An adaptive mesh, GPU-accelerated, and error minimized special relativistic hydrodynamics code

TL;DR

This paper introduces a new GPU-accelerated special relativistic hydrodynamics code with an improved algorithm for variable conversion, capable of handling extreme Lorentz factors and integrated into an adaptive mesh framework for astrophysical simulations.

Contribution

A novel SRHD solver with a realistic equation of state and high Lorentz factor handling, integrated into GAMER for adaptive mesh refinement and GPU acceleration.

Findings

Handles Lorentz factors up to 10^6

Achieves 7×10^7 cell updates/sec on a Tesla P100 GPU

Scales efficiently to 2048 GPUs

Abstract

We present a new special relativistic hydrodynamics (SRHD) code capable of handling coexisting ultra-relativistically hot and non-relativistically cold gases. We achieve this by designing a new algorithm for conversion between primitive and conserved variables in the SRHD solver, which incorporates a realistic ideal-gas equation of state covering both the relativistic and non-relativistic regimes. The code can handle problems involving a Lorentz factor as high as $10^6$ and optimally avoid the catastrophic cancellation. In addition, we have integrated this new SRHD solver into the code GAMER (https://github.com/gamer-project/gamer) to support adaptive mesh refinement and hybrid OpenMP/MPI/GPU parallelization. It achieves a peak performance of $7\times 10^{7}$ cell updates per second on a single Tesla P100 GPU and scales well to 2048 GPUs. We apply this code to two interesting astrophysical applications: (a) an asymmetric explosion source on the relativistic blast wave and (b) the flow acceleration and limb-brightening of relativistic jets.