High-Performance Calculations of Collisionless Plasma for Space-Weather Applications

Performing fundamental scientific research of complex physical phenomena, such as the kinetic dynamics of fluids and plasmas, can require a tremendous amount of computing power to incorporate all relevant length and time scales. Increasingly, this computing power is becoming more available, but leveraging it can require specialized models, algorithms, and frameworks. This is especially true for modern supercomputer architectures, which have placed a heavy focus on GPUs to provide large amounts of floating point operations using relatively low power. In this talk, we present a new tool for simulating the kinetic equations for collisionless plasma and fluid flows, where the velocity space in the Vlasov/Boltzmann equation is discretized using a spectral expansion of asymmetrically weighted Hermite functions. This leads to a high density of degrees of freedom, which leads to interesting consequences for choosing algorithms and implementation strategies, especially when targeted for execution on GPUs. To that end, we focus on the underlying discretization and implementation of this spectrally-expanded Vlasov equation. In particular, we present the use of the open-source FleCSI performance-portable multiphysics infrastructure and Kokkos kernel performance portability library for creating an adaptable and fast physics solver. Results from simulations relevant to space physics will be used to demonstrate capabilities of the tool.