Paper PS-ThP27
Plasma Simulations with Adaptive Mesh Refinement and Hybrid Kinetic-Fluid Models

Thursday, October 22, 2015, 6:00 pm, Room Hall 3

This paper describes the development of plasma simulation tool using adaptive kinetic-fluid models for modern heterogeneous computing systems. The new tool will have a) ability to dynamically switch between fluid and kinetic approaches, b) robust mesh generation and adaptive mesh refinement algorithms, c) implicit solvers adapted for massively parallel CPU-GPU systems.

For gas dynamics in mixed rarefied-continuum regimes, our Unified Flow Solver (UFS) combines a tree-based algorithm of Adaptive Mesh Refinement (AMR) with cell-by-cell selection of kinetic (Boltzmann) and fluid (Navier-Stokes) solvers [1]. This Adaptive Mesh and Algorithm Refinement (AMAR) methodology provides the highest level of adaptation to the solution being computed. The kinetic module in UFS can solve kinetic equations by either the Discrete Velocity Method (DVM) [2] or by the statistical particle methods (such as Direct Simulation Monte Carlo (DSMC)) [3]. We have ported the kinetic solvers to Graphics Processing Units (GPU) and demonstrated double digit speedups on single GPU and good multi-GPU scaling [4].

The extension of AMAR to plasma simulations poses extra challenges due to disparity of electron and ion/neutral masses [5]. In this paper, we will show examples of plasma simulations with AMR capabilities using multi-fluid plasma models and describe the implementation of an implicit electrostatic Particle in Cell with Monte Carlo Collisions (PIC-MCC) module. The capabilities of the new PIC-MCC module will be demonstrated for positive column of DC discharges in weakly-collisional regime. We will describe algorithms for porting the PIC-MCC module to GPU and evaluate acceleration that can be achieved. Finally, we will discuss physics-based criteria for the kinetic-fluid decomposition for hybrid plasma simulations.