Bourbaki

Proposition 9

Tag 0DC9, permanent. Cite it as [Bourbaki, Tag 0DC9].

Commutative Algebra, chapter X, § 9. Modules dualisants

Let B be the integral closure of A, and c=A:Bc = A : B. Suppose that B is a finitely generated A-module. One has the inequality [B/c]≤2[B/A][B/c] \leq 2[B/A] in Z0(A)Z_0(A). For there to be equality, it is necessary and sufficient that A be a Gorenstein ring.

One has [B/c]=[B/A]+[A/c][B/c] = [B/A] + [A/c], so that the inequality under consideration is equivalent to [A/c]≤[B/A][A/c] \leq [B/A].

A) For every maximal ideal mm of A, the integral closure of AmA_m is BmB_m (V, § 1, No. 5, cor. 1), and one has cm=Am:Bmc_m = A_m : B_m. Moreover, by definition, one has [B/c]=∑mlong⁡Am(Bm/cm)[m][B/c] = \sum_m \operatorname{long}_{A_m}(B_m/c_m)[m] and [B/A]=∑mlong⁡Am(Bm/Am)[m][B/A] = \sum_m \operatorname{long}_{A_m}(B_m/A_m)[m]. This reduces us to proving the proposition when the ring A is local, which we shall henceforth assume. In this case the ordered group Z0(A)Z_0(A) identifies canonically with Z\mathbf{Z}, in such a way that the class of a module of finite length is its length. The ring B is semilocal and the B-module BΩB\Omega is free of rank 1 (II, § 5, No. 3, prop. 5).

B) If A is a Gorenstein ring, the A-module Ω\Omega is free of rank 1 (No. 4, cor. 1 of prop. 7); let us choose a generator ω\omega of Ω\Omega. One has Ω=Aω\Omega = A\omega and ΩB=cΩ=cω\Omega_B = c\Omega = c\omega, and consequently long⁡(A/c)=long⁡(Ω/ΩB)=long⁡(B/A)\operatorname{long}(A/c) = \operatorname{long}(\Omega/\Omega_B) = \operatorname{long}(B/A).

C) Suppose the residue field κA\kappa_A is infinite. For every maximal ideal nn of B, let L(n)L(n) denote the B/nB/n-vector space (of dimension 1) BΩ/nBΩB\Omega/nB\Omega, and pr⁡n\operatorname{pr}_n the canonical projection of ⨁nL(n)\bigoplus_n L(n) onto L(n)L(n). Let φ:Ω⟶⨁nL(n)\varphi : \Omega \longrightarrow \bigoplus_n L(n) be the restriction to Ω\Omega of the canonical homomorphism BΩ⟶⨁nL(n)B\Omega \longrightarrow \bigoplus_n L(n). The image of φ\varphi is a κA\kappa_A-vector subspace of ⨁nL(n)\bigoplus_n L(n); it is not contained in Ker⁡pr⁡n\operatorname{Ker} \operatorname{pr}_n, for otherwise one would have Ω⊂nBΩ\Omega \subset nB\Omega and consequently BΩ⊂nBΩB\Omega \subset nB\Omega, which is contradictory. Thus the image of φ\varphi is not contained in the union of the Ker⁡pr⁡n\operatorname{Ker} \operatorname{pr}_n (A, V, p. 40, lemma 1); there therefore exists an element ω\omega of Ω\Omega whose image in BΩ/nBΩB\Omega/nB\Omega is non-zero for every nn, which implies that ω\omega generates the B-module BΩB\Omega (II, § 3, No. 3, prop. 11).

Let a∈Aa \in A; if aωa\omega belong to ΩB\Omega_B, one has aBω⊂ΩBaB\omega \subset \Omega_B, hence aΩ⊂ΩBa\Omega \subset \Omega_B, which implies that a∈ca \in c. The mapping a↦aωa \mapsto a\omega therefore induces an injection of A/cA/c into Ω/ΩB\Omega/\Omega_B; consequently one has long⁡(A/c)≤long⁡(Ω/ΩB)=long⁡(B/A)\operatorname{long}(A/c) \leq \operatorname{long}(\Omega/\Omega_B) = \operatorname{long}(B/A).

If long⁡(A/c)=long⁡(B/A)\operatorname{long}(A/c) = \operatorname{long}(B/A), one has Aω+ΩB=ΩA\omega + \Omega_B = \Omega. One may suppose that the ideal cc is contained in mAm_A (otherwise AA is equal to BB, hence is a Gorenstein ring). Since ΩB=cΩ\Omega_B = c\Omega is contained in mAΩm_A\Omega, it follows from Nakayama's lemma that ω\omega generates Ω\Omega. Thus the AA-module Ω\Omega is monogenous, hence free of rank 1, which means that AA is a Gorenstein ring (No. 4, cor. 1 of prop. 7).

D) Let us treat the general case. Let us denote by A′A' the ring A[X]A[X], that is to say (IX, App., No. 2) the local ring of the polynomial ring A[X]A[X] at the prime ideal mAA[X]m_A A[X]; it is a flat, integral AA-algebra of dimension 1, whose residue field κA′\kappa_{A'} is identified with κA(X)\kappa_A(X) and whose field of fractions is identified with K(X)K(X) (loc. cit.). By the corollary to proposition 5 of No. 3, the A′A'-module A′⊗AΩA' \otimes_A \Omega is dualizing. Put B′=A′⊗ABB' = A' \otimes_A B; this is the integral closure of A′A' in K(X)K(X) (V, § 1, No. 3, proposition 13 and No. 5, proposition 16). The conductor c′=A′:B′c' = A' : B' is equal to cA′cA' (I, § 2, No. 10, formula (11)). For every AA-module MM of finite length, one has long⁡A′(A′⊗AM)=long⁡A(M)\operatorname{long}_{A'}(A' \otimes_A M) = \operatorname{long}_A(M): in fact, since the AA-algebra A′A' is flat, it is enough to prove this relation when MM is simple, that is to say isomorphic to κA\kappa_A; but in this case A′⊗AκAA' \otimes_A \kappa_A is identified with κA′\kappa_{A'}, whence our assertion. Thus one has

long⁡A(B/c)=long⁡A′(B′/c′)andlong⁡A(B/A)=long⁡A′(B′/A′). \operatorname{long}_A(B/c) = \operatorname{long}_{A'}(B'/c') \quad \text{and} \quad \operatorname{long}_A(B/A) = \operatorname{long}_{A'}(B'/A') .

The ring A′A' satisfies the hypotheses of the proposition, and its residue field is infinite. By part C) of the proof, one has long⁡A′(B′/c′)≤2long⁡A′(B′/A′)\operatorname{long}_{A'}(B'/c') \leq 2 \operatorname{long}_{A'}(B'/A'), and equality implies that A′A' is a Gorenstein ring; but this last condition implies that AA is a Gorenstein ring (§ 3, No. 8, corollary 1 to proposition 12).

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Nothing.

Cites

Nothing.

Translated with machine help, not yet reviewed by a person.