guest ::
| Home > Online First > A Massive Space-Time Parallel Particle-In-Fourier Framework for Kinetic Plasma Simulations |
| Home > Online First > A Massive Space-Time Parallel Particle-In-Fourier Framework for Kinetic Plasma Simulations |
| Conference Presentation (Invited) | FZJ-2025-05370 |
;
2025
Abstract: Kinetic plasma simulations play a critical role in applications of societal relevance suchas nuclear fusion and building the next-generation of compact particle accelerators.They are also widely used in studying astrophysical phenomena and industrial plasmaprocesses.Particle-In-Fourier (PIF) schemes are attractive for long-time integration of kineticplasma simulations as they conserve charge, momentum and energy, exhibit avariational structure, do not have aliasing and have excellent stability properties.However, they are typically more expensive than the commonly used Particle-In-Cell(PIC) schemes due to the requirement of non-uniform discrete Fourier transforms (DFT)or fast Fourier transforms (FFT). In this talk, we present a Parareal-based parallel-in-time integration method for PIF schemes by employing a PIF scheme of coarsertolerance for nonuniform FFTs or the standard PIC scheme as coarse propagatorstowards the goal of performing long-time integration simulations. We show an erroranalysis of the algorithm as well as numerical validation of the results with Landaudamping, two-stream instability, and Penning trap test cases in 3D-3V. Finally, we alsopresent massively parallel full-system scaling studies from space-time parallelimplementation of the PIF schemes in the open-source, performance-portable library"Independent Parallel Particle Layer'' (IPPL) on JUWELS and ALPS supercomputerswith A100 and GH200 GPUs. The space-time parallelization provides up to $4-6$ timesspeedup compared to spatial parallelization alone and achieves a push rate of morethan 1 billion particles per second for the benchmark plasma mini-apps considered.
|
The record appears in these collections: |
PDF