Bourbaki

5. Flatness properties of filtered modules

1. Ideally Hausdorff modules

Definition 1

Let A be a commutative ring and g\mathfrak{g} an ideal of A. An A-module M is called ideally Hausdorff with respect to 3 (or simply ideally Hausdorff if there is no ambiguity) if, for every finitely generated ideal a\mathfrak{a} of A, the A-module a⊗AM\mathfrak{a} \otimes_A M is Hausdorff with the 3-adic topology.

Putting a=A\mathfrak{a} = A in this definition, we have already seen that M is necessarily Hausdorff with the 3-adic topology.

Examples

Example 1

If A is Noetherian and g\mathfrak{g} is contained in the Jacobson radical of A (in other words if A is a Zariski ring with the 3-adic topology), every finitely generated A-module is ideally Hausdorff (§3, no. 3, Proposition 6).

Example 2

Every direct sum of ideally Hausdorff modules is an ideally Hausdorff module, by virtue of the relations

gn(a⊗A⨁λ∈LMλ)=gn⨁λ∈L(a⊗AMλ)=⨁λ∈Lgn(a⊗AMλ). \mathfrak{g}^n \left( a \otimes_A \bigoplus_{\lambda \in L} M_\lambda \right) = \mathfrak{g}^n \bigoplus_{\lambda \in L} (a \otimes_A M_\lambda) = \bigoplus_{\lambda \in L} \mathfrak{g}^n (a \otimes_A M_\lambda).

Example 3

If an A-module M is flat and Hausdorff with the g\mathfrak{g}-adic topology it is ideally Hausdorff, for a⊗AMa \otimes_A M is then identified with a submodule of M and the g\mathfrak{g}-adic topology on a⊗AMa \otimes_A M is finer than the topology induced on a⊗AMa \otimes_A M by the g\mathfrak{g}-adic topology on M, which is Hausdorff by hypothesis.

2. Statement of the flatness criterion

Let A be a ring, z\mathfrak{z} a two-sided ideal of A, M a left-module and gr(A) and gr(M) the graded ring and graded gr(A)-module associated respectively with the ring A and with the module M with the g\mathfrak{g}-adic filtrations (§ 2, no. 3). We have seen (loc. cit.) that for every integer n≥0n \geq 0 there is a surjective Z\mathbf{Z}-module homomorphism

γn:(gn/gn+1)⊗A/g(M/gM)→gnM/gn+1M \gamma_n : (\mathfrak{g}^n / \mathfrak{g}^{n+1}) \otimes_{A/\mathfrak{g}} (M / \mathfrak{g}M) \to \mathfrak{g}^n M / \mathfrak{g}^{n+1} M
and a graded homomorphism of degree 0 of graded gr(A)-modules
γM:gr(A)⊗gr0(A)gr0(M)→gr(M) \gamma_M : \mathrm{gr}(A) \otimes_{\mathrm{gr}_0(A)} \mathrm{gr}_0(M) \to \mathrm{gr}(M)
whose restriction to grn(A)⊗gr0(A)gr0(M)\mathrm{gr}_n(A) \otimes_{\mathrm{gr}_0(A)} \mathrm{gr}_0(M) is γn\gamma_n for all nn and which is therefore surjective.

Theorem 1

Let A be a commutative ring, g\mathfrak{g} an ideal of A and M an A-module. Consider the following properties: (i) M is afat A-module. (ii) Tor1A(N,M)=0\mathrm{Tor}_1^A(N, M) = 0 for every A-module N annihilated by z\mathfrak{z}. (iii) M/gMM / \mathfrak{g}M is ajlat (A/g)(A / \mathfrak{g})-module and the canonical mapping z⊗AM→gM\mathfrak{z} \otimes_A M \to \mathfrak{g}M is bijective (the latter condition being equivalent to Tor1A(A/g,M)=0\mathrm{Tor}_1^A(A / \mathfrak{g}, M) = 0 by virtue of the relation Tor1A(A,M)=0\mathrm{Tor}_1^A(A, M) = 0 and the exact sequence

Tor1A(A,M)→Tor1A(A/g,M)→z⊗AM→M). \mathrm{Tor}_1^A(A, M) \to \mathrm{Tor}_1^A(A / \mathfrak{g}, M) \to \mathfrak{z} \otimes_A M \to M ).
(iv) M/gMM / \mathfrak{g}M is ajlat (A/g)(A / \mathfrak{g})-module and the canonical homomorphism
γM:gr(A)⊗gr0(A)gr0(M)→gr(M) \gamma_M : \mathrm{gr}(A) \otimes_{\mathrm{gr}_0(A)} \mathrm{gr}_0(M) \to \mathrm{gr}(M)
is bijective (property (GR) of § 2, no. 8). (v) For all n≥1n \geq 1, M/gnMM / \mathfrak{g}^n M is a flat (A/gn)(A / \mathfrak{g}^n)-module. Then (i) ⇒\Rightarrow (ii) ⇔\Leftrightarrow (iii) ⇒\Rightarrow (iv) ⇔\Leftrightarrow (v). If further g\mathfrak{g} is nilpotent or if A is Noetherian and M is ideally Hausdorff, properties (i), (ii), (iii), (iv) and (v) are equivalent.

Remark

If A/gA / \mathfrak{g} is a field (as often happens in applications) the condition "M/ gM\mathfrak{g}M is a flat (A/g)(A / \mathfrak{g})-module" holds automatically for every A-module M, which simplifies the statement of properties (iii) and (iv); moreover, in this case, property (v) is equivalent to saying that M/gnMM / \mathfrak{g}^n M is a free (A/gn)(A / \mathfrak{g}^n)-module for every integer n≥1n \geq 1 (Chapter 11, § 3, no. 2, Corollary 2 to Proposition 5).

3. Proof of the flatness criterion

(A) The implications (i) ⇒\Rightarrow (ii) ⇔\Leftrightarrow (iii)

The implication (i) ⇒\Rightarrow (ii) is immediate (Chapter I, § 4). The equivalence (ii) ⇔\Leftrightarrow (iii) is a special case of Chapter I, § 4, Proposition 2 applied to R=A,S=A/g,F=M,E=NR = A, S = A/\mathfrak{g}, F = M, E = N, taking account of the fact that being given an (A/g)(A/\mathfrak{g})-module structure on NN is equivalent to being given an AA-module structure under which NN is annihilated by 3.

Remark (1). Condition (ii) is also equivalent to the following:

(ii') Tor⁡1A(N,M)=0\operatorname{Tor}_1^A(N, M) = 0 for every AA*-module* NN annihilated by a power of 3.

Clearly (ii') implies (ii). Conversely, if (ii) holds, then in particular Tor⁡1A(gnN/gn+1N,M)=0\operatorname{Tor}_1^A(\mathfrak{g}^n N / \mathfrak{g}^{n+1} N, M) = 0 for all nn; from the exact sequence

0→gn+1N→gnN→gnN/gn+1N→0 0 \to \mathfrak{g}^{n+1} N \to \mathfrak{g}^n N \to \mathfrak{g}^n N / \mathfrak{g}^{n+1} N \to 0
we derive the exact sequence
Tor⁡1A(gn+1N,M)→Tor⁡1A(gnN,M)→Tor⁡1A(gnN/gn+1N,M) \operatorname{Tor}_1^A(\mathfrak{g}^{n+1} N, M) \to \operatorname{Tor}_1^A(\mathfrak{g}^n N, M) \to \operatorname{Tor}_1^A(\mathfrak{g}^n N / \mathfrak{g}^{n+1} N, M)
and, as there exists an integer mm such that gmN=0\mathfrak{g}^m N = 0, we deduce by descending induction on nn that Tor⁡1A(gnN,M)=0\operatorname{Tor}_1^A(\mathfrak{g}^n N, M) = 0 for all n≤mn \leq m and in particular for n=0n = 0.

It follows from this that if 9 is nilpotent, (ii) implies (i), for (ii') then means that Tor⁡1A(N,M)=0\operatorname{Tor}_1^A(N, M) = 0 for every AA*-module* NN and hence that MM is flat (Chapter I, § 4).

(B) Let us prove the following proposition:

Proposition 1

Let* AA be a commutative ring, 3 an ideal of AA and MM an AA*-module*. The following conditions are equivalent: (a) For all n≥1n \geq 1, Tor⁡1A(A/gn,M)=0\operatorname{Tor}_1^A(A/\mathfrak{g}^n, M) = 0. (b) For all n≥1n \geq 1, the canonical homomorphism

θn:gn⊗AM→gnM \theta_n : \mathfrak{g}^n \otimes_A M \to \mathfrak{g}^n M
is bijective.

Moreover these conditions imply : (c) The canonical homomorphism γM:gr⁡(A)⊗gr⁡0(A)gr⁡r(M)→gr⁡(M)\gamma_M : \operatorname{gr}(A) \otimes_{\operatorname{gr}_0(A)} \operatorname{gr}_r(M) \to \operatorname{gr}(M) is bijective. Conversely, if g\mathfrak{g} is nilpotent, (c) implies (a) and (b).

The equivalence of (a) and (b) follows from the exact sequence

0=Tor⁡1A(A,M)→Tor⁡1A(A/gn,M)→gn⊗AM→M. 0 = \operatorname{Tor}_1^A(A, M) \to \operatorname{Tor}_1^A(A/\mathfrak{g}^n, M) \to \mathfrak{g}^n \otimes_A M \to M.

Consider next the diagram

gn+1⊗AM⟶gn⊗AM⟶(gn/gn+1)⊗A(M/gM)⟶0↓θn+1↓θn↓γn0⟶gn+1M⟶gnM⟶gr⁡n(M)⟶0 \begin{array}{ccccccccc} \mathfrak{g}^{n+1} \otimes_A M & \longrightarrow & \mathfrak{g}^n \otimes_A M & \longrightarrow & (\mathfrak{g}^n / \mathfrak{g}^{n+1}) \otimes_A (M / \mathfrak{g} M) & \longrightarrow & 0 \\ \downarrow \theta_{n+1} & & \downarrow \theta_n & & \downarrow \gamma_n & & \\ 0 & \longrightarrow & \mathfrak{g}^{n+1} M & \longrightarrow & \mathfrak{g}^n M & \longrightarrow & \operatorname{gr}_n(M) & \longrightarrow & 0 \end{array}

where we note that (gn/gn+1)⊗A(M/gM)(\mathfrak{g}^n/\mathfrak{g}^{n+1}) \otimes_A (M/\mathfrak{g}M) is canonically identified with (gn/gn+1)⊗A/g(M/gM)(\mathfrak{g}^n/\mathfrak{g}^{n+1}) \otimes_{A/\mathfrak{g}} (M/\mathfrak{g}M). This diagram is commutative by definition of γn\gamma_n and its rows are exact. If (b) holds, 00, and θn+1\theta_{n+1} are bijective and so therefore is γn\gamma_n by definition of cokernel, hence (b) implies (c). Conversely, assuming that g\mathfrak{g} is nilpotent, let us show that (c) implies (b); we shall argue by descending induction on nn, since gn⊗AM=gnM=0\mathfrak{g}^n \otimes_A M = \mathfrak{g}^n M = 0 for nn sufficiently large. Suppose then that in diagram (1), γn\gamma_n and θn+1\theta_{n+1} are bijective; then so is θn\theta_n by virtue of Chapter I, § 1, no. 4, Corollary 1 to Proposition 2.

(C) The implication (ii) ⇒\Rightarrow (iv)

If (ii) holds, so does (ii') by Remark 1; Proposition 1 then shows that γM\gamma_M is an isomorphism. On the other hand, we already know that (ii) implies (iii) and hence M/gMM/\mathfrak{g}M is a flat (A/g)(A/\mathfrak{g})-module, which completes the proof that (ii) implies (iv).

Remark (2). Proposition 1 shows that, if g\mathfrak{g} is nilpotent, (iv) implies (iii); taking account of Remark 1, we have therefore proved in this case that (i), (ii), (iii) and (iv) are equivalent.

(D) The equivalence (iv) ⇔\Leftrightarrow (v)

For all n≥1n \geq 1, M/gnMM/\mathfrak{g}^n M has a canonical (A/gn)(A/\mathfrak{g}^n)-module structure. If it is filtered by the (g/gn)(\mathfrak{g}/\mathfrak{g}^n)-adic filtration, it is immediate that grm(M/gnM)=grm(M)\mathrm{gr}_m(M/\mathfrak{g}^n M) = \mathrm{gr}_m(M) if m<nm < n and grm(M/gnM)=0\mathrm{gr}_m(M/\mathfrak{g}^n M) = 0 if m≥nm \geq n. For all k≥1k \geq 1, let Ak=A/gkA_k = A/\mathfrak{g}^k, gk=g/gk\mathfrak{g}_k = \mathfrak{g}/\mathfrak{g}^k, Mk=M/gkMM_k = M/\mathfrak{g}^k M; let (iv), (resp. (v)k(v)_k) denote the assertion derived from (iv) (resp. (v)) by replacing A,g,MA, \mathfrak{g}, M by Ak,gk,MkA_k, \mathfrak{g}_k, M_k. It follows from what has just been said that (iv) is equivalent to "for all k≥1k \geq 1, (iv)k(iv)_k" and obviously (v) is equivalent to "for all k≥1k \geq 1, (v)k(v)_k". Then it will suffice to establish the equivalence (iv)k⇔(v)k(iv)_k \Leftrightarrow (v)_k for all kk or also to show that (iv) ⇔\Leftrightarrow (v) when 33 is nilpotent. Now (Remark 2) we have seen that in that case (iv) is equivalent to (i). As M/gnMM/\mathfrak{g}^n M is isomorphic to M⊗A(A/gn)M \otimes_A (A/\mathfrak{g}^n), (i) implies (v) (Chapter I, § 2, no. 7, Corollary 2 to Proposition 8); moreover clearly (v) then implies (i). We have therefore shown the equivalence (iv) ⇔\Leftrightarrow (v) in all cases and also that of all the properties of the theorem in the case where 33 is nilpotent.

(E) The implication (v) ⇒\Rightarrow (i) when AA is Noetherian and MM ideally Hausdorff

It is sufficient to prove that for every ideal aa of AA the canonical mapping j:a⊗AM→Mj : a \otimes_A M \to M is injective (Chapter I, § 2, no. 3, Proposition 1). Let x∈Ker jx \in \mathrm{Ker}\, j; a⊗AMa \otimes_A M is Hausdorff with the g\mathfrak{g}-adic topology, it suffices to verify that, for every integer n>0n > 0, x∈gn(a⊗AM)x \in \mathfrak{g}^n(a \otimes_A M). Let f:gna→af : \mathfrak{g}^n a \to a be the canonical injection; it suffices to show that x∈Im(f⊗1M)x \in \mathrm{Im}(f \otimes 1_M); for if b∈gnb \in \mathfrak{g}^n, a∈aa \in a and m∈Mm \in M, the image under f⊗1f \otimes 1, of the element (ba)⊗m(ba) \otimes m of (gna)⊗AM(\mathfrak{g}^n a) \otimes_A M is the element (ba)⊗m=b(a⊗m)(ba) \otimes m = b(a \otimes m) of a⊗AMa \otimes_A M and hence Im(f⊗1M)⊂gn(a⊗AM)\mathrm{Im}(f \otimes 1_M) \subset \mathfrak{g}^n(a \otimes_A M). By virtue of Krull's Theorem (§ 3, no. 2, Theorem 2), there exists an integer kk such that aj=a∩gk⊂gnaa_j = a \cap \mathfrak{g}^k \subset \mathfrak{g}^n a; if i:aj→ai : a_j \to a is the canonical injection, it will then be sufficient to show that x∈Im⁡(i⊗1M)x \in \operatorname{Im}(i \otimes 1_M). Now, denoting by p:a→a/akp : a \to a/a_k and h:a/ak→A/gkh : a/a_k \to A/\mathfrak{g}^k the canonical mappings, there is a commutative diagram

ak⊗AM→i⊗1Ma⊗AM→p⊗1M(a/ak)⊗AM⟶0↓↓h⊗1MM⟶(A/gk)⊗AM \begin{array}{ccccccccc} a_k \otimes_A M & \xrightarrow{i \otimes 1_M} & a \otimes_A M & \xrightarrow{p \otimes 1_M} & (a/a_k) \otimes_A M & \longrightarrow & 0 \\ & & \downarrow & & \downarrow h \otimes 1_M & & \\ & & M & \longrightarrow & (A/\mathfrak{g}^k) \otimes_A M & & \end{array}

in which the first row is exact. It suffices to prove that x∈Ker⁡(p⊗1M)x \in \operatorname{Ker}(p \otimes 1_M) and, as x∈Ker⁡jx \in \operatorname{Ker} j by hypothesis, it will suffice to verify that the mapping h⊗11h \otimes 1_1, is injective. Now, it may also be written (Algebra, Chapter II, § 3, no. 6, Corollary 3 to Proposition 6)

h⊗1M/gkM:(a/ak)⊗A/gk(M/gkM)→M/gkM h \otimes 1_{M/\mathfrak{g}^k M} : (a/a_k) \otimes_{A/\mathfrak{g}^k} (M/\mathfrak{g}^k M) \to M/\mathfrak{g}^k M

and, as hh is injective and, by (v), M/gkMM/\mathfrak{g}^k M is a flat (A/gk)(A/\mathfrak{g}^k)-module, this completes the proof.

4. Applications

Proposition 2

Let AA be a commutative ring, g\mathfrak{g} an ideal of AA and BB a commutative Noetherian AA-algebra such that BB is contained in the Jacobson radical of BB. Then every finitely generated BB-module MM is an ideally Hausdorff AA-module with respect to g\mathfrak{g}.

We shall see more generally that for every finitely generated AA-module NN, N⊗AMN \otimes_A M is Hausdorff with the 33-adic topology. For N(B)=N⊗ABN_{(B)} = N \otimes_A B is a finitely generated BB-module and the BB-module N⊗AMN \otimes_A M is canonically identified with N,,⊗BMN_{,,} \otimes_B M by virtue of the associativity of the tensor product. Let L\mathfrak{L} be the Jacobson radical of BB; as gB\mathfrak{g} B is contained in L\mathfrak{L}, the g\mathfrak{g}-adic topology on N⊗AMN \otimes_A M is therefore identified with a finer topology than the L\mathfrak{L}-adic topology on N(B)⊗BMN_{(B)} \otimes_B M; but this latter topology is Hausdorff since N,,⊗BMN_{,,} \otimes_B M is a finitely generated BB-module (no. 1, Example 1), whence the conclusion.

Proposition 3

Let AA be a commutative ring, BB a commutative AA-algebra, g\mathfrak{g} an ideal of AA and MM a BB-module. Suppose that BB is a Noetherian ring and a flat AA-module and that MM is ideally Hausdorff with respect to gB\mathfrak{g} B. The following conditions are equivalent: (a) MM is a flat BB-module. (b) MM is a flat AA-module and M/gM=M/(gB)MM/\mathfrak{g} M = M/(\mathfrak{g} B) M is a flat (B/gB)(B/\mathfrak{g} B)-module. If further the canonical homomorphism A/g→B/gBA/\mathfrak{g} \to B/\mathfrak{g} B is bijective, conditions (a) and (b) are also equivalent to: (c) MM is a flat AA-module.

Condition (a) implies (b) by Chapter I, §2, no. 7, Corollaries 2 and 3 to Proposition 8 and the fact that M/gMM/\mathfrak{g} M is isomorphic to M⊗B(B/gB)M \otimes_B (B/\mathfrak{g} B). Suppose condition (b) holds; to show that MM is a flat BB-module, we shall apply Theorem 1 of no. 2 with AA replaced by BB and g\mathfrak{g} by gB\mathfrak{g} B. It will therefore be sufficient to show that the canonical mapping f:JB⊗BM→JMf : \mathfrak{J}B \otimes_B M \to \mathfrak{J}M is injective. Let f1f_1 be the canonical mapping J⊗AB→JB\mathfrak{J} \otimes_A B \to \mathfrak{J}B and f2f_2 the canonical isomorphism J⊗AM→(J⊗AB)⊗BM;f∘(f1∘1M)∘f2\mathfrak{J} \otimes_A M \to (\mathfrak{J} \otimes_A B) \otimes_B M ; f \circ (f_1 \circ 1_M) \circ f_2 is the canonical mapping f′:J⊗AM→JMf' : \mathfrak{J} \otimes_A M \to \mathfrak{J}M, as is easily verified. Now f′f' is an isomorphism since MM is a flat AA-module, whilst f1f_1 is an isomorphism because BB is flat over A;fA ; f is then an isomorphism.

Let p:A/J→B/JBp : A/\mathfrak{J} \to B/\mathfrak{J}B be the canonical homomorphism; the (A/J)(A/\mathfrak{J})-module structure on M/JMM/\mathfrak{J}M derived by means of pp from its (B/JB)(B/\mathfrak{J}B)-module structure is isomorphic to that on M⊗A(A/J)M \otimes_A (A/\mathfrak{J}). Then it follows that, if MM is a flat AA-module, M/JMM/\mathfrak{J}M is a flat (A/J)(A/\mathfrak{J})-module and hence also a flat (B/JB)(B/\mathfrak{J}B)-module if pp is an isomorphism; we have thus proved that (c) ⇒\Rightarrow (b) in that case.

Corollary

Let AA be a commutative Noetherian ring, J\mathfrak{J} an ideal of AA, A^\hat{A} the Hausdorff completion of AA with respect to the J\mathfrak{J}-adic topology and MM an ideally Hausdorff A^\hat{A}-module with respect to JA^\mathfrak{J} \hat{A}. For MM to be a flat AA-module, it is necessary and sufficient that MM be a flat A^\hat{A}-module.

We know in fact that A^\hat{A} is a Noetherian ring (§3§ 3, no. 4, Proposition 8) and a flat AA-module (§3§ 3, no. 4, Theorem 3), that JA^=J^\mathfrak{J} \hat{A} = \hat{\mathfrak{J}} (§2§ 2, no. 12, Proposition 16) and that the canonical homomorphism A/J→A^/J^A/\mathfrak{J} \to \hat{A}/\hat{\mathfrak{J}} is bijective (§2§ 2, no. 12, Proposition 15); Proposition 3 can therefore be applied.

Proposition 4

Let AA and BB be two commutative Noetherian rings, h:A→Bh : A \to B a ring homomorphism, 33 an ideal of AA and 22 an ideal of BB containing JB\mathfrak{J}B and contained in the Jacobson radical of BB. Let A^\hat{A} be the Hausdorff completion of AA with respect to the 33-adic topology and B^\hat{B} the Hausdorff completion of BB with respect to the 22-adic topology; hh is continuous with these topologies and h:A^→B^h : \hat{A} \to \hat{B} therefore makes 8\mathbf{8} into an A^\hat{A}-algebra. Let MM be a finitely generated BB-module and M^\hat{M} its Hausdorff completion with respect to the 22-adic topology; the following properties are equivalent: (a) MM is a flat AA-module. (b) M^\hat{M} is a flat AA-module. (c) M^\hat{M} is a flat A^\hat{A}-module.

As BB with the 22-adic topology is a Zariski ring, B^\hat{B} is a faithfully flat BB-module (§3§ 3, no. 5, Proposition 9) and MM is canonically isomorphic to M⊗BB^M \otimes_B \hat{B} (§3§ 3, no. 4, Theorem 3); it is immediately verified that this canonical isomorphism is an isomorphism of the AA-module structure on M^\hat{M} onto the AA-module structure on M⊗BB^M \otimes_B \hat{B} derived from that on MM. Applying Proposition 4 of Chapter I, §3§ 3, no. 2 with RR replaced by BB, SS by AA, EE by B^\hat{B}, FF by MM, we see that for MM to be a flat AA-module, it is necessary and sufficient that M^\hat{M} be a flat AA-module. Moreover, M^\hat{M} is a finitely generated B^\hat{B}-module and JB^\mathfrak{J} \hat{B} is contained in Q=QB^\mathfrak{Q} = \mathfrak{Q} \hat{B} and hence in the Jacobson radical of B^\hat{B} (§3§ 3, no. 4, Proposition 8); therefore M^\hat{M} is an ideally Hausdorff AA-module with respect to JA^\mathfrak{J} \hat{A} (Proposition 2). Conditions (b) and (c) are therefore equivalent by the Corollary to Proposition 3.

Exercises

See the exercises for § 5.

Transcribed from the printed volume. Not translated.