Koszul modules (Q2570670)

From the authors' introduction: Consider a noetherian local ring \(R\) with maximal ideal \(\mathcal M\) and residue field \(k\). Given a finitely generated \(R\)-module \(M\) and \(F\) its minimal free resolution, denote \(\text{lin}^R(F)\) the associated graded complex of \(F\) arising from the standard \(\mathcal M\)-adic filtration. This is a complex of free modules over the associated graded ring \(\text{gr}_{\mathcal M}(R)\) of \(R\) and its differentials depend only on the linear part of the corresponding differentials of \(F\); whence the notation. We say that \(M\) is Koszul if \(\text{lin}^R(F)\) is acyclic. The local ring \(R\) is said to be a ``Koszul ring'' if the \(R\)-module \(k\) is Koszul. This is tantamount to requiring that the graded \(k\)-algebra \(\text{gr}_{\mathcal M}(R)\) be Koszul in the classical sense of the word. In studying Koszul modules, it is profitable to consider the larger class of modules \(M\) for which the number \(\text{ld}_{R}(M) = \sup\{i/H_i(\text{lin}^R(F)\not=0\}\), called the linearity defect of \(M\), is finite. These are the modules in whose minimal free resolutions the linear part predominates. Observe that \(\text{ld}_R(M) \leq d\) is finite if and only if \(\Omega^d(M)\), the \(d\)th syzygy of \(M\), is Koszul. A basic issue in this context is that of existence of modules of finite linearity defect. Here one encounters a rather surprising phenomenon: if \(\text{ld}_R(k) <\infty\), then \(R\) is Koszul, at least when \(R\) is a standard graded \(k\)-algebra. Nevertheless, even when \(R\) is Koszul, there can be modules \(M\) with \(\text{ld}_R(M) = \infty\). Indeed, if \(\text{ld}_R(M)\) is finite, then the Poincaré series of \(M\) is a rational function, and its denominator term depends only on the Hilbert series of \(\text{gr}_{\mathcal M}(R)\). However, even over the seemingly innocuous Koszul algebra \(k[x_1,x_2,x_3]/(x_1,x_2,x_3)\bigotimes_k k[y_1,y_2,y_3]/(y_1,y_2,y_3)^2\) there are finitely generated modules with transcendental Poincaré series. Furthermore, Roos has constructed a Koszul algebra \(R\) and a family of \(R\)-modules \(\{M_n \}\), \(n\in \mathbb N \) such that the Poincaré series of each \(M_n\) is rational, but these do not share a common denominator. These examples lead one to consider the problem the rings over which each module is of finite (or bounded) linear defect. We prove that if a Koszul local ring \(R\) is either a complete intersection or Golod, then \(\text{ld}_R(M)\) is finite for every finitely generated \(R\)-module \(M\). Modules of finite linear defect are particularly well behaved with regard to their homological properties. One example of this phenomenon is the rationality of Poincaré series mentioned above. More sophisticated manifestation concern the left module structure of \(\text{Ext}_R(M,k)\) over the \(k\)-subalgebra \(R^!\) of \(\text{Ext}_R(k,k)\) generated by \(\text{Ext}^1_R(k,k)\). For instance, we prove that if \(\text{ld}_R(M)\) is finite, then \(\text{Ext}_R(M,k)\) is finitely generated over \(R^!\) by elements with degrees \(\leq \text{ld}_R(M)\). In particular, when the ring \(R\) is Koszul, \(R^! = \text{Ext}_R(k,k)\); this last result is well known when \(R\) is a standard graded \(k\)-algebra. Over Koszul rings such a result admits a vast generalization, best captured by the formula: \[ \text{ld}_R(M) = \text{reg}_{R_!}\text{Ext} _R(M,k). \] It is this equality, in conjunction with work of Backelin and Roos, that allows us to deduce the finiteness of linearity defect of modules over complete intersections and Golod rings, discussed in the preceding paragraphs. It implies also that when \(\text{ld}_R(M)\) is finite, the \(R_!\)-module \(\text{Ext}_R(M,k)\) and all its higher syzygies are finitely generated. The heart of the paper is the theorem that expresses the homology of the linear part of the resolution \(F\) of \(M\), over a Koszul ring \(R\), in terms of the (co)homology of \(\text{Tor} ^R(k,M)\), when viewed as a module over \(R^!\). While the bulk of the results in this paper is stated and proved for modules over local rings, they are applicable also to graded modules over graded \(k\)-algebras.