Department of Computational and Data Sciences
SPEAKER : Dr. Muralikrishnan Sriramkrishnan, ETH Zurich
TITLE : “Novel grid and particle-based methods with applications to plasma physics and fluid dynamics”
Date & Time : April 25, 2022, 11:00 AM.
Venue : Online
Grid and particle-based methods are the common numerical algorithms in the simulation of many physical phenomena including but not limited to fluid dynamics and plasma physics. As we are moving towards the exascale era, robust, scalable and efficient numerical algorithms are the need of the hour to carry out these large scale physical simulations in extreme scale architectures.
In this talk, I will first present a high-order finite element method namely hybridized discontinuous Galerkin (HDG) together with scalable solvers for fluid dynamics and magnetohydrodynamics (MHD) simulations. One of the attractive features of HDG methods is that they have a compact stencil and a lot fewer coupled unknowns at high orders in the context of steady state problems or time dependent problems with implicit time stepping. I will present a novel block preconditioning strategy for HDG schemes and show that for a challenging island coalescence problem in MHD which involves magnetic reconnection the solver is robust and scalable to thousands of cores.
In the second part of my talk, I will present a novel sparse grid-based adaptive noise reduction strategy that can effectively reduce the common problem of high numerical noise associated with the particle-in-cell (PIC) schemes, which are a popular particle-mesh method for kinetic plasma simulations. With certain benchmark problems in plasma physics I show that our approach can provide significant speedup and memory reduction to PIC simulations for achieving comparable accuracy in the charge density.
Finally, I will introduce an open source performance portable library for grid and particle computations targeting exascale architectures. I will show the scaling and performance by means of plasma physics mini-apps on several massively parallel pre-exascale architectures such as Piz Daint, Cori, Summit and Perlmutter up to thousands of CPU cores and GPUs for a problem with more than a billion particles. I will conclude the talk with challenges, research opportunities and future directions of research.
Host Faculty: Prof. Konduri Aditya