Subadditivity of syzygies of Koszul algebras (Q2255287): Difference between revisions

Let \(R\) be a finitely generated graded algebra over the field \(K\). In other words, \(R\cong S/J\) where \(S\) is a standard graded polynomial ring over \(K\) in \(n\) variables and \(J\) an ideal containing no linear forms. Let \((\mathbb{F},\partial )\) be a minimal graded free \(S\)-resolution of \(R\). Each free module \(\mathbb{F}_i\) in the resolution is of the form \(\mathbb{F}_i=\bigoplus_jS(-j)^{\beta_{ij}}\). We set \[ t_i^S(R)=\max \{j\in \mathbb Z: \beta_{ij}\neq 0 \}=\max \{ j\in \mathbb Z: \text{Tor}_i^S(K, R)_j\neq 0 \}. \] Then \(t_i^S(R)\) is the maximal degree of a minimal \(i\)th syzygy of \(R\). We say \(\mathbb{F}\) satisfies the subadditivity condition if \[ t_{a+b}^S(R)\leq t_a^S(R)+t_b^S(R). \] This concept has been studied by many authors. As a case of interest for which the subadditivity condition holds is when \(\dim R\leq 1\) and \(a+b=n\). In this case we have, \(t_n^S(R)\leq t_a^S(R)+t_{n-a}^S(R)\), see Corollary 4.1 in [\textit{D. Eisenbud} et al., Am. J. Math. 128, No. 3, 573--605 (2006; Zbl 1105.13017)]. There is also a bound on the degree of the final syzygies of \(R\). In fact, \(t_p^S(R)\leq \max\{t_i^S(R)+t_{p-i}^S(R): i=1, \dots, p-1 \}\) where \(p=\text{pd} R\), see Theorem 4.4 in [\textit{J. McCullough}, Math. Res. Lett. 19, No. 3, 555--565 (2012; Zbl 1321.13005)]. McCullough's inequality is improved as follows: \(t_p^S(R)\leq t_{p-1}^S(R)+t_{1}^S(R)\) and for monomial ideals one even has \(t_a^S(S/J)\leq t_{a-1}^S(S/J)+t_{1}^S(S/J)\) where \(J\) is a monomial ideal and \(a\geq 1\), see Corollary 3 in [\textit{J. Herzog} and \textit{H. Srinivasan}, ``A note on the subadditivity problem for maximal shifts in free resolutions'', \url{arXiv:1303.6214}]. The paper under review deals with the estimates for the degrees of minimal syzygies of \(R\). The Castelnuovo-Mumford regularity \(\text{reg}^R(N)\) is defined to be \(\sup\{\text{reg}_n^R(N): n\in \mathbb Z \}\) where \(\text{reg}_n^R(N)=\max_{i\leq n}\{ t_i^R(N)-i\}.\) We say \(R\) is Koszul if the minimal free resolution \(K\) over \(R\) is linear. Thus, if \(R\) is Koszul, then \(\text{reg}^R(K)=0.\) For Koszul algebras, \textit{J. Backelin} [``Relations between rates of growth of homologies'', Research Reports in Mathematics 25, Department of Mathematics, Stockholm University (1988)] and \textit{G. R. Kempf} [J. Algebra 134, No. 1, 222--224 (1990; Zbl 0726.13014)] showed \(t_i^S(R)\leq 2i\) for \(1\leq i \leq \text{pd}_S(R)\). The authors here extend and sharpen these results in several directions. Let \(M\) be an \(R\)-module. Suppose \(a, b\) be integers such that \(a+b\leq \text{pd}_SR\) and \({a+b\choose b}\) is invertible in \(K\). It is shown \[ t_{a+b}^S(M)\leq \max \begin{cases} t^S_a(R)+t^S_b(M), \\ \text{reg}^S_{a-1}(R)+\text{reg}^R_{a+b}(M)+a+b,\\ \text{reg}^S_{a-1}(R)+\text{reg}^S_{b-1}(M)+\text{reg}^R_{a+b+1}(K)+a+b+1. \end{cases} \] In particular, if \(\text{reg}_{a+b+1}^R(K)=\text{reg}_{a+b}^R(M)=0\), then \[ t_{a+b}^S(M)\leq \max \begin{cases} t^S_a(R)+t^S_b(M), \\ \text{reg}^S_{a-1}(R)+\text{reg}^S_{b-1}(M)+a+b+1. \end{cases} \] It is also shown that if \(\text{reg}^R_{n+1}(K)=0\) for some \(n\), then \(t_i^S(R)\leq 2i\) for all \(0\leq i \leq n\). When \(R\) is Koszul, these result holds for all \(i\). We set \(t_i(R)=t_i^S(R)\) and \(m(R)=\min \{i\in \mathbb Z: t_i(R)\geq t_{i+1}(R) \}\). One has \(m(R)\leq \text{pd}_S(R).\) Linearity at the start of the resolution of \(K\) over \(R\) gives us the subadditivity of the degree of the syzygies of \(R\) over \(S\). The authors show the following: Let \(a, b\) be integers such that \({a+b\choose b}\) is invertible in \(K\), \(\text{reg}^R_{a+b+1}(K)=0\) and \(\max\{a, b \}\leq m(R)\). Then \[ t_{a+b}(R)\leq \max \begin{cases} t_a(R)+t_b(R), \\ t_{a-1}(R)+t_{b-1}(R)+3. \end{cases} \] In particular, \(t_{a+1}(R)\leq t_a(R)+2\) when \(b=1\) and \(t_{a+b}(R)\leq t_a(R)+t_b(R)+1\) for \(b\geq 2.\) \textit{M. Green} and \textit{R. Lazarsfeld} [Invent. Math. 83, 73--90 (1986; Zbl 0594.14010)] say that \(R\) has property \(N_q\) for a positive integer \(q\) if \(t_i(R)=i+1\) holds for \(i=1, \dots, q\). This is equivalent to say that \(\text{reg}^S_q(R)=1\). The authors prove the following: If \(\text{reg}_{n+1}^R(K)=0\) for some \(n\geq 2\), and \(R\) satisfies \(N_q\) for \(q\geq 2\), then \(t_i(R)\leq 2i-1\) for \(i=2, \dots, n\). In particular, a sharpening of the result of Backelin and Kempf is obtained. Recently, we have the following new result of subadditivity of algebras. We set \(T_a^S(R)=\min \{j\in \mathbb Z: \beta_{aj}\neq 0\}\) to be the minimal shifts in the minimal \(S\)-resolution of graded algebra \(R\) at degree \(a\). It is shown \[ T_a^S(R)\leq T_1^S(R)+t_{a-1}^S(R) \quad \text{for all } a. \] As a consequence, we have subadditivity \[ t_{a+b}^S(R)\leq t_{a}^S(R)+t_{b}^S(R) \] for Gorenstein algebra \(R=S/I\) where \(a+b\geq s-1\) with \(s=\text{height} I=\text{pd} R,\) see [\textit{S. El Khoury} and \textit{H. Srinivasan}, ``A note on the subadditivity of syzygies'', \url{arXiv:1602.02116}].