Development of a parallel multigrid field solver for large-scale particle-in-cell applications

Abstract

Particle-in-cell (PIC) methods are widely used to simulate the motion of charged particles but are limited in their applicability by the increasing computational costs related to systems that are scaled up both in size and particle number/density. Parallelizing the simulation is, therefore, a necessity to expand the range of application. Since the readily available HPC clusters gradually increase in size, increasingly larger systems with more and more particles can then be simulated by utilizing correspondingly more processors for computation. This, however, is only practical in the absence of any significant bottleneck restricting the parallel (weak) scaling of the simulation. I.e., all underlying algorithms of the particle-in-cell method need to be optimized for parallel performance.For this very reason, a new field solver, responsible for calculating the electromagnetic fields in the simulated plasma, needed to be developed and implemented in "PlasmaPIC", a fully three-dimensional parallel plasma simulation code developed in Giessen.This new field solver is based on the geometric multigrid method and is tailored specifically for the requirements of "PlasmaPIC", offering almost ideal weak scaling over a wide range of system sizes and processor numbers.