Bourbaki

§ 1. Depth

Proposition 1

Let AA be a ring, JJ an ideal of AA and 0→M′→M→M′′→00 \to M' \to M \to M'' \to 0 an exact sequence of AA-modules. Put

p′=prof⁡(J;M′),p=prof⁡(J;M),p′′=prof⁡(J;M′′). p' = \operatorname{prof}(J; M') \quad , \quad p = \operatorname{prof}(J; M) \quad , \quad p'' = \operatorname{prof}(J; M'') .

Then one is in one of the following three cases, which are mutually exclusive:

p′=p⩽p′′,p=p′′<p′,p′′=p′−1<p. p' = p \leqslant p'' \quad , \quad p = p'' < p' \quad , \quad p'' = p' - 1 < p .

Consider the exact sequence of extension modules associated with A/JA/J and with the exact sequence above (A, X, p. 92, th. 2). Exclude the case p=p′=p′′=+∞p = p' = p'' = +\infty; there then exists in this sequence a first nonzero module, and the following module is also nonzero. This gives the following three possibilities:

a) The first nonzero module is Ext⁡Ap′(A/J,M′)\operatorname{Ext}_A^{p'}(A/J, M'). Then p′=p⩽p′′p' = p \leqslant p''.

b) The first nonzero module is Ext⁡Ap(A/J,M)\operatorname{Ext}_A^p(A/J, M). Then p=p′′<p′p = p'' < p'.

c) The first nonzero module is Ext⁡Ap′′(A/J,M′′)\operatorname{Ext}_A^{p''}(A/J, M''). Then p′′+1=p′⩽pp'' + 1 = p' \leqslant p.

Remark 5

Suppose that one has p=p′p = p' and that the injection u:M′→Mu : M' \to M which occurs in the exact sequence of prop. 1 belongs to JHom⁡A(M′,M)J \operatorname{Hom}_A(M', M). One then has p′′=p−1p'' = p - 1. In fact, the assumption implies that the mapping Ext⁡Ai(1A/J,u)\operatorname{Ext}_A^i(1_{A/J}, u) is zero for every integer ii; this excludes case a) considered above.

Proposition 2

Let AA be a ring, JJ an ideal of AA, MM an AA-module and NN an AA-module annihilated by a power of JJ. One has Ext⁡Ai(N,M)=0\operatorname{Ext}_A^i(N, M) = 0 for every integer i<prof⁡A(J;M)i < \operatorname{prof}_A(J; M).

Suppose first that JN=0JN = 0 and reason by induction on the integer i<prof⁡A(J;M)i < \operatorname{prof}_A(J; M). The assertion is obvious for i<0i < 0. Consider NN as an (A/J)(A/J)-module and choose an exact sequence of (A/J)(A/J)-modules

0→K⟶(A/J)(I)⟶N→0. 0 \to K \longrightarrow (A/J)^{(I)} \longrightarrow N \to 0 .

One deduces from this an exact sequence of extension modules

Ext⁡Ai−1(K,M)⟶Ext⁡Ai(N,M)⟶Ext⁡Ai((A/J)(I),M). \operatorname{Ext}_A^{i-1}(K, M) \longrightarrow \operatorname{Ext}_A^i(N, M) \longrightarrow \operatorname{Ext}_A^i((A/J)^{(I)}, M) .

The AA-module Ext⁡Ai−1(K,M)\operatorname{Ext}_A^{i-1}(K, M) is zero by the induction hypothesis, and the AA-module Ext⁡Ai((A/J)(I),M)\operatorname{Ext}_A^i((A/J)^{(I)}, M) is isomorphic to Ext⁡Ai(A/J,M)I\operatorname{Ext}_A^i(A/J, M)^I (A, X, p. 89, prop. 7), which is zero by definition of depth. Hence one has Ext⁡Ai(N,M)=0\operatorname{Ext}_A^i(N, M) = 0.

Pass to the general case, and reason by induction on the smallest integer m>0m > 0 such that JmN=0J^m N = 0. We have just treated the case m=1m = 1. Suppose m>1m > 1 and let i<prof⁡A(J;M)i < \operatorname{prof}_A(J; M) be an integer. Consider the exact sequence

Ext⁡Ai(N/JN,M)⟶Ext⁡Ai(N,M)⟶Ext⁡Ai(JN,M) \operatorname{Ext}_A^i(N/JN, M) \longrightarrow \operatorname{Ext}_A^i(N, M) \longrightarrow \operatorname{Ext}_A^i(JN, M)

deduced from the exact sequence 0→JN→N→N/JN→00 \to JN \to N \to N/JN \to 0. The two extreme modules are zero by the induction hypothesis, since N/JNN/JN and JNJN are annihilated by Jm−1J^{m-1}. Therefore one has Ext⁡Ai(N,M)=0\operatorname{Ext}_A^i(N, M) = 0, which was to be proved.

Corollary 1

Let m be an integer > 0 and let J' be an ideal of A which contains J^m. One has prof_A(J ; M) \leq prof_A(J' ; M).

In fact J^m annihilates the A-module A/J', hence Ext^i_A(A/J', M) is zero for every integer i < prof_A(J ; M) (prop. 2).

Corollary 2

Suppose the ideal J to be of finite type, and let J' be an ideal of A such that V(J) \supset V(J').

a) One has prof_A(J ; M) \leq prof_A(J' ; M).

b) If the ideal J' is of finite type and if V(J) = V(J'), one has prof_A(J ; M) = prof_A(J' ; M).

By II, § 4, No. 3, cor. 2 of prop. 11 and § 2, No. 6, prop. 15, there exists an integer m > 0 such that J^m \subset J'. The assertion a) therefore follows from cor. 1 and assertion b) is deduced from it.

Cor. 2 may fail when the ideal J is not of finite type (exercise 2).

2. Depth and Acyclicity

Proposition 3

Let A be a ring, C a complex of A-modules bounded on the left and p an integer. Suppose that for every pair of integers (m, n) with m \geq n \geq p, the depth of the A-module C_m relative to the annihilator of H_n(C) is > m - n. Then one has H_n(C) = 0 for n \geq p.

Since C is bounded on the left, H_n(C) is zero for n sufficiently large. If the conclusion were false, there would exist an integer q \geq p such that H_n(C) = 0 for n > q and H_q(C) \neq 0. Denote by J the annihilator of H_q(C); one then has prof_A(J ; H_q(C)) = 0. Moreover, since Z_q(C) is a submodule of C_q, and since by hypothesis one has prof_A(J ; C_q) > q - q = 0, one has prof_A(J ; Z_q(C)) > 0. One then deduces from the exact sequence

0→Bq(C)→Zq(C)→Hq(C)→0 0 \to B_q(C) \to Z_q(C) \to H_q(C) \to 0

the equality prof_A(J ; B_q(C)) = 1 (No. 1, prop. 1). By the definition of q, B_n(C) is equal to Z_n(C) for every integer n > q. From the canonical exact sequences

0→Bn(C)→Cn→Bn−1(C)→0(n>q) 0 \to B_n(C) \to C_n \to B_{n-1}(C) \to 0 \quad (n > q)

and from the hypothesis prof_A(J ; C_n) > n - q, one obtains by induction the equality prof_A(J ; B_n(C)) = n - q + 1 for every n \geq q (loc. cit.). But this is absurd since B_n(C) is zero for n sufficiently large.

Corollary 1

Let A be a ring, J an ideal of A, C a complex of A-modules bounded on the left and p an integer. Suppose that one has JH_m(C) = 0 and prof_A(J ; C_m) > m - p for m \geq p. Then one has H_n(C) = 0 for n \geq p.

In fact for n \geq p the annihilator J_n of H_n(C) contains J, hence one has prof_A(J_n ; C_m) \geq prof_A(J ; C_m) (No. 1, cor. 1 of prop. 2), so that the hypothesis of the proposition is satisfied.

Corollary 2

Let AA be a local ring, CC a complex of AA-modules bounded to the left, pp an integer. Suppose that for m≥pm \geq p, Hm(C)H_m(C) is of finite length and CmC_m of depth >m−p> m - p. Then Hn(C)=0H_n(C) = 0 for n≥pn \geq p.

The AA-module ⨁m≥pHm(C)\bigoplus_{m \geq p} H_m(C) is of finite length. Let JJ denote its annihilator; by A, VIII, § 1, No. 3, corollary, the ring A/JA/J is artinian, hence JJ contains a power of the maximal ideal of AA (A, VIII, § 10, No. 1, th. 1). It follows that prof⁡A(J;Cm)≥prof⁡(Cm)>m−p\operatorname{prof}_A(J; C_m) \geq \operatorname{prof}(C_m) > m - p for m≥pm \geq p (No. 1, cor. 1 of prop. 2), so that one may apply cor. 1.

3. Depth and Koszul complex

Let AA be a ring, MM an AA-module, x=(xi)i∈Ix = (x_i)_{i \in I} a family of elements of AA. Let u:A(I)→Au : A^{(I)} \to A denote the linear form such that u(ei)=xiu(e_i) = x_i for every i∈Ii \in I, and K∙(x,M)K^\bullet(x, M) the complex KA∙(u,M)K_A^\bullet(u, M) associated with uu (A, X, p. 147). We have Kp(x,M)=0K^p(x, M) = 0 for p<0p < 0; for p≥0p \geq 0 the AA-module Kp(x,M)=Hom⁡A(Λp(A(I)),M)K^p(x, M) = \operatorname{Hom}_A(\Lambda^p(A^{(I)}), M) is canonically identified with the AA-module CIp(M)C_I^p(M) formed by the alternating mappings of IpI^p into MM (A, X, p. 153), the differential ∂p:Kp(x,M)→Kp+1(x,M)\partial^p : K^p(x, M) \to K^{p+1}(x, M) being given by the formula

(∂pm)(α1,…,αp+1)=∑j=1p+1(−1)j+1xαjm(α1,…,αj−1,αj+1,…,αp+1) (\partial^p m)(\alpha_1, \ldots, \alpha_{p+1}) = \sum_{j=1}^{p+1} (-1)^{j+1} x_{\alpha_j} m(\alpha_1, \ldots, \alpha_{j-1}, \alpha_{j+1}, \ldots, \alpha_{p+1})

for m∈Kp(x,M)m \in K^p(x, M) and (α1,…,αp+1)∈Ip+1(\alpha_1, \ldots, \alpha_{p+1}) \in I^{p+1} (A, X, p. 154, formula (12)). It follows in particular that the complex K∙(x,M)K^\bullet(x, M) depends only on the structure of MM as a Z\mathbf{Z}-module and on the endomorphisms (xi)M(x_i)_M.

We denote by H∙(x,M)H^\bullet(x, M) the homology of the complex K∙(x,M)K^\bullet(x, M). The AA-module H0(x,M)H^0(x, M) is identified with Hom⁡A(A/J,M)\operatorname{Hom}_A(A/J, M), where JJ is the ideal of AA generated by the xix_i (A, X, p. 147, lemma 1).

Let (Mα)α∈K(M_\alpha)_{\alpha \in K} be a family of AA-modules, and MM its product; the complex K∙(x,M)K^\bullet(x, M) is canonically isomorphic to the product complex of the K∙(x,Mα)K^\bullet(x, M_\alpha), so that for each integer ss the AA-module Hs(x,M)H^s(x, M) is identified with the product of the Hs(x,Mα)H^s(x, M_\alpha) (A, X, p. 28, Proposition 1).

Theorem 1

Let AA be a ring, JJ an ideal of AA, x=(xi)i∈Ix = (x_i)_{i \in I} a generating family of JJ, MM an AA-module. The depth of MM relative to JJ is the greatest lower bound (in N∪{+∞}\mathbf{N} \cup \{+\infty\}) of the integers nn such that Hn(x,M)≠0H^n(x, M) \neq 0.

Put p=prof⁡A(J;M)p = \operatorname{prof}_A(J; M). Consider the complex K∙(x,M)K^\bullet(x, M). Its homology is annihilated by JJ (A, X, p. 148, Corollary 2), and the depth relative to JJ of each of the modules Ki(x,M)K^i(x, M) is equal to pp or to +∞+\infty (No. 1, Remark 4). It then follows from Corollary 1 of No. 2 that we have Hi(x,M)=0H^i(x, M) = 0 for i<pi < p. It remains to prove that Hp(x,M)H^p(x, M) is non-zero when p<+∞p < +\infty.

The case p=0p = 0 being evident, suppose 0<p<+∞0 < p < +\infty and Hp(x,M)=0H^p(x, M) = 0. Let LL be a free resolution of the AA-module A/JA/J; let CC denote the complex Homgr⁡A(L,M)\operatorname{Homgr}_A(L, M). The AA-module Hi(C)H^i(C) is isomorphic to Ext⁡Ai(A/J,M)\operatorname{Ext}_A^i(A/J, M) (A, X, p. 100, Theorem 1); it is therefore zero for i<pi < p. We then have for i<pi < p canonical exact sequences

0→Bi(C)→Ci→Bi+1(C)→0. 0 \to B^i(C) \to C^i \to B^{i+1}(C) \to 0 .

The AA-module CiC^i is a product of AA-modules isomorphic to MM; therefore Hs(x,Ci)=0H^s(x, C^i) = 0 for s≤ps \leq p. From the preceding exact sequences and from A, X, p. 150, we deduce that the connecting homomorphism ∂s:Hs(x,Bi+1(C))⟶Hs+1(x,Bi(C))\partial^s : H^s(x, B^{i+1}(C)) \longrightarrow H^{s+1}(x, B^i(C)) is injective for s≤ps \leq p and i<pi < p; since B0(C)=0B^0(C) = 0, it follows that Hp−i(x,Bi+1(C))H^{p-i}(x, B^{i+1}(C)) is zero for i<pi < p. In particular, we have H1(x,Bp(C))=0H^1(x, B^p(C)) = 0, so that the exact sequence

0→Bp(C)→Zp(C)→Hp(C)→0 0 \to B^p(C) \to Z^p(C) \to H^p(C) \to 0

yields a surjection H0(x,Zp(C))⟶H0(x,Hp(C))H^0(x, Z^p(C)) \longrightarrow H^0(x, H^p(C)). Since Hp(C)H^p(C) is isomorphic to Ext⁡Ap(A/J,M)\operatorname{Ext}_A^p(A/J, M), which is nonzero and annihilated by J, one has H0(x,Hp(C))≠0H^0(x, H^p(C)) \neq 0, whence H0(x,Zp(C))≠0H^0(x, Z^p(C)) \neq 0 and consequently H0(x,Cp)≠0H^0(x, C^p) \neq 0. But this implies H0(x,M)≠0H^0(x, M) \neq 0, contrary to the hypothesis. Hence Hp(x,M)≠0H^p(x, M) \neq 0, which completes the proof.

Corollary 1

Suppose the ideal J to be of finite type and JM ≠\neq M. Then* prof⁡A(J;M)\operatorname{prof}_A(J; M) is finite and ≤\leq Card(I) ; in order that it be equal to Card(I), it is necessary and sufficient that the family x be completely secant for M (A, X, p. 157, Def. 2).

Suppose first that I is finite, and let r denote its cardinal. The A-module Hr(x,M)H^r(x, M) is canonically isomorphic to H0(x,M)H_0(x, M), itself isomorphic to M/JMM/JM (A, X, p. 155); the inequality prof⁡A(J;M)≤r\operatorname{prof}_A(J; M) \leq r therefore follows from Theorem 1. For equality to hold, it is necessary and sufficient that the A-module Hi(x,M)H^i(x, M) be zero for i<ri < r, which signifies that the family x is completely secant for M (A, X, p. 157).

From the foregoing, prof⁡A(J;M)\operatorname{prof}_A(J; M) is finite; it remains to prove that if the family x is completely secant for M, the set I is finite. Now the condition H1(x,M)=0H_1(x, M) = 0 (A, X, p. 157, Def. 2) implies that one has an exact sequence

Λ2(A(I))⊗AM→∂2M(I)→∂1JM→0, \Lambda^2(A^{(I)}) \otimes_A M \xrightarrow{\partial_2} M^{(I)} \xrightarrow{\partial_1} JM \to 0 ,

where the image of ∂2\partial_2 is contained in JM(I)JM^{(I)}. By tensor product with A/JA/J, one deduces from this an A-linear isomorphism of (M/JM)(I)(M/JM)^{(I)} onto JM/J2MJM/J^2M. Now the latter module is of finite type, since J and M are so; as M/JMM/JM is not zero, it follows that the set I is finite.

Corollary 2

Let A be a local ring, J an ideal of finite type of A distinct from A, M a nonzero A-module of finite type. Put* r=[J/mAJ:κA]r = [J/m_AJ : \kappa_A]. One has prof⁡A(J;M)≤r\operatorname{prof}_A(J; M) \leq r ; there is equality if and only if J is generated by a family completely secant for M. In this case, in order that a generating family of J be completely secant, it is necessary and sufficient that it have r elements.

By Nakayama's lemma, one has JM≠MJM \neq M, and r is the minimal number of generators of J; Corollary 2 therefore follows from Corollary 1.

Proposition 4

Let ρ:A→B\rho : A \to B be a ring homomorphism, JJ an ideal of AA and NN a BB-module. One has the equality prof⁡A(J;N)=prof⁡B(JB;N)\operatorname{prof}_A(J; N) = \operatorname{prof}_B(JB; N).

Let x=(xi)i∈Ix = (x_i)_{i \in I} be a generating family of JJ; the family ρ(x)=(ρ(xi))i∈I\rho(x) = (\rho(x_i))_{i \in I} generates JBJB. By construction the complex K∙(ρ(x),N)K^\bullet(\rho(x), N) is equal to K∙(x,N)K^\bullet(x, N). The proposition therefore follows from Theorem 1.

Corollary

Let AA be a local ring, aa an ideal of AA distinct from AA and MM an AA-module annihilated by aa. One has prof⁡A(M)=prof⁡A/a(M)\operatorname{prof}_A(M) = \operatorname{prof}_{A/a}(M).

Let ρ:A→B\rho : A \to B be a ring homomorphism, x=(xi)i∈Ix = (x_i)_{i \in I} a finite family of elements of AA and MM an AA-module. For every integer pp, let us denote by up:B⊗ACIp(M)→CIp(B⊗AM)u^p : B \otimes_A C_I^p(M) \to C_I^p(B \otimes_A M) the BB-linear homomorphism which associates to b⊗mb \otimes m the alternating mapping (α1,…,αp)↦b⊗m(α1,…,αp)(\alpha_1, \ldots, \alpha_p) \mapsto b \otimes m(\alpha_1, \ldots, \alpha_p). The family (up)(u^p) defines an isomorphism of complexes

u:B⊗AK∙(x,M)→K∙(x,B⊗AM) u : B \otimes_A K^\bullet(x, M) \to K^\bullet(x, B \otimes_A M)

Let us consider the canonical homomorphism

γp(B,K∙(x,M)):B⊗AHp(x,M)→Hp(B⊗AK∙(x,M)) \gamma^p(B, K^\bullet(x, M)) : B \otimes_A H^p(x, M) \to H^p(B \otimes_A K^\bullet(x, M))

(A, X, p. 62); by composition with Hp(u)H^p(u), we deduce from it a homomorphism

vp:B⊗AHp(x,M)→Hp(x,B⊗AM) v^p : B \otimes_A H^p(x, M) \to H^p(x, B \otimes_A M)

Lemma 1

If the AA-module BB is flat, the homomorphism vpv^p is bijective for every integer pp.

This results from A, X, p. 66, Cor. 2.

Proposition 5

Let AA be a ring, JJ an ideal of finite type of AA and MM an AA-module. Let Ω\Omega denote the set of maximal ideals of AA belonging to Supp⁡(M)\operatorname{Supp}(M) and containing JJ. Then

prof⁡A(J;M)=inf⁡p∈Spec⁡(A)prof⁡Ap(Jp;Mp)=inf⁡m∈Ωprof⁡Am(Jm;Mm) \operatorname{prof}_A(J; M) = \inf_{p \in \operatorname{Spec}(A)} \operatorname{prof}_{A_p}(J_p; M_p) = \inf_{m \in \Omega} \operatorname{prof}_{A_m}(J_m; M_m)

Let x=(xi)i∈Ix = (x_i)_{i \in I} be a finite generating family of JJ. Let pp be a prime ideal of AA; the ideal JpJ_p is generated by the image xpx_p of the family (xi)(x_i) in ApA_p. For every p≥0p \geq 0, the ApA_p-module (Hp(x,M))p(H^p(x, M))_p is isomorphic to Hp(xp,Mp)H^p(x_p, M_p) (Lemma 1); in view of Theorem 1, we therefore have prof⁡A(J;M)≤inf⁡p∈Spec⁡(A)prof⁡Ap(Jp;Mp)≤inf⁡m∈Ωprof⁡Am(Jm;Mm)\operatorname{prof}_A(J; M) \leq \inf_{p \in \operatorname{Spec}(A)} \operatorname{prof}_{A_p}(J_p; M_p) \leq \inf_{m \in \Omega} \operatorname{prof}_{A_m}(J_m; M_m).

Let pp be an integer strictly less than prof⁡Am(JAm;Mm)\operatorname{prof}_{A_m}(JA_m; M_m) for every m∈Ωm \in \Omega. Then Hp(xm,Mm)=0H^p(x_m, M_m) = 0 for every maximal ideal mm of AA: this results from Theorem 1 if m∈Ωm \in \Omega, from the fact that Mm=0M_m = 0 if m∉Supp⁡(M)m \not\in \operatorname{Supp}(M), and from the fact that the ideal JAmJA_m, which annihilates Hp(xm,Mm)H^p(x_m, M_m) (A, X, p. 148, Cor. 2), is equal to AmA_m if m∉V(J)m \not\in V(J). Hence (Hp(x,M))m=0(H^p(x, M))_m = 0 for every maximal ideal mm of AA, which implies Hp(x,M)=0H^p(x, M) = 0 (II, § 3, No. 3, Cor. 2 of Theorem 1). The proposition then results from Theorem 1.

Proposition 6

Let AA be a ring, JJ an ideal of finite type of AA and MM an AA-module. Let BB be a ring and ρ:A→B\rho : A \to B a ring homomorphism making BB into a flat AA-module.

a) One has prof⁡A(J;M)≤prof⁡B(JB;B⊗AM)\operatorname{prof}_A(J; M) \leq \operatorname{prof}_B(JB; B \otimes_A M).

b) Suppose moreover that every maximal ideal of Supp⁡(M)\operatorname{Supp}(M) containing JJ belongs to the image of the canonical mapping Spec⁡(B)→Spec⁡(A)\operatorname{Spec}(B) \to \operatorname{Spec}(A). Then one has prof⁡A(J;M)=prof⁡B(JB;B⊗AM)\operatorname{prof}_A(J; M) = \operatorname{prof}_B(JB; B \otimes_A M). This is the case for example if the AA-module BB is faithfully flat.

Assertion a) follows from Theorem 1 and Lemma 1.

Let pp be an integer strictly less than prof⁡B(JB;B⊗AM)\operatorname{prof}_B(JB; B \otimes_A M), and let mm be a maximal ideal of AA belonging to Supp⁡(M)∩V(J)\operatorname{Supp}(M) \cap V(J). Let xx be a finite generating family of the ideal JJ. Under the hypothesis of b), there exists a prime ideal nn of BB lying over mm, and one has a canonical isomorphism

Bn⊗Am(Am⊗AHp(x,M))⟶Bn⊗B(B⊗AHp(x,M)) B_n \otimes_{A_m} (A_m \otimes_A H^p(x, M)) \longrightarrow B_n \otimes_B (B \otimes_A H^p(x, M))

Now B⊗AHp(x,M)B \otimes_A H^p(x, M) is isomorphic to Hp(ρ(x),B⊗AM)H^p(\rho(x), B \otimes_A M) (Lemma 1), hence is zero; moreover BnB_n is faithfully flat over AmA_m (I, § 3, No. 5, Proposition 9 and II, § 3, No. 4, Proposition 14 and Remark). Hence one has Am⊗AHp(x,M)=0A_m \otimes_A H^p(x, M) = 0 and therefore p<prof⁡Am(Jm;Mm)p < \operatorname{prof}_{A_m}(J_m; M_m) (Lemma 1 and Theorem 1). The first assertion of b) then follows from Proposition 5; the second follows from I, § 3, No. 5, Proposition 9).

Corollary

Let AA be a noetherian ring, JJ an ideal of AA, MM a finitely generated AA-module, A^\widehat{A} and M^\widehat{M} the separated completions of AA and MM for the JJ-adic topology. Then one has prof⁡A(J;M)=prof⁡A^(JA^;M^)\operatorname{prof}_A(J; M) = \operatorname{prof}_{\widehat{A}}(J\widehat{A}; \widehat{M}).

In fact, the AA-module A^\widehat{A} is flat and the A^\widehat{A}-module M^\widehat{M} is isomorphic to A^⊗AM\widehat{A} \otimes_A M (III, § 3, No. 4, Theorem 3); moreover, every maximal ideal of AA containing JJ belongs to the image of the mapping Spec⁡(A^)→Spec⁡(A)\operatorname{Spec}(\widehat{A}) \to \operatorname{Spec}(A) (loc. cit., Proposition 8).

4. Depth and regular sequences

Let AA be a ring, MM an AA-module. Recall (A, X, p. 158) that a finite sequence (x1,…,xr)(x_1, \ldots, x_r) of elements of AA is said to be regular for MM or MM-regular if, for i=1,…,ri = 1, \ldots, r, the homothety with ratio xix_i in the AA-module M/(x1M+…+xi−1M)M/(x_1 M + \ldots + x_{i-1} M) is injective. Let (x1,…,xr)(x_1, \ldots, x_r) be an MM-regular sequence; for every flat AA-module NN, the sequence (x1,…,xr)(x_1, \ldots, x_r) is regular for M⊗ANM \otimes_A N. If ρ:A→B\rho : A \to B is a ring homomorphism making BB into a flat AA-module, the sequence (ρ(x1),…,ρ(xr))(\rho(x_1), \ldots, \rho(x_r)) is regular for the BB-module B⊗AMB \otimes_A M. In particular, for every prime ideal pp of AA, the image in ApA_p of the sequence (x1,…,xr)(x_1, \ldots, x_r) is MpM_p-regular.

In the sequel we shall consider chiefly the notion of an MM-regular sequence in the case where the ring AA is local noetherian, the module MM is finitely generated and the elements of the sequence belong to mAm_A; the notion of an MM-regular sequence then coincides with that of a completely secant sequence for MM (A, X, p. 160, cor. 1).

Proposition 7

Let AA be a ring, JJ an ideal of AA, MM an AA-module, and (x1,…,xr)(x_1, \ldots, x_r) an MM-regular sequence of elements of JJ. Then

prof⁡A(J;M)=r+prof⁡A(J;M/(x1M+…+xrM)) \operatorname{prof}_A(J; M) = r + \operatorname{prof}_A(J; M/(x_1M + \ldots + x_rM))
and in particular prof⁡A(J;M)⩾r\operatorname{prof}_A(J; M) \geqslant r.

The case r=1r = 1 follows from Remark 5 of No. 1, applied to the exact sequence

0→M→(x1)MM⟶M/x1M→0. 0 \to M \xrightarrow{(x_1)_M} M \longrightarrow M/x_1M \to 0 .
The general case follows by induction on rr.

Theorem 2

Let AA be a noetherian ring, JJ an ideal of AA and MM a finitely generated AA-module.

a) Suppose that prof⁡A(J;M)\operatorname{prof}_A(J; M) is finite. Then every MM-regular sequence of elements of JJ can be completed to an MM-regular sequence of length prof⁡A(J;M)\operatorname{prof}_A(J; M) of elements of JJ.

b) The depth of MM relative to JJ is the least upper bound of the lengths of MM-regular sequences formed of elements of JJ.

c) For prof⁡A(J;M)\operatorname{prof}_A(J; M) to be finite, it is necessary and sufficient that the support of MM meet V(J)V(J), or again that one have JM≠MJM \neq M.

Let (x1,…,xr)(x_1, \ldots, x_r) be an MM-regular sequence of elements of JJ. One has r⩽prof⁡A(J;M)r \leqslant \operatorname{prof}_A(J; M) (Prop. 7); suppose that the inequality is strict. Let NN denote the AA-module M/(x1M+…+xrM)M/(x_1M + \ldots + x_rM). One has prof⁡A(J;N)>0\operatorname{prof}_A(J; N) > 0 (loc. cit.), so that there exists an element xx of JJ such that the homothety xNx_N is injective (No. 1, Remark 2), that is to say, such that the sequence (x1,…,xr,x)(x_1, \ldots, x_r, x) is MM-regular. It follows by induction that for every integer ss such that r⩽s⩽prof⁡A(J;M)r \leqslant s \leqslant \operatorname{prof}_A(J; M) the sequence (x1,…,xr)(x_1, \ldots, x_r) can be completed to an MM-regular sequence of length ss, which entails assertions a) and b). Assertion c) follows from Remark 1 of No. 1 and Corollary 1 of Theorem 1 of No. 3.

Corollary 1

For every MM-regular sequence (x1,…,xr)(x_1, \ldots, x_r) of elements of JJ, the following properties are equivalent:

(i) one has r=prof⁡A(J;M)r = \operatorname{prof}_A(J; M) ; (ii) the sequence (x1,…,xr)(x_1, \ldots, x_r) is maximal among the MM-regular sequences formed of elements of JJ ; (iii) the AA-module M/(x1M+…+xrM)M/(x_1M + \ldots + x_rM) has a nonzero element annihilated by JJ ; (iv) one has Ass⁡(M/(x1M+…+xrM))∩V(J)≠∅\operatorname{Ass}(M/(x_1M + \ldots + x_rM)) \cap V(J) \neq \varnothing.

The equivalence of (i) and (ii) follows from Theorem 2; the equivalence of (ii), (iii) and (iv) follows from Remark 2 of No. 1 applied to the AA-module M/(x1M+…+xrM)M/(x_1M + \ldots + x_rM).

Corollary 2

Let AA be a noetherian local ring, MM a nonzero finitely generated AA-module. One has

prof⁡A(M)⩽dim⁡A(M)<+∞. \operatorname{prof}_A(M) \leqslant \dim_A(M) < +\infty .

5. Depth along a closed subset

Let A be a noetherian ring, F a closed subset of Spec(A)\mathrm{Spec}(A) and M an A-module. By Corollary 2 of Proposition 2 of No. 1, the element profA(J;M)\mathrm{prof}_A(J;M) of N∪{+∞}\mathbf{N} \cup \{+\infty\} does not depend on the ideal J of A such that F=V(J)F = V(J); it is called the depth of M along F and is denoted by profF(M)\mathrm{prof}_F(M).

Remark 1

Let r be an integer. By Proposition 2 of No. 1 and II, § 4, No. 4, Corollary 2 of Proposition 17, the inequality profF(M)⩾r\mathrm{prof}_F(M) \geqslant r is equivalent to the following property: for every finitely generated A-module N whose support is contained in F, one has ExtAi(N,M)=0\mathrm{Ext}_A^i(N,M) = 0 for i<ri < r.

Remark 2

Suppose that the A-module M is finitely generated. By Remark 2 of No. 1 and Theorem 2 of No. 4, one has the following equivalences

profF(M)=0  ⟺  Ass(M)∩F≠∅profF(M)<+∞  ⟺  Supp(M)∩F≠∅. \mathrm{prof}_F(M) = 0 \iff \mathrm{Ass}(M) \cap F \neq \varnothing \\ \mathrm{prof}_F(M) < +\infty \iff \mathrm{Supp}(M) \cap F \neq \varnothing .

Proposition 8

Let A be a noetherian ring, F a closed subset of Spec(A)\mathrm{Spec}(A), and M a finitely generated A-module. One has

profF(M)=inf⁡p∈FprofAp(Mp)=inf⁡p∈Supp(M)∩FprofAp(Mp). \mathrm{prof}_F(M) = \inf_{p \in F} \mathrm{prof}_{A_p}(M_p) = \inf_{p \in \mathrm{Supp}(M) \cap F} \mathrm{prof}_{A_p}(M_p) .

This is clear if profF(M)=+∞\mathrm{prof}_F(M) = +\infty (Remark 2). If profF(M)=0\mathrm{prof}_F(M) = 0, there exists a prime ideal p∈Ass(M)∩Fp \in \mathrm{Ass}(M) \cap F (Remark 2); one has pAp∈Ass(Mp)pA_p \in \mathrm{Ass}(M_p) (IV, § 1, No. 2, Prop. 5), hence profAp(Mp)=0\mathrm{prof}_{A_p}(M_p) = 0 (Remark 2 of No. 1), whence the proposition in this case.

Suppose 0<profF(M)<+∞0 < \mathrm{prof}_F(M) < +\infty; let J be an ideal of A such that V(J)=FV(J) = F, and let x be an element of J such that the homothety xMx_M is injective (loc. cit.). For each prime ideal p, the homothety xMpx_{M_p} is injective. By Prop. 7 of No. 4, we therefore have

profF(M/xM)=profF(M)−1profAp((M/xM)p)=profAp(Mp)−1. \mathrm{prof}_F(M/xM) = \mathrm{prof}_F(M) - 1 \\ \mathrm{prof}_{A_p}((M/xM)_p) = \mathrm{prof}_{A_p}(M_p) - 1 .

One then concludes by induction on the integer profF(M)\mathrm{prof}_F(M).

Remark 3

If q is a point of Supp(M)\mathrm{Supp}(M), we therefore have profA(q;M)=inf⁡p⊇qprofAp(Mp)\mathrm{prof}_A(q;M) = \inf_{p \supseteq q} \mathrm{prof}_{A_p}(M_p). In particular, one has the inequality profA(q;M)⩽profAq(Mq)\mathrm{prof}_A(q;M) \leqslant \mathrm{prof}_{A_q}(M_q); there is equality when q is maximal. In the general case, one may have profA(q;M)<profAq(Mq)\mathrm{prof}_A(q;M) < \mathrm{prof}_{A_q}(M_q); one may also have profA(q;M)<inf⁡profAm(Mm)\mathrm{prof}_A(q;M) < \inf \mathrm{prof}_{A_m}(M_m) where m runs through the set of maximal ideals of A containing q. Let p, for example, be a nonmaximal prime ideal of A, containing q and distinct from q; put M=A/pM = A/p. One has profA(q;M)=0\mathrm{prof}_A(q;M) = 0, profAq(Mq)=+∞\mathrm{prof}_{A_q}(M_q) = +\infty and profAm(Mm)>0\mathrm{prof}_{A_m}(M_m) > 0 for every maximal ideal m of A.

Proposition 9

Let AA be a noetherian ring, MM and NN two A-modules of finite type, and FF the support of NN. Then profF(M)\mathrm{prof}_F(M) is the greatest lower bound (in N∪{+∞}N \cup \{+\infty\}) of the set of integers nn such that ExtAn(N,M)≠0\mathrm{Ext}_A^n(N, M) \neq 0.

By Remark 1, one has ExtAi(N,M)=0\mathrm{Ext}_A^i(N, M) = 0 for every i<profF(M)i < \mathrm{prof}_F(M). It remains to prove that if profF(M)=n<+∞\mathrm{prof}_F(M) = n < +\infty, one has ExtAn(N,M)≠0\mathrm{Ext}_A^n(N, M) \neq 0. Let JJ be the annihilator of NN; one has F=V(J)F = V(J), hence profF(M)=profA(J;M)\mathrm{prof}_F(M) = \mathrm{prof}_A(J; M). By Cor. 1 of Th. 2 (No. 4), there exists an MM-regular sequence (x1,…,xn)(x_1, \ldots, x_n) of length nn consisting of elements of JJ, and the depth relative to JJ of the A-module M‾=M/(x1M+…+xnM)\overline{M} = M/(x_1 M + \ldots + x_n M) is zero. By A, X, p. 166, Prop. 9, it is enough to prove that HomA(N,M‾)\mathrm{Hom}_A(N, \overline{M}) is nonzero. Now, by Prop. 8, there exists p∈Supp(M)∩Supp(N)p \in \mathrm{Supp}(M) \cap \mathrm{Supp}(N) such that profAp(M‾p)=0\mathrm{prof}_{A_p}(\overline{M}_p) = 0, that is to say HomAp(κ(p),M‾p)≠0\mathrm{Hom}_{A_p}(\kappa(p), \overline{M}_p) \neq 0. Since NpN_p is nonzero, the κ(p)\kappa(p)-vector space Np⊗Apκ(p)N_p \otimes_{A_p} \kappa(p) is nonzero (Nakayama's lemma), and so is its dual; there therefore exists a surjective ApA_p-linear mapping of NpN_p onto κ(p)\kappa(p). It follows that one has HomAp(Np,M‾p)≠0\mathrm{Hom}_{A_p}(N_p, \overline{M}_p) \neq 0, hence HomA(N,M‾)≠0\mathrm{Hom}_A(N, \overline{M}) \neq 0 (II, § 2, No. 7, Prop. 19), which was to be proved.

Remark 4

Let AA be a noetherian ring, NN a finitely generated AA-module. The greatest lower bound in N∪{+∞}N \cup \{+\infty\} of the set of integers nn such that ExtAn(N,A)\mathrm{Ext}_A^n(N, A) is nonzero is sometimes called the grade of NN, and is denoted by grade(N)(N). By Prop. 9, this is also the depth of AA along the support of NN, or again (No. 4, Theorem 2) the least upper bound of the set of lengths of AA-regular sequences of elements of the annihilator of NN. Since for every prime ideal pp of AA the annihilator of NpN_p is equal to Ann(N)p\mathrm{Ann}(N)_p (II, § 2, No. 4, formula (9)), we deduce from Prop. 5 of No. 3 the equality

grade(N)=inf⁡p∈Spec(A)grade(Np)=inf⁡m∈Ωgrade(Nm), \mathrm{grade}(N) = \inf_{p \in \mathrm{Spec}(A)} \mathrm{grade}(N_p) = \inf_{m \in \Omega} \mathrm{grade}(N_m),

where Ω\Omega denotes the set of maximal ideals of AA.

6. Depth of Algebras

Lemma 2

Let ρ:A→B\rho : A \to B be a local homomorphism of noetherian local rings, NN a finitely generated BB-module and yy an element of mBm_B. The following two conditions are equivalent:

(i) the AA-module N/yNN/yN is flat and the homothety yNy_N is injective; (ii) the AA-module NN is flat and the homothety yκA⊗Ny_{\kappa_A \otimes N} is injective.

When they are satisfied, the homothety yM⊗ANy_{M \otimes_A N} is injective for every AA-module MM.

Assume that the hypotheses of (i) are satisfied, and let us prove (ii) as well as the last assertion. Let MM be an AA-module. Since the AA-module N/yNN/yN is flat, we deduce from the exact sequence 0→N→yNN→N/yN→00 \to N \xrightarrow{y_N} N \to N/yN \to 0 the exact sequences

0→M⊗AN→uM⊗AN→M⊗A(N/yN)→0 0 \to M \otimes_A N \xrightarrow{u} M \otimes_A N \to M \otimes_A (N/yN) \to 0
0→Tor1A(M,N)→vTor1A(M,N)→0 0 \to \mathrm{Tor}_1^A(M, N) \xrightarrow{v} \mathrm{Tor}_1^A(M, N) \to 0

where u=1M⊗yNu = 1_M \otimes y_N and v=Tor1A(1M,yN)v = \mathrm{Tor}_1^A(1_M, y_N); it follows that the homothety of ratio yy is injective in M⊗ANM \otimes_A N, and bijective in Tor1A(M,N)\mathrm{Tor}_1^A(M, N). Assume moreover that the A-module MM is finitely generated; then the B-module Tor1A(M,N)\mathrm{Tor}_1^A(M, N) is finitely generated (A, X, p. 107, Prop. 6), hence is zero since yy belongs to mBm_B (Nakayama's lemma), which implies that the A-module NN is flat (A, X, p. 74, Theorem 2).

(ii) ⇒\Rightarrow (i): assume that the hypotheses of (ii) are satisfied. Consider the two exact sequences of B-modules

(1)

0→Ker(yN)⟶N→pIm(yN)→0 0 \to \mathrm{Ker}(y_N) \longrightarrow N \xrightarrow{p} \mathrm{Im}(y_N) \to 0

(2)

0→Im(yN)→iN⟶N/yN→0, 0 \to \mathrm{Im}(y_N) \xrightarrow{i} N \longrightarrow N/yN \to 0,

where pp and ii are the canonical homomorphisms. It follows that the homomorphism 1⊗p:κA⊗AN⟶κA⊗AIm(yN)1 \otimes p : \kappa_A \otimes_A N \longrightarrow \kappa_A \otimes_A \mathrm{Im}(y_N) is surjective, and (since NN is flat) that the kernel of the homomorphism 1⊗i:κA⊗AIm(yN)⟶κA⊗AN1 \otimes i : \kappa_A \otimes_A \mathrm{Im}(y_N) \longrightarrow \kappa_A \otimes_A N is isomorphic to Tor1A(κA,N/yN)\mathrm{Tor}_1^A(\kappa_A, N/yN). But the mapping (1⊗i)∘(1⊗p)(1 \otimes i) \circ (1 \otimes p), equal to yκA⊗ANy_{\kappa_A \otimes_A N}, is injective by hypothesis; it follows that 1⊗p1 \otimes p is bijective and 1⊗i1 \otimes i injective, and consequently that one has Tor1A(κA,N/yN)=0\mathrm{Tor}_1^A(\kappa_A, N/yN) = 0. It follows that the A-module N/yNN/yN is flat (III, § 5, No. 2, Theorem 1 and No. 4, Proposition 2).

Since NN and N/yNN/yN are flat over AA, the same is true of Im(yN)\mathrm{Im}(y_N) (exact sequence (2)). One then deduces from exact sequence (1) that κA⊗AKer(yN)\kappa_A \otimes_A \mathrm{Ker}(y_N) is isomorphic to the kernel of 1⊗p1 \otimes p, hence is zero; thus the homothety yNy_N is injective by Nakayama's lemma.

Proposition 10

Let ρ:A→B\rho : A \to B be a local homomorphism of noetherian local rings, NN a finitely generated B-module and y=(y1,…,ys)y = (y_1, \ldots, y_s) a sequence of elements of mBm_B. Let y\mathfrak{y} denote the ideal of BB generated by this sequence. The following conditions are equivalent:

(i) the A-module N/yNN/\mathfrak{y}N is flat and the sequence yy is N-regular; (ii) the A-module NN is flat and the sequence yy is (κA⊗AN)(\kappa_A \otimes_A N)-regular.

When these conditions are satisfied, for every A-module MM, the sequence yy is M⊗ANM \otimes_A N-regular.

Let us prove the equivalence of (i) and (ii) by induction on ss. The case s=0s = 0 being evident, suppose s⩾1s \geqslant 1; let y′y' denote the sequence (y1,…,ys−1)(y_1, \ldots, y_{s-1}) and y′\mathfrak{y}' the ideal of BB generated by it. By lemma 2 applied to the B-module N/y′NN/\mathfrak{y}'N and to the element ysy_s of BB, condition (i) is equivalent to

(i') the A-module N/y′NN/\mathfrak{y}'N is flat, the sequence y′y' is N-regular, and the homothety of ratio ysy_s in κA⊗A(N/y′N)=(κA⊗AN)/y′(κA⊗AN)\kappa_A \otimes_A (N/\mathfrak{y}'N) = (\kappa_A \otimes_A N)/\mathfrak{y}'(\kappa_A \otimes_A N) is injective.

This condition is equivalent to (ii) by the induction hypothesis.

The last assertion follows likewise by induction on ss from the last assertion of lemma 2.

Proposition 11

Let ρ:A→B\rho : A \to B be a local homomorphism of noetherian local rings, MM a finitely generated A-module and NN a finitely generated B-module; suppose that the A-module NN is flat.

a) Let (x1,…,xr)(x_1, \ldots, x_r) be a sequence of elements of mAm_A which is regular for the A-module MM, and (y1,…,ys)(y_1, \ldots, y_s) a sequence of elements of mBm_B which is regular for the B-module κA⊗AN\kappa_A \otimes_A N; then the sequence (y1,…,ys,ρ(x1),…,ρ(xr))(y_1, \ldots, y_s, \rho(x_1), \ldots, \rho(x_r)) of elements of mBm_B is regular for the B-module M⊗ANM \otimes_A N.

b) One has the equality

prof⁡B(M⊗AN)=prof⁡A(M)+prof⁡B(κA⊗AN) \operatorname{prof}_B(M \otimes_A N) = \operatorname{prof}_A(M) + \operatorname{prof}_B(\kappa_A \otimes_A N)

Let xx denote the ideal of A generated by the sequence xx and yy the ideal of B generated by yy. By Prop. 10, the sequence yy is M⊗ANM \otimes_A N-regular and N/yNN/yN is flat over A, so that the sequence ρ(x)=(ρ(x1),…,ρ(xr))\rho(x) = (\rho(x_1), \ldots, \rho(x_r)) is regular for M⊗A(N/yN)=(M⊗AN)/y(M⊗AN)M \otimes_A (N/yN) = (M \otimes_A N)/y(M \otimes_A N). This proves a).

To prove b), we may suppose that M and N are nonzero. By Nakayama's lemma, κA⊗AN\kappa_A \otimes_A N is likewise nonzero, so that prof⁡A(M)\operatorname{prof}_A(M) and prof⁡B(κA⊗AN)\operatorname{prof}_B(\kappa_A \otimes_A N) are finite (No. 4, Cor. 2 of Th. 2). Take maximal regular sequences xx and yy; then one has r=prof⁡A(M)r = \operatorname{prof}_A(M), s=prof⁡B(κA⊗AN)s = \operatorname{prof}_B(\kappa_A \otimes_A N), and there exists an injective a-linear mapping u:κA→M/xMu : \kappa_A \to M/xM and an injective b-linear mapping v:κB→κA⊗A(N/yN)v : \kappa_B \to \kappa_A \otimes_A (N/yN) (No. 4, Cor. 1 of Th. 2). Since N/yNN/yN is flat over A, the b-linear mapping (u⊗1N/yN)∘v(u \otimes 1_{N/yN}) \circ v of κB\kappa_B into (M/xM)⊗A(N/yN)=(M⊗AN)/(ρ(x)+y)(M⊗AN)(M/xM) \otimes_A (N/yN) = (M \otimes_A N)/(\rho(x) + y)(M \otimes_A N) is injective. This implies the equality prof⁡B(M⊗AN)=r+s\operatorname{prof}_B(M \otimes_A N) = r + s (loc. cit.).

Remark

Recall that, under the preceding assumptions,

dim⁡B(M⊗AN)=dim⁡A(M)+dim⁡B(κA⊗AN) \dim_B(M \otimes_A N) = \dim_A(M) + \dim_B(\kappa_A \otimes_A N)
(VIII, § 3, No. 4, Prop. 7).

Corollary

Let ρ:A→B\rho : A \to B be a local homomorphism of noetherian local rings making B a flat A-module. One has

prof⁡(B)=prof⁡(A)+prof⁡(κA⊗AB), \operatorname{prof}(B) = \operatorname{prof}(A) + \operatorname{prof}(\kappa_A \otimes_A B),
dim⁡(B)=dim⁡(A)+dim⁡(κA⊗AB). \dim(B) = \dim(A) + \dim(\kappa_A \otimes_A B).

In fact the depth (resp. the dimension) of the B-module κA⊗AB\kappa_A \otimes_A B is equal to the depth (resp. the dimension) of the ring κA⊗AB\kappa_A \otimes_A B by the cor. of Prop. 4 (resp. by VIII, § 1, No. 4).

7. Upper Bounds for Depth

Proposition 12

Let A be a noetherian local ring, M a nonzero finitely generated A-module and J an ideal of A distinct from A. One has the sequence of inequalities

prof⁡A(J;M)≤codim⁡(Supp⁡(M)∩V(J),Supp⁡(M))≤dim⁡(M)−dim⁡(M/JM) \operatorname{prof}_A(J; M) \leq \operatorname{codim}(\operatorname{Supp}(M) \cap V(J), \operatorname{Supp}(M)) \leq \dim(M) - \dim(M/JM)
≤[J/mAJ:κA]. \leq [J/m_A J : \kappa_A].

For every element pp of Supp⁡(M)∩V(J)\operatorname{Supp}(M) \cap V(J), prof⁡A(J;M)\operatorname{prof}_A(J; M) is less than or equal to dim⁡Ap(Mp)\dim_{A_p}(M_p) (No. 5, Prop. 8 and No. 4, Cor. 2 of Th. 2), that is to say (VIII, § 1, No. 4, Prop. 9)

Remark 1

Consider the chain of inequalities of Prop. 12.

a) In order that one have prof_A(J ; M) = [J/m_AJ : κ_A], it is necessary and sufficient that J can be generated by an M-regular sequence (No. 3, Cor. 2 of Th. 1 and A, X, p. 160, Cor. 1).

b) The equality dim(M) − dim(M/JM) = [J/m_AJ : κ_A] means that J can be generated by a secant sequence for M (VIII, § 3, No. 2).

c) If M is Macaulay, one has profA(J ; M) = dim(M) − dim(M/JM) (§ 2, No. 2, cor. of Prop. 2).

Lemma 3

Let A be a noetherian ring, p ⊂ p_1 ⊂ ... ⊂ p_{r−1} ⊂ q a saturated chain of length r of prime ideals of A, M a finitely generated Λ-module and n an integer. If the A-module Ext^n_{A_p}(κ(p), M_p) is nonzero, the same is true of Ext^{n+r}_{A_q}(κ(q), M_q).

It is obviously enough to treat the case r = 1; then, replacing A, M, p and q by A_q, M_q, pA_q and qA_q respectively, one is reduced to treating the case where A is local and q = m_A. Let x be an element of m_A − p. The A_p-module Ext^n_A(A/p, M) ⊗_A A_p is isomorphic to Ext^n_{A_p}(κ(p), M_p) (A, X, p. 111, Prop. 10 b)), and is therefore nonzero by assumption; a fortiori Ext^n_A(A/p, M) is nonzero. The exact sequence

0→A/p→xA/pA/p→A/(p+xA)→0 0 \to A/p \xrightarrow{x_{A/p}} A/p \to A/(p + xA) \to 0

yields an exact sequence of extension modules

ExtAn(A/p,M)→uExtAn(A/p,M)→ExtAn+1(A/(p+xA),M), \text{Ext}^n_A(A/p, M) \xrightarrow{u} \text{Ext}^n_A(A/p, M) \to \text{Ext}^{n+1}_A(A/(p + xA), M),

where u is the homothety with ratio x. By Nakayama's lemma, this is not surjective, hence the A-module Ext^{n+1}_A(A/(p + xA), M) is nonzero. But the only prime ideal of A containing p + xA is m_A, so the A-module A/(p + xA) is of finite length (VIII, § 3, No. 2, Lemma 2). If Ext^{n+1}_A(κ_A, M) were zero, one would deduce from this, by induction on the length of N, that Ext^{n+1}_A(N, M) = 0 for every A-module N of finite length. This contradiction completes the proof.

Proposition 13

Let A be a noetherian ring, M a finitely generated A-module, p and q two prime ideals of Supp(M) with p ⊂ q. One has

profAq(Mq)≤profAp(Mp)+dim⁡(Aq/pAq). \text{prof}_{A_q}(M_q) \leq \text{prof}_{A_p}(M_p) + \dim(A_q/pA_q).

More precisely, for every saturated chain of prime ideals p ⊂ p_1 ⊂ ... ⊂ p_{r−1} ⊂ q, one has \text{prof}_{A_q}(M_q) \leq \text{prof}_{A_p}(M_p) + r.

Put p = \text{prof}_{A_p}(M_p), and let us prove the second inequality. It is obvious if p = +∞; in the contrary case one has \text{Ext}^p_{A_p}(κ(p), M_p) ≠ 0, whence

ExtAqp+r(κ(q),Mq)≠0Ext_{A_q}^{p+r}(\kappa(q), M_q) \neq 0 by lemma 3, which implies profAq(Mq)≤p+rprof_{A_q}(M_q) \leq p + r. Since dim⁡(Aq/pAq)\dim(A_q/pA_q) is the least upper bound of the lengths of saturated chains of prime ideals with endpoints pp and qq, the first assertion is a consequence of the second.

Corollary 1

One has the inequality

dim⁡(Mq)−profAq(Mq)≥dim⁡(Mp)−profAp(Mp)≥0. \dim(M_q) - prof_{A_q}(M_q) \geq \dim(M_p) - prof_{A_p}(M_p) \geq 0 .

This follows from prop. 13 and the inequality dim⁡(Mq)≥dim⁡(Mp)+dim⁡(Aq/pAq)\dim(M_q) \geq \dim(M_p) + \dim(A_q/pA_q) (VIII, § 1, No. 4, prop. 9, a) and No. 3, prop. 7, b)).

Corollary 2

Let A be a local noetherian ring and M a finitely generated A-module. One has the inequality

profA(M)≤inf⁡p∈Ass⁡(M)dim⁡(A/p). prof_A(M) \leq \inf_{p \in \operatorname{Ass}(M)} \dim(A/p) .

Let pp be a prime ideal associated with MM; one has profAp(Mp)=0prof_{A_p}(M_p) = 0 (No. 1, remark 2). Prop. 13 applied to the ideals p⊂mAp \subset m_A yields the inequality profA(M)≤dim⁡(A/p)prof_A(M) \leq \dim(A/p), whence the corollary.

Remark 2

One has sup⁡p∈Ass⁡(M)dim⁡(A/p)=dim⁡(M)\sup_{p \in \operatorname{Ass}(M)} \dim(A/p) = \dim(M) (VIII, § 1, No. 4, remark 2), and one recovers the inequality prof(M)≤dim⁡(M)prof(M) \leq \dim(M) for M≠0M \neq 0 (No. 4, cor. 2 of th. 2).

8. Locally Integral Noetherian Rings; Normal Noetherian Rings

Let AA be a noetherian ring. Denote by (Yi)i∈I(Y_i)_{i \in I} the finite family (II, § 4, No. 2, prop. 10 and No. 3, cor. 7 of prop. 11) of connected components of Spec⁡(A)\operatorname{Spec}(A). By II, § 4, No. 3, prop. 15, for each ii there exists a unique idempotent element eie_i of AA such that Yi=V(ei)Y_i = V(e_i), and the canonical mapping of AA into the product of the rings A/AeiA/Ae_i is bijective. The quotient rings A/AeiA/Ae_i of AA are called the canonical components of AA. Put fi=1−eif_i = 1 - e_i. One has ∑i∈Ifi=1\sum_{i \in I} f_i = 1, and (fi)i∈I(f_i)_{i \in I} is an orthogonal family of non-zero idempotents of AA (loc. cit.). It follows that the image of fif_i in A/AejA/Ae_j is equal to 1 if j=ij = i and to 0 if j≠ij \neq i; one therefore deduces from the ring homomorphism A→∏jA/AejA \to \prod_j A/Ae_j a canonical isomorphism Afi→A/AeiA_{f_i} \to A/Ae_i.

By loc. cit., cor. 2 of prop. 14, the following conditions are equivalent: (i) the connected components of Spec⁡(A)\operatorname{Spec}(A) are irreducible; (ii) every prime (resp. maximal) ideal of AA belong to only one irreducible component of Spec⁡(A)\operatorname{Spec}(A); (iii) every prime (resp. maximal) ideal of AA contains only one minimal prime ideal; (iv) for every prime (resp. maximal) ideal pp of AA, the topological space Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible; (v) for every canonical component CC of AA, the topological space Spec⁡(C)\operatorname{Spec}(C) is irreducible.

(i) AA is reduced and the connected components of Spec(A)\mathrm{Spec}(A) are irreducible; (ii) for every prime (resp. maximal) ideal pp of AA, the ring ApA_p is an integral domain; (iii) the canonical components of AA are integral domains.

A noetherian ring is said to be locally integral if it satisfies the equivalent conditions (i) to (iii) above.

Suppose the ring AA locally integral; let uu be an isomorphism of AA onto a (finite) product ∏j∈JAj\prod_{j \in J} A_j of integral domains. There exists a bijection σ:J→I\sigma : J \to I such that the mapping of Spec(∏j∈JAj)\mathrm{Spec}(\prod_{j \in J} A_j) into Spec(A)\mathrm{Spec}(A) associated with uu defines a homeomorphism of Spec(Aj)\mathrm{Spec}(A_j) onto the connected component Yσ(j)Y_{\sigma(j)} of Spec(A)\mathrm{Spec}(A); one then deduces from uu an isomorphism of the canonical component A/Aeσ(j)A/Ae_{\sigma(j)} onto AjA_j.

Proposition 14

Let AA be a noetherian ring. The following conditions are equivalent:

(i) AA is reduced and integrally closed in its total ring of fractions ; (ii) AA is isomorphic to the product of a finite family of integrally closed rings ; (iii) the canonical components of AA are integrally closed ; (iv) for every prime ideal (resp. maximal ideal) pp of AA, the ring ApA_p is integrally closed.

The equivalence of (i) and (ii) follows from V, § 1, No. 2, Cor. 2 of Prop. 9, and that of (ii) and (iii) from the remarks preceding the proposition. Let pp be a prime ideal of AA; there exists a unique canonical component A′A' of AA such that pp belongs to the closed subset Spec(A′)\mathrm{Spec}(A') of Spec(A)\mathrm{Spec}(A), and one has a canonical isomorphism Ap→Ap′A′A_p \to A'_pA'. The equivalence of (iii) and (iv) therefore follows from loc. cit., No. 5, Cor. 1 and Cor. 3 of Prop. 16.

A ring AA is said to be normal if it is noetherian and satisfies the equivalent conditions (i) to (iv) of Proposition 14. A noetherian ring is integrally closed if and only if it is integral and normal. A normal local ring is integrally closed.

9. Depth and Connectedness

Lemma 4

Let AA be a noetherian ring, FF a closed subset of Spec(A)\mathrm{Spec}(A), UU the complementary open set, and u:M→Nu : M \to N a homomorphism of finitely generated AA-modules.

a) Suppose that up:Mp→Npu_p : M_p \to N_p is injective for every p∈Up \in U and that one has profF(M)⩾1\mathrm{prof}_F(M) \geqslant 1. Then uu is injective.

b) Suppose that up:Mp→Npu_p : M_p \to N_p is bijective for every p∈Up \in U and that one has profF(M)⩾2\mathrm{prof}_F(M) \geqslant 2 and profF(N)⩾1\mathrm{prof}_F(N) \geqslant 1. Then uu is bijective.

a) The hypotheses imply Supp(Ker u)⊂F\mathrm{Supp}(\mathrm{Ker}\,u) \subset F, hence HomA(Ker u,M)=0\mathrm{Hom}_A(\mathrm{Ker}\,u, M) = 0 (No. 5, Remark 1); therefore one has Ker u=0\mathrm{Ker}\,u = 0.

b) One already knows that uu is injective, and one has Supp(Coker u)⊂F\mathrm{Supp}(\mathrm{Coker}\,u) \subset F. By loc. cit., one has HomA(Coker u,N)=0\mathrm{Hom}_A(\mathrm{Coker}\,u, N) = 0 and ExtA1(Coker u,M)=0\mathrm{Ext}_A^1(\mathrm{Coker}\,u, M) = 0. From the exact sequence of extension modules

HomA(Coker u,N)→HomA(Coker u,Coker u)→ExtA1(Coker u,M) \mathrm{Hom}_A(\mathrm{Coker}\,u, N) \to \mathrm{Hom}_A(\mathrm{Coker}\,u, \mathrm{Coker}\,u) \to \mathrm{Ext}_A^1(\mathrm{Coker}\,u, M)

one deduces HomA(Coker u,Coker u)=0\mathrm{Hom}_A(\mathrm{Coker}\,u, \mathrm{Coker}\,u) = 0, which implies Coker u=0\mathrm{Coker}\,u = 0.

Remark 1

Let AA be a noetherian ring, FF a closed subset of Spec(A)\mathrm{Spec}(A), UU the complementary open set. In order that profF(A)⩾1\mathrm{prof}_F(A) \geqslant 1, it is necessary and sufficient that Ass(A)⊂U\mathrm{Ass}(A) \subset U (remark 2, No. 5). When this condition is satisfied, each irreducible component of Spec(A)\mathrm{Spec}(A) meets UU, so that UU is dense in Spec(A)\mathrm{Spec}(A).

Theorem 3 (Hartshorne)

Let AA be a noetherian ring, FF a closed subset of Spec(A)\mathrm{Spec}(A) and UU the complementary open set. Suppose that profF(A)⩾2\mathrm{prof}_F(A) \geqslant 2. Then, for every connected component YY of Spec(A)\mathrm{Spec}(A), the set Y∩U‾Y \cap \overline{U} is connected and dense in YY.

Suppose first that Spec(A)\mathrm{Spec}(A) is connected. By remark 1, UU is dense in Spec(A)\mathrm{Spec}(A), and it remains to prove that it is connected. Reasoning by contradiction, suppose that there are given two disjoint open sets U0U_0 and U1U_1 of Spec(A)\mathrm{Spec}(A), non-empty and with union UU. Since the set Ass(A)\mathrm{Ass}(A) is contained in UU by remark 1, it is the disjoint union of Ass(A)∩U0\mathrm{Ass}(A) \cap U_0 and Ass(A)∩U1\mathrm{Ass}(A) \cap U_1. By IV, § 1, No. 1, Prop. 4, there exist ideals J0J_0 and J1J_1 of AA such that Ass(Ji)=Ass(A)∩Ui\mathrm{Ass}(J_i) = \mathrm{Ass}(A) \cap U_i, Ass(A/Ji)=Ass(A)∩U1−i(i=0,1)\mathrm{Ass}(A/J_i) = \mathrm{Ass}(A) \cap U_{1-i} \quad (i = 0, 1). The complement of UiU_i in Spec(A)\mathrm{Spec}(A) contains Ass(A/Ji)\mathrm{Ass}(A/J_i) and Ass(J1−i)\mathrm{Ass}(J_{1-i}); since it is closed, it contains Supp(A/Ji)\mathrm{Supp}(A/J_i) and Supp(J1−i)\mathrm{Supp}(J_{1-i}). For p∈Uip \in U_i, one has thus (Ji)p=Ap(J_i)_p = A_p and (J1−i)p=0(J_{1-i})_p = 0; this implies in particular that J0J_0 and J1J_1 are distinct from AA. Let BB be the AA-module A/J0×A/J1A/J_0 \times A/J_1 and let u:A→Bu : A \to B be the canonical homomorphism. By what precedes, the homomorphism upu_p is bijective for every p∈Up \in U; moreover, one has Ass(B)⊂U0∪U1=U\mathrm{Ass}(B) \subset U_0 \cup U_1 = U, hence profF(B)⩾1\mathrm{prof}_F(B) \geqslant 1 by remark 1. Lemma 4 then implies that uu is bijective, which contradicts the connectedness of Spec(A)\mathrm{Spec}(A).

Let us treat the general case. Let JJ be an ideal of AA such that F=V(J)F = V(J) and let YY be a connected component of Spec(A)\mathrm{Spec}(A). By II, § 4, No. 3, Prop. 15, there exists an idempotent element ff of AA such that YY is identified with the subset Spec(Af)\mathrm{Spec}(A_f) of Spec(A)\mathrm{Spec}(A). Then Y∩FY \cap F is identified with V(Jf)V(J_f); one has profAf(Jf,Af)⩾profA(J;A)⩾2\mathrm{prof}_{A_f}(J_f, A_f) \geqslant \mathrm{prof}_A(J; A) \geqslant 2 by Prop. 6, a) of No. 3. It follows from the first part of the proof that Y∩U=Y−(Y∩F)Y \cap U = Y - (Y \cap F) is connected and dense in YY.

Corollary 1

The mapping which associates to each connected component of UU its closure in Spec(A)\mathrm{Spec}(A) is a bijection of the set of connected components of UU onto the set of connected components of Spec(A)\mathrm{Spec}(A).

Corollary 2

For every noetherian local ring B of depth ≥2\geq 2, the topological space Spec⁡(B)−{mB}\operatorname{Spec}(B) - \{ m_B \} is connected.

Corollary 3

Under the hypotheses of Theorem 3, suppose that Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible (resp. that ApA_p is an integral domain) for every p∈Up \in U; then Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible (resp. ApA_p is an integral domain) for every p∈Spec⁡(A)p \in \operatorname{Spec}(A).

Let (Yi)i∈I(Y_i)_{i \in I} be the (finite) family of irreducible components of Spec⁡(A)\operatorname{Spec}(A). Let p∈Up \in U; since Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible, pp contains only one minimal prime ideal of AA, and therefore belongs to only one of the YiY_i (II, § 4, No. 3, Cor. 2 of Prop. 14). The subsets Yi∩UY_i \cap U are closed subsets of UU, pairwise disjoint, non-empty since UU is dense in Spec⁡(A)\operatorname{Spec}(A), and irreducible by II, § 4, No. 1, Prop. 7; they are therefore the connected components of UU. Their closures YiY_i are the connected components of Spec⁡(A)\operatorname{Spec}(A) (Cor. 1). This proves that the connected components of Spec⁡(A)\operatorname{Spec}(A) are irreducible, and therefore that Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible for every pp (No. 8).

Suppose that AqA_q is an integral domain for every q∈Uq \in U. Let p∈Spec⁡(A)p \in \operatorname{Spec}(A). Since Spec⁡(Ap)\operatorname{Spec}(A_p) is irreducible, the nilradical of ApA_p is the unique minimal prime ideal of ApA_p; it therefore belongs to Ass⁡(Ap)\operatorname{Ass}(A_p) (IV, § 1, No. 3, Cor. 1 of Prop. 7), and consequently is equal to qApqA_p, where qq is a prime ideal associated with AA (IV, § 1, No. 2, Cor. of Prop. 5). One has q∈Uq \in U (Remark 1) and qAq∈Ass⁡(Aq)qA_q \in \operatorname{Ass}(A_q) (loc. cit.); since AqA_q is an integral domain, qq is zero, and therefore ApA_p is an integral domain.

Corollary 4

Let AA be a noetherian ring whose spectrum is connected. Suppose there exists an integer d≥1d \geq 1 such that one has prof⁡(Ap)≥2\operatorname{prof}(A_p) \geq 2 for every prime ideal pp of AA of height >d> d.

a) For every closed subset ZZ of Spec⁡(A)\operatorname{Spec}(A) of codimension >d> d, the space Spec⁡(A)−Z\operatorname{Spec}(A) - Z is connected.

b) Let YY and Y′Y' be irreducible components of Spec⁡(A)\operatorname{Spec}(A). There exists a sequence X1,X2,…,XnX_1, X_2, \ldots, X_n of irreducible components of Spec⁡(A)\operatorname{Spec}(A) such that one has X1=YX_1 = Y, Xn=Y′X_n = Y' and, for i=1,…,n−1i = 1, \ldots, n-1

codim⁡(Xi∩Xi+1,Spec⁡(A))≤d. \operatorname{codim}(X_i \cap X_{i+1}, \operatorname{Spec}(A)) \leq d .

Let Z⊂Spec⁡(A)Z \subset \operatorname{Spec}(A) be a closed subset of codimension >d> d. For every p∈Zp \in Z, one has dim⁡(Ap)>d\dim(A_p) > d, hence prof⁡(Ap)≥2\operatorname{prof}(A_p) \geq 2, which implies that prof⁡Z(A)\operatorname{prof}_Z(A) is ≥2\geq 2 (No. 5, Prop. 8) and therefore that Spec⁡(A)−Z\operatorname{Spec}(A) - Z is connected (Theorem 3).

Let us prove b). Denote by ZZ the union of the sets X′∩X′′X' \cap X'' where (X′,X′′)(X', X'') runs through the (finite) set of pairs of irreducible components of Spec⁡(A)\operatorname{Spec}(A) such that codim⁡(X′∩X′′,Spec⁡(A))>d\operatorname{codim}(X' \cap X'', \operatorname{Spec}(A)) > d. By a), the set Spec⁡(A)−Z\operatorname{Spec}(A) - Z is connected. All the irreducible components of Spec⁡(A)\operatorname{Spec}(A) meet Spec⁡(A)−Z\operatorname{Spec}(A) - Z; their traces on Spec⁡(A)−Z\operatorname{Spec}(A) - Z are the irreducible components of Spec⁡(A)−Z\operatorname{Spec}(A) - Z (II, § 4, No. 1, Prop. 7). Since Spec⁡(A)−Z\operatorname{Spec}(A) - Z is connected, there exists a sequence X1,…,XnX_1, \ldots, X_n of irreducible components of Spec⁡(A)\operatorname{Spec}(A) such that one has X1−Z=Y−ZX_1 - Z = Y - Z, Xn−Z=Y′−ZX_n - Z = Y' - Z and (Xi−Z)∩(Xi+1−Z)≠∅(X_i - Z) \cap (X_{i+1} - Z) \neq \varnothing for 1≤i≤n−11 \leq i \leq n-1; in other words one has X1=YX_1 = Y, Xn=ZX_n = Z and codim⁡(Xi∩Xi+1)≤d\operatorname{codim}(X_i \cap X_{i+1}) \leq d.

Remark 2

When A is a Macaulay ring, one can apply corollary with d=1d = 1 (§ 2, No. 5).

Example (Complete intersection formed by four coordinate planes of an affine space of dimension 4)

Let kk be a field. Let S denote the polynomial ring k[T1,T2,T3,T4]k[T_1, T_2, T_3, T_4]. Recall (VIII, § 2, No. 4, Theorem 3) that every maximal chain of prime ideals of S is of length 4. Let m denote the ideal of S generated by the TiT_i, a the ideal generated by T1T2T_1T_2 and T3T4T_3T_4, and pijp_{ij}, for 1≤i<j≤41 \leq i < j \leq 4, the ideal generated by TiT_i and TjT_j. The ideals pijp_{ij} are prime of height 2, their sum is the maximal ideal m, and one has a=p13∩p14∩p23∩p24a = p_{13} \cap p_{14} \cap p_{23} \cap p_{24}.

a) The ring A=S/aA = S/a is reduced; the irreducible components of Spec(A) are the sets Xij=V(pij/a)X_{ij} = V(p_{ij}/a) for i=1,2, j=3,4i = 1,2,\ j = 3,4, which are of dimension 2 and all contain the point m/am/a. In particular, Spec(A) is connected and of dimension 2. The intersection of two distinct components XijX_{ij} and XklX_{kl} is reduced to {m/a}\{m/a\} if {i,j}∩{k,l}=∅\{i,j\} \cap \{k,l\} = \varnothing, of dimension 1 otherwise. It follows that the conclusion of cor. 4 is satisfied with d=1d = 1 (we shall see later that A is a Macaulay ring, so that the assumption of corollary 4 is itself satisfied for d=1d = 1).

b) Let b denote the ideal of S generated by T1T2T_1T_2, T1T3T_1T_3, T2T4T_2T_4, T3T4T_3T_4, and B the ring S/bS/b. One has b=p14p23=p14∩p23b = p_{14}p_{23} = p_{14} \cap p_{23}. The ring B is reduced. Its spectrum is identified with the closed subset X14∪X23X_{14} \cup X_{23} of Spec(A); it has two irreducible components (of dimension 2) whose intersection is reduced to a point. The depth of B along this point is strictly positive because B is reduced, and less than 1 by Theorem 3, hence equal to 1.* Thus B is not a Macaulay ring. \

10. Depth and normality

Let A be a noetherian ring and pp a prime ideal of A. One has prof(Ap)≤ht(p)\mathrm{prof}(A_p) \leq \mathrm{ht}(p) (No. 4, cor. 2 of Theorem 2). If moreover A is reduced, the local ring ApA_p is reduced (II, § 2, No. 6, prop. 17), whence:

a) if ht(p)=0\mathrm{ht}(p) = 0, ApA_p is a field; b) if ht(p)≥1\mathrm{ht}(p) \geq 1, one has prof(Ap)≥1\mathrm{prof}(A_p) \geq 1 (No. 1, remark 3).

Conversely:

Proposition 15

Let A be a noetherian ring satisfying the following two conditions: (i) for every minimal prime ideal pp of AA, the ring ApA_p is reduced; (ii) for every prime ideal pp of A of height ≥1\geq 1, one has prof(Ap)≥1\mathrm{prof}(A_p) \geq 1. Then A is reduced.

Let nn denote the nilradical of A. For every minimal prime ideal pp of A, one has by (i) np=0n_p = 0, that is, p∉Supp(n)p \not\in \mathrm{Supp}(n) and afortioria fortiori p∉AssA(n)p \not\in \mathrm{Ass}_A(n). For every p∈Spec(A)p \in \mathrm{Spec}(A) of height ≥1\geq 1, one has by (ii) pAp∉AssAp(Ap)pA_p \not\in \mathrm{Ass}_{A_p}(A_p) and afortioria fortiori pAp∉AssAp(np)pA_p \not\in \mathrm{Ass}_{A_p}(n_p), whence p∉AssA(n)p \not\in \mathrm{Ass}_A(n) (IV, § 1, No. 2, prop. 5). Thus the set AssA(n)\mathrm{Ass}_A(n) is empty, which implies that nn is zero.

Proposition 16

Let AA be an integrally closed noetherian ring, JJ an ideal of AA of height ⩾2\geqslant 2, and MM a reflexive AA-module of finite type. One has prof⁡A(J;M)⩾2\operatorname{prof}_A(J; M) \geqslant 2.

Let us choose a finite-dimensional vector space VV over the field of fractions of AA and a lattice NN in VV isomorphic to MM (VII, § 4, No. 2, Remark 1). The prime ideals associated with V/NV/N, being of height 1 (loc. cit., Theorem 2), do not contain JJ; after Remark 2 of No. 1, this implies prof⁡A(J;V/N)⩾1\operatorname{prof}_A(J; V/N) \geqslant 1. On the other hand, the AA-module VV is divisible and torsion-free, hence injective (A, X, p. 17, Corollary 2 of Proposition 10), which implies prof⁡A(J;V)=+∞\operatorname{prof}_A(J; V) = +\infty. The inequality prof⁡A(J;N)⩾2\operatorname{prof}_A(J; N) \geqslant 2 then follows from Proposition 1 of No. 1.

Corollary

A noetherian local integrally closed ring of dimension ⩾2\geqslant 2 has depth ⩾2\geqslant 2*.

Let AA be a normal ring (No. 8) and p\mathfrak{p} a prime ideal of AA. Then: a) if ht⁡(p)=0\operatorname{ht}(\mathfrak{p}) = 0, ApA_{\mathfrak{p}} is a field; b) if ht⁡(p)=1\operatorname{ht}(\mathfrak{p}) = 1, ApA_{\mathfrak{p}} is a discrete valuation ring (VII, § 1, No. 7, Proposition 11); c) if ht⁡(p)⩾2\operatorname{ht}(\mathfrak{p}) \geqslant 2, one has prof⁡(Ap)⩾2\operatorname{prof}(A_{\mathfrak{p}}) \geqslant 2 (corollary to Proposition 16).

Conversely:

Theorem 4 (Serre)

Let AA be a noetherian ring satisfying the following conditions*: (i) for every minimal prime ideal p\mathfrak{p} of AA, the ring ApA_{\mathfrak{p}} is reduced; (ii) for every prime ideal p\mathfrak{p} of AA of height 1, the ring ApA_{\mathfrak{p}} is integrally closed; (iii) for every prime ideal p\mathfrak{p} of AA of height ⩾2\geqslant 2, one has prof⁡(Ap)⩾2\operatorname{prof}(A_{\mathfrak{p}}) \geqslant 2.

Then AA is normal.

It is a question of proving that the ring ApA_{\mathfrak{p}} is integrally closed for every p∈Spec⁡(A)\mathfrak{p} \in \operatorname{Spec}(A), which we shall do by induction on ht⁡(p)\operatorname{ht}(\mathfrak{p}). For ht⁡(p)⩽1\operatorname{ht}(\mathfrak{p}) \leqslant 1 this follows from assumptions (i) and (ii). Suppose therefore that one has ht⁡(p)⩾2\operatorname{ht}(\mathfrak{p}) \geqslant 2 and that AqA_q is integrally closed for every prime ideal qq of AA of height <ht⁡(p)< \operatorname{ht}(\mathfrak{p}). By assumption (iii), one has prof⁡(Ap)⩾2\operatorname{prof}(A_{\mathfrak{p}}) \geqslant 2. By the induction hypothesis and cor. 3 of th. 3 of No. 9, applied to the ring ApA_{\mathfrak{p}} and to the closed subset {pAp}\{\mathfrak{p}A_{\mathfrak{p}}\} of Spec⁡(Ap)\operatorname{Spec}(A_{\mathfrak{p}}), the ring ApA_{\mathfrak{p}} is an integral domain. Let KK be its field of fractions and let BB be a subring of KK, containing ApA_{\mathfrak{p}} and finite over ApA_{\mathfrak{p}}. It is a question of proving that BB is equal to ApA_{\mathfrak{p}}. Let us denote by ii the canonical injection of ApA_{\mathfrak{p}} into BB. Since BB is contained in KK, it is a torsion-free ApA_{\mathfrak{p}}-module, hence of depth ⩾1\geqslant 1. Moreover, for every prime ideal qq of ApA_{\mathfrak{p}} distinct from pAp\mathfrak{p}A_{\mathfrak{p}}, the homomorphism iq:(Ap)q→Bqi_q : (A_{\mathfrak{p}})_q \to B_q is bijective since AqA_q is integrally closed. By lemma 4 of No. 9, applied to the closed subset F={pAp}F = \{\mathfrak{p}A_{\mathfrak{p}}\} of Spec⁡(Ap)\operatorname{Spec}(A_{\mathfrak{p}}), the homomorphism ii is bijective, which completes the proof of the theorem.

Remark 1

A convenient form of th. 4 is the following: let AA be a noetherian ring, such that for every prime ideal pp of AA the ring ApA_p is integrally closed or of depth ⩾2\geqslant 2; then AA is normal. Let us indeed verify the assumptions of th. 4. Let p∈Spec(A)p \in \mathrm{Spec}(A). If ht(p)⩽1\mathrm{ht}(p) \leqslant 1, then one has prof(Ap)⩽1\mathrm{prof}(A_p) \leqslant 1, hence ApA_p is integrally closed. If ht(p)⩾2\mathrm{ht}(p) \geqslant 2, the ring ApA_p is either of depth ⩾2\geqslant 2, or integrally closed, which again implies prof(Ap)⩾2\mathrm{prof}(A_p) \geqslant 2 (No. 1, cor. 2 of prop. 1), whence the desired conclusion. One verifies analogously the following statement: let AA be a noetherian ring such that for every prime ideal pp of AA, the ring ApA_p is reduced or of depth ⩾1\geqslant 1; then AA is reduced.

Corollary 1

Let AA be a noetherian ring, FF a closed subset of Spec(A)\mathrm{Spec}(A), UU the complementary open set. Suppose that profF(A)\mathrm{prof}_F(A) is ⩾2\geqslant 2 (resp. ⩾1\geqslant 1) and that, for every p∈Up \in U, the ring ApA_p is integrally closed (resp. reduced). Then AA is normal* (resp. reduced).

For every p∈Fp \in F, one has prof(Ap)⩾profF(A)\mathrm{prof}(A_p) \geqslant \mathrm{prof}_F(A) (No. 5, Prop. 8); it is therefore enough to apply the preceding remark.

Corollary 2

Let ρ:A→B\rho : A \to B be a homomorphism of noetherian rings making BB into a flat AA-module.

a) *If BB is normal and faithfully flat over AA, AA is normal.

b) Suppose that AA is normal and that the ring κ(p)⊗AB\kappa(p) \otimes_A B is normal when pp is a minimal prime ideal of AA, and reduced when pp is a prime ideal of height 1. Then the ring BB is normal.

a) Suppose BB normal and faithfully flat over AA. Then ρ\rho is injective (I, § 3, No. 5, Prop. 9) and BB is reduced, hence AA is reduced. Let SS be the set of non-zero-divisors of AA. Since BB is flat over AA, ρ(S)\rho(S) consists of non-zero-divisors in BB, so that BB is integrally closed in ρ(S)−1B\rho(S)^{-1}B. Let xx be an element of S−1AS^{-1}A integral over AA; then the element x⊗1Bx \otimes 1_B of the ring S−1A⊗ABS^{-1}A \otimes_A B (which is identified with ρ(S)−1B\rho(S)^{-1}B) is integral over BB, hence belong to BB. Since BB is faithfully flat over AA, one has x∈Ax \in A (I, § 3, No. 5, Prop. 10, (ii)), and AA is normal.

b) Under the hypotheses of b), it is enough, by Remark 1, to prove that for every prime ideal qq of BB, the local ring BqB_q is normal or of depth ⩾2\geqslant 2. Let pp denote the prime ideal ρ−1(q)\rho^{-1}(q) of AA; the local homomorphism Ap→BqA_p \to B_q deduced from ρ\rho makes BqB_q into a faithfully flat ApA_p-module (I, § 3, No. 5, Prop. 9). Let us distinguish four cases :

1) ht(p)=0\mathrm{ht}(p) = 0. Then ApA_p is a field, equal to κ(p)\kappa(p); the ring Ap⊗ABA_p \otimes_A B is normal by assumption. The same is true of BqB_q, which is a ring of fractions thereof.

Remark 2

ht(p)=1\mathrm{ht}(p) = 1 and ht(q)⩽1\mathrm{ht}(q) \leqslant 1. Then ApA_p is a discrete valuation ring; let π\pi be a uniformizing element of ApA_p. Since BqB_q is faithfully flat over ApA_p, the element π1Bq\pi 1_{B_q} of BqB_q is not a zero divisor, so that the local ring Bq/πBqB_q/\pi B_q is of dimension 0 (VIII, § 3, No. 1, Cor. 2 of Prop. 1). It is reduced, since it is a ring of fractions of the reduced ring κ(p)⊗AB\kappa(p) \otimes_A B, and consequently it is a field. Therefore BqB_q is a discrete valuation ring (VI, § 1, No. 4, Prop. 2), hence integrally closed.

Remark 3

ht(p)=1\mathrm{ht}(p) = 1 and ht(q)≥2\mathrm{ht}(q) \geq 2. Then, by the relation

dim⁡(Bq)=dim⁡(Ap)+dim⁡(κ(p)⊗ABq) \dim(B_q) = \dim(A_p) + \dim(\kappa(p) \otimes_A B_q)
(VIII, § 3, No. 4, Cor. 1 of Prop. 7), the ring κ(p)⊗ABq\kappa(p) \otimes_A B_q is of dimension ≥1\geq 1. It is reduced by assumption, hence of depth ≥1\geq 1 (No. 1, Remark 3). We then have (No. 6, Cor. of Prop. 11)
prof(Bq)=prof(Ap)+prof(κ(p)⊗ABq)≥1+1=2. \mathrm{prof}(B_q) = \mathrm{prof}(A_p) + \mathrm{prof}(\kappa(p) \otimes_A B_q) \geq 1 + 1 = 2 .

Remark 4

ht(p)≥2\mathrm{ht}(p) \geq 2. Since ApA_p is of depth ≥2\geq 2 (Cor. of Prop. 16), the same is true of BqB_q by loc. cit.

Corollary 3

Let ρ:A→B\rho : A \to B be a homomorphism of noetherian rings. Assume that BB is a flat AA-module, that AA is normal, and that κ(p)⊗AB\kappa(p) \otimes_A B is normal for every p∈Spec(A)p \in \mathrm{Spec}(A). Then BB is normal.

Exercises

See the exercises for § 1.

Machine translation of the English, by gpt-5.4, against glossary version 34. Not checked by a person.