Bourbaki

§ 2. Cohen rings

Throughout this paragraph, pp denotes a prime number.

1. pp-rings

Definition 1

A ring CC is said to be a pp-ring if the ideal pCpC of CC is maximal, and if CC is separated and complete for the pCpC-adic topology.

Let CC be a ring; if p1Cp1_C is nilpotent and if the ideal pCpC of CC is maximal, CC is a pp-ring, for the pCpC-adic topology of CC is discrete. In particular, every field of characteristic pp is a pp-ring.

Proposition 1

Let CC be a pp-ring.

a) The ring CC is local, with maximal ideal pCpC.

b) Suppose p1Cp1_C nilpotent. Let dd be the smallest positive integer such that pd1C=0p^d1_C = 0. The ideals of CC are of the form pkCp^kC with 0≤k≤d0 \leq k \leq d, and one has pkC≠plCp^kC \neq p^lC when kk and ll are two distinct integers satisfying 0≤k≤d,0≤l≤d0 \leq k \leq d, 0 \leq l \leq d. The CC-module CC is of length dd.

c) Suppose that p1Cp1_C is not nilpotent. Then CC is a discrete valuation ring whose residue field is of characteristic pp, and whose field of fractions is of characteristic 00. The ideals of the form pnCp^nC, with n∈Nn \in \mathbf{N}, are pairwise distinct; they form all the non-zero ideals of CC. The CC-module CC is not of finite length.

Assertion a) follows from prop. 19 of III, §2§ 2, no. 13.

By hypothesis, one has ⋂n⩾0pnC={0}\bigcap_{n \geqslant 0} p^n C = \{0\}. Let x≠0x \neq 0 in CC; there exists an integer n⩾0n \geqslant 0 such that x∈pnCx \in p^n C, x∉pn+1Cx \notin p^{n+1} C; therefore there exists an element yy of CC such that x=pnyx = p^n y; since yy does not belong to pCpC, yy is invertible.

Suppose that p1Cp 1_C is not nilpotent. If xx and x′x' are two non-zero elements of CC, there exist two integers n⩾0n \geqslant 0, n′⩾0n' \geqslant 0 and two invertible elements yy, y′y' of CC such that x=pnyx = p^n y, x′=pn′y′x' = p^{n'} y'. We then have xx′=pn+n′yy′≠0xx' = p^{n+n'} yy' \neq 0, hence CC is an integral domain. Since CC is a local ring, but is not a field, and since the maximal ideal mC=pCm_C = pC of CC is principal, CC is a discrete valuation ring (VI, §3§ 3, no 6, prop. 9). The non-zero ideals of CC are then of the form pnCp^n C by loc. cit., prop. 8, and are pairwise distinct. In particular, the ring CC is not artinian, hence the CC-module CC is not of finite length. The residue field C/pCC/pC of CC is of characteristic pp. Let qq be the characteristic of the field of fractions of CC. We have p1C≠0p 1_C \neq 0, whence p≠qp \neq q. Moreover, if qq were non-zero, we should have q1C=0q 1_C = 0 hence C/pCC/pC would be of characteristic q≠pq \neq p, which is absurd. This proves c).

Suppose that p1Cp 1_C is nilpotent. Let dd be the smallest positive integer such that pd1C=0p^d 1_C = 0. We have a sequence of ideals

(E)

C⊃pC⊃p2C⊃...⊃pd−1C⊃pdC={0}. C \supset pC \supset p^2 C \supset ... \supset p^{d-1} C \supset p^d C = \{0\} .

If kk is an integer such that 0⩽k<d0 \leqslant k < d and pkC=pk+1Cp^k C = p^{k+1} C, we deduce

pd−k−1pkC=pd−k−1pk+1C={0} p^{d-k-1} p^k C = p^{d-k-1} p^{k+1} C = \{0\}

contrary to the hypothesis pd−11C≠0p^{d-1} 1_C \neq 0. Therefore the elements of the sequence (E) are pairwise distinct. Let aa be an ideal of CC and let kk be the smallest positive integer such that a⊃pkCa \supset p^k C. Let xx be a nonzero element of aa; we have seen that xx is of the form pmup^m u with m⩾0m \geqslant 0 and uu invertible in CC. Hence pmC⊂ap^m C \subset a, whence m⩾km \geqslant k, and finally x∈pkCx \in p^k C. In conclusion, one has a=pkCa = p^k C. The sequence (E) is then a Jordan-Hölder sequence of the CC-module CC, which is of length dd.

Corollary 1

If the pp-ring CC is an integral domain, it is a discrete valuation ring, or a field of characteristic pp.

Suppose CC is an integral domain. If p1Cp 1_C is nilpotent, one has p1C=0p 1_C = 0, and {0}\{0\} is a maximal ideal of CC, hence CC is a field of characteristic pp. If p1Cp 1_C is not nilpotent, then CC is a discrete valuation ring by Prop. 1, c).

Corollary 2

Let CC be a pp-ring and aa an ideal of CC distinct from CC. The ring C/aC/a is a pp-ring.

One may suppose a≠{0}a \neq \{0\}. There then exists an integer i⩾1i \geqslant 1 such that a=piCa = p^i C; the ideal pC/apC/a of C/aC/a is maximal and one has pi1C/a=0p^i 1_{C/a} = 0, therefore C/aC/a is a pp-ring.

Let CC be a pp-ring. The length of CC, denoted by l(C)l(C), is the least upper bound in R‾\overline{\mathbf{R}} of the set of integers n⩾1n \geqslant 1 such that pn−11C≠0p^{n-1} 1_C \neq 0. When l(C)l(C) is finite, it is the length of the CC-module CC, and when l(C)l(C) is equal to +∞+ \infty, the CC-module CC is not of finite length (prop. 1).

Example 1

For every integer n⩾1n \geqslant 1, the ring Z/pnZ\mathbf{Z}/p^n\mathbf{Z} is a pp-ring of length nn. The ring Zp\mathbf{Z}_p of pp-adic integers is a pp-ring of infinite length.

Example 2

Let KK be a perfect field of characteristic pp. By prop. 8 of § 1, No. 8, the ring W(K)W(K) of Witt vectors is a pp-ring of infinite length. The mapping (an)n∈N↦a0(a_n)_{n \in \mathbf{N}} \mapsto a_0 induces, by passing to quotients, an isomorphism of W(K)/pW(K)W(K)/pW(K) onto the field KK (loc. cit., prop. 7). For every integer n⩾1n \geqslant 1, the ring

Wn(K)=W(K)/pnW(K) W_n(K) = W(K)/p^nW(K)
is a pp-ring of length nn.

Proposition 2

Let CC and C′C' be two pp-rings and uu a homomorphism of CC into C′C'. Let vv be the homomorphism of κC=C/pC\kappa_C = C/pC into κC′=C′/pC′\kappa_{C'} = C'/pC' deduced from uu by passing to quotients.

a) One has l(C)⩾l(C′)l(C) \geqslant l(C'), and uu is injective if and only if one has l(C)=l(C′)l(C) = l(C').

b) In order that uu be surjective, it is necessary and sufficient that vv be an isomorphism.

c) In order that uu be an isomorphism, it is necessary and sufficient that vv be an isomorphism and that one have l(C)=l(C′)l(C) = l(C').

Let n⩾1n \geqslant 1 be an integer. We have u(pn−11C)=pn−11C′u(p^{n-1}1_C) = p^{n-1}1_{C'}, hence the relation pn−11C′≠0p^{n-1}1_{C'} \neq 0 implies pn−11C≠0p^{n-1}1_C \neq 0 and is equivalent to it if uu is injective. Therefore l(C′)⩽l(C)l(C') \leqslant l(C), with equality if uu is injective. If uu is not injective, there exists an integer i<l(C)i < l(C) such that the kernel of uu is the ideal piCp^iC of CC; then pi1C′=0p^i1_{C'} = 0, whence l(C′)⩽il(C') \leqslant i. This proves a).

Since κC\kappa_C and κC′\kappa_{C'} are fields, the homomorphism vv is injective. If uu is surjective, the same is true of vv, which is therefore an isomorphism. Conversely, suppose vv surjective. Then for every integer n⩾0n \geqslant 0, the mapping vn:pnC/pn+1C→pnC′/pn+1C′v_n : p^nC/p^{n+1}C \to p^nC'/p^{n+1}C' induced by uu is surjective. Since CC is complete for the pCpC-adic filtration and C′C' separated for the pC′pC'-adic filtration, uu is surjective by Cor. 2 of Th. 1 of III, § 2, No. 8. This proves b).

Finally, c) follows from a) and b).

Proposition 3

Let (Cn,πn,m)(C_n, \pi_{n,m}) be a projective system of rings relative to the set of indices N\mathbf{N}. Suppose that CnC_n is a pp-ring for every n∈Nn \in \mathbf{N} and that the homomorphisms πn,m\pi_{n,m} are surjective. Then C=lim←⁡CnC = \varprojlim C_n is a pp-ring, and for every n∈Nn \in \mathbf{N}, the canonical homomorphism πn:C→Cn\pi_n : C \to C_n is surjective and induces an isomorphism of κC\kappa_C onto κCn\kappa_{C_n}.

Since the mappings πn,m\pi_{n,m} are surjective, the same is true of the mappings πn\pi_n (E, III, p. 58, prop. 5). Let us show that CC is a pp-ring. Let dnd_n be the length of CnC_n. By prop. 2, a), the sequence of the elements dnd_n of N∪{+∞}\mathbf{N} \cup \{ + \infty \} is increasing; if it is stationary, there exists an integer n0n_0 such that πn,m\pi_{n,m} is an isomorphism of CmC_m onto CnC_n whenever n0⩽n⩽mn_0 \leqslant n \leqslant m, so that CC, isomorphic to Cn0C_{n_0}, is a pp-ring.

It is therefore enough to consider the case where each dnd_n is finite, and where the sequence (dn)(d_n) tends to +∞+ \infty. Endow the ring CC with the trivial filtration (III, § 2, No. 1, example 5). For n∈Nn \in \mathbf{N}, let InI_n be the kernel of πn\pi_n; put In=CI_n = C if n<0n < 0. Let EE denote the CC-module CC endowed with the filtration (In)n∈Z(I_n)_{n \in \mathbf{Z}}. It is separated and complete, for the topology T\mathscr{T} defined by the filtration (In)n∈Z(I_n)_{n \in \mathbf{Z}} is the projective limit topology of the discrete topologies on the CnC_n.

Let k be an integer ⩾1\geqslant 1. We have pkC⊂lim←⁡(pkCn)p^k C \subset \varprojlim (p^k C_n) (E, III, p. 55, formula (9)). Conversely, if x=(xn)n∈N∈lim←⁡(pkCn)x = (x_n)_{n \in \mathbf{N}} \in \varprojlim (p^k C_n) and if one puts Xn={y∈C∣πn(pky)=xn}X_n = \{ y \in C | \pi_n(p^k y) = x_n \}, the sequence (Xn)n∈N(X_n)_{n \in \mathbf{N}} is a decreasing sequence of nonempty closed affine parts of E. Since E/InE/I_n is an artinian C-module, the intersection of the XnX_n is nonempty (III, § 2, No. 7, prop. 7); for every z∈⋂n∈NXnz \in \bigcap_{n \in \mathbf{N}} X_n, one has pkz=xp^k z = x. We have therefore proved that one has pkC=lim←⁡pkCnp^k C = \varprojlim p^k C_n for every integer k⩾1k \geqslant 1. In particular the ideal pkCp^k C of C is closed for the topology T\mathscr{T}. On C, the pp-adic topology is finer than the topology T\mathscr{T} since one has pdnC⊂Inp^{d_n} C \subset I_n. It then follows from TG, III, p. 26, cor. 1 to prop. 10, that C is separated and complete for the pCpC-adic topology. Moreover one has pC=lim←⁡pCn=π0−1(pC0)pC = \varprojlim pC_n = \pi_0^{-1}(pC_0) and therefore the surjective homomorphism from C/pCC/pC into C0/pC0C_0/pC_0 deduced from π0\pi_0 is an isomorphism. This shows that the ideal pCpC of C is maximal and consequently that C is a pp-ring. The last assertion of prop. 3 follows from prop. 2, b).

2. Cohen Rings

Definition 2

Let A be a separated and complete local ring, whose residue field is of characteristic p. A Cohen subring of A means a subring C of A which is a pp-ring such that A=mA+CA = m_A + C (i.e. A/mA=C/(mA∩C)A/m_A = C/(m_A \cap C)).

If C is a Cohen subring of A, the ideal mA∩Cm_A \cap C of C is maximal, hence equal to pCpC. The canonical mapping of κC=C/pC\kappa_C = C/pC onto κA=A/mA\kappa_A = A/m_A is therefore a field isomorphism.

Example

Let C be a pp-ring. The ring of formal series A=C[[T1,...,Tn]]A = C[[T_1, ..., T_n]] is a noetherian, local, separated and complete ring, whose maximal ideal is generated by the sequence (p,T1,...,Tn)(p, T_1, ..., T_n). It is immediate that C is a Cohen subring of A. This applies in particular when C is equal to Zp\mathbf{Z}_p, to Z/pnZ\mathbf{Z}/p^n \mathbf{Z} or to a field of characteristic pp.

Theorem 1

Let A be a local, separated and complete ring, whose residue field k is of characteristic p. Let π\pi be the canonical mapping of A onto k, and let S be a subset of A, such that π\pi induces a bijection of S onto a pp-basis of k (A, V, p. 95). a) There exists one and only one Cohen subring C of A containing S. b) The subring C of A is closed, and the pCpC-adic topology of C is induced by the mAm_A-adic topology of A. c) Every closed subring A′A' of A, containing S, and such that A=A′+mAA = A' + m_A, contains C.

A) Particular case: mAm_A nilpotent

Let n be a positive integer such that mAn+1={0}m_A^{n+1} = \{0\}. If Φn\Phi_n is the nth Witt polynomial (§1§ 1, No. 1), the mapping u:[a0,...,an]↦Φn(a0,...,an)u : [a_0, ..., a_n] \mapsto \Phi_n(a_0, ..., a_n) is a ring homomorphism of Wn+1(A)W_{n+1}(A) into A (§1§ 1, No. 7). Let BnB_n be the image of uu and let CnC_n be the subring of A generated by Bn∪SB_n \cup S.

Lemma 1

Let A′A' be a subring of AA containing SS. In order that A′A' contain CnC_n, it is necessary and sufficient that one have A′+mA=AA' + m_A = A.

One has pA⊂mApA \subset m_A and BnB_n consists of the elements of the form a0pn+pa1pn−1+⋯+pnana_0^{p^n} + pa_1^{p^{n-1}} + \cdots + p^n a_n with a0,...,ana_0, ..., a_n in AA. Hence one has π(Bn)=kpn\pi(B_n) = k^{p^n}, whence π(Cn)=kpn[π(S)]\pi(C_n) = k^{p^n}[\pi(S)]. But since π(S)\pi(S) is a pp-basis of kk, one has k=kpn[π(S)]k = k^{p^n}[\pi(S)] (A, V, p. 96), whence π(Cn)=k\pi(C_n) = k, that is to say Cn+mA=AC_n + m_A = A.

Let A′A' be a subring of AA containing SS. If A′A' contains CnC_n, we have

A′+mA⊃Cn+mA=A,whenceA′+mA=A. A' + m_A \supset C_n + m_A = A, \quad \text{whence} \quad A' + m_A = A.
Conversely, suppose that one has A′+mA=AA' + m_A = A. Let a0,...,ana_0, ..., a_n be elements of AA; by assumption there exist elements a0′,...,an′a'_0, ..., a'_n of A′A' such that ai≡ai′mod  mAa_i \equiv a'_i \mod m_A for 0≤i≤n0 \leq i \leq n. By prop. 1 of § 1, No. 1 and the assumption mAn+1={0}m_A^{n+1} = \{0\}, we therefore have Φn(a0,...,an)=Φn(a0′,...,an′)∈A′\Phi_n(a_0, ..., a_n) = \Phi_n(a'_0, ..., a'_n) \in A', whence Bn⊂A′B_n \subset A'. Since CnC_n is the ring generated by Bn∪SB_n \cup S, one has Cn⊂A′C_n \subset A'.

In the set SS of subrings A′A' of AA containing SS and such that A′+mA=AA' + m_A = A, there exists by lemma 1 a smallest element CC, and one has Cn=CC_n = C for every integer n≥0n \geq 0 such that mAn+1={0}m_A^{n+1} = \{0\}.

One has C+mA=AC + m_A = A by construction and p1Cp1_C is nilpotent. One obviously has pC⊂C∩mApC \subset C \cap m_A and lemma 2 below therefore shows that pCpC is a maximal ideal of CC and consequently that CC is a Cohen subring of AA.

Lemma 2

One has C∩mA⊂pCC \cap m_A \subset pC.

Choose an integer m≥1m \geq 1 such that mAm={0}m_A^m = \{0\}, whence C=Cm=Cm−1C = C_m = C_{m-1}. Let Λ\Lambda be the subset of N(S)\mathbf{N}^{(S)} formed by the families of integers (αs)s∈S(\alpha_s)_{s \in S} with finite support satisfying 0≤αs<pm0 \leq \alpha_s < p^m for every s∈Ss \in S. Since BmB_m contains spm=Φm(s,0,...,0)s^{p^m} = \Phi_m(s, 0, ..., 0) for every s∈Ss \in S, the monomials Zα=∏s∈SsαsZ_\alpha = \prod_{s \in S} s^{\alpha_s}, where α\alpha runs through Λ\Lambda, generate CmC_m as a BmB_m-module.

Moreover, from the formula

Φm(a0,...,am)=a0pm+pΦm−1(a1,...,am), \Phi_m(a_0, ..., a_m) = a_0^{p^m} + p \Phi_{m-1}(a_1, ..., a_m),
every element of BmB_m is of the form apm+pba^{p^m} + pb with a∈Aa \in A and b∈Bm−1b \in B_{m-1}. Consequently every element of C=CmC = C_m is of the form
x=∑α∈ΛcαpmZα+py x = \sum_{\alpha \in \Lambda} c_\alpha^{p^m} Z_\alpha + py
with cα∈Ac_\alpha \in A for every α∈Λ\alpha \in \Lambda, and y∈Cm−1=Cy \in C_{m-1} = C. If xx belongs to C∩mAC \cap m_A, one has π(x)=0\pi(x) = 0, whence ∑α∈Λπ(cα)pmπ(Zα)=0\sum_{\alpha \in \Lambda} \pi(c_\alpha)^{p^m} \pi(Z_\alpha) = 0. Since π(S)\pi(S) is a pp-base of kk, one has π(cα)=0\pi(c_\alpha) = 0 for every α∈Λ\alpha \in \Lambda from A, V, p. 96. One then has cα∈mAc_\alpha \in m_A, whence cαm=0c_\alpha^m = 0 and a fortiori cαpm=0c_\alpha^{p^m} = 0. From (1), one has x=pyx = py, whence lemma 2.

One has pmC=mAm={0}p^m C = m_A^m = \{0\} for mm sufficiently large, and assertion b) is therefore trivial. Assertion c) results from lemma 1. If C′C' is a Cohen subring of AA containing SS, one has C′⊃CC' \supset C from lemma 1. But since the inclusion of CC in C′C' induces an isomorphism of κC\kappa_C onto κC′\kappa_{C'}, one has C=C′C = C' (No. 1, prop. 2, b)), and this completes the proving of a).

B) General case

For every integer n⩾0n \geqslant 0, let AnA_n denote the local ring A/mAn+1A/m_A^{n+1}, let mn=mA/mAn+1m_n = m_A/m_A^{n+1} denote its maximal ideal, and let πn\pi_n denote the canonical homomorphism of AA onto AnA_n. By A), there exists a unique Cohen subring CnC_n of AnA_n containing πn(S)\pi_n(S). When 0⩽n⩽m0 \leqslant n \leqslant m, we denote by πn,m\pi_{n,m} the canonical homomorphism of AmA_m onto AnA_n. By Corollary 2 of Proposition 1 of No. 1, πn,m(Cm)\pi_{n,m}(C_m) is a pp-ring; one has πn,m(Cm)+mn=An\pi_{n,m}(C_m) + m_n = A_n, hence πn,m(Cm)\pi_{n,m}(C_m) is equal to the Cohen subring CnC_n of AnA_n. By Proposition 3 of No. 1, the subring lim⁡Cn\lim C_n of lim⁡An\lim A_n is a pp-ring. Put C=⋂n∈Nπn−1(Cn)C = \bigcap_{n \in \mathbf{N}} \pi_n^{-1}(C_n). Since CC is the inverse image of lim⁡Cn\lim C_n under the isomorphism a↦(πn(a))n∈Na \mapsto (\pi_n(a))_{n \in \mathbf{N}} of AA onto lim⁡An\lim A_n, it is a closed subring of AA, and a pp-ring. One has πn(C)=Cn\pi_n(C) = C_n for every n∈Nn \in \mathbf{N} (No. 1, Proposition 3), and in particular π0(C)=A0\pi_0(C) = A_0, that is to say π(C)=k\pi(C) = k. Therefore CC is a Cohen subring of AA.

For every integer n⩾0n \geqslant 0, put Jn=C∩mAnJ_n = C \cap m_A^n. Since the local ring AA is separated, one has ⋂n∈NJn={0}\bigcap_{n \in \mathbf{N}} J_n = \{0\}, and in view of the structure of the ideals of a pp-ring (No. 1, Proposition 1), every ideal of CC of the form pkCp^kC contains one of the JnJ_n. Conversely, JnJ_n contains pnCp^nC. Consequently, the pCpC-adic topology of CC is induced by the mAm_A-adic topology of AA. This proves b).

Let A′A' be a closed subring of AA, containing SS and such that A′+mA=AA' + m_A = A. Since A′A' is closed, one has A′=⋂n∈Nπn−1(πn(A′))A' = \bigcap_{n \in \mathbf{N}} \pi_n^{-1}(\pi_n(A')). One has πn(A′)⊃πn(S)\pi_n(A') \supset \pi_n(S) and πn(A′)+mn=An\pi_n(A') + m_n = A_n, whence πn(A′)⊃Cn\pi_n(A') \supset C_n by what was seen in A). Finally, one has πn−1(πn(A′))⊃πn−1(Cn)\pi_n^{-1}(\pi_n(A')) \supset \pi_n^{-1}(C_n), whence A′⊃CA' \supset C. This proves c). One deduces the uniqueness of a Cohen subring as in A).

Remark

Suppose that p1Ap1_A is not nilpotent (this takes place in particular when AA is an integral ring whose field of fractions is of characteristic 0). Then CC is a discrete valuation ring whose field of fractions is of characteristic 0.

3. Existence and uniqueness of pp-rings

Proposition 4

Let CC and C′C' be two pp-rings such that l(C)⩾l(C′)l(C) \geqslant l(C'), and let π\pi (resp. π′\pi') be the canonical homomorphism of CC (resp. C′C') onto κC\kappa_C (resp. κC′\kappa_{C'}). Let (xλ)λ∈Λ(x_\lambda)_{\lambda \in \Lambda} (resp. (xλ′)λ∈Λ(x'_\lambda)_{\lambda \in \Lambda}) be a family of elements of CC (resp. C′C') whose image under π\pi (resp. π′\pi') is a pp-basis of κC\kappa_C (resp. κC′\kappa_{C'}). Let vv be an isomorphism of κC\kappa_C onto κC′\kappa_{C'} such that v(π(xλ))=π′(xλ′)v(\pi(x_\lambda)) = \pi'(x'_\lambda) for every λ∈Λ\lambda \in \Lambda. Then there exists a unique homomorphism uu of CC into C′C', such that v∘π=π′∘uv \circ \pi = \pi' \circ u and u(xλ)=xλ′u(x_\lambda) = x'_\lambda for every λ∈Λ\lambda \in \Lambda. It is surjective. If l(C)=l(C′)l(C) = l(C'), it is an isomorphism.

Let us prove the existence of uu. Let AA be the subring of C×C′C \times C' formed by the pairs (x,x′)(x, x') such that v(π(x))=π′(x′)v(\pi(x)) = \pi'(x'). The mapping (x,x′)↦π(x)(x, x') \mapsto \pi(x) is a surjective homomorphism of rings from A onto κC\kappa_C. Its kernel mm, equal to pC×pC′pC \times pC', is therefore a maximal ideal of A. The topological subspace A of C×C′C \times C' is closed in C×C′C \times C', hence complete, and the topology induced on A by that of C×C′C \times C' is the mm-adic topology. Hence A is a separated complete local ring with maximal ideal mm (III, § 2, No. 13, Prop. 19). For every λ∈Λ\lambda \in \Lambda, we have (xλ,xλ′)∈A(x_\lambda, x'_\lambda) \in A by hypothesis; if ξλ\xi_\lambda is the class of (xλ,xλ′)(x_\lambda, x'_\lambda) modulo mm, the family (ξλ)λ∈Λ(\xi_\lambda)_{\lambda \in \Lambda} is a pp-basis of the field A/mA/m. By theorem 1 of No. 2, there exists a Cohen subring C′′C'' of A, and only one, containing (xλ,xλ′)(x_\lambda, x'_\lambda) for every λ∈Λ\lambda \in \Lambda. We have l(C′′)=l(C)≥l(C′)l(C'') = l(C) \geq l(C'). The restriction to C′′C'' of the projection of C×C′C \times C' onto C is a homomorphism h:C′′→Ch : C'' \to C which induces an isomorphism of κC′′\kappa_{C''} onto κC\kappa_C. By Prop. 2, c) of No. 2, hh is an isomorphism of C′′C'' onto C. One sees analogously that the restriction h′h' to C′′C'' of the projection of C×C′C \times C' onto C′C' is a surjective homomorphism of C′′C'' onto C′C'. Hence, C′′C'' is the graph of a surjective homomorphism u=h′∘h−1u = h' \circ h^{-1} of C onto C′C', and one obviously has v∘π=π′∘uv \circ \pi = \pi' \circ u, u(xλ)=xλ′u(x_\lambda) = x'_\lambda for every λ∈Λ\lambda \in \Lambda. Moreover, if l(C)=l(C′)l(C) = l(C'), uu is an isomorphism.

Let us prove the uniqueness of uu. Let u1u_1 be a homomorphism of C into C′C' such that v∘π=π′∘u1v \circ \pi = \pi' \circ u_1 and u1(xλ)=xλ′u_1(x_\lambda) = x'_\lambda for every λ∈Λ\lambda \in \Lambda, and let C1C_1 be the graph of u1u_1. It is immediate that C1C_1 is a Cohen subring of A, containing (xλ,xλ′)(x_\lambda, x'_\lambda) for every λ∈Λ\lambda \in \Lambda, whence C1=C′′C_1 = C'' (theorem 1 of No. 2) and finally u1=uu_1 = u.

Proposition 5

Let k be a field of characteristic p, and let n be an integer ≥1\geq 1, or +∞+ \infty. There exists a p-ring of length n whose residue field is isomorphic to k.

The ring W(k)W(k) of Witt vectors with coefficients in k is an integral separated complete local ring whose residue field is isomorphic to k (§1§ 1, no. 8, prop. 8), and one has p⋅1W(k)≠0p \cdot 1_{W(k)} \neq 0 (loc. cit., formula (52)). Let C be a Cohen subring of W(k)W(k) (no. 2, th. 1). Then C is a p-ring of length +∞+ \infty whose residue field is isomorphic to k, and, if n is an integer ≥1\geq 1, the quotient C/pnCC/p^nC is a p-ring of length n whose residue field is isomorphic to k.

Remark 1

Let n be an integer ≥1\geq 1 and S a p-basis of k. One can prove that the subring of Wn(k)W_n(k) generated by Wn(kpn)W_n(k^{p^n}) and by the elements [ξ,0,...,0][\xi, 0, ..., 0] (ξ∈S\xi \in S), is a p-ring of length n whose residue field is isomorphic to k (cf. p. 72, exerc. 10).

Remark 2

The reader will find in the Appendix a proof of prop. 5 which uses neither the results of §1§ 1, nor the existence theorem for Cohen subrings (no. 2, th. 1).

Corollary

Let C be a p-ring of finite length n. There exists a p-ring C′C' of infinite length such that C is isomorphic to C′/pnC′C'/p^nC'.

By prop. 5, there exists a p-ring C′C' of infinite length such that κC′\kappa_{C'} is isomorphic to κC\kappa_C. Then C′/pnC′=Cn′C'/p^nC' = C'_n is a p-ring of length n, and the field κCn′\kappa_{C'_n} is isomorphic to κC′\kappa_{C'}, hence to κC\kappa_C. By prop. 4, the rings C and Cn′C'_n are therefore isomorphic.

4. Multiplicative Representatives

Proposition 6

Let C be a p-ring, whose residue field k be perfect. Suppose that C is of finite length n (resp. infinite). There exists a unique isomorphism u:Wn(k)→Cu : W_n(k) \to C (resp. u:W(k)→Cu : W(k) \to C) which induces, by passing to quotients, the identical mapping of k.

Since Wn(k)W_n(k) (resp. W(k)W(k)) is a p-ring with residue field k, and of length n (resp. of infinite length) (No. 1, Example 2), and since ∅\varnothing is a p-basis of the perfect field k, prop. 6 is a particular case of prop. 4 of No. 3.

Theorem 2

Let A be a separated complete local ring, k its residue field and π\pi the canonical homomorphism of A onto k. It is assumed that k is a perfect field of characteristic p.

a) There exists a unique ring homomorphism u:W(k)→Au : W(k) \to A such that π(u(a))=a0\pi(u(a)) = a_0 for a=(an)n∈Na = (a_n)_{n \in \mathbf{N}} in W(k).

b) The homomorphism u is continuous when W(k) is endowed with the pW(k)pW(k)-adic topology, and the image of u is the unique Cohen subring of A.

By th. 1 of No. 2, there exists a unique Cohen subring of A; let us denote it by C. Let u be a homomorphism of W(k) into A such that π(u(a))=a0\pi(u(a)) = a_0 for every a=(an)n∈Na = (a_n)_{n \in \mathbf{N}} in W(k); it is immediate that the image of u is a Cohen subring of A, hence equal to C. The existence and uniqueness of u then result from prop. 6. The pCpC-adic topology of C is induced by the mAm_A-adic topology of A (No. 2, th. 1, b)), whence the continuity of u.

For a direct construction of u, see p. 70, exerc. 6.

Proposition 7

Let us retain the hypotheses and notations of th. 2. There exists a unique multiplicative subset S of A such that π\pi induces a bijection of S onto k. In order that an element a of A belong to S, it is necessary and sufficient that for every n∈Nn \in \mathbf{N}, there exist an element ana_n of A such that a=anpna = a_n^{p^n}. The set S is the set of elements of the form u(x,0,0,...)u(x, 0, 0, ...).

Let us first prove the uniqueness of S. Let S be a multiplicative subset of A, such that π\pi induces a bijection of S onto k. Let T be the set of elements of A which are pnp^n-th powers for every n∈Nn \in \mathbf{N}.

a) One has S⊂TS \subset T : Let a∈Sa \in S and n∈Nn \in \mathbf{N}; since the field k is perfect, there exists an element xnx_n of k such that xnpn=π(a)x_n^{p^n} = \pi(a); since one has π(S)=k\pi(S) = k, there exists an element ana_n of S such that xn=π(an)x_n = \pi(a_n). One then has π(anpn)=π(a)\pi(a_n^{p^n}) = \pi(a) whence anpn=aa_n^{p^n} = a since the restriction of π\pi to S is injective.

b) The restriction of π\pi to T is injective: let a and b be two elements of T such that π(a)=π(b)\pi(a) = \pi(b). Let n∈Nn \in \mathbf{N}; there exist two elements ana_n and bnb_n of A such that a=anpn,b=bnpna = a_n^{p^n}, b = b_n^{p^n}. We then have π(an)pn=π(bn)pn\pi(a_n)^{p^n} = \pi(b_n)^{p^n}, whence π(an)=π(bn)\pi(a_n) = \pi(b_n), that is to say an≡bna_n \equiv b_n mod. mAm_A. By Lemma 1 of § 1, No. 1, we have anpn≡bnpna_n^{p^n} \equiv b_n^{p^n} mod. mAn+1m_A^{n+1} that is to say a≡ba \equiv b mod. mAn+1m_A^{n+1}. Since n is arbitrary, we have a=ba = b.

Properties a) and b) above, together with the formula π(S)=k\pi(S) = k, imply the relation S=TS = T, whence uniqueness.

Let us now prove the existence of S. With the notation of Theorem 2, put φ=u∘τk\varphi = u \circ \tau_k, that is to say (§ 1, No. 6)

(2)

φ(x)=u(x,0,0,...) \varphi(x) = u(x, 0, 0, ...)
for every x∈kx \in k. By Proposition 4 of loc. cit., we have

(3)

φ(1)=1,φ(xy)=φ(x)φ(y) for x,y in k. \varphi(1) = 1,\quad \varphi(xy) = \varphi(x) \varphi(y) \text{ for } x, y \text{ in } k.

It is clear that the mapping π∘φ\pi \circ \varphi is the identity mapping of kk. Therefore the image S of φ\varphi satisfies the conditions of Proposition 7.

The elements of S are often called the multiplicative (or Teichmüller) representatives of A.

Remark 1

Let us retain the preceding hypotheses and notation. We have

a=∑n=0∞pnτk(anp−n)(a=(an)n∈N∈W(k)) a = \sum_{n=0}^{\infty} p^n \tau_k(a_n^{p^{-n}}) \quad (a = (a_n)_{n \in \mathbf{N}} \in \mathbf{W}(k))
by Proposition 7 of § 1, No. 8. Hence (4)
u(a)=∑n=0∞pnφ(anp−n) u(a) = \sum_{n=0}^{\infty} p^n \varphi(a_n^{p^{-n}})
for every a=(an)n∈Na = (a_n)_{n \in \mathbf{N}} in W(k)\mathbf{W}(k), since uu is continuous (Theorem 2, b)). By formula (4), the unique Cohen subring of A consists of the elements of the form ∑n=0∞pnsn\sum_{n=0}^{\infty} p^n s_n with sn∈Ss_n \in S for every integer n≥0n \geq 0.

Remark 2

Let A be a separated and complete local ring, kk its residue field and π\pi the canonical homomorphism of A onto kk. It may be shown that there exists a multiplicative subset S of A (not unique in general) such that π\pi induces a bijection of S onto kk (cf. p. 72, Exercise 11).

Example 1

Let kk be a perfect field of characteristic pp. The multiplicative representatives of the ring W(k)\mathbf{W}(k) are the Witt vectors τ(x)=(x,0,0,...)\tau(x) = (x, 0, 0, ...) for x∈kx \in k.

Example 2

Let A be an integral, separated and complete local ring. Suppose that the residue field kk of A is finite, with q=pfq = p^f elements, hence perfect of characteristic pp. We have xq=xx^q = x for every x∈kx \in k, whence sq=ss^q = s for every multiplicative representative s. It follows that the set of multiplicative representatives consists of 0 and the q−1q - 1 (q−1)(q - 1)-th roots of unity in the field of fractions of A. If the field of fractions of A is locally compact, the existence of multiplicative representatives also follows from VI, § 9, No. 2, Proposition 3 (cf. also VI, § 9, Exercise 5).

Example 3

More particularly, consider the case A=ZpA = \mathbf{Z}_p. Then the multiplicative representatives are 0 and the (p−1)(p - 1)-th roots of unity in the field of fractions Qp\mathbf{Q}_p of Zp\mathbf{Z}_p.

5. Structure of complete noetherian local rings

Let AA and CC be complete noetherian local rings, and let uu be a local homomorphism of CC into AA, inducing by passing to quotients an isomorphism of κC\kappa_C onto κA\kappa_A. Let (p1,...,pm)(p_1, ..., p_m) be a sequence generating the ideal mCm_C of CC, and let t1,...,tnt_1, ..., t_n be elements of mAm_A. Put B=C[[T1,...,Tn]]B = C[[T_1, ..., T_n]].

Lemma 3

a) There exists a unique homomorphism v:B→Av : B \to A extending uu and mapping TiT_i to tit_i for 1≤i≤n1 \leq i \leq n.

b) In order that vv be surjective, it is necessary and sufficient that the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) generate the ideal mAm_A of AA, or again that the classes of these elements modulo mA2m_A^2 generate mA/mA2m_A/m_A^2 as a vector space over the field κA\kappa_A.

c) In order that vv make AA into a finite BB-algebra, it is necessary and sufficient that the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) generate an ideal of definition of (the mAm_A-adic topology of) AA.

Let nn denote the ideal of the ring BB generated by T1,...,TnT_1, ..., T_n. Every homomorphism vv of BB into AA extending uu and such that v(Ti)=tiv(T_i) = t_i maps nn into mAm_A, and is therefore continuous when BB is endowed with the nn-adic topology. The existence and uniqueness of vv then result from A, IV, p. 26, Prop. 4.

The ring B=C[[T1,...,Tn]]B = C[[T_1, ..., T_n]] is a complete noetherian local ring (III, § 2, No. 10, Corollary 6 to Theorem 2 and No. 6, Proposition 6), whose maximal ideal mBm_B is generated by p1,...,pm,T1,...,Tnp_1, ..., p_m, T_1, ..., T_n. Hence v(mB)⊂mAv(m_B) \subset m_A and vv defines a homomorphism gr(v)(v) of gr⁡(B)=⨁n=0∞mBn/mBn+1\operatorname{gr}(B) = \bigoplus_{n=0}^\infty m_B^n/m_B^{n+1} into gr⁡(A)=⨁n=0∞mAn/mAn+1\operatorname{gr}(A) = \bigoplus_{n=0}^\infty m_A^n/m_A^{n+1}. Now the ring gr⁡(A)\operatorname{gr}(A) is generated by A/mA=κAA/m_A = \kappa_A and mA/mA2m_A/m_A^2, gr(v)(v) induces an isomorphism of κB=κC\kappa_B = \kappa_C onto κA\kappa_A, and the classes modulo mB2m_B^2 of the elements p1,...,pm,T1,...,Tnp_1, ..., p_m, T_1, ..., T_n generate mB/mB2m_B/m_B^2 as a vector space over κB\kappa_B; moreover vv is surjective if and only if gr(v)(v) is surjective (III, § 2, No. 8, Corollary 2 to Theorem 1). This proves bb.

The ideal of AA generated by the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) is none other than v(mB)v(m_B) AA. Since mAm_A contains v(mB)v(m_B), AA is a Zariski ring for the v(mB)v(m_B) AA-adic topology. The ring A/v(mB)A/v(m_B) AA is artinian if and only if its length as an AA-module is finite. But since every simple module over AA is annihilated by mAm_A and since, by hypothesis, A/mAA/m_A and B/mBB/m_B are isomorphic, this occurs if and only if the dimension over the field B/mBB/m_B of the vector space A/v(mB)A/v(m_B) AA is finite. By IV, § 2, No. 5, Corollary 2 of Proposition 9, we see therefore that v(mB)v(m_B) AA is an ideal of definition of AA if and only if the dimension of A/v(mB)A/v(m_B) AA over B/mBB/m_B is finite. This is indeed the case if AA is a finite BB-algebra.

Suppose that v(mB)v(m_B) AA is an ideal of definition of AA. The mBm_B-adic topology of the BB-module AA then coincides with the mAm_A-adic topology of the ring AA, and is therefore separated. Since A/v(mB)A/v(m_B) AA is a finitely generated module over B/mBB/m_B, AA is a finitely generated BB-module (III, § 2, No. 3, Example 3 and No. 9, cor. 1 of Prop. 12). This proves cc.

Lemma 4

Suppose that the noetherian local ring C is regular, and that (p1,...,pm)(p_1, ..., p_m) is a system of coordinates of C (VIII, § 5, No. 1, Def. 1).

a) If the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) is secant for A (VIII, § 3, No. 2, Def. 1), the homomorphism v:B→Av : B \to A is injective.

b) In order that v be injective and make A a finite algebra over B, it is necessary and sufficient that (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) be a maximal secant sequence for A. Then A is of dimension m+nm + n.

In order that the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) be a maximal secant sequence for A, it is necessary and sufficient that it generate an ideal of definition of A, and that A be of dimension m+nm + n (VIII, § 3, No. 2, Th. 1). By lemma 3, c), this comes to saying that A is a finite B-algebra, and a ring of dimension m+nm + n. Now C is a noetherian integral ring of dimension mm, hence B=C[[T1,...,Tn]]B = C[[T_1, ..., T_n]] is a noetherian integral ring of dimension m+nm + n (VIII, § 3, No. 4, cor. 3 of Prop. 8). If A is a finite B-algebra, and if aa is the kernel of vv, one has dim⁡(A)=dim⁡(B/a)\dim(A) = \dim(B/a) (VIII, § 2, No. 3, th. 1, c)); since B is an integral ring of finite dimension, one has dim⁡(B/a)<dim⁡(B)\dim(B/a) < \dim(B) if a≠{0}a \neq \{0\} (VIII, § 1, No. 3, prop. 6, e)). Hence, if A is a finite B-algebra, vv is injective if and only if A is of dimension m+nm + n. This proves b).

Suppose that the sequence (u(p1),...,u(pm),t1,...,tn)(u(p_1), ..., u(p_m), t_1, ..., t_n) of elements of mAm_A is secant. One can adjoin to it (VIII, § 3, No. 2, Th. 1) elements tn+1,...,tn+rt_{n+1}, ..., t_{n+r} of mAm_A in order to make it into a maximal secant sequence. From what precedes, there then exists an injective homomorphism ww of C[[T1,...,Tn,Tn+1,...,Tn+r]]=B[[Tn+1,...,Tn+r]]C[[T_1, ..., T_n, T_{n+1}, ..., T_{n+r}]] = B[[T_{n+1}, ..., T_{n+r}]] which extends vv and maps Tn+jT_{n+j} to tn+jt_{n+j} for 1≤j≤r1 \leq j \leq r. Hence vv is injective. This proves a).

Theorem 3

Let A be a local, noetherian and complete ring whose residue field k is of characteristic p. Let C be a p-ring of infinite length, whose residue field is isomorphic to k (No. 3, Prop. 5).

a) Let m be the dimension of the vector space mA/(mA2+pA)m_A/(m_A^2 + pA) over the field k. There exists an ideal aa of the ring C[[T1,...,Tm]]C[[T_1, ..., T_m]] such that A is isomorphic to C[[T1,...,Tm]]/aC[[T_1, ..., T_m]]/a.

b) Let d be the dimension of A. Suppose that p1Ap1_A is not a divisor of 0 in A. Then there exists a subring A′A' of A isomorphic to C[[T1,...,Td−1]]C[[T_1, ..., T_{d-1}]] and such that A is a finite algebra over A′A'.

Let C′C' be a Cohen subring of AA (No. 2, Theorem 1). Since CC is of infinite length, there exists a homomorphism of CC onto C′C' (No. 3, Proposition 4). Consequently, there exists a local homomorphism u:C→Au : C \to A. Let us choose elements t1,...,tmt_1, ..., t_m of mAm_A whose classes form a basis of the vector space mA/(mA2+pA)m_A/(m_A^2 + pA) over the field kk. We have u(p1C)=p1Au(p1_C) = p1_A, and Lemma 3, b) proves the existence of a surjective homomorphism of C[[T1,...,Tm]]C[[T_1, ..., T_m]] onto AA, extending uu and mapping TiT_i to tit_i for 1≤i≤m1 \leq i \leq m. This proves a).

Suppose that p1Ap1_A is not a divisor of 00 in AA, hence is secant for AA (VIII, § 3, No. 2, Proposition 3). Then there exist (VIII, § 3, No. 2, Theorem 1) elements t1,...,td−1t_1, ..., t_{d-1} of mAm_A such that the sequence (p1A,t1,...,td−1)(p1_A, t_1, ..., t_{d-1}) is a maximal secant sequence for AA. The noetherian local ring CC is regular, and (p1C)(p1_C) is a system of coordinates of CC. The assertion b) of Theorem 3 then follows from Lemma 4, b).

Exercises

In the exercises of § 2, pp is a fixed prime number. If aa is an ideal of a ring AA, apa^p denotes the ideal generated by the elements apa^p, where aa ranges over aa.

See the exercises for § 2.

Machine translation of the English, by gpt-5.4, against glossary version 34. Not checked by a person.