is faithfully flat means that is surjective and that makes a faithfully flat -module.