Finding the radical of matrix algebras using Fitting decompositions (Q1295786)

The author addresses the computational problem of finding the Jacobson radical \(\text{Rad}(A)\) of a subalgebra \(A\) of the full matrix algebra \(M_n(K)\) over the field \(K\). It is assumed that \(A\) contains the identity matrix. The author assumes that the input is a finite set of matrices which generate \(A\) as an algebra and the output is a set of matrices which generate \(\text{Rad}(A)\) as an ideal. This task has several applications from modular representation theory of groups to the calculation of the structure of Lie algebras. Until now, all the known methods for computing the radical are based on solving systems of linear or semilinear equations. The coefficients are the traces and other invariants in positive characteristic of the products \(b_ib_j\) where \(b_1,b_2,\ldots,b_s\) is a basis of \(A\). Unfortunately, it is not known how to determine the coefficients in a way more efficient than computing the diagonal elements of the product \(b_ib_j\) of \(O(s^2)\) pairs \(b_i,b_j\). Since \(s\) can be as large as \(n^2\), the existing algorithms require \(\Omega(n^6)\) operations. The approach of this paper is different. The key idea is using the Fitting decomposition with respect to the adjoint actions of appropriate subalgebras to reduce the task to computing the radical of a subalgebra which is nilpotent as a Lie algebra. Factoring by the commutator ideal leads to a reduction to the commutative case. The author also describes a probabilistic version for computing elements which generate the radical as an ideal.