Stability of annihilators of cohomology and closed subsets defined by Jacobian ideals

Kaito Kimura Graduate School of Mathematics, Nagoya University, Furocho, Chikusaku, Nagoya 464-8602, Japan [email protected]

Abstract.

Let $R$ be a commutative Noetherian ring of dimension $d$ . In this paper, we first show that some power of the cohomology annihilator annihilates the $(d+1)$ -th Ext modules for all finitely generated modules when either $R$ admits a dualizing complex or $R$ is local. Next, we study the Jacobian ideal of affine algebras over a field and equicharacteristic complete local rings, and characterize the equidimensionality of the ring in terms of the singular locus and the closed subsets defined by the cohomology annihilator and the Jacobian ideal.

2020 Mathematics Subject Classification. 13D07; 13C15; 13N15

Key words and phrases. cohomology annihilator, singular locus, Jacobian ideal, equidimensional.

The author was partly supported by Grant-in-Aid for JSPS Fellows Grant Number 23KJ1117.

1. Introduction

Throughout the present paper, all rings are assumed to be commutative and Noetherian. For a ring $R$ and an integer $n$ , denote by $\mathsf{ca}^{n}(R)$ the ideal consisting of elements $a$ such that $a\operatorname{Ext}_{R}^{n}(M,N)=0$ for all finitely generated $R$ -modules $M,N$ . The union $\bigcup_{n\geq 0}\mathsf{ca}^{n}(R)$ is called cohomology annihilator of $R$ , which is denoted by $\mathsf{ca}(R)$ . The ascending chain of radicals of $\mathsf{ca}^{n}(R)$ is stable for all large $n$ since $R$ is Noetherian. Iyengar and Takahashi [8] proved that when $R$ is either a localization of an affine algebra over a field or an equicharacteristic excellent local ring, the radical of $\mathsf{ca}^{2d+1}(R)$ is equal to that of $\mathsf{ca}(R)$ , where $d=\operatorname{dim}R$ . They also showed that these are defining ideal of the singular locus $\operatorname{Sing}R$ of $R$ , which is the set of prime ideals $\mathfrak{p}$ of $R$ such that $R_{\mathfrak{p}}$ is not regular. Dey and Takahashi [5] implicitly proved that when $R$ is a Cohen–Macaulay local ring with a canonical module, the radical of $\mathsf{ca}^{n}(R)$ is stable for all $n\geq\operatorname{dim}R+1$ ; see Remark 2.7(1). The main result in this direction refines the results mentioned above.

Theorem 1.1 (Theorem 2.5 and Corollary 2.6).

Let $R$ be a ring of dimension $d$ . Suppose either that $R$ admits a dualizing complex or that $R$ is local. Then $\mathrm{V}(\mathsf{ca}(R))=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ holds. In particular, if $R$ is quasi-excellent, $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ holds.

Theorem 1.1 provides the smallest number at which $\{\sqrt{\smash[b]{{\mathsf{ca}^{n}(R)}}}\}_{n\geq 0}$ stabilizes under several standard assumptions; see Remark 2.7(2). The latter part of Theorem 1.1 is a consequence of results in [4, 8].

Let $k$ be a field, and $R=k[X_{1},\dots,X_{m}]/(f_{1},\ldots,f_{n})$ a quotient of a polynomial ring (resp. $R=k\llbracket X_{1},\dots,X_{m}\rrbracket/(f_{1},\ldots,f_{n})$ a quotient of a formal power series ring) over $k$ . We denote by $J_{r}^{k}(R)$ (resp. $J_{r}(R)$ ) the ideal of $R$ generated by the $(r+s)$ -minors of the $m\times n$ matrix $(\partial f_{j}/\partial X_{i})$ called the Jacobian matrix, where $s=\operatorname{ht}(f_{1},\ldots,f_{n})$ . The supremum of $\operatorname{dim}R-\operatorname{dim}R/\mathfrak{p}$ over all minimal prime ideals $\mathfrak{p}$ of $R$ is denoted by $\operatorname{edd}R$ . The ideal $J_{0}^{k}(R)$ (resp. $J_{0}(R)$ ) is called the Jacobian ideal, and $R$ is called equidimensional if $\operatorname{edd}R=0$ . The following is a well-known characterization of the regularity of rings via the Jacobian ideal, obtained as a corollary of the classical result known as the Jacobian criterion: for any affine $k$ -algebra (resp. an equicharacteristic complete local ring) $R$ with $e=\operatorname{edd}R$ , one has $\operatorname{Sing}R\subseteq\mathrm{V}(J_{e}^{k}(R))$ and $\operatorname{Spec}R=\mathrm{V}(J_{e+1}^{k}(R))$ (resp. $\operatorname{Sing}R\subseteq\mathrm{V}(J_{e}(R))$ and $\operatorname{Spec}R=\mathrm{V}(J_{e+1}(R))$ ). The theorem below refines this fact.

Theorem 1.2 (Corollary 3.13).

Let $k$ be a perfect field and let $n\geq 0$ be an integer.

(1)
For an affine $k$ -algebra $R$ of dimension $d$ , the following are equivalent:
1. (i)
  
  The inequality $\operatorname{edd}R\leq n$ holds;
2. (ii)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}^{l}(S))$ holds;
3. (iii)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}^{l}(S))$ holds;
4. (iv)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}^{l}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ holds.
(2)
For an equicharacteristic complete local ring $(R,\mathfrak{m},k)$ of dimension $d$ , the following are equivalent:
1. (i)
  
  The inequality $\operatorname{edd}R\leq n$ holds;
2. (ii)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}(S))$ ;
3. (iii)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}(S))$ ;
4. (iv)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ .

In (ii), (iii), and (iv) of the above theorem, it is sufficient to consider only the case where the (residue) field is $k$ ; see Corollary 3.13. As mentioned earlier, the implications (i) $\Rightarrow$ (ii) and (i) $\Rightarrow$ (iii) are known. It is worth mentioning that even if $\operatorname{Sing}R\subseteq\mathrm{V}(J_{n}^{k}(R))$ and $\operatorname{Spec}R=\mathrm{V}(J_{n+1}^{k}(R))$ hold for an affine algebra $R$ over a field $k$ , it does not necessarily follow that $\operatorname{edd}R\leq n$ ; see Example 3.11. Our new idea is to view $\operatorname{edd}R$ as a topological invariant of the spectrum of the ring and to show (ii) $\Rightarrow$ (i) and (iii) $\Rightarrow$ (i). It is clear that (iv) $\Rightarrow$ (ii) holds. Thanks to Theorem 1.1, the converse (ii) $\Rightarrow$ (iv) is obtained.

Consider Theorem 1.2 in the case when $n=0$ . In this case, the inclusion relations in (ii) and (iv) can be rewritten as equalities; see Corollary 3.14. Theorem 1.2 provides a necessary and sufficient condition for a ring to be equidimensional in terms of the Jacobian ideal. By Wang [12, 13], under some assumptions, and by Iyengar and Takahashi [9], in the general case, it was proved that $\mathsf{ca}^{d+1}(R)$ contains some power of the Jacobian ideal of $R$ when $R$ is equidimensional, which means that (i) $\Rightarrow$ (iv) holds. Combining this with the trivial implication (iv) $\Rightarrow$ (ii), we have (i) $\Rightarrow$ (ii) that is a fact derived from the Jacobian criterion mentioned earlier. Theorem 1.1 says that (i) $\Rightarrow$ (ii) and (i) $\Rightarrow$ (iv) are equivalent.

Other results of this paper are explained. In Theorem 1.2, we considered all rings for which the spectrum are homeomorphic. On the other hand, in Theorems 3.6 and 3.7, we can compare the inclusion relations in (ii) with the equalities in (iii) for a fixed ring. Using these theorems and the result in [9], Propositions 3.9 and 3.10 provide an elementary proof of (i) $\Rightarrow$ (ii) and (i) $\Rightarrow$ (iii) in Theorem 1.2 without employing the terminology of smoothness used in the original proof of the Jacobian criterion.

The organization of this paper is as follows. In Sections 2, we consider the relationship between the closed subset of the spectrum of a ring defined by the cohomology annihilator and the singular locus, and prove Theorem 1.1. In Section 3, we study the ideal generated by the minors of the Jacobian matrix and give Theorem 1.2. Section 4 is an appendix and provides, using elementary arguments, that the Jacobian ideal is well-defined for equicharacteristic complete local rings. At the end of Section 4, an unresolved question is presented.

2. Asymptotic stability of the radicals of cohomology annihilators

This section studies the number at which the ascending sequence formed by the radicals of the cohomology annihilator stabilizes. The main result of this section provides the smallest number among such ones under several assumptions. First of all, we state the definitions of notions used in this paper.

Definition 2.1.

Let $R$ be a ring and $M$ a finitely $R$ -module. We denote by $\operatorname{mod}R$ the category of finitely generated $R$ -modules For every integer $n\geq 0$ , we denote by $\mathsf{ca}^{n}(R)$ the ideal consisting of elements $a$ such that $a\operatorname{Ext}_{R}^{n}(M,N)=0$ for all $M,N\in\operatorname{mod}R$ . The union $\bigcup_{n\geq 0}\mathsf{ca}^{n}(R)$ is called cohomology annihilator of $R$ , which is denoted by $\mathsf{ca}(R)$ . The singular locus $\operatorname{Sing}(R)$ of $R$ is defined as the set of prime ideals $\mathfrak{p}$ of $R$ such that $R_{\mathfrak{p}}$ is not a regular local ring and the non-Gorenstein locus $\operatorname{NG}R$ of $R$ is defined as the set of prime ideals $\mathfrak{p}$ of $R$ such that $R_{\mathfrak{p}}$ is not Gorenstein. The non-free locus $\operatorname{NF}_{R}(M)$ of $M$ is the set of prime ideals of $R$ such that $M_{\mathfrak{p}}$ is not a free $R_{\mathfrak{p}}$ -module. For each ideal $I$ of $R$ , the set of prime ideals of $R$ which contain $I$ is denoted by $\mathrm{V}(I)$ . The completion of $R$ is denoted by $\widehat{R}$ when $R$ is local.

A similar argument to the proof of [10, Proposition 2.4(1)] shows the lemma below.

Lemma 2.2.

Let $R$ be a local ring. Then there are equalities $\sqrt{\smash[b]{\mathsf{ca}(R)}}=\sqrt{\smash[b]{\mathsf{ca}(\widehat{R})\cap R}}$ and $\sqrt{\smash[b]{\mathsf{ca}^{n}(R)}}=\sqrt{\smash[b]{\mathsf{ca}^{n}(\widehat{% R})\cap R}}$ for all integers $n\geq 0$ .

We prepare a proposition concerning the non-Gorenstein locus, which plays an essential role in the proof of Theorem 2.4. It is well known that the non-Gorenstein locus is a closed set if the ring has a dualizing complex (or equivalently, if it is a homomorphic image of a finite dimensional Gorenstein ring). The following proposition characterizes that closed set in terms of the non-free locus of a finitely generated module and an annihilator ideal.

Proposition 2.3.

Let $R$ be a ring of dimension $d$ with dualizing complex $D=(\cdots\to 0\to D^{0}\to D^{1}\to\cdots\to D^{d}\to 0\to\cdots)$ and let $P=(\cdots\to P^{i-1}\xrightarrow{d^{i-1}}P^{i}\xrightarrow{d^{i}}P^{i+1}\to\cdots)$ be a complex of finitely generated projective $R$ -modules such that $P^{i}=0$ for all $i\gg 0$ and $H^{j}(P)\cong H^{j}(D)$ for every integer $j$ . Then $\operatorname{NG}R=\operatorname{NF}_{R}(\operatorname{Coker}d^{-1})=\mathrm{V% }(\bigcap_{M\in\operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,% R))$ .

Proof.

We put $C=\operatorname{Coker}d^{-1}$ and $I=\bigcap_{M\in\operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)$ . Let $\mathfrak{p}$ be a prime ideal of $R$ . If $\mathfrak{p}$ does not contain $I$ , then it also does not contain $\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(R/\mathfrak{p},R)$ . We see that $\operatorname{Ext}_{R}^{d+1}(R/\mathfrak{p},R)_{\mathfrak{p}}=0$ and that $R_{\mathfrak{p}}$ is Gorenstein. We obtain $\operatorname{NG}R\subset\mathrm{V}(I)$ . Suppose that $R_{\mathfrak{p}}$ is Gorenstein. Then $H^{0}(P_{\mathfrak{p}})\cong H^{0}(D_{\mathfrak{p}})\cong R_{\mathfrak{p}}$ and $H^{j}(P_{\mathfrak{p}})\cong H^{j}(D_{\mathfrak{p}})=0$ for any $j\neq 0$ . For any $i\geq 0$ , there exists a short exact sequence $0\to\operatorname{Ker}d^{i}_{\mathfrak{p}}\to P^{i}_{\mathfrak{p}}\to% \operatorname{Ker}d^{i+1}_{\mathfrak{p}}\to 0$ as $\operatorname{Im}d^{i}_{\mathfrak{p}}\cong\operatorname{Ker}d^{i+1}_{\mathfrak% {p}}$ . For some integer $n>d$ , $P^{n}=0$ and hence $\operatorname{Ker}d^{n-1}\cong P_{n-1}$ . By induction on $i$ , $\operatorname{Ker}d^{i}_{\mathfrak{p}}$ is a free $R_{\mathfrak{p}}$ -module for all $0<i<n$ . In particular, $\operatorname{Im}d^{0}_{\mathfrak{p}}\cong\operatorname{Ker}d^{1}_{\mathfrak{p}}$ is free. The natural short exact sequence $0\to H^{0}(P_{\mathfrak{p}})\to\operatorname{Coker}d^{-1}_{\mathfrak{p}}\to% \operatorname{Im}d^{0}_{\mathfrak{p}}\to 0$ says that $C_{\mathfrak{p}}\cong\operatorname{Coker}d^{-1}_{\mathfrak{p}}$ is free. Therefore we get $\operatorname{NF}_{R}(C)\subset\operatorname{NG}R$ .

Finally, we prove $\mathrm{V}(I)\subset\operatorname{NF}_{R}(C)$ . Fix $M\in\operatorname{mod}R$ . As $\mathrm{V}(\mathsf{Ann}\operatorname{Ext}_{R}^{1}(C,\Omega C))=\operatorname{% NF}_{R}(C)$ , we have only to show that $\mathsf{Ann}\operatorname{Ext}_{R}^{1}(C,\Omega C)\subset\mathsf{Ann}% \operatorname{Ext}_{R}^{d+1}(M,R)$ . There is a short exact sequence

of complexes and the natural quasi-isomorphism $H\to C$ since $H^{j}(P)\cong H^{j}(D)=0$ for all $j<0$ . Applying the derived functor $\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(-,D))$ to the exact triangle $G\to F\to C\rightsquigarrow$ in the (bounded) derived category, we have an exact triangle

\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(C,D))\to% \operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(F,D))\to% \operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(G,D))\rightsquigarrow,

which induces the exact sequence

(2.3.1)

H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(C,% D)))\to\operatorname{Ext}_{R}^{d+1}(M,R)\to H^{d+1}(\operatorname{\mathbf{R}% Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(G,D)))

of $R$ -modules via the quasi-isomorphism $\operatorname{\mathbf{R}Hom}_{R}(F,D)\simeq\operatorname{\mathbf{R}Hom}_{R}(D,% D)\simeq R$ . Noting that $G$ is a bounded complex of finitely generated projective $R$ -modules and $D$ is a bounded complex of injective $R$ -modules, $\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(G,D))$ is quasi-isomorphic to $\operatorname{Hom}_{R}(M,\operatorname{Hom}_{R}(G,D))$ . For any integer $i$ , if $i\leq 0$ , then $G^{i}=0$ ; otherwise, $D^{i+d+1}=0$ . Hence $\operatorname{Hom}_{R}(M,\operatorname{Hom}_{R}(G,D))^{d+1}=\operatorname{Hom}% _{R}(M,\operatorname{Hom}_{R}(G,D)^{d+1})=0$ , which means $H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(G,% D)))=0$ . By (2.3.1), we have $\mathsf{Ann}H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R% }Hom}_{R}(C,D)))\subset\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)$ .

Let $a\in\mathsf{Ann}\operatorname{Ext}_{R}^{1}(C,\Omega C)$ . It follows from [5, Lemma 3.8] that the multiplication by $a$ on $C$ factors through some free module $R^{\oplus m}$ . Applying $H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(-,% D)))$ , the multiplication by $a$ on $H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{\mathbf{R}Hom}_{R}(C,% D)))$ factors through $H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,D))^{\oplus m}$ . Similarly, $\operatorname{\mathbf{R}Hom}_{R}(M,D)$ is quasi-isomorphic to $\operatorname{Hom}_{R}(M,D)$ and $\operatorname{Hom}_{R}(M,D)^{d+1}=\operatorname{Hom}_{R}(M,D^{d+1})=0$ . So $a\in\mathsf{Ann}H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,\operatorname{% \mathbf{R}Hom}_{R}(C,D)))\subset\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)$ as $H^{d+1}(\operatorname{\mathbf{R}Hom}_{R}(M,D))^{\oplus m}=0$ . ∎

In general, for any complex $X=(\cdots\to X^{i}\to X^{i+1}\to\cdots)$ of $R$ -modules such that $H^{i}(X)$ are finitely generated over $R$ and $H^{j}(X)=0$ for all integers $i$ and $j\gg 0$ , there is a complex $P$ of finitely generated projective $R$ -modules and a quasi-isomorphism $P\to X$ such that ${\rm sup}\{i\mid P^{i}\neq 0\}={\rm sup}\{i\mid H^{i}(X)\neq 0\}$ ; see [2, Theorem (A.3.2)(L)] for instance. So, the result below is a direct corollary of Proposition 2.3.

Corollary 2.4.

Let $R$ be a ring of dimension $d$ with dualizing complex. Then

\operatorname{NG}R=\mathrm{V}\Bigl{(}\bigcap\nolimits_{M\in\operatorname{mod}R% }\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)\Bigr{)}.

Suppose that $R$ is a $d$ -dimensional Cohen–Macaulay local ring with canonical module $\omega$ . One has the equality $\operatorname{NG}R=\operatorname{NF}_{R}(\omega)$ . In this case, Corollary 2.4 follows immediately from [3, Theorem 2.3] because it says that $\mathsf{Ann}\operatorname{Ext}_{R}^{1}(\omega,\Omega\omega)=\bigcap_{M\in% \operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)$ holds. On the other hand, Proposition 2.3 can also be viewed as a non-Cohen–Macaulay version of [3, Theorem 2.3]. Indeed, with the notation of the proof of Proposition 2.3, when $R$ is Cohen–Macaulay, $\operatorname{Im}d^{i}\cong\operatorname{Ker}d^{i+1}$ are projective for all $i\geq 0$ and up to a projective summand $\operatorname{Im}d^{0}$ , $C$ is isomorphic to $H^{0}(P)$ , which is a canonical module of $R$ .

The main result of this section is the following theorem. Let $R$ be a finite dimensional ring and let $n>\operatorname{dim}R$ be an integer. Theorem 2.5 asserts that the radical of $\mathsf{ca}(R)$ is equal that of $\mathsf{ca}^{n}(R)$ if either $R$ admits a dualizing complex or $R$ is local.

Theorem 2.5.

Let $R$ be a ring of dimension $d$ . Suppose either that $R$ admits a dualizing complex or that $R$ is local. Then $\mathrm{V}(\mathsf{ca}(R))=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ holds.

Proof.

First, we deal with the case where $R$ admits a dualizing complex. Since $\mathsf{ca}^{d+1}(R)\subset\mathsf{ca}(R)$ , it suffices to prove that $\mathsf{ca}(R)\subset\sqrt{\mathsf{ca}^{d+1}(R)}$ . Put $I=\bigcap_{M\in\operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{d+1}(M,R)$ and take $n\geq d$ such that $\mathsf{ca}^{n+1}(R)=\mathsf{ca}(R)$ . It follows from Corollary 2.4 and [8, Lemma 2.10(2)] that $\mathrm{V}(I)=\operatorname{NG}R\subset\operatorname{Sing}R\subset\mathrm{V}(% \mathsf{ca}(R))$ . Thus $I$ contains $\mathsf{ca}(R)^{l}$ for some $l>0$ . Let $M,N$ be finitely generated $R$ -modules. A short exact sequence $0\to\Omega N\to R^{\oplus m}\to N\to 0$ induces a long exact sequence

\cdots\to\operatorname{Ext}_{R}^{i}(M,R^{\oplus m})\to\operatorname{Ext}_{R}^{% i}(M,N)\to\operatorname{Ext}_{R}^{i+1}(M,\Omega N)\to\cdots.

This means that $I\cdot\mathsf{ca}^{i+1}(R)\subset\mathsf{ca}^{i}(R)$ for every $i>d$ . We obtain $I^{n-d}\cdot\mathsf{ca}^{n+1}(R)\subset\mathsf{ca}^{d+1}(R)$ and hence $\mathsf{ca}(R)^{l(n-d)+1}\subset\mathsf{ca}^{d+1}(R)$ .

Next, We handle the case where $R$ is local. The equality $\sqrt{\smash[b]{\mathsf{ca}(\widehat{R})}}=\sqrt{\smash[b]{\mathsf{ca}^{d+1}(% \widehat{R})}}$ holds since $\widehat{R}$ admits a dualizing complex. The assertion follows from Lemma 2.2. ∎

The cohomology annihilator is defining ideal of the singular locus under several assumptions. Corollary 2.6 plays an important role in proving Theorem 1.2, which is one of the main results of this paper.

Corollary 2.6.

Let $R$ be a quasi-excellent ring of dimension $d$ . Suppose either that $R$ admits a dualizing complex or that $R$ is local. Then $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ holds.

Proof.

It follows from [4, Corollary C] and [8, Theorem 1.1] that $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}(R))$ holds. One has the equality $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ by Theorem 2.5. ∎

Theorem 2.5 and Corollary 2.6 improve and recover several existing results.

Remark 2.7.

(1) When $R$ is a Cohen–Macaulay local ring with a canonical module $\omega$ , Theorem 2.5 can be proved more easily by using existing results instead of Corollary 2.4. Indeed, we put $\mathsf{ca}^{n+1}(R)=\mathsf{ca}(R)$ for some $n\geq d$ . It follows from [5, Propositon 4.2(2)] (or [3, Theorem 2.3]) that

(\operatorname{tr}\omega)^{n}\cdot\bigcap_{X\in\mathsf{CM}R,Y\in\operatorname{% mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{n+1}(X,\Omega^{n}Y)\subset\bigcap_{X% \in\mathsf{CM}R,Y\in\operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{1}% (X,Y)

where $\mathsf{CM}R$ is the subcategory of $\operatorname{mod}R$ consisting of maximal Cohen–Macaulay $R$ -modules and $\operatorname{tr}\omega$ is the trace ideal of $\omega$ ; see [3, 5] for instance. As $\mathrm{V}(\operatorname{tr}\omega)=\operatorname{NG}R$ , similar to the proof of Theorem 2.5, $\mathsf{ca}(R)^{l}\subset\operatorname{tr}\omega$ for some $l>0$ . The relations

	$\displaystyle\mathsf{ca}(R)$	$\displaystyle=\mathsf{ca}^{n+1}(R)\subset\bigcap_{X\in\mathsf{CM}R,Y\in% \operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{n+1}(X,\Omega^{n}Y)% \quad{\rm and}$
	$\displaystyle\bigcap_{X\in\mathsf{CM}R,Y\in\operatorname{mod}R}$	$\displaystyle\mathsf{Ann}\operatorname{Ext}_{R}^{1}(X,Y)\subset\bigcap_{X,Y\in% \operatorname{mod}R}\mathsf{Ann}\operatorname{Ext}_{R}^{1}(\Omega^{d}X,Y)=% \mathsf{ca}^{d+1}(R)$

deduce $\mathsf{ca}(R)^{ln+1}\subset\mathsf{ca}^{d+1}(R)$ , which means $\mathrm{V}(\mathsf{ca}(R))=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ .

(2) Let $R$ be as in Theorem 2.5. Note that $\mathrm{V}(\mathsf{ca}^{d}(R))$ does not necessarily equal to $\mathrm{V}(\mathsf{ca}^{d+1}(R))$ . Indeed, if $R$ is either a polynomial ring or a formal power series ring over a field, then $\mathsf{ca}^{d}(R)=0$ and $\mathsf{ca}^{d+1}(R)=R$ . Moreover, for any ring $R$ of dimension $d$ , we see that $\mathsf{ca}^{d}(R)$ is contained in any prime ideal $\mathfrak{p}$ of $R$ such that $\operatorname{dim}R/\mathfrak{p}=d$ . Now we prove this claim. Take a maximal ideal $\mathfrak{m}$ of $R$ such that it contains $\mathfrak{p}$ and $\operatorname{ht}\mathfrak{m}=d$ . Since $\mathsf{ca}^{d}(R)_{\mathfrak{m}}\subseteq\mathsf{ca}^{d}(R_{\mathfrak{m}})$ , we may assume that $R$ is local. There is a prime ideal $\mathfrak{q}$ of $\widehat{R}$ such that $\operatorname{dim}\widehat{R}/\mathfrak{q}=d$ and $\mathfrak{p}=\mathfrak{q}\cap R$ as $\operatorname{dim}\widehat{R}/\mathfrak{p}\widehat{R}=\operatorname{dim}R/% \mathfrak{p}=d$ . By lemma 2.2, $\mathsf{ca}^{d}(\widehat{R})\subseteq\mathfrak{q}$ implies $\mathsf{ca}^{d}(R)\subseteq\mathfrak{p}$ and thus we may assume that $R$ is complete. There is a Gorenstein local ring $S$ of dimension $d$ such that $R$ is a homomorphic image of $S$ . The inverse image of $\mathfrak{p}$ is a minimal prime ideal of $S$ because $\operatorname{dim}R/\mathfrak{p}=d=\operatorname{dim}S$ . We obtain $\operatorname{Hom}_{S}(R,S)_{\mathfrak{p}}\neq 0$ and hence $\mathsf{Ann}_{R}\operatorname{Hom}_{S}(R,S)\subseteq\mathfrak{p}$ . An analogous argument to the proof of [10, Proposition 2.6] shows that $\mathsf{ca}^{d}(R)$ is contained in $\mathsf{Ann}_{R}H_{\mathfrak{m}}^{d}(R)=\mathsf{Ann}_{R}\operatorname{Hom}_{S}% (R,S)$ . The proof of the claim is now completed. In particular, if $R$ is as in Corollary 2.6 and it is reduced, then we have $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{d+1}(R))\subsetneq\mathrm{V}(% \mathsf{ca}^{d}(R))$ .

(3) Corollary 2.6 improves [8, Theorems 5.3 and 5.4], that is to say, Corollary 2.6 replaces $2d+1$ with $d+1$ . Indeed, if $R$ is an equicharacteristic excellent local ring of dimension $d$ , the equality $\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{d+1}(R))$ immediately follows from [8, Theorem 5.3] and Corollary 2.6 in the case where $R$ is local. On the other hand, If $R$ is a localization of an affine algebra over a field, it is excellent and a homomorphic image of a finite-dimensional Gorenstein ring. So, the same equality is a consequence of [8, Theorem 5.4] and Corollary 2.6 in the case where the ring admits a dualizing complex.

(4) Suppose that $R$ is either an affine algebra over a field or an equicharacteristic complete local ring, of dimension $d$ . It is well-known by the Jacobian criterion that $\mathrm{V}(J)$ contains $\operatorname{Sing}R$ when $R$ is equidimensional, where $J$ is the Jacobian ideal of $R$ ; see Definition 3.3. By this and Corollary 2.6, we have $\mathrm{V}(\mathsf{ca}^{d+1}(R))\subseteq\mathrm{V}(J)$ . Corollary 2.6 deduces a bit weaker version of [9, Theorem 1.1], which asserts that some power of $J$ annihilates $\operatorname{Ext}_{R}^{d+1}(M,N)$ for all $R$ -modules $M,N$ , not necessarily finitely generated. As mentioned in [11], it is necessary to assume in [9, Theorem 1.1] that R is equidimensional.

We close this section by providing an example of a ring $R$ such that $\mathsf{ca}^{d+1}(R)$ is the defining ideal of the singular locus of $R$ , where $d=\operatorname{dim}R$ .

Example 2.8.

Let $R=k\llbracket X,Y\rrbracket/(X^{2},XY)$ be a quotient of a formal power series ring over a field $k$ . The ring $R$ is a complete local ring of dimension 1 that has an isolated singularity. According to Corollary 2.6, the equalities $\mathrm{V}((X,Y)R)=\operatorname{Sing}R=\mathrm{V}(\mathsf{ca}^{2}(R))$ should hold. The equality $\mathsf{ca}^{3}(R)=(X,Y)R$ was proven in [11, Example 5.7]. We see that $\mathsf{ca}^{2}(R)=(X,Y)R$ by using the method employed there. Since $XR$ is contained in the socle of $R$ , it is also contained in $\mathsf{ca}^{1}(R)$ by [8, Example 2.6]. We have only to show $YR\subseteq\mathsf{ca}^{2}(R)$ . For any $m\geq 2$ , $R/(X,Y^{m})R$ is not a submodule of any free $R$ -module. In fact, if a submodule of a free $R$ -module is annihilated by $(X,Y^{m})R$ , then it is also annihilated by $(X,Y)R$ . Let $L,M\in\operatorname{mod}R$ and let $N$ be a first syzygy of $M$ , which is minimal. We get $XN=0$ as $XR$ is contained in the socle of $R$ and $N$ is minimal. This means that $N$ is a finitely generated module over $R/XR\cong k\llbracket Y\rrbracket$ , which is PID. Hence $N$ is a finite direct sum of $R/XR$ and $k$ because $R/(X,Y^{m})R$ is not a submodule of any free $R$ -module for any $m\geq 2$ . We see that $Y\operatorname{Ext}_{R}^{1}(k,L)=0$ and the exact sequence $0\to k\cong XR\to R\to R/XR\to 0$ induces an exact sequence $\operatorname{Hom}_{R}(k,L)\to\operatorname{Ext}_{R}^{1}(R/XR,L)\to% \operatorname{Ext}_{R}^{1}(R,L)=0$ , which implies $Y\operatorname{Ext}_{R}^{1}(R/XR,L)=0$ . We obtain $Y\operatorname{Ext}_{R}^{2}(M,L)=Y\operatorname{Ext}_{R}^{1}(N,L)=0$ and conclude that $YR$ is contained in $\mathsf{ca}^{2}(R)$ .

3. Jacobian ideals and singular loci

In this section, we study the relationship between Jacobian ideals and singular loci for affine algebras over a field and equicharacteristic complete local rings, and characterize the equidimensionality of these rings in those terms. For the time being, we will focus on defining the notations used in the following sections and providing remarks about them.

Definition 3.1.

Let $R$ be a ring, and $m,n\geq 0$ .

(1) Let $A$ be an $m\times n$ matrix over $R$ . We denote by $I_{r}(A)$ the ideal generated by the $r$ -minors of $A$ (i.e., the determinants of the $r\times r$ submatrices) for $1\leq r\leq{\rm min}\{m,n\}$ . We also set $I_{r}(A)=0$ for $r>{\rm min}\{m,n\}$ and $I_{r}(A)=R$ for $r\leq 0$ .

(2) Let $S$ be either the polynomial ring $R[X_{1},\ldots,X_{m}]$ or the formal power series ring $R\llbracket X_{1},\ldots,X_{m}\rrbracket$ . For $f\in S$ and $1\leq i\leq m$ , the partial derivative of $f$ with respect to $X_{i}$ is denoted by $\partial f/\partial X_{i}$ (or $\frac{\partial f}{\partial X_{i}}$ ). That is, if $f=\sum_{k=(k_{1},\cdots,k_{m})}a_{k}X_{1}^{k_{1}}\cdots X_{m}^{k_{m}}\in S$ , then $\partial f/\partial X_{i}=\sum_{k}k_{i}a_{k}X_{1}^{k_{1}}\cdots X_{i}^{k_{i}-1% }\cdots X_{m}^{k_{m}}$ , where $a_{k}\in R$ . For $f_{1},\ldots,f_{n}\in S$ , the $m\times n$ matrix

\displaystyle\left(\frac{\partial f_{j}}{\partial X_{i}}\right)=\begin{pmatrix% }\dfrac{\partial f_{1}}{\partial X_{1}}&\cdots&\dfrac{\partial f_{n}}{\partial X% _{1}}\\ \vdots&\ddots&\vdots\\ \dfrac{\partial f_{1}}{\partial X_{m}}&\cdots&\dfrac{\partial f_{n}}{\partial X% _{m}}\\ \end{pmatrix}

is called the Jacobian matrix of $f_{1},\ldots,f_{n}$ .

(3) Let $k$ be a field. For an affine $k$ -algebra $R$ , that is $R$ is a finitely generated algebra over $k$ , we define $\operatorname{codim}_{k}(R)={\rm inf}\{m-\operatorname{dim}R\mid$ there is a surjective $k$ -algebra homomorphism from $k[X_{1},\ldots,X_{m}]$ to $R\}$ . For an equicharacteristic complete local ring $R$ with residue field $k$ , we also define $\operatorname{codim}(R)={\rm inf}\{m-\operatorname{dim}R\mid$ there is a surjective ring homomorphism from $k\llbracket X_{1},\ldots,X_{m}\rrbracket$ to $R\}$ .

(4) When $R$ is a finite dimensional ring, we denote by $\operatorname{edd}R$ the supremum of $\operatorname{dim}R-\operatorname{dim}R/\mathfrak{p}$ over all minimal prime ideals $\mathfrak{p}$ of $R$ . (Note that $R$ is equidimensional if and only if $\operatorname{edd}R=0$ .)

Remark 3.2.

(1) For an affine algebra $R$ , $\operatorname{codim}_{k}(R)$ depends on a field $k$ . Indeed, $\operatorname{codim}_{\mathbb{R}}(\mathbb{C})=1$ and $\operatorname{codim}_{\mathbb{C}}(\mathbb{C})=0$ , where $\mathbb{R}$ is the field of real numbers and $\mathbb{C}$ is the field of complex numbers. On the other hand, when an equicharacteristic complete local ring $R$ is given, $\operatorname{codim}R$ is uniquely determined, and $\operatorname{codim}R+\operatorname{dim}R$ coincides with the embedding dimension of $R$ .

(2) Let $R$ be a finite dimensional ring, and let $V_{1},\ldots,V_{n}$ be all the maximal irreducible closed subsets of $\operatorname{Spec}R$ . Put $v_{i}={\rm sup}\{m\geq 0\mid$ there exists a chain $V_{i}=W_{0}\supsetneq\cdots\supsetneq W_{m}$ of irreducible closed subsets of $\operatorname{Spec}R\}$ . By definition, we see that $\operatorname{edd}R={\rm sup}\{v_{i}\mid 1\leq i\leq n\}-{\rm inf}\{v_{i}\mid 1% \leq i\leq n\}$ . So, $\operatorname{edd}R$ is characterized in terms of the topology of $\operatorname{Spec}R$ . This means that for any ring $S$ such that $\operatorname{Spec}S$ is homeomorphic to $\operatorname{Spec}R$ , one has the equality $\operatorname{edd}S=\operatorname{edd}R$ .

Definition 3.3.

Let $k$ be a field.

(1) Let $R$ be an affine $k$ -algebra. Suppose that $\phi:S=k[X_{1},\ldots,X_{m}]\to R$ is a surjective $k$ -algebra homomorphism. We set $\operatorname{Ker}\phi=(f_{1},\ldots,f_{r})$ and $d=m-\operatorname{dim}R$ , where $f_{1},\ldots,f_{r}\in S$ . For each integer $n$ , the ideal $\phi(I_{n+d}(\partial f_{j}/\partial X_{i}))$ is uniquely determined regardless of the choices of $S$ , $\phi$ , and $f_{1},\ldots,f_{r}$ ; see Remark 3.4. We denote this ideal by $J_{n}^{k}(R)$ . In particular, $J_{0}^{k}(R)$ is called the Jacobian ideal of $R$ (over $k$ ) and is denoted by $\operatorname{jac}_{k}R$ .

(2) Let $R$ be an equicharacteristic complete local ring with residue field $k$ . Suppose that $\phi:S=k\llbracket X_{1},\ldots,X_{m}\rrbracket\to R$ is a surjective ring homomorphism. We set $\operatorname{Ker}\phi=(f_{1},\ldots,f_{r})$ and $d=m-\operatorname{dim}R$ , where $f_{1},\ldots,f_{r}\in S$ . For each integer $n$ , the ideal $\phi(I_{n+d}(\partial f_{j}/\partial X_{i}))$ is uniquely determined regardless of the choices of $S$ , $\phi$ , and $f_{1},\ldots,f_{r}$ ; see Remark 3.4 and Proposition 4.3. Similarly, we denote this ideal by $J_{n}(R)$ . In particular, $J_{0}(R)$ is called the Jacobian ideal of $R$ and is denoted by $\operatorname{jac}R$ .

Remark 3.4.

Let $k$ be a field.

(1) Let $R=k[X_{1},\ldots,X_{m}]/(f_{1},\ldots,f_{r})$ and $S=k[Y_{1},\ldots,Y_{n}]/(g_{1},\ldots,g_{s})$ . Suppose that there is an isomorphism $\phi:R\xrightarrow{\cong}S$ of $k$ -algebras. Then $\phi$ induces the natural isomorphism $\Omega_{R/k}\xrightarrow{\cong}\Omega_{S/k}$ of modules over $R$ and $S$ , where $\Omega_{R/k}$ and $\Omega_{S/k}$ are the module of Kähler differentials of $R$ and $S$ over $k$ , respectively. There exist exact sequences

R^{\oplus r}\xrightarrow{\bigl{(}\tfrac{\partial f_{j}}{\partial X_{i}}\bigr{)% }}R^{\oplus m}\to\Omega_{R/k}\to 0\ {\rm and}\ S^{\oplus s}\xrightarrow{\bigl{% (}\tfrac{\partial g_{j}}{\partial Y_{i}}\bigr{)}}S^{\oplus n}\to\Omega_{S/k}% \to 0;

see [6, Section 16] and [7, Section 25] for instance. We have $\phi(I_{m-l}(\partial f_{j}/\partial X_{i}))=I_{n-l}(\partial g_{j}/\partial Y% _{i})$ for any integer $l$ since these ideals are $l$ -th Fitting invariants. This means that for an affine $k$ -algebra $T$ and an integer $u$ , the ideal $J_{u}^{k}(T)$ of $T$ is well-defined. The ideal $J_{u}(T)$ is also well-defined for an equicharacteristic complete local ring $T$ with residue field $k$ . A proof of this fact, which does not use Kähler differentials, is provided in Section 4.

(2) Let $(R,\mathfrak{m})$ be a local affine $k$ -algebra with residue field $k$ . Then $R$ is Artinian since the maximal ideal is nilpotent by [7, Theorem 5.5]. Since $R=k+\mathfrak{m}$ , we can choose a surjective homomorphism $\phi:k[X_{1},\ldots,X_{m}]\to R$ of $k$ -algebras such that $m=\operatorname{codim}_{k}(R)+\operatorname{dim}R$ and $\phi(X_{1},\ldots,X_{m})\subseteq\mathfrak{m}$ . Considering the $(X_{1},\ldots,X_{m})$ -adic completion, we see that $\operatorname{codim}(R)\leq\operatorname{codim}_{k}(R)$ . On the other hand, let $\phi:k\llbracket X_{1},\ldots,X_{m}\rrbracket\to R$ be a surjective ring homomorphism such that $m=\operatorname{codim}(R)+\operatorname{dim}R$ . As $R$ is Artinian and $m$ is the embedding dimension of $R$ , $(X_{1},\ldots,X_{m})^{n}\subseteq\operatorname{Ker}\phi\subseteq(X_{1},\ldots,% X_{m})^{2}$ for some $n\geq 2$ . The natural surjection from $k[X_{1},\ldots,X_{m}]/(X_{1},\ldots,X_{m})^{n}=k\llbracket X_{1},\ldots,X_{m}% \rrbracket/(X_{1},\ldots,X_{m})^{n}$ to $R$ is induced by $\phi$ , which means that $k[X_{1},\ldots,X_{m}]/(f_{1},\ldots,f_{r})=k\llbracket X_{1},\ldots,X_{m}% \rrbracket/(f_{1},\ldots,f_{r})\cong R$ for some $f_{1},\ldots,f_{r}\in(X_{1},\ldots,X_{m})^{2}\subseteq k[X_{1},\ldots,X_{m}]$ . This implies that the equalities $\operatorname{codim}_{k}(R)=\operatorname{codim}(R)$ and $J_{u}^{k}(R)=J_{u}(R)$ hold for any integer $u$ .

(3) Let $R$ be a ring, and $A$ an $m\times n$ matrix over $R$ . We see that $I_{r+1}(A)\subseteq I_{r}(A)$ by the Laplace expansion. Hence we have $J_{r+1}^{k}(R)\subseteq J_{r}^{k}(R)$ and $J_{r+1}(S)\subseteq J_{r}(S)$ for any affine $k$ -algebra $R$ , any an equicharacteristic complete local ring $S$ with residue field $k$ , and any integer $r$ .

Recall well-known facts about the tensor product of quotient rings of polynomial rings and the completed tensor product of quotient rings of power series rings. The equalities regarding Krull dimensions of rings are obtained by considering Noether normalization and the dimension formula for flat extensions of local rings, respectively; see [1, Theorems A.11. and A.14.] for instance.

Lemma 3.5.

Let $k$ be a field.

(1)

Let $R=k[X_{1},\ldots,X_{m}]/(f_{1},\ldots,f_{r})$ and $S=k[Y_{1},\ldots,Y_{n}]/(g_{1},\ldots,g_{s})$ . Then the tensor product $R\otimes_{k}S$ of $R$ and $S$ over $k$ is isomorphic to $k[X_{1},\ldots,X_{m},Y_{1},\ldots,Y_{n}]/(f_{1},\ldots,f_{r},g_{1},\ldots,g_{s})$ as $k$ -algebras in a natural way. Also, the natural ring homomorphisms $R\to R\otimes_{k}S$ and $S\to R\otimes_{k}S$ are flat, and the equality $\operatorname{dim}R\otimes_{k}S=\operatorname{dim}R+\operatorname{dim}S$ holds.
(2)

Let $R=k\llbracket X_{1},\ldots,X_{m}\rrbracket/(f_{1},\ldots,f_{r})$ and $S=k\llbracket Y_{1},\ldots,Y_{n}\rrbracket/(g_{1},\ldots,g_{s})$ . Then the completed tensor product $R\ \widehat{\otimes}_{k}\ S$ of $R$ and $S$ over $k$ is isomorphic to $k\llbracket X_{1},\ldots,X_{m},Y_{1},\ldots,Y_{n}\rrbracket/(f_{1},\ldots,f_{r% },g_{1},\ldots,g_{s})$ in a natural way. Also, the natural ring homomorphisms $R\to R\ \widehat{\otimes}_{k}\ S$ and $S\to R\ \widehat{\otimes}_{k}\ S$ are flat and local, and the equality $\operatorname{dim}R\ \widehat{\otimes}_{k}\ S=\operatorname{dim}R+% \operatorname{dim}S$ holds.

Let $R$ be an affine algebra over a field $k$ , and let $n$ be an integer. If $\operatorname{edd}R\leq n$ , then the conditions $\operatorname{Sing}R\subseteq\mathrm{V}(J_{n}^{k}(R))$ and $\operatorname{Spec}(R)=\mathrm{V}(J_{n+1}^{k}(R))$ hold; see Section 1 or Propositions 3.9 and 3.10. Theorem 3.6 states that these conditions are deeply interconnected. In fact, for example, we put $A=k[x]/(x^{2})$ and $R^{\prime}=A\otimes_{k}R$ . Then we see that there is an equality $\operatorname{edd}R=\operatorname{edd}R^{\prime}$ and that $\operatorname{Sing}R^{\prime}\subseteq\mathrm{V}(J_{n}^{k}(R^{\prime}))$ holds if and only if $\operatorname{Spec}(R)=\mathrm{V}(J_{n+1}^{k}(R))$ . Note that $\operatorname{Sing}(R)\subseteq\mathrm{V}(J_{n}^{k}(R))$ is neither a necessary nor a sufficient condition for $\operatorname{Spec}(R)=\mathrm{V}(J_{n+1}^{k}(R))$ to hold.; see Example 3.8.

Theorem 3.6.

Let $k$ be a field, $R$ an affine $k$ -algebra. For any integer $n>0$ , the following are equivalent:

(1)

Suppose that $A\to B$ is a flat homomorphism of affine $k$ -algebras such that $a=\operatorname{codim}_{k}A\leq n$ . If there is $\mathfrak{m}\in\operatorname{Sing}A$ such that $B/\mathfrak{m}B\cong R$ as $k$ -algebras and $\operatorname{ht}\mathfrak{m}=\operatorname{dim}A$ , then $\operatorname{Sing}B\nsubseteq\mathrm{V}(J_{n-a}^{k}(B))$ ;
(2)

Suppose that $A\to B$ is a flat homomorphism of affine $k$ -algebras such that $\operatorname{codim}_{k}A\leq n$ . If there is $\mathfrak{m}\in\operatorname{Sing}A$ such that $B/\mathfrak{m}B\cong R$ as $k$ -algebras and $\operatorname{ht}\mathfrak{m}=\operatorname{dim}A$ , then $\operatorname{Sing}B\nsubseteq\mathrm{V}(\operatorname{jac}_{k}(B))$ ;
(3)

Let $A$ be an affine $k$ -algebra such that $\operatorname{codim}_{k}A\leq n$ . If there is $\mathfrak{m}\in\operatorname{Sing}A$ such that $A/\mathfrak{m}\cong k$ and $\operatorname{ht}\mathfrak{m}=\operatorname{dim}A$ , then $\operatorname{Sing}(A\otimes_{k}R)\nsubseteq\mathrm{V}(\operatorname{jac}_{k}(% A\otimes_{k}R))$ ;
(4)

$\operatorname{Sing}(A\otimes_{k}R)\nsubseteq\mathrm{V}(\operatorname{jac}_{k}(% A\otimes_{k}R))$ for some local affine $k$ -algebra $A$ with residue field $k$ such that $\operatorname{codim}_{k}A$ is equal to $n$ ;
(5)

$\operatorname{Sing}(A\otimes_{k}R)\nsubseteq\mathrm{V}(J_{n-a}^{k}(A\otimes_{k% }R))$ for some local affine $k$ -algebra $A$ with residue field $k$ such that $a=\operatorname{codim}_{k}A\leq n$ ;
(6)

$\operatorname{Spec}(R)\neq\mathrm{V}(J_{n}^{k}(R))$ holds.

Proof.

The implication (1) $\Rightarrow$ (2) holds since $J_{n-a}^{k}(B)\subseteq J_{0}^{k}(B)=\operatorname{jac}_{k}(B)$ by Remark 3.4. Also, (2) $\Rightarrow$ (3) immediately follows from Lemma 3.5. Setting $A=k[X_{1},\ldots,X_{n}]/(X_{1},\ldots,X_{n})^{2}$ , it follows that (3) $\Rightarrow$ (4). It is clear that (4) $\Rightarrow$ (5) holds.

We prove (5) $\Rightarrow$ (6). Suppose that $\operatorname{Sing}(A\otimes_{k}R)\nsubseteq\mathrm{V}(J_{n-a}^{k}(A\otimes_{k% }R))$ for some a local affine $k$ -algebra with residue field $k$ such that $a=\operatorname{codim}_{k}A\leq n$ . Thanks to Remark 3.4(2), $A$ is Artinian and we can put $A=k[X_{1},\ldots,X_{a}]/(f_{1},\ldots,f_{r})$ , for some $f_{1},\ldots,f_{r}\in(X_{1},\ldots,X_{a})^{2}$ . We set $R=k[Y_{1},\ldots,Y_{m}]/(g_{1},\ldots,g_{s})$ . Then $B:=k[X_{1},\ldots,X_{a},Y_{1},\ldots,Y_{m}]/(f_{1},\ldots,f_{r},g_{1},\ldots,g% _{s})\cong A\otimes_{k}R$ and $\operatorname{dim}B=\operatorname{dim}R$ by Lemma 3.5. Since $f_{1},\ldots,f_{r}\in k[X_{1},\ldots,X_{a}]$ and $g_{1},\ldots,g_{s}\in k[Y_{1},\ldots,Y_{m}]$ , $J_{n-a}^{k}(B)$ is generated by the $(n+m-\operatorname{dim}R)$ -minors of the $(a+m)\times(r+s)$ matrix

\displaystyle\begin{pmatrix}\Bigl{(}\dfrac{\partial f_{j}}{\partial X_{i}}% \Bigr{)}&0\\ 0&\Bigl{(}\dfrac{\partial g_{j}}{\partial Y_{i}}\Bigr{)}\\ \end{pmatrix}

over $B$ . For any $1\leq j\leq r$ , $f_{j}$ belongs to $(X_{1},\ldots,X_{a})^{2}$ , and thus $\partial f_{j}/\partial X_{i}$ are in $(X_{1},\ldots,X_{a})$ for all $1\leq i\leq a$ . We obtain $J_{n-a}^{k}(B)\subseteq I_{n+m-\operatorname{dim}R}(\partial g_{j}/\partial Y_% {i})B+(X_{1},\ldots,X_{a})B=J_{n}^{k}(R)B+(X_{1},\ldots,X_{a})B$ . Let $\mathfrak{p}\in\operatorname{Sing}B\setminus\mathrm{V}(J_{n-a}^{k}(B))$ . Since $A$ is Artinian, we get $(X_{1},\ldots,X_{a})B\subseteq\mathfrak{p}$ , which means $J_{n}^{k}(R)B\nsubseteq\mathfrak{p}$ . Hence $\mathfrak{p}\cap R\notin\mathrm{V}(J_{n}^{k}(R))$ .

In order to show (6) $\Rightarrow$ (1), we assume $\operatorname{Spec}(R)\neq\mathrm{V}(J_{n}^{k}(R))$ . Set $A=k[X_{1},\ldots,X_{l}]/(f_{1},\ldots,f_{r})$ , where $l=a+\operatorname{dim}A$ . We can write $B=k[X_{1},\ldots,X_{l},Y_{1},\ldots,Y_{m}]/(f_{1},\ldots,f_{r},g_{1},\ldots,g_% {s})$ for some polynomials $g_{1},\ldots,g_{s}\in k[X_{1},\ldots,X_{l},Y_{1},\ldots,Y_{m}]$ . Since $\mathfrak{m}\in\operatorname{Spec}A$ , we can choose polynomials $h_{1},\ldots,h_{t}\in k[X_{1},\ldots,X_{l}]$ such that $\mathfrak{m}=(h_{1},\ldots,h_{t})A$ and $(f_{1},\ldots,f_{r})\subseteq(h_{1},\ldots,h_{t})$ . Then we see that $R\cong B/\mathfrak{m}B=k[X_{1},\ldots,X_{l},Y_{1},\ldots,Y_{m}]/(h_{1},\ldots,% h_{t},g_{1},\ldots,g_{s})$ . Put $e=l+m-\operatorname{dim}R$ . For

\displaystyle U=\begin{pmatrix}\Bigl{(}\dfrac{\partial h_{j}}{\partial X_{i}}% \Bigr{)}&\Bigl{(}\dfrac{\partial g_{j}}{\partial X_{i}}\Bigr{)}\\ 0&\Bigl{(}\dfrac{\partial g_{j}}{\partial Y_{i}}\Bigr{)}\\ \end{pmatrix},

it is seen that $J_{n}^{k}(R)=I_{n+e}(U)R$ . Note that $(\partial h_{j}/\partial X_{i})$ is an $l\times t$ matrix. The $(n+e)$ -minors of $U$ are of the form

\displaystyle T=\begin{pmatrix}V&\ast\\ 0&W\\ \end{pmatrix},

where $V$ is a $p\times q$ submatrix of $(\partial h_{j}/\partial X_{i})$ , $W$ is an $(n+e-p)\times(n+e-q)$ submatrix of $(\partial g_{j}/\partial Y_{i})$ , $1\leq p\leq l$ and $1\leq q\leq t$ . By the Laplace expansion, $\operatorname{det}T=0$ if $p<q$ , and $\operatorname{det}T\in I_{n+e-p}(\partial g_{j}/\partial Y_{i})$ if $p\geq q$ . Therefore $J_{n}^{k}(R)$ is contained in $I_{n+e-l}(\partial g_{j}/\partial Y_{i})R$ . By $\operatorname{Spec}(R)\neq\mathrm{V}(J_{n}^{k}(R))$ , $J_{n}^{k}(R)$ is nonzero, and hence $n+e-l\leq{\rm min}\{m,s\}$ .

Let $\mathfrak{p}$ be a prime ideal of $R$ which does not contain $J_{n}^{k}(R)$ , and let $\mathfrak{q}$ be a prime ideal of $B$ such that $\mathfrak{p}=\mathfrak{q}/\mathfrak{m}B$ . The ideal $J_{n}^{k}(R)$ is not contained in $\mathfrak{p}$ , and neither is $I_{n+e-l}(\partial g_{j}/\partial Y_{i})R$ , which means that $\mathfrak{q}$ does not contain $I_{n+e-l}(\partial g_{j}/\partial Y_{i})B$ . As $\mathfrak{q}\cap A$ of $A$ contains the maximal ideal $\mathfrak{m}$ , $\mathfrak{q}\cap A=\mathfrak{m}$ and the natural homomorphism $A_{\mathfrak{m}}\to B_{\mathfrak{q}}$ is flat and local. We get $\mathfrak{q}\in\operatorname{Sing}B$ because $\mathfrak{m}\in\operatorname{Sing}A$ .

We see that $\operatorname{dim}B\geq\operatorname{dim}A+\operatorname{dim}R$ . In fact, let $Q$ be a prime ideal of $R$ such that $\operatorname{ht}Q=\operatorname{dim}R$ , and let $P$ be a prime ideal of $B$ such that $Q=P/\mathfrak{m}B$ . Similarly, $P\cap A=\mathfrak{m}$ . Since $A\to B$ is flat, we have $\operatorname{dim}B\geq\operatorname{ht}P=\operatorname{ht}\mathfrak{m}+% \operatorname{ht}P/\mathfrak{m}B=\operatorname{dim}A+\operatorname{dim}R$ . We obtain $d:=l+m-\operatorname{dim}B\leq(l-\operatorname{dim}A)+(m-\operatorname{dim}R)=% a+e-l$ , and hence $d+n-a\leq n+e-l\leq{\rm min}\{m,s\}$ . The ideal $J_{n-a}^{k}(B)$ is generated by the $(d+n-a)$ -minors of the $(n+m)\times(r+s)$ matrix

\displaystyle\begin{pmatrix}\Bigl{(}\dfrac{\partial f_{j}}{\partial X_{i}}% \Bigr{)}&\Bigl{(}\dfrac{\partial g_{j}}{\partial X_{i}}\Bigr{)}\\ 0&\Bigl{(}\dfrac{\partial g_{j}}{\partial Y_{i}}\Bigr{)}\\ \end{pmatrix}

over $B$ . Therefore $I_{n+e-l}(\partial g_{j}/\partial Y_{i})B\subseteq I_{d+n-a}(\partial g_{j}/% \partial Y_{i})B\subseteq J_{n-a}^{k}(B)$ , which means that $\mathfrak{q}$ does not contain $J_{n-a}^{k}(B)$ but belongs to $\operatorname{Sing}B$ . ∎

The following theorem is an equicharacteristic complete local version of Theorem 3.6. The proof is omitted since it is similar to (or simpler than) that of the above theorem.

Theorem 3.7.

Let $k$ be a field, $R$ an equicharacteristic complete local ring with residue field $k$ . For any integer $n>0$ , the following are equivalent:

(1)

Suppose that $(A,\mathfrak{m},k)\to(B,\mathfrak{n},k)$ is a flat and local homomorphism of equicharacteristic complete local rings such that $B/\mathfrak{m}B\cong R$ . If $0<a=\operatorname{codim}A\leq n$ , then $\operatorname{Sing}B\nsubseteq\mathrm{V}(J_{n-a}(B))$ ;
(2)

Suppose that $(A,\mathfrak{m},k)\to(B,\mathfrak{n},k)$ is a flat and local homomorphism of equicharacteristic complete local rings such that $B/\mathfrak{m}B\cong R$ . If $0<\operatorname{codim}A\leq n$ , then $\operatorname{Sing}B\nsubseteq\mathrm{V}(\operatorname{jac}(B))$ ;
(3)

Let $A$ be an equicharacteristic complete local ring with residue field $k$ such that $0<\operatorname{codim}A\leq n$ . Then $\operatorname{Sing}(A\widehat{\otimes}_{k}R)\nsubseteq\mathrm{V}(\operatorname% {jac}(A\widehat{\otimes}_{k}R))$ ;
(4)

$\operatorname{Sing}(A\widehat{\otimes}_{k}R)\nsubseteq\mathrm{V}(\operatorname% {jac}(A\widehat{\otimes}_{k}R))$ for some Artinian equicharacteristic local ring $A$ with residue field $k$ such that $\operatorname{codim}A$ is equal to $n$ ;
(5)

$\operatorname{Sing}(A\widehat{\otimes}_{k}R)\nsubseteq\mathrm{V}(J_{n-a}(A% \widehat{\otimes}_{k}R))$ for some Artinian equicharacteristic local ring $A$ with residue field $k$ such that $a=\operatorname{codim}A\leq n$ ;
(6)

$\operatorname{Spec}(R)\neq\mathrm{V}(J_{n}(R))$ holds.

Example 3.8.

Let $R=k[X,Y_{1},\ldots,Y_{n+1}]/(XY_{1},\ldots,XY_{n+1})$ be a quotient of a polynomial ring over a field $k$ , where $n\geq 1$ . The Jacobian matrix of $XY_{1},\ldots,XY_{n+1}$ is the $(n+1)\times(n+2)$ matrix

\displaystyle T=\begin{pmatrix}Y_{1}&X&0&\cdots&0\\ Y_{2}&0&X&\ddots&\vdots\\ \vdots&\vdots&\ddots&\ddots&0\\ Y_{n+1}&0&\cdots&0&X\\ \end{pmatrix}.

Since $\operatorname{codim}_{k}R=1$ , we obtain the equalities $J_{n-1}^{k}(R)=I_{n}(T)R=(X^{n},X^{n-1}Y_{1},\ldots,X^{n-1}Y_{n+1})R$ and $J_{n}^{k}(R)=I_{n+1}(T)R=(X^{n+1},X^{n}Y_{1},\ldots,X^{n}Y_{n+1})R$ . Hence one has $\operatorname{Sing}R=\mathrm{V}((X,Y_{1},\ldots,Y_{n+1})R)\subseteq\mathrm{V}(% J_{n-1}^{k}(R))$ and $\operatorname{Spec}R\neq\mathrm{V}(J_{n}^{k}(R))$ as the prime ideal $(Y_{1},\ldots,Y_{n+1})R$ of $R$ does not belong to $\mathrm{V}(J_{n}^{k}(R))$ . Fix $m\leq n$ . Let $A=k[Z_{1},\ldots,Z_{m}]/(Z_{1}^{2},\ldots,Z_{m}^{2})$ be a quotient of a polynomial ring over $k$ . We put $S:=R\otimes_{k}A\cong k[X,Y_{1},\ldots,Y_{n+1},Z_{1},\ldots,Z_{m}]/(XY_{1},% \ldots,XY_{n+1},Z_{1}^{2},\ldots,Z_{m}^{2})$ . The Jacobian matrix of $XY_{1},\ldots,XY_{n+1},Z_{1}^{2},\ldots,Z_{m}^{2}$ is the $(m+n+1)\times(m+n+2)$ matrix

\displaystyle U=\begin{pmatrix}T&0&\cdots&0\\ 0&2Z_{1}&\ddots&\vdots\\ \vdots&\ddots&\ddots&0\\ 0&\cdots&0&2Z_{m}\\ \end{pmatrix}.

It is seen that $\operatorname{codim}_{k}A=m$ and $\operatorname{codim}_{k}S=m+1$ hold. We have $J_{n-m}^{k}(S)=I_{n+1}(U)S\supseteq X^{n+1}S$ and $J_{n-m+1}^{k}(S)=I_{n+2}(U)S\subseteq(Z_{1},\ldots,Z_{m})S$ . This means $\operatorname{Sing}S=\operatorname{Spec}S\nsubseteq\mathrm{V}(J_{n-m}^{k}(S))$ and $\operatorname{Spec}S=\mathrm{V}(J_{n-m+1}^{k}(S))$ as the prime ideal $(Y_{1},\ldots,Y_{n+1},Z_{1},\ldots,Z_{m})S$ of $S$ does not belong to $\mathrm{V}(J_{n-m}^{k}(S))$ .

Let $R$ be either an affine algebra over a field or an equicharacteristic complete local ring, of dimension $d$ . The purpose of the remainder of this section is to characterize $\operatorname{edd}R$ in terms of ideals generated by the minors of the Jacobian matrix. The proposition below is a classical result known as a corollary of the Jacobian criterion; see [7, Theorem 30.3] for instance. By [12, Theorem 5.4] and [13, Theorem 3.7], under some assumptions, and by [9, Theorem 1.1], in the general case, it was proved that $\mathsf{ca}^{d+1}(R)$ contains some power of the Jacobian ideal of $R$ when $\operatorname{edd}R=0$ . This fact implies Proposition 3.9 in the case $\operatorname{edd}R=0$ ; see [8, Lemma 2.10(2)]. (Indeed, thanks to Corollary 2.6, these assertions are now equivalent.) The proofs of the above result in [9] do not require the theory of smoothness. Here, we provide an elementary proof of Proposition 3.9 using only the result of Iyengar and Takahashi [9].

Proposition 3.9.

Let $k$ be a field.

(1)

Let $R$ be an affine $k$ -algebra. Then $\operatorname{Sing}R\subseteq\mathrm{V}(J_{\operatorname{edd}R}^{k}(R))$ .
(2)

Let $R$ be an equicharacteristic complete local ring with residue field $k$ . Then $\operatorname{Sing}R\subseteq\mathrm{V}(J_{\operatorname{edd}R}(R))$ .

Proof.

The proof is by induction on $e:=\operatorname{edd}R$ . When $e=0$ , we have $J_{0}^{k}(R)=\operatorname{jac}_{k}(R)$ . It follows from [8, Lemma 2.10(2)] and [9, Theorem 1.1] that $\operatorname{Sing}R\subseteq\mathrm{V}(\mathsf{ca}(R))\subseteq\mathrm{V}(% \operatorname{jac}_{k}(R))$ holds.

Assume $e>0$ . We only prove the affine case. (The local case is proved similarly.) We take a polynomial ring $A=k[X_{1},\ldots,X_{n}]$ over $k$ and an ideal $I=(f_{1},\ldots,f_{r})$ of $A$ such that $A/I=R$ . Let $I=\mathfrak{q}_{1}\cap\cdots\cap\mathfrak{q}_{m}$ be a shortest primary decomposition of $I$ and $\sqrt{\mathfrak{q}_{i}}=\mathfrak{p}_{i}$ for each $1\leq i\leq m$ . We may assume that $\operatorname{dim}R/\mathfrak{q}_{1}=\cdots=\operatorname{dim}R/\mathfrak{q}_{% l}=\operatorname{dim}R$ , $\operatorname{dim}R/\mathfrak{q}_{i}<\operatorname{dim}R$ for all $l+1\leq i\leq m$ , and $\mathfrak{p}_{1},\cdots,\mathfrak{p}_{h}$ are all the minimal prime ideal of $A/I$ . Note that $m\geq h>l$ as $e>0$ . Then $e={\rm sup}\{\operatorname{dim}R-\operatorname{dim}A/\mathfrak{p}_{i}\mid l+1% \leq i\leq h\}$ . Put $\mathfrak{a}=\mathfrak{q}_{l+1}\cap\cdots\cap\mathfrak{q}_{m}=(g_{1},\ldots,g_% {s})$ and $B=k[X_{1},\ldots,X_{n},Y]$ , and let $J=(f_{1},\ldots,f_{r},Yg_{1},\ldots,Yg_{s})$ be an ideal of $B$ . For any ideals $\mathfrak{b},\mathfrak{c}$ of $A$ , it is easy to see that the equalities $(\mathfrak{b},Y)B\cap(\mathfrak{c},Y)B=(\mathfrak{b}\cap\mathfrak{c},Y)B$ , $\mathfrak{b}B\cap\mathfrak{c}B=(\mathfrak{b}\cap\mathfrak{c})B$ , and $(\mathfrak{b},Y)B\cap\mathfrak{c}B=(\mathfrak{b}\cap\mathfrak{c},\mathfrak{c}Y)B$ hold. (Note that $\mathfrak{b}B=\mathfrak{b}[Y]$ and $(\mathfrak{b},Y)B=\mathfrak{b}+YB$ ). Therefore we have $(\mathfrak{q}_{1},Y)B\cap\cdots\cap(\mathfrak{q}_{l},Y)B\cap\mathfrak{q}_{l+1}% B\cap\cdots\cap\mathfrak{q}_{m}B=(I,\mathfrak{a}Y)B=(f_{1},\ldots,f_{r},Yg_{1}% ,\ldots,Yg_{s})=J$ . On the other hand, for any ideal $\mathfrak{b}$ of $A$ , $B/(\mathfrak{b},Y)B$ and $B/\mathfrak{b}B$ are isomorphic to $A/\mathfrak{b}$ and $(A/\mathfrak{b})[Y]$ , respectively. In particular, $\operatorname{dim}B/(\mathfrak{b},Y)B=\operatorname{dim}A/\mathfrak{b}$ , and $\operatorname{dim}B/\mathfrak{b}B=\operatorname{dim}A/\mathfrak{b}+1$ . By this, we can easily see that $(\mathfrak{p}_{1},Y)B,\cdots,(\mathfrak{p}_{l},Y)B,\mathfrak{p}_{l+1}B,\cdots,% \mathfrak{p}_{h}B$ are all the minimal prime ideal of $B/J$ , and that the equalities $\operatorname{dim}(B/J)=\operatorname{dim}R$ and $\operatorname{edd}(B/J)=e-1$ hold. Put $c=n-\operatorname{dim}R$ .

Let $\mathfrak{p}\in\operatorname{Sing}R$ . We prove that $\mathfrak{p}$ belongs to $\mathrm{V}(J_{e}^{k}(R))$ . The ideal $J_{e}^{k}(R)$ is generated by the $(c+e)$ -minors of the Jacobian matrix $(\partial f_{j}/\partial X_{i})$ of $f_{1},\ldots,f_{r}$ . If $c+e>{\rm min}\{n,r\}$ , $J_{e}^{k}(R)=0$ , which means $\mathfrak{p}\in\operatorname{Spec}R=\mathrm{V}(J_{e}^{k}(R))$ . We may assume $c+e\leq{\rm min}\{n,r\}$ . Let $P$ be a prime ideal of $A$ such that $\mathfrak{p}=P/I$ , and let $Q=PB+YB$ be an ideal of $B$ . Since $B/Q\cong A/P$ , $Q$ is a prime ideal containing $J$ . Put $\mathfrak{q}=Q/J$ . We show that $\mathfrak{q}$ is in $\operatorname{Sing}(B/J)$ by considering three cases. First, if $Y$ belongs to $JB_{Q}$ , then $JB_{Q}=(f_{1},\ldots,f_{r},Y)B_{Q}$ . So, $(B/J)_{\mathfrak{q}}\cong B_{Q}/JB_{Q}\cong A_{P}/IA_{P}\cong R_{\mathfrak{p}}$ and thus $(B/J)_{\mathfrak{q}}$ is not regular. In a second step, we assume $JB_{Q}=\mathfrak{a}B_{Q}$ . Then we can easily see that $IA_{P}=\mathfrak{a}A_{P}$ . Consider the following two commutative diagrams:

The right side is a localization of the left side. Since $A/\mathfrak{a}\to B/\mathfrak{a}B$ is flat, $R_{\mathfrak{p}}\to(B/J)_{\mathfrak{q}}$ is flat and local. The ring $R_{\mathfrak{p}}$ is not regular, neither is $(B/J)_{\mathfrak{q}}$ . Finally suppose that $Y\notin JB_{Q}$ and $JB_{Q}\subsetneq\mathfrak{a}B_{Q}$ . Then $g_{j}\notin JB_{Q}$ for some $1\leq j\leq s$ . However, $Yg_{j}\in JB_{Q}$ , which implies that $(B/J)_{\mathfrak{q}}\cong B_{Q}/JB_{Q}$ is not an integral domain, and that it is not a regular local ring.

We obtain $\mathfrak{q}\in\operatorname{Sing}(B/J)$ . By the induction hypothesis, $\operatorname{Sing}(B/J)\subseteq\mathrm{V}(J_{e-1}^{k}(B/J))$ as $\operatorname{edd}(B/J)=e-1$ . By the equality $n+1-\operatorname{dim}(B/J)=n+1-\operatorname{dim}R=c+1$ , the ideal $J_{e-1}^{k}(B/J)$ is generated by the $(c+e)$ -minors of the $(n+1)\times(r+s)$ matrix

\displaystyle\begin{pmatrix}\dfrac{\partial f_{1}}{\partial X_{1}}&\cdots&% \dfrac{\partial f_{r}}{\partial X_{1}}&\dfrac{\partial g_{1}}{\partial X_{1}}Y% &\cdots&\dfrac{\partial g_{s}}{\partial X_{1}}Y\\ \vdots&\ddots&\vdots&\vdots&\ddots&\vdots\\ \dfrac{\partial f_{1}}{\partial X_{n}}&\cdots&\dfrac{\partial f_{r}}{\partial X% _{n}}&\dfrac{\partial g_{1}}{\partial X_{n}}Y&\cdots&\dfrac{\partial g_{s}}{% \partial X_{n}}Y\\ 0&\cdots&0&g_{1}&\cdots&g_{s}\end{pmatrix}

over $B/J$ . Noting that $c+e\leq{\rm min}\{n,r\}$ , $J_{e-1}^{k}(B/J)$ contains $I_{c+e}(\partial f_{j}/\partial X_{i})(B/J)$ , and thus so does $\mathfrak{q}$ . This yields $I_{c+e}(\partial f_{j}/\partial X_{i})B\subset Q=PB+YB=P+YB$ in $B$ , which deduces $I_{c+e}(\partial f_{j}/\partial X_{i})A\subset P$ in $A$ . Therefore we have $\mathfrak{p}\in\mathrm{V}(I_{c+e}(\partial f_{j}/\partial X_{i})R)=\mathrm{V}(% J_{e}^{k}(R))$ . ∎

The result below is also a classical one; see [7, Theorem 30.4] for instance. Using Theorems 3.6 and 3.7, it can also be proven through elementary arguments.

Proposition 3.10.

Let $k$ be a field.

(1)

Let $R$ be an affine $k$ -algebra. Then $\operatorname{Spec}R=\mathrm{V}(J_{\operatorname{edd}R+1}^{k}(R))$ .
(2)

Let $R$ be an equicharacteristic complete local ring with residue field $k$ . Then $\operatorname{Spec}R=\mathrm{V}(J_{\operatorname{edd}R+1}(R))$ .

Proof.

We only prove (1). (The assertion (2) is proved similarly.) Assume $\operatorname{Spec}R\neq\mathrm{V}(J_{\operatorname{edd}R+1}^{k}(R))$ . Take a polynomial ring $S=k[X_{1},\ldots,X_{m}]$ over $k$ and an ideal $I$ of $S$ such that $S/I=R$ . We put $e=\operatorname{edd}R$ , $A=k[Y]/(Y^{2})$ , and $B=k[X_{1},\ldots,X_{m},Y]/(I,Y^{2})\cong A\otimes_{k}R$ . Applying (6) $\Rightarrow$ (1) of Theorem 3.6 to $n=e+1$ , we obtain $\operatorname{Sing}B\nsubseteq\mathrm{V}(J_{e}^{k}(B))$ . Thanks to Proposition 3.9, we get $\operatorname{Sing}B\subseteq\mathrm{V}(J_{\operatorname{edd}B}^{k}(B))$ , which means $\operatorname{edd}B>e$ . Set $T=k[X_{1},\ldots,X_{m},Y]$ and $J=(I,Y^{2})T$ . There exists a minimal prime ideal $\mathfrak{p}/J$ of $B$ such that $\operatorname{dim}B-\operatorname{dim}T/\mathfrak{p}>e$ , where $\mathfrak{p}\in\operatorname{Spec}T$ . Now $Y$ is in $\mathfrak{p}$ . For $\mathfrak{q}:=\mathfrak{p}\cap S$ , we have $J\subseteq\mathfrak{q}T+YT\subseteq\mathfrak{p}$ . The isomorphism $T/(\mathfrak{q}T+YT)\cong S/\mathfrak{q}$ says that $\mathfrak{q}T+YT$ is a prime ideal of $T$ . The equality $\mathfrak{q}T+YT=\mathfrak{p}$ holds since $\mathfrak{p}/J$ is a minimal prime ideal of $T/J$ . By Krull’s height theorem, we have $\operatorname{ht}\mathfrak{p}/\mathfrak{q}T\leq 1$ . Since the ring homomorphisms $S\to T$ and $R=S/I\to A\otimes_{k}B\cong T/J$ are flat, we have $\operatorname{ht}\mathfrak{p}=\operatorname{ht}\mathfrak{q}+\operatorname{ht}% \mathfrak{p}/\mathfrak{q}T\leq\operatorname{ht}\mathfrak{q}+1$ and $0=\operatorname{ht}(\mathfrak{p}/J)\geq\operatorname{ht}(\mathfrak{q}/I)$ ; see [7, Theorem 15.1] for instance. So, $\mathfrak{q}/I$ is a minimal prime ideal of $R$ and thus $\operatorname{dim}R-\operatorname{dim}S/\mathfrak{q}\leq e$ . However $\operatorname{dim}R-\operatorname{dim}S/\mathfrak{q}=(\operatorname{dim}B-% \operatorname{dim}A)-(m-\operatorname{ht}\mathfrak{q})\geq\operatorname{dim}B-% (m+1-\operatorname{ht}\mathfrak{p})=\operatorname{dim}B-\operatorname{dim}T/% \mathfrak{p}>e$ , which is a contradiction. ∎

Propositions 3.9 and 3.10 asserts that for any affine algebra $R$ over a field $k$ and integer $N\geq 0$ , if $\operatorname{edd}R\leq N$ , then $\operatorname{Sing}R\subseteq\mathrm{V}(J_{n}^{k}(R))$ and $\operatorname{Spec}R=\mathrm{V}(J_{n+1}^{k}(R))$ hold for any $n\geq N$ . The example below says that the converse does not hold.

Example 3.11.

Let $S=k[X_{1},\ldots X_{n}]$ be a polynomial ring over a field $k$ , $I$ an ideal of $S$ and $m\geq 2$ . Put $R=S/I^{m}$ and $c=\operatorname{codim}_{k}R$ . Since $\partial f/\partial X_{i}$ belongs to $I^{m-1}$ for any $f\in I^{m}$ and $1\leq i\leq n$ , we have $J_{l}^{k}(R)\subseteq I^{m-1}/I^{m}$ for any $l\geq 1-c$ , which means $\operatorname{Spec}R=\mathrm{V}(J_{l}^{k}(R))$ . On the other hand, $\operatorname{edd}R$ is not necessarily less than or equal to $1-c$ as $I$ is any ideal of $S$ . For example, if $R=k[X,Y,Z]/(XY,XZ)^{2}$ , $\operatorname{Sing}R=\operatorname{Spec}R=\mathrm{V}(J_{l}^{k}(R))$ hold for all $l\geq 0$ , but $\operatorname{edd}R=1$ , which is greater than $0=1-\operatorname{codim}_{k}R$ .

The following theorem is the main result of this section. By Example 3.11, fixing the ring, the converses of Propositions 3.9 and 3.10 does not hold in general. As seen in Remark 3.2(2), $\operatorname{edd}R$ depends only on the spectrum of $R$ . From this perspective, the converses of Propositions 3.9 and 3.10 also hold.

Theorem 3.12.

Let $k$ be a perfect field.

(1)
Let $R=k[X_{1},\ldots,X_{m}]$ be a polynomial ring over $k$ , and $V$ a closed subset of $\operatorname{Spec}R$ of combinatorial dimension $d$ . For every integer $n\geq 0$ , the following are equivalent:
1. (i)
  
  For any mimimal element $\mathfrak{p}$ of $V$ , the inequality $d\leq\operatorname{dim}R/\mathfrak{p}+n$ holds;
2. (ii)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\operatorname{Sing}(R/I)\subseteq\mathrm{V}(J_{n}^{k}(R/I))$ holds;
3. (iii)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\operatorname{Spec}(R/I)=\mathrm{V}(J_{n}^{k}(R/I))$ holds;
4. (iv)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\mathrm{V}(\mathsf{ca}^{d+1}(R/I))\subseteq\mathrm{V}(J_{n}^{k}(R/I))$ holds.
(2)
Let $R=k\llbracket X_{1},\ldots,X_{m}\rrbracket$ be a formal power series ring over $k$ , and $V$ a closed subset of $\operatorname{Spec}R$ of combinatorial dimension $d$ . For every integer $n\geq 0$ , the following are equivalent:
1. (i)
  
  For any mimimal element $\mathfrak{p}$ of $V$ , the inequality $d\leq\operatorname{dim}R/\mathfrak{p}+n$ holds;
2. (ii)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\operatorname{Sing}(R/I)\subseteq\mathrm{V}(J_{n}(R/I))$ holds;
3. (iii)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\operatorname{Spec}(R/I)=\mathrm{V}(J_{n}(R/I))$ holds;
4. (iv)
  
  For any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ , $\mathrm{V}(\mathsf{ca}^{d+1}(R/I))\subseteq\mathrm{V}(J_{n}(R/I))$ holds.

Proof.

Similarly, we only show the assertion (1). For any ideal $I$ of $R$ such that $V=\mathrm{V}(I)$ , $\operatorname{dim}(R/I)=d$ holds. Thanks to Corollary 2.6, there is an equality $\operatorname{Sing}(R/I)=\mathrm{V}(\mathsf{ca}^{d+1}(R/I))$ , which induces (ii) $\Leftrightarrow$ (iv). Suppose that $d\leq\operatorname{dim}R/\mathfrak{p}+n$ holds for any mimimal element $\mathfrak{p}$ of $V$ . Then $e=\operatorname{edd}(R/I)\leq n$ for any ideal $I$ of $R$ satisfying $V=\mathrm{V}(I)$ . It follows from Remark 3.4(3), Propositions 3.9 and 3.10 that $\operatorname{Sing}(R/I)\subseteq\mathrm{V}(J_{e}^{k}(R/I))\subseteq\mathrm{V}% (J_{n}^{k}(R/I))$ and $\operatorname{Spec}(R/I)=\mathrm{V}(J_{e+1}^{k}(R/I))\subseteq\mathrm{V}(J_{n+% 1}^{k}(R/I))$ hold. Hence (i) $\Rightarrow$ (ii) and (i) $\Rightarrow$ (iii) hold.

To prove the converses, assume the existence of a mimimal element $\mathfrak{p}$ of $V$ such that $d>\operatorname{dim}R/\mathfrak{p}+n$ . Put $h=\operatorname{ht}\mathfrak{p}=m-\operatorname{dim}R/\mathfrak{p}$ . Let $\mathfrak{p}=\mathfrak{p}_{0},\mathfrak{p}_{1},\ldots,\mathfrak{p}_{s}$ be all the minimal elements of $V$ . We see that $d=\operatorname{dim}R/\mathfrak{p}_{u}$ for some $1\leq u\leq s$ . Then $\mathfrak{p}\neq 0$ and $\mathfrak{p}_{u}\neq 0$ since $d>\operatorname{dim}R/\mathfrak{p}$ and $\mathfrak{p}$ is minimal. Therefore $h-n=m-\operatorname{dim}R/\mathfrak{p}-n>m-d=m-\operatorname{dim}R/\mathfrak{p% }_{u}=\operatorname{ht}\mathfrak{p}_{u}>0$ . Since $R_{\mathfrak{p}}$ is a regular local ring of dimension $h$ , we can choose $f_{1},\ldots,f_{h}\in\mathfrak{p}$ such that $\operatorname{ht}(f_{1},\ldots,f_{h})=h$ and $(f_{1},\ldots,f_{h})R_{\mathfrak{p}}=\mathfrak{p}R_{\mathfrak{p}}$ ; see the proof of [12, Propositon 4.4]. Put $J=(f_{1},\ldots,f_{h})$ . Then $\mathfrak{p}/J\notin\operatorname{Sing}(R/J)$ and $R/J$ is equidimensional as the heights of all the minimal prime ideals of $J$ are $h$ by Krull’s height theorem. Now $k$ is perfect. It follows from [6, Corollary 16.20] (and in the local case, it follows from [12, Lemma 4.3, and Propositions 4.4 and 4.5]) that $\mathrm{V}(\operatorname{jac}_{k}(R/J))\subseteq\operatorname{Sing}(R/J)$ and thus $\operatorname{jac}_{k}(R/J)\nsubseteq\mathfrak{p}/J$ . We ontain $I_{h}(\partial f_{j}/\partial X_{i})\nsubseteq\mathfrak{p}$ . We take $\gamma\in\bigcap_{i=1}^{s}\mathfrak{p}_{i}\setminus\mathfrak{p}$ .

We prove that condition (ii) does not hold. Let $U_{j}$ be the $m\times(h-1)$ submatrix obtained by removing the $j$ -th column of $(\partial f_{j}/\partial X_{i})$ , that is to say, $U_{j}$ is the Jacobian matrix of $f_{1},\ldots,f_{i-1},f_{i+1},\ldots,f_{h}$ . By the Laplace expansion, we get $I_{h}(\partial f_{j}/\partial X_{i})\subseteq\sum_{j=1}^{h}I_{h-1}(U_{j})$ . We may assume $I_{h-1}(U_{h})\nsubseteq\mathfrak{p}$ as $I_{h}(\partial f_{j}/\partial X_{i})\nsubseteq\mathfrak{p}$ . Set $\mathfrak{q}=(f_{1},\ldots,f_{h-1})+\mathfrak{p}^{2}$ and $I=\mathfrak{q}\cap\mathfrak{p}_{1}\cap\cdots\cap\mathfrak{p}_{s}$ . Then $\mathrm{V}(I)=V$ and $\mathfrak{q}R_{\mathfrak{p}}\subsetneq\mathfrak{p}R_{\mathfrak{p}}$ by [7, Corollary of Theorem 2.2] and $(f_{1},\ldots,f_{h-1})R_{\mathfrak{p}}\subsetneq\mathfrak{p}R_{\mathfrak{p}}$ . There is an isomorphism $R_{\mathfrak{p}}/IR_{\mathfrak{p}}\cong R_{\mathfrak{p}}/\mathfrak{q}R_{% \mathfrak{p}}$ , and the right side is an Artinian local ring that is not a field. This means that $\mathfrak{p}/I$ belongs to $\operatorname{Sing}(R/I)$ . We put $\alpha_{r}=\gamma f_{r}\in I$ for each $1\leq r\leq h-1$ . Calculating the Jacobian matrix of the system of generators of $I$ obtained by extending $\alpha_{1},\ldots,\alpha_{h-1}$ , we see that $I_{n+m-d}(\partial\alpha_{j}/\partial X_{i})(R/I)$ is contained in $J_{n}^{k}(R/I)$ since $h-1\geq n+m-d$ . On the other hand, for any $1\leq r\leq h-1$ and any $1\leq t\leq m$ , the equality

\dfrac{\partial\alpha_{r}}{\partial X_{t}}=\gamma\dfrac{\partial f_{r}}{% \partial X_{t}}+\dfrac{\partial\gamma}{\partial X_{t}}f_{r}

holds. As $f_{r}$ are in $\mathfrak{p}$ for all $1\leq r\leq h-1$ , it is seen that $\gamma^{n+m-d}I_{n+m-d}(U_{h})\subseteq I_{n+m-d}(\partial\alpha_{j}/\partial X% _{i})+\mathfrak{p}$ . By $I_{n+m-d}(U_{h})\supseteq I_{h-1}(U_{h})\nsubseteq\mathfrak{p}$ and $\gamma\notin\mathfrak{p}$ , We conclude that $I_{n+m-d}(\partial\alpha_{j}/\partial X_{i})$ is not contained in $\mathfrak{p}$ , which yields $\mathfrak{p}/I\notin\mathrm{V}(J_{n}^{k}(R/I))$ .

Show that condition (iii) does not hold. We put $J=\mathfrak{p}\cap\mathfrak{p}_{1}\cap\cdots\cap\mathfrak{p}_{s}$ and $\beta_{r}=\gamma f_{r}$ for all $1\leq r\leq h$ . An analogous argument shows that $I_{n+m-d+1}(\partial\beta_{j}/\partial X_{i})(R/J)$ is contained in $J_{n+1}^{k}(R/J)$ and that $\gamma^{n+m-d+1}I_{n+m-d+1}(\partial f_{j}/\partial X_{i})$ is contained in $I_{n+m-d+1}(\partial\beta_{j}/\partial X_{i})+\mathfrak{p}$ . Noting that $I_{h}(\partial f_{j}/\partial X_{i})$ is not contained in $\mathfrak{p}$ , and hence neither $I_{n+m-d+1}(\partial f_{j}/\partial X_{i})$ , similarly, we have $I_{n+m-d+1}(\partial\beta_{j}/\partial X_{i})\nsubseteq\mathfrak{p}$ , which implies $\mathfrak{p}/J\notin\mathrm{V}(J_{n+1}^{k}(R/J))$ . ∎

The following result is a direct corollary of Theorem 3.12.

Corollary 3.13.

Let $k$ be a perfect field and let $n\geq 0$ be an integer.

(1)
For an affine $k$ -algebra $R$ of dimension $d$ , the following are equivalent:
1. (i)
  
  The inequality $\operatorname{edd}R\leq n$ holds;
2. (ii)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}^{l}(S))$ holds;
3. (iii)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}^{l}(S))$ holds;
4. (iv)
  
  For any field $l$ and any affine $l$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}^{l}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ holds;
5. (v)
  
  For any affine $k$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}^{k}(S))$ holds;
6. (vi)
  
  For any affine $k$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}^{k}(S))$ holds;
7. (vii)
  
  For any affine $k$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}^{k}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ holds.
(2)
For an equicharacteristic complete local ring $(R,\mathfrak{m},k)$ of dimension $d$ , the following are equivalent:
1. (i)
  
  The inequality $\operatorname{edd}R\leq n$ holds;
2. (ii)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}(S))$ ;
3. (iii)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}(S))$ ;
4. (iv)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ ;
5. (v)
  
  For any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}(S))$ holds;
6. (vi)
  
  For any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Spec}S=\mathrm{V}(J_{n+1}(S))$ holds;
7. (vii)
  
  For any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $J_{n}(S)\subseteq\sqrt{\mathsf{ca}^{d+1}(S)}$ holds.

Proof.

We only show (i) $\Leftrightarrow$ (ii) $\Leftrightarrow$ (v) in (1). Other implications are proven in a similar methods. Suppose that $\operatorname{edd}R\leq n$ holds. Let $S$ be an affine algebra over a field $l$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ . Remark 3.2(2) yields $\operatorname{edd}(A/I)=\operatorname{edd}R\leq n$ . Proposition 3.9 deduces that $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}^{l}(S))$ holds. We have (i) $\Rightarrow$ (ii). It is clear that (ii) $\Rightarrow$ (v) holds. Assume that $\operatorname{Sing}S\subseteq\mathrm{V}(J_{n}^{k}(S))$ holds for any affine $k$ -algebra $S$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ . Take a polynomial ring $A=k[X_{1},\ldots,X_{m}]$ over $k$ and an ideal $I$ of $A$ such that $A/I=R$ . If an ideal $J$ of $A$ satisfies $\mathrm{V}(J)=\mathrm{V}(I)$ , then $\operatorname{Spec}(A/J)\cong\mathrm{V}(J)=\mathrm{V}(I)\cong\operatorname{% Spec}(A/I)=\operatorname{Spec}R$ . By assumption, $\operatorname{Sing}(A/J)\subseteq\mathrm{V}(J_{n}^{k}(A/J))$ holds. Theorem 3.12 says that for any mimimal element $\mathfrak{q}$ of $\mathrm{V}(I)$ , the inequality $\operatorname{dim}R\leq\operatorname{dim}B/\mathfrak{q}+n$ holds, which means $\operatorname{edd}R\leq n$ . We obtain (v) $\Rightarrow$ (i). ∎

In the case of $n=0$ , the conditions (i), (iv), (v) and (vii) of Corollary 3.13 are rewritten as follows.

Corollary 3.14.

Let $k$ be a perfect field.

(1)
For an affine $k$ -algebra $R$ of dimension $d$ , the following are equivalent:
1. (i)
  
  $R$ is equidimensional;
2. (ii)
  
  For any affine algebra $S$ over a filed with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , there exists in integer $m>0$ such that $\operatorname{jac}_{k}(S)^{m}\operatorname{Ext}^{d+1}_{S}(M,N)=0$ holds for all $S$ -modules $M,N$ ;
3. (iii)
  
  For any affine $k$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S=\mathrm{V}(\operatorname{jac}_{k}(S))$ holds;
4. (iv)
  
  For any affine $k$ -algebra $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\sqrt{\mathsf{ca}^{d+1}(S)}=\sqrt{\operatorname{jac}_{k}(S)}$ holds.
(2)
For an equicharacteristic complete local ring $(R,\mathfrak{m},k)$ of dimension $d$ , the following are equivalent:
1. (i)
  
  $R$ is equidimensional;
2. (ii)
  
  For any equicharacteristic complete local ring $S$ with $\operatorname{Spec}S\cong\operatorname{Spec}R$ , there exists $m>0$ such that $\operatorname{jac}(S)^{m}\operatorname{Ext}^{d+1}_{S}(M,N)=0$ holds for all $S$ -modules $M,N$ ;
3. (iii)
  
  For any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\operatorname{Sing}S=\mathrm{V}(\operatorname{jac}(S))$ holds;
4. (iv)
  
  For any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ , $\sqrt{\mathsf{ca}^{d+1}(S)}=\sqrt{\operatorname{jac}(S)}$ holds.

Proof.

Applying Corollary 3.13 to $n=0$ , the implications (ii) $\Rightarrow$ (i) $\Leftarrow$ (iii) hold. An analogous argument to the former part of the proof of Theorem 3.12 shows $\operatorname{Sing}(S)=\mathrm{V}(\mathsf{ca}^{\operatorname{dim}S+1}(S))$ , where $S$ is either an affine $k$ -algebra or an equicharacteristic complete local ring with residue field $k$ . So, we have (iii) $\Leftrightarrow$ (iv). We see by Remark 3.2(2) and [9, Theorem 1.1] that (i) $\Rightarrow$ (ii) holds. Suppose that $R$ is an equidimensional equicharacteristic complete local ring with residue field $k$ . It follows from Remark 3.2(2) and [12, Lemma 4.3, and Propositions 4.4 and 4.5] that $\operatorname{Sing}(S)\supseteq V(\operatorname{jac}S)$ holds for any equicharacteristic complete local ring $S$ with residue field $k$ such that $\operatorname{Spec}S\cong\operatorname{Spec}R$ . By this and Corollary 3.13, we have $\operatorname{Sing}(S)=V(\operatorname{jac}S)$ . If $R$ is an equidimensional affine $k$ -algebra, the equality $\operatorname{Sing}(S)=\mathrm{V}(\operatorname{jac}_{k}S)$ follows from [6, Corollary 16.20]. Thus we get (i) $\Rightarrow$ (iii). ∎

We close this section by providing an example related to Corollary 3.13.

Example 3.15.

Let $R=k[X,Y_{1},\ldots,Y_{n+1}]/(XY_{1},\ldots,XY_{n+1})$ be a quotient of a polynomial ring over a perfect field $k$ , where $n\geq 1$ . Then $\operatorname{edd}R=n$ . We see that $\operatorname{Sing}R\subseteq\mathrm{V}(J_{n}^{k}(R))$ and $\operatorname{Spec}R=\mathrm{V}(J_{n+1}^{k}(R))$ hold; see Example 3.8. According to Corollary 3.13, there exist affine $k$ -algebras $S$ and $T$ such that $\operatorname{Sing}S\nsubseteq\mathrm{V}(J_{n-1}^{k}(S))$ and $\operatorname{Spec}T\neq\mathrm{V}(J_{n}^{k}(T))$ . By Example 3.8, one has $\operatorname{Spec}R\neq\mathrm{V}(J_{n}^{k}(R))$ . So $T$ can be chosen as $R$ itself. On the other hand, we get $\operatorname{Sing}R\subseteq\mathrm{V}(J_{n-1}^{k}(R))$ by Example 3.8. Now let $S=k[X,Y_{1},\ldots,Y_{n+1}]/(XY_{1}^{2},XY_{2},\ldots,XY_{n+1})$ be a quotient of a polynomial ring over $k$ . It is seen that $\operatorname{Spec}S\cong\operatorname{Spec}R$ and $\operatorname{Sing}S=\mathrm{V}((Y_{1},\ldots,Y_{n+1})S)$ . The ideal $J_{n-1}^{k}(S)$ is generated by the $n$ -minors of

\displaystyle\begin{pmatrix}Y_{1}^{2}&2XY_{1}&0&\cdots&0\\ Y_{2}&0&X&\ddots&\vdots\\ \vdots&\vdots&\ddots&\ddots&0\\ Y_{n+1}&0&\cdots&0&X\\ \end{pmatrix},

which contains $X^{n}S$ . We conclude that $\operatorname{Sing}S\nsubseteq\mathrm{V}(J_{n-1}^{k}(S))$ as $(Y_{1},\ldots,Y_{n+1})S$ does not contain $X^{n}S$ .

4. Uniqueness of the Jacobian ideal

This section focuses on whether the Jacobian ideal is uniquely determined. In the case of a complete local ring in equicharacteristic, when there are two presentations of the ring, the Jacobian ideal can be directly computed through an isomorphism. After presenting two lemmas, we will prove this.

Lemma 4.1.

Let $R$ be a ring, $m,n\geq 0$ , and $A=(a_{ij})_{1\leq i\leq m,1\leq j\leq n}$ an $m\times n$ matrix over $R$ . Let

\displaystyle B=\begin{pmatrix}a_{11}&\cdots&a_{1n}&\sum_{j=1}^{n}c_{j}a_{1j}+% b_{1}\\ \vdots&\ddots&\vdots&\vdots\\ a_{m1}&\cdots&a_{mn}&\sum_{j=1}^{n}c_{j}a_{mj}+b_{m}\\ \end{pmatrix}

be an $(m+1)\times n$ matrix, where $b_{i},c_{j}\in R$ for all $1\leq i\leq m$ and all $1\leq j\leq n$ . Then for any integer $r$ , $I_{r}(A)\subseteq I_{r}(B)\subseteq I_{r}(A)+(b_{1},\ldots,b_{m})R$ .

Proof.

It is clear that $I_{r}(A)\subseteq I_{r}(B)$ . When $r\leq 1$ , we can easily see that $I_{r}(B)\subseteq I_{r}(A)+(b_{1},\ldots,b_{m})R$ . Assume $r\geq 2$ . To prove lemma, we only need to consider the $r$ -th minors that include the $(n+1)$ -th column. That is, we have only to show that for all $1\leq i_{1}<\cdots<i_{r}\leq m$ and $1\leq j_{1}<\cdots<j_{r-1}<j_{r}=n+1$ , the element $\sum_{\sigma\in S_{r}}(\operatorname{sgn}\sigma)\prod_{l=1}^{r-1}a_{i_{\sigma(% l)}j_{l}}\cdot(\sum_{k=1}^{n}c_{k}a_{i_{\sigma(r)}k}+b_{i_{\sigma(r)}})$ belongs to $I_{r}(A)+(b_{1},\ldots,b_{m})R$ . Fix $1\leq k\leq n$ . If $k=j_{l}$ for some $1\leq l\leq r-1$ , $(\operatorname{sgn}\sigma)\prod_{l=1}^{r-1}a_{i_{\sigma(l)}j_{l}}a_{i_{\sigma(% r)}k}+(\operatorname{sgn}\tau)\prod_{l=1}^{r-1}a_{i_{\tau(l)}j_{l}}a_{i_{\tau(% r)}k}=0$ for any $\sigma\in S_{r}$ and $\tau=\sigma\circ(l\ r)$ . Therefore we have

		$\displaystyle\sum_{\sigma\in S_{r}}(\operatorname{sgn}\sigma)\prod_{l=1}^{r-1}% a_{i_{\sigma(l)}j_{l}}\cdot(\sum_{k=1}^{n}c_{k}a_{i_{\sigma(r)}k}+b_{i_{\sigma% (r)}})$
	$\displaystyle=$	$\displaystyle\sum_{k=1}^{n}c_{k}\sum_{\sigma\in S_{r}}(\operatorname{sgn}% \sigma)\prod_{l=1}^{r-1}a_{i_{\sigma(l)}j_{l}}a_{i_{\sigma(r)}k}+\sum_{\sigma% \in S_{r}}(\operatorname{sgn}\sigma)\prod_{l=1}^{r-1}a_{i_{\sigma(l)}j_{l}}b_{% i_{\sigma(r)}}$
	$\displaystyle=$	$\displaystyle\sum_{\begin{subarray}{c}1\leq k\leq n\\ k\notin\{j_{1},\ldots,j_{r-1}\}\end{subarray}}c_{k}\sum_{\sigma\in S_{r}}(% \operatorname{sgn}\sigma)\prod_{l=1}^{r-1}a_{i_{\sigma(l)}j_{l}}a_{i_{\sigma(r% )}k}+\sum_{\sigma\in S_{r}}(\operatorname{sgn}\sigma)\prod_{l=1}^{r-1}a_{i_{% \sigma(l)}j_{l}}b_{i_{\sigma(r)}}$
	$\displaystyle\in$	$\displaystyle\ (c_{1},\ldots,c_{n})I_{r}(A)+(b_{1},\ldots,b_{m})R.$

The proof is now completed. ∎

Lemma 4.2.

Let $R$ be a ring, and let $S$ be either the polynomial ring $R[X_{1},\ldots,X_{l}]$ or the formal power series ring $R\llbracket X_{1},\ldots,X_{l}\rrbracket$ . Let $f_{1},\ldots,f_{m},g_{1},\ldots,g_{n}\in S$ . If $I=(f_{1},\ldots,f_{m})=(g_{1},\ldots,g_{n})$ , then $I_{r}(\partial f_{j}/\partial X_{i})+I=I_{r}(\partial g_{j}/\partial X_{i})+I$ for any integer $r$ .

Proof.

Since $(f_{1},\ldots,f_{m})=(f_{1},\ldots,f_{m},g_{1})=\cdots=(f_{1},\ldots,f_{m},g_{% 1},\ldots,g_{n-1})=(f_{1},\ldots,f_{m},g_{1},\ldots,g_{n})=(f_{2},\ldots,f_{m}% ,g_{1},\ldots,g_{n})=\cdots=(f_{m},g_{1},\ldots,g_{n})=(g_{1},\ldots,g_{n})$ , we may assume that $n=m+1$ and $f_{i}=g_{i}$ for any $1\leq i\leq m$ . Put $g_{m+1}=\sum_{i=1}^{m}a_{i}f_{i}$ . Then

\displaystyle\left(\frac{\partial g_{j}}{\partial X_{i}}\right)=\begin{pmatrix% }\dfrac{\partial f_{1}}{\partial X_{1}}&\cdots&\dfrac{\partial f_{m}}{\partial X% _{1}}&\sum\limits_{i=1}^{m}a_{i}\dfrac{\partial f_{i}}{\partial X_{1}}+\sum% \limits_{i=1}^{m}\dfrac{\partial a_{i}}{\partial X_{1}}f_{i}\\ \vdots&\ddots&\vdots&\vdots\\ \dfrac{\partial f_{1}}{\partial X_{n}}&\cdots&\dfrac{\partial f_{m}}{\partial X% _{n}}&\sum\limits_{i=1}^{m}a_{i}\dfrac{\partial f_{i}}{\partial X_{n}}+\sum% \limits_{i=1}^{m}\dfrac{\partial a_{i}}{\partial X_{n}}f_{i}\\ \end{pmatrix}.

By Lemma 4.1, we have $I_{r}(\partial f_{j}/\partial X_{i})\subseteq I_{r}(\partial g_{j}/\partial X_% {i})\subseteq I_{r}(\partial f_{j}/\partial X_{i})+I$ . ∎

It is well known that the Jacobian ideal of an affine algebra over a field is well-defined, as shown in the argument of Remark 3.4. In the case of complete equicharacteristic local rings, the Jacobian ideal can be computed and compared by using the properties of the local ring.

Proposition 4.3.

Let $R$ be an equicharacteristic complete local ring with residue field $k$ . Suppose that $F:A:=k\llbracket X_{1},\ldots,X_{m}\rrbracket/(f_{1},\ldots,f_{c})\to R$ and $G:B:=\llbracket Y_{1},\ldots,Y_{n}\rrbracket/(g_{1},\ldots,g_{d})\to R$ are ring isomorphisms. Then $F(I_{m+r}(\partial f_{j}/\partial X_{i})A)=G(I_{n+r}(\partial g_{j}/\partial Y% _{i})B)$ holds for any integer $r$ .

Proof.

Put $\psi=G^{-1}F$ , $I=(f_{1},\ldots,f_{c})$ , and $J=(g_{1},\ldots,g_{d})$ . Let $e$ be the embedding dimension of $R$ , which is equal to the dimensions of $(X_{1},\ldots,X_{m})/(I+(X_{1},\ldots,X_{m})^{2})$ and $(Y_{1},\ldots,Y_{n})/(J+(Y_{1},\ldots,Y_{n})^{2})$ as $k$ -vector space. Then $m,n\geq e$ , and $m=e$ holds if and only if $I$ is contained in $(X_{1},\ldots,X_{m})^{2}$ . We prove $\psi(I_{m+r}(\partial f_{j}/\partial X_{i})A)=I_{n+r}(\partial g_{j}/\partial Y% _{i})B$ by induction on $m+n$ .

First we deal with the case $m=n=e$ . For any $1\leq i\leq m$ , we put $\psi(X_{i}+I)=\alpha_{i}+J$ , where $\alpha_{i}\in(Y_{1},\ldots,Y_{n})$ . There is a ring homomorphism $\phi:S:=k\llbracket X_{1},\ldots,X_{m}\rrbracket\to T:=\llbracket Y_{1},\ldots% ,Y_{n}\rrbracket:X_{i}\mapsto\alpha_{i}$ . Then $\phi(X_{1},\ldots,X_{m})+J=(Y_{1},\ldots,Y_{n})$ as $\psi$ is an isomorphism. Since $n=e$ , we have $J\subseteq(Y_{1},\ldots,Y_{n})^{2}$ , and hence $\phi(X_{1},\ldots,X_{m})=(Y_{1},\ldots,Y_{n})$ by [7, Corollary of Theorem 2.2]. This says that $\phi$ is surjective and so it is an isomorphism as $m=n$ . The equality $(\phi(f_{1}),\ldots,\phi(f_{c}))=(g_{1},\ldots,g_{d})$ holds. Thanks to Lemma 4.2, we get $I_{n+r}(\partial\phi(f_{j})/\partial Y_{i})B=I_{n+r}(\partial g_{j}/\partial Y% _{i})B$ . One has the equality

\delta_{ij}=\frac{\partial Y_{j}}{\partial Y_{i}}=\frac{\partial\phi(\phi^{-1}% (Y_{j}))}{\partial Y_{i}}=\sum_{t=1}^{m}\phi\left(\frac{\partial\phi^{-1}(Y_{j% })}{\partial X_{t}}\right)\frac{\partial\phi(X_{t})}{\partial Y_{i}}

where $\delta_{ij}$ is the Kronecker delta. Thus the matrix $(\partial\phi(X_{j})/\partial Y_{i})$ invertible. The equality

\frac{\partial\phi(f_{j})}{\partial Y_{i}}=\sum_{t=1}^{m}\phi\left(\frac{% \partial f_{j}}{\partial X_{t}}\right)\frac{\partial\phi(X_{t})}{\partial Y_{i}}

induces the following commutative diagram:

Hence $\phi(I_{m+r}(\partial f_{j}/\partial X_{i}))=I_{n+r}(\partial\phi(f_{j})/% \partial Y_{i})$ because these are Fitting invariants of the same module. We obtain $\psi(I_{m+r}(\partial f_{j}/\partial X_{i})A)=I_{n+r}(\partial\phi(f_{j})/% \partial Y_{i})B=I_{n+r}(\partial g_{j}/\partial Y_{i})B$ .

Next, we handle the case $m+n>2e$ . We may assume $m>e$ . There is $h\in I\setminus(X_{1},\ldots,X_{m})^{2}$ For some $1\leq l\leq m$ , the coefficient of $X_{i}$ in $h$ is not zero. By this, $X_{1},\ldots X_{l-1},h,X_{l+1}\ldots,X_{m}$ is a regular system of parameters of $k\llbracket X_{1},\ldots,X_{m}\rrbracket$ and $x:=\partial h/\partial X_{l}$ is unit. There is a natural isomomorphism $\chi:k\llbracket X_{1},\ldots X_{l-1},X_{l+1}\ldots,X_{m}\rrbracket\to k% \llbracket X_{1},\ldots,X_{m}\rrbracket/(h)$ . We put $\chi^{-1}(I/(h))=(h_{1},\ldots,h_{e})$ , $\chi$ induces an isomomorphism $\chi:C:=k\llbracket X_{1},\ldots X_{l-1},X_{l+1}\ldots,X_{m}\rrbracket/(h_{1},% \ldots,h_{e})\to k\llbracket X_{1},\ldots,X_{m}\rrbracket/(h_{1},\ldots,h_{e},% h)=A$ . By the induction hypothesis, we have $(\psi\circ\chi)(I_{m-1+r}(U)C)=I_{n+r}(\partial g_{j}/\partial Y_{i})B$ , where

\displaystyle U=\begin{pmatrix}\dfrac{\partial h_{1}}{\partial X_{1}}&\cdots&% \dfrac{\partial h_{e}}{\partial X_{1}}\\ \vdots&\ddots&\vdots\\ \dfrac{\partial h_{1}}{\partial X_{l-1}}&\cdots&\dfrac{\partial h_{e}}{% \partial X_{l-1}}\\ \dfrac{\partial h_{1}}{\partial X_{l+1}}&\cdots&\dfrac{\partial h_{e}}{% \partial X_{l+1}}\\ \vdots&\ddots&\vdots\\ \dfrac{\partial h_{1}}{\partial X_{m}}&\cdots&\dfrac{\partial h_{e}}{\partial X% _{m}}\\ \end{pmatrix}.

On the other hand, it follows from $I=(h_{1},\ldots,h_{e},h)$ and Lemma 4.2 that $I_{m+r}(\partial f_{j}/\partial X_{i})$ is equal to the ideal generated by the $(m+r)$ -minors of

\displaystyle\begin{pmatrix}\dfrac{\partial h_{1}}{\partial X_{1}}&\cdots&% \dfrac{\partial h_{e}}{\partial X_{1}}&\dfrac{\partial h}{\partial X_{1}}\\ \vdots&\ddots&\vdots\\ \dfrac{\partial h_{1}}{\partial X_{l-1}}&\cdots&\dfrac{\partial h_{e}}{% \partial X_{l-1}}&\dfrac{\partial h}{\partial X_{l-1}}\\ 0&\cdots&0&x\\ \dfrac{\partial h_{1}}{\partial X_{l+1}}&\cdots&\dfrac{\partial h_{e}}{% \partial X_{l+1}}&\dfrac{\partial h}{\partial X_{l+1}}\\ \vdots&\ddots&\vdots\\ \dfrac{\partial h_{1}}{\partial X_{m}}&\cdots&\dfrac{\partial h_{e}}{\partial X% _{m}}&\dfrac{\partial h}{\partial X_{m}}\\ \end{pmatrix},

which is equal to $I_{m-1+r}(U)$ . The proof is now completed. ∎

Remark 4.4.

Let $K$ and $L$ be a field, and $R$ a ring. Assume that $R$ is finitely generated over both $K$ and $L$ . In this case, a natural question arises as to whether $J_{n}^{K}(R)=J_{n}^{L}(R)$ holds for any integer $n$ . For example, let $R=\mathbb{C}[X_{1},\dots,X_{m}]/(f_{1},\ldots,f_{n})$ be a quotient of a polynomial ring over the field $\mathbb{C}$ of complex numbers. The natural surjection $\phi:\mathbb{R}[X_{1},\dots,X_{m},Y]\to\mathbb{C}[X_{1},\dots,X_{m}]$ such that $\phi(Y)=i$ induces an isomorphism $\mathbb{R}[X_{1},\dots,X_{m},Y]/(g_{1},\ldots,g_{n},Y^{2}+1)\cong R$ , where $g_{1},\ldots,g_{n}$ are representatives of the inverse images of $f_{1},\ldots,f_{n}$ . The Jacobian matrix of $g_{1},\ldots,g_{n},Y^{2}+1$ is of the form

\displaystyle A:=\begin{pmatrix}B&0\\ \ast&2Y\\ \end{pmatrix}.

Then $\phi(B)$ is the Jacobian matrix of $f_{1},\ldots,f_{n}$ . As $2Y$ is unit in $\mathbb{R}[X_{1},\dots,X_{m},Y]/(g_{1},\ldots,g_{n},Y^{2}+1)$ , we have $J_{n}^{\mathbb{R}}(R)=\phi(I_{n+m+1-\operatorname{dim}R}(A))R=\phi(I_{n+m-% \operatorname{dim}R}(B))R=I_{n+m-\operatorname{dim}R}(\phi(B))R=J_{n}^{\mathbb% {C}}(R)$ for any $n$ . Moreover, a similar argument shows that $J_{n}^{K}(R)=J_{n}^{L}(R)$ holds for any $n$ , any finite (simple) extension $L/K$ , and any affine $L$ -algebra $R$ . However, in general, it is unknown whether equality holds.

Acknowledgments.

The author would like to thank his supervisor Ryo Takahashi, and Yuki Mifune, for providing the motivation for this research. The author is also grateful to Linquan Ma, Yuya Otake, Ryo Takahashi, and Tatsuki Yamaguchi for their valuable comments.

References

[1] W. Bruns; J. Herzog, Cohen–Macaulay rings, revised edition, Cambridge Studies in Advanced Mathematics 39, Cambridge University Press, Cambridge, 1998.
[2] L. W. Christensen, Gorenstein dimensions, Lecture Notes in Math. 1747, Springer-Verlag, Berlin, 2000.
[3] H. Dao; T. Kobayashi; R. Takahashi, Trace ideals of canonical modules, annihilators of Ext modules, and classes of rings close to being Gorenstein, J. Pure Appl. Algebra 225 (2021), no. 9, Paper No. 106655, 18 pp.
[4] S. Dey; P. Lank; R. Takahashi, Strong generation and (co)ghost index for module categories, preprint (2023), arXiv:2307.13675v3.
[5] S. Dey; R. Takahashi, Comparisons between annihilators of Tor and Ext, Acta Math. Vietnam. 47 (2022), no. 1, 123–139.
[6] D. Eisenbud, Commutative algbra, With a view toward algebraic geometry, Grad. Texts in Math., 150 Springer-Verlag, New York, (1995).
[7] H. Matsumura, Commutative ring theory, Translated from the Japanese by M. Reid, Second edition, Cambridge Studies in Advanced Mathematics 8, Cambridge University Press, Cambridge, 1989.
[8] S. B. Iyengar; R. Takahashi, Annihilation of cohomology and strong generation of module categories, Int. Math. Res. Not. IMRN (2016), no. 2, 499–535.
[9] S. B. Iyengar; R. Takahashi, The Jacobian ideal of a commutative ring and annihilators of cohomology, J. Algebra 571 (2021), 280–296.
[10] K. Kimura, Compactness of the Alexandrov topology of maximal Cohen–Macaulay modules, preprint (2024), arXiv:2401.04987.
[11] J. Liu, Annihilators and dimensions of the singularity category, Nagoya Math. J. 250 (2023), 533–548.
[12] H.-J. Wang, On the Fitting ideals in free resolutions, Michigan Math. J. 41 (1994), no. 3, 587–608.
[13] H.-J. Wang, A uniform property of affine domains, J. Algebra 215 (1999), no. 2, 500–508.