Bourbaki

Proposition 4

Tag 04Q4, permanent. Cite it as [Bourbaki, Tag 04Q4].

Commutative Algebra, chapter V, § 2. The lift of prime ideals

*Let A′A' be a ring, G\mathscr{G} a finite group operating on A′A', H\mathscr{H} a subgroup of G\mathscr{G}, AA and BB the rings of invariants of G\mathscr{G} and H\mathscr{H} respectively and p′p' a prime ideal of AA; let us write p=A∩p′p = A \cap p' and p(B)=B∩p′p(B) = B \cap p'.

(i) For H\mathscr{H} to be contained in the decomposition group Gz(p′)\mathscr{G}^z(p'), it is necessary and sufficient that p′p' be the only prime ideal of AA lying above p(B)p(B).

(ii) If H\mathscr{H} contains Gz(p′)\mathscr{G}^z(p'), the following conditions are satisfied: (a) The rings A/pA/p and B/p(B)B/p(B) have the same field of fractions. (b) The maximal ideal of the local ring Bp(B)B_{p(B)} is equal to pBp(B)pB_{p(B)}.

(iii) Suppose further that A′A' is an integral domain and that ⋂n≥0pnAp′=0\bigcap_{n \geq 0} p^nA'_p = 0; then conditions (a) and (b) & (ii) imply that Gz(p′)\mathscr{G}^z(p') leaves invariant the elements of BB.

(i) It follows from Theorem 2 (i) that the prime ideals of AA lying above p(B)p(B) are the ideals of the form σ.p′\sigma.p', where σ∈H\sigma \in \mathscr{H}; whence immediately (i).

(ii) We write S=A−pS = A - p; we know that the rings of invariants of G\mathscr{G} and H\mathscr{H} in S−1A′S^{-1}A' are respectively S−1AS^{-1}A and S−1BS^{-1}B (§ 1, no. 9, Proposition 23) and GZ(S−1p)=GZ(p′)\mathscr{G}^{\mathbf{Z}}(S^{-1}p) = \mathscr{G}^{\mathbf{Z}}(p') (Lemma 3); finally S−1p(B)=S−1p′∩S−1BS^{-1}p(B) = S^{-1}p' \cap S^{-1}B (Chapter II, § 2, no. 4), the local ring of the prime ideal S−1p(B)S^{-1}p(B) of the ring S−1BS^{-1}B is canonically isomorphic to Bp(B)B_{p(B)} and its residue field is isomorphic to the field of fractions of B/p(B)B/p(B) (Chapter II, § 2, no. 5, Proposition 11). We can therefore show (ii) restricting our attention to the case where pp is maximal. To establish (a) it will be sufficient to prove that

B=A+p(B) B = A + p(B)

for this will show that the fields A/pA/p and B/p(B)B/p(B) are canonically isomorphic. By Theorem 2 there is only a finite number of prime ideals of A′A' lying above pp and by Theorem 1 of no. 1 there is at least one prime ideal of A′A' lying above every prime ideal of BB; this implies that there is only a finite number of prime ideals of BB lying above pp; let njn_j (1≤j≤r1 \leq j \leq r) denote those of these ideals which are ≠p(B)\neq p(B). Let xx be an element of BB; as the ideals p(B)p(B) and njn_j are maximal (no. 1, Proposition 1), there exists y∈By \in B such that y≡xy \equiv x (mod. p(B)p(B)) and y∈njy \in n_j for 1≤j≤r1 \leq j \leq r (Chapter 11, § 1, no. 2, Proposition 5). Let y1=y,y2,…,yqy_1 = y, y_2, \ldots, y_q be distinct elements of the orbit of yy under G\mathscr{G}; clearly

z=y1+y2+⋯+yq∈A, z = y_1 + y_2 + \cdots + y_q \in A,

and to establish (3) it will be sufficient to show that yi∈p′y_i \in p' for i≥2i \geq 2, for then we shall deduce that z−y∈p′∩B=p(B)z - y \in p' \cap B = p(B), whence x∈A+p(B)x \in A + p(B) since x≡yx \equiv y (mod. p(B)p(B)). Then let i≥2i \geq 2 and σ∈G\sigma \in \mathscr{G} such that σ.y=yi\sigma.y = y_i; we show that σ−1.p′\sigma^{-1}.p' does not lie above p(B)p(B). For otherwise there would exist τ∈H\tau \in \mathscr{H} such that σ−1.p′=τ.p′\sigma^{-1}.p' = \tau.p' (Theorem 2 (i)), whence (τ−1σ−1).p′=p′(\tau^{-1}\sigma^{-1}).p' = p', in other words τ−1σ−1∈GZ⊂H\tau^{-1}\sigma^{-1} \in \mathscr{G}^{\mathbf{Z}} \subset \mathscr{H} by hypothesis, whence σ∈H\sigma \in \mathscr{H}; but as y∈By \in B and σ.y≠y\sigma.y \neq y, this is absurd. We conclude that σ−1.p′\sigma^{-1}.p' lies above one of the ideals njn_j and, as y∈njy \in n_j by construction, certainly y∈σ−1.p′y \in \sigma^{-1}.p' or yi=σ.y∈p′y_i = \sigma.y \in p'.

To prove (b) it will suffice to establish that p(B)p(B) is contained in the saturation qq of the ideal pBpB with respect to p(B)p(B) (Chapter II, § 2, no. 4, Proposition 10); as p(B)p(B) is contained in none of the njn_j (1≤j≤r1 \leq j \leq r), it will suffice even to prove that

p(B)⊂q∪n1∪⋯∪nr p(B) \subset q \cup n_1 \cup \cdots \cup n_r

by Chapter II, § 1, no. 1, Proposition 2. For this, we consider an element u∈p(B)u \in p(B) belonging to none of the njn_j (1≤j≤r1 \leq j \leq r) (Chapter 11, § 1, no. 1, Proposition 2); let u1=u,u2,…,umu_1 = u, u_2, \ldots, u_m be the distinct elements of the orbit of uu under G\mathscr{G}; we write w=u1u2…um,v=u2…umw = u_1u_2 \ldots u_m, v = u_2 \ldots u_m; clearly w∈Aw \in A; on the other hand, if τ∈H\tau \in \mathscr{H}, τ.u=u\tau.u = u and hence necessarily τ.ui≠u\tau.u_i \neq u for i≥2i \geq 2, which shows that τ.v=v\tau.v = v and hence v∈Bv \in B. It can be shown as in the proof of (a) that, if σ∈G\sigma \in \mathscr{G} is such that σ.u=ui\sigma.u = u_i where i≥2i \geq 2, σ−1.p′\sigma^{-1}.p' lies above one of the njn_j and, as u∉nju \notin n_j, also u∉σ−1.p′u \notin \sigma^{-1}.p', in other words ui∉p′u_i \notin p'. We conclude that v∉p′v \notin p' and therefore v∉p(B)v \notin p(B). On the other hand clearly w∈p′∩A=pw \in p' \cap A = p and the relation w=uvw = uv shows that uu is in the saturation of pBpB with respect to p(B)p(B) and hence establishes (4).

(iii) Suppose that A′A' is an integral domain, that ⋂n≥0pnAp′=0\bigcap_{n \geq 0} p^n A'_p = 0 and that conditions (a) and (b) of (ii) hold. With the same notation as in (ii), clearly S−1A′S^{-1}A' is an integral domain and S−1Ap′=Ap′S^{-1}A'_p = A'_p; it is therefore possible to replace A′A' and p′p' by S−1A′S^{-1}A' and S−1p′S^{-1}p', in other words, suppose also that the ideal pp is maximal. The hypotheses (a) and (b) then imply that

Bp(B)=A+pBp(B) \mathbf{B}_{p(B)} = A + p \mathbf{B}_{p(B)}

By induction on nn we deduce that Bp(B)=A+pnBp(B)\mathbf{B}_{p(B)} = A + p^n \mathbf{B}_{p(B)} for all n>0n > 0. Then let σ\sigma be an element of GZ\mathscr{G}^\mathbf{Z} and xx be an element of BB. For all n>0n > 0, there exists an∈Aa_n \in A such that x−an∈pnBp(B)⊂pnAp′x - a_n \in p^n \mathbf{B}_{p(B)} \subset p^n A'_p; as σ.an=a\sigma.a_n = a, and σ.p′=p′\sigma.p' = p', we deduce that σ.x−x∈pnAp′\sigma.x - x \in p^n A'_p. Since this relation holds for all nn, we conclude from the hypothesis that σ.x=x\sigma.x = x.

h a r k s

(3) If A′A' is an integral domain and Noetherian, the condition ⋂n≥1pnAp′=0\bigcap_{n \geq 1} p^n A'_p = 0 always holds (Chapter III, § 3, no. 2, Corollary to Proposition 5). It can be shown that this condition is also satisfied if A′A' is assumed to be an integral domain and AA to be Noetherian.

(4) If pp is not a maximal ideal of AA relation (3) does not necessarily hold under the hypotheses of (ii) and therefore A/pA/p and B/p(B)B/p(B) are not necessarily isomorphic even if we take H=GZ\mathscr{H} = \mathscr{G}^\mathbf{Z}, whence B=A′′B = A'' (Exercise 10).

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

Cited by

Cites

Transcribed from the printed volume. Not translated.