betaFeedback

WorksMathematics

Coherence and the Hilbert-syzygy identity for power series over valuation domains

Abstract

Let VV be a complete valuation domain of rank 1, and let ΓΓ be its value group. We investigate the power series ring V[[t]]V[[t]], and finally answer the long standing question of when this ring is coherent. Specifically, we prove that it is coherent if and only if VV is almost maximal, and either Γ=RΓ= \mathbb{R} or ΓΓ is discrete. Furthermore, we demonstrate that the weak Hilbert-syzygy identity w.dimV[[X]]=w.dimV+1w. dim V [[X]] = w. dim V + 1 holds if and only if V[[X]]V[[X]] is a coherent ring.

Introduction

In this article, we consider all rings to be commutative with identity. Recall that a ring SS is coherent if every finitely presented SS-module has a finite free resolution (or equivalently, every finitely generated ideal of SS is finitely presented).

One of the most classical examples of a coherent ring is a valuation domain, i.e. commutative domain VV where for all ideals I,J⊴VI, J \trianglelefteq V, either I⊆JI \subseteq J or J⊆IJ \subseteq I. In this case, all finitely generated ideals are principal. Note that all fractional ideals of K=Frac⁡(V)K = \operatorname{Frac}(V) can be indexed by a totally ordered abelian group Γ\Gamma, which we call the value group of VV.

It is clear that any Noetherian ring is coherent. But unlike Noetherianity, it it not generally true that constructions such as polynomial rings or power series rings preserve coherence; there are examples of coherent rings SS such that S[X]S[X] is not coherent.

However, if VV is a valuation ring, then V[X]V[X] is a coherent ring, but the power series ring V[[X]]V[[X]] is not in general. In fact, it was proved by Jøndrup and Small in [6 Section 1] that if VV has rank greater than 11, then V[[X]]V[[X]] is never coherent, and they remarked that it is unknown whether or not V[[X]]V[[X]] is coherent when VV has rank 11 (i.e. when Γ\Gamma is a subgroup of R\mathbb{R}).

All subgroups of R\mathbb{R} are either discrete or dense. If Γ\Gamma is discrete, then VV is Noetherian, so V[[X]]V[[X]] is also Noetherian. However, it was first proved in [1], and later by the second author in [5] that if Γ\Gamma is proper and dense in R\mathbb{R}, then V[[X]]V[[X]] is not coherent, and it was recently proved by the third author in [9] that if Γ=R\Gamma= \mathbb{R} and VV is complete, then V[[X]]V[[X]] is coherent if and only if KK is spherically complete. In general, when Γ=R\Gamma= \mathbb{R}, the question of coherence of V[[X]]V[[X]] remains open.

In this article, we link this question to another open question for power series rings, namely the question of Hilbert’s syzygy theorem: For a ring SS and an SS-module MM, let fd⁡SM\operatorname{fd}_S M denote the flat dimension of MM, and let w. dim⁡S\operatorname{w.\ dim} S denote the weak global dimension of SS, i.e

w. dim⁡S:=sup⁡{fd⁡S(M):M an S-module}\operatorname{w.\ dim} S := \sup\{\operatorname{fd}_S(M) : M\text{ an }S\text{-module}\}

The classical Hilbert’s syzygy theorem states that for any ring RR, w. dim⁡R[X]=w. dim⁡R+1\operatorname{w.\ dim} R[X] = \operatorname{w.\ dim} R + 1, and an ongoing project in commutative algebra is to establish for which rings RR the same result holds for the power series ring, i.e.

w. dim⁡R[[X]]=w. dim⁡R+1(1)\operatorname{w.\ dim} R[[X]] = \operatorname{w.\ dim} R + 1 \tag*{(1)}

For example, VV is a non-field valuation ring, then all ideals in VV are flat, so w. dim⁡(V)=1\operatorname{w.\ dim}(V) = 1. Thus V[[X]]V[[X]] satisfies Hilbert’s syzygy identity (1) if and only if w. dim⁡(V[[X]])=2\operatorname{w.\ dim}(V[[X]]) = 2.

It was proved by Jøndrup and Small that if the power series ring R[[X]]R[[X]] is coherent (e.g. if RR is Noetherian), then (1) is satisfied [6 Lemma 1]. The question was investigated further by Bouchiba in [4], who claimed the identity holds for any coherent ring RR, and moreover that the stronger identity fd⁡R[[X]]M=fd⁡R(M)+1\operatorname{fd}_{R[[X]]} M = \operatorname{fd}_R(M)+1 holds for any R[[X]]R[[X]]-module MM.

However, the second author identified a gap in this argument in [4], and in fact showed that if VV is a rank-one valuation domain VV with proper dense value group [5 Theorem A]. More precisely, there exists a V[[X]]V[[X]]-module CC satisfying

fd⁡VC=0,fd⁡V[[X]]C≥2.\operatorname{fd}_V C = 0,\qquad\operatorname{fd}_{V[[X]]} C \ge2.

It was remarked in the conclusion to [5] that this does not itself disprove Hilbert’s syzygy identity of VV, and that to this end an additional homological step would be required. Thus the question remained open.

The main result of this article provides a full answer to both these open questions. In the below, we let V((X)):=V[[X]][X−1]V((X)) := V[[X]][X^{-1}] be the ring of Laurent series over VV, and we define what it means for VV to be almost maximal in section 2.

Theorem A. Let VV be a valuation ring of rank 1, with value group Γ⊆R\Gamma\subseteq\mathbb{R}. Then the following are equivalent:

i.i. V[[X]]V[[X]] is a coherent ring.

ii.ii. V[[X]]V[[X]] satisfies Hilbert’s syzygy identity (1).

iii.iii. VV is an almost maximal valuation ring, and either Γ=0\Gamma= 0, Γ=R\Gamma= \mathbb{R} or Γ≅Z\Gamma\cong\mathbb{Z}.

In all other cases, w. dim⁡V[[X]]=w. dim⁡V((X))>2\operatorname{w.\ dim} V[[X]] = \operatorname{w.\ dim} V((X)) > 2.

Note: The condition Γ=0\Gamma= 0 just means that VV is a field, while Γ≅Z\Gamma\cong\mathbb{Z} means that VV is discrete. In both these cases, VV is Noetherian, so we know that V[[X]]V[[X]] is Noetherian and satisfies w. dim⁡(V[[X]])=gl. dim⁡(V[[X]])=gl. dim⁡(V)+1\operatorname{w.\ dim}(V[[X]]) = \operatorname{gl.\ dim}(V[[X]]) = \operatorname{gl.\ dim}(V) + 1.

Again, i⟹iii \Longrightarrow ii was proved by Jøndrup-Small in [6 Lemma 1], so this theorem constitutes a partial converse to their result. To complete the proof, the only steps remaining are to prove i⟺iiii \Longleftrightarrow iii and ii⟹iiiii \Longrightarrow iii.

After proving some essential preliminary results in section 2, we go on in section 3 to complete the proof that V[[X]]V[[X]] is a coherent ring if and only if VV satisfies the stated conditions in iiiiii. The proof that i⟹iiii \Longrightarrow iii was given by the third author in [9 Section 5] (with the assumption of maximality rather than almost maximality, but the argument still holds by [9 Remark 4.3]), and the proof of iii⟹iiii \Longrightarrow i is completely analogous to the argument in [9 Section 3].

The remainder of the paper is dedicated to proving ii⟹iiiii \Longrightarrow iii. Specifically, we will prove that if VV does not satisfy the conditions stated in iiiiii, then V[[X]]V[[X]] does not satisfy Hilbert’s syzygy identity. Moreover, the inequality w.dim⁡R[[X]]≥w.dim⁡R+1\operatorname{w.dim} R[[X]] \ge\operatorname{w.dim} R + 1 is satisfied for all coherent rings RR by the proof of [6 Lemma 1.1], so since w.dim⁡V=1\operatorname{w.dim} V = 1 for any non-field valuation ring VV, we only need to prove that w.dim⁡V[[X]]≥3\operatorname{w.dim} V[[X]] \ge3 if VV does not satisfy the stated conditions.

To show that w.dim⁡R[[X]]≥3\operatorname{w.dim} R[[X]] \ge3 for any commutative domain RR, it suffices to show that there exist elements f,r∈R[[X]]f,r \in R[[X]] such that (f)∩(r)(f) \cap(r) is not flat in R[[X]]R[[X]], and apply two dimensional Tor-shifts. So it remains to construct such elements f,r∈V[[X]]f,r \in V[[X]] under the appropriate conditions on VV.

In section 4, we consider the case where Γ\Gamma is a proper, dense subgroup of R\mathbb{R}, and following the approach of the second author in [5] we utilise the coefficient-infimum function

ν(∑n≥0anXn)=inf⁡n≥0v(an)(2)\nu\left(\sum_{n \ge0} a_n X^n\right) = \inf_{n \ge0} v(a_n) \tag*{(2)}

which is a valuation on V[[X]]V[[X]] by [5 Corollary 2.5]. We consider the cut ideals

Is={g∈V[[X]]:ν(g)≥s}.I_s = \{g \in V[[X]] : \nu(g) \ge s\}.

for each s>0s > 0. A fractional-intersection identity satisfied by every nonzero flat ideal then shows that IcI_c is not flat if c∉Γc \notin\Gamma. Thus we construct f,r∈V[[X]]f,r \in V[[X]] such that (f)∩(r)=fIc(f) \cap(r) = fI_c, and hence (f)∩(r)(f) \cap(r) is not flat as required.

In section 5, we turn to the case where Γ=R\Gamma= \mathbb{R}, and focus on the case where K=Frac⁡(V)K = \operatorname{Frac}(V) is complete, but not spherically complete. This was the case considered by the third author in [9 Section 4], where a pair f,r∈V[[X]]f,r \in V[[X]] was constructed from the tails of infinite products, and the intersection (f)∩(r)(f) \cap(r) was proved to be of infinite presentation. We strengthen this construction homologically, and apply a flat-descent lemma to show that, under a suitable coprimality condition in a GCD overring, flatness of the intersection would force invertibility. This suffices to prove that (f)∩(r)(f) \cap(r) is not flat.

Finally, in section 6, we consider the remaining case where Γ=R\Gamma= \mathbb{R} but VV is not almost maximal. Equivalently, the topological completion K^\widehat{K} is not spherically complete. Setting V^=OK^\widehat{V} = \mathcal{O}_{\widehat{K}}, we prove that w.dim⁡V[[X]]=w.dim⁡V^[[X]]\operatorname{w.dim} V[[X]] = \operatorname{w.dim} \widehat{V}[[X]], and hence w.dim⁡(V[[X]])>2\operatorname{w.dim}(V[[X]]) > 2.

The question of whether Hilbert’s syzygy identity (1) is satisfied for any valuation ring VV of rank greater than 1 remains open. However, given that no example of any RR such that R[[X]]R[[X]] is not coherent but the syzygy identity is satisfied seems to appear in the literature, we tentatively make the following conjecture.

Conjecture 1.1. If RR is a coherent ring, then R[[X]]R[[X]] is coherent if and only if w.dim(R[[X]])=w.dim(R)+1\mathrm{w.dim}(R[[X]]) = \mathrm{w.dim}(R) + 1.

Prerequisites

In this section, we will prove some technical results which will be essential in our main argument.

Flatness criteria

For now, let AA be any commutative domain (later we will realise AA as a power series ring). We now isolate some key flatness criteria.

Lemma 2.1. Let L=Frac⁡(A)L = \operatorname{Frac}(A), and let E≠0E \ne0 be a flat ideal of AA. If P,Q⊆LP,Q \subseteq L are AA-submodules, then

E(P∩Q)=EP∩EQ.(3)E(P \cap Q) = EP \cap EQ. \tag*{(3)}

Proof. Tensor the exact sequence

0⟶P∩Q⟶P⊕Q⟶P+Q⟶00 \longrightarrow P \cap Q \longrightarrow P \oplus Q \longrightarrow P + Q \longrightarrow0

with EE. For every AA-submodule P⊆LP \subseteq L, the multiplication map

E⊗AP⟶LE \otimes_A P \longrightarrow L

is injective. Indeed, flatness preserves the injection P↪LP \hookrightarrow L, and, with U=A∖{0}U = A \setminus\{0\},

E⊗AL≅U−1E=LE \otimes_A L \cong U^{-1}E = L

because E≠0E \ne0. Under the multiplication map, the image of E⊗APE \otimes_A P is EPEP. The kernel in the tensored sequence therefore identifies both with E(P∩Q)E(P \cap Q) and with EP∩EQEP \cap EQ, proving (3). □

Lemma 2.2. Suppose f,r∈Af,r \in A and the ideal (f)∩(r)(f) \cap(r) is not a flat AA-module. Then

fd⁡A(A/(f,r))≥3.\operatorname{fd}_A(A/(f,r)) \ge3.

Consequently,

w.dim A≥3(4)\mathrm{w.dim}\,A \ge3 \tag*{(4)}

Proof. Let

L=(f)∩(r),J=(f,r).L = (f) \cap(r), \qquad J = (f,r).

By assumption, LL is not flat. The standard exact sequence

0⟶L⟶(f)⊕(r)⟶J⟶0(5)0 \longrightarrow L \longrightarrow(f) \oplus(r) \longrightarrow J \longrightarrow0 \tag*{(5)}

has flat middle term, since (f)≅A≅(r)(f) \cong A \cong(r). Hence there is an AA-module NN such that Tor⁡1A(L,N)≠0\operatorname{Tor}^{A}_{1}(L,N) \ne0, and the long exact Tor sequence associated with (5) gives

Tor⁡2A(J,N)≅Tor⁡1A(L,N)≠0.\operatorname{Tor}^{A}_{2}(J,N) \cong\operatorname{Tor}^{A}_{1}(L,N) \ne0.

A second dimension shift, applied to

0⟶J⟶A⟶A/J⟶0,0 \longrightarrow J \longrightarrow A \longrightarrow A/J \longrightarrow0,

yields

Tor⁡3A(A/J,N)≅Tor⁡2A(J,N)≠0.\operatorname{Tor}^{A}_{3}(A/J,N) \cong\operatorname{Tor}^{A}_{2}(J,N) \ne0.

Therefore fd⁡A(A/J)≥3\operatorname{fd}_A(A/J) \ge3. □

Lemma 2.3 (Flat descent criterion). Let BB be a domain with A⊆BA \subseteq B, viewed inside Frac⁡(B)\operatorname{Frac}(B), and let EE be a non-zero ideal of AA. We also assume

(i) BB is a GCD domain,

(ii) Frac⁡(A)∩B=A\operatorname{Frac}(A) \cap B = A,

(iii) There exist s,y,z∈Bs,y,z \in B such that E⊆sBE \subseteq sB, sy,sz∈Esy,sz \in E and gcd⁡(y,z)=1\gcd(y,z)=1.

Then if EE is flat over AA, it follows that EE is invertible. In particular, EE is finitely generated.

Proof. Put u=syu=sy and w=szw=sz. Then u,w∈E⊆Au,w \in E \subseteq A. For a nonzero ideal II of AA, write

I−1=(A:I)={q∈Frac⁡(A):qI⊆A}.I^{-1}=(A:I)=\{q \in\operatorname{Frac}(A):qI \subseteq A\}.

We claim that

E−1=u−1A∩w−1A.(6)E^{-1}=u^{-1}A \cap w^{-1}A. \tag*{(6)}

The inclusion from left to right follows from u,w∈Eu,w \in E. Conversely, let qq belong to the right-hand side and set x=qs∈Frac⁡(B)x=qs \in\operatorname{Frac}(B). Then

xy=qu∈A⊆B,xz=qw∈A⊆B.xy=qu \in A \subseteq B,\qquad xz=qw \in A \subseteq B.

For relatively prime elements of a GCD domain one has yB∩zB=yzByB \cap zB=yzB. Hence

xyz∈yB∩zB=yzB,xyz \in yB \cap zB=yzB,

and cancellation gives x=qs∈Bx=qs \in B. If e∈Ee \in E, write e=sbe=sb with b∈Bb \in B. Then

qe=(qs)b∈B∩Frac⁡(A)=Aqe=(qs)b \in B \cap\operatorname{Frac}(A)=A

Thus q∈E−1q \in E^{-1}, proving (6).

Assume that EE is flat. Lemma 2.1, applied to P=u−1AP=u^{-1}A and Q=w−1AQ=w^{-1}A, gives

EE−1=E(u−1A)∩E(w−1A).EE^{-1}=E(u^{-1}A) \cap E(w^{-1}A).

Both ideals on the right contain 11, since 1=u⋅u−1=w⋅w−11=u\cdot u^{-1}=w\cdot w^{-1}. Therefore EE−1=AEE^{-1}=A, so EE is invertible. If

1=∑j=1mejqj,ej∈E,qj∈E−1,1=\sum_{j=1}^{m}e_jq_j,\qquad e_j \in E,\quad q_j \in E^{-1},

then every e∈Ee \in E satisfies e=∑jej(qje)e=\sum_j e_j(q_je) with qje∈Aq_je \in A. Hence E=(e1,…,em)E=(e_1,\ldots,e_m). □

We also record the following known consequence of a regular-element theorem of Rajabi, Torrecillas and Yassemi, and which will complete the proof of the final statement of our main theorem.

Proposition 2.4. For every commutative ring RR,

w.dim⁡R[[X]]=max⁡{w.dim⁡R+1,w.dim⁡R((X))},(7)\operatorname{w.dim} R[[X]]=\max\{\operatorname{w.dim} R+1,\operatorname{w.dim} R((X))\}, \tag*{(7)}

where R((X))=R[[X]][X−1]R((X))=R[[X]][X^{-1}].

Proof. Put S=R[[X]]S=R[[X]]. The element XX is a central nonzerodivisor of SS, S/XS≅RS/XS \cong R, and SX=R((X))S_X=R((X)). If w.dim⁡R<∞\operatorname{w.dim} R<\infty, the assertion is exactly [8 Theorem 3.7]. If w.dim⁡R=∞\operatorname{w.dim} R=\infty, the unconditional lower bound w.dim⁡R[[X]]≥w.dim⁡R+1\operatorname{w.dim} R[[X]] \ge\operatorname{w.dim} R+1 from the proof of [6 Lemma 1] gives w.dim⁡R[[X]]=∞\operatorname{w.dim} R[[X]]=\infty, and (7) follows as well. □

Almost maximal valuation rings

We will now fix a valuation ring VV of rank 11, with fraction field KK and associated valuation v:K×→Rv: K^{\times} \to\mathbb{R}. We write m\mathfrak{m} for the maximal ideal of VV.

Recall that a valued field KK is spherically complete if any descending chain of closed balls in KK has non-empty intersection (which implies that KK is complete). If K=Frac⁡(V)K = \operatorname{Frac}(V) is spherically complete then we say VV is a maximal valuation ring. Furthermore, we say that KK is almost spherically complete if for any R>0R > 0 and any descending chain of closed balls in KK of radius at least RR, the intersection of this chain is non-empty. Any spherically complete field KK is almost spherically complete, and if K=Frac⁡(V)K = \operatorname{Frac}(V) is almost spherically complete, then we say VV is an almost maximal valuation ring.

Example: Any discrete valuation ring is almost maximal, and the ring of Hahn series over any totally ordered abelian group is maximal.

The following elementary result will be essential when alternating between almost maximal valuation rings and their completions.

Lemma 2.5. For any valuation ring VV of rank 11, the completion V^\widehat{V} of VV is the valuation ring of K^\widehat{K}. Moreover, VV is almost maximal if and only if V^\widehat{V} is maximal.

Proof. Since the valuation vv extends to K^\widehat{K}, it is clear that V⊆OK^={x∈K^:v(x)≥0}V \subseteq\mathcal{O}_{\widehat{K}} = \{x \in\widehat{K}: v(x) \geq0\}. Moreover, suppose x∈OK^x \in\mathcal{O}_{\widehat{K}}, then xx is the limit of a sequence (xn)(x_n) with each xn∈Vx_n \in V, and hence for sufficiently nn, v(xn)=v(x)≥0v(x_n) = v(x) \geq0, thus x∈V^x \in\widehat{V} as required.

For the second statement, note that a closed ball in KK has the form y+Iαy + I_\alpha or y+Iα+y + I_\alpha^+, where Iα={x∈K:v(x)≥α}I_\alpha= \{x \in K: v(x) \geq\alpha\} and Iα+={x∈K:v(x)>α}I_\alpha^+ = \{x \in K: v(x) > \alpha\}. We can take the radius of the ball to be 2−α2^{-\alpha}. Closed balls in K^\widehat{K} have a similar form. But for any y∈K^y \in\widehat{K}, by density there exists y′∈Ky' \in K such that v(y−y′)>αv(y-y') > \alpha, so y+Iα=y′+Iαy + I_\alpha= y' + I_\alpha and y+Iα+=y′+Iα+y + I_\alpha^+ = y' + I_\alpha^+, so a closed ball in K^\widehat{K} contains a closed ball in KK with the same centre and radius.

Therefore, a descending sequence of closed balls with bounded radius has non-empty intersection in KK if and only if it has non-empty intersection in K^\widehat{K}, i.e. KK is almost spherically closed if and only if K^\widehat{K} is.

Moreover, a descending sequence of closed balls whose radii converge to 00 corresponds to a Cauchy sequence, so naturally it converges in K^\widehat{K} by completeness. Thus K^\widehat{K} is spherically complete if and only if KK is almost spherically complete as required. □

Almost maximal valuation rings were studied by Matlis in [7]. There, it was defined to be a valuation ring VV such that V/IV/I is linearly compact for all nonzero ideals II. Here, a module is linearly compact if every family of cosets of submodules with the finite intersection property has nonempty intersection. It is well known that finite direct sums and quotients of linearly compact modules are linearly compact.

It is immediately verified that these definitions are equivalent, but the formulation in terms of linearly compact modules will be essential in the proof of the following technical results (analogous to [9 Lemma 2.1 and Lemma 2.2]), and in the proof of our main theorem.

Lemma 2.6. Let F=VnF = V^n be a finite free VV-module, and let Z1⊇Z2⊇⋯Z_1 \supseteq Z_2 \supseteq\cdots be nonempty cosets of submodules LiL_i of FF. If cF⊆LicF \subseteq L_i for every ii, for some 0≠c∈V0 \ne c \in V, then ⋂iZi≠∅\bigcap_i Z_i \ne\varnothing.

Proof. The images of the ZiZ_i in the linearly compact module F/cF≅(V/cV)nF/cF \cong(V/cV)^n have nonempty intersection. Every lift of a point in this intersection belongs to all the ZiZ_i, since cF⊆LicF \subseteq L_i. □\square

Proposition 2.7. If v(K×)=Rv(K^\times) = \mathbb{R}, then the intersection of any nonempty family of finitely generated VV-submodules of a finite free VV-module is finite free.

Proof. We adapt the rank induction of [3 Lemma 6.14], using Matlis’s theorem [7 Theorem 4], which states that the module K/VK/V is injective.

The module KK is also injective: a homomorphism from a nonzero ideal of VV to KK is multiplication by an element of KK, and hence extends to VV. Thus applying the functor Hom⁡V(V/m,−)\operatorname{Hom}_V(V/\mathfrak{m}, -) to the sequence 0→V→K→K/V→00 \to V \to K \to K/V \to0 and taking the associated long-Ext exact sequence we see that

Ext⁡V2(V/m,V)=0\operatorname{Ext}^{2}_{V}(V/\mathfrak{m}, V) = 0

and applying dimension shifting gives

Ext⁡V1(m,V)≅Ext⁡V2(V/m,V)=0.\operatorname{Ext}^{1}_{V}(\mathfrak{m}, V) \cong\operatorname{Ext}^{2}_{V}(V/\mathfrak{m}, V) = 0.

For a submodule N⊆KdN \subseteq K^d, identify Hom⁡V(m,N)\operatorname{Hom}_V(\mathfrak{m}, N) with

N#={z∈Kd:mz⊆N}.N^\# = \{z \in K^d : \mathfrak{m}z \subseteq N\}.

Indeed, a homomorphism extends upon localization to a KK-linear map K→KdK \to K^d, so is multiplication by a unique vector. We claim that a bounded submodule N⊆KdN \subseteq K^d satisfying N=N#N = N^\# is finite free. Here bounded means N⊆aVdN \subseteq aV^d for some a∈K×a \in K^\times.

For d=1d = 1, a nonzero bounded submodule has a real lower bound γ=inf⁡v(N∖{0})\gamma= \inf v(N \setminus\{0\}) and is either {z:v(z)≥γ}\{z : v(z) \ge\gamma\} or {z:v(z)>γ}\{z : v(z) > \gamma\}, with zero included. The second possibility contradicts N=N#N = N^\#; the first is principal because v(K×)=Rv(K^\times) = \mathbb{R}. The zero module causes no difficulty.

For the induction step, let N′=N∩Kd−1N' = N \cap K^{d-1} and let N′′⊆KN'' \subseteq K be the image of NN under the last-coordinate projection. Both are bounded, and (N′)#=N′(N')^\# = N'. Induction makes N′N' finite free, so Ext⁡V1(m,N′)=0\operatorname{Ext}^{1}_{V}(\mathfrak{m}, N') = 0. Apply Hom⁡V(m,−)\operatorname{Hom}_V(\mathfrak{m}, -) to 0→N′→N→N′′→00 \to N' \to N \to N'' \to0. The natural multiplication maps identify the first two modules with their Hom modules; exactness then gives (N′′)#=N′′(N'')^\# = N''. The rank-one case makes N′′N'' finite free. The sequence splits, proving the claim.

Now let L=⋂iLi⊆VdL = \bigcap_i L_i \subseteq V^d, with each LiL_i finitely generated. It follows from [2 Lemma 3.1.1] that LiL_i is finite free, so since the value group is R\mathbb{R}, one has V#=VV^\# = V: if v(z)<0v(z) < 0, choose t∈mt \in\mathfrak{m} with 0<v(t)<−v(z)0 < v(t) < -v(z). Using a basis of LiL_i in its KK-span gives Li#=LiL_i^\# = L_i.

::: Consequently

L♯=⋂iLi♯=⋂iLi=L.\mathcal{L}^{\sharp}=\bigcap_i\mathcal{L}_i^{\sharp}=\bigcap_i\mathcal{L}_i=\mathcal{L}.

The claim applies to the bounded module L\mathcal{L}. □

Coherence over an almost maximal valuation domain

In this section, we will answer Jøndrup and Smalls question from [6], and once and for all determine when a power sereis ring over a valuation ring is coherent. Again, fix an almost maximal valuation ring VV of rank 1, with associated valuation v:K×→Rv:K^{\times}\to\mathbb{R}, and we will assume that v(K×)=Rv(K^{\times})=\mathbb{R}.

Let A:=V[[X]]A:=V[[X]]. We will adapt the kernel-lifting argument of [9 Section 3] to prove that AA is a coherent ring.

For each r∈Nr\in\mathbb{N}, let Ar=A/XrAA_r=A/X^rA. As a VV-module, Ar≅VrA_r\cong V^r is finite free. For a finite matrix Φ:Am→An\Phi:A^m\to A^n, write

K=ker⁡Φ,Kr=ker⁡(Φr:Arm→Arn),\mathcal{K}=\ker\Phi,\qquad\mathcal{K}_r=\ker(\Phi_r:A_r^m\to A_r^n),

and let ρs,r:Ks→Kr\rho_{s,r}:\mathcal{K}_s\to\mathcal{K}_r denote reduction, for s≥rs\ge r. Each Kr\mathcal{K}_r is finite free over VV. Completeness of AA gives

K≃lim←⁡rKr.(8)\mathcal{K}\simeq\varprojlim_r\mathcal{K}_r. \tag*{(8)}

Set

Er=⋂s≥rim⁡ρs,r.E_r=\bigcap_{s\ge r}\operatorname{im}\rho_{s,r}.

An element of ErE_r lifts to every finite level separately; the next proposition supplies compatible lifts.

The ideal (X)(X) of AA is prime. Every nonzero element of D=A(X)D=A_{(X)} is a power of XX times a unit, so DD is a DVR with uniformizer XX and residue field KK. Moreover

Dr=D/XrD≃K[X]/(Xr)≃Ar⊗VK.D_r=D/X^rD\simeq K[X]/(X^r)\simeq A_r\otimes_VK.

Localization is flat, and hence

KD:=ker⁡(ΦD:Dm→Dn)=K⊗AD.(9)\mathcal{K}_D:=\ker(\Phi_D:D^m\to D^n)=\mathcal{K}\otimes_A D. \tag*{(9)}

As a submodule of a finite free module over a DVR, KD\mathcal{K}_D is finite free. It can therefore be generated over DD by finitely many vectors coming from K\mathcal{K}; this does not presuppose finite generation of K\mathcal{K}.

Proposition 3.1. Every map Er+1→ErE_{r+1}\to E_r is surjective. In particular,

im⁡(K→K1)=⋂r≥1im⁡(Kr→K1).\operatorname{im}(\mathcal{K}\to\mathcal{K}_1)=\bigcap_{r\ge1}\operatorname{im}(\mathcal{K}_r\to\mathcal{K}_1).

Proof. Fix xr∈Erx_r\in E_r. For s≥r+1s\ge r+1, put

Ys=im⁡ρs,r+1,Zs={z∈Ys:ρr+1,r(z)=xr},Nr=ker⁡ρr+1,r.Y_s=\operatorname{im}\rho_{s,r+1},\qquad Z_s=\{z\in Y_s:\rho_{r+1,r}(z)=x_r\},\qquad N_r=\ker\rho_{r+1,r}.

The ZsZ_s form a descending sequence of nonempty cosets with directions Ls=Ys∩NrL_s=Y_s\cap N_r (i.e. Zs=z+LsZ_s=z+L_s for any z∈Zsz\in Z_s). The modules YsY_s and NrN_r are finitely generated VV-submodules of Ar+1mA^{m}_{r+1}, so their intersections LsL_s are finite free by coherence of VV. Also NrN_r lies in the coefficient layer XrAm/Xr+1Am≃VmX^r A^m/X^{r+1}A^m \simeq V^m.

Tensoring by KK is exact, so it preserves these kernels, images and intersections. Choose Smith normal form for ΦD\Phi_D, with nonzero diagonal entries Xa1,…,XakX^{a_1},\ldots,X^{a_k}. Let a=max⁡iaia=\max_i a_i when k>0k>0, and a=0a=0 otherwise. In the corresponding source basis e1,…,eme_1,\ldots,e_m, for s≥max⁡{1,a}s\geq\max\{1,a\},

ker⁡ΦDs=⨁i=1kXs−aiDsei⊕⨁i=k+1mDsei.\ker\Phi_{D_s}=\bigoplus_{i=1}^{k}X^{s-a_i}D_s e_i\oplus\bigoplus_{i=k+1}^{m}D_s e_i.

Thus, for s≥r+1+as\geq r+1+a, the image of this kernel modulo Xr+1X^{r+1} equals the image of KD\mathcal{K}_D. Write

q:Dm⟶Dr+1m,H=XrDm/Xr+1Dm.q:D^m\longrightarrow D^m_{r+1},\qquad H=X^rD^m/X^{r+1}D^m.

For these ss, flat base change consequently gives

Ls⊗VK=q(KD)∩H.L_s\otimes_V K=q(\mathcal{K}_D)\cap H.

Since q−1(H)=XrDmq^{-1}(H)=X^rD^m and KD\mathcal{K}_D is XX-saturated,

q(KD)∩H=q(KD∩XrDm)=q(XrKD)=:W.(10)q(\mathcal{K}_D)\cap H=q(\mathcal{K}_D\cap X^rD^m)=q(X^r\mathcal{K}_D)=:W. \tag*{(10)}

Choose u1,…,ut∈Ku_1,\ldots,u_t\in\mathcal{K} generating KD\mathcal{K}_D over DD, using (9). Let L0⊆NrL_0\subseteq N_r be the VV-span of the classes of XrujX^r u_j. These are reductions of actual kernel vectors at every level, and (10) gives

L0⊆Lsfor all s,L0⊗VK=Ls⊗VK=W(s≥r+1+a).L_0\subseteq L_s\quad\text{for all }s,\qquad L_0\otimes_V K=L_s\otimes_V K=W\quad(s\geq r+1+a).

Choose s0≥r+1+as_0\geq r+1+a. The finite modules L0L_0 and Ls0L_{s_0} have the same KK-span. Clearing denominators for generators of Ls0L_{s_0} gives 0≠c∈V0\ne c\in V such that

cLs0⊆L0⊆Ls⊆Ls0(s≥s0).cL_{s_0}\subseteq L_0\subseteq L_s\subseteq L_{s_0}\qquad(s\geq s_0).

Choose z0∈Zs0z_0\in Z_{s_0}. The translated sets Zs−z0Z_s-z_0, for s≥s0s\geq s_0, are nested nonempty cosets in the finite free module Ls0L_{s_0}, with directions containing cLs0cL_{s_0}. Applying Lemma 2.6, there exists a point in their intersection. Translating back gives an element of Er+1E_{r+1} mapping to xrx_r.

Starting from any x1∈E1x_1\in E_1, recursively choose compatible xr∈Erx_r\in E_r. (8) produces the required global kernel vector. The reverse inclusion in the asserted equality follows by reduction of a global vector. □

Theorem 3.2. AA is a coherent domain.

Proof. For each rr, the module Λr=im⁡(Kr→K1)\Lambda_r=\operatorname{im}(\mathcal{K}_r\to\mathcal{K}_1) is a finitely generated VV-submodule of VmV^m. Proposition 3.1 and Proposition 2.7 show that

im⁡(K→K1)=⋂rΛr\operatorname{im}(\mathcal{K}\to\mathcal{K}_1)=\bigcap_r\Lambda_r

is finite free. The kernel K\mathcal{K} is XX-saturated in AmA^m, so

K/XK≃im⁡(K→K1).\mathcal{K}/X\mathcal{K}\simeq\operatorname{im}(\mathcal{K}\to\mathcal{K}_1).

It is closed in the complete module AmA^m, and saturation gives K∩XrAm=XrK\mathcal{K}\cap X^rA^m=X^r\mathcal{K}. Hence K\mathcal{K} is complete and separated for its XX-adic topology. Lifts of finitely many generators

of K/XKK/XK generate KK: successive subtraction of their linear combinations and division by XX expresses any vector using convergent power-series coefficients. Thus every finite matrix over AA has finitely generated kernel, and AA is coherent. □

Corollary 3.3. Let VV be any valuation ring. Then the power series ring V[[X]]V[[X]] is coherent if and only if

  • VV has rank 1,

  • VV is almost maximal, and

  • VV is discrete, or has value group R\mathbb{R}.

Proof. Firstly, if VV has rank 1, then V[[X]]V[[X]] is Noetherian, and hence coherent, if VV is discrete. On the other hand, if VV is almost maximal with value group R\mathbb{R}, then V[[X]]V[[X]] is coherent by Theorem 3.2.

For the converse, if VV has rank greater than 1, then we know that V[[X]]V[[X]] is not coherent by [6 Section 1]. So we may assume that VV has rank 1, and let v:K×→Rv : K^\times\to\mathbb{R} be the associated valuation.

Since the value group Γ=v(K×)\Gamma= v(K^\times) is a subgroup of R\mathbb{R}, either Γ\Gamma is discrete, or Γ\Gamma is dense in R\mathbb{R}. If Γ≠R\Gamma\ne\mathbb{R} and Γ\Gamma is dense, it follows from [1 Section 3 Corollary] and [5 Theorem 3.1] that V[[X]]V[[X]] is not coherent.

Finally, if Γ=R\Gamma= \mathbb{R} but VV is not almost maximal, then the proof in [9 Section 4 and 5] shows that V[[X]]V[[X]] is not coherent.

Note: The proof in [9] assumed that VV was complete, but the only place in section 4 or 5 that uses completeness is the proof of [9 Construction Lemma 4.2], where it was required to assert that a descending chain of balls with empty intersection has bounded radius (as noted in [9 Remark 4.3]). Since assuming VV is not almost maximal implies the existence of such a chain immediately, the proof goes through without any changes. □

Cut ideals and flatness

We now turn our attention to Hilbert’s syzygy. Again, fix a valuation domain VV of rank 1, with fraction field KK, valuation v:K×→Rv : K^\times\to\mathbb{R}, and let Γ=v(K×)⊆R\Gamma= v(K^\times) \subseteq\mathbb{R} be the value group.

In this section, we will assume that Γ\Gamma is a proper, dense subgroup of R\mathbb{R}. Set A:=V[[X]]A := V[[X]], and consider the valuation ν\nu on AA given by (2) (and also denote by ν\nu its extension to Frac⁡(A)\operatorname{Frac}(A)).

For s≥0s \ge0, set

Is={a∈A:ν(a)≥s}.I_s = \{a \in A : \nu(a) \ge s\}.

Since ν(a)≥0\nu(a) \ge0 for a∈Aa \in A, each IsI_s is an ideal of AA. Moreover, if s∈Γs \in\Gamma then s=v(a)s = v(a) for some a∈Va \in V, so Is=(a)I_s = (a) is a principal ideal, and hence flat. Fixing c>0c > 0 with c∉Γc \notin\Gamma, we will first prove that IcI_c is not a flat ideal of AA.

Choose ρ∈Γ\rho\in\Gamma with ρ>c\rho> c, and set

d=ρ−c.d = \rho- c.

Then

c,d>0,c,d∉Γ,c+d=ρ.(11)c,d>0,\qquad c,d\notin\Gamma,\qquad c+d=\rho. \tag*{(11)}

Choose r∈Vr\in V with v(r)=ρv(r)=\rho. By density, there are sequences (cn)(c_n) and (dn)(d_n) in Γ\Gamma such that

cn>c,cn⟶c,dn>d,dn⟶d.c_n>c,\qquad c_n\longrightarrow c,\qquad d_n>d,\qquad d_n\longrightarrow d.

Choose un,wn∈Vu_n,w_n\in V with v(un)=cnv(u_n)=c_n and v(wn)=dnv(w_n)=d_n, and set

h=∑n≥0unXn,f=∑n≥0wnXn.(12)h=\sum_{n\geq0}u_nX^n,\qquad f=\sum_{n\geq0}w_nX^n. \tag*{(12)}

Then

ν(h)=c,ν(f)=d.(13)\nu(h)=c,\qquad\nu(f)=d. \tag*{(13)}

Moreover,

Iρ=rA.(14)I_\rho=rA. \tag*{(14)}

Indeed, ν(g)≥ρ=v(r)\nu(g)\geq\rho=v(r) if and only if every coefficient of gg is divisible by rr in VV.

Lemma 4.1. With the notation above,

IcId⊆Iρ.I_cI_d\subseteq I_\rho.

Proof. Multiplicativity of ν\nu gives IcId⊆IρI_cI_d\subseteq I_\rho. If a=∑anXn∈Ica=\sum a_nX^n\in I_c, every nonzero coefficient ana_n has value strictly greater than cc, because v(an)∈Γv(a_n)\in\Gamma while c∉Γc\notin\Gamma. The analogous assertion holds for elements of IdI_d. Hence the constant coefficient of every finite sum of products of an element of IcI_c and an element of IdI_d is either zero or has value strictly greater than c+d=ρc+d=\rho. The constant series rr lies in IρI_\rho and has constant coefficient of value exactly ρ\rho. Therefore r∉IcIdr\notin I_cI_d. □

Proposition 4.2. The ideal IcI_c is not flat over AA.

Proof. Put

q=hr∈Frac⁡(A).q=\frac{h}{r}\in\operatorname{Frac}(A).

By (11) and (13),

ν(q)=c−ρ=−d.\nu(q)=c-\rho=-d.

We first compute

A∩qA=qId.(15)A\cap qA=qI_d. \tag*{(15)}

If qa∈Aqa\in A, then

0≤ν(qa)=ν(a)−d,0\leq\nu(qa)=\nu(a)-d,

so a∈Ida\in I_d. Conversely, if a∈Ida\in I_d, then ν(ha)≥c+d=ρ\nu(ha)\geq c+d=\rho. Every coefficient of haha is divisible by rr, and therefore qa=ha/r∈Aqa=ha/r\in A. This proves (15).

We also have

Ic∩qIc=qIρ.(16)I_c\cap qI_c=qI_\rho. \tag*{(16)}

Indeed, if z=qa∈Icz=qa\in I_c with a∈Ica\in I_c, then

ν(a)=ν(z)+d≥c+d=ρ.\nu(a)=\nu(z)+d\geq c+d=\rho.

so a∈Iρa \in I_{\rho}. Conversely, if a∈Iρa \in I_{\rho}, then a∈Ida \in I_d, whence qa∈Aqa \in A by (15); moreover, a∈Ica \in I_c and ν(qa)≥ρ−d=c\nu(qa) \ge\rho- d = c. Thus qa∈Ic∩qIcqa \in I_c \cap qI_c, proving (16).

Suppose that IcI_c were flat. Lemma 2.1, applied to E=IcE = I_c, P=AP = A and Q=qAQ = qA, would give

Ic(A∩qA)=Ic∩qIc.I_c(A \cap qA) = I_c \cap qI_c.

Using (15) and (16), we obtain

qIcId=qIρ,qI_cI_d = qI_{\rho},

and hence IcId=IρI_cI_d = I_{\rho}, contrary to Lemma 4.1. □

Lemma 4.3. For the elements ff and rr chosen above,

(f)∩(r)=fIc.(17)(f) \cap(r) = fI_c. \tag*{(17)}

In particular, (f)∩(r)(f) \cap(r) is not flat over AA.

Proof. Let fg=rafg = ra with g,a∈Ag, a \in A. By multiplicativity of ν\nu,

d+ν(g)=ρ+ν(a)≥ρ,d + \nu(g) = \rho+ \nu(a) \ge\rho,

so ν(g)≥ρ−d=c\nu(g) \ge\rho- d = c and g∈Icg \in I_c. Thus (f)∩(r)⊆fIc(f) \cap(r) \subseteq fI_c.

Conversely, if g∈Icg \in I_c, then

ν(fg)≥d+c=ρ.\nu(fg) \ge d + c = \rho.

Every coefficient of fgfg is divisible by rr, and hence fg∈(r)fg \in(r). This proves (17). Since AA is a domain and f≠0f \ne0, multiplication by ff identifies IcI_c with fIcfI_c. Proposition 4.2 gives the final assertion. □

Theorem 4.4. If Γ=v(K×)\Gamma= v(K^{\times}) is a proper dense subgroup of R\mathbb{R}, then w. dim V[[X]]>2V[[X]] > 2.

Proof. Using Lemma 4.3 we know that there exist f,r∈V[[X]]f, r \in V[[X]] such that (f)∩(r)(f) \cap(r) is not flat as a V[[X]]V[[X]]-module, and hence w. dim V[[X]]≥3V[[X]] \ge3 by Theorem 2.2. □

The full real value group

We now return to valued fields with full real value group. The cut argument of Sections 2–3 is unavailable in this case, since there is no real cut outside the value group.

Let KK be a complete valued field with v(K×)=Rv(K^{\times}) = \mathbb{R}, let V=OKV = \mathcal{O}_K, and put A=V[[X]]A = V[[X]]. Assume that KK is not spherically complete. Following [9], fix a maximal immediate extension K~/K\widetilde{K}/K, set V~=OK~\widetilde{V} = \mathcal{O}_{\widetilde{K}}, and put

B=V~[[X]].B = \widetilde{V}[[X]].

The field K~\widetilde{K} is spherically complete and v(K~×)=Rv(\widetilde{K}^{\times}) = \mathbb{R}. Thus BB is a GCD domain by [9 Theorem 1.1(i)].

We record the precise consequences of the construction in [9] that are needed below, replacing the variable TT by XX.

Proposition 5.1. [9 Section 4 and 5] There are nonzero elements ai∈mVa_i \in\mathfrak{m}_V, real numbers αi=v(ai)>0\alpha_i = v(a_i) > 0, and a nonzero c∈mVc \in\mathfrak{m}_V such that

α1>α2>⋯ ,αi⟶0,v(c)=∑i≥1αi.\alpha_1 > \alpha_2 > \cdots,\qquad\alpha_i \longrightarrow0,\qquad v(c) = \sum_{i\geq1} \alpha_i.

Set

ξi=ai−X,Pn=∏i=1nξi.\xi_i = a_i-X,\qquad P_n=\prod_{i=1}^{n}\xi_i.

There are F∈BF \in B and nonzero f,g∈Af,g \in A with the following properties:

(i) F(0)=cF(0)=c and Pn∣FP_n \mid F in BB for every nn;

(ii) Pn∣f,gP_n \mid f,g in AA for every nn, and FF is a greatest common divisor of ff and gg in BB;

(iii) the ideal

E=(f)A∩(g)AE=(f)A\cap(g)A

is not finitely generated over AA;

(iv) there is a strictly increasing sequence of positive integers (Ni)(N_i) and an element c1∈K×c_1\in K^\times such that every NiN_i is prime to the residue characteristic when it is positive and

f(X)=c1∏i≥1(1−(Xai)Ni)(18)f(X)=c_1\prod_{i\geq1}\left(1-\left(\frac{X}{a_i}\right)^{N_i}\right) \tag*{(18)}

in K[[X]]K[[X]].

Proof. The valuation data are given by [9 Construction Lemma 4.2]. The remaining assertions follow from [9 Propositions 5.1–5.3 and the proof of Theorem 4.1(i)]; the normalized product (18) appears in the proof of [9 Proposition 5.2]. □

We shall use evaluation at elements of positive value.

Lemma 5.2. Let CC be the valuation ring of a complete nonarchimedean valued field, and let 0≠a∈mC0\ne a\in\mathfrak{m}_C. Evaluation at aa defines a surjective homomorphism

ev⁡a:C[[X]]⟶C\operatorname{ev}_a:C[[X]]\longrightarrow C

whose kernel is (X−a)(X-a). Moreover, if H∈XNC[[X]]H\in X^N C[[X]], then

v(H(a))≥Nv(a).(19)v\bigl(H(a)\bigr)\geq Nv(a). \tag*{(19)}

In particular, evaluation is continuous from the XX-adic topology to the valuation topology, and X−aX-a is prime in C[[X]]C[[X]].

Proof. For H(X)=∑n≥0hnXnH(X)=\sum_{n\geq0}h_nX^n, the series H(a)=∑n≥0hnanH(a)=\sum_{n\geq0}h_na^n converges in CC. Passage to the limit from polynomial truncations shows that evaluation is a homomorphism. The estimate (19) is immediate. The coefficient of XrX^r in

Q(X)=∑n≥1hn∑j=0n−1Xn−1−jajQ(X)=\sum_{n\geq1}h_n\sum_{j=0}^{n-1}X^{n-1-j}a^j

is a convergent series in CC, so Q∈C[[X]]Q\in C[[X]]. The identity

H(X)−H(a)=(X−a)Q(X)H(X)-H(a)=(X-a)Q(X)

shows that the kernel of evaluation is (X−a)(X-a). Surjectivity follows from the constant series. The quotient is isomorphic to the domain CC, and hence X−aX-a is prime. □

For n≥1n \ge1, define

Yn=FPn∈B.(20)Y_n=\frac{F}{P_n}\in B. \tag*{(20)}

Then

Yn=ξn+1Yn+1(21)Y_n=\xi_{n+1}Y_{n+1} \tag*{(21)}

and, since Pn(0)=∏i=1naiP_n(0)=\prod_{i=1}^{n}a_i,

v(Yn(0))=v(c)−∑i=1nαi=∑i>nαi⟶0.(22)v\left(Y_n(0)\right)=v(c)-\sum_{i=1}^{n}\alpha_i=\sum_{i>n}\alpha_i\longrightarrow0. \tag*{(22)}

Lemma 5.3. For every n≥1n \ge1,

ξn2∤F\xi_n^2\nmid F in BB. Consequently,

Ym(an)≠0(m≥n).(23)Y_m(a_n)\ne0\qquad(m\ge n). \tag*{(23)}

Proof. For m≥nm\ge n, let

fm=c1∏i=1m(1−(Xai)Ni)∈A.f_m=c_1\prod_{i=1}^{m}\left(1-\left(\frac{X}{a_i}\right)^{N_i}\right)\in A.

These are the partial products used in the construction [9 Proposition 5.2]; they converge XX-adically to ff, and, for each fixed nn, one has ξn∣fm\xi_n\mid f_m for all sufficiently large mm. Since ξn mod X=an≠0\xi_n\bmod X=a_n\ne0, division by ξn\xi_n on (ξn)(\xi_n) is XX-adically continuous by [9 Lemma 2.3(i)]. Hence fm/ξnf_m/\xi_n converges XX-adically to f/ξnf/\xi_n. Applying the continuous evaluation map of Lemma 5.2 and using

1−(X/an)Nnan−X=1anNn∑r=0Nn−1anNn−1−rXr,\frac{1-(X/a_n)^{N_n}}{a_n-X} = \frac{1}{a_n^{N_n}}\sum_{r=0}^{N_n-1}a_n^{N_n-1-r}X^r,

we obtain

(fξn)(an)=c1Nnan∏j≥1j≠n(1−(anaj)Nj).(24)\left(\frac{f}{\xi_n}\right)(a_n) = \frac{c_1N_n}{a_n}\prod_{\substack{j\ge1\\j\ne n}}\left(1-\left(\frac{a_n}{a_j}\right)^{N_j}\right). \tag*{(24)}

This element is nonzero. The finitely many factors with j<nj<n are nonzero because v(an/aj)=αn−αj<0v(a_n/a_j)=\alpha_n-\alpha_j<0. For j>nj>n,

v((anaj)Nj)=Nj(αn−αj)⟶+∞,v\left(\left(\frac{a_n}{a_j}\right)^{N_j}\right)=N_j(\alpha_n-\alpha_j)\longrightarrow+\infty,

so the tail product and its inverse converge. By construction, NnN_n is a unit of VV; this is automatic in residue characteristic zero and follows from the prime-to-residue-characteristic condition in positive residue characteristic. Thus the right-hand side of (24) is nonzero. Hence ξn2∤f\xi_n^2\nmid f, and therefore ξn2∤F\xi_n^2\nmid F because F∣fF\mid f in BB.

Let m≥nm\ge n. If Ym(an)=0Y_m(a_n)=0, Lemma 5.2 gives ξn∣Ym\xi_n\mid Y_m. Since PmP_m contains the factor ξn\xi_n, the equality F=PmYmF=P_mY_m would imply ξn2∣F\xi_n^2\mid F, a contradiction. □

The next construction produces the coprime quotients required by Lemma 2.3.

Proposition 5.4. There is an element Z∈BZ \in B such that

gcd⁡B(Y1,Z)=1(25)\gcd_{B}(Y_{1},Z)=1 \tag*{(25)}

and, with

s=fgF∈B,s=\frac{fg}{F}\in B,

one has

sY1,sZ∈E.(26)sY_{1},sZ\in E. \tag*{(26)}

Proof. Choose integers

0=k1<k2<k3<⋯0=k_{1}<k_{2}<k_{3}<\cdots

recursively so that, whenever 1≤n<m1\le n<m,

(km−kn)αn+v(Ym(an))>v(Yn(an))+m.(27)(k_{m}-k_{n})\alpha_{n}+v\left(Y_{m}(a_{n})\right)>v\left(Y_{n}(a_{n})\right)+m. \tag*{(27)}

At the mm-th stage there are only finitely many inequalities, and αn>0\alpha_{n}>0, so such an integer kmk_{m} exists.

For n≥1n\ge1, put

Zn=∑m≥nXkm−knYm∈B.(28)Z_{n}=\sum_{m\ge n}X^{k_{m}-k_{n}}Y_{m}\in B. \tag*{(28)}

The series converges XX-adically and satisfies

Zn=Yn+Xkn+1−knZn+1.(29)Z_{n}=Y_{n}+X^{k_{n+1}-k_{n}}Z_{n+1}. \tag*{(29)}

By Lemma 5.2 evaluation at ana_{n} is continuous for the XX-adic topology. Hence

Zn(an)=∑m≥nankm−knYm(an).Z_{n}(a_{n})=\sum_{m\ge n}a_{n}^{k_{m}-k_{n}}Y_{m}(a_{n}).

The first term is nonzero by (23), while (27) gives, for m>nm>n,

v(ankm−knYm(an))>v(Yn(an))+m.v\left(a_{n}^{k_{m}-k_{n}}Y_{m}(a_{n})\right)>v\left(Y_{n}(a_{n})\right)+m.

Thus the sum of the later terms has valuation strictly larger than v(Yn(an))v(Y_{n}(a_{n})), and the first term cannot be cancelled. Therefore

Zn(an)≠0,ξn∤Zn.(30)Z_{n}(a_{n})\ne0,\qquad\xi_{n}\nmid Z_{n}. \tag*{(30)}

We prove that Y1Y_{1} and Z1Z_{1} are relatively prime. Let d∈Bd\in B be a common divisor. Since Y1(0)≠0Y_{1}(0)\ne0, one has X∤dX\nmid d. The element XX is prime because B/XB≅V~B/XB\cong\widetilde{V} is a domain, so gcd⁡B(d,X)=1\gcd_{B}(d,X)=1. We claim inductively that

d∣Ynandd∣Zn(n≥1).(31)d\mid Y_{n}\qquad\text{and}\qquad d\mid Z_{n}\qquad(n\ge1). \tag*{(31)}

Suppose this holds for nn. Equation (29) gives

Xkn+1−knZn+1∈dB∩Xkn+1−knB.X^{k_{n+1}-k_{n}}Z_{n+1}\in dB\cap X^{k_{n+1}-k_{n}}B.

:::%

Since dd is relatively prime to Xkn+1−knX^{k_{n+1}-k_n} and relatively prime elements a,ba,b of a GCD domain satisfy aB∩bB=abBaB \cap bB = abB, cancellation gives d∣Zn+1d \mid Z_{n+1}. By (30), ξn+1∤Zn+1\xi_{n+1} \nmid Z_{n+1}, and hence ξn+1∤d\xi_{n+1} \nmid d. Since ξn+1\xi_{n+1} is prime by Lemma 5.2, dd and ξn+1\xi_{n+1} are relatively prime. Now

Yn=ξn+1Yn+1∈dB∩ξn+1B=dξn+1B,Y_n = \xi_{n+1}Y_{n+1} \in dB \cap\xi_{n+1}B = d\xi_{n+1}B,

and cancellation gives d∣Yn+1d \mid Y_{n+1}. This proves (31).

Write Yn=denY_n = de_n with en∈Be_n \in B. Since Yn(0)≠0Y_n(0) \ne0, both d(0)d(0) and en(0)e_n(0) are nonzero. Taking constant terms gives

0≤v(d(0))≤v(Yn(0))(n≥1).0 \le v\bigl(d(0)\bigr) \le v\bigl(Y_n(0)\bigr) \qquad(n \ge1).

By (22), v(d(0))=0v(d(0)) = 0. Hence d(0)d(0) is a unit of V~\widetilde{V}, so dd is a unit of BB. This proves (25) with Z=Z1Z = Z_1.

It remains to prove (26). Since PmP_m divides both ff and gg in AA,

sY1=fgP1∈E.sY_1 = \frac{fg}{P_1} \in E.

Moreover,

sZ=∑m≥1XkmfgPm=f(∑m≥1XkmgPm)=g(∑m≥1XkmfPm).\begin{aligned} sZ = \sum_{m \ge1} X^{k_m}\frac{fg}{P_m} \\ &= f\left(\sum_{m \ge1} X^{k_m}\frac{g}{P_m}\right) = g\left(\sum_{m \ge1} X^{k_m}\frac{f}{P_m}\right). \end{aligned}

The two series in parentheses converge XX-adically in AA. Consequently, sZ∈(f)A∩(g)A=EsZ \in(f)A \cap(g)A = E.

Theorem 5.5. Let KK be a complete nonarchimedean valued field with v(K×)=Rv(K^\times) = \mathbb{R}, let V=OKV = \mathcal{O}_K, and assume that KK is not spherically complete. Set A=V[[X]]A = V[[X]]. Then there are nonzero f,g∈Af,g \in A such that E:=(f)∩(g)E := (f) \cap(g) is not flat over AA. Moreover,

fd⁡A(A/(f,g))≥3,w.dim⁡A≥3.\operatorname{fd}_A\bigl(A/(f,g)\bigr) \ge3,\qquad\operatorname{w.dim} A \ge3.

Proof. We take f,g∈Af,g \in A to be the elements in Proposition 5.1, so that E=(f)∩(g)E = (f) \cap(g) is not finitely generated over AA. We will construct elements y,z,s∈By,z,s \in B such that the data A,B,E,s,y,zA,B,E,s,y,z satisfies the conditions of the Flat Descent criterion (Lemma 2.3), and it will follow that EE is not flat.

Write f=Ff0f = Ff_0 and g=Fg0g = Fg_0 in BB. Since FF is a greatest common divisor, f0f_0 and g0g_0 are relatively prime in BB, so f0B∩g0B=f0g0Bf_0B \cap g_0B = f_0g_0B. So writing s:=Ff0g0=fgFs := Ff_0g_0 = \frac{fg}{F}, we see that

(f)B∩(g)B=F(f0B∩g0B)=Ff0g0BsB.(f)B \cap(g)B = F(f_0B \cap g_0B) = Ff_0g_0BsB.

It follows that E⊆sBE \subseteq sB.

The natural embedding Frac⁡(A)⊆K((X))\operatorname{Frac}(A) \subseteq K((X)) gives

Frac⁡(A)∩B⊆K((X))∩V~[[X]].\operatorname{Frac}(A) \cap B \subseteq K((X)) \cap\widetilde{V}[[X]].

A Laurent series in the last intersection has no negative powers and all its coefficients lie in K∩V^=VK \cap\widehat{V} = V. Thus Frac⁡(A)∩B⊆A\operatorname{Frac}(A) \cap B \subseteq A, and we see immediately that

Frac⁡(A)∩B=A.(32)\operatorname{Frac}(A) \cap B = A. \tag*{(32)}

Let y:=Y1=FP1y := Y_1 = \frac{F}{P_1}, and take z∈Bz \in B to be the element ZZ given by Proposition 5.4, and it follows that this data satisfies the conditions of Lemma 2.3 as required. □

Completion invariance

In this section, we will complete the proof of our main theorem. In light of Theorem 5.5, the only remaining case to consider is when VV is non-complete, with value group R\mathbb{R}. The valuation on K^\widehat{K} is again denoted by vv.

Again, let K^\widehat{K} be the metric completion of K=Frac⁡(V)K = \operatorname{Frac}(V) and V^=OK^\widehat{V} = \mathcal{O}_{\widehat{K}} for its valuation ring. It follows from Lemma 2.5 that K^\widehat{K} is spherically complete if and only if VV is almost maximal.

Lemma 6.1. The extension K^/K\widehat{K}/K is immediate. Moreover, for every 0≠a∈V0 \ne a \in V, the natural map

V/aV⟶V^/aV^(33)V/aV \longrightarrow\widehat{V}/a\widehat{V} \tag*{(33)}

is an isomorphism.

Proof. Let 0≠x∈K^0 \ne x \in\widehat{K}. By density, there is y∈Ky \in K such that v(x−y)>v(x)v(x-y) > v(x), and then v(y)=v(x)v(y) = v(x). Thus completion does not enlarge the value group. If x∈V^x \in\widehat{V}, choose y∈Ky \in K with v(x−y)>0v(x-y) > 0. Then y∈Vy \in V and x,yx,y have the same residue class. Hence the residue fields also agree, so K^/K\widehat{K}/K is immediate.

For surjectivity in (33), let x∈V^x \in\widehat{V}. Choose y∈Ky \in K with v(x−y)>v(a)v(x-y) > v(a). Then y∈Vy \in V and x−y∈aV^x-y \in a\widehat{V}. For injectivity, if y∈V∩aV^y \in V \cap a\widehat{V}, then y/a∈K∩V^=Vy/a \in K \cap\widehat{V} = V, and hence y∈aVy \in aV. □

We will now consider Laurent series rings over VV and V^\widehat{V}. Put

R=V((X)),S=V^((X)).R = V((X)), \qquad S = \widehat{V}((X)).

Proposition 6.2. For every 0≠t∈R0 \ne t \in R, the natural map

R/tR⟶S/tS(34)R/tR \longrightarrow S/tS \tag*{(34)}

is an isomorphism. The inclusion R⟶SR \longrightarrow S is faithfully flat.

Proof. Since XX is a unit in both rings, after multiplying tt by a power of XX we may write

t=∑i≥0tiXi∈V[[X]],t0≠0.t = \sum_{i \ge0} t_i X^i \in V[[X]], \qquad t_0 \ne0.

Let s=∑n≥NsnXn∈R^s = \sum_{n \ge N} s_n X^n \in\widehat{R}. We construct bn∈Vb_n \in V and yn∈V^y_n \in\widehat{V} recursively so that, with b=∑n≥NbnXnb = \sum_{n \ge N} b_n X^n and y=∑n≥NynXny = \sum_{n \ge N} y_n X^n, one has s=b+tys = b + ty. Set yj=0y_j = 0 for j<Nj < N. Once yN,…,yn−1y_N,\ldots,y_{n-1} have been chosen, put

rn=sn−∑i=1n−Ntiyn−i∈V^.r_n = s_n - \sum_{i=1}^{n-N} t_i y_{n-i} \in\widehat{V}.

Lemma 6.1, applied to a=t0a=t_0, provides bn∈Vb_n\in V such that rn−bn∈t0V^r_n-b_n\in t_0\widehat{V}. Define

yn=rn−bnt0∈V^.y_n=\frac{r_n-b_n}{t_0}\in\widehat{V}.

The coefficient recursion gives s=b+tys=b+ty. Hence S=R+tSS=R+tS, proving surjectivity of (34).

For injectivity, suppose that b=tyb=ty with b∈Rb\in R and y∈Sy\in S. If y≠0y\ne0, write y=∑n≥NynXny=\sum_{n\geq N}y_nX^n with yN≠0y_N\ne0. Comparing coefficients and arguing inductively gives

t0yn=bn−∑i=1n−Ntiyn−i∈K(n≥N).t_0y_n=b_n-\sum_{i=1}^{n-N}t_i y_{n-i}\in K\qquad(n\geq N).

Thus every yny_n lies in K∩V^=VK\cap\widehat{V}=V, so y∈Ry\in R. Therefore

R∩tS=tR,(35)R\cap tS=tR, \tag*{(35)}

and (34) is an isomorphism.

Let Q=S/RQ=S/R. The identities S=R+tSS=R+tS and (35) show that multiplication by every 0≠t∈R0\ne t\in R is bijective on QQ. Hence QQ is naturally a vector space over Frac⁡(R)\operatorname{Frac}(R), and is therefore flat over RR. The exact sequence

0⟶R⟶S⟶Q⟶00\longrightarrow R\longrightarrow S\longrightarrow Q\longrightarrow0

then shows that SS is flat over RR. Since QQ is flat, tensoring this sequence with any RR-module MM yields an injection

M⟶M⊗RS.M\longrightarrow M\otimes_R S.

Thus the extension is faithfully flat.

Lemma 6.3. Every nonzero element of SS is associated in SS to an element of RR. Consequently, every finitely generated ideal of SS is extended from a finitely generated ideal of RR.

Proof. Let 0≠c∈S0\ne c\in S. Multiplication by a power of XX, which is a unit in both rings, reduces us to

c=∑n≥0cnXn∈V^[[X]],c0≠0.c=\sum_{n\geq0}c_nX^n\in\widehat{V}[[X]],\qquad c_0\ne0.

Since K^/K\widehat{K}/K is immediate, choose 0≠b0∈V0\ne b_0\in V with v(b0)=v(c0)v(b_0)=v(c_0). Then b0V^=c0V^b_0\widehat{V}=c_0\widehat{V} and u0=b0/c0u_0=b_0/c_0 is a unit of V^\widehat{V}.

We construct bn∈Vb_n\in V and un∈V^u_n\in\widehat{V} recursively so that

(∑i≥0ciXi)(∑j≥0ujXj)=∑n≥0bnXn.\left(\sum_{i\geq0}c_iX^i\right)\left(\sum_{j\geq0}u_jX^j\right)=\sum_{n\geq0}b_nX^n.

Suppose u0,…,un−1u_0,\ldots,u_{n-1} have been chosen and set

rn=∑i=1nciun−i∈V^.r_n=\sum_{i=1}^{n}c_i u_{n-i}\in\widehat{V}.

By Lemma 6.1, choose bn∈Vb_n \in V such that bn−rn∈b0V^=c0V^b_n-r_n \in b_0\widehat{V}=c_0\widehat{V}, and put

un=bn−rnc0∈V^.u_n=\frac{b_n-r_n}{c_0}\in\widehat{V}.

Then, for u=∑n≥0unXnu=\sum_{n\geq0}u_nX^n and b=∑n≥0bnXnb=\sum_{n\geq0}b_nX^n, one has cu=bcu=b. Since u0u_0 is a unit, u∈S×u\in S^\times. Hence cc and b∈Rb\in R are associated in SS.

If J=(c1,…,cm)SJ=(c_1,\ldots,c_m)S, choose bi∈Rb_i\in R associated to cic_i. Then

J=(b1,…,bm)S,J=(b_1,\ldots,b_m)S,

which proves the final assertion.

Theorem 6.4. w.dim⁡V((X))=w.dim⁡V^((X))\operatorname{w.dim}V((X))=\operatorname{w.dim}\widehat{V}((X)).

Proof. Again, write R=V((X))R=V((X)) and S=V^((X))S=\widehat{V}((X)). Since SS is faithfully flat over RR by Proposition 6.2, flat base change gives, for RR-modules M,NM,N,

Tor⁡nR(M,N)⊗RS≅Tor⁡nS(M⊗RS,N⊗RS).\operatorname{Tor}^{R}_{n}(M,N)\otimes_R S\cong\operatorname{Tor}^{S}_{n}(M\otimes_R S,N\otimes_R S).

Faithfulness therefore implies

w.dim⁡R≤w.dim⁡S.(36)\operatorname{w.dim}R\leq\operatorname{w.dim}S. \tag*{(36)}

Suppose first that d=w.dim⁡R<∞d=\operatorname{w.dim}R<\infty. Let MM be an SS-module and let J⊆SJ\subseteq S be a finitely generated ideal. By Lemma 6.3, there is a finitely generated ideal I⊆RI\subseteq R such that J=ISJ=IS. A free resolution of R/IR/I, after tensoring with the flat RR-algebra SS, gives

Tor⁡nS(S/J,M)≅Tor⁡nR(R/I,M)(n≥0),(37)\operatorname{Tor}^{S}_{n}(S/J,M)\cong\operatorname{Tor}^{R}_{n}(R/I,M)\qquad(n\geq0), \tag*{(37)}

where MM on the right is viewed as an RR-module. Thus Tor⁡d+1S(S/J,M)=0\operatorname{Tor}^{S}_{d+1}(S/J,M)=0 for every finitely generated ideal JJ of SS. Let LL be the dd-th syzygy of MM in a free resolution, with the convention L=ML=M when d=0d=0. Dimension shifting gives

Tor⁡1S(S/J,L)=0\operatorname{Tor}^{S}_{1}(S/J,L)=0

for every finitely generated JJ. The equational criterion for flatness now shows that LL is flat. Hence fd⁡SM≤d\operatorname{fd}_S M\leq d, and w.dim⁡S≤w.dim⁡R\operatorname{w.dim}S\leq\operatorname{w.dim}R. Together with (36), this proves that w.dim⁡R=w.dim⁡S\operatorname{w.dim}R=\operatorname{w.dim}S when w.dim⁡R\operatorname{w.dim}R is finite. If w.dim⁡R=∞\operatorname{w.dim}R=\infty, (36) forces w.dim⁡S=∞\operatorname{w.dim}S=\infty, so the same equality holds.

Corollary 6.5. For any valuation ring VV of rank 11, with value group R\mathbb{R}, w.dim⁡V[[X]]=2\operatorname{w.dim}V[[X]]=2 if and only if VV is almost maximal, otherwise w.dim⁡V[[X]]=w.dim⁡V((X))>2\operatorname{w.dim}V[[X]]=\operatorname{w.dim}V((X))>2.

Proof. If VV is almost maximal, then V[[X]]V[[X]] is coherent by Theorem 3.2, so w.dim⁡V[[X]]=2\operatorname{w.dim}V[[X]]=2 by [6 Lemma 1].

Conversely, if VV is not almost maximal, then K^\widehat{K} is spherically complete by Lemma 2.5, so it follows from Theorem 5.5 that w.dim⁡V^[[X]]≥3\operatorname{w.dim}\widehat{V}[[X]]\geq3. But since w.dim⁡V^=1\operatorname{w.dim}\widehat{V}=1, this means that w.dim⁡V^[[X]]>w.dim⁡V^+1\operatorname{w.dim}\widehat{V}[[X]]>\operatorname{w.dim}\widehat{V}+1, and hence w.dim⁡V^[[X]]=w.dim⁡V^((X))>2\operatorname{w.dim}\widehat{V}[[X]]=\operatorname{w.dim}\widehat{V}((X))>2 by Proposition 2.4.

But w.dim⁡V^((X))=w.dim⁡V((X))>2\operatorname{w.dim}\widehat{V}((X))=\operatorname{w.dim}V((X))>2 by Theorem 6.4, and applying Proposition 2.4 again gives

w.dim⁡V[[X]]=max⁡{w.dim⁡V+1,w.dim⁡V((X))}=max⁡{2,w.dim⁡V((X))}=w.dim⁡V((X)).\operatorname{w.dim}V[[X]]=\max\{\operatorname{w.dim}V+1,\operatorname{w.dim}V((X))\}=\max\{2,\operatorname{w.dim}V((X))\}=\operatorname{w.dim}V((X)).

Combining all our results so far allow us to finally complete the proof of our main result.

Proof of Theorem A. Using Corollary 3.3, we know i⟺iiii \Longleftrightarrow iii, and i⟹iii \Longrightarrow ii is given by [6 Lemma 1.1], so it remains to prove ii⟹iiiii \Longrightarrow iii. We will assume iiiiii is false, i.e. either VV is not almost maximal or Γ≠0,R,Z\Gamma\ne0, \mathbb{R}, \mathbb{Z}, and show that w.dim⁡V[[X]]=w.dim⁡V((X))>2\operatorname{w.dim} V[[X]] = \operatorname{w.dim} V((X)) > 2, which will complete the proof.

If Γ≠0,R\Gamma\ne0, \mathbb{R} and Γ≇Z\Gamma\not\cong\mathbb{Z}, then Γ\Gamma is a proper dense subgroup of R\mathbb{R}, so it follows from Theorem 4.4 that w.dim⁡V[[X]]>2\operatorname{w.dim} V[[X]] > 2. Since w.dim⁡V=1\operatorname{w.dim} V = 1, this means that w.dim⁡V[[X]]>w.dim⁡V+1\operatorname{w.dim} V[[X]] > \operatorname{w.dim} V + 1, and hence w.dim⁡V[[X]]=w.dim⁡V((X))>2\operatorname{w.dim} V[[X]] = \operatorname{w.dim} V((X)) > 2 by Proposition 2.4.

On the other hand, if VV is not almost maximal, then VV is not discrete, so we may assume that Γ=R\Gamma= \mathbb{R}, and hence w.dim⁡V[[X]]=w.dim⁡V((X))>2\operatorname{w.dim} V[[X]] = \operatorname{w.dim} V((X)) > 2 by Theorem 6.5. □

References

Department of Mathematics, Faculty of Sciences of Monastir, University of Monastir, Monastir, Tunisia

Email address: [email protected]

Department of Pure Mathematics and Mathematical Statistics, Centre for Mathematical Sciences, University of Cambridge, Cambridge United Kingdom

Email address: [email protected]

Quizhen College, Tsinghua University, Beijing, China

Email address: [email protected]

References

  1. [1]M. Anderson; J. Watkins, Coherence of power series rings over pseudo-bezout domains, J. Algebra 107 (1987) 187–194. doi:10.1016/0021-8693(87)90085-8DOI
  2. [2]B. Antieau and R. Datta, Valuation rings are derived splinters, Math. Z. 299 (2021), 827–851. doi:10.1007/s00209-020-02683-6.DOI
  3. [3]B. Bhatt; P. Scholze, Projectivity of the Witt vector affine Grassmannian, Invent. Math. 209 (2017), no. 2, 329–423. doi:10.1007/s00222-016-0710-4.
  4. [4]S. Bouchiba, Global dimensions of power series rings, J. Pure Appl. Algebra 219 (2015), no. 11, 5149–5157. doi:10.1016/j.jpaa.2015.05.010.DOI
  5. [5]A. Jones, Weak dimension of power series rings over valuation rings, J. Pure Appl. Algebra 229 (2025), no. 1, Paper No. 107778. doi:10.1016/j.jpaa.2024.107778.DOI
  6. [6]S. Jøndrup and L. W. Small, Power series over coherent rings, Math. Scand. 35 (1974), 21–24. doi:10.7146/math.scand.a-11530.DOI
  7. [7]E. Matlis, Injective modules over Prüfer rings, Nagoya Math. J. 15 (1959), 57–69. doi:10.1017/S002776300000667X.DOI
  8. [8]S. Rajabi; B. Torrecillas; S. Yassemi, Homological dimensions and special base changes, J. Pure Appl. Algebra 219 (2015), no. 3, 646–651. doi:10.1016/j.jpaa.2014.05.018.DOI
  9. [9]Y. Wang, Spherical completeness, coherence, and GCD properties of formal power series and Witt vector rings, preprint, arXiv:2608.04897v1 [math.AC] (2026). doi:10.48550/arXiv.2608.04897.DOI

Paper details

vlnc:W_4vmxs4qk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents