Past Event: Oden Institute Seminar

Exascale particle-mesh schemes for kinetic plasma simulations

Sriramkrishnan Muralikrishnan, Forschungszentrum Jülich GmbH

Abstract

Particle-in-cell (PIC) scheme first introduced in the context of fluid dynamics in 1957 has been the method of choice for the simulation of charged particles dynamics in kinetic plasma simulations. It is conceptually intuitive, highly parallelizable, and robust for a wide range of physical scenarios. In the first part of the talk I will introduce the basics of the PIC scheme, its advantages and disadvantages. I will also present a open-source, performance portable C++ library IPPL being developed by us for large scale PIC simulations. The primary use cases of IPPL are two-fold. On the one hand, it is used to develop novel numerical algorithms in a performance portable way targeting exascale architectures. The second important use case of IPPL is the production level particle accelerator library OPAL which uses IPPL for all the numerical and HPC components. OPAL has more than a hundred users in various parts of the world and is widely used for the design and analysis of particle accelerators. 

One of the primary disadvantages of PIC schemes is the lack of energy conservation which leads to numerical instabilities in kinetic plasma simulations. In the second part of the talk, I will introduce the so-called particle-in-Fourier (PIF) scheme which is a structure preserving scheme and hence possesses excellent conservation and stability properties. I will present a novel space-time parallelization strategy for PIF schemes with extreme-scale runs involving millions of Fourier modes and billions of particles in 3D-3V scaling up to 6144 A100 NVIDIA GPUs (full system) on Perlmutter supercomputer at NERSC, USA. Finally, I will conclude the presentation with some of the grand challenging problems in particle accelerator simulations and our strategies for tackling them with the novel particle-mesh schemes. 

Biography

Sriramkrishnan Muralikrishnan is currently a research staff in the Department of Mathematics and Education at the Jülich Supercomputing Centre, Forschungszentrum Jülich GmbH in Germany. Prior to this he was a Marie-Curie postdoctoral fellow at the Paul Scherrer Institut in Switzerland for three years.He finished his PhD from the Department of Aerospace Engineering and Engineering Mechanics at the UT Austin in 2019.