Algorithmic computation of local cohomology modules and the local cohomological dimension of algebraic varieties (Q1295794)

Let \(K\) be a (computable) field of characteristic zero, \(R:=K[X_1\ldots X_n]\) the ring of polynomials in \(n\) variables and let \(A_n:=K<X_1,\partial_1,\dots,X_n,\partial_n>\) be the Weyl algebra over \(K\) in \(n\) variables. Let \(I=(f_1,\dots,f_r)\) be an ideal in \(R\) given by a finite set of generators, say \(f_1\ldots f_r\), and let \(m:=(X_1,\dots,X_n)\). The main outcome of the paper is an algorithm which computes for given non-negative integers \(i\), \(j\), \(k\) the abstract structure of the local cohomology modules \(H_{I}^k(R)\), \(H_m^i\left(H_{I}^j(R)\right)\), the cohomological dimension of \(I\) in \(R\) and the \(K\)--vector space dimension of the socle of \(H_m^i(H_I^j(R))\). The cohomology modules \(H_I^k(R)\) and \(H_m^i(H_I^j(R))\) are finitely generated over \(A_n\) and described by generators and relations (section 5). The procedure may be extended to any \textit{regular} affine ring \(R\) given by a polynomial ideal but not to more general situations. The algorithm of the paper are rewriting procedures, typically Gröbner basis algorithms in suitable Weyl algebras (section 3 and 4). No complexity bounds are given and it is not clear now that they could be read-off from the paper. It seems that the worst case complexity of the main algorithm is at least doubly exponential in \(n\). Consequently the author could only check the practical performance of his algorithm by examples of very moderate size (section 6). The only previously known algorithmic result in the direction of this paper was the determination of the cohomological dimension of \(I\) in \(R\) if \(K\) is of \textit{positive} characteristic. The paper starts from the description of the local cohomology groups of \(I\) in \(R\) by means of the Čech complex associated to the open set of the affine \(n\)-space defined by the polynomials \(f_1,\dots,f_r\). Unfortunately, these local cohomology groups are seldom finite \(R\)-modules and this makes their description in the language of Gröbner basis theory particularly cumbersome. In order to overcome this difficulty the author considers the localizations of \(R\) by certain products of the generators \(f_1,\dots,f_r\) as holonomic \(A_n\)-modules. This implies that all modules computed by the algorithms are finite \(A_n\)-modules which can be described by generators and relations. This is done by a skillful extension of Oaku's work on Gröbner bases in \(D\)-modules.