This talk will introduce a new implementation of the frontal method for direct solution of large sparse linear systems of equations derived from finite element models. We start by motivating the use of the frontal method with finite element modeling including an informal overview of the mechanics of the frontal method. Along the way, we indicate design opportunities for minimizing floating point arithmetic operations, improving memory access patterns and reducing memory storage requirements. We conclude by comparing the performance of our new solver with the performance of two contemporary direct equation-based solvers.
Seminars are open to alumni, friends of the Department, and the general public.