Implementation of Implicit Methods in PicassoMPM

Lance Bullerwell, Austin Isner, Kwitae Chong, Sam Reeve, Stuart Slattery, Duan Z. Zhang

Keywords: Implicit methods, High Performance Computing, Performance Portability

TL;DR: An implicit solver for momentum was implemented in a performance portable material point method code

Abstract: PicassoMPM is a material point method solver that has been developed with a focus on performance portability, the ability to run the code on a variety of different hardware without seeing significant losses in code performance. PicassoMPM is built upon the Kokkos C++ performance portability ecosystem and the Cabana performance portability library for particle simulations, with simulation performance on the exascale supercomputer Frontier being a goal of the software. Previously, PicassoMPM has been able to solve the material momentum and enthalpy equations in a fully explicit methodology. In this work, an implicit methodology was implemented to solve the momentum equation to overcome the strict sonic Courant limits of the fully explicit method. For the implicit momentum solution, a Newton solve was implemented where the momentum update on the grid was performed using the semi-structured solver interface provided by HYPRE. The grid forces are then updated via updating the particle stress. The implicit solver implemented will be compared to analytic results for a variety of test cases, and the time step size and performance on Frontier will be compared to the existing explicit methods in Picasso MPM

Submission Number: 15