betaFeedback

WorksMathematics

Reduction numbers under flat extensions

Abstract

The reduction number of the maximal ideal of a localization (or even faithfully flat extension) need not increase under specialization. Nevertheless, for each fixed local ring the reduction numbers of the maximal ideals of all its localization have a finite common bound. A connection to (generalized) Loewy length is given.

Introduction

If II is an ideal and J⊆IJ \subseteq I, then JJ is called a reduction of II if Ia+1=JIaI^{a+1}=JI^a for some a∈Z≥0a \in\mathbb{Z}_{\geq0}. The least such aa is denoted by rJ(I)r_J(I). A minimal reduction is a reduction minimal under inclusion, and throughout this note

r(I):=min⁡{rJ(I):J is a minimal reduction of I}.r(I):=\min\{r_J(I): J\text{ is a minimal reduction of }I\}.

The restriction to minimal reductions matters: allowing every reduction in this minimum would give zero by taking J=IJ=I. In the local setting, it controls the Hilbert–Samuel function, the structure of the associated graded ring, and the Cohen–Macaulayness of blow-up algebras. There is an extensive literature on bounds for reduction numbers. In particular, Vasconcelos obtained general bounds for reduction numbers of ideals and multiplicity-dependent bounds for primary ideals. Related estimates have been developed using the Eakin–Sathaye theorem and, under additional hypotheses, the theory of Sally modules and associated graded rings. For more details, see the book of Vasconcelos [12]. A natural question, going back to Vasconcelos, is how this invariant behaves under localization, and whether it is uniformly bounded on the prime spectrum of a fixed ring.

Question 1.1 (Vasconcelos, see [12 Page 83]). Let RR be local and let p⊊q\mathfrak{p} \subsetneq\mathfrak{q} be prime ideals of RR.

(i) Is r(pRp)≤r(qRq)r(\mathfrak{p}R_{\mathfrak{p}}) \leq r(\mathfrak{q}R_{\mathfrak{q}})?

(ii) Can sup⁡{r(pRp):p∈Spec⁡(R)}\sup\{r(\mathfrak{p}R_{\mathfrak{p}}):\mathfrak{p}\in\operatorname{Spec}(R)\} be infinite?

Part (i) asks whether reduction numbers are monotone under specialization, i.e. whether passing from the larger prime q\mathfrak{q} to the smaller prime p\mathfrak{p} can only decrease the reduction number. Part (ii) asks whether, for a fixed Noetherian local ring, the reduction numbers of the maximal ideals of all its localizations admit a common finite bound, independently of the prime. Recall that a positive answer to (i) gives a positive answer to (ii), and not necessarily the converse.

Theorem 1.2. (i) The implication “p⊊q⟹r(pRp)≤r(qRq)\mathfrak{p} \subsetneq\mathfrak{q} \Longrightarrow r(\mathfrak{p}R_{\mathfrak{p}}) \leq r(\mathfrak{q}R_{\mathfrak{q}})” is not the case.

(ii) sup⁡p∈Spec⁡Rr(pRp)<∞\sup_{\mathfrak{p}\in\operatorname{Spec} R}r(\mathfrak{p}R_{\mathfrak{p}})<\infty.

Without the noetherian hypothesis, sup⁡p∈Spec⁡Rr(pRp)\sup_{\mathfrak{p}\in\operatorname{Spec} R}r(\mathfrak{p}R_{\mathfrak{p}}) can be infinite, even when dim⁡R=1\dim R=1, and every term in the supremum is finite (see § 5.3). Section 3 deals with the following question from the exercise [12 1.193] attributed to Hanes (see [3 Conjecture 3.1]):

Question 1.3. Let f:(R,m)→(S,n)f:(R,\mathfrak{m})\to(S,\mathfrak{n}) be a flat local morphism. Is r(m)≤r(n)r(\mathfrak{m})\leq r(\mathfrak{n})?

We present progress on the monotonicity property of the reduction number with respect to faithfully flat extensions, under additional Cohen–Macaulay hypotheses (see Proposition 3.4). In Proposition 3.5, we establish a positive answer under suitable graded assumptions. We note that this was recently proved by Pollitz and KC [11] via a different argument; our proof appears simpler and yields a sharp bound, although the assumptions in [11] are sharper than ours. On the negative side, we present Proposition 3.8. This is inspired by [4], and of course the motivational example from the book of Huneke-Swanson [6].

In Section 4 we present one-dimensional Cohen–Macaulay rings with r(m)=1r(\mathfrak{m}) = 1 and r(pRp)=Nr(\mathfrak{p}R_{\mathfrak{p}}) = N for every N≥2N \ge2. The proof of Theorem 1.2(ii) occupies Section 5 and combines three ingredients. The first is an elementary estimate, proved in Section 5.1, which bounds the number of generators of a power nh\mathfrak{n}^{h} of the maximal ideal of a dd-dimensional local ring in terms of hh, dd, and the colength of a parameter ideal; this estimate does not require the Cohen–Macaulay property. Combined with the Eakin–Sathaye theorem, it yields a local bound r(n)≤d! ℓA(A/Q)−1r(\mathfrak{n}) \le d!\,\ell_A(A/Q) - 1 for every parameter ideal QQ, valid whenever the residue field is infinite. The second ingredient, proved in Section 5.2, is a uniform Briançon–Skoda statement for complete reduced local rings, which supplies a single exponent CC such that nC⊆J\mathfrak{n}^{C} \subseteq J for every prime p\mathfrak{p} and every reduction JJ of pRp\mathfrak{p}R_{\mathfrak{p}}. This converts the local colength bound into a uniform bound on the colength ℓA(A/Q)\ell_A(A/Q), hence on r(pRp)r(\mathfrak{p}R_{\mathfrak{p}}). The third ingredient is a uniform embedding-dimension bound, obtained from the Cohen structure theorem and Lech’s inequality for flat local maps, which makes the resulting bound depend only on the fixed ring RR.

Recall from previous remarks that the reduction number is connected with the generalized Loewy length; our reference for this is [11] and the references therein. Pollitz and Nawaj KC [11] considered the following, known as the Loewy Length Conjecture with a contribution to an unpublished manuscript of Corso, Huneke, Polini and Ulrich:

Conjecture 1.4. For a local ring RR and a nonzero RR-module MM with finite projective dimension, the following inequality holds:

ℓℓR(M)≥gℓℓ(R):=min⁡{ℓℓR(R/(x)):x ranges over systems of parameters}.\ell\ell_R(M) \ge g\ell\ell(R) := \min\{\ell\ell_R(R/(x)) : x\text{ ranges over systems of parameters}\}.

Here, ℓℓR(−)\ell\ell_R(-) denotes the Loewy length. A conjectural connection between this and the reduction number was hinted to us by Pollitz and KC. It may be worth mentioning that

m(R):=min⁡{ℓR(R/(x)):x ranges over systems of parameters},\mathfrak{m}(R) := \min\{\ell_R(R/(x)) : x\text{ ranges over systems of parameters}\},

is also considered in the literature; see [1] and references therein, we call it Matsumura’s number. In the final section we show that Conjecture 1.4 needs some additional assumptions, for instance that the residue field is infinite. This follows by an analysis of a ring given by Hashimoto and Shida [4], Example 3.2.

Conveniences

All rings are commutative with identity. Unless explicitly stated otherwise, rings are Noetherian. For a local ring (R,m)(R,\mathfrak{m}), write μR(M)\mu_R(M) for the minimal number of generators of a finite module MM, and ℓR(M)\ell_R(M) for its length when finite. For a finite-length module NN, write ℓℓR(N)\ell\ell_R(N) for the least j≥0j \ge0 with mjN=0\mathfrak{m}^{j}N = 0.

Remark 2.1. The convention Ia+1=JIaI^{a+1} = JI^a, taking the minimum over minimal reductions, is essential to the formulation of the problem. With the larger class of all reductions the minimum would always be zero, since J=IJ = I is a reduction of II, and both questions would trivialize. Minimal reductions exist for finitely generated ideals in Noetherian local rings regardless of the residue field; for completeness we recall the argument here. Indeed, minimal reductions exist for finitely generated ideals in Noetherian local rings, also when the residue field is finite. For (S,n)(S,\mathfrak{n}) and an ideal II, choose a reduction JJ for which dim⁡S(S/(n+I)/n)\dim_S(S/(\mathfrak{n} + I)/\mathfrak{n}) is as small as possible. Lift a basis to y1,…,ys∈Jy_1,\ldots,y_s \in J and put Q=(y1,…,ys)Q = (y_1,\ldots,y_s). Since J⊆Q+nIJ \subseteq Q + \mathfrak{n}I, a reduction equality for JJ, followed by Nakayama’s lemma applied to Ia+1/QIaI^{a+1}/QI^a, shows that QQ is a reduction. Independence of the chosen initial classes gives Q∩nI=nQQ \cap\mathfrak{n}I = \mathfrak{n}Q. If K⊆QK \subseteq Q is a reduction, minimality of ss gives K+nI=Q+nIK + \mathfrak{n}I = Q + \mathfrak{n}I, hence Q=K+nQQ = K + \mathfrak{n}Q and K=QK = Q by Nakayama. Thus QQ is minimal. In particular, the reduction number is a finite integer for the ideals considered in Theorem 1.2(ii). The books of Matsumura [10], Huneke–Swanson [6] and Vasconcelos [12] serve as our standard references.

Dealing with an exercise

Some positive side

We start with:

Lemma 3.1. Let f:(R,m)→(S,n)f:(R,\mathbf{m})\to(S,\mathbf{n}) be flat and local morphism of artinian rings. Then r(m)≤r(n)r(\mathbf{m})\leq r(\mathbf{n}). The equality holds, if in addition ff is unramified (or at least n=mS\mathbf{n}=\mathbf{m}S).

Proof. Let rr be an integer. The condition mr=0\mathbf{m}^{r}=0 is equivalent with mr⊗RS=0\mathbf{m}^{r}\otimes_{R}S=0. But, mr⊗RS≅mrS⊆n\mathbf{m}^{r}\otimes_{R}S\cong\mathbf{m}^{r}S\subseteq\mathbf{n}. Now, let s:=r(n)s:=r(\mathbf{n}). Since the zero ideal is a minimal reduction of n\mathbf{n}, we know ns+1=0\mathbf{n}^{s+1}=0. Let r:=s+1r:=s+1. This, in turns, imply that mr=0\mathbf{m}^{r}=0, and consequently, r(m)≤s=r(n)r(\mathbf{m})\leq s=r(\mathbf{n}). Now, suppose n=mS\mathbf{n}=\mathbf{m}S. Then mr=0\mathbf{m}^{r}=0 iff nr=0\mathbf{n}^{r}=0, and so r(m)=r(n)r(\mathbf{m})=r(\mathbf{n}). □

The second assertion is not true in the higher dimension:

Example 3.2. Let R=F2[[X,Y]]/(XY(X+Y))R=\mathbb{F}_{2}[[X,Y]]/(XY(X+Y)) and S=F[[X,Y]]/(XY(X+Y))S=\mathbb{F}[[X,Y]]/(XY(X+Y)), where F\mathbb{F} denotes the algebraic closure of F2\mathbb{F}_{2}. Let m=(X,Y)R\mathbf{m}=(X,Y)R and n=(X,Y)S\mathbf{n}=(X,Y)S. The natural homomorphism R→SR\to S is faithfully flat, n=mS\mathbf{n}=\mathbf{m}S and r(m)≠r(n)r(\mathbf{m})\neq r(\mathbf{n}).

Proof. By [6 Example 8.3.2] (also see [4]), the ideal m\mathbf{m} is basic: it has no proper reductions. Consequently, m\mathbf{m} is its own minimal reduction, and, with our convention, r(m)=0r(\mathbf{m})=0. On the other hand, choose u∈Fu\in\mathbb{F} with u≠0,1u\neq0,1, and put J=(X+uY)SJ=(X+uY)S. The calculation in [6 Example 8.3.2] shows that Jn2=n3J\mathbf{n}^{2}=\mathbf{n}^{3}. Moreover, JJ is a minimal reduction of n\mathbf{n}. Since Gr⁡n(S)≅F[X,Y]/(XY(X+Y))\operatorname{Gr}_{\mathbf{n}}(S)\cong\mathbb{F}[X,Y]/(XY(X+Y)) and Gr⁡n(S)(X+uY)≅F[Y]/(Y3)\frac{\operatorname{Gr}_{\mathbf{n}}(S)}{(X+uY)}\cong\mathbb{F}[Y]/(Y^{3}), the quotient has a nonzero degree-two component and vanishes in all degrees greater than two. Hence 0<r(n)≤20<r(\mathbf{n})\leq2, as n\mathbf{n} is not basic. □

Remark 3.3. In general, the reduction number does not behave well with respect to a superficial element x∈Ix\in I; that is, the equality rJ/(x)(I/(x))=rJ(I)r_{J/(x)}(I/(x))=r_{J}(I) need not hold. However, when depth⁡Gr⁡(I)≥1\operatorname{depth}\operatorname{Gr}(I)\geq1, one has rJ/(x)(I/(x))=rJ(I)r_{J/(x)}(I/(x))=r_{J}(I), see [9 Lemma 2.14].

Proposition 3.4. Let f:(R,m)→(S,n)f:(R,\mathbf{m})\to(S,\mathbf{n}) be a flat local homomorphism of Cohen–Macaulay local rings of the same dimension, with infinite residue fields.

(i) If Gr⁡m(R)\operatorname{Gr}_{\mathbf{m}}(R) is Cohen–Macaulay, then r(m)≤r(n)r(\mathbf{m})\leq r(\mathbf{n}).

(ii) Assume in addition to part i) that n=mS\mathbf{n}=\mathbf{m}S. Then r(m)=r(n)r(\mathbf{m})=r(\mathbf{n}).

Proof. i) Reduction by regular sequence shows that r(m‾)≤r(m)r(\overline{\mathbf{m}})\leq r(\mathbf{m}), where R‾=R/(x1,…,xd)\overline{R}=R/(x_{1},\ldots,x_{d}) and m‾=mR‾\overline{\mathbf{m}}=\mathbf{m}\overline{R}, with x1,…,xdx_{1},\ldots,x_{d} a suitable superficial system of parameters for RR. Similarly, r(n‾)≤r(n)r(\overline{\mathbf{n}})\leq r(\mathbf{n}). Combining with Lemma 3.1 it is enough to show that r(m‾)=r(m)r(\overline{\mathbf{m}})=r(\mathbf{m}). This follows by the Cohen–Macaulay assumption of Gr⁡m(R)\operatorname{Gr}_{\mathbf{m}}(R) where we take reduction by xx so that x+I2∈I/I2=(Gr⁡m(R))1x+I^{2}\in I/I^{2}=(\operatorname{Gr}_{\mathbf{m}}(R))_{1} is (Gr⁡m(R))(\operatorname{Gr}_{\mathbf{m}}(R))-regular (see Remark 3.3).

ii) Since n:=mS\mathbf{n}:=\mathbf{m}S, and by flatness, we have

Gr⁡m(R)⊗RS≅⨁n(mn)⊗RS(mn+1)⊗RS≅⨁nnnSnn+1S=Gr⁡n(S)\operatorname{Gr}_{\mathbf{m}}(R)\otimes_{R}S\cong\bigoplus_{n}\frac{(\mathbf{m}^{n})\otimes_{R}S}{(\mathbf{m}^{n+1})\otimes_{R}S}\cong\bigoplus_{n}\frac{\mathbf{n}^{n}S}{\mathbf{n}^{n+1}S}=\operatorname{Gr}_{\mathbf{n}}(S)

This implies that Gr⁡n(S)\operatorname{Gr}_{\mathbf{n}}(S) is also Cohen–Macaulay. Then by Remark 3.3 r(n‾)=r(n)r(\overline{\mathbf{n}})=r(\mathbf{n}). By arguing as part i), we have r(m)=r(n)r(\mathbf{m})=r(\mathbf{n}). □

Proposition 3.5. Let R→SR\to S be a degree-preserving flat homomorphism of standard graded Cohen–Macaulay kk-algebras over an infinite field kk, where R0=S0=kR_{0}=S_{0}=k and dim⁡R=dim⁡S\dim R=\dim S. Let m=R+\mathbf{m}=R_{+} and n=S+\mathbf{n}=S_{+}.

(i) One has r(m)≤r(n)r(\mathbf{m})\leq r(\mathbf{n}).

(ii) Assume in addition to part i) that n=mS\mathbf{n}=\mathbf{m}S. Then r(m)=r(n)r(\mathbf{m})=r(\mathbf{n}).

Proof. i) Set F=S/mSF=S/\mathbf{m}S. Since R→SR\to S is flat and the two rings have the same dimension, FF is zero-dimensional. Hence FF is an Artinian graded kk-algebra.

Claim 3.6. One has HS(t)=HR(t)HF(t)H_S(t)=H_R(t)H_F(t).

Proof. Since F=S/mSF=S/mS is a finite-dimensional graded kk-vector space, choose a homogeneous kk-basis {uˉ1,…,uˉN}\{\bar{u}_1,\ldots,\bar{u}_N\} with deg⁡uˉi=ai\deg\bar{u}_i=a_i. Lift each uˉi\bar{u}_i to a homogeneous element ui∈Su_i\in S of the same degree. Because SS is finite and flat over RR, these lifts form a homogeneous RR-basis of SS. Thus, as graded RR-modules, S≅⨁i=1NR(−ai)S\cong\bigoplus_{i=1}^{N}R(-a_i), where R(−ai)R(-a_i) denotes the graded shift of RR. Taking Hilbert series gives

HS(t)=∑i=1NHR(−ai)(t)=∑i=1NtaiHR(t)=HR(t)∑i=1Ntai.H_S(t)=\sum_{i=1}^{N}H_{R(-a_i)}(t)=\sum_{i=1}^{N}t^{a_i}H_R(t)=H_R(t)\sum_{i=1}^{N}t^{a_i}.

The degrees of the homogeneous basis elements give HF(t)=∑i=1NtaiH_F(t)=\sum_{i=1}^{N}t^{a_i}. The claim follows.

As RR and SS are Cohen–Macaulay of the same dimension dd, we can write

HR(t)=hR(t)(1−t)d,HS(t)=hS(t)(1−t)d.H_R(t)=\frac{h_R(t)}{(1-t)^d},\qquad H_S(t)=\frac{h_S(t)}{(1-t)^d}.

where hR(t)h_R(t) and hS(t)h_S(t) are the respective hh-polynomials. It follows that hS(t)=hR(t)HF(t)h_S(t)=h_R(t)H_F(t). Since HF(t)H_F(t) is a nonzero polynomial with nonnegative coefficients, we obtain deg⁡hS(t)≥deg⁡hR(t)\deg h_S(t)\geq\deg h_R(t). Finally, for a standard graded Cohen–Macaulay algebra over an infinite field, the reduction number of the homogeneous maximal ideal equals the degree of its hh-polynomial. Thus

r(m)=deg⁡hR(t)≤deg⁡hS(t)=r(n).r(\mathfrak m)=\deg h_R(t)\leq\deg h_S(t)=r(\mathfrak n).

ii) Under this additional assumption, FF is just residue field and so hS(t)=hR(t)HF(t)=hR(t)h_S(t)=h_R(t)H_F(t)=h_R(t), and so the claim follows.

Claim 3.6.

Observation 3.7. One has r(m)=r(m^)r(\mathfrak m)=r(\widehat{\mathfrak m}).

Proof. For any ideal II, consider the special fiber ring FR(I)=⨁n≥0InmIn\mathcal{F}_R(I)=\bigoplus_{n\geq0}\frac{I^n}{\mathfrak m I^n}. Since completion is faithfully flat, we have a natural graded isomorphism FR(I)≅FR^(IR^)\mathcal{F}_R(I)\cong\mathcal{F}_{\widehat{R}}(I\widehat{R}). Recall from [6 Proposition 8.2.4] that an ideal J⊆IJ\subseteq I is a reduction of II if and only if its image J∗=J+mImI⊆FR(I)1J^*=\frac{J+\mathfrak m I}{\mathfrak m I}\subseteq\mathcal{F}_R(I)_1 satisfies FR(I)n+1=J∗FR(I)n\mathcal{F}_R(I)_{n+1}=J^*\mathcal{F}_R(I)_n for all sufficiently large nn. Moreover,

rJ(I)=min⁡{n≥0:FR(I)n+1=J∗FR(I)n}.r_J(I)=\min\{n\geq0:\mathcal{F}_R(I)_{n+1}=J^*\mathcal{F}_R(I)_n\}.

The point is that m^/m^2=m/m2\widehat{\mathfrak m}/\widehat{\mathfrak m}^{2}=\mathfrak m/\mathfrak m^{2}, i.e., any reduction of m^\widehat{\mathfrak m} is algebraic, that it is extended from an ideal of RR.

A negative side

The following example uses finite residue fields and non-Cohen–Macaulay rings.

Proposition 3.8. Let k:=F2k:=\mathbb F_2, K:=F4K:=\mathbb F_4. Let also R:=k[[x,y]](x4,xy(x+y))⊆S:=K[[x,y]](x4,xy(x+y))R:=\frac{k[[x,y]]}{(x^4,xy(x+y))}\subseteq S:=\frac{K[[x,y]]}{(x^4,xy(x+y))}. Then ff is flat and n=mS\mathfrak n=\mathfrak mS. Also, r(m)=3>2=r(n)r(\mathfrak m)=3>2=r(\mathfrak n).

Proof. For a coefficient field FF, the associated graded ring is AF=F[X,Y]/(X4,XY(X+Y))A_F=F[X,Y]/(X^4,XY(X+Y)). Recall that ff is local and finite free of rank 22, hence flat, with m=(x,y)\mathfrak m=(x,y) and n=mS\mathfrak n=\mathfrak mS. Let G:=Gr⁡m(AF)≅F[X,Y]/(X4,XY(X+Y))G:=\operatorname{Gr}_{\mathfrak m}(A_F)\cong F[X,Y]/(X^4,XY(X+Y)), see [10 Ex. 14.2(iii)]. Let JJ be a minimal reduction of m\mathfrak m. If R/mR/\mathfrak m were infinite, then every minimal reduction of m\mathfrak m is minimally generated by exactly the analytic spread of m\mathfrak m elements (see [6 Proposition 8.3.7]). Here, we have no such cardinality assumption. Then, we argue directly, i.e., there exists an integer r≥0r\geq0 such that mr+1=Jmr\mathfrak m^{r+1}=J\mathfrak m^r. Consider the quotient map π:AF→AF/(x)≅F[[y]]\pi:A_F\to A_F/(x)\cong F[[y]]. Since π(m)=(y)\pi(\mathfrak m)=(y), applying π\pi to the reduction equality gives (y)r+1=π(J)(y)r(y)^{r+1}=\pi(J)(y)^r. Consequently, π(J)=(y)\pi(J)=(y). Hence there exists f∈Jf\in J whose image modulo xx has a nonzero linear term in yy. In particular, the initial form of ff is f∗=aX+bYf^*=aX+bY with a,b∈Fa,b\in F.

Claim 3.9. One has b≠0b\ne0.

Proof. First, f∉(x)f \notin(x). Indeed, if f∈(x)f \in(x), then (f)⊆(x)(f) \subseteq(x), and a reduction would require mn+1=(f)mn\mathfrak{m}^{n+1}=(f)\mathfrak{m}^{n} for some nn. Reducing modulo xx would give (y)n+1=0(y)^{n+1}=0 in F[[y]]F[[y]], which is impossible. Hence f∉(x)f \notin(x), so the image g(y)g(y) of ff in F[[x,y]]/(x)≅F[[y]]F[[x,y]]/(x) \cong F[[y]] is nonzero. Next, g(y)g(y) must generate the maximal ideal (y)(y) of the discrete valuation ring F[[y]]F[[y]]. Indeed, the image of m\mathfrak{m} modulo xx is (y)(y), and a reduction of (y)(y) in a DVR is generated by an element of yy-adic valuation exactly 1. Hence g(y)g(y) has a nonzero linear term. Write f=a′x+b′y+terms of order≥2f=a'x+b'y+\text{terms of order}\geq2, so that f∗=a′X+b′Yf^{*}=a'X+b'Y. Reducing modulo xx, g(y)≡b′y+higher order(modx)g(y)\equiv b'y+\text{higher order}\pmod{x}. Since g(y)g(y) has nonzero linear term, we conclude b′≠0b'\ne0. Thus b≠0b\ne0 in f∗=aX+bYf^{*}=aX+bY.

Since b≠0b\ne0, the relation aX+bY=0aX+bY=0 expresses YY as a scalar multiple of XX. Therefore

G/(f∗)≅F[X,Y]/(X4,XY(X+Y),aX+bY)G/(f^{*})\cong F[X,Y]/(X^{4},XY(X+Y),aX+bY)

is finite-dimensional over FF, because it is a quotient of F[X]/(X4)F[X]/(X^{4}). Thus there exists an integer N≥0N\geq0 such that (G/(f∗))n+1=0(G/(f^{*}))_{n+1}=0 for all n≥Nn\geq N, i.e., Gn+1=f∗GnG_{n+1}=f^{*}G_n. By the definition of the associated graded ring, this implies mn+1=fmn+mn+2\mathfrak{m}^{n+1}=f\mathfrak{m}^{n}+\mathfrak{m}^{n+2}. Set M=mn+1/fmnM=\mathfrak{m}^{n+1}/f\mathfrak{m}^{n}. The preceding equality yields M=mMM=\mathfrak{m}M. Since MM is a finitely generated AFA_F-module and m\mathfrak{m} is the maximal ideal, Nakayama’s lemma gives M=0M=0. Hence mn+1=fmn\mathfrak{m}^{n+1}=f\mathfrak{m}^{n}. It follows that (f)(f) is a reduction of m\mathfrak{m}. Finally, f∈Jf\in J implies (f)⊆J(f)\subseteq J. Since JJ is a minimal reduction and (f)(f) is itself a reduction, we conclude that J=(f)J=(f). This proves that every minimal reduction of m\mathfrak{m} is principal.

Claim 3.10. r(f)(mF)=max⁡{j:(AF/(f∗))j≠0}r_{(f)}(\mathfrak{m}_F)=\max\{j:(A_F/(f^{*}))_j\ne0\}.

Proof. Let f∈mFf\in\mathfrak{m}_F be a principal reduction of mF\mathfrak{m}_F so that (f)(f) is a minimal reduction. Write f∗∈AFf^{*}\in A_F for the initial form of ff in the associated graded ring Gr⁡mF(A)=F[X,Y]/(X4,XY(X+Y))\operatorname{Gr}_{\mathfrak{m}_F}(A)=F[X,Y]/(X^{4},XY(X+Y)). For a minimal reduction J=(f)J=(f) of mF\mathfrak{m}_F, the reduction number is rJ(mF)=min⁡{n:mFn+1=JmFn}r_J(\mathfrak{m}_F)=\min\{n:\mathfrak{m}_F^{n+1}=J\mathfrak{m}_F^n\}. Passing to the associated graded ring, the condition mFn+1=JmFn\mathfrak{m}_F^{n+1}=J\mathfrak{m}_F^n is equivalent to the surjectivity of the multiplication map ×f∗:(AF)n−1→(AF)n\times f^{*}:(A_F)_{n-1}\to(A_F)_n, i.e. to the vanishing (AF/(f∗))n=0(A_F/(f^{*}))_n=0. Since ff is a reduction, AF/(f∗)A_F/(f^{*}) is a finite-dimensional artinian graded FF-algebra, so it vanishes in all sufficiently high degrees. Because f∗f^{*} is a nonzerodivisor of degree 1, the quotient (AF/(f∗))j(A_F/(f^{*}))_j is nonzero precisely for j≤rj\leq r, where rr is the largest index with nonzero component. Hence (AF/(f∗))j≠0(A_F/(f^{*}))_j\ne0 iff j≤r(f)(mF)j\leq r_{(f)}(\mathfrak{m}_F).

Let f∈m=(x,y)f\in\mathfrak{m}=(x,y) be such that (f)(f) is a minimal reduction of m\mathfrak{m}, and write f∗=aX+bYf^{*}=aX+bY for its initial form in AF=F[X,Y]/(X4,XY(X+Y))A_F=F[X,Y]/(X^{4},XY(X+Y)). Equivalently, b≠0b\ne0 says that ff is not contained in (x)+m2(x)+\mathfrak{m}^{2}: a reduction of m\mathfrak{m} must detect the yy-direction modulo xx, and the only way to do so at first order is to have a nonzero YY-coefficient in f∗f^{*}. Over F2F_2, the only possible initial forms are YY and X+YX+Y, and Ak/(Y)≅k[X]/(X4)≅Ak/(X+Y)A_k/(Y)\cong k[X]/(X^{4})\cong A_k/(X+Y). Consequently, r(m)=3r(\mathfrak{m})=3. Over KK, choose α∉F2\alpha\notin F_2. Then AK/(Y+αX)≅K[X]/(X4,α(1+α)X3)≅K[X]/(X3)A_K/(Y+\alpha X)\cong K[X]/(X^{4},\alpha(1+\alpha)X^{3})\cong K[X]/(X^{3}), because α(1+α)\alpha(1+\alpha) has inverse over KK. Thus r(m)≤2r(\mathfrak{m})\leq2. Equality holds because AKA_K has no quadratic relations: its degree-two component has dimension 3, while multiplication by one linear form from its degree-one component has rank at most 2.

Something is not monotone

Let kk be any field. Let us compute reduction number in our favorite non Cohen–Macaulay ring.

Example 4.1. Let R0:=k[[x,y]]/(x2,xy)R_0:=k[[x,y]]/(x^2,xy) with m=(x,y)R0\mathfrak{m}=(x,y)R_0. Since x2=xy=0x^2=xy=0, we have m2=(y2)=ym\mathfrak{m}^2=(y^2)=y\mathfrak{m}. Thus J=(y)J=(y) is a reduction of m\mathfrak{m}. Moreover, JJ is a minimal reduction, and rJ(m)=1r_J(\mathfrak{m})=1, since m≠J\mathfrak{m}\ne J but m2=Jm\mathfrak{m}^2=J\mathfrak{m}. Hence r(m)=1r(\mathfrak{m})=1. Now localize at the prime ideal p=(x)\mathfrak{p}=(x). Since y∉py\notin\mathfrak{p}, the element yy is a unit in (R0)p(R_0)_{\mathfrak{p}}. The relation xy=0xy=0 therefore implies x=0x=0 in (R0)p(R_0)_{\mathfrak{p}}. Consequently, p(R0)p=(0)\mathfrak{p}(R_0)_{\mathfrak{p}}=(0). Taking the zero ideal as a reduction of itself, we obtain rp(R0)p(p(R0)p)=0r_{\mathfrak{p}(R_0)_{\mathfrak{p}}}(\mathfrak{p}(R_0)_{\mathfrak{p}})=0. Thus r(m)=1r(\mathfrak{m})=1 and rp(R0)p(p(R0)p)=0r_{\mathfrak{p}(R_0)_{\mathfrak{p}}}(\mathfrak{p}(R_0)_{\mathfrak{p}})=0.

Our main ring in this section follows by adding a binomial relation y2−txy^2-tx to R0R_0. Namely, R:=k[[t,x,y]]/(x2,xy,y2−tx)R:=k[[t,x,y]]/(x^2,xy,y^2-tx) (and its generalization), and we want to compare the reduction number of localization at p:=(x,y)\mathfrak{p}:=(x,y) with m:=(t,x,y)\mathfrak{m}:=(t,x,y).

Proposition 4.2. The ring RR is a complete one-dimensional Cohen–Macaulay and local, p⊊m\mathfrak{p} \subsetneq\mathfrak{m}, and r⁡(Rp)=2>1=r⁡(m)\operatorname{r}(R_{\mathfrak{p}})=2>1=\operatorname{r}(\mathfrak{m}).

Proof. Let D:=k[[t]]D:=k[[t]]. The defining relations give a DD-basis 1,x,y1,x,y for RR. For completeness, independence follows by realizing this algebra as the subalgebra with basis 1,tu2,tu1,tu^{2},tu of D[u]/(u3)D[u]/(u^{3}), with x↦tu2x\mapsto tu^{2} and y↦tuy\mapsto tu. The relations reduce every power series to a DD-linear combination of 1,x,y1,x,y, and their images are independent in D[u]/(u3)D[u]/(u^{3}). Thus RR is free of rank three over DD, and tt is a nonzerodivisor. Relations x2=0x^{2}=0 and y2=txy^{2}=tx say that y4=0y^{4}=0, hence p=(x,y)\mathfrak{p}=(x,y) is nilpotent and R/p=DR/\mathfrak{p}=D; consequently dim⁡R=1\dim R=1 and RR is Cohen–Macaulay. One has m2=(t2,tx,ty)=tm\mathfrak{m}^{2}=(t^{2},tx,ty)=t\mathfrak{m}. The principal ideal (t)(t) is a reduction. Let us verify minimality directly. If a reduction JJ of m\mathfrak{m} is contained in (t)(t), write J=tIJ=tI. For a sufficiently large a≥1a\geq1, the identities ma+1=tma=Jma=tIma\mathfrak{m}^{a+1}=t\mathfrak{m}^{a}=J\mathfrak{m}^{a}=tI\mathfrak{m}^{a} imply, by cancellation of tt, that ma=Ima\mathfrak{m}^{a}=I\mathfrak{m}^{a}. If II were proper, Nakayama’s lemma would give ma=0\mathfrak{m}^{a}=0, contrary to ta≠0t^{a}\ne0. Hence I=RI=R and J=(t)J=(t). Also m≠(t)\mathfrak{m}\ne(t). A minimal reduction with reduction number zero would have to equal m\mathfrak{m} itself, which is impossible since m\mathfrak{m} contains the proper reduction (t)(t). Therefore r⁡(m)=1\operatorname{r}(\mathfrak{m})=1. Inverting tt gives R[1/t]≅K[y]/(y3)R[1/t]\cong K[y]/(y^{3}) with the convenience that K:=k((t))K:=k((t)) and x=t−1y2x=t^{-1}y^{2}.

Claim 4.3. This ring is local, and its maximal ideal contracts to p\mathfrak{p}.

Proof. Consider the localization map R→S:=R[1/t]R\to S:=R[1/t]. One has (y)S∩R=(x,y)=p(y)S\cap R=(x,y)=\mathfrak{p}. To this end, observe that R/p≅k[[t]]R/\mathfrak{p}\cong k[[t]] and localizing at tt gives (R/p)[1/t]≅k((t))=K(R/\mathfrak{p})[1/t]\cong k((t))=K. Thus the kernel of the composite map R→R[1/t]≅K[y]/(y3)→K[y]/(y)≅KR\to R[1/t]\cong K[y]/(y^{3})\to K[y]/(y)\cong K is exactly p\mathfrak{p}. This kernel is also (y)S∩R(y)S\cap R, since (y)(y) is the kernel of S→S/(y)S\to S/(y). Therefore, the maximal ideal (y)(y) of SS contracts to p\mathfrak{p}. Thus Rp≅R[1/t]R_{\mathfrak{p}}\cong R[1/t], because both invert precisely the elements outside p\mathfrak{p}, up to the resulting localization.

Thanks to Claim 4.3, Rp=R[1/t]R_{\mathfrak{p}}=R[1/t]. In an artinian local ring the zero ideal is a reduction of the nilpotent maximal ideal, so it is its unique minimal reduction. Here (y)2≠0(y)^{2}\ne0 and (y)3=0(y)^{3}=0, which gives r⁡(pRp)=2\operatorname{r}(\mathfrak{p}R_{\mathfrak{p}})=2.

Proposition 4.4. For every integer N≥2N\geq2 there is a complete one-dimensional Cohen–Macaulay local ring (RN,mN)(R_{N},\mathfrak{m}_{N}) with a prime pN⊊mN\mathfrak{p}_{N}\subsetneq\mathfrak{m}_{N} such that r⁡(mN)=1\operatorname{r}(\mathfrak{m}_{N})=1 and r⁡(pN(RN)pN)=N\operatorname{r}(\mathfrak{p}_{N}(R_{N})_{\mathfrak{p}_{N}})=N. Moreover, the failure can be made arbitrarily large by varying the ring, so no inequality of the form r⁡(pRp)≤f(r⁡(qRq))\operatorname{r}(\mathfrak{p}R_{\mathfrak{p}})\leq f(\operatorname{r}(\mathfrak{q}R_{\mathfrak{q}})) with a fixed function ff can hold in general.

Proof. Let D:=k[[t]]D:=k[[t]]. Inside D[u]/(uN+1)D[u]/(u^{N+1}), take the subring R:=RN:=D⋅1⊕⨁i=1ND⋅(tui)R:=R_{N}:=D\cdot1\oplus\bigoplus_{i=1}^{N}D\cdot(tu^{i}). It is finite free over DD, hence Noetherian, and p:=pN:=⨁i=1ND⋅(tui)\mathfrak{p}:=\mathfrak{p}_{N}:=\bigoplus_{i=1}^{N}D\cdot(tu^{i}) is nilpotent with quotient DD. Its unique maximal ideal is m=tD+p\mathfrak{m}=tD+\mathfrak{p}. Thus m\mathfrak{m} is generated by {t,tu,tu2,…,tuN}\{t,tu,tu^{2},\ldots,tu^{N}\}. Squaring, the generators of m2\mathfrak{m}^{2} are all products of two of these:

t2,t⋅tui=t2ui (1≤i≤N),(tui)(tuj)=t2ui+j,t^{2},\qquad t\cdot tu^{i}=t^{2}u^{i}\ (1\leq i\leq N),\qquad(tu^{i})(tu^{j})=t^{2}u^{i+j},

and ui+j=0u^{i+j}=0 whenever i+j≥N+1i+j\geq N+1. Hence m2=(t2,t2u,t2u2,…,t2uN)\mathfrak{m}^{2}=(t^{2},t^{2}u,t^{2}u^{2},\ldots,t^{2}u^{N}). On the other hand

tm=t(tD+p)=t2D⊕⨁i=1ND(t2ui)=(t2,t2u,t2u2,…,t2uN).t\mathfrak{m}=t(tD+\mathfrak{p})=t^{2}D\oplus\bigoplus_{i=1}^{N}D(t^{2}u^{i})=(t^{2},t^{2}u,t^{2}u^{2},\ldots,t^{2}u^{N}).

Therefore m2=tm\mathfrak{m}^{2}=t\mathfrak{m}. The tt-adic and m\mathfrak{m}-adic topologies agree, so RR is complete; freeness over DD makes it Cohen–Macaulay of dimension one. The same minimality argument as in Proposition 4.2 gives r⁡(m)=1\operatorname{r}(\mathfrak{m})=1. Finally, (R)p=R[1/t]≅k((t))[u]/(uN+1)(R)_{\mathfrak{p}}=R[1/t]\cong k((t))[u]/(u^{N+1}), whose maximal ideal has reduction number NN.

Remark 4.5

Remark 4.5. i) Proposition 4.4 generalizes Proposition 4.2: the latter is the case N=2N=2. Indeed, with x↦tu2x\mapsto tu^{2}, y↦tuy\mapsto tu,

R=k[[t,x,y]]/(x2,xy,y2−tx)≅k[[t]]⋅1⊕k[[t]](tu2)⊕k[[t]](tu)=D⋅1⊕⨁i=12D(tui)⊆D[u]/(u3),R=k[[t,x,y]]/(x^{2},xy,y^{2}-tx)\cong k[[t]]\cdot1\oplus k[[t]](tu^{2})\oplus k[[t]](tu)=D\cdot1\oplus\bigoplus_{i=1}^{2}D(tu^{i})\subseteq D[u]/(u^{3}),

which is exactly RNR_{N} for N=2N=2.

ii) In both cases p\mathfrak{p} is the nilpotent ideal with R/p≅DR/\mathfrak{p} \cong D, r⁡(m)=1\operatorname{r}(\mathfrak{m})=1, and after inverting tt,

R[1/t]≅K[u]/(u3),(RN)[1/t]≅K[u]/(uN+1),R[1/t] \cong K[u]/(u^3), \qquad(R_N)[1/t] \cong K[u]/(u^{N+1}),

with reduction number 22, resp. NN, at the maximal ideal (u)(u). The pair shows that the gap r⁡(pRp)−r⁡(m)\operatorname{r}(\mathfrak{p}R_{\mathfrak{p}})-\operatorname{r}(\mathfrak{m}) is unbounded.

iii) The rings we constructed were not be reduced.

Observation 4.6. If (R,m)(R,\mathfrak{m}) is 11-dimensional reduced ring, then for any p⊂m\mathfrak{p}\subset\mathfrak{m}, r⁡(pRp)≤r⁡(m)\operatorname{r}(\mathfrak{p}R_{\mathfrak{p}})\leq\operatorname{r}(\mathfrak{m}).

Proof. We may assume p⊊m\mathfrak{p}\subsetneq\mathfrak{m}. Since RR is reduced it satisfies Serre’s condition (R0)(R_0). But, dim⁡(R)=1\dim(R)=1, it follows that RpR_{\mathfrak{p}} is a field and so pRp=0\mathfrak{p}R_{\mathfrak{p}}=0. Consequently, r⁡(pRp)≤r⁡(m)\operatorname{r}(\mathfrak{p}R_{\mathfrak{p}})\leq\operatorname{r}(\mathfrak{m}).

A finite bound for every fixed local ring

A bound from parameter colength

The next estimate works without a Cohen–Macaulay assumption (compare with [12, Theorem 2.44).

Lemma 5.1. Let (A,n)(A,\mathfrak{n}) have dimension d≥1d\geq1, and let Q=(x1,…,xd)Q=(x_1,\ldots,x_d) be a parameter ideal. Set L=ℓA(A/Q)L=\ell_A(A/Q). For every integer h≥1h\geq1, μA(nh)≤hd−1L+d−1\mu_A(\mathfrak{n}^h)\leq h^{d-1}L+d-1.

Proof. First suppose d=1d=1. If x=x1x=x_1 and II is any n\mathfrak{n}-primary ideal, and

0⟶I⟶A⟶A/I⟶00\longrightarrow I\longrightarrow A\longrightarrow A/I\longrightarrow0

a short exact sequence of AA-modules. Apply multiplication by xx to each term:

0⟶I⟶A⟶A/I⟶00↓x⟶I↓x⟶A↓x0⟶I⟶A⟶A/I⟶0\begin{aligned} 0&\longrightarrow I\longrightarrow A\longrightarrow A/I\longrightarrow0\\ &\phantom{0}\downarrow x\phantom{\longrightarrow I}\quad\downarrow x\phantom{\longrightarrow A}\quad\downarrow x\\ 0&\longrightarrow I\longrightarrow A\longrightarrow A/I\longrightarrow0 \end{aligned}

The snake lemma gives the exact kernel–cokernel sequence

0⟶(0:Ax)⟶(0:Ax)⟶(0:A/Ix)⟶I/xI⟶A/xA⟶(A/I)/x(A/I)⟶0.0\longrightarrow(0:_A x)\longrightarrow(0:_A x)\longrightarrow(0:_{A/I}x)\longrightarrow I/xI\longrightarrow A/xA\longrightarrow(A/I)/x(A/I)\longrightarrow0.

Set M=A/IM=A/I. For the endomorphism x:M→Mx:M\to M of the finite-length module MM,

ℓ(M)=ℓ(ker⁡x)+ℓ(im⁡x)=ℓ((0:Mx))+ℓ(M)−ℓ(M/xM),\ell(M)=\ell(\ker x)+\ell(\operatorname{im}x)=\ell((0:_M x))+\ell(M)-\ell(M/xM),

hence ℓ((0:Mx))=ℓ(M/xM)\ell((0:_M x))=\ell(M/xM). Thus the kernel term ℓ((0:A/Ix))\ell((0:_{A/I}x)) and the cokernel term ℓ((A/I)/x(A/I))\ell((A/I)/x(A/I)) are equal and cancel in the length identity. In order to all terms be of finite length, we rewrite:

0⟶(0:Ax)/(0:Ix)⟶(0:A/Ix)⟶I/xI⟶A/xA⟶(A/I)/x(A/I)⟶0.0\longrightarrow(0:_A x)/(0:_I x)\longrightarrow(0:_{A/I}x)\longrightarrow I/xI\longrightarrow A/xA\longrightarrow(A/I)/x(A/I)\longrightarrow0.

Taking lengths along this exact sequence,

ℓ(I/xI)−ℓ(A/xA)+ℓ((A/I)/x(A/I))−ℓ((0:A/Ix))+ℓ((0:Ax))−ℓ((0:Ix))=0.\ell(I/xI)-\ell(A/xA)+\ell((A/I)/x(A/I))-\ell((0:_{A/I}x))+\ell((0:_A x))-\ell((0:_I x))=0.

This gives

ℓA(I/xI)=ℓA(A/xA)−ℓA((0:Ax)/(0:Ix))≤ℓA(A/xA).\ell_A(I/xI)=\ell_A(A/xA)-\ell_A((0:_A x)/(0:_I x))\leq\ell_A(A/xA).

Since xI⊆nIxI\subseteq\mathfrak{n}I, it follows that μA(I)≤ℓA(I/xI)≤L\mu_A(I)\leq\ell_A(I/xI)\leq L. Taking I=nhI=\mathfrak{n}^h proves the assertion for d=1d=1. For d>1d>1, we look at B:=A/(x1h,…,xd−1h)B:=A/(x_1^h,\ldots,x_{d-1}^h) and I=nhI=\mathfrak{n}^h. Then dim⁡B=1\dim B=1, xdx_d is a parameter of BB, and (x1h,…,xd−1h)⊆I(x_1^h,\ldots,x_{d-1}^h)\subseteq I. Lifting generators from the image IBIB and adjoining these d−1d-1 elements gives

μA(I)≤μB(IB)+d−1≤ℓA(A/(x1h,…,xd−1h,xd))+d−1.\mu_A(I)\leq\mu_B(IB)+d-1\leq\ell_A(A/(x_1^h,\ldots,x_{d-1}^h,x_d))+d-1.

To bound the last length, we increase the exponents of x1,…,xd−1x_1,\ldots,x_{d-1} one at a time. The key observation is that, for any finite-length AA-module CC and any x∈Ax\in A, the filtration C⊇xC⊇x2C⊇⋯⊇xhCC\supseteq xC\supseteq x^2C\supseteq\cdots\supseteq x^hC has hh successive factors xjC/xj+1Cx^jC/x^{j+1}C, for 0≤j<h0\leq j<h. Each factor is a homomorphic image of C/xCC/xC, so ℓA(C/xhC)≤hℓA(C/xC)\ell_A(C/x^hC)\leq h\ell_A(C/xC). Applying this inequality successively to x1,…,xd−1x_1,\ldots,x_{d-1}, we obtain

ℓA(A/(x1h,…,xd−1h,xd))≤hd−1ℓA(A/Q)=hd−1L.\ell_A(A/(x_1^h,\ldots,x_{d-1}^h,x_d))\leq h^{d-1}\ell_A(A/Q)=h^{d-1}L.

Consequently, μA(nh)≤hd−1L+d−1\mu_A(\mathfrak{n}^h)\leq h^{d-1}L+d-1, as desired.

We use the following form of the Eakin–Sathaye theorem [6 Theorem 8.6.8]:

Fact 5.2. If (A,n)(A,\mathfrak{n}) has infinite residue field and μA(Ih)<(h+dd)\mu_A(I^h)<\binom{h+d}{d}, then there exist a1,…,ad∈Ia_1,\ldots,a_d\in I such that Ih=(a1,…,ad)Ih−1I^h=(a_1,\ldots,a_d)I^{h-1}. For an n\mathfrak{n}-primary ideal in dimension dd, this is a minimal reduction. Here and below d≥1d\ge1 in applications of this statement.

Corollary 5.3. If (A,n)(A,\mathfrak{n}) has infinite residue field and dimension d≥1d\ge1, then for every parameter ideal QQ, r(n)≤d!ℓA(A/Q)−1r(\mathfrak{n})\le d!\ell_A(A/Q)-1. More generally, any integer h≥d!ℓA(A/Q)h\ge d!\ell_A(A/Q) satisfies r(n)≤h−1r(\mathfrak{n})\le h-1.

Proof. Set L=ℓA(A/Q)L=\ell_A(A/Q) and take h≥d!Lh\ge d!L. If d=1d=1, Lemma 5.1 gives μA(nh)≤L≤h<h+1\mu_A(\mathfrak{n}^h)\le L\le h<h+1. For d≥2d\ge2, it gives

μA(nh)≤Lhd−1+d−1≤hdd!+d−1<(h+dd).\mu_A(\mathfrak{n}^h)\le Lh^{d-1}+d-1\le\frac{h^d}{d!}+d-1<\binom{h+d}{d}.

For the strict inequality, expand (h+1)⋯(h+d)/d!(h+1)\cdots(h+d)/d!. In addition to hd/d!h^d/d!, its hd−1h^{d-1} term alone is d(d+1)2d!hd−1>d−1\frac{d(d+1)}{2d!}h^{d-1}>d-1, since h≥d!h\ge d!. Now apply Fact 5.2 with I=nI=\mathfrak{n}. □

A finite bound for sup {r(pRp)}\{r(\mathfrak{p}R_{\mathfrak{p}})\}

Here is the precise standard uniformity input used below. Huneke’s uniform Briançon–Skoda theorem implies that (see [5]). The bar denotes integral closure of an ideal.

Fact 5.4. If UU is a complete reduced Noetherian local ring, there is an integer b≥0b\ge0 such that Ib+1‾⊆I\overline{I^{b+1}}\subseteq I for every ideal I⊆UI\subseteq U.

Lemma 5.5. For a Noetherian local ring (R,m)(R,\mathfrak{m}) there is an integer C≥1C\ge1 such that, for every p∈Spec⁡R\mathfrak{p}\in\operatorname{Spec}R and every reduction JJ of n=pRp\mathfrak{n}=\mathfrak{p}R_{\mathfrak{p}}, nC⊆J\mathfrak{n}^C\subseteq J.

Proof. Let T=R^T=\widehat{R}, let N=(0)TN=\sqrt{(0)_T}, and choose s≥1s\ge1 with Ns=0N^s=0. Apply Fact 5.4 to the complete reduced local ring E=T/NE=T/N, obtaining bb; set C=s(b+1)C=s(b+1). Fix p\mathfrak{p} and JJ as in the statement, put S=R∖pS=R\setminus\mathfrak{p}, and set U=S−1TU=S^{-1}T. The map Rp→UR_{\mathfrak{p}}\to U is faithfully flat, being a base change of R→TR\to T. The reduction relation says that n\mathfrak{n} is integral over JJ; this property persists in UU (see [6, Proposition 8.1.1]) and in U/NU=S−1EU/NU=S^{-1}E (see [6, Lemma 8.1.3 (3)]). Consequently, in U/NUU/NU the (b+1)(b+1)st power of the image of nU\mathfrak{n}U is contained in the integral closure of the (b+1)(b+1)st power of the image of JUJU, hence in the image of JUJU by Fact 5.4. Thus (nU)b+1⊆JU+NU(\mathfrak{n}U)^{b+1}\subseteq JU+NU. Raising to the ssth power yields

nCU⊆(nU)C⊆(JU+NU)s⊆JU+(NU)s=JU.\mathfrak{n}^CU\subseteq(\mathfrak{n}U)^C\subseteq(JU+NU)^s\subseteq JU+(NU)^s=JU.

Contract along the faithfully flat map to obtain the desired claim (see [10, Theorem 7.5(ii)]). □

Remark 5.6. We need a uniform embedding-dimension bound. By the Cohen structure theorem choose a surjection V↠T=R^V\twoheadrightarrow T=\widehat{R} from a complete regular local ring VV, and write v=dim⁡Vv=\dim V. Then edim⁡(Rp)≤v\operatorname{edim}(R_{\mathfrak{p}})\le v for all p∈Spec⁡R\mathfrak{p}\in\operatorname{Spec}R. Indeed, choose a prime PP of TT over p\mathfrak{p}. The map Rp→TPR_{\mathfrak{p}}\to T_P is flat and local. Lech’s embedding-dimension inequality for flat local maps gives edim⁡(Rp)≤edim⁡(TP)\operatorname{edim}(R_{\mathfrak{p}})\le\operatorname{edim}(T_P); see [8]. If P^\widehat{P} is the preimage of PP in VV, then TPT_P is a quotient of the regular local ring VP^V_{\widehat{P}}, whose dimension is at most vv. This proves the desired claim.

Our next main result is as follows.

Theorem 5.7. Let (R,m)(R,\mathfrak{m}) be a Noetherian local ring of dimension D≥1D\ge1. Choose CC as in Lemma 5.5 and vv as in Remark 5.6. Finally, let M:=(v+C−1v)M:=\binom{v+C-1}{v} and H:=D!M−1H:=D!M-1. Then, for every prime p\mathfrak{p} for which κ(p)\kappa(\mathfrak{p}) is infinite, we have r(pRp)≤Hr(\mathfrak{p}R_{\mathfrak{p}})\le H. In particular, the supremum in Theorem 1.2(ii) is finite, for arbitrary residue field of RR.

Proof. Fix such a p\mathfrak{p}, and set (A,n)=(Rp,pRp)(A,\mathfrak{n})=(R_{\mathfrak{p}},\mathfrak{p}R_{\mathfrak{p}}) and d=dim⁡Ad=\dim A. If d=0d=0, AA is Artinian and the unique minimal reduction of n\mathfrak{n} is zero. Lemma 5.5 therefore gives nC=0\mathfrak{n}^C=0 and r(n)≤C−1r(\mathfrak{n})\le C-1.

Suppose d≥1d\ge1. Since the residue field of AA is infinite, n\mathfrak{n} has a minimal reduction QQ generated by dd elements; it is a parameter ideal. By Lemma 5.5, nC⊆Q\mathfrak{n}^C\subseteq Q. Since μA(n)≤v\mu_A(\mathfrak{n})\le v (see Remark 5.6), the graded pieces of the n\mathfrak{n}-adic filtration satisfy ℓA(ni/ni+1)≤(v+i−1i)\ell_A(\mathfrak{n}^i/\mathfrak{n}^{i+1}) \le\binom{v+i-1}{i}. Summing these inequalities, with the degree-zero term equal to one, gives

L:=ℓA(A/Q)≤ℓA(A/nC)≤∑i=0C−1(v+i−1i)=(v+C−1v)=M.L := \ell_A(A/Q) \le\ell_A(A/\mathfrak{n}^C) \le\sum_{i=0}^{C-1} \binom{v+i-1}{i} = \binom{v+C-1}{v} = M.

Corollary 5.3 now implies r(n)≤d!L−1≤D!M−1=Hr(\mathfrak{n}) \le d!L-1 \le D!M-1=H. The dimension-zero case also satisfies this bound, because v≥1v \ge1 and hence M≥CM \ge C.

Finally, every nonmaximal prime of a local ring has infinite residue field. Otherwise R/pR/\mathfrak{p}, which embeds into its finite fraction field κ(p)\kappa(\mathfrak{p}), would be a finite domain and thus a field; this would make p\mathfrak{p} maximal. If R/mR/\mathfrak{m} is infinite, r(pRp)≤Hr(\mathfrak{p}R_{\mathfrak{p}}) \le H. If it is finite, only m\mathfrak{m} is omitted, and Remark 2.1 gives r(m)<∞r(\mathfrak{m}) < \infty. In that case the supremum is at most max⁡{H,r(m)}\max\{H,r(\mathfrak{m})\}.

If dim⁡R=0\dim R=0, then Spec⁡R={m}\operatorname{Spec} R=\{\mathfrak{m}\} and the assertion is immediate from nilpotence of m\mathfrak{m}. This completes the proof of Theorem 1.2(ii).

An infinite supremum without noetherianity

Definition 5.8. Let kk be any field, D=k[[t]]D=k[[t]], K=k((t))K=k((t)), and for n≥1n\ge1 set An=D[u]/(un+1)A_n=D[u]/(u^{n+1}). Inside ∏n≥1An\prod_{n\ge1} A_n, define B=D⋅1+⨁n≥1AnB=D\cdot1+\bigoplus_{n\ge1} A_n.

Thus an element of BB is a sequence eventually equal to a single element of DD. Write ene_n for the idempotent supported at coordinate nn.

Notation 5.9. Set R=k⋅1+tBR=k\cdot1+tB.

The diagonal element tt is a nonzerodivisor in BB and in RR.

Remark 5.10. For this subsection only, allow non-Noetherian rings, and use the same definitions of reduction and minimal reduction whenever they exist.

Observation 5.11. The ring RR is quasi-local and non-Noetherian, with maximal ideal m:=tB\mathfrak{m}:=tB.

Proof. Every element of RR has the form c+tbc+tb with c∈kc\in k and b∈Bb\in B. If c≠0c\ne0, each coordinate is a unit in AnA_n, and its inverse sequence is eventually the inverse of a single unit of DD. Write c+tb=c(1+c−1tb)c+tb=c(1+c^{-1}tb). For any element xx since (1+x)(1−x(1+x)−1)=(1+x)−x=1(1+x)(1-x(1+x)^{-1})=(1+x)-x=1, we have (1+x)−1=1−x(1+x)−1(1+x)^{-1}=1-x(1+x)^{-1}. Applying this with x=c−1tbx=c^{-1}tb and multiplying by c−1c^{-1} gives

(c+tb)−1=c−1(1−c−1tb(1+c−1tb)−1)=c−1−c−2tb(1+c−1tb)−1.(c+tb)^{-1}=c^{-1}(1-c^{-1}tb(1+c^{-1}tb)^{-1})=c^{-1}-c^{-2}tb(1+c^{-1}tb)^{-1}.

Using (1+c−1tb)−1=c(c+tb)−1(1+c^{-1}tb)^{-1}=c(c+tb)^{-1}, the second term becomes c−1tb(c+tb)−1c^{-1}tb(c+tb)^{-1}, so (c+tb)−1=c−1−c−1tb(c+tb)−1(c+tb)^{-1}=c^{-1}-c^{-1}tb(c+tb)^{-1}, i.e. (c+tb)−1−c−1=−tc−1b(c+tb)−1(c+tb)^{-1}-c^{-1}=-tc^{-1}b(c+tb)^{-1}. Thus (c+tb)−1∈B(c+tb)^{-1}\in B, and (c+tb)−1−c−1=−tc−1b(c+tb)−1∈tB(c+tb)^{-1}-c^{-1}=-tc^{-1}b(c+tb)^{-1}\in tB. So c+tbc+tb is a unit of RR. Since R/tB≅kR/tB\cong k, this proves that RR is quasi-local with maximal ideal m=tB\mathfrak{m}=tB. For non-Noetherianity: multiplication by tt identifies m/m2≅B/tB\mathfrak{m}/\mathfrak{m}^2\cong B/tB as kk-vector spaces. The classes of the ene_n are linearly independent in B/tBB/tB, so this space is infinite-dimensional. Hence m\mathfrak{m} is not finitely generated. □

:::

Lemma 5.12. The primes of RR consist of m\mathfrak{m}, primes pn\mathfrak{p}_n for n≥1n\ge1, and one additional prime p∞\mathfrak{p}_\infty. So, dim⁡R=1\dim R=1. Their localizations satisfy Rpn≅K[u]/(un+1)R_{\mathfrak{p}_n}\cong K[u]/(u^{n+1}), and Rp∞≅KR_{\mathfrak{p}_\infty}\cong K.

Proof. Inverting tt yields C:=R[1/t]=B[1/t]=K⋅1+⨁n≥1K[u]/(un+1)C:=R[1/t]=B[1/t]=K\cdot1+\bigoplus_{n\ge1}K[u]/(u^{n+1}). Let FF be the direct-sum ideal in this expression. If a prime PP of CC omits some ene_n, then localization forces en=1e_n=1 and 1−en=01-e_n=0. It therefore selects the nnth factor, whose unique prime is (u)(u). Let PnP_n denote this prime of CC. Then CPn≅K[u]/(un+1)C_{P_n}\cong K[u]/(u^{n+1}). If PP contains every ene_n, it contains FF; since C/F=KC/F=K, it must be FF itself. Every element of FF has finite support and is annihilated by 1−∑n∈Een1-\sum_{n\in E}e_n for a suitable finite set EE. This annihilator lies outside FF, and so CF=KC_F=K. Contracting PnP_n and FF to RR defines pn\mathfrak{p}_n and p∞\mathfrak{p}_\infty, respectively, and proves the asserted localization formulas. All these primes omit tt and lie strictly below m\mathfrak{m}. Conversely, any prime of RR containing tt equals m\mathfrak{m}, because m2⊂tR\mathfrak{m}^2\subset tR. We have therefore listed every prime. The primes of CC are all maximal, so dim⁡R=1\dim R=1. □

Corollary 5.13. The reduction numbers are r(m)=1r(\mathfrak{m}) = 1, r(pnRpn)=nr(\mathfrak{p}_nR_{\mathfrak{p}_n}) = n and r(p∞Rp∞)=0r(\mathfrak{p}_{\infty}R_{\mathfrak{p}_{\infty}}) = 0. In particular, every displayed reduction number is finite and their supremum is infinite.

Proof. We have m2=t2B=tm\mathfrak{m}^2 = t^2B = t\mathfrak{m}, and m≠tR\mathfrak{m} \ne tR because ten∉tRt e_n \notin tR. We verify that tRtR is a minimal reduction without using a finite-generation assumption. If J⊆tRJ \subseteq tR is a reduction of m\mathfrak{m}, write J=tIJ = tI with II an ideal of RR. For some a≥1a \ge1, ta+1B=ma+1=Jma=ta+1IBt^{a+1}B = \mathfrak{m}^{a+1} = J\mathfrak{m}^a = t^{a+1}IB. Cancellation inside BB gives IB=BIB = B. A proper ideal II of the local ring RR lies in tBtB, which would force IB⊆tB≠BIB \subseteq tB \ne B. Hence I=RI = R and J=tRJ = tR. Thus r(m)=1r(\mathfrak{m}) = 1, since reduction number zero would force m\mathfrak{m} itself to be minimal. According to Lemma 5.12, localizing at pn\mathfrak{p}_n, the maximal ideal is (u)(u), with nonzero nnth power and zero (n+1)(n+1)st power. Its unique minimal reduction is zero and its reduction number is nn. At p∞\mathfrak{p}_{\infty} the maximal ideal is zero and the reduction number is zero. Hence all displayed reduction numbers are finite, while their supremum over nn is infinite. □

Loewy Length Conjecture

We start with a modification of Conjecture 1.4.

Question 6.1. Let (R,m)(R,\mathfrak{m}) be a noetherian local ring and MM a finitely generated RR-module of finite projective dimension with mnM=0\mathfrak{m}^nM = 0. Does mn⊆(x)\mathfrak{m}^n \subseteq(x) for some system of parameters xx?

For every finite field k:=Fqk := \mathbb{F}_q there is a one-dimensional hypersurface RR with residue field kk and a nonzero module of projective dimension one for which the containment fails at n=q+1n = q + 1:

Notation 6.2. Let A:=k[[x,y]]A := k[[x,y]], f:=y∏λ∈k(x−λy)f := y\prod_{\lambda\in k}(x-\lambda y), and R:=Rq:=A/(f)R := R_q := A/(f), with m=(x,y)R\mathfrak{m} = (x,y)R.

Also, we need an auxiliary ring:

Notation 6.3. Let K/kK/k be a quadratic extension, α∈K∖k\alpha\in K\setminus k and set S:=K[[x,y]]/(f)=R⊗kKS := K[[x,y]]/(f) = R\otimes_k K.

Lemma 6.4. Any SS-module of finite projective dimension, has finite projective dimension as an RR-module.

Proof. The ring SS is free of rank two over RR. So, the claim is clear. □

Example 6.5. There is an RR-module MM of finite projective dimension and Loewy length q+1q+1.

Proof. Let u:=x−αyu := x-\alpha y and define M:=S/(u)M := S/(u). Substitution gives M≅K[[y]]/(yq+1)M \cong K[[y]]/(y^{q+1}) and f(αy,y)=yq+1∏λ∈k(α−λ)f(\alpha y,y) = y^{q+1}\prod_{\lambda\in k}(\alpha-\lambda). In K[[x,y]]K[[x,y]], uu generates a prime ideal and f∉(u)f \notin(u) as α∈K∖k\alpha\in K\setminus k. If ub=fcub = fc, then u∣cu \mid c, and cancellation gives b∈(f)b \in(f). Thus multiplication by uu is injective on SS, i.e., there is a short exact sequence $0 \to S \xrightarrow{\cdot u} S \to M \to 0$ of SS-modules. In K[[x,y]]/(u)K[[x,y]]/(u) we have x≡αyx \equiv\alpha y, so substitution gives f(αy,y)=yq+1∏λ∈k(α−λ)f(\alpha y,y) = y^{q+1}\prod_{\lambda\in k}(\alpha-\lambda). Because α∈K∖k\alpha\in K\setminus k, none of the factors α−λ\alpha-\lambda vanishes, hence ∏λ∈k(α−λ)≠0\prod_{\lambda\in k}(\alpha-\lambda) \ne0 is a unit in K[[y]]K[[y]]. Therefore M≅K[[y]]/(yq+1)M \cong K[[y]]/(y^{q+1}), as claimed. Under this isomorphism, the class of yy maps to the generator yy of K[[y]]/(yq+1)K[[y]]/(y^{q+1}), while the class of xx maps to αy\alpha y. Since MM is an RR-module via R→SR \to S, the maximal ideal m=(x,y)R\mathfrak{m} = (x,y)R acts through the images of xx and yy in MM. But in MM the class of xx equals α\alpha times the class of yy. Hence, as operators on MM, x=αyx = \alpha y, y=yy = y. Thus every element of m\mathfrak{m} acts as a multiple of yy, and m2M=m(mM)=m(yM)=y2M\mathfrak{m}^2M = \mathfrak{m}(\mathfrak{m}M) = \mathfrak{m}(yM) = y^2M, and inductively mnM=ynM\mathfrak{m}^nM = y^nM for all n≥1n \ge1. Since M=K[[y]]/(yq+1)M = K[[y]]/(y^{q+1}), we have mq+1M=0\mathfrak{m}^{q+1}M = 0, and therefore mqM=yqM≠0\mathfrak{m}^qM = y^qM \ne0. Similarly mqM=yqM≠0\mathfrak{m}^qM = y^qM \ne0, because yq∉(yq+1)y^q \notin(y^{q+1}) in K[[y]]K[[y]]. Thus, Loewy length of MM is q+1q+1. □

For q=2q = 2, RR is the ring of Hashimoto and Shida [4 Example 3.2]. Bartels proved the generalized Loewy length of RqR_q is q+2q+2, see [2 Proposition 3.5]. This gives:

Corollary 6.6. Adopt the above notation. Then ℓℓR(M)=q+1≠q+2=gℓℓ(R)\ell\ell_R(M) = q+1 \ne q+2 = g\ell\ell(R).

KC and Pollitz proved their inequality for infinite residue fields when Gr⁡m(R)\operatorname{Gr}_{\mathfrak{m}}(R) is Cohen–Macaulay, see [11 Conjecture A and Theorem B].

Acknowledgments

The author is grateful to Josh Pollitz and Nawaj KC for useful comments.

References

M. ASGHARZADEH

Email address: [email protected]

References

  1. [1]M. Asgharzadeh, On the Rees, Dilworth and Matsumura number, arXiv:1712.03828.arxiv.org/abs/1712.03828
  2. [2]R. F. Bartels, Generalized Loewy length of Cohen–Macaulay local and graded rings, J. Algebra 653 (2024), 28–41.DOI
  3. [3]D. Hanes, Length approximations for independently generated ideals, J. Algebra 237 (2) (2001) 708–718.DOI
  4. [4]M. Hashimoto and A. Shida, Some remarks on index and generalized Loewy length of a Gorenstein local ring, J. Algebra 187 (1997), 150–162.DOI
  5. [6]I. Swanson and C. Huneke, Integral closure of ideals, rings, and modules, London Mathematical Society Lecture Note Series, 336, Cambridge University Press, Cambridge, 2006.
  6. [8]C. Lech, Inequalities related to certain couples of local rings, Acta Math. 112 (1964), 69–89.DOI
  7. [9]T. Marley, Hilbert functions of ideals in Cohen-Macaulay rings, PhD Thesis (1989).
  8. [10]H. Matsumura, Commutative ring theory, Cambridge Studies in Advanced Math, 8, (1986).DOI
  9. [11]Nawaj KC, J. Pollitz, Lower bounds on Loewy lengths of modules of finite projective dimension, Advances in Mathematics 473 (2025), 110309.arxiv.org/abs/2404.18368
  10. [12]W. V. Vasconcelos, Integral Closure: Rees Algebras, Multiplicities, Algorithms, Springer Monographs in Mathematics, Springer-Verlag, Berlin, 2005.

Paper details

vlnc:W_5z2bgydr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

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