Let ρ:A→B be a homomorphism of noetherian rings, making B into a flat A-module. Assume that for every maximal ideal n of B, the ring κ(ρ−1(n))⊗AB is a Gorenstein ring. Let Ω be a dualizing A-module ; the B-module Ω(B) is dualizing.
Let n be a maximal ideal of B, and p its inverse image in A. The Ap-module Bn is flat, the Ap-module Ωp is dualizing, Ω(B)⊗BBn is identified with Ωp⊗ApBn and κAp⊗ApBn, which is identified with a ring of fractions of κ(p)⊗AB, is a Gorenstein ring. It is therefore enough to prove the proposition when ρ is a local homomorphism of local rings, which we shall henceforth assume.
Let us first treat the case where the rings A and B are artinian. Put C=B/mAB. Since B is flat over A, the B-module HomB(C,Ω(B)) is isomorphic to HomA(κA,Ω)⊗AB (I, § 2, n° 10, prop. 11), hence to HomA(κA,Ω)⊗κAC. One deduces a sequence of isomorphisms
HomB(κB,Ω(B))⟶HomC(κC,HomB(C,Ω(B)))⟶HomC(κC,HomA(κA,Ω)⊗κAC)
⟶HomA(κA,Ω)⊗κAHomC(κC,C).
The κA-vector space HomA(κA,Ω) is of dimension 1 since Ω is dualizing, and the same holds for the κC-vector space HomC(κC,C) since C is a Gorenstein ring; consequently the κB-vector space HomB(κB,Ω(B)) is of dimension 1.
Let M be a B-module of finite length; let us prove by induction on longB(M) that one has longB(HomB(M,Ω(B)))⩽longB(M). The assertion is clear if M=0, and it follows from the preceding if M=κB. Suppose longB(M)⩾2. There exists an exact sequence of B-modules
0→M′→M→κB→0 with
longB(M′)<longB(M). One deduces from this an exact sequence
0→HomB(κB,Ω(B))→HomB(M,Ω(B))→HomB(M′,Ω(B)), and one concludes by applying the induction hypothesis to
M′.
Let N be the kernel of the canonical surjection of κA⊗AB onto κB. Put m=longB(κA⊗AB); one has longB(N)=m−1. Consider the exact sequence of B-modules
0→HomB(κB,Ω(B))⟶HomB(κA⊗AB,Ω(B))⟶HomB(N,Ω(B)) ⟶ExtB1(κB,Ω(B))⟶ExtB1(κA⊗AB,Ω(B)).
The B-modules HomB(κA⊗AB,Ω(B)) and ExtB1(κA⊗AB,Ω(B)) are respectively isomorphic to HomA(κA,Ω)⊗AB and ExtA1(κA,Ω)⊗AB, that is to say to κA⊗AB and to 0. The lengths of the B-modules HomB(κB,Ω(B)) and HomB(κA⊗AB,Ω(B)) are 1 and m, and that of HomB(N,Ω(B))) is ⩽m−1; one deduces from this that the B-module ExtB1(κB,Ω(B)) is zero. By Prop. 6 of § 3, No. 3, the B-module Ω(B) is injective; consequently it is a dualizing module (No. 1, Example 1).
Let us pass to the general case. Put C=κA⊗AB; by hypothesis this is a Gorenstein ring, hence a Macaulay ring (§3,n°7, prop. 10). By prop. 1 of No. 1, A is a Macaulay ring, and the A-module Ω is Macaulay. Hence B is a Macaulay ring, and the B-module Ω(B) is Macaulay (§2,n°7, cor. 1 of prop. 9). Put r=dim(A), s=dim(C). There exists a sequence (x1,…,xr) of elements of mA regular for the A-modules A and Ω, and a sequence (y1,…,ys) of elements of mB regular for the B-module C; let x denote the ideal of A and g the ideal of B generated by them respectively. The sequence (y1,…,ys,ρ(x1),…,ρ(xr)) is regular for the B-modules B and Ω(B) (§1,n°6, prop. 11), and the A-module B/g is flat (loc. cit., prop. 10). Put A′=A/x, B′=B/(xB+g) and let ρ′:A′→B′ denote the homomorphism deduced from ρ by passing to quotients. The rings A′ and B′ are artinian, the A′-module B′ is flat, the ring κA′⊗A′B′, which is identified with C/g, is a Gorenstein ring (§3,n°7, example 2) and the A′-module Ω(A′) is dualizing (No. 2, prop. 4). By the first part of the proof, the B′-module Ω(B′) is dualizing. It then follows from loc. cit. that the B-module Ω(B) is dualizing.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.
Nothing.