Bourbaki

2. Primary decomposition

1. Primary s bmodules

Proposition 1

Let AA be a Noetherian ring and MM an AA-module. The following conditions are equivalent: (a) Ass⁡(M)\operatorname{Ass}(M) is reduced to a single element.

(b) M≠0M \neq 0 and every homothety of MM is either injective or almost nilpotent (§1§ 1, no. 4). If these conditions are fulfilled and pp is the set of a∈Aa \in A such that the homothety aMa_M is almost nilpotent, then Ass⁡(M)={p}\operatorname{Ass}(M) = \{p\}.

This follows immediately from §1§ 1, no. 4, Corollary to Proposition 9 and no. 1, Corollary 2 to Proposition 2.

Definition 1

Let AA be a Noetherian ring, NN an AA-module and QQ a submodule of NN. If the module M=N/QM = N/Q satisfies the conditions of Proposition 1, QQ is called p-primary with respect to NN (or in NN).

When there is no ambiguity, we shall simply say that QQ is "p-primary" or "primary"; clearly for every submodule N′≠QN' \neq Q of NN containing QQ, QQ is p-primary in N′N'.

Definition 1 applies in particular to the case N=AN = A; the submodules of NN are then the ideals of AA and hence an ideal qq of AA is called primary if Ass⁡(A/q)\operatorname{Ass}(A/q) has a single element or, what amounts to the same, if A≠qA \neq q and every divisor of zero in the ring A/qA/q is nilpotent. If qq is p-primary, it follows from Definition 1 that pp is the radical (Chapter II, §2§ 2, no. 6) of the ideal qq.

Remark

Let qq be a p-primary submodule of an AA-module NN. If N/QN/Q is finitely generated, there exists an integer k>0k > 0 such that pkN⊂Qp^k N \subset Q by §1§ 1, no. 4, Proposition 9.

Examples

Example 1

If pp is a prime ideal of AA, pp is p-primary (§1§ 1, no. 1, Proposition 1).

Example 2

Let qq be an ideal of AA such that there exists a single prime ideal mm (necessarily maximal) containing qq; then, if MM is an AA-module such that qM≠MqM \neq M, qMqM is m-primary with respect to MM. For every element of Ass⁡(M/qM)\operatorname{Ass}(M/qM) contains qq, hence is equal to mm and Ass⁡(M/qM)≠0\operatorname{Ass}(M/qM) \neq 0 (§1§ 1, no. 1, Corollary 1 to Proposition 2). In particular qq is an m-primary ideal in AA.

Example 3

Let mm be a maximal ideal of AA; the m-primary ideals are then the ideals qq of AA for which there exists an integer n≥1n \geq 1 such that mn⊂q⊂mm^n \subset q \subset m. For if mn⊂q⊂mm^n \subset q \subset m, mm is the only prime ideal containing qq (Chapter 11, §1§ 1, no. 1, Corollary to Proposition 1) and the conclusion follows from Example 2; conversely, if qq is m-primary, mm is the radical of qq and there therefore exists n≥1n \geq 1 such that mn⊂qm^n \subset q (Chapter II, §2§ 2, no. 6, Proposition 15).

Example 4

In a principal ideal domain AA, the primary ideals are (0)(0) and the ideals of the form ApnAp^n, where pp is an extremal element and n≥1n \geq 1; this follows immediately from Example 3.

Example 5

The powers of any prime ideal are not necessarily primary ideals (Exercise 1). On the other hand, there exist primary ideals which are not powers of prime ideals (Exercise 1).

Proposition 2

Let M be a module over a Noetherian ring A, p a prime ideal of A and (Qi)i∈I(\mathscr{Q}_i)_{i \in I} a non-empty finite family of submodules of M which are p-primary with respect to M. Then ⋂i∈IQi\bigcap_{i \in I} \mathscr{Q}_i is p-primary with respect to M.

M/(⋂i∈IQi)M / (\bigcap_{i \in I} \mathscr{Q}_i) is isomorphic to a submodule ≠0\neq 0 of the direct sum ⨁i∈I(M/Qi)\bigoplus_{i \in I} (M / \mathscr{Q}_i). Now

Ass⁡(⨁i∈I(M/Qi))=⋃i∈IAss⁡(M/Qi)={p} \operatorname{Ass}\left( \bigoplus_{i \in I} (M / \mathscr{Q}_i) \right) = \bigcup_{i \in I} \operatorname{Ass}(M / \mathscr{Q}_i) = \{p\}

(§ 1, no. 1, Corollary 1 to Proposition 3). Hence Ass⁡(M/(⋂i∈IQi))={p}\operatorname{Ass}\left(M / (\bigcap_{i \in I} \mathscr{Q}_i)\right) = \{p\} (§ 1, no. 1, Proposition 3 and Corollary 1 to Proposition 2).

Proposition 3

Let A be a Noetherian ring, S a multiplicative subset of A, p a prime ideal of A, M an A-module, N a submodule of M and i=iASi = i_A^S the canonical mapping of M to S−1MS^{-1}M.

(i) Suppose that p∩S≠∅p \cap S \neq \varnothing. If N is p-primary with respect to M, then S−1N=S−1MS^{-1}N = S^{-1}M.

(ii) Suppose that p∩S=∅p \cap S = \varnothing. For N to be p-primary with respect to M, it is necessary and sufficient that N be of the form i1(N′)i_1(N'), where N′N' is a sub-S−1AS^{-1}A-module of S−1MS^{-1}M which is (S−1p)(S^{-1}p)-primary with respect to S−1MS^{-1}M; then N′=S−1NN' = S^{-1}N.

(i) If p∩S≠∅p \cap S \neq \varnothing and N is p-primary with respect to M, then

Ass⁡S−1A(S−1(M/N))=∅ \operatorname{Ass}_{S^{-1}A}(S^{-1}(M/N)) = \varnothing

(§ 1, no. 2, Corollary to Proposition 5) and hence S−1(M/N)=0S^{-1}(M/N) = 0 (§ 1, no. 1, Corollary 1 to Proposition 2), whence S−1M/S−1N=0S^{-1}M/S^{-1}N = 0.

(ii) Suppose that p∩S=∅p \cap S = \varnothing. If N is p-primary with respect to M, then Ass⁡S−1A(S−1(M/N))={S−1p}\operatorname{Ass}_{S^{-1}A}(S^{-1}(M/N)) = \{S^{-1}p\} (§ 1, no. 2, Corollary to Proposition 5) and hence the submodule N′=S−1NN' = S^{-1}N of S−1MS^{-1}M is (S−1p)(S^{-1}p)-primary; moreover, if s∈Ss \in S and m∈Mm \in M are such that sm∈Nsm \in N, then m∈Nm \in N, for the homothety with ratio s in M/N is injective, whence N=i1(N′)N = i_1(N') (Chapter II, § 2, no. 4, Proposition 10). Conversely, let N′N' be a submodule of S−1MS^{-1}M which is (S−1p)(S^{-1}p)-primary with respect to S−1MS^{-1}M; let us write N=i1(N′)N = i_1(N'); then N′=S−1NN' = S^{-1}N (Chapter II, § 2, no. 4, Proposition 10) and Ass⁡S−1A(S−1(M/N))=Ass⁡S−1A((S−1M)/N′)={S−1p}\operatorname{Ass}_{S^{-1}A}(S^{-1}(M/N)) = \operatorname{Ass}_{S^{-1}A}((S^{-1}M)/N') = \{S^{-1}p\}. As the canonical mapping M/N→S(M/N)M/N \to S(M/N) is injective, no prime ideal of A associated with M/N meets S (§ 1, no. 2, Proposition 6); it follows that Ass⁡(M/N)={p}\operatorname{Ass}(M/N) = \{p\} (§ 1, no. 2, Corollary to Proposition 5), so that N is p-primary with respect to M.

2. The existence of a primary decomposition

Definition 2

Let AA be a Noetherian ring, MM an AA-module and NN a submodule of MM. A finite family (Qi)i∈I(\mathscr{Q}_i)_{i \in I} of submodules of MM which are primary with respect to MM and such that N=⋂i∈IQiN = \bigcap_{i \in I} \mathscr{Q}_i is called a primary decomposition of NN in MM.

Example

Let us take A=ZA = \mathbf{Z}, M=ZM = \mathbf{Z}, N=nZN = n\mathbf{Z} for some integer n>0n > 0. If n=p1α1⋯pkαkn = p_1^{\alpha_1} \cdots p_k^{\alpha_k} is the decomposition of nn into prime factors,

nZ=(p1α1Z)∩…∩(pkαkZ) n\mathbf{Z} = (p_1^{\alpha_1}\mathbf{Z}) \cap \ldots \cap (p_k^{\alpha_k}\mathbf{Z})
is a primary decomposition of nZn\mathbf{Z} in Z\mathbf{Z} by Example 4 of no. 1.

By an abuse of language, the relation N=⋂i∈IQiN = \bigcap_{i \in I} \mathscr{Q}_i is called a primary decomposition of NN in MM. It amounts to the same to say that {0}=⋂i∈I(Qi/N)\{0\} = \bigcap_{i \in I} (\mathscr{Q}_i/N) is a primary decomposition of {0}\{0\} in M/NM/N. If (Qi)i∈I(\mathscr{Q}_i)_{i \in I} is a primary decomposition of NN in MM, the canonical mapping from M/NM/N to ⨁i∈I(M/Qi)\bigoplus_{i \in I} (M/\mathscr{Q}_i) is injective. Conversely let NN be a submodule of MM and ff an injective homomorphism from M/NM/N to a finite direct sum P=⨁i∈IPiP = \bigoplus_{i \in I} P_i, where each set Ass⁡(Pi)\operatorname{Ass}(P_i) is reduced to a single element pip_i; let fif_i be the homomorphism M/N→PiM/N \to P_i obtained by taking the composition off with the projection P→PiP \to P_i, and let Qi/NQ_i/N be the kernel off.; then the QiQ_i distinct from MM are primary with respect to MM (no. 1, Definition 1) and N=⋂i∈IQiN = \bigcap_{i \in I} Q_i. Moreover, Ass⁡(M/N)⊂⋃i∈I{pi}\operatorname{Ass}(M/N) \subset \bigcup_{i \in I} \{p_i\} by virtue of § 1, no. 1, Proposition 3.

Theorem 1

Let MM be a finitely generated module over a Noetherian ring and let NN be a submodule of MM. There exists a primary decomposition of NN of the form

N=⋂p∈Ass⁡(M/N)Q(p) N = \bigcap_{p \in \operatorname{Ass}(M/N)} Q(p)
where, for all p∈Ass⁡(M/N)p \in \operatorname{Ass}(M/N), Q(p)Q(p) is pp-primary with respect to MM.

We may replace MM by M/NM/N and therefore suppose that N=0N = 0. By § 1, no. 4, Corollary to Theorem 2, Ass⁡(M)\operatorname{Ass}(M) is finite; by § 1, no. 1, Proposition 4, there exists, for each p∈Ass⁡(M)p \in \operatorname{Ass}(M), a submodule Q(p)Q(p) of MM such that Ass⁡(M/Q(p))={p}\operatorname{Ass}(M/Q(p)) = \{p\} and Ass⁡(Q(p))=Ass⁡(M)−{p}\operatorname{Ass}(Q(p)) = \operatorname{Ass}(M) - \{p\}. Let us write P=⋂p∈Ass⁡(M)Q(p)P = \bigcap_{p \in \operatorname{Ass}(M)} Q(p); for all p∈Ass⁡(M)p \in \operatorname{Ass}(M), Ass⁡(P)⊂Ass⁡(Q(p))\operatorname{Ass}(P) \subset \operatorname{Ass}(Q(p)) and hence Ass⁡(P)=∅\operatorname{Ass}(P) = \varnothing, which implies P=0P = 0 (§ 1, no. 1, Corollary 1 to Proposition 2) and therefore proves the theorem.

3. Uniqueness properties in the primary decomposition

Definition 3

Let MM be a module over a Noetherian ring and NN a submodule of MM. A primary decomposition N=⋂i∈IQiN = \bigcap_{i \in I} Q_i of NN in MM is called reduced if the following conditions are fulfilled:

(a) there exists no index i∈Ii \in I such that ⋂j≠iQj⊂Qi;\bigcap_{j \neq i} Q_j \subset Q_i;

(b) if Ass⁡(M/Qi)={pi}\operatorname{Ass}(M/Q_i) = \{p_i\}, the pip_i (i ∈I\in I) are distinct.

From every primary decomposition N=⋂i∈IQiN = \bigcap_{i \in I} Q_i of NN in MM a reduced primary decomposition of MM in NN can be deduced as follows: let JJ be a minimal element of the set of subsets I′I' of II such that N=⋂i∈I′QiN = \bigcap_{i \in I'} Q_i. Clearly (Qi)i∈J(Q_i)_{i \in J} satisfies condition (a). Then let Φ\Phi be the set of pip_i for i∈Ji \in J; for all p∈Φp \in \Phi, let H(p)H(p) be the set of i∈Ji \in J such that pi=pp_i = p and let Q(p)=⋂i∈H(p)QiQ(p) = \bigcap_{i \in H(p)} Q_i; it follows from Proposition 2 of no. 1 that Q(p)Q(p) is pp-primary with respect to MM; further N=⋂p∈ΦQ(p)N = \bigcap_{p \in \Phi} Q(p) and the family Q((p))p∈ΦQ((p))_{p \in \Phi} is therefore a reduced primary decomposition of NN in MM.

We shall see that the primary decomposition defined in the proof of Theorem 1 of no. 2 is reduced; this follows from the following proposition:

Proposition 4

Let* MM *be a module over a Noetherian ring, N a submodule of M, N=⋂i∈IQiN = \bigcap_{i \in I} Q_i a primary decomposition of N in M and, for all i∈Ii \in I, let {pi}=Ass⁡(M/Qi)\{p_i\} = \operatorname{Ass}(M/Q_i). For this decomposition to be reduced, it is necessary and sufficient that the pip_i be distinct and belong to Ass⁡(M/N)\operatorname{Ass}(M/N); *then

(2)

Ass⁡(M/N)=⋃i∈I{pi} \operatorname{Ass}(M/N) = \bigcup_{i \in I} \{p_i\}

(3)

Ass⁡(Qi/N)=⋃j≠i{pj}for all i∈I. \operatorname{Ass}(Q_i/N) = \bigcup_{j \neq i} \{p_j\} \quad \text{for all } i \in I.

If the condition of the statement is fulfilled, N=⋂j≠iQjN = \bigcap_{j \neq i} Q_j cannot hold, for we would deduce that Ass⁡(M/N)⊂⋃j≠i{pj}\operatorname{Ass}(M/N) \subset \bigcup_{j \neq i} \{p_j\} (§ 1, no. 1, Corollary 2 to Proposition 3) contrary to the hypothesis; the primary decomposition (Qi)i∈I(Q_i)_{i \in I} of NN is then certainly reduced. Conversely, Ass⁡(M/N)⊂⋃i∈I{pi}\operatorname{Ass}(M/N) \subset \bigcup_{i \in I} \{p_i\} always holds (§ 1, no. 1, Corollary 2 to Proposition 3); on the other hand, for all i∈Ii \in I, let us write Pi=⋂jQjP_i = \bigcap_j Q_j; then Pi∩Qi=NP_i \cap Q_i = N and Pi≠NP_i \neq N if (Qi)i∈I(Q_i)_{i \in I} is reduced, hence Pi/NP_i/N is non-zero and is isomorphic to the submodule (Pi+Qi)/Qi(P_i + Q_i)/Q_i of M/QiM/Q_i, whence {pi}=Ass⁡(Pi/N)\{p_i\} = \operatorname{Ass}(P_i/N) (§ 1, no. 1, Proposition 3 and Corollary 1 to Proposition 2); as Pi/N⊂M/N,pi∈Ass⁡(M/N)P_i/N \subset M/N, p_i \in \operatorname{Ass}(M/N), which completes the proof of the necessity of the condition in the statement and formula (2). Finally, as N=⋂j≠i(Qj∩Qi),Ass⁡(Qi/N)⊂⋃j≠iAss⁡(Qi/(Qj∩Qi))N = \bigcap_{j \neq i} (Q_j \cap Q_i), \operatorname{Ass}(Q_i/N) \subset \bigcup_{j \neq i} \operatorname{Ass}(Q_i/(Q_j \cap Q_i)) (§ 1, no. 1, Corollary 2 to Proposition 3); but Qi/(Qj∩Qi)Q_i/(Q_j \cap Q_i) is isomorphic to the submodule (Qi+Qj)/Qj(Q_i + Q_j)/Q_j of M/QjM/Q_j, hence Ass⁡(Qi/(Qj∩Qi))⊂{pj}\operatorname{Ass}(Q_i/(Q_j \cap Q_i)) \subset \{p_j\} and

Ass⁡(Qi/N)⊂⋃j≠i{pj}; \operatorname{Ass}(Q_i/N) \subset \bigcup_{j \neq i} \{p_j\};

Corollary

Let AA be a Noetherian ring, MM an AA-module, NN a submodule of MM and (Qi)i∈I(Q_i)_{i \in I} a primary decomposition of NN in MM. Then Card(I)≥Card(Ass(M/N))\mathrm{Card}(I) \geq \mathrm{Card}(\mathrm{Ass}(M/N)); for (Qi)i∈I(Q_i)_{i \in I} to be a reduced primary decomposition, it is necessary and sufficient that Card(I)=Card(Ass(M/N))\mathrm{Card}(I) = \mathrm{Card}(\mathrm{Ass}(M/N)).

It follows from the remarks preceding Proposition 4 that there exists a reduced primary decomposition (Rj)j∈J(R_j)_{j \in J} of NN in MM such that Card(J)≤Card(I)\mathrm{Card}(J) \leq \mathrm{Card}(I); the first assertion then follows from the second and the latter is a consequence of Proposition 4.

Proposition 5

Let AA be a Noetherian ring, MM an AA-module, NN a submodule of MM, N=⋂i∈IQiN = \bigcap_{i \in I} Q_i a reduced primary decomposition of NN in MM and, for all i∈Ii \in I, let {pi}=Ass(M/Qi)\{p_i\} = \mathrm{Ass}(M/Q_i). If pip_i is a minimal element of Ass(M/N)\mathrm{Ass}(M/N), QiQ_i is equal to the saturation on NN with respect to pip_i (Chapter 11, § 2, no. 4) (cf. Exercise 2).

We can obviously restrict our attention to the case where N=0N = 0, replacing if need be MM by M/NM/N. If pip_i is minimal in Ass(M)\mathrm{Ass}(M), the set of elements of Ass(M)\mathrm{Ass}(M) which do not meet A−piA - p_i reduces to pip_i; the proposition then follows from formula (3) above and § 1, no. 2, Proposition 6, the kernel of the canonical mapping M→M0M \to M_0, being equal to the saturation of 0 with respect to pip_i (Chapter 11, § 2, no. 4).

Remark

The prime ideals pi∈Ass(M/N)p_i \in \mathrm{Ass}(M/N) which are not minimal elements of this set are sometimes called the immersed prime ideals associated with M/NM/N; if M/NM/N is finitely generated, for p0∈Ass(M/N)p_0 \in \mathrm{Ass}(M/N) to be immersed, it is necessary and sufficient that V(p0)V(p_0) be not an irreducible component of Supp(M/N)\mathrm{Supp}(M/N) (Chapter II, § 4, no. 3, Corollary 2 to Proposition 14); if (Q(p))p∈Ass(M/N)(Q(p))_{p \in \mathrm{Ass}(M/N)} and (Q′(p))p∈Ass(M/N)(Q'(p))_{p \in \mathrm{Ass}(M/N)} are two reduced primary decompositions of NN in MM, it may be that Q′(p0)≠Q(p0)Q'(p_0) \neq Q(p_0) (Exercise 24 (c)); a canonical reduced primary decomposition of NN in MM may always be defined by imposing supplementary condition on the primary submodules which appear in it (Exercise 4).

4. The localization of a primary decomposition

Given a submodule NN of a module MM over a Noetherian ring AA, to simplify we shall denote by DI(M/N)D_I(M/N), in this no., the set of reduced primary decompositions of NN in MM whose indexing set is II (equipotent to Ass(M/N)\mathrm{Ass}(M/N)).

Proposition 6

Let AA be a Noetherian ring, MM an AA-module, NN a submodule of MM and I=Ass(M/N)I = \mathrm{Ass}(M/N). Let SS be a multiplicative subset of AA and JJ the subset of II consisting of the indices i such that S∩pi=∅S \cap p_i = \varnothing. Let N′N' be the saturation of NN with respect to SS in MM. Then:

(i) If (Qi)i∈I(Q_i)_{i \in I} is an element of DI(M/N)D_I(M/N), the family (Qi)i∈J(Q_i)_{i \in J} is an element of DJ(M/N′)D_J(M/N') and the family (S−1Qi)i∈J(S^{-1}Q_i)_{i \in J} is an element of DJ(S−1M/S−1N)D_J(S^{-1}M/S^{-1}N).

(ii) The mapping (Qi)i∈J→(S−1Qi)i∈J(Q_i)_{i \in J} \to (S^{-1}Q_i)_{i \in J} is a bijection of DJ(M/N′)D_J(M/N') onto DJ(S−1M/S−1N)D_J(S^{-1}M/S^{-1}N).

(iii) If (Qi)i∈J(Q_i)_{i \in J} is an element of DJ(M/N′)D_J(M/N') and (Ri)i∈I(R_i)_{i \in I} an element of DI(M/N)D_I(M/N), the family (Ti)i∈I(T_i)_{i \in I} such that Ti=QiT_i = Q_i for i∈Ji \in J and Ti=RiT_i = R_i for i∈I−Ji \in I - J is an element of DI(M/N)D_I(M/N).

(i) We know (no. 1, Proposition 3) that for i∈J,S−1Qii \in J, S^{-1}Q_i is primary for S−1piS^{-1}p_i and that for i∈I−J,S−1Qi=S−1Mi \in I - J, S^{-1}Q_i = S^{-1}M; as S−1N=⋂S−1QiS^{-1}N = \bigcap S^{-1}Q_i (Chapter II, § 2, no. 4), then also S−1N=⋂i∈JS−1QiS^{-1}N = \bigcap_{i \in J} S^{-1}Q_i. The S−1piS^{-1}p_i for i∈Ji \in J are distinct and their set is Ass⁡(S−1M/S−1N)\operatorname{Ass}(S^{-1}M/S^{-1}N) (§1§ 1, no. 2, Corollary to Proposition 5); then (Proposition 4) (S−1Qi)i∈J(S^{-1}Q_i)_{i \in J} is a reduced primary decomposition of S−1NS^{-1}N. Moreover, Qi=(iMS)−1(S−1Qi)Q_i = (i_M^S)^{-1}(S^{-1}Q_i) (no. 1 Proposition 3), hence N′=(iMS)−1(S−1N)=⋂i∈JQiN' = (i_M^S)^{-1}(S^{-1}N) = \bigcap_{i \in J} Q_i and (Qi)i∈J(Q_i)_{i \in J} is obviously a reduced primary decomposition of N′N' in MM.

(ii) As S−1N′=S−1NS^{-1}N' = S^{-1}N, we may replace NN by N′N', that is suppose that J=IJ = I. Let (Pi)i∈I(P_i)_{i \in I} be a reduced primary decomposition of S−1NS^{-1}N in S−1MS^{-1}M and let us write Qi=(iMS)−1(Pi)Q_i = (i_M^S)^{-1}(P_i); it follows from no. 1, Proposition 3 that QiQ_i is primary for pip_i (i∈Ii \in I) and (Qi)i∈I(Q_i)_{i \in I} is then a reduced primary decomposition of NN in MM by virtue of no. 3, Corollary to Proposition 4. Finally, as, for all i∈Ii \in I and every submodule Qi′Q'_i of which MM is pip_i-primary with respect to M,Qi′=(iMS)−1(S−1Qi′)M, Q'_i = (i_M^S)^{-1}(S^{-1}Q'_i) by virtue of no. 1, Proposition 3 and the hypothesis J=IJ = I, we see that two mappings DI(M/N)→DI(S−1M/S−1N)D_I(M/N) \to D_I(S^{-1}M/S^{-1}N) and DI(S−1M/S−1N)→DI(M/N)D_I(S^{-1}M/S^{-1}N) \to D_I(M/N) have been defined whose compositions are the identities on DI(M/N)D_I(M/N) and DI(S−1M/S−1N)D_I(S^{-1}M/S^{-1}N), which proves (ii).

(iii) By virtue of (i), N′=⋂i∈JRiN' = \bigcap_{i \in J} R_i, whence

N=N′∩⋂i∈I−JRi=(⋂i∈JQi)∩(⋂i∈I−JRi) N = N' \cap \bigcap_{i \in I - J} R_i = \left( \bigcap_{i \in J} Q_i \right) \cap \left( \bigcap_{i \in I - J} R_i \right)

and it follows immediately from no. 3, Corollary to Proposition 4 that this primary decomposition is reduced.

Corollary

The mappings

DI(M/N)→DJ(M/N′)andDI(M/N)→DJ(S−1M/S−1N) D_I(M/N) \to D_J(M/N') \quad \text{and} \quad D_I(M/N) \to D_J(S^{-1}M/S^{-1}N)

defined in Proposition 6 (i) are surjective.

Proposition 6 (iii) shows that the mapping DI(M/N)→DJ(M/N′)D_I(M/N) \to D_J(M/N') is surjective and Proposition 6 (ii) then shows that the mapping

DI(M/N)→DJ(S−1M/S−1N) D_I(M/N) \to D_J(S^{-1}M/S^{-1}N)
is surjective.

5. Rings and modules of finite length

If an AA-module MM is of finite length, we shall denote this length by long⁡A(M)\operatorname{long}_A(M) or long⁡(M)\operatorname{long}(M). Recall that every Artinian ring is Noetherian (Algebra, Chapter VIII, § 6, no. 5, Corollary 3 to Proposition 12) and that every finitely generated module over an Artinian ring is of finite length (loc. cit., Corollary 1 to Proposition 12). Moreover, every Artinian integral domain is a field (Algebra, Chapter VIII, § 6, no. 4, Proposition 9).

Proposition 7

Let MM be a finitely generated module over a Noetherian ring AA. The following properties are equivalent:* (a) MM is of finite length. (b) Every ideal p∈Ass⁡(M)p \in \operatorname{Ass}(M) is a maximal ideal of AA. (c) *Every ideal p∈Supp⁡(M)p \in \operatorname{Supp}(M) is a maximal ideal of AA.

Let (Mi)0≤i≤n(M_i)_{0 \leq i \leq n} be a composition series of MM such that, for 0≤i≤n−10 \leq i \leq n - 1, Mi/Mi+1M_i/M_{i+1} is isomorphic to A/piA/p_i, where pip_i is prime (§1§ 1, no. 4, Theorem 1). If MM is of finite length, so is each of the AA-modules A/piA/p_i, which implies that each of the rings A/piA/p_i is Artinian; but as A/piA/p_i is an integral domain, it is therefore a field, in other words pip_i is maximal; we conclude that (a) implies (b) (§1§ 1, no. 4, Theorem 2). Condition (b) implies (c) by §1§ 1, no. 3, Proposition 7. Finally, if all the ideals of Supp⁡(M)\operatorname{Supp}(M) are maximal, so are the pip_i (§1§ 1, no. 4, Theorem 2), hence the A/piA/p_i are simple AA-modules and MM is of finite length, which completes the proof.

Corollary 1

For every module of finite length MM over a Noetherian Zing AA, Ass⁡(M)=Supp⁡(M)\operatorname{Ass}(M) = \operatorname{Supp}(M).

Every element of Supp⁡(M)\operatorname{Supp}(M) is then minimal in Supp⁡(M)\operatorname{Supp}(M) and the conclusion follows from §1§ 1, no. 3, Corollary 1 to Proposition 7.

Corollary 2

Let MM be a finitely generated module over a Noetherian ring AA and pp a prime ideal of AA. For MpM_p to be a non-zero ApA_p-module of finite length, it is necessary and sufficient that pp be a minimal element of Ass⁡(M)\operatorname{Ass}(M).

By §1§ 1, no. 2, Corollary to Proposition 5, Ass⁡Ap(Mp)\operatorname{Ass}_{A_p}(M_p) is the set of ideals qpq_p, where qq runs through the set of ideals of Ass⁡(M)\operatorname{Ass}(M) which are contained in pp. On the other hand, ppp_p is the unique maximal ideal of ApA_p; by Proposition 7, for MpM_p to be an ApA_p-module of finite length, it is necessary and sufficient that no element of Ass⁡(M)\operatorname{Ass}(M) be strictly contained in pp. On the other hand, for Mp≠0M_p \neq 0, it is necessary and sufficient by definition that p∈Supp⁡(M)p \in \operatorname{Supp}(M) (Chapter 11, § 4, no. 4), that is that pp contain an element of Ass⁡(M)\operatorname{Ass}(M) (§1§ 1, no. 3, Proposition 7). This proves the corollary.

Remark (1). Let MM be a finitely generated module over a Noetherian ring AA; let (M1),…(M_1), \ldots be a composition series of MM such that, for 0≤i≤n−10 \leq i \leq n - 1, Mi/Mi+1M_i / M_{i+1} is isomorphic to A/piA/p_i, where pip_i is a prime ideal of AA (§1§ 1, no. 4, Theorem 1). If pp is a minimal element of Ass⁡(M)\operatorname{Ass}(M), the length long⁡Ap(Mp)\operatorname{long}_{A_p}(M_p) is equal to the number of indices ii such that pi=pp_i = p. For the (Mi)p(M_i)_p form a composition series of MM, and (Mi)p/(Mi+1)p(M_i)_p / (M_{i+1})_p is isomorphic to (A/pi)p(A/p_i)_p and hence to {0}\{0\} if pi≠pp_i \neq p (since pp is minimal in the set of pip_i by §1§ 1, no. 4, Theorem 2) and to (A/p)p(A/p)_p which is a field, if pi=pp_i = p.

Proposition 8

Let MM be a module of finite length over a Noetherian ring AA.

(i) There only exists a single primary decomposition of {0}\{0\} with respect to MM indexed by Ass⁡(M)\operatorname{Ass}(M) (necessarily reduced) — let {0}=⋂p∈Ass⁡(M)Q(p)\{0\} = \bigcap_{p \in \operatorname{Ass}(M)} Q(p) be this decomposition, where Q(p)Q(p) is pp-primary with respect to MM.

(ii) There exists an integer n0n_0 such that, for all n≥n0n \geq n_0 and all p∈Ass⁡(M)p \in \operatorname{Ass}(M), Q(p)=pnMQ(p) = p^n M.

(iii) For all p∈Ass⁡(M)p \in \operatorname{Ass}(M), the canonical mapping of MM to MpM_p is surjective and its kernel is Q(p)Q(p).

(iv) The canonical injection of MM into ⨁p∈Ass⁡(M)(M/Q(p))\bigoplus_{p \in \operatorname{Ass}(M)} (M/Q(p)) is bijective.

As every element p∈Ass⁡(M)p \in \operatorname{Ass}(M) is minimal in Ass⁡(M)\operatorname{Ass}(M) (Proposition 7), assertion (i) follows from no. 3, Proposition 5. As MM is finitely generated, there exists n0n_0 such that pnM⊂Q(p)p^n M \subset Q(p) for all p∈Ass⁡(M)p \in \operatorname{Ass}(M) and all n≥n0n \geq n_0 (no. 1, Remark); but as pp is a maximal ideal, pnMp^n M is pp-primary with respect to MM (no. 1, Examples 2 and 3) and, as ⋂p∈Ass⁡(M)pnM={0}\bigcap_{p \in \operatorname{Ass}(M)} p^n M = \{0\}, it follows from (i) that necessarily pnM=Q(p)p^n M = Q(p) for all p∈Ass⁡(M)p \in \operatorname{Ass}(M); whence (ii). As the pnp^n, for p∈Ass⁡(M)p \in \operatorname{Ass}(M), are relatively prime in pairs (Chapter 11, § 1, no. 2, Proposition 3), the canonical mapping M→⨁p∈Ass⁡(M)(M/pnM)M \to \bigoplus_{p \in \operatorname{Ass}(M)} (M/p^n M) is surjective (Chapter 11, § 1, no. 2, Proposition 6), whence (iv). Then Ass⁡(Q(p))=Ass⁡(M)−{p}\operatorname{Ass}(Q(p)) = \operatorname{Ass}(M) - \{p\} and Ass⁡(M/Q(p))={p}\operatorname{Ass}(M/Q(p)) = \{p\} (no. 3, Proposition 4); as the elements of Ass⁡(M)\operatorname{Ass}(M) are maximal ideals, pp is the only element of Ass⁡(M)\operatorname{Ass}(M) which does not meet A−pA - p; Q(p)Q(p) is therefore the kernel of the canonical mapping j:M→Mpj : M \to M_p (§1§ 1, no. 2, Proposition 6). If s∈A−ps \in A - p, the homothety of M/Q(p)M/Q(p) with ratio ss is injective by virtue of the relation Ass⁡(M/Q(p))={p}\operatorname{Ass}(M/Q(p)) = \{p\} (no. 1, Proposition 1); since M/Q(p)M/Q(p) is Artinian, this homothety is bijective (Algebra, Chapter VIII, § 1, no. 2, Lemma 3). The canonical mapping M→M/Q(p)M \to M/Q(p) is then written f∘jf \circ j, where f:Mp→M/Q(p)f : M_p \to M/Q(p) is an AA-homomorphism (Chapter II, § 2, no. 2, Proposition 3); as Ker⁡(j)=Ker⁡(f∘j)=Q(p)\operatorname{Ker}(j) = \operatorname{Ker}(f \circ j) = Q(p), ff is injective; we conclude that jj is surjective and ff bijective.

Corollary

If M is a module of finite length over a Noetherian ring A, then

long⁡A(M)=∑p∈Ass⁡(M)long⁡Ap(Mp). \operatorname{long}_A(M) = \sum_{p \in \operatorname{Ass}(M)} \operatorname{long}_{A_p}(M_p).

This will follow from Proposition 8 (iv) if we prove that

long⁡A(M/Q(p))=long⁡Ap(Mp). \operatorname{long}_A(M/\mathscr{Q}(p)) = \operatorname{long}_{A_p}(M_p).

Now, it follows from Proposition 1 of no. 1 that for all s∈A−ps \in A - p the homothety with ratio s on M/Q(p)M/\mathscr{Q}(p) is injective; the homothety with ratio s on every submodule R of M/Q(p)M/\mathscr{Q}(p) is therefore injective and, as R is Artinian, it is bijective (Algebra, Chapter VIII, § 2, no. 2, Lemma 3); we conclude that the sub-A-modules of M/Q(p)M/\mathscr{Q}(p) are the images under the bijection f:M,→M/Q(p)f : M, \to M/\mathscr{Q}(p) of the sub-A,-modules of M, (Chapter 11, § 2, no. 3), whence our assertion.

Proposition 9

Let A be a Noetherian ring. The following conditions are equivalent :* (a) A is Artinian. (b) All the prime ideals a\mathfrak{a} in A are maximal ideals. (c) *All the elements of Ass(A) are maximal ideals.

If these conditions are fulfilled, A has only a finite number of prime ideals, which are all maximal and associated with the A-module A; further, A is a semi-local ring and its Jacobson radical is nilpotent.

To say that A is Artinian is equivalent to saying that A is an A-module of finite length; hence (a) and (c) are equivalent by Proposition 7. Clearly (b) implies (c). Finally, (a) implies (b) since every Artinian integral domain is a field. The properties (a), (b) and (c) are therefore equivalent.

Suppose they hold. As every prime ideal of A belongs to Supp(A) and every element of Supp(A) contains an element of Ass(A) (§\S 1, no. 3, Proposition 7), it follows from (c) that Ass(A) is the set of all prime ideals of A; then A has only a finite number of prime ideals, all of them maximal and associated with the A-module A. This obviously implies that A is semi-local; finally, we know that the Jacobson radical of an Artinian ring is nilpotent (Algebra, Chapter VIII, § 6, no. 4, Theorem 3).

Remark (2) The conditions of Proposition 9 for a Noetherian ring A imply that the spectrum of A is finite and discrete, every point of Spec(A) being therefore closed (Chapter 11, § 4, no. 3, Corollary 6 to Proposition 11). Conversely, for a Noetherian ring A, to say that every point a\mathfrak{a} of Spec(A) is closed means that every prime ideal of A is maximal (loc. cit..) and hence this condition is equivalent to that of Proposition 9.

Corollary 1

Every Artinian ring A is isomorphic to the direct composition of a finite family of Artinian local rings.

It follows from Proposition 9 and Proposition 8 (iii) and (iv) that, if (mi)1≤i≤n(\mathfrak{m}_i)_{1 \leq i \leq n} is the family of maximal ideals of AA, the canonical mapping A→∏iAmiA \to \prod_i A_{\mathfrak{m}_i} is bijective.

Remark (3). This corollary can also be deduced from the fact that Spec⁡(A)\operatorname{Spec}(A) is finite and discrete and Chapter 11, § 4, no. 3, Proposition 15.

Corollary 2

Let AA be a Noetherian ring and mm an ideal of AA. The following conditions are equivalent:

(a) AA is a semi-local ring and mm is a defining ideal of AA. (b) AA is a Zariski ring with the mm-adic topology and A/mA/m is Artinian.

If (a) holds, AA is a Zariski ring with the mm-adic topology (Chapter III, § 3, no. 3, Example 3); further, as by hypothesis mm contains a power of the Jacobson radical rr of AA, every prime ideal of AA which contains mm also contains rr (Chapter II, § 1, no. 1, Proposition 1); it is therefore maximal, since rr is a finite intersection of maximal ideals (\emph{loc. cit.}); Proposition 9 then shows that A/mA/m is Artinian. Conversely, if (b) holds, every maximal ideal pp of AA contains the Jacobson radical of AA and hence contains mm (Chapter III, § 3, no. 3, Proposition 6); as A/mA/m is Artinian, the ideals p/mp/m are finite in number (Proposition 9) and hence AA has only a finite number of maximal ideals, which implies that it is semi-local.

Corollary 3

Let A,A′A, A' be two rings and hh a homomorphism from AA to A′A'. Suppose that AA is semi-local and Noetherian and that A′A' is a finitely generated AA-module. Then the ring A′A' is semi-local and Noetherian; if mm is a defining ideal of AA, mA′mA' is a defining ideal of A′A'.

We know that A′A' is a Zariski ring with the mA′mA'-adic topology (Chapter III, § 3, no. 3, Proposition 7). As A/mA/m is Artinian (Corollary 2) and A′/mA′A'/mA' is a finitely generated (A/m)(A/m)-module, A′/mA′A'/mA' is an Artinian ring, hence A′A' is semi-local and mA′mA' is a defining ideal of A′A' (Corollary 2).

Corollary 4

Let AA be a complete semi-local Noetherian ring, mm a defining ideal of AA, EE a finitely generated AA-module and (Fn)(F_n) a decreasing sequence of submodules of EE such that ⋂nFn=0\bigcap_n F_n = 0. Then, for all p>0p > 0, there exists n>0n > 0 such that Fn⊂mpEF_n \subset m^p E.

As AA is a Zariski ring, EE is Hausdorff and the FnF_n are closed under the mm-adic topology. On the other hand, EE is complete (Chapter III, § 2, no. 12, Corollary 1 to Proposition 16). Finally, E/mpEE/m^p E is a finitely generated module over the ring A/mpA/m^p, which is Artinian (Corollary 2); then E/mpEE/m^p E is an Artinian (A/mp)(A/m^p)-module and hence an Artinian AA-module. The corollary then follows from Chapter III, § 2, no. 7, Proposition 8.

Corollary 5

In a complete semi-local Noetherian ring every decreasing sequence ϕ\phi ideals whose intersection is 0 is a filter base which converges to 0.

It is sufficient to apply Corollary 4 to the A-module A\mathbf{A}.

Proposition 10

Let AA be a Noetherian ring and p1,…,pnp_1, \ldots, p_n the prime ideals associated with the A-module AA, where pi≠pjp_i \neq p_j for i≠ji \neq j.

(i) The set S=⋂i=1n(A−pi)S = \bigcap_{i=1}^n (A - p_i) is the set of elements which are not divisors of 0 in AA.

(ii) If all the pip_i are minimal elements of Ass⁡(A)\operatorname{Ass}(A), the total ring of fractions S−1AS^{-1}A of AA is Artinian.

(iii) If the ring AA is reduced, all the pip_i are minimal elements of Ass⁡(A)\operatorname{Ass}(A) (and therefore are the minimal elements of Spec⁡(A)\operatorname{Spec}(A)) and each of the ApiA_{p_i} is a field; for each index ii, the canonical homomorphism S−1A→ApiS^{-1}A \to A_{p_i} (Chapter II, § 2, no. 1, Corollary 1 to Proposition 2) is surjective and its kernel is S−1piS^{-1}p_i; finally the canonical homomorphism from S−1AS^{-1}A to ∏i=1n(S−1A/S−1pi)\prod_{i=1}^n (S^{-1}A/S^{-1}p_i) is bijective.

The fact that SS is the set of elements which are not divisors of 0 in A\mathbf{A} has already been seen (§ 1, no. 1, Corollary 3 to Proposition 2). The prime ideals of S−1AS^{-1}A are of the form S−1pS^{-1}p, where pp is a prime ideal of AA contained in ⋃i=1npi\bigcup_{i=1}^n p_i (Chapter 11, § 2, no. 5, Proposition 10) that is contained in one of the pip_i (Chapter II, § 1, no. 1, Proposition 2). If pip_i is a minimal element of Ass⁡(A)\operatorname{Ass}(A), it is a minimal element of Spec⁡(A)\operatorname{Spec}(A) (§ 1, no. 3, Corollary to Proposition 7); if each of the pip_i is a minimal element of Ass⁡(A)\operatorname{Ass}(A), we then see that the prime ideals of S−1AS^{-1}A are the S−1piS^{-1}p_i and they are therefore all maximal, which proves that S−1AS^{-1}A is Artinian (Proposition 9).

Suppose finally that the ring AA is reduced. Then ⋂i=1npi={0}\bigcap_{i=1}^n p_i = \{0\} (§ 1, no. 3, Corollary 2 to Proposition 7). We deduce that {0}=⋂i=1npi\{0\} = \bigcap_{i=1}^n p_i is a reduced primary decomposition of the ideal {0}\{0\} (no. 3, Corollary to Proposition 4); in particular, none of the pip_i can contain a pjp_j of index j≠ij \neq i and therefore the pip_i are all minimal elements of Ass⁡(A)\operatorname{Ass}(A). The ring S−1AS^{-1}A is then Artinian by (ii). The S−1piS^{-1}p_i are prime ideals associated with the S−1AS^{-1}A-module S−1AS^{-1}A (§ 1, no. 2, Corollary to Proposition 5) and {0}=S−1(⋂i=1npi)=⋂i=1nS−1pi\{0\} = S^{-1} \left( \bigcap_{i=1}^n p_i \right) = \bigcap_{i=1}^n S^{-1}p_i (Chapter II, § 2, no. 4); as the S−1piS^{-1}p_i are distinct, (S−1pi)1≤i≤n(S^{-1}p_i)_{1 \leq i \leq n} is a reduced primary decomposition of {0}\{0\} in S−1AS^{-1}A (no. 3, Corollary to Proposition 4). Proposition 8 then shows that the canonical homomorphismg,: S−1A→(S−1A)piS^{-1}A \to (S^{-1}A)_{p_i} is surjective and has kernel S−1piS^{-1}p_i and the canonical homomorphism S−1A→∏i=1n(S−1A/S−1pi)S^{-1}A \to \prod_{i=1}^n (S^{-1}A/S^{-1}p_i) is bijective. We know moreover that the canonical homomorphism S−1A→AS^{-1}A \to A, is composed of g, and an isomorphism (S−1A)S−1pi→A(S^{-1}A)_{S^{-1}p_i} \to A, (Chapter II, § 2, no. 3, Proposition 7). Finally, it follows from Proposition 8 that (S−1A)S−1pi(S^{-1}A)_{S^{-1}p_i} is isomorphic to S−1A/S−1piS^{-1}A/S^{-1}p_i and hence is a field since S−1piS^{-1}p_i is a maximal ideal.

6. Primary decomposition and extension of scalars

In this no., A and B will denote two rings and we shall consider a ring homomorphism ρ:A→B\rho : A \to B which makes BB into an A-algebra; recall that, for every B-module F, ρ∗(F)\rho_*(F) is the commutative group F with the A-module structure defined by a.y=ρ(a)ya.y = \rho(a)y for all a∈A,y∈Fa \in A, y \in F.

Lemma 1

Let A be a Noetherian ring, pp a prime ideal of A, E an A-module whose annihilator contains a power of pp and such that Ass⁡(E)={p}\operatorname{Ass}(E) = \{p\} and F a B-module such that ρ∗(F)\rho_*(F) is a flat A-module. The condition P∈Ass⁡B(E⊗AF)\mathfrak{P} \in \operatorname{Ass}_B(E \otimes_A F) then implies ρ−1(P)=p\rho^{-1}(\mathfrak{P}) = p.

If n is such that pnE=0p^nE = 0, then pnB⊂Ann⁡(E⊗AF)p^nB \subset \operatorname{Ann}(E \otimes_A F), whence pnB⊂Pp^nB \subset \mathfrak{P}, which implies pn⊂ρ1(P)p^n \subset \rho_1(\mathfrak{P}) and therefore p⊂ρ1(P)p \subset \rho_1(\mathfrak{P}) since ρ1(P)\rho_1(\mathfrak{P}) is prime. Moreover, if a∈A−pa \in A - p, the homothety h with ratio a on E is injective (§1§ 1, no. 1, Corollary 2 to Proposition 2); as h⊗1Fh \otimes 1_F is the homothety h′h' with ratio ρ(a)\rho(a) on E⊗AFE \otimes_A F and ρ∗(F)\rho_*(F) is flat, h′h' is injective (Chapter I, § 2, no. 2, Definition 1); this proves that ρ(a)∉P\rho(a) \notin \mathfrak{P}, whence ρ−1(P)=p\rho^{-1}(\mathfrak{P}) = p.

Theorem 2

Let ρ:A→B\rho : A \to B be a ring homomorphism, E an A-module and F a B-module such that ρ∗(F)\rho_*(F) is a flat A-module. Then

Ass⁡B(E⊗AF)⊃⋃p∈Ass⁡A(E)Ass⁡B(F/pF). \operatorname{Ass}_B(E \otimes_A F) \supset \bigcup_{p \in \operatorname{Ass}_A(E)} \operatorname{Ass}_B(F/pF).

When A is Noetherian, the two sides of (5) are equal.

Let p∈Ass⁡A(E)p \in \operatorname{Ass}_A(E); by definition there exists an exact sequence

0→A/p→E. 0 \to A/p \to E.

Since F is a flat A-module we derive an exact sequence

0→F/pF→E⊗AF 0 \to F/pF \to E \otimes_A F

whence Ass⁡B(F/pF)⊂Ass⁡B(E⊗AF)\operatorname{Ass}_B(F/pF) \subset \operatorname{Ass}_B(E \otimes_A F), which proves the inclusion (5).

Suppose now that A is Noetherian and let us prove the opposite inclusion.

We proceed in stages:

(i) Suppose first that E is a finitely generated A-module and that Ass_A(E) is reduced to a single element pp. By § 1, no. 4, Theorem 1 there exists a composition series (Ei)0≤i≤n(E_i)_{0 \leq i \leq n} of E such that Ei/Ei+1E_i/E_{i+1} is isomorphic to A/piA/p_i, where pip_i is a prime ideal of A; moreover (§1§ 1, no. 4, Theorem 2 and no. 3, Proposition 7) all the pip_i contain pp. As F is a flat A-module, the Ei⊗AFE_i \otimes_A F form a composition series of E⊗AFE \otimes_A F and (Ei⊗AF)/(Ei+1⊗AF)(E_i \otimes_A F)/(E_{i+1} \otimes_A F) is identified with

(A/pi)⊗AF=F/piF. (A/p_i) \otimes_A F = F/p_i F.
Then by virtue of §1§ 1, no. 1, Proposition 3
Ass⁡B(E⊗AF)⊂⋃i=0n−1Ass⁡B(F/piF). \operatorname{Ass}_B(E \otimes_A F) \subset \bigcup_{i=0}^{n-1} \operatorname{Ass}_B(F/p_i F).
We know that E is annihilated by a power of pp (no. 1, Remark); Lemma 1 then shows that, for all P∈Ass⁡B(E⊗AF)\mathfrak{P} \in \operatorname{Ass}_B(E \otimes_A F), ρ−1(P)=p\rho^{-1}(\mathfrak{P}) = p. As F/piFF/p_i F is isomorphic to (A/pi)⊗AF(A/p_i) \otimes_A F, ρ−1(P′)=pi\rho^{-1}(\mathfrak{P}') = p_i for all P′∈Ass⁡B(F/piF)\mathfrak{P}' \in \operatorname{Ass}_B(F/p_i F) by Lemma 1, whence Ass⁡B(E⊗AF)∩Ass⁡(F/piF)=∅\operatorname{Ass}_B(E \otimes_A F) \cap \operatorname{Ass}(F/p_i F) = \varnothing if pi≠pp_i \neq p, which proves the theorem in the case considered.

(ii) Suppose only that E is a finitely generated A-module. Let pip_i (1≤i≤n1 \leq i \leq n) be the elements of Ass⁡A(E)\operatorname{Ass}_A(E) and let {0}=⋂nQi\{0\} = \bigcap^n Q_i be a corresponding reduced primary decomposition (no. 3); E is then isomorphic to a submodule of the direct sum of the Ei=E/QiE_i = E/Q_i and, as F is a flat A-module, E⊗AFE \otimes_A F is isomorphic to a submodule of the direct sum of B-modules Ei⊗AFE_i \otimes_A F. We deduce (§1§ 1, no. 1, Proposition 3 and Corollary 1 to Proposition 3) that

Ass⁡B(E⊗AF)⊂⋃i=1nAss⁡B(Ei⊗AF). \operatorname{Ass}_B(E \otimes_A F) \subset \bigcup_{i=1}^n \operatorname{Ass}_B(E_i \otimes_A F).
But EiE_i is a finitely generated A-module such that Ass⁡A(Ei)\operatorname{Ass}_A(E_i) is reduced to a single element pip_i (no. 1, Definition 1). By (i), Ass⁡B(Ei⊗AF)=Ass⁡B(F/piF)\operatorname{Ass}_B(E_i \otimes_A F) = \operatorname{Ass}_B(F/p_i F), whence the theorem in this case.

(iii) General case. The B-module E⊗AFE \otimes_A F is the union of the submodules E′⊗AFE' \otimes_A F, where E′E' runs through the set of finitely generated submodules of the A-module E. If P\mathfrak{P} belongs to Ass⁡B(E⊗AF)\operatorname{Ass}_B(E \otimes_A F), then there exists a finitely generated submodule E′E' of E such that P∈Ass⁡B(E′⊗AF)\mathfrak{P} \in \operatorname{Ass}_B(E' \otimes_A F). By (ii), there exists p∈Ass⁡A(E′)p \in \operatorname{Ass}_A(E') such that P∈Ass⁡B(F/pF)\mathfrak{P} \in \operatorname{Ass}_B(F/pF); as Ass⁡A(E′)⊂Ass⁡A(E)\operatorname{Ass}_A(E') \subset \operatorname{Ass}_A(E), this completes the proof of Theorem 2.

Corollary 1

If A is Noetherian and P∈Ass⁡B(E⊗AF)\mathfrak{P} \in \operatorname{Ass}_B(E \otimes_A F), then ρ−1(P)∈Ass⁡A(E)\rho^{-1}(\mathfrak{P}) \in \operatorname{Ass}_A(E) and ρ−1(P)\rho^{-1}(\mathfrak{P}) is the only prime ideal pp of A such that P∈Ass⁡B(F/pF)\mathfrak{P} \in \operatorname{Ass}_B(F/pF).

Corollary 2

Suppose that AA and BB are Noetherian and that BB is a flat AA-module. Let pp be a prime ideal of AA, Q⊂EQ \subset E a pp-primary submodule and P\mathfrak{P} a prime ideal of BB. For Q⊗ABQ \otimes_A B to be a qq-primary submodule of E⊗ABE \otimes_A B, it is necessary and sufficient that pBpB be a 'p-primary ideal' of BB.

Let us apply Theorem 2 to the AA-module E/QE/Q and the BB-module BB; then Ass⁡A(E/Q)={p}\operatorname{Ass}_A(E/Q) = \{p\} and (E/Q)⊗AB(E/Q) \otimes_A B is isomorphic to (E⊗AB)/(Q⊗AB)(E \otimes_A B)/(Q \otimes_A B) and hence Ass⁡B((E⊗AB)/(Q⊗AB))=Ass⁡B(B/pB)\operatorname{Ass}_B((E \otimes_A B)/(Q \otimes_A B)) = \operatorname{Ass}_B(B/pB). To say that Q⊗ABQ \otimes_A B is P\mathfrak{P}-primary in E⊗ABE \otimes_A B therefore means that Ass⁡B(B/pB)\operatorname{Ass}_B(B/pB) is reduced to P\mathfrak{P}, whence the corollary.

Remark

Suppose that AA and BB are Noetherian. Let P\mathfrak{P} be a prime ideal of BB and let p=ρ‾−1(P)p = \overline{\rho}^{-1}(\mathfrak{P}); let us write S=A−pS = A - p and let k(p)=S−1(A/p)k(p) = S^{-1}(A/p) be the field of fractions of A/pA/p. Since P\mathfrak{P} contains pBpB, P/pB\mathfrak{P}/pB is a prime ideal of B/pBB/pB. If ρ′\rho' is the composite homomorphism A→ρB→B/pBA \xrightarrow{\rho} B \to B/pB, we know that S−1(B/pB)S^{-1}(B/pB) is identified with the ring (ρ′(S))−1(B/pB)(\rho'(S))^{-1}(B/pB) and P′=S−1(P/pB)\mathfrak{P}' = S^{-1}(\mathfrak{P}/pB) with an ideal of this ring (Chapter II, § 2, no. 2, Proposition 6); as P/pB\mathfrak{P}/pB does not meet ρ′(S)\rho'(S), P′\mathfrak{P}' is a prime ideal of S−1(B/pB)S^{-1}(B/pB) (Chapter 11, § 2, no. 5, Proposition 11); moreover there are canonical isomorphisms between S−1(B/pB)S^{-1}(B/pB), S−1((A/p)⊗AB)S^{-1}((A/p) \otimes_A B) and (S−1(A/p))⊗AB=k(p)⊗AB(S^{-1}(A/p)) \otimes_A B = k(p) \otimes_A B; similarly S−1(F/pF)S^{-1}(F/pF) is canonically identified with k(p)⊗AFk(p) \otimes_A F. This being so, under the hypotheses of Theorem 2, in order that P∈Ass⁡B(E⊗AF)\mathfrak{P} \in \operatorname{Ass}_B(E \otimes_A F), it is necessary and sufficient that p∈Ass⁡A(E)p \in \operatorname{Ass}_A(E) and

P′∈Ass⁡k(p)⊗AB(k(p)⊗AF). \mathfrak{P}' \in \operatorname{Ass}_{k(p) \otimes_A B}(k(p) \otimes_A F).
For by Theorem 2 and its Corollary 1, it amounts to verifying that the conditions
“P∈Ass⁡B(F/pF)”and“P′∈Ass⁡k(p)⊗AB(k(p)⊗AF)” \text{``}\mathfrak{P} \in \operatorname{Ass}_B(F/pF)\text{''} \quad \text{and} \quad \text{``}\mathfrak{P}' \in \operatorname{Ass}_{k(p) \otimes_A B}(k(p) \otimes_A F)\text{''}
are equivalent; but, as BB is Noetherian, this follows from § 1, no. 2, Corollary to Proposition 5 and the above identifications.

Proposition 11

Suppose that AA and BB are Noetherian and that BB is a flat AA-module. Let EE be an AA-module and E′E' a submodule of EE such that, for every ideal p∈Ass⁡A(E/E′)p \in \operatorname{Ass}_A(E/E'), pBpB is a prime ideal of BB or equal to BB. Let E′=⋂p∈Ass⁡(E/E′)Q(p)E' = \bigcap_{p \in \operatorname{Ass}(E/E')} Q(p) be a reduced primary decomposition of E′E' in EE, Q(p)Q(p) being pp-primary for all p∈Ass⁡(E/E′)p \in \operatorname{Ass}(E/E'). (i) If p∈Ass⁡(E/E′)p \in \operatorname{Ass}(E/E') and pB=BpB = B, then Q(p)⊗AB=E⊗ABQ(p) \otimes_A B = E \otimes_A B. (ii) If p∈Ass⁡(E/E′)p \in \operatorname{Ass}(E/E') and pBpB is prime, Q(p)Q(p) is pBpB*-primary* in E⊗ABE \otimes_A B. (iii) If Φ\Phi is the set p∈Ass⁡(E/E′)\mathfrak{p} \in \operatorname{Ass}(E/E') such that pBpB is prime, then

E′⊗AB=⋂p∈Φ(Q(p)⊗AB) E' \otimes_A B = \bigcap_{p \in \Phi} (Q(p) \otimes_A B)
and this relation is a reduced primary decomposition of E′⊗ABE' \otimes_A B in E⊗ABE \otimes_A B.

If pB=BpB = B, Theorem 2 applied to E/Q(p)E/\mathbf{Q}(p) and BB shows that

Ass⁡B((E/Q(p))⊗AB)=∅ \operatorname{Ass}_B((E/\mathbf{Q}(p)) \otimes_A B) = \varnothing
and, as BB is Noetherian and is a flat AA-module, we conclude (§ 1, no. 1, Corollary 1 to Proposition 2) that Q(p)⊗AB=E⊗AB\mathbf{Q}(p) \otimes_A B = E \otimes_A B. Assertion (ii) follows from Corollary 2 to Theorem 2, taking P=pB\mathfrak{P} = pB. Finally the relation E′⊗AB=⋂p∈Φ(Q(p)⊗AB)E' \otimes_A B = \bigcap_{p \in \Phi} (\mathbf{Q}(p) \otimes_A B) follows from the fact that BB is a flat AA-module (Chapter I, § 2, no. 6, Proposition 6); as p=ρ‾−1(pB)p = \overline{\rho}^{-1}(pB) for p∈Φp \in \Phi (Lemma 1), pB≠p′BpB \neq p'B for two distinct ideals p,p′p, p' of the set Φ\Phi; on the other hand,
Ass⁡((E⊗AB)/(E′⊗AB))=Φ \operatorname{Ass}((E \otimes_A B)/(E' \otimes_A B)) = \Phi
by Theorem 2; we conclude from no. 3, Proposition 4 that
E′⊗AB=⋂p∈Φ(Q(p)⊗AB) E' \otimes_A B = \bigcap_{p \in \Phi} (\mathbf{Q}(p) \otimes_A B)
is a reduced primary decomposition.

Corollary

Suppose that* pBpB is prime for all p∈Ass⁡A(E/E′)p \in \operatorname{Ass}_A(E/E'). Then, if p1,…,pnp_1, \ldots, p_n are the minimal elements of Ass⁡A(E/E′)\operatorname{Ass}_A(E/E'), the piBp_iB are minimal elements of

Ass⁡A((E⊗AB)/(E′⊗AB)). \operatorname{Ass}_A((E \otimes_A B)/(E' \otimes_A B)).
It follows from Proposition 11 that in this case piB≠pjBp_iB \neq p_jB for i#ji \# j.

Examples (1) Let us take B=S−1AB = S^{-1}A, where SS is a multiplicative subset of AA; if AA is Noetherian, the hypotheses of Proposition 11 are satisfied and we recover a part of Proposition 6 of no. 4. (2) Let AA be a Noetherian ring, mm an ideal of AA and BB the Hausdorff completion of AA with respect to the mm-adic topology; then BB is a flat AA-module and Theorem 2 may be applied with F=BF = B; but in general the hypotheses of Proposition 11 are not satisfied for the prime ideals of AA (Chapter III, § 2, Exercise 15 (b)). (3) Let AA be a Noetherian ring and BB the polynomial algebra A[X1,…,Xn]A[X_1, \ldots, X_n]; BB is Noetherian and is a free AA-module and therefore flat. Also, if pp is a prime ideal of AA, B/pBB/pB is isomorphic to (A/p)[X1,…,Xn](A/p)[X_1, \ldots, X_n], which is an integral domain, and hence pBpB is prime; the hypotheses of Proposition 11 are therefore satisfied for every AA-module EE and every submodule E′E' of EE. (4) Let AA be a finitely generated algebra over a field kk, KK an extension of kk and B=A⊗kKB = A \otimes_k K the algebra over KK obtained by extension of scalars; AA and BB are Noetherian and BB is a free AA-module and hence Theorem 2 may be applied to F=BF = B. In certain cases (for example if kk is algebraically closed) it can be shown that for every prime ideal pp of AA, pBpB is prime or equal to BB; we shall return later to this example.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.