Resistive and multi-fluid RMHD on graphics processing units

TL;DR

This paper demonstrates that CUDA-accelerated finite volume IMEX schemes for resistive and multi-fluid RMHD models can significantly speed up complex simulations, enabling more efficient neutron star merger studies.

Contribution

It introduces the first GPU implementation of finite volume IMEX schemes for resistive and two-fluid RMHD models, achieving substantial speed-ups.

Findings

Over 21× parallel speed-up with double precision

Effective optimization strategies for resistive RMHD schemes

Potential impact on next-generation neutron star merger simulations

Abstract

In this work we present a proof of concept of CUDA-capable, resistive, multi-fluid models of relativistic magnetohydrodynamics (RMHD). Resistive and multi-fluid codes for simulating models of RMHD suffer from stiff source terms, so it is common to implement a set of semi-implicit time integrators to maintain numerical stability. We show, for the first time, that finite volume IMEX schemes for resistive and two-fluid models of RMHD can be accelerated by execution on graphics processing units, significantly reducing the demand set by these kinds of problems. We report parallel speed-ups of over 21$\times$ using double precision floating-point accuracy, and highlight the optimisation strategies required for these schemes, and how they differ from ideal RMHD models. The impact of these results is discussed in the context of the next-generation simulations of neutron star mergers.