Two examples of the impact of partitioning with Chaco and Metis on the convergence of additive Schwarz-preconditioned FGMRES

Algebraic graph partitioning packages that require no geometric or physical information are often used to partition sparse linear systems for solution on parallel computers using Krylov-space iterative methods with domain decomposition-type preconditioning methods. The authors show through a couple of examples that this can have a substantial impact on the convergence of the iterative methods.