Direct Solution of Large, Sparse Systems of Equations on a Distributed Memory Computer

摘要

A distributed LU factorization is implemented for sparse, irregular systems of equations. A nested dissection of the system graph is done as a serial preprocessing step resulting in a border-block diagonal system. The system is mapped to a distributed memory parallel computer, where it is factored and solved. In this paper, we briefly discuss the algorithm and present timing results using matrices generated from the tetrahedral finite-element discretization of the three-dimensional vector equation for electromagnetic scattering problems.