betaFeedback

WorksMathematics

Iterated socles and closure operations

Abstract

Let (R,m)(R,\mathfrak m) be a dd-dimensional Cohen–Macaulay local ring, d≥2d\geq2 containing a field, and let J⊆msJ\subseteq\mathfrak m^s be a parameter ideal with s≥2s\geq2, also, in the mixed case via some mild assumptions. We prove for that J:m(d−1)(s−1)⊆J‾J:\mathfrak m^{(d-1)(s-1)}\subseteq\overline{J}, and that J:m(d−1)(s−1)+1⊆J‾J:\mathfrak m^{(d-1)(s-1)+1}\subseteq\overline{J} when RR is not regular. We show that Cohen–Macaulayness cannot be removed. We also prove the characteristic-free exact formula g(J)=(d−1)(s−1)+rg(J)=(d-1)(s-1)+r for parameter ideals of order ss with a level tangent cone, giving singular mixed-characteristic Gorenstein and non-Gorenstein examples in which the singular bound is attained.

Introduction

Let (R,m)(R,\mathfrak{m}) be a Noetherian local ring and let JJ be an m\mathfrak{m}-primary ideal. The Goto number of JJ is g(J):=max⁡{t≥0∣J:mt⊆J‾}g(J):=\max\{t\geq0\mid J:\mathfrak{m}^{t}\subseteq\overline{J}\}, where J‾\overline{J} denotes the integral closure of JJ. Thus, the Goto number measures how far one can enlarge JJ by taking colons with powers of the maximal ideal while remaining inside its integral closure. Motivated by a result of Wang [7], Watanabe and Yoshida established a remarkable lower bound for this invariant:

Theorem 1.1 (Watanabe–Yoshida [8 Theorem 6.2]). Let (R,m)(R,\mathfrak{m}) be a dd-dimensional Cohen–Macaulay local ring of characteristic p>0p>0, where d≥2d\geq2. Let s≥2s\geq2 and let J⊆msJ\subseteq\mathfrak{m}^{s} be a parameter ideal. Then

(i) (J:m(d−1)(s−1))⊆J‾(J:\mathfrak{m}^{(d-1)(s-1)})\subseteq\overline{J};

(ii) if RR is not regular, then (J:m(d−1)(s−1)+1)⊆J‾(J:\mathfrak{m}^{(d-1)(s-1)+1})\subseteq\overline{J}.

In characteristic zero, Watanabe and Yoshida obtained related results under additional hypotheses, including normality and conditions on the canonical divisor. They also raised the question of whether the characteristic assumption in the above theorem could be removed. More precisely, they asked the following.

Question 1.2 (See [8 Question 8.1]). Does the containment in Theorem 1.1 hold without any restriction on the characteristic?

They also asked whether the Cohen–Macaulay hypothesis could be removed.

Question 1.3 (See [8 Question 8.2]). Can the Cohen–Macaulay hypothesis in Theorem 1.1 be removed?

A further question, posed in [8 Question 1.2], concerns the determination of the Goto number itself. Related questions about the containment of colon ideals in integral closures were also considered in [3]. The purpose of this paper is to investigate these questions from three complementary perspectives: characteristic-free containments, mixed-characteristic methods, and exact computations of Goto numbers. Our first result is:

Theorem 1.4. Let (R,m)(R,\mathfrak{m}) be a Cohen–Macaulay local ring containing a field, with d=dim⁡R≥2d=\dim R\geq2. Let s≥2s\geq2 and let J⊆msJ\subseteq\mathfrak{m}^{s} be a parameter ideal. Then (J:m(d−1)(s−1))⊆J‾(J:\mathfrak{m}^{(d-1)(s-1)})\subseteq\overline{J}. If RR is not regular, then also (J:m(d−1)(s−1)+1)⊆J‾(J:\mathfrak{m}^{(d-1)(s-1)+1})\subseteq\overline{J}.

The proof in characteristic zero uses the Lefschetz approximation theory of Aschenbrenner and Schoutens. The basic idea is to approximate a complete local ring of characteristic zero by complete local rings of positive characteristic, apply the Watanabe–Yoshida theorem to these approximating rings, and descend the resulting containment by faithful flatness.

We next turn to mixed characteristic. Here ultraproduct constructions provide a framework for comparing certain test ideals in mixed characteristic with their positive characteristic counterparts. In particular, we consider rings of the form

Ri=W(ki)[[X2,…,Xd]]andR♯=W(ulim⁡ki)[[X2,…,Xd]].R_i = W(k_i)[[X_2,\ldots,X_d]] \quad\text{and}\quad R^\sharp= W(\operatorname{ulim} k_i)[[X_2,\ldots,X_d]].

where the kik_i are perfect fields of a fixed positive characteristic. For monomial ideals and fixed rational exponents, we establish the comparison (ulim⁡τ(Ri,ait))R♯=τ(R♯,(a♯)t)(\operatorname{ulim}\tau(R_i,\mathfrak{a}_i^{t}))R^\sharp= \tau(R^\sharp,(\mathfrak{a}^{\sharp})^t). The test ideals here are understood in the sense of [2 Definition 8.13]. We also establish several ultraproduct transfer results for trace-image descriptions, Skoda-type identities, and colon containments.

Our next objective is to determine the Goto number more precisely. The lower bound in Theorem 1.1 depends only on the dimension and the order of the parameter ideal. The associated graded ring, however, contains additional information that can determine the exact value.

Recall that a standard graded Cohen–Macaulay algebra is called level if its canonical module is generated in a single degree. Suppose that gr⁡m(R)\operatorname{gr}_{\mathfrak{m}}(R) is a standard graded Cohen–Macaulay level algebra with Hilbert series h(t)(1−t)d\frac{h(t)}{(1-t)^d}, r:=deg⁡h(t)r := \deg h(t). For a parameter ideal J=(f1,…,fd)J=(f_1,\ldots,f_d), fi∈ms∖ms+1f_i\in\mathfrak{m}^s\setminus\mathfrak{m}^{s+1}, whose initial forms constitute a homogeneous system of parameters, we obtain the following exact formulas:

(J:mt)=J+mr+d(s−1)+1−t,J‾=ms,(J:\mathfrak{m}^t)=J+\mathfrak{m}^{r+d(s-1)+1-t},\qquad\overline{J}=\mathfrak{m}^s,

and, consequently, g(J)=(d−1)(s−1)+rg(J)=(d-1)(s-1)+r.

Thus, the Goto number is governed not only by the dimension and the order of the parameter ideal, but also by the degree of the hh-polynomial of the tangent cone. In particular, the difference between the general singular lower bound and the actual Goto number can be arbitrarily large.

We give examples illustrating this phenomenon, including singular mixed-characteristic Gorenstein and non-Gorenstein rings. In the regular case, the same computation recovers Q‾=ns\overline{Q}=\mathfrak{n}^s, g(Q)=(d−1)(s−1)g(Q)=(d-1)(s-1), for Q:=(u1s,…,uds)Q:=(u_1^s,\ldots,u_d^s) in a regular local ring. Then, we show that the Cohen–Macaulay hypothesis in Question 1.3 cannot be omitted. The final section considers further questions from [8] concerning the interaction between test ideals, colon ideals, and the associated graded ring, see Questions 6.1 and 6.4. In particular, we investigate the containments J+τ(R,Jd−1)⊆(J:J)J+\tau(R,J^{d-1})\subseteq(J:J) and τ(R,mt)⊆mt+a(gr⁡m(R))+1\tau(R,\mathfrak{m}^t)\subseteq\mathfrak{m}^{t+a(\operatorname{gr}_{\mathfrak{m}}(R))+1}, in the settings specified below.

Iterated socles under ultraproducts

Rings in this section are containing a field.

The Lefschetz approximation setup

Definition 2.1. Let WW be a set. An ultrafilter on WW is a collection U⊆P(W)\mathcal{U}\subseteq\mathcal{P}(W) of subsets of WW such that:

(1) ∅∉U\varnothing\notin\mathcal{U} and W∈UW\in\mathcal{U};

(2) if A∈UA\in\mathcal{U} and A⊆B⊆WA\subseteq B\subseteq W, then B∈UB\in\mathcal{U};

(3) if A,B∈UA,B\in\mathcal{U}, then A∩B∈UA\cap B\in\mathcal{U};

(4) for every A⊆WA\subseteq W, exactly one of A∈UA\in\mathcal{U} or W∖A∈UW\setminus A\in\mathcal{U} holds.

We say that a property holds for almost all ww (or U\mathcal{U}-almost everywhere) if the set of ww for which it holds belongs to U\mathcal{U}.

Definition 2.2. Let (Xw)w∈W(X_w)_{w\in W} be a family of nonempty sets. On the product ∏w∈WXw\prod_{w\in W}X_w define (xw)∼(yw)⟺{w∈W:xw=yw}∈U(x_w)\sim(y_w)\Longleftrightarrow\{w\in W:x_w=y_w\}\in\mathcal{U}. This is an equivalence relation. The ultraproduct is ulim⁡wXw:=(∏w∈WXw)/∼\operatorname{ulim}_w X_w:=\left(\prod_{w\in W}X_w\right)/\sim. The class of (xw)(x_w) is denoted [xw][x_w] or ulim⁡wxw\operatorname{ulim}_w x_w. If all Xw=XX_w=X, one writes XUX^{\mathcal{U}} and speaks of the ultrapower.

Definition 2.3. If each RwR_w is a ring, the componentwise operations on ∏w∈WRw\prod_{w\in W}R_w descend to the ultraproduct D(R)=ulim⁡wRwD(R)=\operatorname{ulim}_w R_w, which is again a ring with

[xw]+[yw]:=[xw+yw],[xw]⋅[yw]:=[xw⋅yw],0:=[0w],1:=[1w].[x_w]+[y_w]:=[x_w+y_w],\qquad[x_w]\cdot[y_w]:=[x_w\cdot y_w],\qquad0:=[0_w],\qquad1:=[1_w].

More generally the same construction applies to groups, modules, fields and first-order structures in a fixed language.

Fact 2.4 (Łos’s theorem). A first-order sentence in the language of rings holds in ulim⁡wRw\operatorname{ulim}_{w} R_{w} if and only if it holds in RwR_{w} for almost all ww.

Let (R,m)(R,\mathfrak m) be a local ring of equicharacteristic zero. Aschenbrenner–Schoutens [1 Theorem 4.3, Theorem 5.2, Lemmas 5.3–5.4] construct a faithfully flat extension R→D(R)=ulim⁡wRwR \to D(R)=\operatorname{ulim}_{w} R_{w}, an ultraproduct of complete local rings RwR_{w} of positive characteristic. Here ww runs over an index set WW equipped with an ultrafilter U\mathcal{U}; Each RwR_{w} is a complete local ring with maximal ideal mw\mathfrak m_{w}, and the maximal ideal m\mathfrak m of RR is approximated by the mw\mathfrak m_{w}: the image of m\mathfrak m in D(R)D(R) is ulim⁡wmw\operatorname{ulim}_{w}\mathfrak m_{w}. The transfer principle says: given a fixed finite list of elements a1,…,as∈Ra_{1},\ldots,a_{s}\in R and ideals I1,…,It⊆RI_{1},\ldots,I_{t}\subseteq R, one can choose compatible approximations ai,w∈Rwa_{i,w}\in R_{w} and Ij,w⊆RwI_{j,w}\subseteq R_{w} such that, for almost all ww,

  • dim⁡Rw=dim⁡R\dim R_{w}=\dim R and embdim⁡Rw=embdim⁡R\operatorname{embdim} R_{w}=\operatorname{embdim} R (when finite);

  • RwR_{w} is Cohen–Macaulay iff RR is;

  • a fixed tuple approximating a parameter sequence of RR is a parameter sequence of RwR_{w};

  • all fixed ideal relations among the aia_{i} and IjI_{j} hold in RwR_{w};

Let x∈Rx\in R and I⊆RI\subseteq R an ideal. We want to transfer statements such as:

  • xx is integral over an ideal II, i.e. xx satisfies an equation xd+c1xd−1+⋯+cd=0x^{d}+c_{1}x^{d-1}+\cdots+c_{d}=0 with ci∈Iic_{i}\in I^{i};

  • an element lies in the integral closure I‾\overline{I} of II.

Dealing with Question 1.2

We give the additional numerical argument needed to transfer integral dependence. In particular, we do not assume a uniform degree for integral equations in the approximating rings.

Lemma 2.5. If HH is an m\mathfrak m-primary ideal of RR, then, for almost all ww, ℓRw(Rw/Hw)=ℓR(R/H)\ell_{R_{w}}(R_{w}/H_{w})=\ell_{R}(R/H).

Proof. Let the length on the right be hh, and lift a composition series of R/HR/H to ideals H=H0⊊H1⊊⋯⊊Hh=RH=H_{0}\subsetneq H_{1}\subsetneq\cdots\subsetneq H_{h}=R. Each factor is cyclic, nonzero, and annihilated by m\mathfrak m. Choose its cyclic generator and finite generating sets of the HiH_{i}. The corresponding generation relations, inclusions mHi⊆Hi−1\mathfrak mH_{i}\subseteq H_{i-1}, and nonmembership of the cyclic generator in Hi−1H_{i-1} transfer to almost all RwR_{w}. Nonmembership is preserved because the extension to D(R)D(R) is faithfully flat. Since mw\mathfrak m_{w} is maximal, the resulting chain has exactly hh simple factors. □

Lemma 2.6. Let RR be a complete Cohen–Macaulay local ring of equicharacteristic zero, and let II be an m\mathfrak m-primary ideal. Then e(Iw,Rw)=e(I,R)e(I_{w},R_{w})=e(I,R) for almost all ww.

Proof. The residue field of RR is infinite. Choose a dd-generated minimal reduction K⊆IK\subseteq I, where d=dim⁡Rd=\dim R. Thus KK is a parameter ideal and, for some fixed integer N≥0N\geq0, IN+1=KINI^{N+1}=KI^{N}. This equality is a finite collection of ideal-membership relations. Consequently KwK_{w} is a reduction of IwI_{w} for almost all ww. For these ww, RwR_{w} is Cohen–Macaulay of dimension dd and KwK_{w} is a parameter ideal.

e(Iw,Rw)=e(Kw,Rw)=ℓRw(Rw/Kw)=ℓR(R/K)=e(K,R)=e(I,R).e(I_{w},R_{w})=e(K_{w},R_{w})=\ell_{R_{w}}(R_{w}/K_{w})=\ell_{R}(R/K)=e(K,R)=e(I,R).

The middle equality is Lemma 2.5. □

Lemma 2.7. Let RR be as in Lemma 2.6. Let J⊆IJ\subseteq I be m\mathfrak m-primary ideals. If Iw⊆Jw‾I_{w}\subseteq\overline{J_{w}} for almost all ww, then I⊆J‾I\subseteq\overline{J}.

Proof. Integral dependence gives e(Iw,Rw)=e(Jw,Rw)e(I_{w},R_{w})=e(J_{w},R_{w}). Apply Lemma 2.6 to II and JJ to obtain e(I,R)=e(J,R)e(I,R)=e(J,R). A complete Cohen–Macaulay local ring is formally equidimensional. Rees’s multiplicity theorem [4 Theorem 11.3.1] now gives I⊆J‾I\subseteq\overline{J}. □

Theorem 2.8. Let (R,m)(R,\mathfrak m) be a Cohen–Macaulay local ring containing a field, with d=dim⁡R≥2d=\dim R\geq2. If s≥2s\geq2 and J⊆msJ\subseteq\mathfrak m^{s} is a parameter ideal, then (J:m(d−1)(s−1))⊆J‾(J:\mathfrak m^{(d-1)(s-1)})\subseteq\overline{J}. If RR is not regular, then also (J:m(d−1)(s−1)+1)⊆J‾(J:\mathfrak m^{(d-1)(s-1)+1})\subseteq\overline{J}.

Proof. In positive characteristic these are precisely [8 Theorem 1.1]. Suppose that RR has characteristic zero. First pass to the m\mathfrak m-adic completion. This preserves dimension, Cohen–Macaulayness, and regularity. It also preserves the colon condition, and integral dependence descends by faithful flatness. For completeness, if zz becomes integral over JR‾J\overline{R} and I=J+(z)I = J + (z), then for some nn In+1R‾=JInR‾I^{n+1}\overline{R} = JI^n\overline{R}. Faithful flatness gives In+1=JInI^{n+1} = JI^n, and hence z∈Jz \in J. We may therefore assume that R‾\overline{R} is complete. Take t=t0t = t_0, or t=t0+1t = t_0 + 1 in the nonregular case, and fix z∈J:mtz \in J : \mathfrak{m}^t. Choose compatible approximations of RR, JJ, m\mathfrak{m}, and zz. For almost all ww,

dim⁡Rw=d,Jw is a parameter ideal,Jw⊆mws,zwmwt⊆Jw,\dim R_w = d,\qquad J_w\text{ is a parameter ideal},\qquad J_w \subseteq\mathfrak{m}_w^s,\qquad z_w\mathfrak{m}_w^t \subseteq J_w,

and RwR_w is Cohen–Macaulay. In the nonregular case, edim⁡R>d\operatorname{edim} R > d; dimension and embedding dimension transfer, so that almost all RwR_w are nonregular. The positive-characteristic theorem yields zw∈J‾wz_w \in\overline{J}_w for almost all ww. In particular zw∈mwz_w \in\mathfrak{m}_w for almost all ww, so z∈mz \in\mathfrak{m}. Thus J+(z)J + (z) is a proper m\mathfrak{m}-primary ideal. Apply Lemma 2.7 to J⊆J+(z)J \subseteq J + (z). It follows that z∈J‾z \in\overline{J}, as required. □

Mixed char.

Low dimensional approach

Proposition 3.1. Let (R,m)(R,\mathfrak{m}) be a two-dimensional Cohen–Macaulay local ring. For every parameter ideal J⊆msJ \subseteq\mathfrak{m}^s, s≥2s \ge2, (J:ms−1)⊆J‾(J : \mathfrak{m}^{s-1}) \subseteq\overline{J}. If RR is not regular, then (J:ms)⊆J‾(J : \mathfrak{m}^s) \subseteq\overline{J}.

Proof. If RR is regular, Heinzer–Swanson prove, without a characteristic restriction, g(J)=ord⁡m(J)−1g(J) = \operatorname{ord}_{\mathfrak{m}}(J) - 1, see [5 Theorem 2.2]. Since ord⁡m(J)≥s\operatorname{ord}_{\mathfrak{m}}(J) \ge s, this proves the first assertion. If RR is nonregular, Wang’s theorem [7 Theorems 1.1–1.2], also reproduced as [8 Theorem 1.1], gives

(J:ms)2=J(J:ms).(J : \mathfrak{m}^s)^2 = J(J : \mathfrak{m}^s).

Thus (J:ms)(J : \mathfrak{m}^s) is integral over JJ. The inclusion (J:ms−1)⊆(J:ms)(J : \mathfrak{m}^{s-1}) \subseteq(J : \mathfrak{m}^s) gives the other assertion. □

Proposition 3.2. Let RR be Cohen–Macaulay and either nonregular of dimension at least two, or regular of dimension at least three. Let JJ be a parameter ideal and put r=ord⁡m(J)≥2r = \operatorname{ord}_{\mathfrak{m}}(J) \ge2. Then, in any characteristic, g(J)≥rg(J) \ge r.

Proof. Apply Wang’s theorem with exponent rr and use the increasing chain (J:ma)⊆(J:mb)(J : \mathfrak{m}^a) \subseteq(J : \mathfrak{m}^b) for a≤ba \le b. For d=3d = 3, s=2s = 2, one has t0=2≤rt_0 = 2 \le r. □

The containment J+τ(R,J‾d−1)⊆(J:J‾)J + \tau(R,\overline{J}^{d-1}) \subseteq(J : \overline{J})

Notation 3.3. Fix an ultrafilter UU on an index set and let D:=ulim⁡RiD := \operatorname{ulim} R_i. If Ii⊆RiI_i \subseteq R_i are ideals, write IU:=ulim⁡Ii:={[zi]∈D:zi∈Ii for U-almost all i}I_U := \operatorname{ulim} I_i := \{[z_i] \in D : z_i \in I_i\text{ for }U\text{-almost all }i\}. This is an ideal of DD, called the internal ideal of the family. It need not be finitely generated.

Lemma 3.4. Suppose Ii=(f1,i,…,fr,i)I_i = (f_{1,i}, \ldots, f_{r,i}) with the same finite rr for every ii, and put fj=[fj,i]f_j = [f_{j,i}]. Then IU=(f1,…,fr)DI_U = (f_1, \ldots, f_r)D. For any other family HiH_i of ideals, one has

ulim⁡(IiHi)=IUHU,ulim⁡(Ii:Hi)=(IU:HU).\operatorname{ulim}(I_iH_i) = I_UH_U,\qquad\operatorname{ulim}(I_i : H_i) = (I_U : H_U).

The equality of colons holds even without the bound on the number of generators of IiI_i.

Proof. For the first assertion, express each element of IiI_i as a sum of the same rr generators with coefficients in RiR_i, and take their ultraproducts. For the product assertion, every element of IiHiI_iH_i is ∑j=1rfj,ihj,i\sum_{j=1}^r f_{j,i}h_{j,i} with hj,i∈Hih_{j,i} \in H_i. The fixed bound on the number of summands permits passage to DD. For the colon assertion, one inclusion is immediate. Conversely, if zi∉(Ii:Hi)z_i \notin(I_i : H_i) on a set in UU, choose hi∈Hih_i \in H_i with zihi∉Iiz_ih_i \notin I_i on that set. Then [hi]∈HU[h_i] \in H_U and [zi][hi]∉IU[z_i][h_i] \notin I_U. This proves the other inclusion. □

Definition 3.5. For a family on which the mixed-characteristic test ideals are defined, set TU(at):=ulim⁡τ(Ri,ait)T_U(\mathfrak{a}^t) := \operatorname{ulim}\tau(R_i,\mathfrak{a}_i^t).

Proposition 3.6. Suppose that, for each ii, a specified RiR_i-linear map ϕi:Mi→Ri\phi_i : M_i \to R_i satisfies im⁡(ϕi)=τ(Ri,ait)\operatorname{im}(\phi_i) = \tau(R_i,\mathfrak{a}_i^t). Then the coordinatewise map ΦU:ulim⁡Mi→D\Phi_U : \operatorname{ulim} M_i \to D defined by [mi]↦[ϕi(mi)][m_i] \mapsto[\phi_i(m_i)] is DD-linear and satisfies TU(at)=im⁡(ΦU)T_U(\mathfrak{a}^t) = \operatorname{im}(\Phi_U).

Proof. The map is well-defined and linear by its coordinatewise definition. If [zi]∈TU(at)[z_i] \in T_U(\mathfrak{a}^t), choose mi∈Mim_i \in M_i with ϕi(mi)=zi\phi_i(m_i) = z_i on the set in UU where zi∈τ(Ri,ait)z_i \in\tau(R_i,\mathfrak{a}_i^t). These choices give a preimage of [zi][z_i]. The reverse containment is immediate. □

Proposition 3.7. Let the RiR_i be complete normal Q\mathbb{Q}-Gorenstein local domains of mixed characteristic. Suppose ai\mathfrak{a}_i is generated by a fixed number rr of specified elements. Then, for every rational t≥rt \ge r, TU(at)=aUTU(at−1)\mathcal{T}_U(\mathfrak{a}^t)=\mathfrak{a}_U\mathcal{T}_U(\mathfrak{a}^{t-1}).

Proof. Apply Skoda’s theorem in each factor [2 Theorem 8.7 and Corollary 8.20(b)] and then apply Lemma 3.4 to the products. □

Proposition 3.8. Let RR be a complete normal Q\mathbb{Q}-Gorenstein local domain of mixed characteristic and dimension d≥2d \ge2. For every parameter ideal JJ, J+τ(R,Jd−1)⊆(J:J‾)J+\tau(R,J^{d-1})\subseteq(J:\overline{J}). Consequently, for a family of such rings of fixed dimension dd and parameter ideals JiJ_i, let J_U:=\ulim J_i and C_U:=\ulim J_i. Then JU+TU(Jd−1)⊆(JU:CU)J_U+\mathcal{T}_U(J^{d-1})\subseteq(J_U:C_U).

Proof. We use monotonicity, invariance under integral closure of the ideal, and Skoda’s theorem. The integral-closure invariance for the unified ideal follows from [2 Proposition 8.6(c), Theorem 8.11, and Definition 8.13], using a common sufficiently small perturbation for the two ideals in this single ring. For any ideal LL and t≥0t\ge0, Lτ(R,Lt)⊆τ(R,Lt+1)L\tau(R,L^t)\subseteq\tau(R,L^{t+1}). Indeed, for f∈Lf\in L, principal Skoda gives fτ(R,Lt)=τ(R,(f)Lt)f\tau(R,L^t)=\tau(R,(f)L^t), and monotonicity gives the containment in τ(R,Lt+1)\tau(R,L^{t+1}). Thus

J‾ τ(R,Jd−1)=J‾ τ(R,J‾d−1)⊆τ(R,J‾d)=τ(R,Jd)=Jτ(R,Jd−1)⊆J.\overline{J}\,\tau(R,J^{d-1})=\overline{J}\,\tau(R,\overline{J}^{d-1})\subseteq\tau(R,\overline{J}^{d})=\tau(R,J^d)=J\tau(R,J^{d-1})\subseteq J.

Here Skoda applies to the dd generators of JJ. This proves the single-ring assertion. Take ultraproducts and use the colon equality in Lemma 3.4 for the second assertion. □

An unconditional mixed-characteristic comparison

Fix a prime pp, an integer d≥1d\ge1, and perfect fields kik_i of characteristic pp.

Notation 3.9. Let Ri:=W(ki)[[X2,…,Xd]]R_i:=W(k_i)[[X_2,\ldots,X_d]], mi=(p,X2,…,Xd)\mathfrak{m}_i=(p,X_2,\ldots,X_d), and K:=\ulim k_i. Let D:=\ulim R_i and m:=(p,X2,…,Xd)D\mathfrak{m}:=(p,X_2,\ldots,X_d)D.

The symbols XjX_j on DD denote the classes of the corresponding variables. The field KK is perfect. In the formulas below, set u1=pu_1=p and uj=Xju_j=X_j for 2≤j≤d2\le j\le d.

Lemma 3.10. There is a natural isomorphism R♯:=lim⁡N←D/mN≅W(K)[[X2,…,Xd]]R^\sharp:=\lim\limits_{\overleftarrow{N}}D/\mathfrak{m}^N\cong W(K)[[X_2,\ldots,X_d]]. It identifies the images of p,X2,…,Xdp,X_2,\ldots,X_d with the indicated regular parameters. In particular, R♯R^\sharp is a complete Noetherian regular local ring of mixed characteristic (0,p)(0,p).

Proof. For every fixed NN, Lemma 3.4 gives D/\mathfrak{m}^N\cong\ulim(R_i/\mathfrak{m}_i^N). Only finitely many monomials XαX^\alpha with ∣α∣<N|\alpha|<N are needed to describe Ri/miNR_i/\mathfrak{m}_i^N; their coefficients are respectively taken modulo pN−∣α∣p^{N-|\alpha|}. Thus they are truncated Witt vectors of bounded lengths. For each fixed nn, addition and multiplication on Wn(ki)W_n(k_i) are given by the same finite list of Witt polynomials, because pp is fixed. Consequently \ulim W_n(k_i)=W_n(K). Applying this simultaneously to the finitely many coefficient slots gives compatible ring isomorphisms D/mN≅W(K)[[X2,…,Xd]]/(p,X2,…,Xd)ND/\mathfrak{m}^N\cong W(K)[[X_2,\ldots,X_d]]/(p,X_2,\ldots,X_d)^N. Take the inverse limit over NN. This argument constructs the map D→R♯D\to R^\sharp; no approximation theorem asserting faithful flatness of that map is needed or claimed. □

Let E⊂Z≥0dE\subset\mathbb{Z}_{\ge0}^d be a fixed finite nonempty set. Assume the monomial ideals

ai=(uα:α∈E)Ri,aD=(uα:α∈E)D,a♯=(uα:α∈E)R♯\mathfrak{a}_i=(u^\alpha:\alpha\in E)R_i,\qquad\mathfrak{a}_D=(u^\alpha:\alpha\in E)D,\qquad\mathfrak{a}^\sharp=(u^\alpha:\alpha\in E)R^\sharp

are proper. Their common Newton polyhedron is P=conv⁡(E)+R≥0dP=\operatorname{conv}(E)+\mathbb{R}_{\ge0}^d. We record a version of Dickson’s lemma that we need:

Lemma 3.11 (Dickson’s lemma). Let d≥1d\ge1 and let S⊆NdS\subseteq\mathbb{N}^d. Then SS has only finitely many minimal elements with respect to the coordinatewise partial order. Moreover, if SS is upward closed, then every element of SS lies above a minimal element of SS.

Theorem 3.12. For every rational t>0t>0, TU(at)=(uβ:β∈Z≥0d,β+1∈Int⁡(tP))D\mathcal{T}_U(\mathfrak{a}^t)=(u^\beta:\beta\in\mathbb{Z}_{\ge0}^d,\beta+\mathbf{1}\in\operatorname{Int}(tP))D and TU(at)R♯=τ(R♯,(a♯)t)\mathcal{T}_U(\mathfrak{a}^t)R^\sharp=\tau(R^\sharp,(\mathfrak{a}^\sharp)^t). In particular, the first ideal is finitely generated by a set of monomials independent of ii.

Proof. The mixed-characteristic monomial formula [2 Proposition 8.21] says that

τ(Ri,ait)=(uβ:β+1∈Int⁡(tP), β∈Z≥0d)Ri.\tau(R_i,\mathfrak{a}_i^t)=(u^\beta:\beta+\mathbf{1}\in\operatorname{Int}(tP),\ \beta\in\mathbb{Z}_{\geq0}^d)R_i.

The exponent set is upward closed for the coordinatewise order. Dickson’s lemma gives a finite set M⊂Z≥0dM\subset\mathbb{Z}_{\geq0}^d of minimal exponents generating it. Both PP and MM are independent of ii. The first assertion now follows from Lemma 3.4. Lemma 3.10 identifies R♯R^\sharp as another complete regular local ring with the same regular parameters. Applying the same monomial formula to this ring proves the second assertion.

Corollary 3.13. For fixed EE and t>0t>0 as above, there exists ϵ0>0\epsilon_0>0 such that, for every ii and rational 0<ϵ<ϵ00<\epsilon<\epsilon_0, τ(Ri,ait)=τ(Ri,ait+ϵ)\tau(R_i,\mathfrak{a}_i^t)=\tau(R_i,\mathfrak{a}_i^{t+\epsilon}).

Proof. Write the nontrivial supporting inequalities of PP as ⟨vj,x⟩≥cj\langle v_j,x\rangle\geq c_j, with integral vectors vjv_j and positive integers cjc_j; there are finitely many of them. For β∈Z≥0d\beta\in\mathbb{Z}_{\geq0}^d, the numbers ⟨vj,β+1⟩\langle v_j,\beta+\mathbf{1}\rangle are integers. Choose

0<ϵ0<min⁡j⌊tcj⌋+1−tcjcj.0<\epsilon_0<\min_j\frac{\lfloor tc_j\rfloor+1-tc_j}{c_j}.

Then replacing tt by t+ϵt+\epsilon changes none of the strict integer inequalities defining the monomial test ideal. Coordinate inequalities with zero right-hand side do not change.

The Watanabe–Yoshida identity for equal parameter powers

We continue with the fixed-prime family in § 3.3, and assume d,s≥2d,s\geq2. Set

Qi:=(ps,X2s,…,Xds)Ri,Q:=(u1s,…,uds)D,Q♯=QR♯,h:=(d−1)(s−1).Q_i:=(p^s,X_2^s,\ldots,X_d^s)R_i,\qquad Q:=(u_1^s,\ldots,u_d^s)D,\qquad Q^\sharp=QR^\sharp,\qquad h:=(d-1)(s-1).

Theorem 3.14. With ordinary ideal integral closure on DD, one has Q‾=ms=⋃iQi‾\overline{Q}=\mathfrak{m}^s=\overline{\bigcup_i Q_i}, Tu(Qd−1)=mhT_u(Q^{d-1})=\mathfrak{m}^h and (Q:Q‾)=Q+Tu(Qd−1)(Q:\overline{Q})=Q+T_u(Q^{d-1}). Also, (Q♯:Q♯‾)=Q♯+τ(R♯,(Q♯)d−1)(Q^\sharp:\overline{Q^\sharp})=Q^\sharp+\tau(R^\sharp,(Q^\sharp)^{d-1}). Furthermore, (Q:mh)=ms=Q‾(Q:\mathfrak{m}^h)=\mathfrak{m}^s=\overline{Q}, and (Q:mh+1)=Q+ms−1⊈Q‾(Q:\mathfrak{m}^{h+1})=Q+\mathfrak{m}^{s-1}\not\subseteq\overline{Q}.

Proof. We first record computations valid in every regular local ring (S,n)(S,\mathfrak{n}) of dimension dd with parameters v1,…,vdv_1,\ldots,v_d, in any characteristic. Let QS=(v1s,…,vds)Q_S=(v_1^s,\ldots,v_d^s). Every monomial vαv^\alpha of total degree ss is integral over QSQ_S, since (vα)s∈QSs(v^\alpha)^s\in Q_S^s. Also ns\mathfrak{n}^s is integrally closed. Indeed, gr⁡nS\operatorname{gr}_{\mathfrak{n}}S is a polynomial domain, so the n\mathfrak{n}-adic order is multiplicative. An integral equation over ns\mathfrak{n}^s cannot hold for an element of order less than ss, because its highest power would be the unique term of smallest order. Therefore QS‾=ns\overline{Q_S}=\mathfrak{n}^s. The associated graded ring of S/QSS/Q_S is k[V1,…,Vd]/(V1s,…,Vds)k[V_1,\ldots,V_d]/(V_1^s,\ldots,V_d^s), an artinian graded Gorenstein ring with socle degree L=d(s−1)L=d(s-1). Its perfect pairings into degree LL give (QS:nr)=QS+nL+1−r(Q_S:\mathfrak{n}^r)=Q_S+\mathfrak{n}^{L+1-r} for 0≤r≤L+10\leq r\leq L+1. To justify the filtered step, an element of order j≤L−rj\leq L-r has nonzero initial form which can be multiplied by a degree-rr form to give a nonzero product. It therefore does not annihilate nr\mathfrak{n}^r modulo QSQ_S. The opposite inclusion follows from nL+1⊆QS\mathfrak{n}^{L+1}\subseteq Q_S. Let us apply these computations in each RiR_i. The fixed finite generating sets for powers of mi\mathfrak{m}_i allow all the displayed colon identities to pass to DD by Lemma 3.4. It remains to check that the internal integral closure is the ordinary integral closure in DD. If z∈Q‾z\in\overline{Q}, choose a single integral equation for zz over QQ. Its degree is finite, and membership of its coefficients in the fixed powers of QiQ_i is witnessed by finitely many equations. It follows that zi∈Qi‾=misz_i\in\overline{Q_i}=\mathfrak{m}_i^s almost everywhere, so z∈msz\in\mathfrak{m}^s. Conversely, the degree-ss monomial equations above work directly in DD, and the integral closure of an ideal is an ideal. Hence ms⊆Q‾\mathfrak{m}^s\subseteq\overline{Q}. The Newton polyhedron of QiQ_i is {x∈R≥0d:x1+⋯+xd≥s}\{x\in\mathbb{R}_{\geq0}^d:x_1+\cdots+x_d\geq s\}. Theorem 3.12 with t=d−1t=d-1 shows that its test ideal is generated by monomials with ∣β∣+d>s(d−1)|\beta|+d>s(d-1), i.e., ∣β∣≥h|\beta|\geq h. Thus Tu(Qd−1)=mhT_u(Q^{d-1})=\mathfrak{m}^h. Setting r=sr=s in the colon formula gives (Q:ms)=Q+mh(Q:\mathfrak{m}^s)=Q+\mathfrak{m}^h, proving the first boxed identity. The same calculations apply to R♯R^\sharp, proving the second one. Finally set r=hr=h and r=h+1r=h+1 in that formula. The element u1s−1u_1^{s-1} belongs to the latter colon and not to ms\mathfrak{m}^s, in each factor and hence also in DD.

Remark 3.15. We will present in Proposition 5.5 another argument of Theorem 3.14 without any use of mixed-characteristic test ideal from [2].

Proposition 3.16. Let (R,m)(R,\mathfrak{m}) have mixed characteristic (0,p)(0,p). There is no faithfully flat unital map from RR to an ultraproduct of nonzero rings of prime positive characteristics.

Proof. If the factors have characteristic pp on a set in UU, then pp maps to zero, contradicting the injectivity of a faithfully flat map. Otherwise pp is a unit in almost every factor, and hence in the ultraproduct. This contradicts faithful flatness, since the proper ideal pRpR would extend to the unit ideal. □

The role of Cohen–Macaulayness

We show the basic containment of Theorem 1.1 fails for TT. The ring TT is reduced, complete, two-dimensional, and not Cohen–Macaulay; moreover, π\pi is a nonzerodivisor. Choosing V=k[[π]]V=k[[\pi]] gives examples over every field.

Proposition 4.1. Let (V,π)(V,\pi) be a complete discrete valuation ring. Define T:=V[[x,y]]/(xy,y2−πy)T:=V[[x,y]]/(xy,y^2-\pi y) and t=(π,x,y)\mathfrak{t}=(\pi,x,y). For every s≥3s\geq3, the ideal J=(πs,xs)TJ=(\pi^s,x^s)T is a parameter ideal contained in ts\mathfrak{t}^s, but ys−1∈(J:ts−1)∖J‾y^{s-1}\in(J:\mathfrak{t}^{s-1})\setminus\overline{J}.

Proof. In V[[x,y]]V[[x,y]], one has (xy,y2−πy)=(y)∩(x,y−π)(xy,y^2-\pi y)=(y)\cap(x,y-\pi). Both ideals on the right are prime. Their quotient rings are V[[x]]V[[x]] and VV, of dimensions two and one. Thus TT is reduced and has minimal components of different dimensions. Since it is complete, it cannot be Cohen–Macaulay. Neither minimal prime contains π\pi, so π\pi is a nonzerodivisor. In mixed characteristic pp is a unit times a positive power of π\pi, so TT is also pp-torsionfree.

The relations in TT give yn=πn−1yy^n=\pi^{n-1}y for n≥1n\geq1. In particular ys+1=πsy∈Jy^{s+1}=\pi^s y\in J. Together with πs,xs∈J\pi^s,x^s\in J, this proves that JJ is t\mathfrak{t}-primary and hence a parameter ideal. Let a:=ys−1=πs−2ya:=y^{s-1}=\pi^{s-2}y. Any degree-(s−1)(s-1) monomial in π,x,y\pi,x,y containing xx is killed by aa. For a monomial πiyj\pi^i y^j with i+j=s−1i+j=s-1, one has aπiyj=π2s−3y∈(πs)T⊆Ja\pi^i y^j=\pi^{2s-3}y\in(\pi^s)T\subseteq J, since 2s−3≥s2s-3\geq s for s≥3s\geq3. Thus ats−1⊆Ja\mathfrak{t}^{s-1}\subseteq J.

Under the quotient map T→T/(x,y−π)T≅VT\to T/(x,y-\pi)T\cong V the ideal JJ maps to (πs)(\pi^s) and aa maps to πs−1\pi^{s-1}. An integral equation descends under any ring homomorphism. But (πs)(\pi^s) is integrally closed in the discrete valuation ring VV, and does not contain πs−1\pi^{s-1}. Consequently a∉J‾a\notin\overline{J}. □

Bounds on Goto number

We start with some lemmas.

Lemma 5.1. Let (A,q)(A,\mathfrak{q}) be artinian so that B:=gr⁡q(A)B:=\operatorname{gr}_{\mathfrak{q}}(A) is a standard graded artinian algebra whose socle is concentrated in degree LL. Then (0:Aqt)=qL+1−t(0:_A\mathfrak{q}^t)=\mathfrak{q}^{L+1-t} for 0≤t≤L+10\leq t\leq L+1.

Proof. Because Hi=0H_i=0 for i>Li>L, Nakayama’s lemma gives qL+1=0\mathfrak{q}^{L+1}=0. Thus the right-hand side is contained in the left-hand side. For the reverse inclusion let 0≠a∈A0\ne a\in A with j=ord⁡q(a)≤L−tj=\operatorname{ord}_{\mathfrak{q}}(a)\leq L-t. Its initial form a∗∈Bja^*\in B_j is nonzero. Since the socle of BB is concentrated in degree LL, one can successively multiply a∗a^* by degree-one elements without obtaining zero until reaching degree LL. In particular, there is b∗∈Btb^*\in B_t with a∗b∗≠0a^*b^*\ne0. Lift b∗b^* to b∈qtb\in\mathfrak{q}^t. The initial form of abab is a∗b∗≠0a^*b^*\ne0, so ab≠0ab\ne0. Hence a∉0:Aqta\notin0:_A\mathfrak{q}^t, as required. The endpoint cases t=0,L+1t=0,L+1 follow by the same argument. □

Lemma 5.2. Let (R,m)(R,\mathfrak{m}) be a local ring of positive dimension, and let s≥2s\geq2. Then ms−1⊈ms‾\mathfrak{m}^{s-1}\not\subseteq\overline{\mathfrak{m}^s}.

Proof. Otherwise, the ideal ms\mathfrak{m}^s would be a reduction of ms−1\mathfrak{m}^{s-1}. For sufficiently large NN, (ms−1)N+1=ms(ms−1)N(\mathfrak{m}^{s-1})^{N+1}=\mathfrak{m}^s(\mathfrak{m}^{s-1})^N, which says m(s−1)(N+1)=ms+(s−1)N\mathfrak{m}^{(s-1)(N+1)}=\mathfrak{m}^{s+(s-1)N}. Nakayama’s lemma would force a positive power of m\mathfrak{m} to vanish, contradicting dim⁡R>0\dim R>0. □

The following has application to Question 1.2 when the ring is of mixed char.

Theorem 5.3. Let (R,m,k)(R,\mathfrak{m},k) be a dd-dimensional Cohen–Macaulay local ring, d≥2d\geq2. Suppose G:=gr⁡m(R)G:=\operatorname{gr}_{\mathfrak{m}}(R) is a Cohen–Macaulay, standard graded level kk-algebra, with Hilbert series Hilb⁡G(t)=h(t)(1−t)d\operatorname{Hilb}_G(t)=\frac{h(t)}{(1-t)^d} with r:=deg⁡h(t)r:=\deg h(t). Fix s≥2s\geq2 and a parameter ideal J=(f1,…,fd)J=(f_1,\ldots,f_d), where each fi∈ms∖ms+1f_i\in\mathfrak{m}^s\setminus\mathfrak{m}^{s+1} and the initial forms f1∗,…,fd∗∈Gsf_1^*,\ldots,f_d^*\in G_s constitute a homogeneous system of parameters in GG. Then for 0≤t≤r+d(s−1)+10\leq t\leq r+d(s-1)+1, (i) (J:mt)=J+mr+d(s−1)+1−t(J:\mathfrak{m}^t)=J+\mathfrak{m}^{r+d(s-1)+1-t}, (ii) J‾=ms\overline{J}=\mathfrak{m}^s and g(J)=(d−1)(s−1)+rg(J)=(d-1)(s-1)+r.

Proof. The initial forms form a regular sequence in the Cohen–Macaulay ring GG. The standard lifting theorem for regular initial sequences gives gr⁡m(R/J)≅G/(f1∗,…,fd∗)\operatorname{gr}_{\mathfrak m}(R/J) \cong G/(f_1^*,\ldots,f_d^*). This Artinian graded ring is level: for a Cohen–Macaulay level graded ring, quotienting by a homogeneous regular sequence preserves levelness. One way to see this is to use the graded canonical module: it is generated in a single degree, and the canonical module of the quotient is the quotient of the canonical module, up to a grading shift. The Hilbert series of the quotient is h(t)(1+t+⋯+ts−1)dh(t)(1+t+\cdots+t^{s-1})^d, so its top socle degree is L=r+d(s−1)L=r+d(s-1). Lemma 5.1, applied to R/JR/J, proves (i).

We next verify that JJ is a reduction of ms\mathfrak m^s. Since G/(f1∗,…,fd∗)G/(f_1^*,\ldots,f_d^*) has finite length, for all sufficiently large nn one has Gn=∑ifi∗Gn−sG_n=\sum_i f_i^*G_{n-s}. Lifting to RR gives mn=Jmn−s+mn+1\mathfrak m^n=J\mathfrak m^{n-s}+\mathfrak m^{n+1}. Nakayama’s lemma implies mn=Jmn−s\mathfrak m^n=J\mathfrak m^{n-s}. Taking n=qsn=qs for large qq, we obtain (ms)q=J(ms)q−1(\mathfrak m^s)^q=J(\mathfrak m^s)^{q-1}, as required. Consequently J‾=ms‾\overline{J}=\overline{\mathfrak m^s}. Let t∗:=r+(d−1)(s−1)t_*:=r+(d-1)(s-1). Formula (i) gives (J:mt∗)=J+ms⊆J‾(J:\mathfrak m^{t_*})=J+\mathfrak m^s\subseteq\overline{J}. At the next exponent the same formula gives (J:mt∗+1)=J+ms−1(J:\mathfrak m^{t_*+1})=J+\mathfrak m^{s-1}, which is not contained in J‾\overline{J} by Lemma 5.2. Thus g(J)=t∗g(J)=t_*. Finally r≥0r\geq0. If RR is nonregular, then r≥1r\geq1: indeed, r=0r=0 would imply e(R)=h(1)=1e(R)=h(1)=1, forcing the Cohen–Macaulay local ring RR to be regular. □

Corollary 5.4. Under the tangent-cone hypotheses of Theorem 5.3, if the residue field kk is infinite, the dd-tuples (f1∗,…,fd∗)∈(Gs)d(f_1^*,\ldots,f_d^*)\in(G_s)^d that satisfy the theorem form a nonempty Zariski-open subset. Thus the exact formula 5.3(ii) holds for a general dd-generated parameter ideal of order ss.

Proof. A standard graded Cohen–Macaulay algebra over an infinite field admits a linear homogeneous system of parameters ℓ1,…,ℓd\ell_1,\ldots,\ell_d. Their ss-th powers are a homogeneous system of parameters of degree ss, so the asserted locus is nonempty. The homogeneous-system-of-parameters condition is open in (Gs)d(G_s)^d (equivalently, the condition that the quotient have dimension zero is open). Theorem 5.3 applies to every lift of such a tuple. □

Proposition 5.5. Suppose (V,π)(V,\pi) is a complete mixed-characteristic discrete valuation ring. Let (A,n)(A,\mathfrak n) be any regular local ring of dimension d≥2d\geq2, with regular system of parameters u1:=π,…,udu_1:=\pi,\ldots,u_d. For s≥2s\geq2, let Q:=(u1s,…,uds)Q:=(u_1^s,\ldots,u_d^s). Then Q‾=ns\overline{Q}=\mathfrak n^s and g(Q)=(d−1)(s−1)g(Q)=(d-1)(s-1).

Proof. The associated graded ring of A/QA/Q is k[U1,…,Ud]/(U1s,…,Uds)k[U_1,\ldots,U_d]/(U_1^s,\ldots,U_d^s), an Artinian graded complete intersection of socle degree L=d(s−1)L=d(s-1). The local ring A/QA/Q is Gorenstein. Lemma 5.1 gives (Q:nt)=Q+nd(s−1)+1−t(Q:\mathfrak n^t)=Q+\mathfrak n^{d(s-1)+1-t} for all 0≤t≤d(s−1)+10\leq t\leq d(s-1)+1. Every degree-ss parameter monomial uαu^\alpha is integral over QQ, since (uα)s∈Qs(u^\alpha)^s\in Q^s. Hence ns\mathfrak n^s is integral over QQ. Also ns\mathfrak n^s is integrally closed: the n\mathfrak n-adic order is multiplicative because gr⁡nA\operatorname{gr}_{\mathfrak n}A is a polynomial domain, and taking orders in an integral equation excludes elements of order less than ss. Thus Q‾=ns\overline{Q}=\mathfrak n^s. For t0=(d−1)(s−1)t_0=(d-1)(s-1) the displayed colon is ns\mathfrak n^s. For t=t0+1t=t_0+1 it is Q+ns−1Q+\mathfrak n^{s-1} and contains u1s−1∉Q‾u_1^{s-1}\notin\overline{Q}. This proves the exact value of g(Q)g(Q). □

Theorem 1.1 holds for these parameter ideals in every residue characteristic. For e=2e=2, the exponent (d−1)(s−1)+1(d-1)(s-1)+1 is best possible even among singular mixed-characteristic Cohen–Macaulay domains:

Example 5.6. Let AA be either k[[u1,…,ud]]k[[u_1,\ldots,u_d]] over any field, or V[[u2,…,ud]]V[[u_2,\ldots,u_d]] over a complete mixed-characteristic discrete valuation ring (V,π)(V,\pi), with u1=πu_1=\pi. Fix d,s,e≥2d,s,e\geq2. Set (R,m):=A[[z]]/(ze−u1e−1u2)(R,\mathfrak m):=A[[z]]/(z^e-u_1^{e-1}u_2) and J:=(u1s,…,uds)RJ:=(u_1^s,\ldots,u_d^s)R. Then RR is a singular complete Cohen–Macaulay domain, dim⁡R=d\dim R=d, J‾=ms\overline{J}=\mathfrak m^s and g(J)=(d−1)(s−1)+e−1g(J)=(d-1)(s-1)+e-1.

Proof. The ring is a hypersurface of dimension dd and embedding dimension d+1d+1, so it is Cohen–Macaulay and nonregular. Its associated graded ring is

G:=gr⁡m(R)=k[U1,…,Ud,Z]/(Ze−U1e−1U2).G:=\operatorname{gr}_{\mathfrak m}(R)=k[U_1,\ldots,U_d,Z]/(Z^e-U_1^{e-1}U_2).

The displayed binomial is irreducible in every characteristic: view it as a primitive polynomial of degree one in U2U_2 over the unique factorization domain k[U1,U3,…,Ud,Z]k[U_1,U_3,\ldots,U_d,Z]. Consequently, GG is a domain. Separatedness of the m\mathfrak m-adic filtration shows that RR is a domain, and its m\mathfrak m-adic order is multiplicative. The ring GG is free over k[U1,…,Ud]k[U_1,\ldots,U_d], with basis {1,Z,…,Ze−1}\{1,Z,\ldots,Z^{e-1}\}. Thus the initial forms U1s,…,UdsU_1^s,\ldots,U_d^s form a regular sequence in GG, and

gr⁡m(R/J)=k[U1,…,Ud,Z]/(U1s,…,Uds,Ze−U1e−1U2).(1)\operatorname{gr}_{\mathfrak m}(R/J)=k[U_1,\ldots,U_d,Z]/(U_1^s,\ldots,U_d^s,Z^e-U_1^{e-1}U_2). \tag*{(1)}

This is an artinian graded complete intersection with Hilbert series (1+t+⋯+ts−1)d(1+t+⋯+te−1)(1+t+\cdots+t^{s-1})^{d}(1+t+\cdots+t^{e-1}) and socle degree L=d(s−1)+e−1L=d(s-1)+e-1. The local ring R/JR/J is also artinian Gorenstein. It follows from Lemma 5.1 that

(∗):(J:mt)=J+md(s−1)+e−t(0≤t≤d(s−1)+e).(\ast):\quad(J:\mathfrak{m}^{t})=J+\mathfrak{m}^{d(s-1)+e-t}\qquad(0\leq t\leq d(s-1)+e).

Write n=(u1,…,ud)A\mathfrak{n}=(u_{1},\ldots,u_{d})A. The equation for zz shows that zz is integral over nR\mathfrak{n}R, so m\mathfrak{m} is integral over nR\mathfrak{n}R. Also nsR\mathfrak{n}^{s}R is integral over JJ, by the same monomial equations as in Proposition 5.5. Powers and transitivity of integral dependence therefore show that ms\mathfrak{m}^{s} is integral over JJ. Since GG is a domain, the order argument used above shows that ms\mathfrak{m}^{s} is integrally closed. Hence J‾=ms\overline{J}=\mathfrak{m}^{s}. For g=(d−1)(s−1)+e−1g=(d-1)(s-1)+e-1, formula (∗)(\ast) gives (J:mg)=ms(J:\mathfrak{m}^{g})=\mathfrak{m}^{s}. At g+1g+1 it gives J+ms−1J+\mathfrak{m}^{s-1}, which contains u1s−1∉Ju_{1}^{s-1}\notin J. The claimed Goto number follows. □\square

The singular bound in Theorem 1.1 is sharp in mixed characteristic even for non-Gorenstein domains, in every dimension d≥2d\geq2:

Example 5.7. Let (V,π)(V,\pi) be a complete discrete valuation ring of mixed characteristic, let A:=V[[x2,…,xd]]A:=V[[x_{2},\ldots,x_{d}]] and (R,m):=A[[u,v]](u2−πv, uv−πx2, v2−ux2)(R,\mathfrak{m}):=\dfrac{A[[u,v]]}{(u^{2}-\pi v,\,uv-\pi x_{2},\,v^{2}-ux_{2})} and J:=(πs,x2s,…,xds)RJ:=(\pi^{s},x_{2}^{s},\ldots,x_{d}^{s})R. Then RR is a complete, Cohen–Macaulay domain of dimension dd, but RR is not Gorenstein. For these parameter ideals, J‾=ms\overline{J}=\mathfrak{m}^{s} and g(J)=(d−1)(s−1)+1g(J)=(d-1)(s-1)+1.

Proof. The three relations reduce all monomials in u,vu,v to AA-linear combinations of 1,u,v1,u,v. To prove independence, consider the finite AA-algebra S=A[a,b]/(a3−π,b3−x2)S=A[a,b]/(a^{3}-\pi,b^{3}-x_{2}). It is a complete regular local ring of dimension dd, with regular parameters a,b,x3,…,xda,b,x_{3},\ldots,x_{d}; in particular it is a domain. As an AA-module it is free with basis aibja^{i}b^{j} for 0≤i,j<30\leq i,j<3. The substitutions u↦a2bu\mapsto a^{2}b, v↦ab2v\mapsto ab^{2} respect the displayed relations, and 1,a2b,ab21,a^{2}b,ab^{2} are AA-linearly independent. Consequently RR is a domain and is free over the regular local ring AA with basis 1,u,v1,u,v. In particular RR is Cohen–Macaulay of dimension dd. The relations are homogeneous of degree two for the maximal-ideal filtration, and the displayed free basis is compatible with that filtration. Writing P,X2,…,Xd,U,VP,X_{2},\ldots,X_{d},U,V for initial forms, we obtain

G=gr⁡m(R)≅k[P,X2,…,Xd,U,V](U2−PV, UV−PX2, V2−UX2).G=\operatorname{gr}_{\mathfrak{m}}(R)\cong\frac{k[P,X_{2},\ldots,X_{d},U,V]}{(U^{2}-PV,\,UV-PX_{2},\,V^{2}-UX_{2})}.

It is the homogeneous coordinate ring of the rational normal cubic curve, with the extra polynomial variables X3,…,XdX_{3},\ldots,X_{d}. In particular GG is a standard Gorenstein-free (non-Gorenstein), Cohen–Macaulay level domain, and as a module over k[P,X2,…,Xd]k[P,X_{2},\ldots,X_{d}] it has homogeneous basis 1,U,V1,U,V of degrees 0,1,10,1,1. More explicitly, the quotient of GG by the linear regular sequence P,X2,…,XdP,X_{2},\ldots,X_{d} is k[U,V]/(U,V)2k[U,V]/(U,V)^{2}, with socle concentrated in degree one; this verifies levelness. Thus Hilb⁡G(t)=1+2t+t2(1−t)d\operatorname{Hilb}_{G}(t)=\dfrac{1+2t+t^{2}}{(1-t)^{d}}, and so r=1r=1. The initial forms Ps,X2s,…,XdsP^{s},X_{2}^{s},\ldots,X_{d}^{s} are a homogeneous system of parameters, so Theorem 5.3 gives g(J)=(d−1)(s−1)+1g(J)=(d-1)(s-1)+1 and J‾=ms\overline{J}=\mathfrak{m}^{s}. As GG is a domain, the m\mathfrak{m}-adic order on RR is multiplicative, so each ms\mathfrak{m}^{s} is integrally closed; hence J‾=ms\overline{J}=\mathfrak{m}^{s}. Finally, R/(π,x2,…,xd)RR/(\pi,x_{2},\ldots,x_{d})R is k[u,v]/(u,v)2k[u,v]/(u,v)^{2}, whose socle has dimension two. Therefore RR has Cohen–Macaulay type two and is not Gorenstein. Its embedding dimension is d+2>dd+2>d, so it is also nonregular.

Additional investigations

A non containment τ(R,mt)⊈mt+a(gr⁡mR)+1\tau(R,\mathfrak{m}^{t})\not\subseteq\mathfrak{m}^{t+a(\operatorname{gr}_{\mathfrak{m}}R)+1}

In this subsection, τ\tau denotes the classical positive-characteristic Hara–Yoshida test ideal used in [8]. They asked:

Question 6.1. ([8], Question 5.8.) When τ(R,mt)⊆mt+a(gr⁡mR)+1\tau(R,\mathfrak{m}^{t})\subseteq\mathfrak{m}^{t+a(\operatorname{gr}_{\mathfrak{m}}R)+1}?

Normality is not assumed in the question. The following example disproves the containment, even for a complete Gorenstein domain with Cohen–Macaulay associated graded ring.

Proposition 6.2. Let k=Fpk=\mathbb{F}_{p} for any prime pp, and set (R,m):=k[[u5,u6,u9]](R,\mathfrak{m}):=k[[u^{5},u^{6},u^{9}]]. Then RR is a one-dimensional complete Gorenstein domain, G:=gr⁡m(R)G:=\operatorname{gr}_{\mathfrak{m}}(R) is Cohen–Macaulay, and Hilb⁡G(z)=1+2z+z2+z31−z\operatorname{Hilb}_{G}(z)=\dfrac{1+2z+z^{2}+z^{3}}{1-z}, a(G)=2a(G)=2. For every integer t≥0t\geq0, τ(R,mt)=u5t+14k[[u]]\tau(R,\mathfrak{m}^{t})=u^{5t+14}k[[u]], and u5t+14∉mt+3u^{5t+14}\notin\mathfrak{m}^{t+3}. In particular, u19∈τ(R,m)∖m4u^{19}\in\tau(R,\mathfrak{m})\setminus\mathfrak{m}^{4}. Proof. Let S:=k[[u]]S := k[[u]]. Let x:=u5x := u^{5}, y:=u6y := u^{6}, and z:=u9z := u^{9}. The numerical semigroup H:=⟨5,6,9⟩H := \langle5,6,9\rangle. Recall that if HH is a numerical semigroup and n∈H∖{0}n \in H \setminus\{0\}, the Apéry set of HH with respect to nn is defined by Ap⁡(H,n):={h∈H∣h−n∉H}\operatorname{Ap}(H,n) := \{h \in H \mid h-n \notin H\}. Equivalently, it consists of the least elements of HH in each residue class modulo nn. Since H=⟨5,6,9⟩H = \langle5,6,9\rangle, we have Ap⁡(H,5)={0,6,12,18,9}\operatorname{Ap}(H,5) = \{0,6,12,18,9\}, where the displayed elements are the least elements of HH in the residue classes 0,1,2,3,40,1,2,3,4, respectively. In residue classes 0,1,2,3,40,1,2,3,4 modulo 55, these are the least members of HH. Consequently RR is free of rank 55 over k[[x]]k[[x]], with basis 1,y,y2,y3,z1,y,y^{2},y^{3},z. There is a presentation R≅k[[X,Y,Z]]/(X3−YZ,Z2−Y3)R \cong k[[X,Y,Z]]/(X^{3}-YZ,Z^{2}-Y^{3}). Here is a verification that this presentation has no extra component. The quotient on the right has dimension one and is a complete intersection: modulo XX its quotient is k[[Y,Z]]/(YZ,Z2−Y3)k[[Y,Z]]/(YZ,Z^{2}-Y^{3}), of length 55. It is therefore Cohen–Macaulay, XX is a nonzerodivisor, and it is free of rank 55 over k[[X]]k[[X]]. Its natural surjection onto RR, also free of rank 55, is an isomorphism. In particular, RR is Gorenstein. For each v∈Hv \in H, let ord⁡m(uv)=max⁡{r≥0∣uv∈mr}\operatorname{ord}_{\mathfrak m}(u^{v}) = \max\{r \ge0 \mid u^{v} \in\mathfrak m^{r}\}. Recall that Ap⁡(H,5)={0,6,12,18,9}\operatorname{Ap}(H,5) = \{0,6,12,18,9\}. Every v∈Hv \in H has a unique expression

v=5q+w,q≥0,w∈Ap⁡(H,5).v = 5q+w,\qquad q \ge0,\quad w \in\operatorname{Ap}(H,5).

The m\mathfrak m-adic orders of the corresponding Apéry monomials are

ww00661212181899
ord⁡m(uw)\operatorname{ord}_{\mathfrak m}(u^{w})0011222211

Indeed, these orders follow directly from

u6=y,u12=y2,u18=z2,u9=z,u^{6}=y,\qquad u^{12}=y^{2},\qquad u^{18}=z^{2},\qquad u^{9}=z,

and the fact that u18∉m3u^{18} \notin\mathfrak m^{3}. We claim that ord⁡m(u5q+w)=q+ord⁡m(uw)\operatorname{ord}_{\mathfrak m}(u^{5q+w}) = q+\operatorname{ord}_{\mathfrak m}(u^{w}). To see this, multiplication by x=u5x=u^{5} increases the order by at least one. Conversely, the valuation of every element of mr\mathfrak m^{r} is at least 5r5r, and the displayed Apéry orders give the required upper bound. Thus equality follows. Every element f∈Rf \in R has a unique expression

f=∑w∈Ap⁡(H,5)uwfw(x),fw(x)∈k[[x]].f=\sum_{w\in\operatorname{Ap}(H,5)}u^{w}f_{w}(x),\qquad f_{w}(x)\in k[[x]].

Since the five summands have distinct valuations modulo 55, there is no cancellation between their terms of lowest valuation. It follows that ord⁡m(xf)=ord⁡m(f)+1\operatorname{ord}_{\mathfrak m}(xf)=\operatorname{ord}_{\mathfrak m}(f)+1 for every nonzero f∈Rf \in R. Consequently, multiplication by x∗x^{*} is injective on G=gr⁡m(R)G=\operatorname{gr}_{\mathfrak m}(R). Since dim⁡G=1\dim G=1, we conclude that GG is Cohen–Macaulay. Since x∗x^{*} is a homogeneous nonzerodivisor of degree one on GG, we have Hilb⁡G/(x∗)(z)=(1−z)Hilb⁡G(z)\operatorname{Hilb}_{G/(x^{*})}(z)=(1-z)\operatorname{Hilb}_{G}(z). Moreover, R/(x)≅k[[Y,Z]]/(YZ,Z2−Y3)R/(x)\cong k[[Y,Z]]/(YZ,Z^{2}-Y^{3}), whose kk-basis is {1,y,y2,y3,z}\{1,y,y^{2},y^{3},z\}. These elements have m/(x)\mathfrak m/(x)-adic orders 0,1,2,3,10,1,2,3,1, respectively. Hence Hilb⁡G/(x∗)(z)=1+2z+z2+z3\operatorname{Hilb}_{G/(x^{*})}(z)=1+2z+z^{2}+z^{3}. It follows that Hilb⁡G(z)=1+2z+z2+z31−z=hG(z)1−z\operatorname{Hilb}_{G}(z)=\frac{1+2z+z^{2}+z^{3}}{1-z}=\frac{h_{G}(z)}{1-z}. Recall that if AA is a standard graded Cohen–Macaulay ring of dimension dd, its aa-invariant is defined by a(A)=deg⁡hA(z)−da(A)=\deg h_{A}(z)-d. In particular, since GG is one-dimensional and Cohen–Macaulay, its aa-invariant is a(G)=deg⁡(1+2z+z2+z3)−1=2a(G)=\deg(1+2z+z^{2}+z^{3})-1=2.

The gaps of HH are 1,2,3,4,7,8,131,2,3,4,7,8,13. Thus SS is the normalization of RR and its conductor is

c:=(R:S)=u14S=(u14,u15,u16,u17,u18)R.\mathfrak c:=(R:S)=u^{14}S=(u^{14},u^{15},u^{16},u^{17},u^{18})R.

The one-dimensional test-ideal/conductor formula gives τ(R)=c\tau(R)=\mathfrak c; see [9 Theorem 3.3]. For completeness, this formula has an elementary verification here. For every nonzero ideal I⊆RI\subseteq R, the valuation criterion and the conductor give I∗=IS∩RI^{*}=IS\cap R. For the nontrivial inclusion, choose f∈If\in I of least valuation. If w∈IS∩Rw\in IS\cap R, then w/f∈Sw/f\in S; for a fixed nonzero c∈cc\in\mathfrak c, cwq=cf(w/f)q∈I[q]cwq=c f(w/f)q\in I[q] for every q=peq=p^{e}. Conversely, a tight-closure equation implies v(w)≥v(I)v(w)\ge v(I) by letting qq tend to infinity. It follows that cI∗⊆I\mathfrak c I^{*}\subseteq I for every II, so c⊆τ(R)\mathfrak c\subseteq\tau(R). Taking I=uNRI=u^{N}R with N≥14N\ge14 gives I∗=uNSI^{*}=u^{N}S, and τ(R)I∗⊆I\tau(R)I^{*}\subseteq I forces τ(R)S⊆R\tau(R)S\subseteq R. Hence τ(R)=c\tau(R)=\mathfrak c.

The ideal (x)(x) is a reduction of m\mathfrak m. Indeed, using the relations {yz=x3,z2=y3}\{yz=x^{3},z^{2}=y^{3}\} we can verify that every monomial of degree four in x,y,zx,y,z belongs to xm3x\mathfrak m^{3}. For the monomials not involving xx, we have

y4=x⋅xy3,y3z=x3y2,y2z2=x⋅xy4,yz3=x3z2,z4=x⋅xy5.y^{4}=x\cdot xy^{3},\qquad y^{3}z=x^{3}y^{2},\qquad y^{2}z^{2}=x\cdot xy^{4},\qquad yz^{3}=x^{3}z^{2},\qquad z^{4}=x\cdot xy^{5}.

Consequently, m4⊆xm3\mathfrak{m}^{4} \subseteq x\mathfrak{m}^{3}. The reverse inclusion is immediate, and hence m4=xm3\mathfrak{m}^{4}=x\mathfrak{m}^{3}. Therefore, (x)(x) is a reduction of m\mathfrak{m}. Reduction invariance and the principal multiplication formula for test ideals [6 Proposition 1.11(1)–(2)] therefore give, for every integer t≥0t \geq0,

τ(R,mt)=τ(R,(x)t)=xtτ(R)=u5t+14S.\tau(R,\mathfrak{m}^{t})=\tau(R,(x)^{t})=x^{t}\tau(R)=u^{5t+14}S.

Alternatively, just the containment xtτ(R)⊆τ(R,mt)x^{t}\tau(R)\subseteq\tau(R,\mathfrak{m}^{t}) suffices for the counterexample; it follows from [8 Lemma 2.6(1)]. Since m=(u5,u6,u9)\mathfrak{m}=(u^{5},u^{6},u^{9}), every generator of mt+3\mathfrak{m}^{t+3} is a product of t+3t+3 elements, each of which has valuation at least 5. Consequently, mt+3⊆u5(t+3)S=u5t+15S\mathfrak{m}^{t+3}\subseteq u^{5(t+3)}S=u^{5t+15}S. On the other hand, u5t+14∉u5t+15Su^{5t+14}\notin u^{5t+15}S. Therefore, u5t+14∉mt+3u^{5t+14}\notin\mathfrak{m}^{t+3}. Since u5t+14∈τ(R,mt)u^{5t+14}\in\tau(R,\mathfrak{m}^{t}), we conclude that τ(R,mt)⊈mt+3\tau(R,\mathfrak{m}^{t})\not\subseteq\mathfrak{m}^{t+3} for every t≥0t\geq0. In particular, taking t=1t=1, we obtain u19∈τ(R,m)∖m4u^{19}\in\tau(R,\mathfrak{m})\setminus\mathfrak{m}^{4}. This proves the claim. □

The proposed exponent in Question 6.1 is too large by the following:

Proposition 6.3. For every integer n≥2n\geq2, let (Rn,mn):=k[[u2n+1,u2n+2,u3n+3]](R_n,\mathfrak{m}_n):=k[[u^{2n+1},u^{2n+2},u^{3n+3}]]. Then RnR_n is a one-dimensional complete intersection, Gn:=gr⁡mn(Rn)G_n:=\operatorname{gr}_{\mathfrak{m}_n}(R_n) is Cohen–Macaulay, and

a(Gn)=n,Cn=(Rn:k[[u]])=u2n2+2n+2k[[u]].a(G_n)=n,\qquad\mathfrak{C}_n=(R_n:k[[u]])=u^{2n^2+2n+2}k[[u]].

For every integer t≥0t\geq0, τ(Rn,mnt)=uet+2n2+2n+2k[[u]]⊆mnt+n\tau(R_n,\mathfrak{m}_n^{t})=u^{et+2n^2+2n+2}k[[u]]\subseteq\mathfrak{m}_n^{t+n} and τ(Rn,mnt+n)⊈mnt+n+1\tau(R_n,\mathfrak{m}_n^{t+n})\not\subseteq\mathfrak{m}_n^{t+n+1}.

Proof. Recall that the previous result is the special case n=2n=2. Also, e(R)=e=2n+1e(R)=e=2n+1. Let b:=2n+2b:=2n+2 and c:=3n+3c:=3n+3. A presentation is given by Rn≅k[[X,Y,Z]]/(Xn+1−Yn−1Z,Z2−Y3)R_n\cong k[[X,Y,Z]]/(X^{n+1}-Y^{n-1}Z,Z^2-Y^3). Similar to Proposition 6.2, we use the following basis {1,Y,…,Yn+1,Z,YZ,…,Yn−2Z}\{1,Y,\ldots,Y^{n+1},Z,YZ,\ldots,Y^{n-2}Z\} over k[[ue]]k[[u^{e}]]. Its valuations form the Apéry set

{jb:0≤j≤n+1}∪{c+jb:0≤j≤n−2}.\{jb:0\leq j\leq n+1\}\cup\{c+jb:0\leq j\leq n-2\}.

These have all the distinct residues modulo ee. Their orders are respectively 0,1,…,n+10,1,\ldots,n+1 and 1,2,…,n−11,2,\ldots,n-1, and each equals the floor of its valuation divided by ee. As before, ord⁡mn(uv)=⌊v/e⌋\operatorname{ord}_{\mathfrak{m}_n}(u^{v})=\lfloor v/e\rfloor for v∈⟨e,b,c⟩v\in\langle e,b,c\rangle. Hence GnG_n is Cohen–Macaulay and its Hilbert numerator is

1+2z+⋯+2zn−1+zn+zn+1.1+2z+\cdots+2z^{n-1}+z^n+z^{n+1}.

It follows that a(Gn)=na(G_n)=n. The largest Apéry element is (n+1)b(n+1)b; the conductor exponent is therefore (n+1)b−e+1=2n2+2n+2(n+1)b-e+1=2n^2+2n+2. The conductor formula and the principal reduction (ue)(u^e) give the stated test ideal, exactly as above. Finally, 2n2+2n+2=en+(n+2)2n^2+2n+2=en+(n+2) with 0<n+2<e0<n+2<e. The first element of that test ideal has order t+nt+n, and all of its elements have order at least t+nt+n. □

J(R,Jd−1)J(R,J^{d-1}) for monomial parameter ideals

Question 6.4. (See [8 Question 8.6]). When J⊆τ(R,Jd−1)J\subseteq\tau(R,J^{d-1})?

The next theorem gives a complete answer for monomial parameter ideals in regular parameters. The same answer holds for the mixed-characteristic test ideal used earlier in this note.

Proposition 6.5. Let (R,m)(R,\mathfrak{m}) be a complete regular ring of dimension d≥2d\geq2 and positive residue characteristic, and let v1,…,vdv_1,\ldots,v_d be a regular system of parameters. Let Q:=(v1a1,…,vdad)Q:=(v_1^{a_1},\ldots,v_d^{a_d}) with ai≥1a_i\geq1. In equal characteristic, assume also that RR is FF-finite. Then

τ(R,Qd−1)=(vβ:β∈Z≥0d,∑i=1dβi+1ai>d−1).\tau(R,Q^{d-1})=\left(v^{\boldsymbol{\beta}}:\boldsymbol{\beta}\in\mathbb{Z}_{\geq0}^{d},\sum_{i=1}^{d}\frac{\beta_i+1}{a_i}>d-1\right).

In particular, $Q\subseteq\tau(R,Q^{d-1})\quad\Longleftrightarrow\quad\sum_{i=1}^{d}\frac{1}{a_i}>d-2.

Equation not extracted

Proof. The Newton polyhedron of QQ is P(Q)={x∈R≥0d:∑ixiai≥1}P(Q)=\{x\in\mathbb{R}_{\geq0}^{d}:\sum_i\frac{x_i}{a_i}\geq1\}. The monomial test-ideal formula gives (*); see [6 Theorem 4.8], [8 Example 2.8], and [2 Proposition 8.21]. For the generator vjajv_j^{a_j}, the strict inequality in that formula becomes 1+∑i1ai>d−11+\sum_i\frac{1}{a_i}>d-1. It is independent of jj, and is equivalent to the boxed condition. Monomial membership in ideals generated by monomials in a regular system of parameters is determined by divisibility of their exponent vectors, so this checks every generator of QQ. □

Corollary 6.6. If all entries with ai=1a_i=1 deleted, and arrange the remaining rr entries in increasing order, then the containment in Proposition 6.5 holds exactly in the following cases:

(1) r≤2r\leq2, with arbitrary remaining exponents;

(2) r=3r=3 and the triple is one of {(2,2,c),(2,3,3),(2,3,4),(2,3,5):c≥2}\{(2,2,c),(2,3,3),(2,3,4),(2,3,5):c\geq2\}

Also, it never holds if r≥4r\geq4.

Proof. After removing the entries equal to 1, the criterion is ∑j=1r1/aj>r−2\sum_{j=1}^{r}1/a_j>r-2. This is automatic for r≤2r\leq2. For r≥4r\geq4, the left side is at most r/2≤r−2r/2\leq r-2. If r=3r=3 and 2≤a≤b≤c2\leq a\leq b\leq c, the inequality forces a=2a=2. If b=2b=2, it always holds; if b=3b=3, it forces c=3,4,5c=3,4,5; and if b≥4b\geq4, it fails. □

Corollary 6.7. Adopt the notation from subsection 3.3, and fix a1,…,ad≥1a_1,\ldots,a_d\geq1. Set Qi:=(pa1X2a2,…,Xdad)RiQ_i:=(p^{a_1}X_2^{a_2},\ldots,X_d^{a_d})R_i, QD:=ulimQiQ_D:=\mathop{\mathrm{ulim}} Q_i, and Q♯:=QDR♯Q^\sharp:=Q_D R^\sharp. Then the following conditions are equivalent:

QD⊆TU(Qd−1),Q♯⊆τ(R♯,(Q♯)d−1),∑i=1d1ai>d−2.Q_D\subseteq T_U(Q^{d-1}),\qquad Q^\sharp\subseteq\tau(R^\sharp,(Q^\sharp)^{d-1}),\qquad\sum_{i=1}^{d}\frac{1}{a_i}>d-2.

Moreover, TU(Qd−1)R♯=τ(R♯,(Q♯)d−1)T_U(Q^{d-1})R^\sharp=\tau(R^\sharp,(Q^\sharp)^{d-1}).

Proof. In view of Theorem 3.12, we see the same finite monomial generators in every factor and in R♯R^\sharp. It remains to apply the membership test in Proposition 6.5. □

M. ASGHARZADEH

Email address: [email protected]

References

  1. [1]M. Aschenbrenner and H. Schoutens, Lefschetz extensions, tight closure, and big Cohen–Macaulay algebras, Israel J. Math. 161 (2007), 221–310.DOI
  2. [2]B. Bhatt, L. Ma, Z. Patakfalvi, K. Schwede, K. Tucker, J. Waldron, and J. Witaszek, with an appendix by R. Datta, Test ideals in mixed characteristic: a unified theory up to perturbation, arXiv:2401.00615v2arxiv.org/abs/2401.00615
  3. [3]A. Corso, S. Goto, Shiro; C. Huneke, C. Polini, B.Ulrich, Iterated socles and integral dependence in regular rings, Trans. Amer. Math. Soc. 370 (2018), no. 1, 53-72.arxiv.org/abs/1409.5481
  4. [4]C. Huneke and I. Swanson, Integral Closure of Ideals, Rings, and Modules, London Mathematical Society Lecture Note Series, vol. 336, Cambridge University Press, Cambridge, 2006.
  5. [5]W. Heinzer and I. Swanson, The Goto numbers of parameter ideals, J. Algebra 321 (2009), 152–166.arxiv.org/abs/0801.0685
  6. [6]N. Hara and K. Yoshida, A generalization of tight closure and multiplier ideals, Trans. Amer. Math. Soc. 355 (2003), 3143–3174.DOI
  7. [7]H.-J. Wang, Links of symbolic powers of prime ideals, Math. Z. 256 (2007), 749–756.
  8. [8]K.-i. Watanabe and K. Yoshida, A variant of Wang’s theorem, J. Algebra 369 (2012), 129–145.
  9. [9]J. C. Vassilev and A. N. Vraciu, When is tight closure determined by the test ideal?, J. Commut. Algebra 1 (2009), no. 3, 591–602.arxiv.org/abs/0809.1891

Paper details

vlnc:W_cv7m59ab
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