Upcoming Event: PhD Dissertation Defense

GPU Accelerated Particle-in-Cell Methods for Low-Temperature Plasmas

James Almgren-Bell, CSEM Student, Oden Institute

Abstract

We present a GPU-accelerated hybrid PIC-MCC (particle-in-cell with Monte Carlo collisions) solver for low-temperature plasmas. Our work focuses on the design and implementation of numerical algorithms for solving the collisional electron Boltzmann transport equation.

Our first contribution is a 0D3V (0-D physical space, 3-D velocity space) direct simulation Monte Carlo (DSMC) solver for the electron Boltzmann equation. This solver supports a variety of electron collision types including electron-electron Coulombic interactions and makes no approximations of the collision step. Our solver removes the commonly used two-term approximation of the Boltzmann equation, allowing us to present high-fidelity simulation results for higher order terms in spherical harmonic representations of the electron distribution function.

Our second contribution focuses on design, implementation, and performance analysis of fast GPU algorithms for a 1D3V (1-D physical space, 3-D velocity space) hybrid electron Boltzmann solver. Our Python-driven solver leverages compiled GPU kernels and demonstrates an approximate 5x speedup over similar C++ codes. Using MPI, we present what to our knowledge is the first multi-GPU solver for low-temperature plasmas and the first to report details for recombination and Coulombic collisions.

Our final contribution is Energy-Weighted Particle-in-Cell (EW-PIC), a novel methodology for leveraging variable weight macroparticles to improve the simulation of low-temperature plasmas. EW-PIC uses custom, physics-informed binning discretizations of energy space to increase sampling resolution of typically under-resolved regions of the distribution function. For each grid cell, particle number control methods are used to maintain the desired particle counts within each energy bin.

We integrate EW-PIC with our hybrid electron Boltzmann solver and present results for a series of argon glow discharge simulations with different pressures and chemistry models. To our knowledge, such high-fidelity simulations are the first of their kind for argon glow discharge devices.

Biography

Jimmy is a PhD candidate working under the supervision of Dr. George Biros at the Oden Institute. His research focuses on high-performance computing for kinetic solvers, with particular focus on the design of fast numerical methods and the implementation of optimized GPU kernels. He received his BA in Applied Mathematics with a focus on Scientific Computing from Harvard University.