Computing representation matrices for the action of Frobenius on cohomology groups (Q820979)

Let \(K\) be a perfect field of characteristic \(p>0\). For a positive integer \(r\), let \({\mathbb{P}}^r=\mathrm{Proj}(S)\) denote the projective \(r\)-space, where \(S=K[x_0,\cdots,x_r]\). For a projective scheme \(X\subset{\mathbb{P}}^r\), let \(H^q(X,{\mathcal{O}}_X)\) be the \(q\)-th cohomology group for any \(q\in{\mathbb{Z}}\). Let \(F\) be the Frobenius map and let \(F^*\) be the induced Frobenius map on \(H^q(X,{\mathcal{O}}_X)\). A goal of this paper is to present a general algorithm for computing \(F^*\). Toward this goal, the first step is to compute an explicit basis of \(H^q(X,{\mathcal{O}}_X)\), and secondly to obtain the represention matrix for \(F^*\) with respect to the bases computed in the first step. The paper presents two explicit algorithms. The first algorithm works for arbitrary projective schemes, and is based on the following three techniques: (i) decompose \(F^*\) into two computable maps, (ii) compute a matrix corresponding to one of these two maps, whose entries are homogeneous polynomials in \(S\), and (iii) multiplying each basis element by the matrix computed in (ii), express each image by \(F^*\) as a \(K\)-linear combination of the basis. The second algorithm is a special case of the first, applied, to complete intersections. Algorithm 1: Fix the dimension \(r\) of \({\mathbb{P}}^r\). Given \(1\leq q\leq r-1\), and a projective scheme \(X\subset{\mathbb{P}}^r=\mathrm{Proj}(S)\) with the defining homogeneous polynomials \(f_1,\cdots f_m\in S=K[x_0,\cdots, x_r]\), there exists an algorithm to compute the representation matrix for \(F^*\). The algorithm terminates in \(\tilde{0}(D^4+D^2\cdot P(p)+D^3p^r)\) arithmetic operations over \(K\). Here \(D\) is the maximal value of the dimensions of the cohomology groups over \({\mathbb{P}}^r\) appearing in the computation of \(H^q(X,{\mathcal{O}}_X)\). For each \(e\in{\mathbb{Z}}_{\geq 1}\), \(P(e)\) denotes the number of arithmetic operations over \(K\) for computing the \(e\)-th power. Note that Algorithm 1 requires to compute a (minimal) free resolution, however, the cost of this computation is not included in the algorithm. Algorithm 2: Retaining the notation of Algorithm 1, let \(X=V(f_1,\cdots, f_m)\) be a complete intersection embedded in \({\mathbb{P}}^r\) with an \(S\)-regular sequence \((f_1,\cdots, f_m)\in S^m\). Assume \(d_{j_1\cdots j_{m-1}}:=\sum_{k=1}^{m-1} \deg(f_k)\leq r\) for all \(1\leq j_1\leq\cdots\leq j_{m-1}\leq m\) and \(\mathrm{gcd}(f_i,f_j)=1\) in \(S\) for \(i\neq j\). Then there exists an algorithm to compute the representation matrix for \(F^*\) with \(q=\dim(X)=r-m\). The complexity of the algorithm is bounded by the cost of computing \((f_1\cdots f_m)^{p-1}\).