Graded Cohen-Macaulay rings of wild Cohen-Macaulay type (Q2443275)

Let \(k\) be an infinite field and \(R = \bigoplus_{i = 0}^\infty R_i\) a graded commutative \(k\)-algebra with \(R_0 = k\), \(\dim_k R_1 < \infty\) and such that \(R = k[R_1]\). One assumes that \(R\) is Cohen-Macaulay (CM for short). In this case the projective variety \(X = \text{Proj}\, R\) is arithmetically CM (ACM for short) and the category of graded maximal Cohen-Macaulay \(R\)-modules is equivalent to the category of ACM coherent sheaves over \(X\). If \(X\) is regular then any ACM sheaf is locally free. \(R\) is called \textit{algebraically CM wild} if, for any (not necessarily commutative) finitely generated \(k\)-algebra \(\Lambda\), there exists a finitely generated graded \(R\text{-}\Lambda\)-bimodule \(\mathcal M\) (the grading of \(R\otimes_k\Lambda\) being defined by \((R\otimes_k\Lambda)_i = R_i\otimes_k\Lambda\)) which is \textit{flat} over \(\Lambda\) and such that: \begin{itemize} \item[(0)] for any \(\Lambda\)-module \(L\) with \(\dim_k L < \infty\), \({\mathcal M}\otimes_\Lambda L\) is a maximal CM \(R\)-module; \item[(1)] \({\mathcal M}\otimes_\Lambda L \simeq {\mathcal M}\otimes_\Lambda L^\prime\) implies \(L \simeq L^\prime\); \item[(2)] if \(L\) is indecomposable then so is \({\mathcal M}\otimes_\Lambda L\). \end{itemize} \textit{Y. A. Drozd} [C. R. Math. Acad. Sci., Soc. R. Can. 23, No. 4, 97--125 (2001; Zbl 1031.16010)] noticed that in order to show that \(R\) is CM wild it is enough to construct such a module \(\mathcal M\) for only one algebra \(\Lambda\) which is wild in the sense of representation theory. According to \textit{I. M. Gel'fand} and \textit{V. A. Ponomarev} [Funkts. Anal. Prilozh. 3, No. 4, 81--82 (1969; Zbl 0204.45301)], the localizations of \(k[x,y]\) are examples of such algebras. The main result of the paper under review asserts that if \(\underline{y} = (y_1, \ldots ,y_d)\) is a maximal \(R\)-regular sequence consisting of homogeneous elements \(y_i\) of degree \(m_i > 0\), \(i = 1, \ldots , d\), and if \(\dim_k(R/\underline{y}R)_c > 2\) for some integer \(c > m - d + 1\), where \(m = \sum_{i = 1}^d m_i\), then \(R\) is CM wild. Actually, putting \(\Lambda = k[x,y]\), the authors construct, for some \(0 \neq a \in \Lambda\), a graded \(R\text{-}\Lambda[a^{-1}]\)-bimodule \({\mathcal M}^\prime\) which is flat over \(\Lambda[a^{-1}]\) and satisfies the conditions (0)--(2) above. The construction is based on a generalization of an idea of \textit{D. Eisenbud} and \textit{J. Herzog} [Math. Ann. 280, No. 2, 347--352 (1988; Zbl 0616.13011)]. More precisely, let \(\overline{R} := R/\underline{y}R\), let \(e_1, e_2, e_3\) be three linearly independent elements of \(\overline{R}_c\) and let \({\mathcal N} := \overline{R}\otimes_k\Lambda /(e_1\otimes 1 + e_2\otimes x + e_3 \otimes y)\). Using Grothendieck's Generic Freeness Lemma one shows that there exists \(0 \neq a \in \Lambda\) such that \({\mathcal N}^\prime := {\mathcal N}[a^{-1}]\) is a free \(\Lambda^\prime = \Lambda[a^{-1}]\)-module and admits a graded resolution: \[ \begin{tikzcd} {\mathcal F}_{d-1} \rar["\delta_{d-1}"] & \cdots \rar["\delta_1"] & \mathcal F_0 \rar["\delta_0"] & \mathcal N^\prime \rar & 0 \end{tikzcd} \] such that \({\mathcal F}_i\) is a direct sum of modules of the form \(R(-l)\otimes_k\Lambda^\prime\), \(\text{Im}\, \delta_i \subseteq R_1{\mathcal F}_{i-1}\) and \(\text{Ker}\, \delta_i\) is a free \(\Lambda^\prime\)-module, \(i \leq d-1\). One takes \({\mathcal M}^\prime := \text{Ker}\, \delta_{d-1}\). In order to verify the conditions (0)--(2) one notices, firstly, that a \(\Lambda^\prime\)-module \(L\) with \(\dim_k L = n < \infty\) is given by two \(n \times n\) matrices \(A_x\) and \(A_y\), that \(N := {\mathcal N}^\prime \otimes_{\Lambda^\prime}L\) is isomorphic to \(n\overline{R}/\overline{R}W\) where \(n\overline{R}\) is the direct sum of \(n\) copies of \(\overline{R}\) and \(W\) is the subspace of \(n\overline{R}_c\) generated by the columns of the matrix \(I_ne_1 + A_xe_2 + A_ye_3\), and that \(M := {\mathcal M}^\prime \otimes_{\Lambda^\prime}L\) is a minimal \(d\)-th syzygy of the graded \(R\)-module \(N\). Then the basic observation is that \(N(-m)\) is isomorphic to the submodule of \(M/\underline{y}M\) generated by its elements of degree \(m\). This, in turn, follows from the following observation\(\, :\) putting \(F_0 := nR\), lift \(W\) to a subspace \(\widetilde W\) of \((F_0)_c\) of the same dimension. Let \(\cdots F_2^\prime \rightarrow F_1^\prime \rightarrow \overline{R}W \rightarrow 0\) be a minimal graded free resolution of \(\overline{R}W \subset \overline{R}\) over \(R\). Then \(N \simeq F_0/(R{\widetilde W} + \underline{y}F_0)\) and using the exact sequence\(\, :\) \[\begin{tikzcd} 0 \rar & \underline{y}F_0 \rar & R\widetilde{W} + \underline{y}F_0 \rar & \overline{R}W \rar & 0 \end{tikzcd}\] one sees that \(N\) admits a graded free resolution \(\cdots F_1 \rightarrow F_0 \rightarrow N \rightarrow 0\) with \(F_i = nK_i \oplus F_i^\prime\) for \(i \geq 1\), where \(K_\bullet\) is the Koszul resolution of \(\overline{R}\) over \(R\). The resolution \(F_\bullet\) turns out to be \textit{minimal} because the twist of any of the rank 1 direct summands of \(F_i^\prime\) is strictly less than the twist of any of the rank 1 direct summands of \(K_{i-1}\) for \(i \geq 1\). As an (immediate) application, the authors show that any hypersurface of degree \(\geq 4\) in \({\mathbb P}^n\), \(n \geq 2\), is ACM wild and the same is true for any complete intersection of type \((d_1, d_2, \ldots , d_l)\) in \({\mathbb P}^n\) if either \(l \geq 3\) or \(l = 2\) and \(d_1 \geq 3\). \textit{M. Casanellas} and \textit{R. Hartshorne} [J. Eur. Math. Soc. (JEMS) 13, No. 3, 709--731 (2011; Zbl 1245.14044)] constructed families of arbitrary large dimensions of non-isomorphic, indecomposable ACM bundles on a cubic surface in \({\mathbb P}^3\). The authors conclude the paper under review by bringing to the reader's attention the open question asking whether such a cubic surface is (algebraically) ACM or not.