The International Conference for High Performance Computing, Networking, Storage and Analysis
Parallel High-Order Geometric Multigrid Methods on Adaptive Meshes for Highly Heterogeneous Nonlinear Stokes Flow Simulations of Earth's Mantle.
Authors: Johann Rudi (University of Texas at Austin), Hari Sundar (University of Utah), Tobin Isaac (University of Texas at Austin), Georg Stadler (University of Texas at Austin), Michael Gurnis (California Institute of Technology), Omar Ghattas (University of Texas at Austin)
Best Poster Finalist
Abstract: The simulation of Earth's mantle flow with associated plate motion at global
scale is challenging due to (1) the severe nonlinear rheology, (2) the
viscosity variations, and (3) the widely-varying spatial scales inherent in the
problem. We discretize the governing nonlinear incompressible Stokes equations
using high-order finite elements on highly adapted meshes, which allow us to
resolve plate boundaries down to a few hundred meters. Crucial components in
our scalable solver are an efficient, collective communication-free, parallel
implementation of geometric multigrid for high-order elements, the use of a
Schur complement preconditioner that is robust with respect to extreme
viscosity variations, and the use of an inexact Newton solver combined with
grid-continuation methods. We present results based on real Earth data that
are likely the most realistic global scale mantle flow simulations conducted to
date.
