Title: Efficient Setup Phase Algorithms for Parallel Algebraic Multigrid

Speaker: David Alber, University of Illinois, Urbana-Champaign, Postdoctoral Candidate

Date/Time: Monday, February 26, 2007, 9:00 am – 10:00 am

Location: CSRI Building, Room 279 (Sandia NM)

Brief Abstract: Linear solvers play an important role in scientific computing by solving linear equations arising from the discretization of partial differential equations. As we approach the petaflop computing era, scalable parallel linear solvers continue to gain importance in many scientific computing applications. Algebraic multigrid (AMG) is a linear solver that exhibits attractive scalability features in parallel. AMG relies on setup phase algorithms, such as coarse grid selection, to construct a hierarchy of linear systems.

This talk begins with an introduction on multigrid and coarse grid selection. New techniques leading to improvements in coarsening performance are discussed, and experimental results are presented to illustrate the performance of the new algorithms.