Some results on asymptotic regularity of ideal sheaves (Q371400)

This paper is placed in the wide topic of the asymptotic regularity (i.e. the regularity of the powers of an ideal or an ideal sheaf). The fundamental notions are the following. \noindent Let \(M\) be a finitely generated graded module over a polynomial ring and let \[ 0 \rightarrow E_{n+1} \rightarrow \cdots \rightarrow E_1 \rightarrow E_0 \rightarrow M \rightarrow 0 \] be a minimal graded free resolution of \(M\). Then \(M\) is said \textit{\(m\)-regular} if \(E_i\) is generated in degrees less than or equal to \(m+i\) and the \textit{regularity} of \(M\), denoted by \((M)\), is the least \(m\) for which this holds. There is a corresponding notion in the geometric context: if \(\mathcal F\) is a coherent sheaf on \(\mathbb P^n\), then \(\mathcal F\) is called \textit{\(m\)-regular} if \[ H^i (\mathbb P^n, {\mathcal F}(m-i)) =0 \] for all \(i > 0\), and we say that the \textit{regularity} of \(\mathcal F\) is the least \(m\) for which these vanishings hold. An effective result (see [\textit{V. Kodiyalam}, Proc. Am. Math. Soc. 128, No. 2, 407--411 (2000; Zbl 0929.13004)] and [\textit{S. D. Cutkosky} et al., Compos. Math. 118, No. 3, 243--261 (1999; Zbl 0974.13015)]) about the asymptotic regularity of a homogeneous ideal \(I\) of a polynomial ring is the following: \[ I^t = dt + e, \quad t \gg0 \] where \(d\) is the so called ``slope'' and much less is known about the constant \(e\). In the geometric setting, if \(\mathcal I\) is an ideal sheaf on \(\mathbb P^n\), then (see [\textit{S. D. Cutkosky} et al., Math. Ann. 321, No. 2, 213--234 (2001; Zbl 1029.14022)]): \[ \lim _{t \rightarrow \infty} \; {{{\mathcal I}^t } \over t} = s \] where \(s\) (the so called \(s\)-invariant of \(\mathcal I\)) could be an irrational number. Therefore, the best one can hope is only to bound the regularity of \({\mathcal I}^t\). But, surprisingly, [\textit{M. Chardin}, Algebra Number Theory 7, No. 1, 1--18 (2013; Zbl 1270.13008)] shows that \( {\mathcal I}^t \) could be linear for large \(t\). In the present paper, the author shows in the main theorem (1.1 or 2.11) that the cited result can be improved in the following way: if \(d\) is the minimal number such that \({\mathcal I}(d)\) is generated by its global sections and \(s=d\), then \[ {\mathcal I}^t = dt + e, \quad t \gg 0. \] A relevant point of this result is a precise description of the ``asymptotic regularity constant'' \(e\). Many consequences are given: first, varieties having linear asymptotic regularity are not so rare (Corollary 1.2). Second, previous results (see [\textit{P. Vermeire}, Compos. Math. 131, No. 2, 161--172 (2002; Zbl 1083.14503)]) about the asymptotic regularity of a curve \(X\) embedded by line bundle of large degree and having ideal sheaf \({\mathcal I}_X\), establish that if \(\deg L \geq 2g+3\) (where \(g\) is the genus of \(X\)) then \({\mathcal I}_X^t\) is \((2t+1)\)-regular, for \(t \geq 1\). The author here considers the case of higher dimensional non singular projective variety \(X\) (where \(\dim(X) \geq 3\)) and proves that, under suitable assumptions on the adjoint line bundle \(L_d\), then \({\mathcal I}_X^t\) is \((2t+2 \dim(X) -2)\)-regular, for \(t \geq 1\) (see Theorem 1.3 or 4.5). Other interesting results are proved when \(X\) is a locally complete intersection. Third, new properties are obtained about curves and surfaces, where a relation between asymptotic regularity and the Gauss map is established. The cohomology techniques and the new notion of biregularity are clearly used and explained. So this paper is a rich and well written study on the subject.