Bourbaki

§ 2. General results on filtered rings and modules

1. Filtered rings and modules

Definition 1

An increasing (resp. decreasing) sequence (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} of subgroups of a group GG is called an increasing (resp. decreasing) filtration on GG. A group with a filtration is called a filtered group.

If (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} is an increasing (resp. decreasing) filtration on a group GG and we write Gn′=G−nG'_n = G_{-n}, clearly (Gn′)n∈Z(G'_n)_{n \in \mathbf{Z}} is a decreasing (resp. increasing) filtration on GG. We may therefore restrict our study to decreasing filtrations and hence forth when we speak of a filtration, we shall mean a decreasing filtration, unless otherwise stated.

Given a decreasing filtration (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} on a group GG, clearly ⋂n∈ZGn\bigcap_{n \in \mathbf{Z}} G_n and ⋃n∈ZGn\bigcup_{n \in \mathbf{Z}} G_n are subgroups of GG; the filtration is called separated if ⋃n∈ZGn\bigcup_{n \in \mathbf{Z}} G_n is reduced to the identity element and exhaustive if ⋃n∈ZGn=G\bigcup_{n \in \mathbf{Z}} G_n = G.

Definition 2

Given a ring AA, a filtration (An)n∈Z(A_n)_{n \in \mathbf{Z}} over the additive group AA is called compatible with the ring structure on AA if

(1)

AmAn⊂Am+nform∈Z,n∈Z A_m A_n \subset A_{m+n} \quad \text{for} \quad m \in \mathbf{Z}, \quad n \in \mathbf{Z}

(2)

1∈A0. 1 \in A_0.
The ring AA with this filtration is then called a filtered ring.

Conditions (1) and (2) show that AnA_n is a subring of AA and the AnA_n (left and right) A0A_0-modules. The set B=⋃n∈ZAnB = \bigcup_{n \in \mathbf{Z}} A_n is a subring of AA and the set n=⋂n∈ZAnn = \bigcap_{n \in \mathbf{Z}} A_n, a two-sided ideal of BB; for if x∈nx \in n and a∈Aka \in A_k for all k∈Zk \in \mathbf{Z}, x∈Ak−px \in A_{k-p}, whence ax∈Aax \in A, and xa∈Axa \in A, by (1); therefore ax∈nax \in n and xa∈nxa \in n.

An important particular case is that in which An=AA_n = A; then An=AA_n = A for n≤0n \leq 0 and all the AnA_n are two-sided ideals of AA.

Definition 3

Let AA be a filtered ring, (An)n∈Z(A_n)_{n \in \mathbf{Z}} its filtration and EE an AA-module. A filtration (En)n∈Z(E_n)_{n \in \mathbf{Z}} on EE is called compatible with its module structure over the filtered ring AA if

AmEn⊂Em+nform∈Z,n∈Z. A_m E_n \subset E_{m+n} \quad \text{for} \quad m \in \mathbf{Z}, \quad n \in \mathbf{Z}.
The AA-module EE with this filtration is called a filtered module.

The EnE_n are all AnA_n-modules; if B=⋃n∈ZAnB = \bigcup_{n \in \mathbf{Z}} A_n, clearly ⋃n∈ZEn\bigcup_{n \in \mathbf{Z}} E_n is a BB-module and so is ⋂n∈ZEn\bigcap_{n \in \mathbf{Z}} E_n by the same argument as above for ⋂n∈ZAn\bigcap_{n \in \mathbf{Z}} A_n. If An=AA_n = A, all the EnE_n are submodules of EE.

Examples

(1) Let AA be a graded ring of type Z\mathbf{Z}; for all i∈Zi \in \mathbf{Z}, let A(i)A_{(i)} be the subgroup of homogeneous elements of degree ii in AA. Let us write An=∑i≥nA(i)A_n = \sum_{i \geq n} A_{(i)}; then it is immediate that (An)(A_n) is an exhaustive and separated decreasing filtration which is compatible with the ring structure on AA; this filtration is said to be associated with the graduation (A(i))i∈Z(A_{(i)})_{i \in \mathbf{Z}} and the filtered ring AA is said to be associated with the given graded ring AA.

Now let EE be a graded module of type Z\mathbf{Z} over the graded ring AA and for all i∈Zi \in \mathbf{Z} let E(i)E_{(i)} be the subgroup of homogeneous elements of degree ii of EE. Let us write En=∑i≥nE(i)E_n = \sum_{i \geq n} E_{(i)}; then (En)(E_n) is an exhaustive and separated decreasing filtration which is compatible with the module structure on EE over the filtered ring AA; this filtration is said to be associated with the graduation (E(i))i∈Z(E_{(i)})_{i \in \mathbf{Z}} and the filtered module EE is said to be associated with the given graded module EE.

(2) Let AA be a filtered ring, (An)n∈Z(A_n)_{n \in \mathbf{Z}} its filtration and EE an AA-module. Let us write En=AnEE_n = A_n E; it follows from (1) that

AmEn=AmAnE⊂Am+nE=Em+n, A_m E_n = A_m A_n E \subset A_{m+n} E = E_{m+n},
and from (2) that E0=EE_0 = E; then (En)(E_n) is an exhaustive filtration which is compatible with the AA-module structure on EE. This filtration is said to be derived from the given filtration (An)(A_n) on AA; note that it is not necessarily separated, even if (An)(A_n) is separated and EE and the AnA_n are finitely generated AA-modules (cf. , Exercise 2; see however § 3, no. 3, Proposition 5 and no. 2, Corollary to Proposition 4)

(3) Let AA be a ring and mm a two-sided ideal of AA. Let us write An=mnA_n = m^n for n≥0n \geq 0, An=AA_n = A for n<0n < 0; it is immediate that (An)(A_n) is an exhaustive filtration on AA, called the mm-adic filtration. Let EE be an AA-module; the filtration (En)(E_n) derived from the mm-adic filtration on AA is called the mm-adic filtration on EE; in other words, En=mnEE_n = m^n E for n≥0n \geq 0 and En=EE_n = E for n<0n < 0.

If AA is commutative and BB is an AA-algebra, n=mBn = mB is a two-sided ideal of BB and, for every BB-module FF, nkF=mkFn^k F = m^k F and hence the nn-adic filtration on FF coincides with the mm-adic filtration (if FF is considered as an AA-module).

(4) If AA is a filtered ring and (An)(A_n) its filtration, the left AA-module AnA_n is a filtered AA-module with the filtration (An)(A_n). On the other hand, clearly (An)(A_n) is a filtration which is compatible with the ring structure on the opposite ring A0A^0 and AnA_n is a filtered (left) A0A^0-module with the filtration (An)(A_n).

(5) On a ring AA the sets AnA_n such that An=0A_n = 0 for n>0n > 0, An=AA_n = A for n≤0n \leq 0 form what is called a trivial filtration associated (Example 1) with the trivial graduation on AA; on an AA-module EE, every filtration (En)(E_n) consisting of sub-AA-modules is then compatible with the module structure on EE over the filtered ring AA. Then it is possible to say that every filtered commutative group GG is a filtered Z\mathbf{Z}-module, if Z\mathbf{Z} is given the trivial filtration.

Let GG be a filtered group and (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} its filtration; clearly, for every subgroup HH of GG, (H∩Gn)n∈Z(H \cap G_n)_{n \in \mathbf{Z}} is a filtration said to be induced by that on GG; it is exhaustive (resp. separated) if that on GG is. Similarly, if HH is a normal subgroup of GG, the family ((H.Gn)/H)n∈Z((H.G_n)/H)_{n \in \mathbf{Z}} is a filtration on the group G/HG/H, called the quotient under HH of the filtration on GG; it is exhaustive if (Gn)(G_n) is. If G′G' is another filtered group and (Gn′)n∈Z(G'_n)_{n \in \mathbf{Z}} its filtration, (Gn×Gn′)(G_n \times G'_n) is a filtration on G×G′G \times G' called the product of the filtrations on GG and G′G', which is exhaustive (resp. separated) if (Gn)(G_n) and (Gn′)(G'_n) are.

Now let AA be a filtered ring and (An)(A_n) its filtration; on every subring BB of AA, clearly the filtration induced by that on AA is compatible with the ring structure on BB. If bb is a two-sided ideal of AA, the quotient filtration on A/bA/b of that on AA is compatible with the structure of this ring, for

(An+b)(Am+b)⊂An+m+b. (A_n + b)(A_m + b) \subset A_{n+m} + b.
If A′A' is another filtered ring, the product filtration on A×A′A \times A' is compatible with the structure of this ring.

Finally let EE be a filtered AA-module and (En)(E_n) its filtration; on every submodule FF of EE, the filtration induced by that on EE is compatible with the AA-module structure on FF and, on the quotient module E/FE/F, the quotient filtration of that on EE is compatible with the AA-module structure, as

An(F+Em)⊂F+AnEm⊂F+Em+n. A_n(F + E_m) \subset F + A_n E_m \subset F + E_{m+n}.

Note that if the filtration on E is derived from that on A (Example 2), so is the quotient filtration on E/F, but not in general the filtration induced on F (Exercise 1; see however , no. 2, Theorem 2)

If E' is another filtered A-module, the product filtration on E × E' is compatible with its A-module structure. If the filtrations on E and E' are derived from that on A (Example 2), so is their product filtration.

2. The order function

Let A be a filtered ring, E a filtered A-module and (E_n) the filtration of E. For all x ∈ E let v(x) denote the least upper bound in R of the set on integers n ∈ Z such that x ∈ E_n. Then the following equivalences hold:

{v(x)=−∞⇔x∉⋃n∈ZEnv(x)=p⇔x∈Ep and x∉Ep+1v(x)=+∞⇔x∈⋂n∈ZEn \begin{cases} v(x) = -\infty \Leftrightarrow x \notin \bigcup_{n \in \mathbf{Z}} E_n \\ v(x) = p \Leftrightarrow x \in E_p \text{ and } x \notin E_{p+1} \\ v(x) = +\infty \Leftrightarrow x \in \bigcap_{n \in \mathbf{Z}} E_n \end{cases}

The mapping v : E → \overline{\mathbf{R}} is called the order function of the filtered module E. If v is known then so are the E_n, for E_n is the set of x ∈ E such that v(x) ≥ n; the fact that the E_n are additive subgroups of E implies the relation

v(x−y)≥inf⁡(v(x),v(y)). v(x - y) \geq \inf(v(x), v(y)).

The above definition applies in particular to the filtered A-module A; let w be its order function. It follows from equation (3) of no. 1 that for a ∈ A and x ∈ E,

v(ax)≥w(a)+v(x) v(ax) \geq w(a) + v(x)
whenever the right-hand side is defined; in particular, for a ∈ A and b ∈ A,

w(ab)≥w(a)+w(b) w(ab) \geq w(a) + w(b)
whenever the right hand side is defined.

The order function is defined similarly on a filtered group G which is not necessarily commutative; the corresponding relation to (5) is then written

v(yx−1)=v(xy−1)≥inf⁡(v(x),v(y)). v(yx^{-1}) = v(xy^{-1}) \geq \inf(v(x), v(y)).

3. The graded module associated with a filtered module

Let G be a commutative group (written additively) and (G_n) a filtration on G. Let us write

gr⁡n(G)=Gn/Gn+1for n∈Zgr⁡(G)=⨁n∈Zgr⁡n(G). \begin{align*} \operatorname{gr}_n(G) &= G_n / G_{n+1} \quad \text{for } n \in \mathbf{Z} \\ \operatorname{gr}(G) &= \bigoplus_{n \in \mathbf{Z}} \operatorname{gr}_n(G). \end{align*}

The commutative group gr(G) is then a graded group of type Z\mathbf{Z}, called the graded group associated with the filtered group G, the homogeneous elements of degree nn of gr(G) being those of grn(G)\mathrm{gr}_n(G).

Now let A be a filtered ring, (A_n) its filtration, E a filtered A-module and (E_n) its filtration. For all p∈Zp \in \mathbf{Z}, q∈Zq \in \mathbf{Z}, a mapping

grp(A)×grq(E)→grp+q(E) \mathrm{gr}_p(A) \times \mathrm{gr}_q(E) \to \mathrm{gr}_{p+q}(E)

is defined as follows: given a∈grp(A)a \in \mathrm{gr}_p(A), ξ∈grq(E)\xi \in \mathrm{gr}_q(E), two representatives a,a′a, a' of aa and two representatives x,x′x, x' of ξ\xi, ax∈Ep+qax \in E_{p+q}, a′x′∈Ep+qa'x' \in E_{p+q} and ax≡a′x′ax \equiv a'x' (mod. Ep+q+1E_{p+q+1}), for

ax−a′x′=(a−a′)x+a′(x−x′) ax - a'x' = (a - a')x + a'(x - x')

and a−a′∈Ap+1a - a' \in A_{p+1} and x−x′∈Eq+1x - x' \in E_{q+1} and hence our assertion follows from formula (3) of no. 1. We may therefore denote by αξ\alpha\xi the class in

Ep+1/Ep+q+1=grp+q(E) E_{p+1}/E_{p+q+1} = \mathrm{gr}_{p+q}(E)

of the product axax of any representative a∈aa \in a and any representative x∈ξx \in \xi. It is immediate that the mapping (9) is Z-bilinear; by linearity, we derive a Z\mathbf{Z}-bilinear mapping

gr(A)×gr(E)→gr(E). \mathrm{gr}(A) \times \mathrm{gr}(E) \to \mathrm{gr}(E).

If this definition is first applied to the case E=A0E = A_0, the mapping (10) is an internal law of composition on gr(A)\mathrm{gr}(A), which it is immediately verified is associative and has an identity element which is the canonical image in gr0(A)\mathrm{gr}_0(A) of the unit element of A; it therefore defines on gr(A)\mathrm{gr}(A) a ring structure and the graduation (grn(A))n∈Z(\mathrm{gr}_n(A))_{n \in \mathbf{Z}} is by definition compatible with this structure. The graded ring gr(A)\mathrm{gr}(A) (of type Z\mathbf{Z}) thus defined is called the graded ring associated with the filtered ring A; it is obviously commutative if A is commutative; gr0(A)\mathrm{gr}_0(A) is a subring of gr(A)\mathrm{gr}(A). The mapping (10) is on the other hand a gr(A)\mathrm{gr}(A)-module external law on gr(E)\mathrm{gr}(E), the module axioms being trivially satisfied, and the graduation (grn(E))n∈Z(\mathrm{gr}_n(E))_{n \in \mathbf{Z}} on gr(E)\mathrm{gr}(E) is obviously compatible with this module structure. The graded gr(A)\mathrm{gr}(A)-module gr(E)\mathrm{gr}(E) (of type Z\mathbf{Z}) thus defined is called the graded module associated with the filtered A-module E.

Examples

Example 1

Let A be a commutative ring and tt an element of A which is not a divisor of 0. Let us give A the (t)-*adic filtration* (no. 1, Example 3). Then the associated graded ring gr(A)\mathrm{gr}(A) is canonically isomorphic to the polynomial ring (A/(t))[X](A/(t))[X]. For grn(A)=0\mathrm{gr}_n(A) = 0 for n<0n < 0 and by definition the ring gr0(A)\mathrm{gr}_0(A) is the ring A/(t)A/(t). We now note that by virtue of the hypothesis on tt the relation atn≡0at^n \equiv 0 (mod. tn+1t^{n+1}) is equivalent to a≡0a \equiv 0 (mod. tt); if τ\tau is the canonical image of tt in gr1(A)\mathrm{gr}_1(A), every element of grn(A)\mathrm{gr}_n(A) may then be written uniquely in the form ατn\alpha \tau^n where α∈gr0(A)\alpha \in \mathrm{gr}_0(A); whence our assertion.

Example 2

Let KK be a commutative ring, AA the ring of formal power series

K[[X1,…,Xr]] K[[X_1,\ldots,X_r]]

(Algebra, Chapter IV, § 5) and mm the ideal of AA whose elements are the formal power series with no constant term. Let us give AA the mm-adic filtration (no. 1, Example 3); if M1,…,MsM_1,\ldots,M_s are the distinct monomials in X1,…,XrX_1,\ldots,X_r of total degree n−1n-1, clearly every formal power series uu of total order ω(u)≥n\omega(u)\geq n

(loc. cit., no. 2)* may be written as ∑k=1sukMk\displaystyle\sum_{k=1}^{s}u_kM_k, where the uku_k belong to mm; it is seen that mnm^n is the set of formal power series uu such that ω(u)≥n\omega(u)\geq n, which shows that ω\omega is the orderfunction for the mm-adic filtration. Then clearly, for every formal power series u∈mnu\in m^n, there exists a unique homogeneous polynomial of degree nn in the XiX_i which is congruent to uu mod. mn+1m^{n+1}, namely the sum of terms of degree nn of uu; we conclude that gr⁡(A)\operatorname{gr}(A) is canonically isomorphic to the polynomial ring K[X1,…,Xr]K[X_1,\ldots,X_r].

Example 3

More generally, let AA be a commutative ring, bb an ideal of AA and AA be given the bb-adic filtration. If we write B=gr⁡0(A)B=\operatorname{gr}_0(A), F=gr⁡1(A)=b/b2F=\operatorname{gr}_1(A)=b/b^2, we know (Algebra, Chapter III), that the identity mapping of the BB-module FF onto itself can be extended uniquely to a homomorphism uu from the symmetric algebra S(F)S(F) of FF to the BB-algebra gr⁡(A)\operatorname{gr}(A); it follows from the definition of gr⁡(A)\operatorname{gr}(A) that uu is a surjective homomorphism of graded algebras; for n≥1n\geq1, every element of gr⁡n(A)\operatorname{gr}_n(A) is a sum of classes mod bn+1b^{n+1} of elements of the form y=x1x2⋯xny=x_1x_2\cdots x_n, where xi∈bx_i\in b (1≤i≤n)(1\leq i\leq n); if ξi\xi_i is the class of xix_i mod. b2b^2, clearly the class of yy mod. bn+1b^{n+1} is the element u(ξ1)⋯u(ξn)u(\xi_1)\cdots u(\xi_n), whence our assertion. In particular, every system of generators of the BB-module FF is a system of generators of the BB-algebra gr⁡(A)\operatorname{gr}(A).

If now EE is an AA-module and EE is given the bb-adic filtration, it is seen similarly that the graded gr⁡(A)\operatorname{gr}(A)-module gr⁡(E)\operatorname{gr}(E) is generated by gr⁡0(E)=E/bE\operatorname{gr}_0(E)=E/bE. To be precise, the restriction ϕ\phi to gr⁡(A)×gr⁡0(E)\operatorname{gr}(A)\times\operatorname{gr}_0(E) of the external law on the gr⁡(A)\operatorname{gr}(A)-module gr⁡(E)\operatorname{gr}(E) is a Z\mathbf{Z}-bilinear mapping of gr⁡(A)×gr⁡0(E)\operatorname{gr}(A)\times\operatorname{gr}_0(E) to gr⁡(E)\operatorname{gr}(E); moreover gr⁡(A)\operatorname{gr}(A) is a (gr⁡0(A),gr⁡0(A))(\operatorname{gr}_0(A),\operatorname{gr}_0(A))-bimodule and gr⁡0(E)\operatorname{gr}_0(E) a gr⁡0(A)\operatorname{gr}_0(A)-module; it is immediately verified that, for a∈gr⁡(A)a\in\operatorname{gr}(A), α0∈gr⁡0(A)\alpha_0\in\operatorname{gr}_0(A), ξ∈gr⁡0(E)\xi\in\operatorname{gr}_0(E),

ϕ(aα0,ξ)=ϕ(a,α0ξ) \phi(a\alpha_0,\xi)=\phi(a,\alpha_0\xi)

and hence ϕ\phi defines a surjective gr⁡1(A)\operatorname{gr}_1(A)-linear mapping

(11)

γE:gr⁡(A)⊗gr⁡0(A)gr⁡0(E)⟶gr⁡(E) \gamma_E:\operatorname{gr}(A)\otimes_{\operatorname{gr}_0(A)}\operatorname{gr}_0(E)\longrightarrow\operatorname{gr}(E)

which is called canonical.

Example 4

\* Let KK be a commutative ring, g\mathfrak g a Lie algebra over KK and UU the enveloping algebra of g\mathfrak g. An increasing filtration (Un)n∈Z(U_n)_{n\in\mathbf{Z}} is defined on UU by taking Un={0}U_n=\{0\} for n<0n<0 and denoting by UnU_n, for n≥0n\geq0 the set of elements of UU which can be expressed as a sum of products of at most nn elements of g\mathfrak g, then

U_0 = K and gr(U) is a commutative K-algebra (Lie Groups and Lie Algebras, Chapter I, § 2, no. 6). The canonical mapping of g to gr_1(U) = U_1/U_0 can be extended uniquely to a homomorphism h of the symmetric algebra S(g)S(g) of the K-module g to the K-algebra gr(U); the homomorphism h is surjective and, if the K-module g is free, h is bijective (loc. cit., no. 7, Theorem 1).

Example 5

Let A be a graded ring of type Z\mathbf{Z} and E a graded A-module of type Z\mathbf{Z}; let A(i)A_{(i)} (resp. E(i)E_{(i)}) be the subgroup of homogeneous elements of degree i of A (resp. E). Let A and E be given the filtrations associated with their graduations (no. 1, Example 1) and let A' and E' denote the filtered ring and filtered A-module thus obtained. Then it is immediate that the Z-linear mapping A→gr(A′)A \to \mathrm{gr}(A') which maps an element of A(n)A_{(n)} to its canonical image in

grn(A)=(⨁i≥nA(i))/(⨁i≥n+1A(i)) \mathrm{gr}_n(A) = \left( \bigoplus_{i \geq n} A_{(i)} \right) / \left( \bigoplus_{i \geq n+1} A_{(i)} \right)
is a graded ring isomorphism. A canonical graded A-module isomorphism E→gr(E′)E \to \mathrm{gr}(E') is defined similarly.

Proposition 1

Let A be a filtered ring, (An)n∈Z(A_n)_{n \in \mathbf{Z}} its filtration and v its orderfunction. Suppose that gr(A)\mathrm{gr}(A) is a ring with no divisor of zero. Then, for every ordered pair of elements a, b of the ring B=⋃n∈ZAnB = \bigcup_{n \in \mathbf{Z}} A_n, v(ab)=v(a)+v(b)v(ab) = v(a) + v(b).

As n=⋂n∈ZAnn = \bigcap_{n \in \mathbf{Z}} A_n is a two-sided ideal of the ring B, the formula holds if v(a)v(a) or v(b)v(b) is equal to +∞+\infty. If not, v(a)=rv(a) = r and v(b)=sv(b) = s are integers; the classes α\alpha of a mod. A, +, and β\beta of b mod. A, +_1 are ≠0\neq 0 by definition, whence by hypothesis αβ≠0\alpha \beta \neq 0 in gr(A)\mathrm{gr}(A) and therefore ab∉Ar+s+1ab \notin A_{r+s+1}, as ab∈Ar+sab \in A_{r+s},

v(ab)=v(a)+v(b). v(ab) = v(a) + v(b).

Corollary

Let A be a filtered ring and (An)n∈Z(A_n)_{n \in \mathbf{Z}} its filtration; let us set B=⋃n∈ZAnB = \bigcup_{n \in \mathbf{Z}} A_n, n=⋂n∈ZAnn = \bigcap_{n \in \mathbf{Z}} A_n. If the ring gr(A)\mathrm{gr}(A) has no divisors of zero, neither has the ring B/nB/n. If a and b are elements of B not belonging to n, then v(a)≠+∞v(a) \neq +\infty and v(b)≠+∞v(b) \neq +\infty, whence v(ab)≠+∞v(ab) \neq +\infty and therefore ab∉nab \notin n.

Note that the ring A can be an integral domain and the filtration (A_n) exhaustive and separated without gr(A)\mathrm{gr}(A) being an integral domain (Exercise 2).

Remark

Let G be a group which is not necessarily commutative with a filtration (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} such that Gn+1G_{n+1} is normal in G, for all n∈Zn \in \mathbf{Z}; again let grn(G)=Gn/Gn+1\mathrm{gr}_n(G) = G_n/G_{n+1}. The restricted product of the family (grn(G))n∈Z(\mathrm{gr}_n(G))_{n \in \mathbf{Z}}, that is the subgroup of the product ∏n∈Zgrn(G)\prod_{n \in \mathbf{Z}} \mathrm{gr}_n(G) consisting of the elements (ξn)(\xi_n) all of whose components, except at most a finite number, are equal to the identity element, is also called the graded group associated with G and denoted by gr(G)\mathrm{gr}(G).

4. Homomorphisms compatible with filtrations

Let G, G' be two commutative groups (written additively), (G_n) a filtration on G and (G'_n) a filtration on G'; a homomorphism h : G → G' is called compatible with the filtrations on G and G' if h(G_n) ⊂ G'_n for all n ∈ Z. The composite homomorphism G_n →hGn′⟶Gn′/Gn+1′\xrightarrow{h} G'_n \longrightarrow G'_n/G'_{n+1} is zero on G_{n+1} and hence defines by taking quotients a homomorphism h_n : G_n/G_{n+1} → G'_n/G'_{n+1}; there is therefore a unique additive group homomorphism gr(h) : gr(G) → gr(G') such that, for all n ∈ Z, gr(h) coincides with h_n on gr_n(G) = G_n/G_{n+1}. gr(h) is called the graded group homomorphism associated with h. If G'' is a third filtered group and h' : G' → G'' a homomorphism which is compatible with the filtrations, h' ∘ h is a homomorphism which is compatible with the filtrations and

(12)

gr⁡(h′∘h)=gr⁡(h′)∘gr⁡(h) \operatorname{gr}(h' \circ h) = \operatorname{gr}(h') \circ \operatorname{gr}(h)

Proposition 2

Let G be a filtered commutative group and H a subgroup of G; let H be given the induced filtration and G/H the quotient filtration. If j : H → G is the canonical injection and p : G → G/H the canonical surjection, j and p are compatible with the filtrations and the sequence

(13)

0⟶gr⁡(H)→gr⁡(j)gr⁡(G)→gr⁡(p)gr⁡(G/H)⟶0 0 \longrightarrow \operatorname{gr}(H) \xrightarrow{\operatorname{gr}(j)} \operatorname{gr}(G) \xrightarrow{\operatorname{gr}(p)} \operatorname{gr}(G/H) \longrightarrow 0
is exact.

The first assertion is obvious; if (G_n) is the filtration on G, then

(H∩Gn)∩Gn+1=H∩Gn+1 (H \cap G_n) \cap G_{n+1} = H \cap G_{n+1}
and hence gr(j) is injective; moreover the canonical mapping
Gn→(H+Gn)/H G_n \to (H + G_n)/H
is surjective, hence so is gr(p) and gr(p) ∘ gr(j) = 0 by (12). Finally, let ξ∈Gn/Gn+1\xi \in G_n/G_{n+1} belong to the kernel of gr(p); then there exists x ∈ ξ such that x ∈ H + G_{n+1}; but as G_{n+1} ⊂ G,
Gn∩(H+Gn+1)=(H∩Gn)+Gn+1 G_n \cap (H + G_{n+1}) = (H \cap G_n) + G_{n+1}
and hence x = y + z where y ∈ H ∩ G, and z ∈ G, +_1; this proves that ξ is the class mod. G, +_1 of j(y), in other words it belongs to the image of gr(H) under gr(j).

Note that, given an exact sequence 0 → G' \xrightarrow{u} G \xrightarrow{v} G'' → 0 of filtered commutative groups, where u and v are compatible with the filtrations, the sequence 0 \longrightarrow \operatorname{gr}(G') \xrightarrow{\operatorname{gr}(u)} \operatorname{gr}(G) \xrightarrow{\operatorname{gr}(v)} \operatorname{gr}(G'') \longrightarrow 0 is not necessarily exact (Exercise 4).

If now A and B are two filtered rings and h: A → B a ring homomorphism which is compatible with the filtrations, it is immediately verified that the graded group homomorphism gr(h): gr(A) → gr(B) is also a ring homomorphism. In particular, if A' is a subring of A with the induced filtration, gr(A') is canonically identified with a graded subring of gr(A) (Proposition 2); if b is a two-sided ideal of A and A/b is given the quotient filtration, gr(A/b) is canonically identified with the quotient graded ring gr(A)/gr(b) (Proposition 2).

Finally, let A be a filtered ring, E, F two filtered A-modules and u : E → F an homomorphism compatible with the filtrations. Then it is immediate that gr(u) : gr(E) → gr(F) is a gr(A)-linear mapping and hence a homogeneous homomorphism of degree 0 of graded gr(A)-modules. Moreover, if u' : E → F is another A-homomorphism compatible with the filtrations, so is u + u' and

gr⁡(u+u′)=gr⁡(u)+gr⁡(u′). \operatorname{gr}(u + u') = \operatorname{gr}(u) + \operatorname{gr}(u').

Remarks

(1) Clearly filtered ring homomorphisms (resp. homomorphisms of filtered modules over a given filtered ring A) compatible with the filtrations can be taken as morphisms for the filtered ring structure (resp. filtered A-module structure) (Set Theory, Chapter IV, § 2, no. 1).

(2) Let E and F be two modules over a filtered ring A and let them have the filtrations derived from the filtration (A_n) on A (no. 1, Example 2). Then every A-linear mapping u : E → F is compatible with the filtrations, since

u(AnE)=Anu(E)⊂AnF. u(A_nE) = A_nu(E) \subset A_nF.

(3) Note that a filtered A-module homomorphism u : E → F which is compatible with the filtrations may satisfy gr(u) = 0 without being zero; this is so for example of the endomorphism x ↦ nx of the additive group Z\mathbf{Z} with the (n)-adic filtration (for any integer n > 1). The relation gr(u) = gr(v) for two homomorphisms u, v of E to F, compatible with the filtrations, does not therefore imply necessarily u = u.

(4) The definitions at the beginning of this no. extend immediately to two groups G, G', which are not necessarily commutative and are filtered by subgroups Gn,Gn′G_n, G'_n such that Gn+1G_{n+1} (resp. Gn+1′G'_{n+1}) is normal in G, (resp. Gn′G'_n). Proposition 2 is also valid with the same hypotheses on the G, and assuming that H is invariant in G, the proof remaining unchanged except for notation.

5. The topology defined by a filtration

Let G be a group filtered by a family (Gn)n∈Z(G_n)_{n \in \mathbf{Z}} of normal subgroups of G. There exists a unique topology on G which is compatible with the group structure and for which the G_n constitute a fundamental system of neighbourhoods of the identity element e of G (General Topology, Chapter III, § 1, no. 2, Example); it is called the topology on G defined by the filtration (G_n). When we use topological notions concerning a filtered group, we shall mean, unless otherwise stated, with the topology defined by the filtration. Note that the G_n, being subgroups of G, are both open and closed (General Topology, Chapter III, § 2, no. 1, Corollary to Proposition 4).

As each G, is normal in G, the entourages of the left and right uniformities on G coincide; we deduce that G admits a Hausdorff completion group G^\hat{G} (General Topology, Chapter III, § 3, no. 4, Theorem 1 and no. 1, Proposition 2).

For every subset M of G, the closure of M in G is equal to

⋂n∈Z(M.Gn)=⋂n∈Z(Gn.M) \bigcap_{n \in \mathbf{Z}} (M . G_n) = \bigcap_{n \in \mathbf{Z}} (G_n . M)

(General Topology, Chapter III, § 3, no. 1, formula (1)); in particular ⋂n∈ZG\bigcap_{n \in \mathbf{Z}} G, is the closure of {e}\{e\}; thus it is seen that for the topology on G to be Hausdorff it is necessary and sufficient that the filtration (G_n) be separated. For the topology on G to be discrete, it is necessary and sufficient that there exist n∈Zn \in \mathbf{Z} such that G+={e}G_+ = \{e\} (in which case Gm={e}G_m = \{e\} for m≥nm \geq n); then the filtration (G_n) is called discrete.

Since the Hausdorff group associated with G is H=G/(⋂n∈ZGn)H = G / \left( \bigcap_{n \in \mathbf{Z}} G_n \right), the associated graded groups gr(G) and gr(H) (if H is given the quotient filtration) are canonically identified.

Not let G' be another filtered group and u:G→G′u : G \to G' a homomorphism compatible with the filtrations; the definition of the topologies on G and G' shows immediately that uu is continuous (*). If H is a subgroup (resp. normal subgroup) of G, the topology induced on H by that on G (resp. the quotient topology with respect to H of that on G) is the topology on H (resp. G/H) defined by the filtration induced by that on G (resp. quotient topology of that on G). The product topology of those on G and G' is the topology defined by the product of filtrations on G and G'.

Let v be the order function (no. 2) on G. The hypothesis on the G, implies that v(xyx−1)=v(y)v(xy x^{-1}) = v(y) and hence v(xy−1)=v(yx−1)=v(x−1y)=v(y−1x)v(xy^{-1}) = v(yx^{-1}) = v(x^{-1}y) = v(y^{-1}x) for all x, y in G. Let ρ\rho be a real number such that 0<ρ<10 < \rho < 1 (for example take

(*) Throughout this chapter we shall use the words "continuous homomorphism" in the sense of what is called "continuous representation" in General Topology, Chapter III, § 2, no. 8; the word "homomorphism" will never be used in the sense of General Topology, Chapter III, § 2, no. 8, Definition 1; for this notion we shall always use the term "strict morphism" in order to avoid any confusion.

ρ=1/e\rho = 1/e and let d(x,y)=ρv(xy−1)d(x, y) = \rho^{v(xy^{-1})} for all x,yx, y in G. Then d(x,x)=0,d(x,y)=d(y,x)d(x, x) = 0, d(x, y) = d(y, x) and inequality (5') of no. 2 gives

(15)d(x,y)≤sup⁡(d(x,z),d(y,z)) (15) \quad d(x, y) \leq \sup(d(x, z), d(y, z))

for all x,y,zx, y, z in G, which implies the triangle inequality

d(x,y)≤d(x,z)+d(y,z). d(x, y) \leq d(x, z) + d(y, z).

Thus dd is a pseudometric on G which is invariant under left and right translations and G_n is the set of x∈Gx \in G such that d(e,x)<ρnd(e, x) < \rho^n; the uniform structure defined by dd is then the uniform structure on the topological group G. If G is Hausdorff, G is a zero-dimensional metrizable topological space (General Topology, Chapter IX, § 6, no. 4); dd is a distance on G if also the filtration (G_n) is exhaustive.

Given a topological ring A, recall that a left topological A-module is an A-module E with a topology compatible with its additive group structure and such that the mapping (a,x)↦ax(a, x) \mapsto ax from A×EA \times E to E is continuous (General Topology, Chapter III, § 6, no. 6).

Proposition 3

Let A be a filtered ring, (A_n) its filtration, B the subring ⋃n∈ZAn\bigcup_{n \in \mathbf{Z}} A_n of A, E a filtered B-module, (E_n) its filtration and F the sub-B-module ⋃n∈ZEn\bigcup_{n \in \mathbf{Z}} E_n of E. Then the mapping (a,x)↦ax(a, x) \mapsto ax from B×FB \times F to F is continuous.

Let a,∈B,x0∈Fa, \in B, x_0 \in F; there exists by hypothesis integers r,sr, s such that a0∈Ara_0 \in A_r and x0∈Esx_0 \in E_s. The relation

ax−a0x0=(a−a0)x0+a0(x−x0)+(a−a0)(x−x0) ax - a_0x_0 = (a - a_0)x_0 + a_0(x - x_0) + (a - a_0)(x - x_0)

shows that if a−a0∈Aia - a_0 \in A_i and x−x0∈Ejx - x_0 \in E_j, ax−a0x0ax - a_0x_0 belongs to

Ei+s+Ej+r+Ei+j. E_{i+s} + E_{j+r} + E_{i+j}.

Then, given an integer nn, ax−a0x0∈Enax - a_0x_0 \in E_n provided i≥n−s,j≥n−ri \geq n - s, j \geq n - r and i+j≥ni + j \geq n, that is so long as ii and jj are sufficiently large.

Corollary

The ring B is a topological ring and the B-module F a topological B-module.

The first assertion is obtained by applying Proposition 3 to F=BF = B.

It is seen in particular that a filtered ring A whose filtration is exhaustive is a topological ring; if this is so every filtered A-module whose filtration is exhaustive is a topological A-module.

Proposition 4

Let A be a commutative ring filtered by an exhaustive filtration (A_n) and p an ideal of A. Suppose that the ideal gr⁡(p)=⨁n∈Z(p∩An)/(p∩An+1)\operatorname{gr}(p) = \bigoplus_{n \in \mathbf{Z}} (p \cap A_n)/(p \cap A_{n+1}) of the ring gr⁡(A)\operatorname{gr}(A) is prime. Then the closure of p in A is a prime ideal.

We know that gr(A/p) is isomorphic to gr(A)/gr(p) (no. 4, Proposition 2) and hence an integral domain; we conclude that A/⋂n∈Z(p+An)A / \bigcap_{n \in \mathbf{Z}} (p + A_n) is an integral domain (no. 3, Corollary to Proposition 1). Then the closure ⋂n∈Z(p+An)\bigcap_{n \in \mathbf{Z}} (p + A_n) of pp is a prime ideal.

Let AA be a ring and mm a two-sided ideal of AA; the topology defined on AA by the mm-adic filtration (no. 1, Example 3) is called the m-adic topology; as the m-adic filtration is exhaustive, AA is a topological ring with this topology (Corollary to Proposition 3). Similarly, for every AA-module EE, the topology defined by the mm-adic filtration is called the m-adic topology on EE; EE is a topological AA-module under this topology.

Let m′m' be another two-sided ideal of AA; for the m′m'-adic topology on AA to be finer than the mm-adic topology, it is necessary and sufficient that there exist an integer n>0n > 0 such that m′n⊂m{m'}^n \subset m; the condition is necessary and, if it is fulfilled, m′hn⊂mh{m'}^{hn} \subset m^h for all h>0h > 0 and hence the condition is sufficient. If AA is a commutative Noetherian ring, it amounts to the same to say that V(m)⊂V(m′)V(m) \subset V(m') in the prime spectrum of AA (Chapter II, § 4, no. 3, Corollary 2 to Proposition 11 and § 2, no. 6, Proposition 15).

6. Complete filtered modules

Proposition 5

Let GG be a filtered group whose filtration (Gn)(G_n) consists of invariant subgroups of GG. The following conditions are equivalent:

(a) GG is a complete topological group.

(b) The associated Hausdorff group G′=G/(⋂n∈ZGn)G' = G / (\bigcap_{n \in \mathbf{Z}} G_n) is complete.

(c) Every Cauchy sequence in GG is convergent.

If GG is commutative and written additively, these conditions are also equivalent to the following:

(d) Every family (xλ)λ∈L(x_\lambda)_{\lambda \in L} of elements of G′G' which converges to 0 with respect to the filter F\mathscr{F} of complements of finite subsets of LL is summable in G′G'.

For a filter on GG to be a Cauchy filter (resp. a convergent filter), it is necessary and sufficient that its image under the canonical mapping G→G′G \to G' be a Cauchy (resp. convergent) filter (General Topology, Chapter II, § 3, no. 1, Proposition 4); whence first of all the equivalence of (a) and (b); on the other hand, as G′G' is metrizable, the equivalence of (b) and (c) follows from Proposition 9 of General Topology, Chapter IX, § 2, no. 6.

Suppose now that GG is commutative. Suppose that G′G' is complete and let (xλ)λ∈L(x_\lambda)_{\lambda \in L} be a family of elements of G′G' which converge to 0 with respect to F\mathscr{F}. For every neighbourhood V′V' of 0 in GG which is a subgroup of G′G', there exists a finite subset JJ of LL such that the condition λ∈L−J\lambda \in L - J implies xλ∈V′x_\lambda \in V'; then ∑λ∈Hxλ∈V′\sum_{\lambda \in H} x_\lambda \in V' for every finite subset HH of LL not meeting JJ, which shows that the family (xλ)λ∈L(x_\lambda)_{\lambda \in L} is summable (General Topology, Chapter III, § 5, no. 2, Theorem 1).

Conversely, suppose that condition (d) holds and let (xn)(x_n) be a Cauchy sequence on G′G'; the family (xn+1−xn)(x_{n+1} - x_n) is then summable and in particular the series with general term xn+1−xnx_{n+1} - x_n is convergent and hence the sequence (xn)(x_n) is convergent.

Let GG be a filtered group whose filtration (Gr)(G_r) consists of normal subgroups of GG; the quotient groups G/GnG/G_n are discrete and hence complete, since the GrG_r are open in GG. Let fnf_n be the canonical mapping G→G/GnG \to G/G_n and for m≤nm \leq n let fmnf_{mn} be the canonical mapping G/Gn→G/GmG/G_n \to G/G_m; (G/Gn,fmn)(G/G_n, f_{mn}) is an inverse system of discrete groups with Z\mathbf{Z} as indexing set (General Topology, Chapter III, § 7, no. 3). Let G~\tilde{G} be the topological group the inverse limit of this inverse system and for all nn let gn:G~→G/Gng_n : \tilde{G} \to G/G_n be the canonical mapping; let f:G→G~f : G \to \tilde{G} be the inverse limit of the inverse system of mappings (fn)(f_n) such that fn=gn∘ff_n = g_n \circ f for all nn; finally, let jj be the canonical mapping of GG to its Hausdorff completion G^\hat{G}; as the G/GnG/G_n are complete, there exists a unique topological group isomorphism i:G^→G~i : \hat{G} \to \tilde{G} such that f=i∘jf = i \circ j (loc. cit., Corollary 1 to Proposition 2); we shall call it the canonical isomorphism of G^\hat{G} onto G~\tilde{G}.

Let HH be another filtered group whose filtration (Hr)(H_r) consists of normal subgroups of HH and let u:G→Hu : G \to H be a homomorphism compatible with the filtrations (no. 4). Set H~=lim⁡H/Hn\tilde{H} = \lim H/H_n; for all nn, uu defines by taking quotients a homomorphism un:G/Gn→H/Hnu_n : G/G_n \to H/H_n and the unu_n obviously form an inverse system of mappings; set u~=lim⁡un\tilde{u} = \lim u_n. Moreover let H^\hat{H} be the Hausdorff completion of HH and ti:G^→Hti : \hat{G} \to H the homomorphism derived from uu by passing to the Hausdorff completions (General Topology, Chapter II, § 3, no. 7, Proposition 15). It follows immediately from the definitions that if G^\hat{G} is identified with G~\tilde{G} and H^\hat{H} with H~\tilde{H} by means of the canonical isomorphisms, titi is identified with u~\tilde{u}. We conclude in particular that, if, for all nn, unu_n is an isomorphism, then titi is an isomorphism of topological groups.

Examples of complete filtered groups and rings

Example 1

Let GG be a complete filtered group. Every closed subgroup of GG with the induced filtration is complete (General Topology, Chapter II, § 3, no. 4, Proposition 8). Every quotient group of GG with the quotient filtration is complete (General Topology, Chapter IX, § 3, no. 1, Remark 1).

Example 2

Let AA be a filtered commutative ring whose filtration we denote by (an)n∈Z(a_n)_{n \in \mathbf{Z}}; let A′A' be the ring of formal power series A[[X1,…,Xs]]A[[X_1, \ldots, X_s]]. For all e=(e1,…,es)∈Nse = (e_1, \ldots, e_s) \in \mathbf{N}^s, we write ∣e∣=∑i=1sei|e| = \sum_{i=1}^s e_i, Xe=∏i=1sXieiX^e = \prod_{i=1}^s X_i^{e_i} so that every element P∈A′P \in A' can be written uniquely P=∑e∈Nsαe,PXeP = \sum_{e \in \mathbf{N}^s} \alpha_{e,P} X^e where αe,P∈A\alpha_{e,P} \in A. For all n∈Zn \in \mathbf{Z}, let an′a'_n denote the set of P∈A′P \in A' such that αe,P∈an−∣e∣\alpha_{e,P} \in a_{n-|e|} for all e∈Nse \in \mathbf{N}^s; we show that an′a'_n is an ideal of A′A'. Clearly an′a'_n is an additive subgroup of A′A'; on the other hand, if P∈an′P \in a'_n and Q∈A′Q \in A', then, for all e∈Nse \in \mathbf{N}^s, αe,PQ=∑e′+e′′=eαe′,Qαe′′,P\alpha_{e,PQ} = \sum_{e'+e''=e} \alpha_{e',Q}\alpha_{e'',P}; as the relation e′+e′′=ee' + e'' = e implies ∣e′′∣≤∣e∣|e''| \leq |e|, PQ∈an′PQ \in a'_n. Moreover, if Q∈am′Q \in a'_m, then, for e′+e′′=ee' + e'' = e, αe′,Qαe′′,P∈am−∣e′∣an−∣e′′∣⊂am+n−∣e∣\alpha_{e',Q}\alpha_{e'',P} \in a_{m-|e'|}a_{n-|e''|} \subset a_{m+n-|e|}, which proves that (an′)n∈Z(a'_n)_{n \in \mathbf{Z}} is a filtration compatible with the ring structure on A′A' (for obviously 1∈a0′1 \in a'_0). When in future we speak of A′A' as a filtered ring, we shall mean, unless otherwise stated, with the filtration (ah). Clearly ⋂n∈Zan′\bigcap_{n \in \mathbf{Z}} a'_n is the set of formal power series all of whose coefficients belong to ⋂n∈Zan\bigcap_{n \in \mathbf{Z}} a_n; then, if AA is Hausdorff, so is A′A'. If a0=Aa_0 = A, then a0′=A′a'_0 = A'.

Proposition 6

With the above notation, suppose that a0=Aa_0 = A and let hh denote the mapping P↦(αe,P)e∈NsP \mapsto (\alpha_{e,P})_{e \in \mathbf{N}^s}. Then hh is an isomorphism of the additive topological group AA onto the additive topological group ANsA^{\mathbf{N}^s}. The polynomial ring A[X1,…,Xs]A[X_1, \ldots, X_s] is dense in A′A'; if AA is complete, so is A′A'.

Clearly hh is bijective; Vn=h(an′)V_n = h(a'_n) is the set of (ae)∈ANs(a_e) \in A^{\mathbf{N}^s} such that ae∈an−∣e∣a_e \in a_{n-|e|} for all e∈Nse \in \mathbf{N}^s such that ∣e∣≤n|e| \leq n; as these elements ee are finite in number, VnV_n is a neighbourhood of 0 in ANsA^{\mathbf{N}^s}. Conversely, if VV is a neighbourhood of 0 in ANsA^{\mathbf{N}^s}, there is a finite subset EE of Ns\mathbf{N}^s and an integer vv such that the conditions ae∈ava_e \in a_v for all e∈Ee \in E imply (ae)∈V(a_e) \in V; if then nn is the greatest of the integers v+∣e∣v + |e| for e∈Ee \in E, then h(an′)⊂Vh(a'_n) \subset V, which proves the first assertion of Proposition 6.

Moreover, with nn and EE defined as above, h(P−∑e∈Eαe,PXe)∈Vh(P - \sum_{e \in E} \alpha_{e,P}X^e) \in V for all P∈A′P \in A', which shows that A[X1,…,Xs]A[X_1, \ldots, X_s] is dense in A′A'. The last assertion follows from the first and the fact that a product of complete spaces is complete.

Let mm be an ideal of AA and suppose that (ai)(a_i) is the mm-adic filtration; then, if nn is the ideal of A′A' generated by mm and the XiX_i (1≤i≤s1 \leq i \leq s), the filtration (ai)(a_i) is the nn-adic filtration. For clearly, for all k≥0k \geq 0, nkn^k is generated by the elements aXeaX^e such that a∈mk−∣e∣a \in m^{k-|e|} for all e∈Nse \in \mathbf{N}^s such that ∣e∣≤k|e| \leq k, whence nk⊂a&n^k \subset a \&. Let us prove conversely that ak′⊂nka'_k \subset n^k. For all P∈ak′P \in a'_k, P=P′+P′′P = P' + P'', where P′=∑∣e∣<kαe,PXeP' = \sum_{|e| < k} \alpha_{e,P}X^e, P′′=∑∣e∣≥kαe,PXeP'' = \sum_{|e| \geq k} \alpha_{e,P}X^e. Clearly it is possible to write P′′=∑∣e∣=kXeQeP'' = \sum_{|e|=k} X^e Q_e, where the QeQ_e are elements of A′A', whence P′′∈nkP'' \in n^k; on the other hand, clearly αe,PXe∈nk\alpha_{e,P}X^e \in n^k for all e∈Nse \in \mathbf{N}^s, whence P′∈nkP' \in n^k. Then nk=ak′n^k = a'_k.

Corollary

Let AA be a commutative ring,

A′=A[[X1,…,Xs]] A' = A[[X_1, \ldots, X_s]]

the ring of formal power series in ss indeterminates over AA and nn the ideal of A′A' consisting of the formal power series with no constant term. The ring A′A' is Hausdorff and complete with the n-adic topology and the polynomial ring A[X1,…,Xs]A[X_1, \ldots, X_s] is everywhere dense in A′A'.

It is sufficient to apply what has just been said to the case m={0}m = \{0\}.

7. Linear compactness properties of complete filtered modules

Recall that, if E is an A-module, an affine subset (or an affine linear variety) of E is any subset F which is empty or of the form a+Ma + M, where a∈Ea \in E and M is a submodule of E called the direction of F (Algebra, Chapter II, § 9, nos. 1 and 3).

Proposition 7

Let A be a filtered ring, E a filtered A-module and (En)(E_n) filtration of E; suppose that E0=EE_0 = E, that the EnE_n are submodules of E, that the A-modules E/EnE/E_n are Artinian and finally that the topological group E is Hausdorff and complete. Then the intersection of a decreasing sequence of non-empty closed affine subsets of E is non-empty.

We have seen in no. 6 that, since E is Hausdorff and complete, it is identified with E~=lim⁡←E/En\tilde{E} = \lim_{\leftarrow} E/E_n. Let (Wp)(W_p) be a decreasing sequence of non-empty closed affine subsets of E and, for all n≥0n \geq 0, let Wp,nW_{p,n} be the canonical image of WpW_p in E/EnE/E_n; we are going to construct a sequence x=(xn)∈E~x = (x_n) \in \tilde{E} such that xn∈Wp,nx_n \in W_{p,n} for all pp and all nn; hence x∈Wp+Enx \in W_p + E_n for all pp and all nn and, as the WpW_p are closed, x∈Wpx \in W_p for all pp (no. 5), which will prove the proposition.

As E/E0=0E/E_0 = 0, we shall take x0=0x_0 = 0. Suppose that the xix_i are defined for 0≤i≤n−10 \leq i \leq n - 1 and let Wp,n′W'_{p,n} be the set of elements of Wp,nW_{p,n} whose canonical image in E/En−1E/E_{n-1} is xn−1x_{n-1}; as xn−1∈Wp,n−1x_{n-1} \in W_{p,n-1} and Wp,n−1W_{p,n-1} is the canonical image of Wp,nW_{p,n}, Wp,n′W'_{p,n} is non-empty and is obviously an affine subset of E/EnE/E_n; moreover the sequence (Wp,n′)p∈N(W'_{p,n})_{p \in \mathbf{N}} is decreasing. As E/EnE/E_n is Artinian, this sequence is stationary (otherwise the sequence of submodules of E/EnE/E_n which are the directions of the Wp,n′W'_{p,n} would be strictly decreasing, which is absurd). It is sufficient then to take xnx_n to be an element of ⋂p∈NWp,n′\bigcap_{p \in \mathbf{N}} W'_{p,n} and the construction of (xn)(x_n) can then be performed by induction.

Proposition 8

Suppose that A and E satisfy the hypotheses of Proposition 7. Let (Fp)(F_p) be a decreasing sequence of closed submodules of E such that ⋂Fp=0\bigcap F_p = 0. Then, for every neighbourhood V of 0 in E, there exists pp such that Fp⊂VF_p \subset V (in other words, the base of the filter (Fp)(F_p) converges to 0).

We may assume that V is one of the EnE_n, in which case E/VE/V is Artinian. Let us write Fp′=(Fp+V)/VF'_p = (F_p + V)/V; as the Fp′F'_p form a decreasing sequence of submodules of E/VE/V, there exists an integer jj such that Fp′=Fj′F'_p = F'_j for all p>jp > j. We shall see that Fj′={0}F'_j = \{0\}, which will complete the proof. Let x∈Fj′x \in F'_j and let W, be the set of elements of F, whose image in E/VE/V is xx (p≥jp \geq j); by definition of j, the W, are non-empty closed affine subsets of E and obviously Wp+1⊂WW_{p+1} \subset W;

then it follows from Proposition 7 that there exists an element y belonging to all the W,. As W, ⊂F,\subset F, and ⋂p∈NFp={0},y=0\bigcap_{p \in \mathbf{N}} F_p = \{0\}, y = 0; since x is the canonical image of y in E/V, x=0x = 0 (cf. Exercises 15 to 21).

8. The lift of homomorphisms of associated graded modules

Theorem 1

Let X, Y be two filtered groups whose filtrations (Xn),(Yn)(X_n), (Y_n) consist of normal subgroups; let u:X→Yu : X \to Y be a homomorphism compatible with the filtrations.

(i) Suppose that the filtration (Xn)(X_n) is exhaustive. For gr(u)\mathrm{gr}(u) to be injective, it is necessary and sufficient that uˉ−1(Yn)=Xn\bar{u}^{-1}(Y_n) = X_n for all n∈Zn \in \mathbf{Z}.

(ii) Suppose that one of the following hypotheses holds: (α) X is complete and Y Hausdorff; (β) Y is discrete. Then, for gr(u)\mathrm{gr}(u) to be surjective, it is necessary and sufficient that Yn=u(Xn)Y_n = u(X_n) for all n∈Zn \in \mathbf{Z}.

(i) To say that the mapping grn(u)\mathrm{gr}_n(u) is injective means that

Xn∩uˉ−1(Yn+1)⊂Xn+1. X_n \cap \bar{u}^{-1}(Y_{n+1}) \subset X_{n+1}.
This is obviously the case if (Yn+1)=Xn+1(Y_{n+1}) = X_{n+1}. Conversely, if
Xn∩uˉ−1(Yn+1)⊂Xn+1 X_n \cap \bar{u}^{-1}(Y_{n+1}) \subset X_{n+1}
for all nn, we deduce by induction on kk that Xn−k∩uˉ−1(Yn+1)⊂Xn+1X_{n-k} \cap \bar{u}^{-1}(Y_{n+1}) \subset X_{n+1} for all n∈Zn \in \mathbf{Z} and all k≥0k \geq 0. As the filtration (Xn)(X_n) is exhaustive, we see that, for all nn, XnX_n is the union of the Xn−kX_{n-k} (k≥0k \geq 0), hence uˉ−1(Yn+1)⊂Xn+1\bar{u}^{-1}(Y_{n+1}) \subset X_{n+1} for all nn and therefore Xn+1⊂uˉ−1(Yn+1)X_{n+1} \subset \bar{u}^{-1}(Y_{n+1}), which completes the proof.

(ii) To say that the mapping grn(u)\mathrm{gr}_n(u) is surjective means that

Yn=u(Xn)Yn+1. Y_n = u(X_n)Y_{n+1}.
This is obviously the case if Yn=u(Xn)Yn+1Y_n = u(X_n)Y_{n+1} for all n∈Zn \in \mathbf{Z}. Let nn be an integer and yy an element of YnY_n; we shall define a sequence (xk)k≥0(x_k)_{k \geq 0} of elements of XnX_n such that xk∈Xnx_k \in X_n, xk+1≡xk(modXn+k)x_{k+1} \equiv x_k \pmod{X_{n+k}} and u(xk)≡y(modYn+k)u(x_k) \equiv y \pmod{Y_{n+k}} for all k≥0k \geq 0. We shall take x0x_0 equal to the identity element of X, which certainly gives u(x0)≡y(modYn)u(x_0) \equiv y \pmod{Y_n}. Suppose that an xk∈Xnx_k \in X_n has been constructed such that u(xk)≡y(modYn+k)u(x_k) \equiv y \pmod{Y_{n+k}}; then (u(xk))−1y∈Yn+k(u(x_k))^{-1}y \in Y_{n+k}; the hypothesis implies that there exists t∈Xn+kt \in X_{n+k} such that u(t)≡(u(xk))−1y(modYn+k+1)u(t) \equiv (u(x_k))^{-1}y \pmod{Y_{n+k+1}} and hence u(xkt)≡y(modYn+k+1)u(x_k t) \equiv y \pmod{Y_{n+k+1}}; it is sufficient to take xk+1=xktx_{k+1} = x_k t to carry out the induction. This being so, if Y is discrete, there exists k≥0k \geq 0 such that Yn+k={e′}Y_{n+k} = \{e'\} (the identity element of Y), whence u(xk)=yu(x_k) = y and hence in this case it has been proved that u(Xn)=Yu(X_n) = Y, for all nn. Suppose now that X is complete and Y Hausdorff. As xh−1xk∈Xn+kx_h^{-1} x_k \in X_{n+k} for h≥k≥0h \geq k \geq 0, (xk)(x_k) is a Cauchy sequence in XnX_n; as

X_n is closed in X and hence complete, this sequence has at least one limit x in X_n. By virtue of the continuity of u, u(x) is the unique limit of the sequence (u(x_k)) in Y, Y being Hausdorff. But the relations u(x_k) \equiv y (\text{mod. } Y_{n+k+1}) show that y is also a limit of this sequence, whence u(x)=yu(x) = y and it has also been proved that u(Xn)=Ynu(X_n) = Y_n.

Corollary 1

Suppose that X is Hausdorff and its filtration exhaustive. Then, if gr(u) is injective, u is injective.

Let e, e' be the identity elements of X and Y respectively. Then

uˉ−1(e′)⊂⋂nuˉ−1(Yn)=⋂nXn={e} \bar{u}^{-1}(e') \subset \bigcap_n \bar{u}^{-1}(Y_n) = \bigcap_n X_n = \{e\}
by hypothesis, whence the corollary.

Corollary 2

Suppose that one of the following hypotheses holds :* (α)(\alpha) X is complete, Y is Hausdorff and its filtration is exhaustive; (β)(\beta) Y is discrete and its filtration is exhaustive. *Then, if gr(u) is surjective, u is surjective.

In this case Y=⋃nYn=⋃nu(Xn)⊂u(X)Y = \bigcup_n Y_n = \bigcup_n u(X_n) \subset u(X).

Corollary 3

Suppose that X and Y are Hausdorff, the filtrations of X and Y exhaustive and X complete. Then, if gr(u) is bijective, u is bijective.

Let A be a local ring, m its maximal ideal and M an A-module; let A and M be given the m-adic filtrations and let gr(A) and gr(M) be the graded ring and the graded gr(A)-module associated with A and M. We have seen (no. 3, Example 3) that the canonical mapping (11) is always surjective; we are going to consider the following property of M: (GR) The canonical mapping

γM:gr⁡(A)⊗gr⁡0(A)gr⁡0(M)→gr⁡(M) \gamma_M : \operatorname{gr}(A) \otimes_{\operatorname{gr}_0(A)} \operatorname{gr}_0(M) \to \operatorname{gr}(M)
is bijective.

Proposition 9

Let A be a local ring, m its maximal ideal, M, N two A-modules and u : N \to M an A-homomorphism. M and N are given the m-adic filtrations and suppose that: (1) M satisfies property (GR); (2) \operatorname{gr}_0(u) : \operatorname{gr}_0(N) \to \operatorname{gr}_0(M) is injective. Then \operatorname{gr}(u) : \operatorname{gr}(N) \to \operatorname{gr}(M) is injective, N and P = \operatorname{Coker}(u) satisfy property (GR) and mnN=uˉ−1(mnM)m^n N = \bar{u}^{-1}(m^n M) for every integer n>0n > 0.

It is immediately verified that the diagram

gr⁡(A)⊗gr⁡0(A)gr⁡0(N)→1⊗gr⁡0(u)gr⁡(A)⊗gr⁡0(A)gr⁡0(M)↓γN↓γMgr⁡(N)→gr⁡(u)gr⁡(M) \begin{array}{ccc} \operatorname{gr}(A) \otimes_{\operatorname{gr}_0(A)} \operatorname{gr}_0(N) & \xrightarrow{1 \otimes \operatorname{gr}_0(u)} & \operatorname{gr}(A) \otimes_{\operatorname{gr}_0(A)} \operatorname{gr}_0(M) \\ \downarrow \gamma_N & & \downarrow \gamma_M \\ \operatorname{gr}(N) & \xrightarrow{\operatorname{gr}(u)} & \operatorname{gr}(M) \end{array}

is commutative. As gr0(A)=A/m\mathrm{gr}_0(A) = A/m is a field, the hypothesis implies that 1⊗gr0(u)1 \otimes \mathrm{gr}_0(u) is injective; as by hypothesis γM\gamma_M is injective, so is γM∘(1⊗gr0(u))\gamma_M \circ (1 \otimes \mathrm{gr}_0(u)). This implies first that γN\gamma_N is injective and hence bijective and therefore that gr(u)\mathrm{gr}(u) is injective. The formula u~1(mnM)=mnN\tilde{u}^1(m^nM) = m^nN is then a consequence of Theorem 1 (i).

Also, let us write N′=u(N)N' = u(N) and let j:N′→Mj : N' \to M be the canonical injection. If p:M→P=M/N′p : M \to P = M/N' is the canonical homomorphism, then in the commutative diagram

gr(A)⊗gr0(N′)→1⊗gr0(j)gr(A)⊗gr0(M)→1⊗gr0(p)gr(A)⊗gr0(P)→0↓γN′↓γM↓γPgr(N′)→gr(j)gr(M)→gr(p)gr(P)→0 \begin{array}{ccccccccc} \mathrm{gr}(A) \otimes \mathrm{gr}_0(N') & \xrightarrow{1 \otimes \mathrm{gr}_0(j)} & \mathrm{gr}(A) \otimes \mathrm{gr}_0(M) & \xrightarrow{1 \otimes \mathrm{gr}_0(p)} & \mathrm{gr}(A) \otimes \mathrm{gr}_0(P) & \to & 0 \\ \downarrow \gamma_{N'} & & \downarrow \gamma_M & & \downarrow \gamma_P & & \\ \mathrm{gr}(N') & \xrightarrow{\mathrm{gr}(j)} & \mathrm{gr}(M) & \xrightarrow{\mathrm{gr}(p)} & \mathrm{gr}(P) & \to & 0 \end{array}

the lower row is exact (no. 4, Proposition 2) and so is the upper row by virtue of Proposition 2 of no. 4 and the fact that gr0(A)\mathrm{gr}_0(A) is a field. Moreover, gr(j)\mathrm{gr}(j) is injective (no. 4, Proposition 2) and hence gr0(j)\mathrm{gr}_0(j) is injective, The first part of the argument applied to jj shows that γN′\gamma_{N'} is bijective; as γM\gamma_M is also bijective by hypothesis, we conclude that γP\gamma_P is bijective (Chapter I, § 1, no. 4, Corollary 2 to Proposition 2).

Corollary

Under the hypotheses of Proposition 9, if we assume also that NN is Hausdorf with the mm-adic filtration, then uu is injective.

This follows from the fact that gr(u)\mathrm{gr}(u) is injective (Corollary 1 to Theorem 1).

\* Remark. Suppose that the hypotheses of Proposition 9 hold and also one of the following conditions: (1) mm is nilpotent; (2) AA is Noetherian and PP is ideally Hausdorff (cf. § 5, no. 1); then PP is a flat AA-module. This follows from the fact that γP\gamma_P is bijective and § 5, no. 2, Theorem 1 (iv), since A/mA/m is a field. \*

9. The lift of families of elements of an associated graded module

Let AA be a filtered commutative ring, (An)n∈Z(A_n)_{n \in \mathbf{Z}} its filtration and CC a subring of AA, such that C∩A1={0}C \cap A_1 = \{0\}. The restriction to CC of the canonical mapping A0→A0/A1=gr0(A)A_0 \to A_0/A_1 = \mathrm{gr}_0(A) is then injective, which allows us to identify CC with a submodule of gr0(A)\mathrm{gr}_0(A); this is what we shall usually do in similar cases. If A1≠AA_1 \neq A, and KK is any subfield of AA, then K∩A1=(0)K \cap A_1 = (0) since K∩A1K \cap A_1 is an ideal of KK not containing 1 ; then KK may be identified with a subfield of gr0(A)\mathrm{gr}_0(A).

Proposition 10

Let AA be a filtered commutative ring and (An)(A_n) its filtration; suppose that there exists a subring CC of A0A_0 such that C∩An={0}C \cap A_n = \{0\} and CC is identified with a subring of gr0(A)\mathrm{gr}_0(A). Let (xi)1≤i≤q(x_i)_{1 \leq i \leq q} be a finite family of elements of AA; suppose that xi∈Anx_i \in A_n, for 1≤i≤q1 \leq i \leq q and let ξi\xi_i be the class of xix_i in grni(A)\mathrm{gr}_{n_i}(A) for 1≤i≤q1 \leq i \leq q.

(i) If the family (ξi)(\xi_i) of elements of gr(A)\mathrm{gr}(A) is algebraically free over CC, the family (xi)(x_i) is algebraically free over CC.

(ii) If the filtration on AA is exhaustive and discrete and (ξi)(\xi_i) is a system of generators of the CC-algebra gr(A)\mathrm{gr}(A), then (xi)(x_i) is a system of generators of the CC-algebra AA.

Let A′A' be the polynomial algebra C[X1,…,Xq]\mathbf{C}[X_1, \ldots, X_q] over C\mathbf{C}; let A′A' be given the graduation (An′)(A'_n) of type Z\mathbf{Z} where AnA_n is the set of C\mathbf{C}-linear combinations of the monomials X1s(1)…Xqs(q)X_1^{s(1)} \ldots X_q^{s(q)} such that ∑i=1qnis(i)=n\sum_{i=1}^q n_i s(i) = n. Let uu be the homomorphism f↦f(x1,…,xq)f \mapsto f(x_1, \ldots, x_q) from the C\mathbf{C}-algebra A′A' to the C\mathbf{C}-algebra AA; by definition, u(An′)⊂Anu(A'_n) \subset A_n for all n∈Zn \in \mathbf{Z} and hence uu is compatible with the filtrations (A′)(A') being given its filtered ring structure associated with its graded ring structure, cf. no. 1, Example 1). This being so, the hypothesis of (i) means that

gr(u):A′=gr(A′)→gr(A) \mathrm{gr}(u): A' = \mathrm{gr}(A') \to \mathrm{gr}(A)

is injective; as the filtration on A′A' is exhaustive and separated, Corollary 1 to Theorem 1 of no. 8 may be applied and uu is injective, which proves the conclusion of (i). Similarly, the hypothesis (ii) on the (\&) means that gr(u)\mathrm{gr}(u) is surjective; as AA is discrete and its filtration is exhaustive, Corollary 2 to Theorem 1 of no. 8 may be applied and uu is surjective, which proves the conclusion of (ii).

Proposition 11

Let AA be a complete Hausdorff filtered commutative ring, CC a subring of A0A_0 such that C∩An=0C \cap A_n = 0 and (xi)1≤i≤q(x_i)_{1 \leq i \leq q} a finite family of elements of AA such that xi∈Anix_i \in A_{n_i} where ni>0n_i > 0 for 1≤i≤q1 \leq i \leq q; let ξi\xi_i be the class of xix_i in grni(A)\mathrm{gr}_{n_i}(A) for 1≤i≤q1 \leq i \leq q.

(i) There exists a unique CC-homomorphism vv from the algebra of formal power series A′′=C[[X1,…,Xq]]A'' = \mathbf{C}[[X_1, \ldots, X_q]] to AA such that v(Xi)=xiv(X_i) = x_i for 1≤i≤q1 \leq i \leq q.

(ii) If the family (ξi)(\xi_i) is algebraically free over CC, the homomorphism vv is injective.

(iii) If the filtration on AA is exhaustive and the family (ξi)(\xi_i) is a system of generators of the CC-algebra gr(A)\mathrm{gr}(A), the homomorphism vv is surjective.

As ni≥1n_i \geq 1 for all ii, ∑i=1qnis(i)≥∑i=1qs(i)\sum_{i=1}^q n_i s(i) \geq \sum_{i=1}^q s(i) for every monomial X1s(1)…Xqs(q)X_1^{s(1)} \ldots X_q^{s(q)} and on the other hand ∑i=1qnis(i)≤r⋅∑i=1qs(i)\sum_{i=1}^q n_i s(i) \leq r \cdot \sum_{i=1}^q s(i) if rr is the greatest of the nin_i. If An′′A''_n denotes the set of formal power series whose non-zero terms asX1s(1)…Xqs(q)a_s X_1^{s(1)} \ldots X_q^{s(q)} satisfy ∑i=1qnis(i)≥n\sum_{i=1}^q n_i s(i) \geq n, it follows from no. 6, Corollary to Proposition 6 that A′′A'' is Hausdorff and complete with the exhaustive filtration (An′′)(A''_n) and that

A′=C[X1,…,Xq] A' = \mathbf{C}[X_1, \ldots, X_q]

is dense in A′′A''; moreover the homomorphism uu defined in the proof of Proposition 10 is continuous on A′A' and can be extended uniquely to a continuous homomorphism v:A→Av : A \to A, since AA is Hausdorff and complete (General Topology, Chapter III, § 3, no. 3, Proposition 5), which proves (i); also, gr(A′′)=gr(A′)\mathrm{gr}(A'') = \mathrm{gr}(A') and gr(v)=gr(u)\mathrm{gr}(v) = \mathrm{gr}(u); (ii) and (iii) then follow respectively from Corollaries 1 and 2 to Theorem 1 of no. 8 in view of the hypotheses on AA.

The conclusion of (ii) (resp. (iii)) of the proposition is sometimes expressed by saying that the family (xi)(x_i) is formally free over CC (resp. a formal system of generators of AA).

Proposition 12

Let AA be a filtered ring, EE a filtered AA-module and (An)(A_n) and (En)(E_n) the respective filtrations on AA and EE. Suppose that AA is complete and the filtration (En)(E_n) is exhaustive and separated. Let (xi)i∈I(x_i)_{i \in I} be a finite family of elements of EE and for i∈Ii \in I let n(i)n(i) be an integer such that xi∈En(i)x_i \in E_{n(i)}; finally let ξi\xi_i be the class of xix_i in grn(i)(E)\mathrm{gr}_{n(i)}(E). Then, if (ξi)(\xi_i) is a system of generators of the gr(A)\mathrm{gr}(A)-module gr(E)\mathrm{gr}(E), (xi)(x_i) is a system of generators of the AA-module EE.

In the AA-module L=A′L = A', let L0L_0 denote the set (ai)(a_i) such that ai∈An−n(i)a_i \in A_{n - n(i)} for all i∈Ii \in I; if pp and qq are the least and greatest of the n(i)n(i), then An−qI⊃L0⊃An−pIA^I_{n-q} \supset L_0 \supset A^I_{n-p} and the topology defined on LL by the definition (Ln)(L_n) is the same as the product topology; hence LL is a complete filtered AA-module. As LL is free, there exists an AA-linear mapping u:L→Eu : L \to E such that u((ai))=∑i∈Iaixiu((a_i)) = \sum_{i \in I} a_i x_i and it is obviously compatible with the filtrations; we must prove that uu is surjective and for this it is sufficient, by virtue of Corollary 2 to Theorem 1, no. 8, to show that

gr(u):gr(L)→gr(E) \mathrm{gr}(u) : \mathrm{gr}(L) \to \mathrm{gr}(E)

is surjective or also that, for all x∈Enx \in E_n, there exist a family (ai)(a_i) such that ai∈An−n(i)a_i \in A_{n-n(i)} for all i∈Ii \in I and x≡∑i∈Iaixi(modEn+1)x \equiv \sum_{i \in I} a_i x_i \pmod{E_{n+1}}. Let ξ\xi be the class of xx in grn(E)\mathrm{gr}_n(E); since the ξi\xi_i generate the gr(A)\mathrm{gr}(A)-module gr(E)\mathrm{gr}(E), there exist αi∈gr(A)\alpha_i \in \mathrm{gr}(A) such that ξ=∑i∈Iαiξi\xi = \sum_{i \in I} \alpha_i \xi_i and we may assume that αi∈grn−n(i)(A)\alpha_i \in \mathrm{gr}_{n-n(i)}(A) by replacing if need be αi\alpha_i by its homogeneous component of degree n−n(i)n - n(i). Then αi\alpha_i is the image of an element ai∈An−n(i)a_i \in A_{n-n(i)} and the family (ai)(a_i) has the required property.

Corollary 1

Let AA be a complete filtered ring and EE a filtered AA-module whose filtration is exhaustive and separated. If gr(E)\mathrm{gr}(E) is a finitely generated (resp. Noetherian) gr(A)\mathrm{gr}(A)-module, then EE is a finitely generated (resp. Noetherian) AA-module.

If gr(E)\mathrm{gr}(E) is finitely generated, there is a finite system of homogeneous generators and Proposition 12 shows that EE is finitely generated. Suppose now that gr(E)\mathrm{gr}(E) is Noetherian and let FF be a submodule of EE; the filtration induced on FF by that on EE is exhaustive and separated and gr(F)\mathrm{gr}(F) is identified with a sub-gr(A)-module of gr(E) (no. 4, Proposition 2) and hence is finitely generated by hypothesis; we conclude that F is a finitely generated A-module and hence E is Noetherian.

Corollary 2

Let A be a complete Hausdorff filtered ring whose filtration is exhaustive. If gr(A) is a left Noetherian ring, so is A.

It is sufficient to apply Corollary 1 with E = A.

Corollary 3

Let A be a complete filtered ring, (A_n) its filtration, E a Hausdorff filtered A-module, (E_n) its filtration and F a finitely generated submodule of E; suppose that A_∞ = A and E_0 = E.

(i) If, for all k ≥ 0, E_k = E_{k+1} + A_k F, then F = E. (ii) If it is further supposed that the filtration on E is derived from that on A (no. 1, Example 2), the relation E = E_1 + F implies F = E.

Let ξ_i (1 ≤ i ≤ n) be the classes mod. E_1 of a finite system of generators of F. It follows from the given hypothesis that for all k ≥ 0 every element of gr_k(E) can be expressed in the form ∑i=1nαiξi\sum_{i=1}^n \alpha_i \xi_i where αi∈gr(A)\alpha_i \in \mathrm{gr}(A); the ξi\xi_i therefore generate the gr(A)-module gr(E), which proves (i) by virtue of Proposition 12. If the filtration on E is derived from that on A, the relation E = E_1 + F implies

Ek=AkE=AkE1+AkF=AkA1E+AkF⊂Ak+1E+AkF=Ek+1+AkF⊂Ek, E_k = A_k E = A_k E_1 + A_k F = A_k A_1 E + A_k F \subset A_{k+1} E + A_k F \\ = E_{k+1} + A_k F \subset E_k,

whence (ii).

Proposition 13

Let A be a ring, m a two-sided ideal of A contained in the Jacobson radical of A and E an A-module. Let A and E be given the m-adic filtrations (no. 1, Example 3). Suppose that one of the following conditions holds: (a) E is a finitely generated A-module and A is Hausdorff; (b) m is nilpotent.

For E to be a free A-module, it is necessary and sufficient that E/mE be a free (A/m)-module and that E satisfy property (GR) (no. 8).

If E is a free A-module and (eλ)(e_\lambda) a basis of E, mkEm^k E is the direct sum of the submodules mkeλm^k e_\lambda of E for all k>0k > 0 (Algebra, Chapter II, § 3, no. 7, Remark); then mkE/mk+1Em^k E / m^{k+1} E is identified with the direct sum of the mkeλ/mk+1eλm^k e_\lambda / m^{k+1} e_\lambda (Algebra, Chapter 11, § 1, no. 6, Proposition 7). We deduce first (for k=0k = 0) that the classes 1⊗eλ1 \otimes e_\lambda of the eλe_\lambda in E/mE=(A/m)⊗AEE / mE = (A / m) \otimes_A E form a basis of the (A/m)-module E/mEE / mE, since the canonical mapping

(mk/mk+1)⊗A(E/mE)→mkE/mk+1E (m^k / m^{k+1}) \otimes_A (E / mE) \to m^k E / m^{k+1} E

is bijective for all k≥0k \geq 0; hence γE\gamma_E is bijective. Note that this part of the proof uses neither condition (a) nor condition (b).

Suppose conversely that the conditions of the statement hold and let (xi)i∈I(x_i)_{i \in I} be a family of elements of E whose classes mod. mEmE form a basis of the (A/m)(A/m)-module E/mEE/mE; let L be the free A-module As(I)A_s^{(I)}, (fi)i∈I(f_i)_{i \in I} its canonical basis and u:L→Eu : L \to E the A-linear mapping such that u(fi)=xiu(f_i) = x_i for all i∈Ii \in I. The hypotheses already imply that uu is surjective (Chapter II, § 3, no. 2, Corollary 1 to Proposition 4) and it remains to prove that uu is injective. Now, each of the hypotheses (a) and (b) implies that A is Hausdorff and hence so is L with the m-adic filtration, since mkL=(mk)(I)m^kL = (m^k)^{(I)} (Algebra, Chapter II, § 3, no. 7, Remark) and gr⁡(L)\operatorname{gr}(L) is identified with gr⁡(A)⊗A/m(L/mL)\operatorname{gr}(A) \otimes_{A/m} (L/mL) from the first part of the proof; the homomorphism uu is compatible with the filtrations and it is possible to write gr⁡(u)=γE∘v\operatorname{gr}(u) = \gamma_E \circ v where vv is the bijection of gr⁡(L)\operatorname{gr}(L) onto

gr⁡(A)⊗A/m(E/mE) \operatorname{gr}(A) \otimes_{A/m} (E/mE)
mapping the class off, mod. mM onto 1⊗xˉi1 \otimes \bar{x}_i, where xˉi\bar{x}_i is the class of xix_i mod. mEmE. The hypothesis then implies that gr⁡(u)\operatorname{gr}(u) is injective and the conclusion follows with the aid of Corollary 1 to Theorem 1, no. 8.

10. Application: examples of noetherian rings

Lemma 1

Let A be a graded ring of type Z\mathbf{Z}, whose graduation (An)(A_n) is such that An=0A_n = 0 for all n<0n < 0 or An=0A_n = 0 for all n>0n > 0. Let M be a graded A-module of type Z\mathbf{Z}. For M to be a Noetherian A-module, it is necessary and sufficient that every graded submodule of M be finitely generated.

As n↦−nn \mapsto -n is an automorphism of the group Z\mathbf{Z}, we may restrict our attention to the case An=0A_n = 0 for all n>0n > 0. Let A' and M' denote the ring A and the module M with the filtrations associated with their respective graduations (no. 1, Example 1), which are exhaustive and separated; the hypothesis on A implies that A' is discrete and hence complete. If E is a sub-A-module of M, the filtered A'-module E' obtained by giving E the induced filtration is Hausdorff and its filtration is exhaustive; moreover gr⁡(E′)\operatorname{gr}(E') is identified with a graded sub-A-module of M=gr⁡(M′)M = \operatorname{gr}(M') and hence is finitely generated by hypothesis. The conclusion then follows from Corollary 1 to Proposition 12 of no. 9.

Theorem 2

Let A be a graded ring of type N\mathbf{N}, M a graded A-module of type N\mathbf{N} and (An)(A_n) and (Mn)(M_n) their respective graduations. Suppose that there exists an element a∈A0a \in A_0 such that An=A0anA_n = A_0 a^n and Mn=anM0M_n = a^n M_0 for all n>0n > 0. Then, if M, is a Noetherian A0A_0-module, M is a Noetherian A-module.

By virtue of Lemma 1 it is sufficient to prove that every graded submodule N of M is finitely generated. For all r≥0r \geq 0, let Nr=N∩MrN_r = N \cap M_r and let L_r be the set of m∈M0m \in M_0 such that arm∈Nra^r m \in N_r. As

arA0⊂Ar=A0ar,arA0Lr⊂A0arLr⊂A0Nr⊂Nr a^r A_0 \subset A_r = A_0 a^r, \quad a^r A_0 L_r \subset A_0 a^r L_r \subset A_0 N_r \subset N_r

and hence the L, are sub-A,-modules of M,; moreover,

aNr⊂N∩aMr=N∩Mr+1=Nr+1 aN_r \subset N \cap aM_r = N \cap M_{r+1} = N_{r+1}

and hence the sequence (Lr)r≥0(L_r)_{r \geq 0} is increasing. The hypothesis implies that there exists an integer n≥0n \geq 0 such that Lr=LnL_r = L_n for r≥nr \geq n. For each r≤nr \leq n, let (mr,s)1≤s≤kr(m_{r,s})_{1 \leq s \leq k_r} be a system of generators of the A,-module LrL_r. We shall show that the elements armr,sa^r m_{r,s} for 1≤s≤kr1 \leq s \leq k_r, 0≤r≤n0 \leq r \leq n form a system of generators of the A-module N. As Mr=arM0M_r = a^r M_0 for all rr, Nr=arLrN_r = a^r L_r for all rr by definition of LrL_r. Then, for r≤nr \leq n,

Nr=arLr=∑s=1krarA0mr,s⊂∑s=1krA0armrs, N_r = a^r L_r = \sum_{s=1}^{k_r} a^r A_0 m_{r,s} \subset \sum_{s=1}^{k_r} A_0 a^r m_{rs},

and, for r>nr > n,

Nr=arLn=∑s=1knarA0mn,s⊂∑s=1knA0armn,s⊂∑s=1knA0ar−n⋅(anmn,s) N_r = a^r L_n = \sum_{s=1}^{k_n} a^r A_0 m_{n,s} \subset \sum_{s=1}^{k_n} A_0 a^r m_{n,s} \subset \sum_{s=1}^{k_n} A_0 a^{r-n} \cdot (a^n m_{n,s})

which completes the proof (cf. Exercise 10).

Corollary 1 (Hilbert’s Theorem)

For every commutative Noetherian ring C, the polynomial ring C[X] is Noetherian* (cf. Exercise 10).

Corollary 2

For every commutative Noetherian ring C and every integer n>0n > 0, the polynomial ring C[X_1, ..., X_n] is Noetherian.

This follows from Corollary 1 by induction on nn.

Corollary 3

If C is a commutative Noetherian ring, every finitely generated commutative C-algebra is a Noetherian ring.

Such an algebra is isomorphic to a quotient of a polynomial algebra C[X_1, ..., X_n] (§ 1, no. 1).

Corollary 4

Let A be a graded commutative ring of type N\mathbf{N} and let (An)(A_n) be its graduation. For A to be Noetherian, it is necessary and sufficient that AnA_n be Noetherian and that A be ajinitely generated A,-algebra.

The condition is sufficient by Corollary 3. Conversely, suppose A is Noetherian; m=∑n≥1Anm = \sum_{n \geq 1} A_n which is an ideal of A, is then finitely generated; then A is a finitely generated A,-algebra (§ 1, no. 2, Corollary to Proposition 1); on the other hand AnA_n, which is isomorphic to A/mA/m, is a Noetherian ring.

Corollary 5

Let A be a commutative ring and m an ideal of A such that A/mA/m is Noetherian, m/m2m/m^2 is ajinitely generated (A/m)(A/m)-module and A is Hausdorff and complete with the m-adic topology. Then gr(A) and A are Noetherian.

gr(A) is an (A/m)(A/m)-module generated by m/m2m/m^2 (no. 3, Example 3) and hence the ring gr(A) is Noetherian by Corollary 3. From this we deduce that A itself is Noetherian (no. 9, Corollary 2 to Proposition 12).

Corollary 6

For every commutative Noetherian ring C and every integer n > 0, the ring of formal power series C[[X1,…,Xn]]\mathbf{C}[[X_1, \ldots, X_n]] is Noetherian.

This follows from Corollary 5 and no. 6, Corollary to Proposition 6, for if m is the ideal of A=C[[X1,…,Xn]]A = \mathbf{C}[[X_1, \ldots, X_n]] consisting of the formal power series with no constant term, A/mA/m is isomorphic to C and m/m2m/m^2 to the C-module C"\mathbf{C}".

Remarks

(1) Corollaries 2, 3 and 6 apply in particular if C is a commutative field.

\* (2) Let g be a Lie algebra over a commutative Noetherian ring C and suppose that g is a finitely generated C-module. Let the enveloping algebra U of g be given the increasing filtration (Un)(U_n) defined in no. 3, Example 4. With the corresponding topology, U is discrete and hence Hausdorff and complete; the associated graded ring gr(U) is a finitely generated C-algebra, being a quotient of the symmetric algebra S(g)\mathscr{S}(g), hence gr(U) is a Noetherian ring (Corollary 3) and we deduce that U is a left and right Noetherian ring (no. 9, Corollary 2 to Proposition 12). \*

11. Complete m-adic rings and inverse limits

We have seen in no. 6 that, if A is a commutative ring and m an ideal of A such that A is Hausdorf and complete with the m-adic topology, then the topological ring AA is canonically identified with the inverse limit of the discrete rings Ai=A/mi+1 (i∈N)A_i = A/m^{i+1} \ (i \in \mathbf{N}) with respect to the canonical mappings

hij:A/mj+1→A/mi+1(i≤j); h_{ij}: A/m^{j+1} \to A/m^{i+1} \quad (i \leq j);

note that hijh_{ij} is surjective and that, if nijn_{ij} is its kernel, then

nij=mi+1/mj+1=(m/mj+1)i+1= n_{ij} = m^{i+1}/m^{j+1} = (m/m^{j+1})^{i+1} =

in particular (n0j)j+1=0(n_{0j})^{j+1} = 0. Conversely:

Proposition 14

Let (Ai,hij)(A_i, h_{ij}) be an inverse system of discrete commutative rings, whose indexing set is N\mathbf{N} and let (Mi,uij)(M_i, u_{ij}) be an inverse system of modules over the inverse system of rings (Ai,hij)(A_i, h_{ij}). Let njn_j denote the kernel of h0j:A→Ah_{0j}: A \to A, and set A=lim⁡AiA = \lim A_i, M=lim⁡MiM = \lim M_i. Suppose that

(a) for all i∈Ni \in \mathbf{N}, hiih_{ii} is the identity mapping on A, and, for i≤ji \leq j, hijh_{ij} and uiju_{ij} are surjective; (b) for i≤ji \leq j, the kernels of hijh_{ij} and uiju_{ij} are nji+1n_j^{i+1} and nji+1Mjn_j^{i+1} M_j respectively.

Then: (i) AA is a complete Hausdorff topological ring, MM is a complete Hausdorff topological A-module and the canonical mappings hi:A→Aih_i : A \to A_i, ui:M→Miu_i : M \to M_i are surjective. (ii) If M0M_0 is a finitely generated A,-module, MM is a finitely generated A-module;

to be precise, every finite subset S of M such that u0(S)u_0(S) generates M, is a system of generators of M.

The assertions in (i) follow from General Topology, Chapter II, § 3, no. 5, Corollary to Proposition 10 and Corollary 1 to Theorem 1.

For all i∈Ni \in \mathbf{N}, let us write mi+1=Ker(hi),Ni+1=Ker(ui)m_{i+1} = \mathrm{Ker}(h_i), N_{i+1} = \mathrm{Ker}(u_i); then

mi+1=lim⁡k≥0hi,t+k−1(0)=lim⁡←ni+1t+k m_{i+1} = \lim_{k \geq 0} h_{i,t+k}^{-1}(0) = \lim_{\leftarrow} n_{i+1}^{t+k}

and Ni+1=lim⁡←ni+kt+1Mi+kN_{i+1} = \lim_{\leftarrow} n_{i+k}^{t+1} M_{i+k}; as hi+kh_{i+k} and ui+ku_{i+k} are surjective,

hi+k(mi+1)=ni+kt+1,ui+k(Ni+1)=ni+kt+1Mi+k. h_{i+k}(m_{i+1}) = n_{i+k}^{t+1}, \quad u_{i+k}(N_{i+1}) = n_{i+k}^{t+1} M_{i+k}.

Let us show that miNj⊂Ni+jm_i N_j \subset N_{i+j} for i≥1i \geq 1 and j≥1j \geq 1, which amounts to proving that ui+j−1(miNj)=0u_{i+j-1}(m_i N_j) = 0; now

ui+j−1(miNj)=hi+j−1(mi)ui+j−1(Nj) u_{i+j-1}(m_i N_j) = h_{i+j-1}(m_i) u_{i+j-1}(N_j)

is equal to ni+j−1t′(ni+j−1t′Mi+j−1)=0n_{i+j-1}^{t'}(n_{i+j-1}^{t'} M_{i+j-1}) = 0, as, for all k≥0k \geq 0, nkk+1n_k^{k+1}, which is the kernel of hkkh_{kk}, is equal to 0. We see similarly that mimj⊂mi+jm_i m_j \subset m_{i+j}. If for i≤0i \leq 0 we set mi=Am_i = A and Ni=MN_i = M, (mi)i∈Z(m_i)_{i \in \mathbf{Z}} is a filtration of A and (Ni)i∈Z(N_i)_{i \in \mathbf{Z}} a filtration of M compatible with the filtration on A; the topologies on A and M are obviously those defined by these filtrations. This being so, let aa be an ideal of A such that h1(a)=n1h_1(a) = n_1 and M′M' the submodule of M generated by S; we are going to prove that

Ni=aiM′+Ni+1for i≥0. N_i = a^i M' + N_{i+1} \quad \text{for } i \geq 0.

Let us write ai=hi(a),Mi′=ui(M′)a_i = h_i(a), M'_i = u_i(M'); it is sufficient to show that

ui(Ni)=aiiMi′. u_i(N_i) = a_i^i M'_i.

This is true if i=0i = 0, for N0=MN_0 = M and M0′=MM'_0 = M, by hypothesis. If i≥1i \geq 1, then ui(Ni)=niiMiu_i(N_i) = n_i^i M_i by (16). As h1ih_{1i} is surjective and h0i=h01∘h1ih_{0i} = h_{01} \circ h_{1i}, h1ih_{1i} maps the kernel nin_i of hih_i, onto the kernel n1n_1 of h1h_1, and ni=h1i(n1)n_i = h_{1i}(n_1); then

h1i(ai)=h1(a)=n1=h1i(ni) h_{1i}(a_i) = h_1(a) = n_1 = h_{1i}(n_i)

and, as the kernel of h1ih_{1i} is ni2n_i^2, ni⊂ai+ni2n_i \subset a_i + n_i^2 and ai⊂nia_i \subset n_i, whence ni=ai+ni2n_i = a_i + n_i^2. Moreover u0i(Mi′)=u0(M′)=M,=u0i(Mi)u_{0i}(M'_i) = u_0(M') = M, = u_{0i}(M_i) and, as Ker(u0i)=niMi\mathrm{Ker}(u_{0i}) = n_i M_i, Mi=Mi′+niMiM_i = M'_i + n_i M_i; whence

niiMi=(ai+ni2)i(Mi′+niMi). n_i^i M_i = (a_i + n_i^2)^i (M'_i + n_i M_i).

Now, aiknii+1−k⊂nii+1=0a_i^k n_i^{i+1-k} \subset n_i^{i+1} = 0 for 0≤k≤i0 \leq k \leq i; then it certainly follows that ui(Ni)=niMi=aiiMi′u_i(N_i) = n_i M_i = a_i^i M'_i, which proves (17).

Moreover m1=h1(n1)m_1 = h_1(n_1), whence a⊂m1a \subset m_1 and therefore ai⊂m1i⊂mia^i \subset m_1^i \subset m_i, whence

Ni⊂miM′+Ni+1N_i \subset m_i M' + N_{i+1}; on the other hand obviously miM⊂Nim_i M \subset N_i and hence Ni=miM′+Ni+1N_i = m_i M' + N_{i+1} for all i≥0i \geq 0; then it follows from Corollary 3 to Proposition 12 of no. 9 that M′=MM' = M, which completes the proof.

Corollary 1

With the notation and hypotheses of Proposition 14 suppose further that M0M_0 is a finitely generated AA-module and that the ideal n1n_1 of AA, is finitely generated. Let m1m_1 be the kernel of h1h_1; the topologies on AA and MM are then the m1m_1-adic topologies on this ring and this module respectively; to be precise, for all i≥0i \geq 0, the kernels of hih_i and uiu_i are m1i+1m_1^{i+1} and m1i+1Mm_1^{i+1} M respectively; further m1/m12m_1 / m_1^2 is a finitely generated AA-module.

We preserve the notation of the proof of Proposition 14; the hypotheses here allow us to assume that the ideal aa is finitely generated. Let i≥0i \geq 0 be any integer; for all j≥0j \geq 0, by (17), Ni+j=aj(aiM)+Ni+j+1⊂mj(aiM)+Ni+j+1N_{i+j} = a^j(a^i M) + N_{i+j+1} \subset m_j(a^i M) + N_{i+j+1}; conversely, mj(aiM)⊂mjmiM⊂mi+jM⊂Ni+jm_j(a^i M) \subset m_j m_i M \subset m_{i+j} M \subset N_{i+j}, whence

Ni+j=mj(aiM)+Ni+j+1. N_{i+j} = m_j(a^i M) + N_{i+j+1}.

As aa and MM are finitely generated AA-modules, so is aiMa^i M. Applying Corollary 3 to Proposition 12 of no. 9 to the module NiN_i with the filtration (Nij)j∈Z(N_{ij})_{j \in \mathbf{Z}} defined by Nij=NiN_{ij} = N_i, if j<0j < 0, Nij=Ni+jN_{ij} = N_{i+j} if j≥0j \geq 0, we obtain Ni=aiMN_i = a^i M, whence Ni⊂m1iMN_i \subset m_1^i M. But also m1iM⊂miM⊂Nim_1^i M \subset m_i M \subset N_i, whence Ni=m1iMN_i = m_1^i M. Applying this to the case where M=AM = A, uij=hiju_{ij} = h_{ij}, we obtain mi=m1im_i = m_1^i. Moreover, m1=a+m12m_1 = a + m_1^2 by (17), which proves the last assertion of the corollary.

Corollary 2

Under the hypotheses of Corollary 1, for AA to be Noetherian, it is necessary and sufficient that AA, be so.

The condition is necessary since AA, is isomorphic to a quotient of AA; it is sufficient by virtue of no. 10, Corollary 5 to Theorem 2.

12. The Hausdorff completion of a filtered module

Let GG be a filtered group whose filtration (Gn)(G_n) consists of normal subgroups of GG; we have already recalled (no. 6) that the Hausdorff completion G^\hat{G} of the topological group GG is canonically identified with the inverse limit lim⁡←G/Gn\lim \leftarrow G/G_n of the discrete groups G/GnG/G_n, the canonical homomorphism i:G→G^i : G \to \hat{G} having image the Hausdorff group associated with GG (everywhere dense in G^\hat{G}) and kernel the closure ⋂Gn\bigcap G_n of {0}\{0\} in GG. The Hausdorff completion G^n\hat{G}_n of the subgroup GnG_n of GG is identified with the closure of i(Gn)i(G_n) in G^\hat{G} (General Topology, Chapter II, § 3, no. 9, Corollary 1 to Proposition 18) and, since GG, is closed in G^\hat{G},

G,=i−1(G^n)=i−1(G^n∩i(G)). G, = i^{-1}(\hat{G}_n) = i^{-1}(\hat{G}_n \cap i(G)).

Moreover, the G^n\hat{G}_n form a fundamental system of neighbourhoods of 0 in G^\hat{G} (General Topology, Chapter 111, § 3, no. 4, Proposition 7) and are therefore normal open subgroups of G^\hat{G} (General Topology, Chapter 111, § 2, no. 3, Proposition 8); the topology on G^\hat{G} is defined by the filtration (G^n)(\hat{G}_n), which is always separated by definition. As i(G)i(G) is dense in G^\hat{G} and G^n\hat{G}_n is open,

G^=i(G)⋅G^n \hat{G} = i(G) \cdot \hat{G}_n

and similarly

=i(Gn−1)⋅G^n. = i(G_{n-1}) \cdot \hat{G}_n.

We deduce from (18) and (19) that the filtration (G^n)(\hat{G}_n) is exhaustive if and only if (Gn)(G_n) is.

The second isomorphism theorem (Algebra, Chapter I, § 6, no. 13, Theorem 6 (d)) and equations (18), (19) and (20) show that the canonical homomorphisms

Gn−1/Gn→G^n−1/G^n,G/Gn→G^/G^n, G_{n-1}/G_n \to \hat{G}_{n-1}/\hat{G}_n, \quad G/G_n \to \hat{G}/\hat{G}_n,

are bijective and hence so is the canonical homomorphism

gr⁡(G)→gr⁡(G^). \operatorname{gr}(G) \to \operatorname{gr}(\hat{G}).

Now let AA be a filtered ring, EE a filtered AA-module and (An)(A_n) and (En)(E_n) the respective filtrations of AA and EE; we shall assume that these filtrations are exhaustive so that for the corresponding topologies AA is a topological ring and EE a topological AA-module (no. 5, Proposition 3). Then we have defined (General Topology, Chapter 111, §6, nos. 5 and 6) A^\hat{A} as a topological ring and E^\hat{E} as a topological A^\hat{A}-module. If i:A→A^i : A \to \hat{A} is the canonical homomorphism, then i(Am)i(An)⊂i(Am+n)i(A_m)i(A_n) \subset i(A_{m+n}), whence by the continuity of multiplication in AA,

A^mA^n⊂A^m+n \hat{A}_m \hat{A}_n \subset \hat{A}_{m+n}

since AnA_n is the closure of i(An)i(A_n) in A^\hat{A}. It can be similarly shown that

A^mE^n⊂E^m+n; \hat{A}_m \hat{E}_n \subset \hat{E}_{m+n};

in other words:

Proposition 15

Let AA be a filtered ring and EE a filtered AA-module, the respective filtrations (An),(En)(A_n), (E_n) of AA and EE being exhaustive. Then (A^n)(\hat{A}_n) is a filtration compatible with the ring structure on A^\hat{A} and (E^n)(\hat{E}_n) a filtration compatible with the module structure on E^\hat{E} over the filtered ring A^\hat{A}; moreover these filtrations are exhaustive and define respectively the topologies on A^\hat{A} and E^\hat{E}. Finally, the canonical mappings gr⁡(A)→gr⁡(A^)\operatorname{gr}(A) \to \operatorname{gr}(\hat{A}) and gr⁡(E)→gr⁡(E^)\operatorname{gr}(E) \to \operatorname{gr}(\hat{E}) of graded Z\mathbf{Z}-modules are respectively a graded ring isomorphism and a graded gr⁡(A)\operatorname{gr}(A)-module isomorphism.

In what follows, for every uniform space XX, jXj_X will denote the canonical mapping from XX to its Hausdorff completion X^\hat{X} and X0=jX(X)X_0 = j_X(X) the uniform subspace of X^\hat{X}, which is the Hausdorff space associated with XX. Recall that the topology on XX is the inverse image under jXj_X of that on X0X_0 (General Topology,

Chapter II, § 3, no. 7, Proposition 12). Recall also that, for every uniformly continuous mapping f:X→Yf : X \to Y, f^\hat{f} denotes the uniformly continuous mapping from X^\hat{X} to Y^\hat{Y} such that f^∘jX=jY∘f\hat{f} \circ j_X = j_Y \circ f (loc. cit., Proposition 15); if XX is a uniform subspace of YY and E\mathscr{E} the canonical injection, X^\hat{X} is identified with a uniform subspace of Y^\hat{Y} and f^\hat{f} is the canonical injection of XX into Y^\hat{Y} (loc. cit., no. 9. Corollary 1 to Proposition 18).

Lemma 2

Let X→fY→gZX \xrightarrow{f} Y \xrightarrow{g} Z be an exact sequence of strict morphisms of topological groups (Algebra, Chapter II, § 1, no. 4, Remark). Suppose that X,Y,ZX, Y, Z admit Hausdorff completion groups and that the identity elements EX,Y,Z\mathscr{E} X, Y, Z admit countable fundamental systems of neighbourhoods. Then X^→f^Y^→g^Z^\hat{X} \xrightarrow{\hat{f}} \hat{Y} \xrightarrow{\hat{g}} \hat{Z} is an exact sequence of strict morphisms.

Let N1,NgN_1, N_g be the respective kernels off and gg; let us write

f=f3∘f2∘f1 f = f_3 \circ f_2 \circ f_1

where f1f_1 is the canonical mapping X→X/NfX \to X/N_f, f2f_2 is an isomorphism of X/NfX/N_f onto NN, and f3f_3 is the canonical injection N→YN \to Y. We already know that f2f_2 is an isomorphism of (X/Nf)∗(X/N_f)^* onto N^g\hat{N}_g and we have just recalled that f1f_1 is an injective strict morphism of NN, to Y^\hat{Y}; if we show that f^1\hat{f}_1 is a surjective strict morphism, it will follow that f^\hat{f} is a strict morphism (General Topology, Chapter III, § 2, no. 8, Remark 2). Let g1g_1 be the canonical mapping Y→Y/NgY \to Y/N_g; if we show that g^1\hat{g}_1 is a surjective strict morphism of the kernel N^g\hat{N}_g, we shall see as above that g^\hat{g} is a strict morphism and the sequence X^→f^Y^→g^Z^\hat{X} \xrightarrow{\hat{f}} \hat{Y} \xrightarrow{\hat{g}} \hat{Z} will be exact. Thus we have reduced the problem to proving that, if Y=X/NY = X/N (where NN is a normal subgroup of XX) and f:X→Yf : X \to Y is the canonical mapping, f^\hat{f} is a surjective strict morphism with kernel N^\hat{N}.

Let f0:X0→Y0f_0 : X_0 \to Y_0 be the mapping which coincides with ff on X0X_0; as jXj_X (resp. jYj_Y) is a surjective strict morphism of XX onto XX, (resp. YY onto Y0Y_0), f0f_0 is a surjective strict morphism (General Topology, Chapter III, § 2, no. 8, Remark 3). Now X0X_0 and Y0Y_0 are metrizable (General Topology, Chapter IX, § 3, no. 1, Proposition 1); then it follows from General Topology, Chapter IX, § 3, no. 1, Corollary 1 to Proposition 4 and Lemma 1, that f^0=f^\hat{f}_0 = \hat{f} is a surjective strict morphism and has kernel the closure N^0′\hat{N}_0' in XX of the kernel N0′N_0' of f0f_0. Then it will be sufficient for us to prove that N^0′=N^\hat{N}_0' = \hat{N}. Now N0′N_0' obviously contains N0=jX(N)N_0 = j_X(N); it will be sufficient to show that N0′N_0' is contained in the closure N‾0\overline{N}_0 of N0N_0 in XX. Now,

u=jX−1(X0−N‾0)=X−jX−1(N‾0) u = j_X^{-1}(X_0 - \overline{N}_0) = X - j_X^{-1}(\overline{N}_0)

is an open set in XX which does not meet NN; as ff is a surjective strict morphism, V=f(U)V = f(U) is an open set in YY not containing the identity element e′e' of YY and hence not meeting the closure of e′e'; then jY(V)j_Y(V) does not contain the identity element of Y0Y_0. But jY(V)=f0(X0−N‾0)j_Y(V) = f_0(X_0 - \overline{N}_0) and hence N0′⊂N‾0N'_0 \subset \overline{N}_0, which complete the proof of Lemma 2.

Proposition 16

Let AA be a filtered ring, (A,A,) its filtration, EE an AA-module and (EnE_n) the filtration on EE derived from that on AA consisting of the En=AnEE_n = A_n E. Suppose that the filtration (A,A,) is exhaustive and the module EE is finitely generated. If i:E→E^i : E \to \hat{E} is the canonical mapping, then, for all n∈Zn \in \mathbf{Z},*

8n=A^nE^=A^ni(E)andE^=A^.i(E). 8_n = \hat{A}_n \hat{E} = \hat{A}_n i(E) \quad \text{and} \quad \hat{E} = \hat{A}.i(E).
*In particular E^\hat{E} is a finitely generated A^\hat{A}-module.

The equation AnE=EnA_n E = E_n implies, by virtue of the continuity of the external law on the AA-module 8,A^nE^⊂88, \hat{A}_n \hat{E} \subset 8, and obviously A^nE^⊃A^ni(E)\hat{A}_n \hat{E} \supset \hat{A}_n i(E). By hypothesis there exists a surjective homomorphism u:L→Eu : L \to E, where L=AsIL = A_s^I, II being a finite set; let LL be given the product filtration, consisting of the Ln=AnIL_n = A_n^I, which define on LL the product topology; then L^=A^sI\hat{L} = \hat{A}_s^I and L^n=A^nI\hat{L}_n = \hat{A}_n^I (General Topology, Chapter II, § 3, no. 9, Corollary 2 to Proposition 18). Let j:L→L^j : L \to \hat{L} be the canonical mapping and (ei)i∈I(e_i)_{i \in I} the canonical basis of LL; for an element ∑i∈Iaij(ei)\sum_{i \in I} a_{ij}(e_i) (where ai∈A^a_i \in \hat{A} for all i∈Ii \in I) to belong to L^n\hat{L}_n, it is necessary and sufficient that ai∈A^na_i \in \hat{A}_n for all ii; then L^n=A^n.j(L)\hat{L}_n = \hat{A}_n . j(L). This being so, by definition u(L^n)=AnE=Enu(\hat{L}_n) = A_n E = E_n and hence uu is a strict morphism of LL onto EE (General Topology, Chapter III, § 2, no. 8, Proposition 24). Lemma 2 then shows that u^:L^→E^\hat{u} : \hat{L} \to \hat{E} is a surjective strict morphism. As L^n\hat{L}_n is an open subgroup off, u^(L^n)\hat{u}(\hat{L}_n) is an open (and therefore closed) subgroup of E^\hat{E}; but u^(L^n)=A^nu^(j(L))=A^ni(E)\hat{u}(\hat{L}_n) = \hat{A}_n \hat{u}(j(L)) = \hat{A}_n i(E) and, as i(En)⊂Ani(E)⊂A^ni(E)i(E_n) \subset A_n i(E) \subset \hat{A}_n i(E), finally E^n⊂A^ni(E)⊂A^nE^⊂E^n\hat{E}_n \subset \hat{A}_n i(E) \subset \hat{A}_n \hat{E} \subset \hat{E}_n and therefore E^n=A^nE^=A^ni(E)\hat{E}_n = \hat{A}_n \hat{E} = \hat{A}_n i(E); setting n=0n = 0, we obtain the second formula of (25).

Corollary 1

Under the conditions of Proposition 16, if AA is complete, so is EE.

As the canonical mapping A→A^A \to \hat{A} is then surjective (no. 6, Proposition 5), E^=i(E)\hat{E} = i(E) by (25) and the conclusion follows by Proposition 5 of no. 6.

Corollary 2

Let AA be a commutative ring, mm a finitely generated ideal of AA and A^\hat{A} the Hausdorff completion of AA with respect to the mm-adic topology. Then m^n=(m^)n=mn.A^\hat{m}^n = (\hat{m})^n = m^n . \hat{A} for every integer n>0n > 0 and the topology on A^\hat{A} is the m^\hat{m}-adic topology.

Let us write An=mnA_n = m^n, which is a finitely generated ideal of AA. The formula mpAn=mn+pm^p A_n = m^{n+p} shows that the topology induced on AA, by the mm-adic topology coincides with the mm-adic topology on the AA-module AnA_n (no. 1, Example 3).

By Proposition 16 applied to E=AnE = A_n, A^n=A^.An\hat{A}_n = \hat{A}.A_n in other words m^n=mn.A^\hat{m}^n = m^n . \hat{A}. In particular m=m.A^m = m . \hat{A}, whence

(m^)n=mn.A^=A^.An (\hat{m})^n = m^n . \hat{A} = \hat{A}.A_n
(cf. Exercise 12).

Examples of Hausdorff completions of filtered rings

(1) Let AA be a graded ring of type N\mathbf{N} and let (An)n≥0(A_n)_{n \geq 0} be its graduation; let it be given the associated filtration which is separated and exhaustive (no. 1, Example 1). The additive group AA is canonically identified with a subgroup of B=∏n∈NAnB = \prod_{n \in \mathbf{N}} A_n; if BB is given the topology the product of the discrete topologies, the topology induced on AA is the topology defined by the filtration on AA; also BB is a complete topological group and AA is dense in BB (General Topology, Chapter III, §2, no. 9, Proposition 25). The additive topological group BB is then identified with the completion A^\hat{A} of the Hausdorff additive group AA and it follows from Proposition 15 that it has a unique ring structure which makes it the completion of the topological ring AA. To define multiplication in this ring, note that, if we write An′=∑i>nAiA'_n = \sum_{i > n} A_i, the closure in BB of the two-sided ideal An′A'_n is the set BB, of x=(xi)∈Bx = (x_i) \in B such that xi=0x_i = 0 for i≤ni \leq n. Then let x=(xi),y=(yi)x = (x_i), y = (y_i) be two elements of BB and z=(zi)z = (z_i) their product. Then, for all n>0n > 0, x≡xn′x \equiv x'_n (mod. BnB_n), y≡yn′y \equiv y'_n (mod. BnB_n), where xn′=(xi)0≤i≤nx'_n = (x_i)_{0 \leq i \leq n}, yn′=(yi)0≤i≤ny'_n = (y_i)_{0 \leq i \leq n}, whence z≡xn′yn′z \equiv x'_n y'_n (mod. BnB_n). But xn′x'_n and yn′y'_n belong to AA and it is therefore seen that, for all n∈Nn \in \mathbf{N},

zn=∑j=0nxjyn−j. z_n = \sum_{j=0}^n x_j y_{n-j}.

In particular, we again obtain the Corollary to Proposition 6 of no. 6 : if CC is a commutative ring, the completion of the polynomial ring C[X1,…,Xr]C[X_1, \ldots, X_r], with the filtration associated with its usual graduation (by total degree) is canonically identified with the ring of formal power series C[[X1,…,Xr]]C[[X_1, \ldots, X_r]] (cf. Algebra, Chapter IV, §5, no. 10).

(2) Let KK be a complete commutative field with a valuation. The completion of the ring of convergent series in rr variables over KK is canonically identified with the ring of formal power series K[[X1,…,Xr]]K[[X_1, \ldots, X_r]]. \

(3) Let aa be a non-zero non-invertible element of a principal ideal domain; the (a)-adic topology on AA is also called the aa-adic topology; it is Hausdorff, for the intersection of the ideals (an)(a^n) reduces to 0 (Algebra, Chapter VII, §1, no. 3). Note that the completion of AA with respect to this topology is not necessarily an integral domain (cf. no. 13, Remark 3). The associated graded ring gr(A)=gr(A^)\mathrm{gr}(A) = \mathrm{gr}(\hat{A}) is canonically isomorphic to (A/a)[X](A/a)[X] (no. 3, Example 1). If A=ZA = \mathbf{Z}, the completion of Z\mathbf{Z} with respect to the n-adic topology (n>1n > 1) is denoted by Zn\mathbf{Z}_n and its elements are called nn-adic integers.

Every element of Z/nkZ\mathbf{Z}/n^k \mathbf{Z} admits a unique representative of the form ∑i=0k−1aini\sum_{i=0}^{k-1} a_i n^i where 0≤ai≤n−10 \leq a_i \leq n - 1 for all ii; moreover, its canonical image in Z/nk−1Z\mathbf{Z}/n^{k-1} \mathbf{Z} is the class of ∑i=0k−2aini\sum_{i=0}^{k-2} a_i n^i. These remarks and the fact that ZZ, is canonically identified with the inverse limit lim⁡←Z/nkZ\lim_{\leftarrow} \mathbf{Z}/n^k \mathbf{Z} show immediately that every element of ZZ, can be written uniquely in the form ∑i=0∞aini\sum_{i=0}^{\infty} a_i n^i where 0≤ai<n0 \leq a_i < n and conversely that such a series is convergent in Z\mathbf{Z}.

13. The Hausdorff completion of a semi-local ring

Proposition 17

Let AA be a commutative ring and (mλ)λ∈L(m_{\lambda})_{\lambda \in L} a family of ideals of AA, distinct from AA, such that mλm_{\lambda} and mμm_{\mu} are relatively prime for λ≠μ\lambda \neq \mu. For every family s=(s(λ))λ∈Ls = (s(\lambda))_{\lambda \in L} of integers ≥0\geq 0, of finite support, set as=⋂λ∈Lmλs(λ)a_s = \bigcap_{\lambda \in L} m_{\lambda}^{s(\lambda)} (equal to the product of the mλs(λ)m_{\lambda}^{s(\lambda)} for the λ\lambda such that s(λ)≠0s(\lambda) \neq 0; cf. Chapter 11, § 1, no. 2, Propositions 3 and 5); the asa_s form a fundamental system of neighbourhoods of 0 with respect to a topology T\mathscr{T} compatible with the ring structure on AA; let A^\hat{A} be the Hausdorff completion of AA with respect to this topology. On the other hand, for all λ∈L\lambda \in L, let AλA_{\lambda} be the ring AA with the mλm_{\lambda}-adic topology and let A^λ\hat{A}_{\lambda} be its Hausdorff completion. If u:A→∏λ∈LAλu : A \to \prod_{\lambda \in L} A_{\lambda} denotes the diagonal homomorphism, uu is continuous and the corresponding homomorphism u^:\hat{u} :

A^→(∏λ∈LAλ)^=∏λ∈LA^λ \hat{A} \to \left( \prod_{\lambda \in L} A_{\lambda} \right)^{\hat{}} = \prod_{\lambda \in L} \hat{A}_{\lambda}

(General Topology, Chapter III, § 6, no. 5 and Chapter II, § 3, no. 9, Corollary 2 to Proposition 18) is a topological ring isomorphism.

The first assertion follows from General Topology, Chapter III, § 6, no. 3, Example 3. Let us set B=∏λ∈LAλB = \prod_{\lambda \in L} A_{\lambda}; as the topology on AA is finer than each of the mλm_{\lambda}-adic topologies, the mappings prλ∘u\mathrm{pr}_{\lambda} \circ u are continuous and hence uu is continuous. Also, u(as)u(a_s) is the intersection of the diagonal AA of BB and the open set ⋂λ∈Lprλ−1(mλs(λ))\bigcap_{\lambda \in L} \mathrm{pr}_{\lambda}^{-1}(m_{\lambda}^{s(\lambda)}) of BB; it follows that uu is a strict morphism from the additive group AA to BB with image AA. Now Δ\Delta is dense in BB. For let b=(aλ)λ∈Lb = (a_{\lambda})_{\lambda \in L} be an element of BB; every neighbourhood of bb in BB contains a set of the form b+Vb + V, where V=⋂λ∈Lprλ−1(mλs(λ))V = \bigcap_{\lambda \in L} \mathrm{pr}_{\lambda}^{-1}(m_{\lambda}^{s(\lambda)}) for a family s=(s(λ))λ∈Ls = (s(\lambda))_{\lambda \in L} with finite support of integers ≥0\geq 0. As the mλs(λ)m_{\lambda}^{s(\lambda)} are relatively prime in pairs (Chapter II, § 1, no. 2, Proposition 3), there exists x∈Ax \in A such that x≡aλ(modmλs(λ))x \equiv a_{\lambda} \pmod{m_{\lambda}^{s(\lambda)}} for all λ\lambda (loc. cit., Proposition 5) and hence (b+V)∩A≠∅(b + V) \cap A \neq \varnothing. The Hausdorff completion of the group B/ΔB/\Delta is then {0}\{0\}; applying Lemma 2 of no. 12 to the exact sequences 0→A→uB,A→uB→B/Δ0 \to A \xrightarrow{u} B, A \xrightarrow{u} B \to B/\Delta, we see that u^\hat{u} is an isomorphism of A^\hat{A} onto B^\hat{B}.

Corollary

Let AA be a principal ideal domain and PP a representative system of extremal elements of AA (Algebra, Chapter VII, § 1, no. 3). The topology on AA with respect to which the ideals ≠0\neq 0 of A\mathbf{A} form a fundamental system of neighbourhoods of 0, which is compatible with the ring structure on A\mathbf{A}, is Hausdorff and the completion of A\mathbf{A} with this topology is canonically isomorphic to the product of the completions of A\mathbf{A} with respect to the xx-adic topologies, where xx runs through PP.

The principal ideals (x)(x) where π∈P\pi \in P are maximal and distinct and hence relatively prime, we have already seen (no. 12, Example 3) that the xx-adic topologies are Hausdorff and hence so is the topology defined in the statement of Proposition 17, which is finer than each of the xx-adic topologies.

If the Corollary to Proposition 17 is applied when A=Z\mathbf{A} = \mathbf{Z}, we denote by Z^\hat{\mathbf{Z}} the completion of Z\mathbf{Z} with respect to the topology for which all the ideals ≠0\neq 0 of Z\mathbf{Z} form a fundamental system of neighbourhoods of 0, the ring isomorphic to the product ∏p∈PZp\prod_{p \in P} \mathbf{Z}_p of the rings of pp-adic integers (P being the set of prime numbers).

Remarks

(1) Clearly, under the conditions of Proposition 17, the topology T\mathscr{T} is the least upper bound of the mλm_\lambda-adic topologies on A\mathbf{A}.

(2) Every closed ideal aa of ∏λ∈LA^λ\prod_{\lambda \in L} \hat{A}_\lambda is identical with the product of its projections aλ=prλ(a)a_\lambda = \mathrm{pr}_\lambda(a), which are closed ideals in the AλA_\lambda; for A^λ\hat{A}_\lambda is canonically identified with a closed ideal Aλ′A'_\lambda of ∏λA^λ\prod_{\lambda} \hat{A}_\lambda and aλa_\lambda with a∩Aλ′a \cap A'_\lambda (Algebra, Chapter I, §8, no. 10, Proposition 6), the sum of the aλa_\lambda is dense in the product ∏λaλ\prod_\lambda a_\lambda (General Topology, Chapter III, §2, no. 9, Proposition 25) and the latter is closed in ∏λA^λ\prod_\lambda \hat{A}_\lambda, whence our assertion.

Proposition 18

Let A\mathbf{A} be a commutative ring, (mi)1≤i≤q(m_i)_{1 \leq i \leq q} a finite family of distinct maximal ideals of A\mathbf{A}, t the product ideal m1m2…mq=m1∩m2∩…∩mqm_1 m_2 \ldots m_q = m_1 \cap m_2 \cap \ldots \cap m_q and SS the multiplicative subset ⋂i=1q(A−mi)\bigcap_{i=1}^q (\mathbf{A} - m_i). Let A\mathbf{A} be given the rr-adic topology, the ring B=S−1AB = S^{-1} \mathbf{A} the rBrB-adic topology and each of the local rings AnA_n the (miAmi)(m_i A_{m_i})-adic topology. Let u:A→Bu : \mathbf{A} \to B, vi:B→Amiv_i : B \to A_{m_i} be the canonical homomorphism (Chapter II, §2, no.1, Corollary 2 to Proposition 2) and vv the homomorphism (vi):B→∏i=1qAn(v_i) : B \to \prod_{i=1}^q A_n. The homomorphisms uu and vv are continuous and the corresponding homomorphism u^:A^→\hat{u} : \hat{\mathbf{A}} \to and v^:B^→∏n(An)∗\hat{v} : \hat{B} \to \prod_{n} (A_n)^* are topological ring isomorphisms.

mi∩S=∅m_i \cap S = \varnothing for 1≤i≤q1 \leq i \leq q, hence the ideal mi′=miBm'_i = m_i B of BB is maximal (Chapter II, §2, no. 5, Proposition 11) and

rB=m1′∩m2′∩⋯∩mq′ rB = m'_1 \cap m'_2 \cap \cdots \cap m'_q

(Chapter II, § 2, no. 4); finally, Bmi′=AiB_{m'_i} = A_i, up to a canonical isomorphism (Chapter 11, § 2, no. 5, Proposition 11). As uˉ1(rB)=r\bar{u}^1(rB) = r and vˉi1(miAmi)⊃rB\bar{v}_i^1(m_iA_{m_i}) \supset rB, uu and vv are continuous. Then it is sufficient to prove that,

w=v∘u:A→∏i=14Ami, w = v \circ u : A \to \prod_{i=1}^4 A_{m_i},

w^\hat{w} is an isomorphism of A^\hat{A} onto ∏i=1qA^mi\prod_{i=1}^q \hat{A}_{m_i}, for this result applied to BB and the mi′m'_i will show that v^\hat{v} is an isomorphism and therefore also u^\hat{u}. Note that every product of powers of the mim_i contains a power of cc and hence the rr-adic topology is the least upper bound of the mim_i-adic topologies; moreover, if AiA_i denotes the ring AA with the mim_i-adic topology and ϕ:A→∏i=1qA\phi : A \to \prod_{i=1}^q A, the diagonal mapping, ϕ^:A^→∏i=14A^i\hat{\phi} : \hat{A} \to \prod_{i=1}^4 \hat{A}_i is an isomorphism (Proposition 17). Then it all amounts to proving that, if ui:Ai→Amiu_i : A_i \to A_{m_i} is the canonical mapping, u^i:A^i→A^mi\hat{u}_i : \hat{A}_i \to \hat{A}_{m_i} is an isomorphism. Now, for all nn, the mapping

ui,n:A/min→Ami/minAmi u_{i,n} : A/m_i^n \to A_{m_i}/m_i^n A_{m_i}

derived from uiu_i by taking quotients is an isomorphism (Chapter II, §3, no. 3, Proposition 9); our assertion follows from the fact that A^i\hat{A}_i (resp. A^mi\hat{A}_{m_i}) is the inverse limit of the discrete rings A/minA/m_i^n (resp. Ami/minAmiA_{m_i}/m_i^n A_{m_i}) (cf. no. 6).

Remark (3). We see that an integral domain AA can be such that its Hausdorff completion A^\hat{A} admits non-zero divisors of zero.

Proposition 19

Let AA be a commutative ring and mm a maximal ideal of AA. The Hausdorff completion A^\hat{A} of AA with respect to the mm-adic topology is a local ring whose maximal ideal is mm.

If a=⋂k≥1mka = \bigcap_{k \geq 1} m^k, A^\hat{A} is the completion of the Hausdorff ring A/aA/a associated with AA and, as m/am/a is maximal in A/aA/a, we may assume that AA is Hausdorff with respect to the mm-adic topology. As A/mA/m and A^/m^\hat{A}/\hat{m} are isomorphic rings (no. 12, formula (21)), mm is maximal in A^\hat{A}. As the topology on A^\hat{A} is defined by the filtration (mn)∗(m^n)^* (no. 12), the proposition will be a consequence of the following lemma:

Lemma 3

Let AA be a complete Hausdorf topological ring, in which there exists a fundamental system G\mathscr{G} of neighbourhoods of 0 consisting of additive subgroups of AA. (i) For all x∈Ax \in A such that lim⁡n→mxn=0\lim_{n \to m} x^n = 0, 1−x1 - x is invertible in AA and its inverse is equal to ∑n=0∞xn\sum_{n=0}^\infty x^n. (ii) Let aa be a two-sided ideal of AA such that lim⁡n→∞xn=0\lim_{n \to \infty} x^n = 0 for all x∈ax \in a. For an element y of A\mathbf{A} to be invertible, it is necessary and sufficient that its class mod. a be invertible in A/a\mathbf{A}/a; in particular a is contained in the Jacobson radical of A\mathbf{A}.

(i) As

(1−x)(1+x+⋯+xn)=(1+x+⋯+xn)(1−x)=1−xn+1, (1 - x)(1 + x + \cdots + x^n) = (1 + x + \cdots + x^n)(1 - x) = 1 - x^{n+1},

it all amounts to proving that the series with general term xnx^n is convergent in A\mathbf{A}; now, by hypothesis, for every neighbourhood V∈SV \in \mathfrak{S} of 0 in A\mathbf{A}, there exists an integer p>0p > 0 such that xn∈Vx^n \in V for all n≥pn \geq p. We conclude that

xp+xp+1+…+xq∈V x^p + x^{p+1} + \ldots + x^q \in V

for all q≥pq \geq p and our assertion then follows from Cauchy's criterion (General Topology, Chapter III, § 5, no. 2, Theorem 1).

(ii) Suppose that there exists y′∈Ay' \in \mathbf{A} such that yy′≡1yy' \equiv 1 (mod. a) and y′y≡1y'y \equiv 1 (mod. a). The hypothesis on a implies, by (i), that yy′yy' and y′yy'y are invertible in A\mathbf{A} and hence yy is invertible in A\mathbf{A}. In particular, every x∈ax \in a is such that 1−x1 - x is invertible in A\mathbf{A} and, as a is a two-sided ideal of A\mathbf{A}, it is contained in the Jacobson radical of A\mathbf{A} (Algebra, Chapter VIII, § 6, no. 3, Theorem 1).

Having established this lemma, it is sufficient to apply it to the topological ring A^\hat{\mathbf{A}} and the ideal m^\hat{\mathfrak{m}}, as, for all x∈m^x \in \hat{\mathfrak{m}}, xn∈(m^)n⊂(mn)∧x^n \in (\hat{\mathfrak{m}})^n \subset (\mathfrak{m}^n)^{\wedge} and the sequence (xn)(x^n) therefore tends to 0.

If we take A=Z\mathbf{A} = \mathbf{Z}, every maximal ideal of Z\mathbf{Z} is of the form pZp\mathbf{Z} where pp is prime. The ring of pp-adic numbers Zp\mathbf{Z}_p is then a local ring of which pZpp\mathbf{Z}_p is the maximal ideal (Corollary 2 to Proposition 16) and whose residue field is isomorphic to Z/pZ=Fp\mathbf{Z}/p\mathbf{Z} = \mathbf{F}_p, and Z(p)\mathbf{Z}_{(p)} with the pZ(p)p\mathbf{Z}_{(p)}-adic topology is identified with a topological subring of Z\mathbf{Z}, containing Z\mathbf{Z}.

Corollary

Let A\mathbf{A} be a semi-local ring (Chapter II, § 3, no. 5), mim_i its distinct maximal ideals (1≤i≤q)(1 \leq i \leq q) and

r=m1∩m2∩⋯∩mq r = m_1 \cap m_2 \cap \cdots \cap m_q

its Jacobson radical. The Hausdorff completion A^\hat{\mathbf{A}} of A\mathbf{A} with respect to the r-adic topology is a semi-local ring, canonically isomorphic to the product ∏i=1qA^mi\prod_{i=1}^q \hat{\mathbf{A}}_{m_i}, where A^mi\hat{\mathbf{A}}_{m_i} is the Hausdorff completion ring of the local ring Ami\mathbf{A}_{m_i} with respect to the (miAmi)(m_i \mathbf{A}_{m_i})-adic topology.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.