Development of GPU accelerated kinetic modeling and simulation code for hypersonic non-equilibrium flow

Kinetic theory studies the physical system using a statistical approach, describing the collective microscopical motion of atoms, molecules, ions, electrons, and their interactions. Kinetic modeling represents a more accurate method compared to the continuum approach, which is usually governed by the Boltzmann equation. Particle-based simulation methods are being widely used in kinetic modeling to solve this integro-differential equation accurately and efficiently. These methods can largely benefit from the advancement in GPU accelerated high performance computing, due to their unique solution procedures. In this presentation, we discuss our work developing a particle simulation code, Comet, working on either CPU or GPU accelerators. Comet is a direct simulation Monte Carlo (DSMC) code for modeling and simulating hypersonic non-equilibrium flow, rarefied gas flows, and micro-scale and nano-scale flows. Central to Comet is the unstructured mesh infrastructure Omega_h, and mesh-centric particle data structure PUMIPic. Comet simulation results of hypersonic rarefied gas flows will be presented and compared with existing experimental data. The parallel performance of the code will also be presented, including both strong and weak scaling.