Randomized Asynchronous Iterative Methods to Enhance Multigrid Algorithms for Elliptic Systems in Viscous Flow

Ghafirlia Istafa, Max Heldman, Johann Rudi

Abstract

Multigrid algorithms are widely used in computational science. In viscous fluid models using Stokes equations, multigrid methods are used to precondition elliptic operators. The efficiency of multigrid algorithms in parallel can be hindered by communication overheads. We will present a review of multigrid techniques in the context of Stokes systems. Key performance indicators will include convergence rates. We will further investigate the application of randomized asynchronous iterative methods, used within multigrid, to improve the performance of multigrid algorithm in parallel by aiming to reduce communication time.