This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Remarks on a conjecture of Huneke and Wiegand
and the vanishing of (co)homology

Olgur Celikbas Olgur Celikbas
Department of Mathematics
West Virginia University
Morgantown, WV 26506-6310, U.S.A
[email protected]
Uyen Le Uyen Le
Department of Mathematics
West Virginia University
Morgantown, WV 26506-6310, U.S.A
[email protected]
Hiroki Matsui Hiroki Matsui
Department of Mathematical Sciences, Faculty of Science and Technology, Tokushima University, 2-1 Minamijosanjima-cho, Tokushima 770-8506, JAPAN
[email protected]
 and  Arash Sadeghi Arash Sadeghi
Dokhaniat 49179-66686, Gorgan, IRAN
[email protected]
Abstract.

In this paper we study a long-standing conjecture of Huneke and Wiegand which is concerned with the torsion submodule of certain tensor products of modules over one-dimensional local domains. We utilize Hochster’s theta invariant and show that the conjecture is true for two periodic modules. We also make use of a result of Orlov and formulate a new condition which, if true over hypersurface rings, forces the conjecture of Huneke and Wiegand to be true over complete intersection rings of arbitrary codimension. Along the way we investigate the interaction between the vanishing of Tate (co)homology and torsion in tensor products of modules, and obtain new results that are of independent interest.

Key words and phrases:
complexity, Tate (co)homology, Tor-rigidity, vanishing of Ext and Tor, tensor products, torsion
2010 Mathematics Subject Classification:
Primary 13D07; Secondary 13H10, 13D05, 13C12
Celikbas was partly supported by WVU Mathematics Excellence and Research Funds (MERF). Matsui was partly supported by JSPS Grant-in-Aid for JSPS Fellows 19J00158.

1. Introduction

Throughout RR denotes a commutative Noetherian local ring with unique maximal ideal 𝔪\mathfrak{m} and residue field kk, and all RR-modules are assumed to be finitely generated.

In commutative algebra there are several questions about tensor products of modules that are notoriously difficult to solve; see, for example, [27]. A fine example of such a question is the following long-standing conjecture of Huneke and Wiegand:

Conjecture 1.1.

(Huneke - Wiegand; see [35, page 473]) Let RR be a one-dimensional local ring and let MM be a nonfree and torsion-free RR-module. Assume MM has rank (e.g., RR is a domain). Then the torsion submodule of MRMM\otimes_{R}M^{\ast} is nonzero, i.e., MRMM\otimes_{R}M^{\ast} has (nonzero) torsion, where M=HomR(M,R)M^{\ast}=\operatorname{Hom}_{R}(M,R).

Conjecture 1.1, over Gorenstein rings, is in fact a special case of a celebrated conjecture of Auslander and Reiten [6] and is wide open in general, even for two generated ideals over complete intersection domains of codimension two; see Remark 3.10. On the other hand, besides some other special cases, Conjecture 1.1 is known to be true over hypersurface rings; see [35] and [33] for the details. In fact, since maximal Cohen-Macaulay modules (that have no free direct summand) are two-periodic over hypersurface rings, the hypersurface case of Conjecture 1.1 is subsumed by the following result:

Theorem 1.2.

([17, 4.17]) Let RR be a one-dimensional local complete intersection domain and let MM be a nonzero RR-module. Assume MM is two-periodic, i.e., MΩR2MM\cong\Omega_{R}^{2}M. Then MRMM\otimes_{R}M^{\ast} has torsion.

In this paper we study Conjecture 1.1 for the case where RR is a domain. Our aim concerning Theorem 1.2 is twofold. In section 2 we give a proof of Theorem 1.2 – which is entirely different from the one given in [17] – by using Tate (co)homology; see Corollary 2.17. This approach motivates us to seek, and hence obtain, new results about the vanishing of Tate (co)homology and torsion in tensor products that are independent of Conjecture 1.1; see, for example, Corollary 2.11 and Theorem 2.13. Furthermore, we generalize Theorem 1.2 in section 3. More precisely, we remove the complete intersection hypothesis from Theorem 1.2 and hence prove:

Theorem 1.3.

Let RR be a one-dimensional local domain and let MM be a nonzero RR-module that is two-periodic. Then MRMM\otimes_{R}M^{\ast} has torsion.

The proof of Theorem 1.3 relies upon an invariant, referred to as the Hochster’s theta invariant, and is established in the paragraph following Proposition 3.9. In section 3, motivated by Theorems 1.2 and 1.3, we also provide an example of a two-periodic module over a one-dimensional local ring that is not a complete intersection; see Example 3.11.

A remarkable theorem of Orlov [39] determines an equivalence between the singularity category of RR and that of ProjA\operatorname{Proj}A for the generic hypersurface AA of RR. The gist of our work in Section 4 is to exploit Orlov’s theorem and show that Conjecture 1.1 holds over all one-dimensional complete intersection domains in case a certain condition we formulate holds for all hypersurface domains. Our approach to use Orlov’s theorem to attack Conjecture 1.1 seems to be new and it establishes the following theorem; see the paragraph following Remark 4.13.

Theorem 1.4.

Conjecture 1.1 is true over each one-dimensional local complete intersection domain provided that the following condition holds:

Whenever RR is a local hypersurface domain, cc is a positive integer, MM and MRExtRc1(M,R)M\otimes_{R}\operatorname{Ext}_{R}^{c-1}(M,R) are Cohen-Macaulay RR-modules, both of which have grade c1c-1, it follows MM has finite projective dimension.

Note that the case where c=1c=1 of the condition stated in Theorem 1.4 is nothing but the condition of Conjecture 1.1. We do not know whether or not each hypersurface domain satisfies the condition stated in Theorem 1.4, but we are now able to translate the problem of Conjecture 1.1 to a problem over hypersurface rings; the new advantage we have is that homological algebra is better understood over hypersurface rings than over complete intersection rings.

2. A proof of Theorem 1.2 via Tate homology

In this section we use Tate (co)homology and give a proof of Theorem 1.2 that is distinct from the one obtained in [17]. Along the way we obtain general results that should be useful to further understand the vanishing of Tate (co)homology and torsion; see, for example, Corollary 2.10 and Proposition 2.13.

We start by recording several preliminary results some of which will also be used in this section. For the definitions and basic properties of homological dimensions, such as the Gorenstein dimension G-dim\operatorname{\textnormal{G-dim}} and the complete intersection dimension CI-dim\operatorname{\textnormal{CI-dim}}, we refer the reader to [3, 9, 10].

2.1.

Let RR be a local ring, MM be an RR-module and let nn be an integer. If n>0n>0, then ΩRnM\Omega^{n}_{R}M denotes the nnth syzygy of MM, that is, the image of the nnth differential map in a minimal free resolution of MM. Also, if MM is totally reflexive and n<0n<0, then ΩRnM\Omega^{n}_{R}M denotes the nnth cosyzygy of MM, that is, the image of the RR-dual of the nnth differential map in a minimal free resolution of MM^{\ast}. Note, by convention, we have ΩR0(M)=M\Omega^{0}_{R}(M)=M.

2.2.

Let RR be a local ring and let MM be an RR-module. Then the Auslander transpose of MM, denoted by 𝖳𝗋M\mathsf{Tr}\hskip 0.72229ptM, is the cokernel of the map f=HomR(f,R)f^{\ast}=\operatorname{Hom}_{R}(f,R), where F1fF0M0F_{1}\stackrel{{\scriptstyle f}}{{\longrightarrow}}F_{0}\to M\to 0 is part of the minimal free resolution of MM; see [3]. Note that MΩR2𝖳𝗋MM^{\ast}\cong\Omega^{2}_{R}\mathsf{Tr}\hskip 0.72229ptM. Note also that 𝖳𝗋M\mathsf{Tr}\hskip 0.72229ptM is unique, up to isomorphism, since so are minimal free resolutions.

The following fact is used for 2.9 and Propositon 3.9,

2.3.

If RR is a local ring and MM is an RR-module such that Tor1R(M,𝖳𝗋M)=0\operatorname{Tor}_{1}^{R}(M,\mathsf{Tr}\hskip 0.72229ptM)=0, then MM is free; see [5, A1] and also [41, 3.9].

Next we recall the definitions of Tate homology and cohomology. Although their definitions do not require the ring to be local, we keep the local setting for simplicity.

2.4.

Let RR be a local ring and let MM be an RR-module. A complex 𝐓\mathbf{T} of free RR-modules is said to be totally acyclic provided that Hn(𝐓)=0=Hn(HomR(𝐓,R))\operatorname{H}_{n}(\mathbf{T})=0=\operatorname{H}_{n}(\operatorname{Hom}_{R}(\mathbf{T},R)) for all nn\in\mathbb{Z}. A complete resolution of MM is a diagram 𝐓ϑ𝐏𝜋M\mathbf{T}\overset{\vartheta}{\longrightarrow}\mathbf{P}\overset{\pi}{\longrightarrow}M, where 𝐏\mathbf{P} is a projective resolution, 𝐓\mathbf{T} is a totally acyclic complex and ϑ\vartheta is a morphism of complexes such that ϑi\vartheta_{i} is an isomorphism for all i0i\gg 0.

Assume 𝐓𝐏M\mathbf{T}\rightarrow\mathbf{P}\rightarrow M is a complete resolution of MM. Then, for an RR-module NN and for ii\in\mathbb{Z}, the Tate homology Tor^iR(M,N)\widehat{\operatorname{Tor}}_{i}^{R}(M,N) and the Tate cohomology Ext^Ri(M,N)\widehat{\operatorname{Ext}}_{R}^{i}(M,N) of MM and NN over RR are defined as Hi(𝐓RN)\operatorname{H}_{i}(\mathbf{T}\otimes_{R}N) and Hi(HomR(𝐓,N))\operatorname{H}^{i}(\operatorname{Hom}_{R}(\mathbf{T},N)), respectively.

It is known that G-dimR(M)<\operatorname{\textnormal{G-dim}}_{R}(M)<\infty if and only if MM has a complete resolution; see [11, 3.1]. Hence Tor^iR(M,N)\widehat{\operatorname{Tor}}_{i}^{R}(M,N) and Ext^Ri(M,N)\widehat{\operatorname{Ext}}_{R}^{i}(M,N) are defined for each RR-module NN in case G-dimR(M)<\operatorname{\textnormal{G-dim}}_{R}(M)<\infty. ∎

The following are some of the fundamental properties of Tate (co)homology modules; see, for example [9, 11, 23] or [21, 2.11 and 2.12].

2.5.

Let RR be a local ring, and let MM and NN be RR-modules such that G-dimR(M)<\operatorname{\textnormal{G-dim}}_{R}(M)<\infty.

  1. (i)

    If i>G-dimR(M)i>\operatorname{\textnormal{G-dim}}_{R}(M), then it follows Tor^iR(M,N)ToriR(M,N)\widehat{\operatorname{Tor}}_{i}^{R}(M,N)\cong\operatorname{Tor}_{i}^{R}(M,N) and Ext^Ri(M,N)ExtRi(M,N)\widehat{\operatorname{Ext}}^{i}_{R}(M,N)\cong\operatorname{Ext}^{i}_{R}(M,N).

  2. (ii)

    If 0MMM′′00\to M^{\prime}\to M\to M^{\prime\prime}\to 0 is a short exact sequence of RR-modules, where G-dimR(M)<\operatorname{\textnormal{G-dim}}_{R}(M^{\prime})<\infty or G-dimR(M′′)<\operatorname{\textnormal{G-dim}}_{R}(M^{\prime\prime})<\infty, then, for each ii\in\mathbb{Z}, there is an exact sequence of cohomology of the form:

    Ext^Ri(M′′,N)Ext^Ri(M,N)Ext^Ri(M,N)Ext^Ri+1(M′′,N),\widehat{\operatorname{Ext}}_{R}^{i}(M^{\prime\prime},N)\to\widehat{\operatorname{Ext}}_{R}^{i}(M,N)\to\widehat{\operatorname{Ext}}_{R}^{i}(M^{\prime},N)\to\widehat{\operatorname{Ext}}_{R}^{i+1}(M^{\prime\prime},N),

    and also of homology of the form:

    Tor^iR(M,N)Tor^iR(M,N)Tor^iR(M′′,N)Tor^i1R(M,N).\widehat{\operatorname{Tor}}_{i}^{R}(M^{\prime},N)\to\widehat{\operatorname{Tor}}_{i}^{R}(M,N)\to\widehat{\operatorname{Tor}}_{i}^{R}(M^{\prime\prime},N)\to\widehat{\operatorname{Tor}}_{i-1}^{R}(M^{\prime},N).
  3. (iii)

    Tor^i+nR(M,N)Tor^iR(ΩnM,N)\widehat{\operatorname{Tor}}_{i+n}^{R}(M,N)\cong\widehat{\operatorname{Tor}}_{i}^{R}(\Omega^{n}M,N) and Ext^Ri+n(M,N)Ext^Ri(ΩnM,N)\widehat{\operatorname{Ext}}^{i+n}_{R}(M,N)\cong\widehat{\operatorname{Ext}}^{i}_{R}(\Omega^{n}M,N) for all i,ni,n\in\mathbb{Z}.

  4. (iv)

    If pdR(M)<\operatorname{pd}_{R}(M)<\infty, then Ext^Ri(M,N)=0=Tor^iR(M,N)\widehat{\operatorname{Ext}}^{i}_{R}(M,N)=0=\widehat{\operatorname{Tor}}_{i}^{R}(M,N) for all ii\in\mathbb{Z}.

  5. (v)

    If MM is totally reflexive, then Tor^iR(M,N)Ext^Ri1(M,N)Ext^Ri+1(𝖳𝗋M,N)\widehat{\operatorname{Tor}}_{i}^{R}(M,N)\cong\widehat{\operatorname{Ext}}^{-i-1}_{R}(M^{*},N)\cong\widehat{\operatorname{Ext}}^{-i+1}_{R}(\mathsf{Tr}\hskip 0.72229ptM,N) for all ii\in\mathbb{Z}.

In the following we collect some results that are used in the proof of Theorem 2.8.

2.6.

Let RR be a local ring, and let MM and NN be RR-modules such that n1n\geq 1.

  1. (i)

    If depthR(MRN)n\operatorname{depth}_{R}(M\otimes_{R}N)\geq n, depthR(N)n1\operatorname{depth}_{R}(N)\geq n-1 and lengthR(ExtRi(𝖳𝗋M,N))<\operatorname{\textup{length}}_{R}(\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptM,N))<\infty for all i=1,,ni=1,\ldots,n, then it follows that ExtRi(𝖳𝗋M,N)=0\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptM,N)=0 for all i=1,,ni=1,\ldots,n; see [21, 4.2(ii)].

  2. (ii)

    If ExtRi(𝖳𝗋M,N)=0\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptM,N)=0 for all i=1,,ni=1,\ldots,n, then it follows that depthR(MRN)min{n,depthR(N)}\operatorname{depth}_{R}(M\otimes_{R}N)\geq\min\{n,\operatorname{depth}_{R}(N)\}; see [21, 4.2(iii)].

  3. (iii)

    If pdR(M)depthR(N)\operatorname{pd}_{R}(M)\leq\operatorname{depth}_{R}(N) and lengthR(ToriR(M,N))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(M,N))<\infty for all i1i\geq 1, then it follows ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i1i\geq 1; see [36, 2.2].

  4. (iv)

    If pdR(M)<\operatorname{pd}_{R}(M)<\infty and ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i1i\geq 1, then the depth formula for MM and NN holds, that is, depthR(MRN)=depthR(N)pdR(M)\operatorname{depth}_{R}(M\otimes_{R}N)=\operatorname{depth}_{R}(N)-\operatorname{pd}_{R}(M); see [2, 1.2]. ∎

The following approximation results are classical; see [4, 1.1] and also [22, 3.1 and 3.3].

2.7.

Let RR be a local ring and let MM be an RR-module such that G-dimR(M)=n<\operatorname{\textnormal{G-dim}}_{R}(M)=n<\infty.

  1. (i)

    There is a short exact sequence of RR-modules 0MXG00\rightarrow M\rightarrow X\rightarrow G\rightarrow 0, where pdR(X)=n\operatorname{pd}_{R}(X)=n and GG is totally reflexive. Such an exact sequence is called a finite projective hull of MM.

  2. (ii)

    There is a short exact sequence of RR-modules 0YXM00\rightarrow Y\rightarrow X\rightarrow M\rightarrow 0, where pdR(Y)=n1\operatorname{pd}_{R}(Y)=n-1 and XX is totally reflexive. Such an exact sequence is called a Cohen-Macaulay approximation of MM. ∎

The next result can be deduced from [1, 7.1(a)], which is a more general result. Here we provide a different and self-contained proof for the convenience of the reader.

Proposition 2.8.

Let RR be a local ring and let MM and NN be RR-modules. Assume n1n\geq 1 is an integer and the following conditions hold:

  1. (i)

    lengthR(ToriR(M,N))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(M,N))<\infty for all i1i\geq 1.

  2. (ii)

    lengthR(Tor^iR(M,N))<\operatorname{\textup{length}}_{R}(\widehat{\operatorname{Tor}}_{i}^{R}(M,N))<\infty for all ii\in\mathbb{Z}.

  3. (iii)

    G-dimR(M)depthR(N)n\operatorname{\textnormal{G-dim}}_{R}(M)\leq\operatorname{depth}_{R}(N)-n.

Then Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all i=n+1,,0i=-n+1,\ldots,0 if and only if depthR(MRN)n\operatorname{depth}_{R}(M\otimes_{R}N)\geq n.

Proof.

We start by considering a finite projective hull of MM, that is, a short exact sequence of RR-modules

(2.8.1) 0MXG0,0\rightarrow M\rightarrow X\rightarrow G\rightarrow 0,

where pdR(X)=G-dimR(M)\operatorname{pd}_{R}(X)=\operatorname{\textnormal{G-dim}}_{R}(M) and GG is totally reflexive; see 2.7(i). Then, by 2.5(v), the following holds for all ii\in\mathbb{Z}:

(2.8.2) Tor^iR(G,N)Ext^Ri1(G,N)Ext^Ri+1(𝖳𝗋G,N).\widehat{\operatorname{Tor}}_{i}^{R}(G,N)\cong\widehat{\operatorname{Ext}}^{-i-1}_{R}(G^{*},N)\cong\widehat{\operatorname{Ext}}^{-i+1}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N).

It follows, since pdR(X)<\operatorname{pd}_{R}(X)<\infty, that Tor^iR(X,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(X,N)=0 for all ii\in\mathbb{Z}; see 2.5(iv). So, in view of 2.5(ii), the short exact sequence in (2.8.1) yields the following isomorphisms for all ii\in\mathbb{Z}:

(2.8.3) Tor^iR(M,N)Tor^i+1R(G,N).\widehat{\operatorname{Tor}}_{i}^{R}(M,N)\cong\widehat{\operatorname{Tor}}_{i+1}^{R}(G,N).

Note that, as lengthR(Tor^iR(M,N))<\operatorname{\textup{length}}_{R}(\widehat{\operatorname{Tor}}_{i}^{R}(M,N))<\infty for all ii\in\mathbb{Z}, it follows from (2.8.2) and (2.8.3) that both lengthR(Tor^iR(G,N))\operatorname{\textup{length}}_{R}(\widehat{\operatorname{Tor}}_{i}^{R}(G,N)) and lengthR(Ext^Ri(𝖳𝗋G,N))\operatorname{\textup{length}}_{R}(\widehat{\operatorname{Ext}}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)) are finite for each ii\in\mathbb{Z}. Therefore, since GG is totally reflexive, 2.5(i) shows, for each i1i\geq 1, that:

(2.8.4) length(ToriR(G,N))< and lengthR(ExtRi(𝖳𝗋G,N))<.\operatorname{\textup{length}}(\operatorname{Tor}_{i}^{R}(G,N))<\infty\text{ and }\operatorname{\textup{length}}_{R}(\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N))<\infty.

Recall that we assume lengthR(ToriR(M,N))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(M,N))<\infty for all i1i\geq 1. So, by (2.8.1) and (2.8.4), it follows that lengthR(ToriR(X,N))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(X,N))<\infty for all i1i\geq 1. As pdR(X)=G-dimR(M)depthR(N)n<depthR(N)\operatorname{pd}_{R}(X)=\operatorname{\textnormal{G-dim}}_{R}(M)\leq\operatorname{depth}_{R}(N)-n<\operatorname{depth}_{R}(N), we conclude by 2.6(iii) that ToriR(X,N)=0\operatorname{Tor}_{i}^{R}(X,N)=0 for all i1i\geq 1. Hence, by tensoring (2.8.1) with NN, we obtain the following exact sequence:

(2.8.5) 0Tor1R(G,N)MRNXRNGRN0.0\rightarrow\operatorname{Tor}_{1}^{R}(G,N)\rightarrow M\otimes_{R}N\rightarrow X\otimes_{R}N\rightarrow G\otimes_{R}N\rightarrow 0.

Furthermore, the depth formula 2.6(iv), in view of the vanishing of ToriR(X,N)\operatorname{Tor}_{i}^{R}(X,N) for all i1i\geq 1, shows:

(2.8.6) depthR(XRN)=depthR(N)G-dimR(M)n.\operatorname{depth}_{R}(X\otimes_{R}N)=\operatorname{depth}_{R}(N)-\operatorname{\textnormal{G-dim}}_{R}(M)\geq n.

Now we assume depthR(MRN)n\operatorname{depth}_{R}(M\otimes_{R}N)\geq n and proceed to prove Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all i=n+1,,0i=-n+1,\ldots,0. Note length(Tor1R(G,N))<\operatorname{\textup{length}}(\operatorname{Tor}_{1}^{R}(G,N))<\infty; see (2.8.4). Therefore, (2.8.5) shows that Tor1R(G,N)=0\operatorname{Tor}_{1}^{R}(G,N)=0 and also depthR(GRN)n1\operatorname{depth}_{R}(G\otimes_{R}N)\geq n-1 since both depthR(XRN)\operatorname{depth}_{R}(X\otimes_{R}N) and depthR(MRN)\operatorname{depth}_{R}(M\otimes_{R}N) are at least nn; see (2.8.6). We know from (2.8.4) that lengthR(ExtRi(𝖳𝗋G,N))<\operatorname{\textup{length}}_{R}(\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N))<\infty for all i=1,,n1i=1,\ldots,n-1. Hence 2.6(i) implies that Ext^Ri(𝖳𝗋G,N)ExtRi(𝖳𝗋G,N)=0\widehat{\operatorname{Ext}}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)\cong\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)=0 for all i=1,,n1i=1,\ldots,n-1. Consequently, (2.8.2) and (2.8.3) yield the required vanishing of Tate Tor modules.

Next we assume Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all i=n+1,,0i=-n+1,\ldots,0 and proceed to prove depthR(MRN)n\operatorname{depth}_{R}(M\otimes_{R}N)\geq n. Note Tor1R(G,N)Tor^1R(G,N)Tor^0R(M,N)\operatorname{Tor}_{1}^{R}(G,N)\cong\widehat{\operatorname{Tor}}_{1}^{R}(G,N)\cong\widehat{\operatorname{Tor}}_{0}^{R}(M,N) and ExtRi(𝖳𝗋G,N)Ext^Ri(𝖳𝗋G,N)Tor^iR(M,N)\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)\cong\widehat{\operatorname{Ext}}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)\cong\widehat{\operatorname{Tor}}_{-i}^{R}(M,N) for all i=1,,n1i=1,\ldots,n-1; see (2.8.2) and (2.8.3). Therefore, we conclude Tor1R(G,N)=0=ExtRi(𝖳𝗋G,N)\operatorname{Tor}_{1}^{R}(G,N)=0=\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N) for all i=1,,n1i=1,\ldots,n-1. Now, since depthR(N)n\operatorname{depth}_{R}(N)\geq n and ExtRi(𝖳𝗋G,N)=0\operatorname{Ext}^{i}_{R}(\mathsf{Tr}\hskip 0.72229ptG,N)=0 for all i=1,,n1i=1,\ldots,n-1, we use 2.6(ii) and deduce that depthR(GRN)min{n1,depthR(N)}n1\operatorname{depth}_{R}(G\otimes_{R}N)\geq\min\{n-1,\operatorname{depth}_{R}(N)\}\geq n-1. Recall that depthR(XRN)n\operatorname{depth}_{R}(X\otimes_{R}N)\geq n; see (2.8.6). Thus, since Tor1R(G,N)=0\operatorname{Tor}_{1}^{R}(G,N)=0, we obtain, by the depth lemma applied to the exact sequence (2.8.5), that depthR(MRN)n\operatorname{depth}_{R}(M\otimes_{R}N)\geq n. ∎

2.9.

Let RR be a local ring and let MM and NN be RR-modules. Assume CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty or CI-dimR(N)<\operatorname{\textnormal{CI-dim}}_{R}(N)<\infty. Then the following conditions are equivalent; see [9, 4.9].

  1. (i)

    ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i0i\gg 0.

  2. (ii)

    ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i>CI-dimR(M)i>\operatorname{\textnormal{CI-dim}}_{R}(M).

  3. (iii)

    Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all ii\in\mathbb{Z}.

Moreover, if CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0 and N=MN=M^{\ast}, then MM is free if and only if one of the above equivalent conditions holds: this is because, if CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0 and Tor^iR(M,M)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,M^{\ast})=0 for all ii\in\mathbb{Z}, then, since MΩR2𝖳𝗋MM^{\ast}\cong\Omega_{R}^{2}\mathsf{Tr}\hskip 0.72229ptM, it follows that Tor1R(M,𝖳𝗋M)=0\operatorname{Tor}_{1}^{R}(M,\mathsf{Tr}\hskip 0.72229ptM)=0, which forces MM to be free; see 2.3 and 2.5(i).

We should also note, if CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty, then the conditions (i), (ii) and (iii) stated above are also equivalent for Ext and Tate cohomology modules; see [9, 4.7].

Corollary 2.10.

Let RR be a dd-dimensional Gorenstein local ring and let MM and NN be maximal Cohen-Macaulay RR-modules such that pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}. Then MRNM\otimes_{R}N is maximal Cohen-Macaulay if and only if Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all i=d+1,,0i=-d+1,\ldots,0.

Proof.

Let 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}. Then it follows that CI-dimR𝔭(M𝔭)=0\operatorname{\textnormal{CI-dim}}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})=0. Therefore, by 2.9, we have ToriR(M,N)𝔭=0\operatorname{Tor}_{i}^{R}(M,N)_{\mathfrak{p}}=0 and Tor^jR(M,N)𝔭=0\widehat{\operatorname{Tor}}_{j}^{R}(M,N)_{\mathfrak{p}}=0 for all i1i\geq 1 and for all jj\in\mathbb{Z}. Now the claim follows from Theorem 2.8. ∎

In passing we record an immediate consequence of Corollary 2.10; it yields a new characterization of torsionfreeness of tensor products in terms of the vanishing of Tate homology:

Corollary 2.11.

Let RR be a one-dimensional Gorenstein local domain and let MM and NN be torsion-free RR-modules. Then MRNM\otimes_{R}N is torsion-free if and only if Tor^0R(M,N)=0\widehat{\operatorname{Tor}}_{0}^{R}(M,N)=0.

Let us note that, in view of Corollary 2.11, Conjecture 1.1 can be stated over Gorenstein rings as follows: if RR is a one-dimensional Gorenstein domain and MM is a torsion-free RR-module such that Tor^0R(M,M)=0\widehat{\operatorname{Tor}}_{0}^{R}(M,M^{\ast})=0, then MM is free.

Our main result in this section is Theorem 2.13 which yields a new criterion for the vanishing of Tate (co)homology. More precisely, Theorem 2.13 is an extension of [19, 4.8], which considers the vanishing of (absolute) cohomology under the same hypotheses. To prove the theorem, we record a few more preliminary definitions and results.

2.12.

Let RR be a local ring and let MM be an RR-module. Then the complexity cxR(M)\operatorname{cx}_{R}(M) of MM is defined as inf{r{0}:A such that dimk(ExtRn(M,k))Anr1 for all n0}\inf\{r\in\mathbb{N}\cup\{0\}:A\in\mathbb{R}\text{ such that }\dim_{k}(\operatorname{Ext}^{n}_{R}(M,k))\leq A\cdot n^{r-1}\text{ for all }n\gg 0\}; see [8].

Note, it follows from the definition that, cxR(M)=0\operatorname{cx}_{R}(M)=0 if and only if pdR(M)<\operatorname{pd}_{R}(M)<\infty, and cxR(M)1\operatorname{cx}_{R}(M)\leq 1 if and only if MM has bounded Betti numbers. Furthermore, the following properties hold:

  1. (i)

    If CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty, then it follows that cxR(M)embdim(R)depth(R)\operatorname{cx}_{R}(M)\leq\operatorname{embdim}(R)-\operatorname{depth}(R); see [10, 5.6].

  2. (ii)

    If CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0, then it follows CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast})=0 and cxR(M)=cxR(M)\operatorname{cx}_{R}(M)=\operatorname{cx}_{R}(M^{\ast}); see [12, 4.2]. ∎

In the following G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}} denotes the reduced Grothendieck group with rational coefficients, that is, G¯(R)=(G(R)/[R])\overline{\operatorname{G}}(R)_{\mathbb{Q}}=(\operatorname{G}(R)/\mathbb{Z}\cdot[R])\otimes_{\mathbb{Z}}\mathbb{Q}, where G(R)\operatorname{G}(R) is the Grothendieck group of (finitely generated) RR-modules. Also, [N][N] denotes the class of a given RR-module NN in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}.

Theorem 2.13.

Let RR be a local ring and let MM and NN be RR-modules. Assume:

  1. (i)

    cxR(M)=c\operatorname{cx}_{R}(M)=c.

  2. (ii)

    CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty.

  3. (iii)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

  4. (iv)

    [N]=0[N]=0 in G¯(R)\overline{G}(R)_{\mathbb{Q}}.

Then, given an integer nn, the following hold:

  1. (a)

    If Ext^Ri(M,N)=0\widehat{\operatorname{Ext}}^{i}_{R}(M,N)=0 for all i=n,,n+c1i=n,\ldots,n+c-1, then Ext^Ri(M,N)=0\widehat{\operatorname{Ext}}^{i}_{R}(M,N)=0 for all ii\in\mathbb{Z}.

  2. (b)

    If Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all i=n,,n+c1i=n,\ldots,n+c-1, then Tor^iR(M,N)=0\widehat{\operatorname{Tor}}_{i}^{R}(M,N)=0 for all ii\in\mathbb{Z}.

Proof.

We proceed and prove the statement in part (a) first. Set dim(R)=d\dim(R)=d and X=ΩRnd1ΩRdMX=\Omega^{n-d-1}_{R}\Omega^{d}_{R}M. Then XX is totally reflexive and 2.5(iii) yields the following isomorphisms for all ii\in\mathbb{Z}:

(2.13.1) Ext^Ri(M,N)Ext^Rid(ΩRdM,N)Ext^Rin+1(X,N).\widehat{\operatorname{Ext}}^{i}_{R}(M,N)\cong\widehat{\operatorname{Ext}}^{i-d}_{R}(\Omega^{d}_{R}M,N)\cong\widehat{\operatorname{Ext}}^{i-n+1}_{R}(X,N).

Therefore, for all j1j\geq 1, we obtain:

(2.13.2) ExtRj(X,N)Ext^Rj(X,N)Ext^Rj+n1(M,N).\operatorname{Ext}^{j}_{R}(X,N)\cong\widehat{\operatorname{Ext}}^{j}_{R}(X,N)\cong\widehat{\operatorname{Ext}}^{j+n-1}_{R}(M,N).

Here the first and the second isomorphisms are due to 2.5(i) and (2.13.1), respectively. Hence, (2.13.2) and our assumption give:

(2.13.3) ExtRj(X,N)=0 for all j=1,,c.\operatorname{Ext}^{j}_{R}(X,N)=0\text{ for all }j=1,\ldots,c.

Note that CI-dimR(X)=0\operatorname{\textnormal{CI-dim}}_{R}(X)=0 and cxR(X)=cxR(M)=c\operatorname{cx}_{R}(X)=\operatorname{cx}_{R}(M)=c; see [10, 1.9.1] and [40, 3.6]. As XX is locally free on the punctured spectrum of RR, we use [19, 4.8] and conclude from (2.13.3) that ExtRj(X,N)=0\operatorname{Ext}^{j}_{R}(X,N)=0 for all j1j\geq 1, that is, ExtRj(M,N)=0\operatorname{Ext}^{j}_{R}(M,N)=0 for all j0j\gg 0. Now the required vanishing follows from 2.9.

Next we prove part (b). Note 2.5(iii) yields the following isomorphisms for all ii\in\mathbb{Z}:

(2.13.4) Tor^iR(M,N)Tor^idR(ΩRdM,N)Tor^in+1R(X,N).\widehat{\operatorname{Tor}}_{i}^{R}(M,N)\cong\widehat{\operatorname{Tor}}_{i-d}^{R}(\Omega^{d}_{R}M,N)\cong\widehat{\operatorname{Tor}}_{i-n+1}^{R}(X,N).

Then our assumption and (2.13.4) yield:

(2.13.5) Tor^jR(X,N)=0 for all j=1,,c.\widehat{\operatorname{Tor}}_{j}^{R}(X,N)=0\text{ for all }j=1,\ldots,c.

Note, as XX is totally reflexive, 2.5(v) gives the following isomorphism for all ii\in\mathbb{Z}:

(2.13.6) Ext^Rj1(X,N)Tor^jR(X,N).\widehat{\operatorname{Ext}}_{R}^{-j-1}(X^{\ast},N)\cong\widehat{\operatorname{Tor}}_{j}^{R}(X,N).

Now (2.13.5) and (2.13.6) show that:

(2.13.7) Ext^Rj1(X,N)=0 for all j=1,,c.\widehat{\operatorname{Ext}}_{R}^{-j-1}(X^{\ast},N)=0\text{ for all }j=1,\ldots,c.

It follows, since CI-dimR(X)=0\operatorname{\textnormal{CI-dim}}_{R}(X)=0, that CI-dimR(X)=CI-dimR(X)=0\operatorname{\textnormal{CI-dim}}_{R}(X^{*})=\operatorname{\textnormal{CI-dim}}_{R}(X)=0 and cxR(X)=cxR(X)=c\operatorname{cx}_{R}(X^{*})=\operatorname{cx}_{R}(X)=c; see 2.12(ii). Now we use part (a) of the theorem with (2.13.7) and conclude that Ext^Rj1(X,N)=0\widehat{\operatorname{Ext}}_{R}^{-j-1}(X^{\ast},N)=0 for all jj\in\mathbb{Z}. Then (2.13.4) and (2.13.6) establish the required vanishing of Tate homology. ∎

We proceed and collect several examples of rings for which hypothesis (iv) of Theorem 2.13 holds; see [19, 2.5, 2.6 and 4.11] and also Example 3.6.

2.14.

Let RR be a local ring and let NN be an RR-module. Then [N]=0[N]=0 in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}} for each one of the following cases:

  1. (i)

    N=ΩRn(M)N=\Omega^{n}_{R}(M) for some n0n\geq 0 and for some RR-module MM such that lengthR(M)<\operatorname{\textup{length}}_{R}(M)<\infty.

  2. (ii)

    RR is Artinian.

  3. (iii)

    RR has dimension one and NN has rank.

  4. (iv)

    RR is a one-dimensional domain.

  5. (v)

    RR is a two-dimensional normal domain with torsion class group.

  6. (vi)

    RR is a two-dimensional complete rational singularity such that kk is algebraically closed of characteristic zero, or kk is finite of positive characteristic, or kk is the algebraic closure of a finite field which has positive characteristic.

Corollary 2.15.

Let RR be a local ring of positive depth and let MM be a nonfree RR-module. Assume the following conditions hold:

  1. (i)

    CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0 and cx(M)1\operatorname{cx}(M)\leq 1.

  2. (ii)

    lengthR(ToriR(M,M))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(M,M^{\ast}))<\infty for all i0i\gg 0.

  3. (iii)

    [M][M] or [M][M^{\ast}] is zero in G¯(R)\overline{G}(R)_{\mathbb{Q}}.

Then it follows that depthR(MRM)=0\operatorname{depth}_{R}(M\otimes_{R}M^{\ast})=0.

Proof.

We start by noting that cxR(M)=1\operatorname{cx}_{R}(M)=1: this is because we assume MM is a nonfree RR-module such that depth(M)=depth(R)\operatorname{depth}(M)=\operatorname{depth}(R) and cxR(M)1\operatorname{cx}_{R}(M)\leq 1. Note also we have that CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast})=0 and cxR(M)=1\operatorname{cx}_{R}(M^{\ast})=1; see 2.12(ii). Furthermore, if 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}, then our assumption (ii) and 2.9 imply that M𝔭M_{\mathfrak{p}} is free over R𝔭R_{\mathfrak{p}}. In other words, MM is locally free on the punctured spectrum of RR.

Now suppose depthR(MRM)1\operatorname{depth}_{R}(M\otimes_{R}M^{\ast})\geq 1. Then, setting n=1n=1, we conclude from Theorem 2.8 that Tor^0R(M,M)=0\widehat{\operatorname{Tor}}_{0}^{R}(M,M^{\ast})=0. So Theorem 2.13 yields the vanishing of Tor^iR(M,M)\widehat{\operatorname{Tor}}_{i}^{R}(M,M^{\ast}) for all ii\in\mathbb{Z}. As we know CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0, this implies that MM is free; see 2.9. Therefore, the depth of MRMM\otimes_{R}M^{\ast} must be zero. ∎

We should note that, concerning the hypothesis in part (iii) of Corollary 2.15, we are not aware of an example of a ring RR and an RR-module MM such that CI-dimR(M)=0\operatorname{\textnormal{CI-dim}}_{R}(M)=0 and [M][M]=0[M^{\ast}]\neq[M]=0 in G¯(R)\overline{G}(R)_{\mathbb{Q}}. However, we observe next that such an example cannot occur over Gorenstein rings.

Remark 2.16.

Let RR be a Gorenstein local ring. It is well-known that the Grothendieck group G(R)\operatorname{G}(R) is generated by maximal Cohen-Macaulay RR-modules. In fact, given an RR-module MM with minimal a free resolution 𝐅=(Fi)\mathbf{F}=(F_{i}), it follows that [M]=i=0d1(1)i[Fi]+(1)d[ΩdM][M]=\sum_{i=0}^{d-1}(-1)^{i}[F_{i}]+(-1)^{d}[\Omega^{d}M], where dd is the dimension of RR.

Now we define a map Φ:G(R)G(R)\Phi:\operatorname{G}(R)\to\operatorname{G}(R) as follows:

Φ([M])=i=0d(1)i1[ExtRi(M,R)].\Phi([M])=\sum_{i=0}^{d}(-1)^{i-1}[\operatorname{Ext}_{R}^{i}(M,R)].

One can check that Φ\Phi is a well-defined group homomorphism. Let MM be a maximal Cohen-Macaulay RR-module. As ExtRi(M,R)=0\operatorname{Ext}^{i}_{R}(M,R)=0 for all i1i\geq 1, it follows that Φ([M])=[M]\Phi([M])=[M^{*}] and so Φ2([M])=[M]=[M]\Phi^{2}([M])=[M^{**}]=[M]. This shows that Φ\Phi is an isomorphism as G(R)\operatorname{G}(R) is generated by maximal Cohen-Macaulay RR-modules. Since Φ\Phi sends [R][R] to [R][R], it induces an isomorphism Φ¯:G¯(R)G¯(R)\overline{\Phi}:\overline{G}(R)_{\mathbb{Q}}\to\overline{G}(R)_{\mathbb{Q}}. Hence, if MM is a maximal Cohen-Macaulay RR-module, then [M]=0[M]=0 in G¯(R)\overline{G}(R)_{\mathbb{Q}} if and only if Φ¯([M])=[M]=0\overline{\Phi}([M])=[M^{*}]=0 in G¯(R)\overline{G}(R)_{\mathbb{Q}}. ∎

If RR is a one-dimensional local ring and MM is a torsion-free RR-module such that CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty and cxR(M)1\operatorname{cx}_{R}(M)\leq 1 (for example, RR is a hypersurface ring, or RR is a complete intersection ring and MM has bounded Betti numbers), then it follows that MΩR2MM\cong\Omega_{R}^{2}M (for our purpose we may assume MM has no free direct summand); see [10, 7.3] for the details. Hence Theorem 1.2 is subsumed by the next result which was proved in [17] via different techniques.

Corollary 2.17.

([18, 4.10]) Let RR be a one-dimensional local ring and let MM be a nonfree torsion-free RR-module. Assume MM has rank (e.g., RR is a domain). Assume further CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty and cxR(M)1\operatorname{cx}_{R}(M)\leq 1. Then it follows that MRMM\otimes_{R}M^{\ast} has torsion.

Proof.

It follows, as RR is a one-dimensional local ring and MM has rank, that [M]=0[M]=0 in G¯(R)\overline{G}(R)_{\mathbb{Q}}; see 2.14. Hence the claim follows from Corollary 2.15. ∎

Note that, in the next section we will prove Theorem 1.3 and hence generalize Corollary 2.17. We finish this section by pointing out some flexibility about the hypotheses of Corollary 2.17.

Remark 2.18.

Let RR be a one-dimensional local domain and let MM be a nonzero torsion-free RR-module. If CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty or CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast})<\infty, then it follows that CI-dimR(M)=0=CI-dimR(M)\operatorname{\textnormal{CI-dim}}_{R}(M)=0=\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast}); see 2.12(ii) and [18, 4.5(i)]. So, if CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty or CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast})<\infty, and cxR(M)1\operatorname{cx}_{R}(M)\leq 1 or cxR(M)1\operatorname{cx}_{R}(M^{\ast})\leq 1 (for example, if CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M^{\ast})<\infty and cxR(M)1\operatorname{cx}_{R}(M)\leq 1, or CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty and cxR(M)1\operatorname{cx}_{R}(M^{\ast})\leq 1), then it follows that CI-dimR(M)<\operatorname{\textnormal{CI-dim}}_{R}(M)<\infty and cxR(M)1\operatorname{cx}_{R}(M)\leq 1.

3. Conjecture 1.1 for two-periodic modules over one-dimensional domains

In this section we give a proof of Theorem 1.3; see the paragraph following Proposition 3.9. The primary result of the section is Theorem 3.2 which relies upon a slightly modified version of the theta invariant. This invariant was initially defined by Hochster [32] to study the direct summand conjecture; it was further developed by Dao, for example, to study the vanishing of Tor; see [24, 25, 26].

We start by recalling the definition of the theta invariant; see also [25, section 2].

Definition 3.1.

([32]) Let RR be a local ring and let MM be an RR-module. Assume the following hold:

  1. (i)

    MΩR2MM\cong\Omega^{2}_{R}M.

  2. (ii)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

Then, for each i1i\geq 1, it follows that ToriR(M,N)Tori+2R(M,N)\operatorname{Tor}_{i}^{R}(M,N)\cong\operatorname{Tor}_{i+2}^{R}(M,N) and lengthR(ToriR(M,N))<\operatorname{\textup{length}}_{R}(\operatorname{Tor}_{i}^{R}(M,N))<\infty. Therefore, given an RR-module NN, the theta invariant for the pair (M,N)(M,N) is defined as follows:

(3.1.1) θR(M,N)=lengthR(Tor2nR(M,N))lengthR(Tor2n1R(M,N)) for some n1.\theta^{R}(M,N)=\operatorname{\textup{length}}_{R}\big{(}\operatorname{Tor}_{2n}^{R}(M,N)\big{)}-\operatorname{\textup{length}}_{R}\big{(}\operatorname{Tor}_{2n-1}^{R}(M,N)\big{)}\text{ for some }n\geq 1.

Note that θR(M,N)\theta^{R}(M,N) is well-defined, i.e., its value is independent of the integer nn used in the definition.

Next we show that the theta function is additive on short exact sequences.

Theorem 3.2.

Let RR be a local ring and let MM be an RR-module. Assume the following hold:

  1. (i)

    MΩR2MM\cong\Omega_{R}^{2}M.

  2. (ii)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

  3. (iii)

    0XfYgZ00\to X\stackrel{{\scriptstyle f}}{{\longrightarrow}}Y\stackrel{{\scriptstyle g}}{{\longrightarrow}}Z\to 0 is a short exact sequence of RR-modules.

Then it follows that θR(M,Y)=θR(M,X)+θR(M,Z)\theta^{R}(M,Y)=\theta^{R}(M,X)+\theta^{R}(M,Z).

Proof.

Note that, since MΩR2MM\cong\Omega^{2}_{R}M, we can pick a minimal free resolution F=(Fn,n)F_{\bullet}=(F_{n},\partial_{n}) of MM such that Fn=Fn+2F_{n}=F_{n+2} for each n0n\geq 0 and n=n+2\partial_{n}=\partial_{n+2} for all n1n\geq 1.

Fix an integer n1n\geq 1. Then, by tensoring ff with FnF_{n}, we get the following commutative diagram:

FnRX\textstyle{F_{n}\otimes_{R}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}1Fnf\scriptstyle{1_{F_{n}}\otimes f}FnRY\textstyle{F_{n}\otimes_{R}Y\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Fn+2RX\textstyle{F_{n+2}\otimes_{R}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}1Fn+2f\scriptstyle{1_{F_{n+2}}\otimes f}Fn+2RY\textstyle{F_{n+2}\otimes_{R}Y}

The diagram above yields the following commutative diagram on homologies:

TornR(M,X)\textstyle{\operatorname{Tor}_{n}^{R}(M,X)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}TornR(M,f)\scriptstyle{\operatorname{Tor}_{n}^{R}(M,f)}\scriptstyle{\cong}TornR(M,Y)\textstyle{\operatorname{Tor}_{n}^{R}(M,Y)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}\scriptstyle{\cong}Torn+2R(M,X)\textstyle{\operatorname{Tor}_{n+2}^{R}(M,X)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Torn+2R(M,f)\scriptstyle{\operatorname{Tor}_{n+2}^{R}(M,f)}Torn+2R(M,Y)\textstyle{\operatorname{Tor}_{n+2}^{R}(M,Y)}

Now set Xn=ker(TornR(M,f))X_{n}=\ker\big{(}\operatorname{Tor}_{n}^{R}(M,f)\big{)}. Then it follows, since the diagram above involving the Tor modules, is commutative with vertical maps being isomorphisms, that XnXn+2X_{n}\cong X_{n+2}.

Note that the short exact sequence 0XfYgZ00\to X\stackrel{{\scriptstyle f}}{{\longrightarrow}}Y\stackrel{{\scriptstyle g}}{{\longrightarrow}}Z\to 0 gives rise to the following exact sequence:

(3.2.1) 0Xn+2Torn+2R(M,X)Torn+2R(M,Y)Torn+2R(M,Z)\displaystyle 0\to X_{n+2}\to\operatorname{Tor}^{R}_{n+2}(M,X)\to\operatorname{Tor}^{R}_{n+2}(M,Y)\to\operatorname{Tor}^{R}_{n+2}(M,Z)\to
Torn+1R(M,X)Torn+2R(M,Y)Torn+1R(M,Z)Xn0.\displaystyle{}\operatorname{Tor}^{R}_{n+1}(M,X)\to\operatorname{Tor}^{R}_{n+2}(M,Y)\to\operatorname{Tor}^{R}_{n+1}(M,Z)\to X_{n}\to 0.

As we have XnXn+2X_{n}\cong X_{n+2}, by taking the alternating sum of lengths of the terms in the exact sequence (3.2.1), we conclude that θR(M,Y)=θR(M,X)+θR(M,Z)\theta^{R}(M,Y)=\theta^{R}(M,X)+\theta^{R}(M,Z). ∎

Recall that G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}} denotes the reduced Grothendieck group with rational coefficients.

Corollary 3.3.

Let RR be a local ring and let MM be an RR-module. Assume the following hold:

  1. (i)

    MΩR2MM\cong\Omega_{R}^{2}M.

  2. (ii)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

Then θR(M,):G¯(R)\theta^{R}(M,-):\overline{G}(R)_{\mathbb{Q}}\to\mathbb{Q} is a well-defined function.

Proof.

It is clear from Theorem 3.2 that θR(M,)\theta^{R}(M,-) is a well-defined function on G(R)\operatorname{G}(R). As θR(M,R)=0\theta^{R}(M,R)=0, the result follows; see Definition 3.1. ∎

3.4.

If MM and NN are RR-modules, then the pair (M,N)(M,N) is said to be nn-Tor-rigid for some n1n\geq 1 provided that the following condition holds: if ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i=r,r+1,,r+n1i=r,r+1,\ldots,r+n-1 for some r1r\geq 1, then TorjR(M,N)=0\operatorname{Tor}_{j}^{R}(M,N)=0 for all jrj\geq r.

An RR-module MM is called Tor-rigid if (M,N)(M,N) is 11-Tor-rigid for each RR-module NN. For example, if RR is a regular local ring, or a hypersurface which is quotient of an unramified regular local ring, then each RR-module of finite projective dimension is Tor-rigid; see [2, 2.2] and [37, Cor. 1 and Thm. 3]. ∎

Corollary 3.5.

Let RR be a local ring and let MM and NN be RR-modules. Assume:

  1. (i)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

  2. (ii)

    [N]=0[N]=0 in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}.

  3. (iii)

    MΩRqMM\cong\Omega_{R}^{q}M for some even integer q2q\geq 2.

Then the pair (M,N)(M,N) is (q1)(q-1)-Tor-rigid.

Proof.

Set X=MΩR2(M)ΩRq2(M)X=M\oplus\Omega^{2}_{R}(M)\oplus\cdots\oplus\Omega^{q-2}_{R}(M). Then, since MΩRqMM\cong\Omega_{R}^{q}M, it follows that XΩR2XX\cong\Omega_{R}^{2}X. Note also, if r1r\geq 1, then by the definition of XX, we have:

(3.5.1) TorrR(X,N)=0 if and only if Torr+2iR(M,N)=0 for all i=0,,(q/2)1.\operatorname{Tor}_{r}^{R}(X,N)=0\text{ if and only if }\operatorname{Tor}_{r+2i}^{R}(M,N)=0\text{ for all }i=0,\ldots,(q/2)-1.

Next, to show (M,N)(M,N) is (q1)(q-1)-Tor-rigid, suppose ToriR(M,N)=0\operatorname{Tor}_{i}^{R}(M,N)=0 for all i=n,n+1,,n+q2i=n,n+1,\ldots,n+q-2 for some n1n\geq 1. Then, by (3.5.1), it follows that TornR(X,N)=0\operatorname{Tor}_{n}^{R}(X,N)=0. Hence, if we show Torn+1R(X,N)=0\operatorname{Tor}_{n+1}^{R}(X,N)=0, then, as XΩR2XX\cong\Omega_{R}^{2}X, we conclude that ToriR(X,N)=0\operatorname{Tor}_{i}^{R}(X,N)=0 for all i1i\geq 1; this yields the vanishing of TorjR(M,N)\operatorname{Tor}_{j}^{R}(M,N) for all j1j\geq 1 and establishes that (M,N)(M,N) is (q1)(q-1)-Tor-rigid.

Notice pdR𝔭(X𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(X_{\mathfrak{p}})<\infty, for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}. So, Corollary 3.3 shows θR(X,):G¯(R)\theta^{R}(X,-):\overline{G}(R)_{\mathbb{Q}}\to\mathbb{Q} is well-defined. Thus θR(X,N)=0\theta^{R}(X,N)=0 because [N]=0[N]=0 in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}. Consequently, as we know TornR(X,N)\operatorname{Tor}_{n}^{R}(X,N) vanishes, we deduce by the definition of the theta function that Torn+1R(X,N)=0\operatorname{Tor}^{R}_{n+1}(X,N)=0; see Definition 3.1. ∎

The following example points out that hypothesis (ii) of Corollary 3.5 is needed, even over isolated hypersurface singularities.

Example 3.6.

Let R=[[x,y,z,w]]/(xwyz)R=\mathbb{C}[\![x,y,z,w]\!]/(xw-yz), M=R/(x,z)M=R/(x,z) and N=R/(x,y)N=R/(x,y). Then one can check that RR is a three-dimensional (A1)(A_{1}) singularity (and hence is a domain) and Tor2R(M,N)=0Tor3R(M,N)\operatorname{Tor}_{2}^{R}(M,N)=0\neq\operatorname{Tor}_{3}^{R}(M,N). Thus we have θR(M,N)=1\theta^{R}(M,N)=-1. This implies that [N]0[N]\neq 0 in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}. Let us also note, since G¯(S)\overline{\operatorname{G}}(S)_{\mathbb{Q}}\cong\mathbb{Q} for each one-dimensional (A1)(A_{1}) singularity SS, it follows from Knörrer periodicity that G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}\cong\mathbb{Q}; see [41, 12.10 and 13.10]. ∎

As Example 3.6 shows, the conclusion of Corollary 3.5 may fail if hypothesis (ii) is not satisfied. However, even without this assumption, the corollary can be useful to produce Tor-rigid modules:

Corollary 3.7.

Let RR be a local ring, MM be an RR-module, and let x𝔪x\in\mathfrak{m} be a non zero-divisor on RR. Assume that the following hold:

  1. (i)

    pdR𝔭(M𝔭)<\operatorname{pd}_{R_{\mathfrak{p}}}(M_{\mathfrak{p}})<\infty for all 𝔭Spec(R){𝔪}\mathfrak{p}\in\operatorname{Spec}(R)-\{\mathfrak{m}\}.

  2. (ii)

    MΩR2MM\cong\Omega_{R}^{2}M.

Then the pair (M/xM,N/xN)(M/xM,N/xN) is Tor-rigid over R/xRR/xR for each torsion-free RR-module NN. Therefore, the pair (M/xM,M/xM)(M/xM,M/xM) is Tor-rigid over R/xRR/xR.

Proof.

Let NN be a torsion-free RR-module. As there is an exact sequence 0NxNN/xN00\to N\stackrel{{\scriptstyle x}}{{\rightarrow}}N\to N/xN\to 0, it follows that [N/xN]=[N][N]=0[N/xN]=[N]-[N]=0 in G¯(R)\overline{\operatorname{G}}(R)_{\mathbb{Q}}. Hence Corollary 3.5 shows that the pair (M,N/xN)(M,N/xN) is Tor-rigid over RR. Now, for an integer n0n\geq 0, it follows TornR/xR(M/xM,N/xN)=0\operatorname{Tor}_{n}^{R/xR}(M/xM,N/xN)=0 if and only if TornR(M,N/xN)=0\operatorname{Tor}_{n}^{R}(M,N/xN)=0 if and only if ToriR/xR(M/xM,N/xN)=0\operatorname{Tor}_{i}^{R/xR}(M/xM,N/xN)=0 for all ini\geq n; see [38, Lemma 2, page 140]. ∎

In the following we record some classes of rings over which periodic modules have certain Tor-rigidity property; see also Corollary 4.16 for a characterization of Tor-rigid modules over one-dimensional Gorenstein domains.

Corollary 3.8.

Let RR be a local ring. Assume that one of the following holds:

  1. (i)

    RR is Artinian.

  2. (ii)

    RR is a one-dimensional domain.

  3. (iii)

    RR is a two-dimensional normal domain with torsion class group.

If MM is an RR-module such that MΩRqMM\cong\Omega_{R}^{q}M for some even integer q2q\geq 2, then MM is (q1)(q-1)-Tor-rigid. Therefore, if MΩR2MM\cong\Omega_{R}^{2}M, then MM is Tor-rigid.

Proof.

It is known that, if RR is as in part (i), (ii) or (iii), then G¯(R)=0\overline{G}(R)_{\mathbb{Q}}=0; see 2.14. Therefore the result follows from Corollary 3.5.∎

Proposition 3.9.

Let RR be a local ring and let MM be a nonzero RR-module. Assume:

  1. (i)

    M𝔭M_{\mathfrak{p}} is free for each associated prime ideal of RR.

  2. (ii)

    MΩRqMM\cong\Omega_{R}^{q}M for some even integer q2q\geq 2.

  3. (iii)

    MM is Tor-rigid.

Then MRMM\otimes_{R}M^{\ast} has torsion.

Proof.

Note that M0M^{\ast}\neq 0 since MM is nonzero and torsion-free. Hence there exists a short exact sequence of RR-modules 0MFC00\to M^{\ast}\to F\to C\to 0, where FF is free. Tensoring this exact sequence with MM, we see that Tor1R(C,M)MRM\operatorname{Tor}^{R}_{1}(C,M)\hookrightarrow M\otimes_{R}M^{\ast}. Note that Tor1R(C,M)\operatorname{Tor}^{R}_{1}(C,M) is a torsion module since M𝔭M_{\mathfrak{p}} is free for each associated prime ideal of RR.

Now assume MRMM\otimes_{R}M^{\ast} is torsion-free. Then Tor1R(C,M)\operatorname{Tor}^{R}_{1}(C,M), and hence ToriR(C,M)\operatorname{Tor}^{R}_{i}(C,M) for each i1i\geq 1, vanishes. This yields ToriR(M,M)=0\operatorname{Tor}_{i}^{R}(M,M^{\ast})=0 for all i1i\geq 1. As MΩR2𝖳𝗋MM^{\ast}\cong\Omega^{2}_{R}\mathsf{Tr}\hskip 0.72229ptM (up to free summands), we conclude Tor1R(M,𝖳𝗋M)Tor1R(ΩRqM,𝖳𝗋RM)Tor1R(M,ΩRq𝖳𝗋M)Tor1+q2R(M,M)=0\operatorname{Tor}_{1}^{R}(M,\mathsf{Tr}\hskip 0.72229ptM)\cong\operatorname{Tor}_{1}^{R}(\Omega^{q}_{R}M,\mathsf{Tr}\hskip 0.72229pt_{R}M)\cong\operatorname{Tor}_{1}^{R}(M,\Omega^{q}_{R}\mathsf{Tr}\hskip 0.72229ptM)\cong\operatorname{Tor}_{1+q-2}^{R}(M,M^{\ast})=0. This implies MM is free and hence M=0M=0 since MΩRqMM\cong\Omega_{R}^{q}M; see 2.3. Therefore, MRMM\otimes_{R}M^{\ast} must have torsion. ∎

We are now ready to prove Theorem 1.3 advertised in the introduction. Recall that our argument removes the complete intersection hypothesis from Theorem 1.2.

Proof of Theorem 1.3.

Note that the module MM considered in the theorem is Tor-rigid; see Corollary 3.8. Therefore, MRMM\otimes_{R}M^{\ast} has torsion by Proposition 3.9. ∎

Remark 3.10.

Let RR be a one-dimensional local domain and let MM be a torsion-free RR-module.

  1. (i)

    If RR is Gorenstein, then it follows that MRMM\otimes_{R}M^{\ast} is torsion-free if and only if ExtR1(M,M)=0\operatorname{Ext}^{1}_{R}(M,M)=0; see [34, 5.9]. So Conjecture 1.1 predicts that the celebrated conjecture of Auslander and Reiten is true over one-dimensional Gorenstein rings, even if only one Ext module vanishes; see [6] and [34] for the details. However, if RR is not Gorenstein, then torsionfreeness and the vanishing of Ext may not be equivalent: for example, if RR is Cohen-Macaulay with a canonical module ωR\omega\ncong R and if RR has minimal multiplicity, then it follows that ExtR1(ω,ω)=0\operatorname{Ext}^{1}_{R}(\omega,\omega)=0, but ωRω\omega\otimes_{R}\omega^{\ast} has torsion; see [33, 3.6].

  2. (ii)

    It seems interesting that, even if RR is not Gorenstein, when MΩR2MM\cong\Omega_{R}^{2}M, it still follows that MRMM\otimes_{R}M^{\ast} is torsion-free if and only if ExtR1(M,M)=0\operatorname{Ext}^{1}_{R}(M,M)=0, since either of these two conditions forces MM to be zero; this follows from Theorem 1.3 and the fact that, when MM is a Tor-rigid module over a local ring RR and ExtRn(M,M)=0\operatorname{Ext}^{n}_{R}(M,M)=0 for some n0n\geq 0, then pdR(M)<n\operatorname{pd}_{R}(M)<n; see, for example, [27, 3.1.2]. ∎

In general there exist two-periodic modules that do not have finite complete intersection dimension. We build on an example of Gasharov and Peeva and construct such a module in the next example (note the ring considered in the example is not a complete intersection); cf. Theorems 1.2 and 1.3.

Example 3.11.

([28, 3.10]) Let kk be a field and fix αk\alpha\in k such that α4=1α3\alpha^{4}=1\neq\alpha^{3}. Let R=k[[x1,x2,x3,x4]]/IαR=k[\![x_{1},x_{2},x_{3},x_{4}]\!]/I_{\alpha}, where IαI_{\alpha} is the ideal of RR generated by the elements x12x22,x32,x42,x3x4,x1x4+x2x4,αx1x3+x2x3x_{1}^{2}-x_{2}^{2},\,\,\,x_{3}^{2},\,\,\,x_{4}^{2},\,\,\,x_{3}x_{4},\,\,\,x_{1}x_{4}+x_{2}x_{4},\,\,\,\alpha\cdot x_{1}x_{3}+x_{2}x_{3}. Then the complex

3R22R21R2N0\cdots\xrightarrow{\partial_{3}}R^{\oplus 2}\xrightarrow{\partial_{2}}R^{\oplus 2}\xrightarrow{\partial_{1}}R^{\oplus 2}\to N\to 0

is exact, where

N=coker(1) and n=(x1αnx3+x40x2) for each n1.N=\operatorname{coker}(\partial_{1})\text{ and }\partial_{n}=\begin{pmatrix}x_{1}&\alpha^{n}\cdot x_{3}+x_{4}\\ 0&x_{2}\\ \end{pmatrix}\text{ for each }n\geq 1.

It follows ΩR4Ncoker(5)=coker(1)N\Omega_{R}^{4}N\cong\operatorname{coker}(\partial_{5})=\operatorname{coker}(\partial_{1})\cong N and that NΩRiNN\ncong\Omega_{R}^{i}N for each i=1,2,3i=1,2,3. Then, by setting X=NΩR2NX=N\oplus\Omega^{2}_{R}N, we see that XΩR2XX\cong\Omega^{2}_{R}X and hence cxR(X)1\operatorname{cx}_{R}(X)\leq 1. Moreover, we have CI-dimR(X)=\operatorname{\textnormal{CI-dim}}_{R}(X)=\infty: otherwise we would have CI-dimR(N)<\operatorname{\textnormal{CI-dim}}_{R}(N)<\infty and so NΩR2NN\cong\Omega^{2}_{R}N since cxR(N)1\operatorname{cx}_{R}(N)\leq 1.

Next we set T=R[[t]]T=R[\![t]\!] and M=XRTM=X\otimes_{R}T. Then, since TT is a faithfully flat extension of RR, we conclude that TT is a one-dimensional Gorenstein ring, MΩT2MM\cong\Omega_{T}^{2}M, and also CI-dimT(M)=\operatorname{\textnormal{CI-dim}}_{T}(M)=\infty. ∎

The ring TT in Example 3.11 is not a domain and also the TT-module MM does not have rank. This raises the following question which, in view of Conjecture 1.1 and Theorem 1.3, should be important:

Question 3.12.

Is there a one-dimensional local ring TT and a TT-module MM such that MM has rank over TT (e.g., TT is a domain), MΩT2MM\cong\Omega_{T}^{2}M, and CI-dimT(M)=\operatorname{\textnormal{CI-dim}}_{T}(M)=\infty? ∎

We finish this section by recording some further observations concerning the torsion submodule of tensor products of the form MRMM\otimes_{R}M^{\ast}. Let us point out that the conclusion of Theorem 1.3 may fail if the ring in question is not one-dimensional; for example, [20, 3.5] provides such an example of a two-dimensional local hypersurface domain. In the aforementioned example, the module considered is torsion-free but it is not maximal Cohen-Macaulay. On the other hand, when maximal Cohen-Macaulay modules are considered, there is a partial result, which we recall next:

3.13.

If RR is an even-dimensional local hypersurface and MM is a nonfree maximal Cohen-Macaulay RR-module that is locally free on the punctured spectrum of RR, then MRMM\otimes_{R}M^{\ast} has torsion; see [20, 3.7].

Note that the module MM in 3.13 is two-periodic since it is maximal Cohen-Macaulay over a hypersurface ring. As we are concerned with periodic modules of even period, we proceed and investigate whether there is an extension of 3.13 for such modules over rings that are not necessarily hypersurfaces. For that we first prove:

Proposition 3.14.

Let RR be a dd-dimensional Cohen-Macaulay local ring with canonical module ω\omega and let MM be a nonzero RR-module that is locally free on the punctured spectrum of RR. Assume d=2nrd=2nr and MΩR2nMM\cong\Omega_{R}^{2n}M for some positive integers nn and rr. Then MRMM\otimes_{R}M^{\dagger} has torsion, where M=HomR(M,ω)M^{\dagger}=\operatorname{Hom}_{R}(M,\omega).

Proof.

Note that, if MRMM\otimes_{R}M^{\dagger} is torsion-free, then MM is zero due to the following isomorphisms:

0=H𝔪0(MRM)\displaystyle{}0=\operatorname{H}_{\mathfrak{m}}^{0}(M\otimes_{R}M^{\dagger})^{\vee} ExtRd(MRM,ω)\displaystyle\cong\operatorname{Ext}^{d}_{R}(M\otimes_{R}M^{\dagger},\omega)
ExtRd(M,M)\displaystyle{}\cong\operatorname{Ext}^{d}_{R}(M,M)
ExtR2n(ΩR2n(r1)M,M)\displaystyle\cong{}\operatorname{Ext}^{2n}_{R}\big{(}\Omega_{R}^{2n(r-1)}M,M\big{)}
ExtR2n(M,M)\displaystyle\cong{}\operatorname{Ext}^{2n}_{R}(M,M)
ExtR1(ΩR2n1M,ΩR2nM).\displaystyle{}{}\cong\operatorname{Ext}^{1}_{R}(\Omega_{R}^{2n-1}M,\Omega_{R}^{2n}M).

Here the first isomorphism follows by the local duality [14, 3.5.9] and the second is due to [29, 2.3]. ∎

The next corollary of Proposition 3.14 yields an extension of 3.13.

Corollary 3.15.

Let RR be an even-dimensional Gorenstein local ring and let MM be a nonzero RR-module that is locally free on the punctured spectrum of RR. If MΩR2MM\cong\Omega_{R}^{2}M, then MRMM\otimes_{R}M^{\ast} has torsion.

We should note that Corollary 3.15 may fail if the dimension of RR is odd; see [20, 3.12].

4. A condition implying Conjecture 1.1 over complete intersection rings

In this section each complete intersection ring RR of codimension cc is of the form S/(x¯)S/(\underline{x}) for some regular local ring (S,𝔫)(S,\mathfrak{n}) and for some SS-regular sequence x¯=x1,,xc𝔫2\underline{x}=x_{1},\ldots,x_{c}\subseteq\mathfrak{n}^{2}. Such a complete intersection ring is called a hypersurface when c=1c=1.

As mentioned in the introduction, although Conjecture 1.1 is true over hypersurface rings, it is wide open for complete intersection rings that have codimension at least two. The aim of this section is to formulate a condition over hypersurfaces, which, if true, forces Conjecture 1.1 to be true over all complete intersection rings.

We start with a setup and then recall a theorem of Orlov [39, Section 2] which plays a key role for our argument. For the basic definitions that are not defined in this section, we refer the reader to [31].

4.1.

Throughout, given a commutative Noetherian ring RR and a scheme TT, 𝗆𝗈𝖽R\operatorname{\operatorname{\mathsf{mod}}}R and 𝖼𝗈𝗁T\operatorname{\operatorname{\mathsf{coh}}}T denote the category of (finitely generated) RR-modules and the category of coherent 𝒪T\mathcal{O}_{T}-modules, respectively. It follows that there is a category equivalence: 𝗆𝗈𝖽R𝖼𝗈𝗁(Spec(R))\operatorname{\operatorname{\mathsf{mod}}}R\cong\operatorname{\operatorname{\mathsf{coh}}}(\operatorname{Spec}(R)) given by MM~M\mapsto\widetilde{M}, see, for example, [31, II 5.5].

4.2.

Let XX be a scheme. A perfect complex on XX is a bounded complex of coherent sheaves on XX which has finite flat dimension as a complex.

The singularity category is defined as the quotient:

𝖣sg(X)=𝖣b(𝖼𝗈𝗁X)/𝖣perf(X),\operatorname{\operatorname{\mathsf{D}_{sg}}}(X)=\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}X)/\operatorname{\operatorname{\mathsf{D}^{perf}}}(X),

where 𝖣perf(X)\operatorname{\operatorname{\mathsf{D}^{perf}}}(X) denotes the full subcategory of the bounded derived category 𝖣b(𝖼𝗈𝗁X)\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}X) consisting of all perfect complexes on XX.

Note that a perfect complex on Spec(R)\operatorname{Spec}(R) is isomorphic in 𝖣b(𝗆𝗈𝖽R)\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{mod}}}R) to a bounded complex of finitely generated projective modules via the equivalence 𝗆𝗈𝖽R𝖼𝗈𝗁(Spec(R))\operatorname{\operatorname{\mathsf{mod}}}R\cong\operatorname{\operatorname{\mathsf{coh}}}(\operatorname{Spec}(R)) mentioned in 4.1. Moreover, if RR is Gorenstein, then the singularity category of RR is equivalent to the stable category of maximal Cohen-Macaulay RR-modules by [15, 4.4.1].

4.3.

Let RR be a complete intersection ring of codimension cc and let t¯=t1,,tc\underline{t}=t_{1},\ldots,t_{c} be indeterminates over SS. Then we define a graded hypersurface ring

A=S[t¯]/(i=1cxiti)\displaystyle{A=S[\underline{t}]/\bigg{(}\sum_{i=1}^{c}x_{i}t_{i}\bigg{)}}

where the grading is given by degs=0\deg s=0 for all sSs\in S, and degti=1\deg t_{i}=1 for each ii.

Now we consider the natural surjections:

S[t¯]AA/(x¯)=R[t¯].S[\underline{t}]\twoheadrightarrow A\twoheadrightarrow A/(\underline{x})=R[\underline{t}].

These surjections yield the following commutative diagram of schemes:

Z:=Rc1\textstyle{Z:=\mathbb{P}^{c-1}_{R}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}i\scriptstyle{i}p\scriptstyle{p}Y:=Proj(A)\textstyle{Y:=\operatorname{Proj}(A)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}u\scriptstyle{u}X:=Sc1\textstyle{X:=\mathbb{P}^{c-1}_{S}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}q\scriptstyle{q}Spec(R)\textstyle{\operatorname{Spec}(R)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}j\scriptstyle{j}Spec(S).\textstyle{\operatorname{Spec}(S).}

In the above diagram, the morphisms ii, uu, and jj are closed immersions that are induced by the surjections AR[t¯]A\twoheadrightarrow R[\underline{t}], S[t¯]AS[\underline{t}]\twoheadrightarrow A, SRS\twoheadrightarrow R, respectively. Also, the morphisms pp and qq are canonical. We note that ii is a regular closed immersion of codimension c1c-1, that is, the ideal sheaf of ii is locally generated by a regular sequence of length c1c-1. Furthermore, the morphism pp is flat.

We consider two functors p:𝗆𝗈𝖽R𝖼𝗈𝗁Zp^{*}:\operatorname{\operatorname{\mathsf{mod}}}R\to\operatorname{\operatorname{\mathsf{coh}}}Z and i:𝖼𝗈𝗁Z𝖼𝗈𝗁Yi_{*}:\operatorname{\operatorname{\mathsf{coh}}}Z\to\operatorname{\operatorname{\mathsf{coh}}}Y, which are defined as follows:

  1. (i)

    p:𝗆𝗈𝖽R𝖼𝗈𝗁(Spec(R))𝖼𝗈𝗁Zp^{*}:\operatorname{\operatorname{\mathsf{mod}}}R\cong\operatorname{\operatorname{\mathsf{coh}}}(\operatorname{Spec}(R))\to\operatorname{\operatorname{\mathsf{coh}}}Z is the pullback along pp, where p(M)p^{*}(M) is the 𝒪Z\mathcal{O}_{Z}-module M[t¯]~\widetilde{M[\underline{t}]} associated with a graded R[t¯]R[\underline{t}]-module M[t¯]=MRR[t¯](MRA)/x(MRA)M[\underline{t}]=M\otimes_{R}R[\underline{t}]\cong(M\otimes_{R}A)/x(M\otimes_{R}A); see [31, p116, Definition].

  2. (ii)

    i:𝖼𝗈𝗁Z𝖼𝗈𝗁Yi_{*}:\operatorname{\operatorname{\mathsf{coh}}}Z\to\operatorname{\operatorname{\mathsf{coh}}}Y is the pushout along ii. Note that, every object of 𝖼𝗈𝗁Z\operatorname{\operatorname{\mathsf{coh}}}Z is isomorphic to L~\widetilde{L} for some graded R[t¯]R[\underline{t}]-module LL. Then i(L~)i_{*}(\widetilde{L}) is isomrphic to LA~\widetilde{L_{A}}, where LAL_{A} is the graded R[t¯]R[\underline{t}]-module LL considered as a graded AA-module via the ring map AR[t¯]A\twoheadrightarrow R[\underline{t}].

Notice, since pp is flat and ii is closed immersion, it follows that pp^{\ast} and ii_{*} are exact functors, see [7, 02N4 and 01QY]. Therefore, by deriving these functors, we also obtain triangle functors:

i:𝖣b(𝖼𝗈𝗁Z)𝖣b(𝖼𝗈𝗁Y) and p:𝖣b(𝗆𝗈𝖽R)𝖣b(𝖼𝗈𝗁Z).{}i_{*}:\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Z)\to\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Y)\text{ and }p^{*}:\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{mod}}}R)\to\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Z).

Here, the functors are given by applying pp^{*} and ii_{*} component-wise. ∎

4.4.

Note that Y=ProjAY=\operatorname{Proj}A is an integral scheme and hence every ring of section is an integral domain. Indeed, since S[t1,,tc]S[t_{1},\ldots,t_{c}] is an integral domain, we can easily check that i=1cxiti\sum_{i=1}^{c}x_{i}t_{i} is an irreducible element. Therefore, it is a prime element as S[t1,,tc]S[t_{1},\ldots,t_{c}] is a UFD and hence AA is a domain; see 4.3. ∎

Throughout this section we keep the notations and the setting of 4.3. The following result of Orlov [39] plays a key role in the proof Theorem 4.12.

4.5.

([39, 2.1], see also [16, A.4]) The triangle functor Φ=ip:𝖣b(𝗆𝗈𝖽R)𝖣b(𝖼𝗈𝗁Y)\Phi=i_{*}p^{*}:\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{mod}}}R)\to\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Y) induces a triangle equivalence Φ¯:𝖣sg(R)𝖣sg(Y)\overline{\Phi}:\operatorname{\operatorname{\mathsf{D}_{sg}}}(R)\xrightarrow{\cong}\operatorname{\operatorname{\mathsf{D}_{sg}}}(Y).

Note, Φ(M)(MRA)/x¯(MRA)~\Phi(M)\cong\widetilde{(M\otimes_{R}A)/\underline{x}(M\otimes_{R}A)} for any RR-module MM. Moreover, we have

pdRM<M0 in 𝖣sg(R)Φ(M)0 in 𝖣sg(Y)\displaystyle\operatorname{pd}_{R}M<\infty\Longleftrightarrow M\cong 0\text{ in }\operatorname{\operatorname{\mathsf{D}_{sg}}}(R)\Longleftrightarrow\Phi(M)\cong 0\text{ in }\operatorname{\operatorname{\mathsf{D}_{sg}}}(Y)
fd𝒪YΦ(M)<pd𝒪Y,yΦ(M)y< for all yY.\displaystyle\qquad\Longleftrightarrow\operatorname{fd}_{\mathcal{O}_{Y}}\Phi(M)<\infty\Longleftrightarrow\operatorname{pd}_{\mathcal{O}_{Y,y}}\Phi(M)_{y}<\infty\mbox{ for all }y\in Y.

Here, the first and third equivalences follow by the definition of singularity categories, the second equivalence follows from Orlov’s theorem 4.5, and [31, III 9.2(e)] proves the last equivalence.

4.6.

Let MM and NN be RR-modules. Then we have the following:

Φ(M)𝒪YΦ(N)\displaystyle\Phi(M)\otimes_{\mathcal{O}_{Y}}\Phi(N) =((MRA)/x¯(MRA))~𝒪Y((NRA)/x¯(NRA))~\displaystyle=\widetilde{\big{(}(M\otimes_{R}A)/\underline{x}(M\otimes_{R}A)\big{)}}\otimes_{\mathcal{O}_{Y}}\widetilde{\big{(}(N\otimes_{R}A)/\underline{x}(N\otimes_{R}A)\big{)}}
([(MRA)/x¯(MRA)]A[(NRA)/x¯(NRA))])~\displaystyle\cong\widetilde{\bigg{(}\big{[}(M\otimes_{R}A)/\underline{x}(M\otimes_{R}A)\big{]}\otimes_{A}\big{[}(N\otimes_{R}A)/\underline{x}(N\otimes_{R}A))\big{]}\bigg{)}}
=([(MRN)RA]/x¯[(MRN)RA])~\displaystyle=\widetilde{\bigg{(}\big{[}(M\otimes_{R}N)\otimes_{R}A\big{]}/\underline{x}\big{[}(M\otimes_{R}N)\otimes_{R}A\big{]}\bigg{)}}
=Φ(MRN).\displaystyle=\Phi(M\otimes_{R}N).

where the isomorphism follows from [31, Proof of II 5.12(b)]. ∎

Next we proceed to determine Φ(M)\Phi(M^{*}). For this, we use the Grothendieck duality theorem [30].

Lemma 4.7.

Let 𝖼𝗈𝗁Z\mathcal{F}\in\operatorname{\operatorname{\mathsf{coh}}}Z and nn\in\mathbb{Z}. Then there is a natural isomorphism as follows:

xt𝒪Yn(i,𝒪Y)ixt𝒪Znc+1(,𝒪Z)(1).\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{n}(i_{*}\mathcal{F},\mathcal{O}_{Y})\cong i_{*}\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Z}}^{n-c+1}(\mathcal{F},\mathcal{O}_{Z})(1).
Proof.

Note, by the Grothendieck duality theorem [30, III 6.7], there is a natural isomorphism

Rom𝒪Y(i,𝒪Y)iRom𝒪Z(,i!𝒪Y),\operatorname{R\mathcal{H}om}_{\mathcal{O}_{Y}}(i_{*}\mathcal{F},\mathcal{O}_{Y})\cong i_{*}\operatorname{R\mathcal{H}om}_{\mathcal{O}_{Z}}(\mathcal{F},i^{!}\mathcal{O}_{Y}),

where i!:𝖣b(𝖼𝗈𝗁Y)𝖣b(𝖼𝗈𝗁Z)i^{!}:\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Y)\to\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Z) denotes the right adjoint functor of i:𝖣b(𝖼𝗈𝗁Z)𝖣b(𝖼𝗈𝗁Y)i_{*}:\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Z)\to\operatorname{\operatorname{\mathsf{D}^{b}}}(\operatorname{\operatorname{\mathsf{coh}}}Y). We proceed to prove that i!𝒪Yi^{!}\mathcal{O}_{Y} is isomorphic to 𝒪Z(1)[c+1]\mathcal{O}_{Z}(1)[-c+1].

Note that, by [30, III 7.3], there is an isomorphism i!𝒪YωZ/Y[c+1]i^{!}\mathcal{O}_{Y}\cong\omega_{Z/Y}[-c+1]. Here, ωZ/Y\omega_{Z/Y} is the relative canonical sheaf of the regular closed immersion i:ZYi:Z\hookrightarrow Y; see [30, III §1] for its definition. On the other hand, by [30, III 1.5], we have an isomorphism of the form ωZ/YωZ/X𝒪Z(iωY/X)\omega_{Z/Y}\cong\omega_{Z/X}\otimes_{\mathcal{O}_{Z}}(i^{*}\omega_{Y/X})^{\vee}. The ideal sheaves of ui:ZXui:Z\hookrightarrow X and u:YXu:Y\hookrightarrow X are globally generated by degree 0 and 11 regular sequences, respectively. Therefore, the following isomorphisms hold:

ωZ/X𝒪Z and ωY/X𝒪Y(1).\omega_{Z/X}\cong\mathcal{O}_{Z}\,\,\ \text{ and }\,\,\,\omega_{Y/X}\cong\mathcal{O}_{Y}(-1).

Hence we conclude that i!𝒪YωZ/Y[c+1]𝒪Z𝒪Zi(𝒪Y(1))[c+1]𝒪Z(1)[c+1]i^{!}\mathcal{O}_{Y}\cong\omega_{Z/Y}[-c+1]\cong\mathcal{O}_{Z}\otimes_{\mathcal{O}_{Z}}i^{*}(\mathcal{O}_{Y}(-1))^{\vee}[-c+1]\cong\mathcal{O}_{Z}(1)[-c+1]. ∎

4.8.

Let MM be an RR-module. Then there are natural isomorphisms

xt𝒪Yn(Φ(M),𝒪Y)\displaystyle\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{n}(\Phi(M),\mathcal{O}_{Y}) =xt𝒪Yn(ip(M),𝒪Y)\displaystyle=\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{n}(i_{*}p^{*}(M),\mathcal{O}_{Y})
ixt𝒪Znc+1(p(M),𝒪Z)(1)\displaystyle\cong i_{*}\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Z}}^{n-c+1}(p^{*}(M),\mathcal{O}_{Z})(1)
Φ(ExtRnc+1(M,R))(1)\displaystyle\cong\Phi(\operatorname{Ext}_{R}^{n-c+1}(M,R))(1)

Here, the first isomorphism uses 4.7 and the last isomorphism is due to [31, III 6.5] together with the following fact: p(P)p^{*}(P_{\bullet}) is a resolution of p(M)p^{*}(M) by locally free sheaves of finite rank for a given resolution PP_{\bullet} by finite free module of MM.

Assume further that MM is a totally reflexive RR-module. Then, by definition, if n0n\neq 0, it follows that ExtRn(M,R)0\operatorname{Ext}_{R}^{n}(M,R)\cong 0. Therefore, the above isomorphisms yield:

xt𝒪Yi(Φ(M),𝒪Y){0(ic1)Φ(M)(1)(i=c1).\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{i}(\Phi(M),\mathcal{O}_{Y})\cong\begin{cases}0&(i\neq c-1)\\ \Phi(M^{*})(1)&(i=c-1).\end{cases}

Next we recall the definition of the codimension of a module:

4.9.

Let AA be a local ring and let NN be an AA-module. Then the codimension codimA(N)\operatorname{codim}_{A}(N) of NN is defined as the codimension of its support SuppA(N)\operatorname{Supp}_{A}(N) as a closed set in Spec(A)\operatorname{Spec}(A). More precisely, we have

codimA(N)=inf{heightA(𝔭)𝔭SuppA(N)}.{}\operatorname{codim}_{A}(N)=\inf\{\operatorname{height}_{A}(\mathfrak{p})\mid\mathfrak{p}\in\operatorname{Supp}_{A}(N)\}.

In the following we record some properties of the codimension that are needed for our argument; see [14, 2.1.2 and 3.3.10]. Among those is the fact that the codimension of a module does not change when localizing at a prime ideal in its support.

4.10.

Let AA be a Cohen-Macaulay local ring and let NN be a nonzero AA-module.

  1. (i)

    It follows that codimA(N)=dim(A)dimA(N)=gradeA(N)=inf{iExtAi(N,A)0}\operatorname{codim}_{A}(N)=\dim(A)-\dim_{A}(N)=\operatorname{grade}_{A}(N)=\inf\{i\in\mathbb{Z}\mid\operatorname{Ext}_{A}^{i}(N,A)\neq 0\}, where gradeA(N)\operatorname{grade}_{A}(N) denotes the grade of NN over AA.

  2. (ii)

    Assume AA admits a canonical module ωA\omega_{A}.

    1. (a)

      Then NN is Cohen-Macaulay of codimension tt if and only if ExtAi(N,ωA)=0\operatorname{Ext}_{A}^{i}(N,\omega_{A})=0 for iti\neq t.

    2. (b)

      If NN is Cohen-Macaulay of codimension tt and 𝔭SuppA(N)\mathfrak{p}\in\operatorname{Supp}_{A}(N), then N𝔭N_{\mathfrak{p}} is Cohen-Macaulay over A𝔭A_{\mathfrak{p}} such that codimA𝔭(N𝔭)=t\operatorname{codim}_{A_{\mathfrak{p}}}(N_{\mathfrak{p}})=t.

    3. (c)

      If NN is Cohen-Macaulay of codimension tt, then it follows SuppA(N)=SuppA(ExtAt(N,ωA))\operatorname{Supp}_{A}(N)=\operatorname{Supp}_{A}(\operatorname{Ext}_{A}^{t}(N,\omega_{A})).

We define the following conditions for local domains RR:

4.11.

Let c1c\geq 1 be an integer.

  1. (i)

    If MM and MRExtRc1(M,R)M\otimes_{R}\operatorname{Ext}_{R}^{c-1}(M,R) are Cohen-Macaulay RR-modules of codimension c1c-1, then it follows that pdR(M)<\operatorname{pd}_{R}(M)<\infty.

  2. (ii)

    If MM and MRMM\otimes_{R}M^{\ast} are maximal Cohen-Macaulay (i.e., Cohen-Macaulay RR-modules of codimension 0), then it follows MM is free. ∎

Next is the main result of this section; recall that we keep the notations and the setting of 4.3.

Theorem 4.12.

Let c1c\geq 1 be an integer. If each local hypersurface domain satisfies condition (i) of 4.11, then each local complete intersection domain of codimension cc satisfies condition (ii) of 4.11.

Remark 4.13.

Prior to giving a proof for Theorem 4.12, we remark that condition (ii) of 4.11, or equivalently condition (i) for the case where c=1c=1, is nothing but the condition stated in Conjecture 1.1 for one-dimensional local domains. Recall that each hypersurface local domain satisfies condition (ii) of 4.11; see [35, 3.1].

Proof of Theorem 4.12.

We assume, for the given integer cc, that each local hypersurface domain satisfies condition (i) of 4.11.

Let R=S/(x¯)R=S/(\underline{x}) be a domain, where SS is a regular local ring and x¯\underline{x} is an SS-regular sequence of length cc. Let MM be a maximal Cohen-Macaulay RR-module such that MRMM\otimes_{R}M^{*} is maximal Cohen-Macaulay. We proceed to prove that MM is free.

First we prove that SuppY(Φ(M))=SuppY(xt𝒪Yc1(Φ(M),𝒪Y))\operatorname{Supp}_{Y}(\Phi(M))=\operatorname{Supp}_{Y}(\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{c-1}(\Phi(M),\mathcal{O}_{Y})). Fix ySuppY(Φ(M))y\in\operatorname{Supp}_{Y}(\Phi(M)). The combination of 4.8 and 4.10(ii)(a) show that Φ(M)y\Phi(M)_{y} is a Cohen-Macaulay 𝒪Y,y\mathcal{O}_{Y,y}-module of codimension c1c-1. Therefore, by 4.10(ii)(c), we conclude ySuppY(xt𝒪Yc1(Φ(M),𝒪Y))y\in\operatorname{Supp}_{Y}(\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{c-1}(\Phi(M),\mathcal{O}_{Y})) and hence it follows that SuppY(Φ(M))SuppY(xt𝒪Yc1(Φ(M),𝒪Y))\operatorname{Supp}_{Y}(\Phi(M))\subseteqq\operatorname{Supp}_{Y}(\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{c-1}(\Phi(M),\mathcal{O}_{Y})). The converse inclusion is trivial. Moreover, the support of Φ(M)\Phi(M) equals to the support of 𝒳\mathcal{X}, where 𝒳=Φ(M)𝒪Yxt𝒪Yc1(Φ(M),𝒪Y)(1)\mathcal{X}=\Phi(M)\otimes_{\mathcal{O}_{Y}}\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{c-1}(\Phi(M),\mathcal{O}_{Y})(-1).

Note, it follows from 4.6 and 4.8 that Φ(MRM)Φ(M)𝒪YΦ(M)𝒳\Phi(M\otimes_{R}M^{*})\cong\Phi(M)\otimes_{\mathcal{O}_{Y}}\Phi(M^{*})\cong\mathcal{X}. As MM and MRMM\otimes_{R}M^{*} are totally reflexive RR-module, by 4.8, we obtain the following:

(4.12.1) xt𝒪Yi(𝒳,𝒪Y)xt𝒪Yi(Φ(MRM),𝒪Y)Φ(ExtRic+1(MRM,R))(1)=0\displaystyle\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{i}(\mathcal{X},\mathcal{O}_{Y})\cong\operatorname{\mathcal{E}xt}_{\mathcal{O}_{Y}}^{i}(\Phi(M\otimes_{R}M^{*}),\mathcal{O}_{Y})\cong\Phi(\operatorname{Ext}_{R}^{i-c+1}(M\otimes_{R}M^{*},R))(1)=0

for ic1i\neq c-1.

Let ySuppY(Φ(M))y\in\operatorname{Supp}_{Y}(\Phi(M)). Then the module 𝒳y=Φ(M)y𝒪Y,yExt𝒪Y,yc1(Φ(M)y,𝒪Y,y)\mathcal{X}_{y}=\Phi(M)_{y}\otimes_{\mathcal{O}_{Y,y}}\operatorname{Ext}_{\mathcal{O}_{Y,y}}^{c-1}(\Phi(M)_{y},\mathcal{O}_{Y,y}) is Cohen-Macaulay of codimension c1c-1 over the local hypersurface domain 𝒪Y,y\mathcal{O}_{Y,y} by 4.10(ii)(a) and (4.12.1). Now our assumption shows that Φ(M)y\Phi(M)_{y} has finite projective dimension over 𝒪Y,y\mathcal{O}_{Y,y}. So, by 4.5, we have that pdR(M)<\operatorname{pd}_{R}(M)<\infty

Now the proof of Theorem 1.4 follows as an immediate consequence of Theorem 4.12:

Proof of Theorem 1.4.

Let c1c\geq 1 be an integer. If each local hypersurface domain satisfies condition (i) of 4.11 (for the given cc), then Theorem 4.12 implies that Conjecture 1.1 is true over each one-dimensional local complete intersection domain. ∎

Remark 4.14.

We note a fact that follows from the proof of Theorem 4.12: if each hypersurface domain which is quotient of an equi-characteristic regular local ring satisfies condition (i) of Theorem 4.12, then Conjecture 1.1 holds over each one-dimensional complete intersection domain which is quotient of an equi-characteristic regular local ring. ∎

We finish this section by noting that, if we consider Conjecture 1.1 over one-dimensional complete intersection domains that have algebraically closed residue fields, then the proof of Theorem 4.12 is simplified significantly due to a result in [13]:

Remark 4.15.

Let R=S/IR=S/I be a one-dimensional local complete intersection domain of codimension cc with algebraically closed residue field, and let MM be a torsion-free RR-module such that I𝔫2I\subseteq\mathfrak{n}^{2} and MRMM\otimes_{R}M^{*} is torsion-free.

Let fI𝔫If\in I-\mathfrak{n}I. Then we can extend {f}\{f\} to a minimal generating set {f1,,fc}\{f_{1},\ldots,f_{c}\} of II, with f=f1f=f_{1}, which is necessarily a regular sequence on SS. Hence MS/(f1)ExtS/(f1)c1(M,S/(f1))MRMM\otimes_{S/(f_{1})}\operatorname{Ext}_{S/(f_{1})}^{c-1}(M,S/(f_{1}))\cong M\otimes_{R}M^{*} is a Cohen-Macaulay S/(f1)S/(f_{1})-module of codimension c1c-1. Assuming the condition 4.11(ii), it follows that pdS/(f1)(M)<\operatorname{pd}_{S/(f_{1})}(M)<\infty. Now [13, 3.3] implies that pdR(M)<\operatorname{pd}_{R}(M)<\infty.

Appendix: a remark on the rigidity of Tor

It is known that Tor-rigidity, a subtle property, is a sufficient condition for Conjecture 1.1 to hold over one-dimensional Gorenstein domains; see 3.4 and Remark 3.10(ii). Motivated by this fact, we examine the vanishing of Tor more closely over Gorenstein rings. The observation we aim to establish in this appendix is the following, which may be helpful for further study Tor-rigidity.

4.16.

Let RR be a one-dimensional local Gorenstein domain and let MM be an RR-module. Then the following conditions are equivalent:

  1. (i)

    MM is Tor-rigid over RR.

  2. (ii)

    (M,C)(M,C) is Tor-rigid for each torsion (or equivalently, finite length) RR-module CC.

  3. (iii)

    (M,C)(M,C) is Tor-rigid for each torsion-free (or equivalently, maximal Cohen-Macaulay) RR-module CC.

We deduce 4.16 from the following more general result:

Proposition 4.17.

Let RR be a dd-dimensional Gorenstein local ring and let MM be an RR-module. Then the following conditions are equivalent:

  1. (i)

    If CC is a torsion RR-module and TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0 for some ndn\geq d, then it follows ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n.

  2. (ii)

    If CC is a maximal Cohen-Macaulay RR-module that has rank and TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0 for some ndn\geq d, then it follows ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n.

  3. (iii)

    If CC is an RR-module with rank and TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0 for some ndn\geq d, then it follows ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n.

Proof.

First we show that parts (ii) and (iii) are equivalent, that is, part (ii) implies part (iii). For that assume part (ii) holds. Let CC be an RR-module with rank such that TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0 for some ndn\geq d. We want to show that ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n.

We may assume CC is not maximal Cohen-Macaulay. Then we consider a Cohen-Macaulay approximation of CC, that is, a short exact sequence of RR-modules 0LXC00\to L\to X\to C\to 0, where pdR(L)<\operatorname{pd}_{R}(L)<\infty and XX is maximal Cohen-Macaulay; see 2.7(ii). Note pdR(L)<d\operatorname{pd}_{R}(L)<d so that TornR(M,X)=0\operatorname{Tor}_{n}^{R}(M,X)=0. Thus, as XX has rank, it follows from the hypothesis that ToriR(M,X)=0\operatorname{Tor}_{i}^{R}(M,X)=0 for all ini\geq n. This yields ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n, and hence establishes part (iii).

Next we show that part (i) implies (ii). Assume part (i) holds and let CC be a maximal Cohen-Macaulay RR-module with rank such that TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0 for some ndn\geq d. We want to show that ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n. As CC has rank, there exists a short exact sequence 0CGY00\to C\to G\to Y\to 0, where GG is free and YY is torsion; see [35, 1.3]. As TornR(M,C)=0\operatorname{Tor}_{n}^{R}(M,C)=0, we have that Torn+1R(M,Y)=0\operatorname{Tor}_{n+1}^{R}(M,Y)=0; now the hypothesis implies that ToriR(M,Y)=0\operatorname{Tor}_{i}^{R}(M,Y)=0 for all in+1i\geq n+1. Consequently, ToriR(M,C)=0\operatorname{Tor}_{i}^{R}(M,C)=0 for all ini\geq n, as required.

Finally we note a module is torsion if and only if it has rank zero. So part (iii) implies part (i). ∎

Acknowledgements

The authors thank Kenta Sato for pointing to them a simpler proof of Lemma 4.7 than the one in a previous version of the manuscript.

References

  • [1] Mohsen Asgharzadeh, Olgur Celikbas, and Arash Sadeghi, A study of the cohomological rigidity property, preprint (2020), posted at arXiv:2009.06481.
  • [2] Maurice Auslander, Modules over unramified regular local rings, Illinois J. Math. 5 (1961), 631–647.
  • [3] Maurice Auslander and Mark Bridger, Stable module theory, Memoirs of the American Mathematical Society, No. 94, American Mathematical Society, Providence, R.I., 1969.
  • [4] Maurice Auslander and Ragnar-O. Buchweitz, The homological theory of maximal Cohen-Macaulay approximations, Mém. Soc. Math. France (N.S.) (1989), no. 38, 5–37, Colloque en l’honneur de Pierre Samuel (Orsay, 1987).
  • [5] Maurice Auslander and Oscar Goldman, Maximal orders, Trans. Amer. Math. Soc. 97 (1960), 1–24.
  • [6] Maurice Auslander and Idun Reiten, On a generalized version of the Nakayama conjecture, Proc. Amer. Math. Soc. 52 (1975), 69–74.
  • [7] The Stacks Project Authors, Stacks project, available at https://stacks.math.columbia.edu.
  • [8] Luchezar L. Avramov, Modules of finite virtual projective dimension, Invent. Math. 96 (1989), no. 1, 71–101.
  • [9] Luchezar L. Avramov and Ragnar-O. Buchweitz, Support varieties and cohomology over complete intersections, Invent. Math. 142 (2000), no. 2, 285–318.
  • [10] Luchezar L. Avramov, Vesselin N. Gasharov, and Irena V. Peeva, Complete intersection dimension, Inst. Hautes Études Sci. Publ. Math. (1997), no. 86, 67–114 (1998).
  • [11] Luchezar L. Avramov and Alex Martsinkovsky, Absolute, relative, and Tate cohomology of modules of finite Gorenstein dimension, Proc. London Math. Soc. (3) 85 (2002), no. 2, 393–440.
  • [12] Petter Andreas Bergh and David A. Jorgensen, On growth in minimal totally acyclic complexes, J. Commut. Algebra 6 (2014), no. 1, 17–31.
  • [13] Petter Andreas Bergh and David A. Jorgensen, A generalized Dade’s lemma for local rings, Algebr. Represent. Theory 21 (2018), 1369–1380.
  • [14] Winfried Bruns and Jürgen Herzog, Cohen-Macaulay rings, Cambridge Studies in Advanced Mathematics, vol. 39, Cambridge University Press, Cambridge, 1993.
  • [15] R.-O. Buchweitz, Maximal Cohen-Macaulay modules and Tate cohomology over Gorenstein rings, preprint, http://hdl.handle.net/1807/16682, 1986.
  • [16] Jesse Burke and Mark E. Walker, Matrix factorizations in higher codimension, Trans. Amer. Math. Soc. 367 (2015), 3323–3370.
  • [17] Olgur Celikbas, Vanishing of Tor over complete intersections, J. Commut. Algebra 3 (2011), no. 2, 169–206. MR 2813471
  • [18] by same author, On the vanishing of the theta invariant and a conjecture of Huneke and Wiegand, Pacific J. Math. 309 (2020), no. 1, 103–144.
  • [19] Olgur Celikbas and Hailong Dao, Asymptotic behavior of Ext functors for modules of finite complete intersection dimension, Math. Z. 269 (2011), no. 3-4, 1005–1020.
  • [20] by same author, Necessary conditions for the depth formula over Cohen-Macaulay local rings, J. Pure Appl. Algebra 218 (2014), no. 3, 522–530.
  • [21] Olgur Celikbas and Arash Sadeghi, Maximal Cohen-Macaulay tensor products, Ann. Mat. Pura Appl. (4) 200 (2021), no. 3, 923–944.
  • [22] Lars Winther Christensen and Srikanth Iyengar, Gorenstein dimension of modules over homomorphisms, J. Pure Appl. Algebra 208 (2007), no. 1, 177–188.
  • [23] Lars Winther Christensen and David A. Jorgensen, Vanishing of Tate homology and depth formulas over local rings, J. Pure Appl. Algebra 219 (2015), no. 3, 464–481.
  • [24] Hailong Dao, Asymptotic behaviour of Tor over complete intersections and applications, preprint (2008), posted at arxiv:07105818.
  • [25] by same author, Some observations on local and projective hypersurfaces, Math. Res. Lett. 15 (2008), no. 2, 207–219.
  • [26] by same author, Decent intersection and tor-rigidity for modules over local hypersurfaces, Transactions of the American Mathematical Society 365 (2013), no. 6, 2803–2821.
  • [27] by same author, Some homological properties of modules over a complete intersection, with applications, Commutative algebra, Springer, New York, 2013, pp. 335–371.
  • [28] Vesselin N. Gasharov and Irena V. Peeva, Boundedness versus periodicity over commutative local rings, Trans. Amer. Math. Soc. 320 (1990), no. 2, 569–580.
  • [29] Shiro Goto and Ryo Takahashi, On the Auslander-Reiten conjecture for Cohen-Macaulay local rings, Proc. Amer. Math. Soc. 145 (2017), no. 8.
  • [30] Robin Hartshorne, Residues and duality, Lecture Notes in Math., vol. 20, Springer-Verlag, 1966.
  • [31] by same author, Algebraic geometry, Graduate Texts in Mathematics, No. 52, Springer-Verlag, New York-Heidelberg, 1977.
  • [32] Melvin Hochster, The dimension of an intersection in an ambient hypersurface, Algebraic geometry (Chicago, Ill., 1980), Lecture Notes in Math., vol. 862, Springer, Berlin, 1981, pp. 93–106.
  • [33] Craig Huneke, Srikanth B. Iyengar, and Roger Wiegand, Rigid ideals in Gorenstein rings of dimension one, Acta Math. Vietnam. 44 (2019), no. 1, 31–49.
  • [34] Craig Huneke and David A. Jorgensen, Symmetry in the vanishing of Ext over Gorenstein rings, Math. Scand. 93 (2003), no. 2, 161–184.
  • [35] Craig Huneke and Roger Wiegand, Tensor products of modules and the rigidity of Tor, Math. Ann. 299 (1994), no. 3, 449–476.
  • [36] David A. Jorgensen, A generalization of the Auslander-Buchsbaum formula, J. Pure Appl. Algebra 144 (1999), no. 2, 145–155.
  • [37] Stephen Lichtenbaum, On the vanishing of Tor in regular local rings, Illinois J. Math. 10 (1966), 220–226.
  • [38] Hideyuki Matsumura, Commutative ring theory, second ed., Cambridge Studies in Advanced Mathematics, vol. 8, Cambridge University Press, Cambridge, 1989, Translated from the Japanese by M. Reid.
  • [39] Dmitri O. Orlov, Triangulated categories of singularities, and equivalences between landau-ginzburg models, Mat. Sb. 197 (2006), no. 12, 117–132.
  • [40] Sean Sather-Wagstaff, Complete intersection dimensions for complexes, J. Pure Appl. Algebra 190 (2004), no. 1-3, 267–290.
  • [41] Yuji Yoshino, Cohen-Macaulay modules over Cohen-Macaulay rings, London Mathematical Society Lecture Note Series, vol. 146, Cambridge University Press, Cambridge, 1990.