CSE Seminar Announcement

A New Solver for the Generalized Eigenvalue Problem at Large Scales Combining Jacobi-Davidson and Multigrid Methods

Wednesday, December 13, 2017
1030 National Center for Supercomputing Applications (NCSA)
12:00pm - 1:00pm
Lunch Provided

Darin Peetz
CSE Fellow
Civil and Environmental Engineering

Abstract

Several methods exist for solving the generalized eigenvalue problem, but each has its own flaws, especially at large scales. We describe a new method for solving the symmetric generalized eigenvalue problem that combines features of the Jacobi-Davidson method and multigrid solvers. The combined method achieves excellent scaling both in terms of problem size and degree of parallelization. We compare the cost and scalability of this method against more established solvers such as Lanczos, LOBPCG, or traditional Jacobi-Davidson. Finally, the method is incorporated in a topology optimization scheme to demonstrate its versatility in handling challenging problems.