Physics-Informed Koopman Autoencoders for Dimensionality Reduction and Forecasting of Nonlinear Kinetic Plasma-Wave Interactions

Accurate simulations of nonlinear kinetic plasma-wave interactions are essential for the design and analysis of high-power microwave devices, fusion energy devices, terahertz and directed energy devices, as well as for studying a host of ionosphere and magnetosphere phenomena and their impact on space assets. However, large-scale kinetic plasma simulations that rely on popular electromagnetic particle-in-cell (EMPIC) algorithms face a significant computational burden due to the need to individually simulate thousands to millions of charged particles (ions, electrons). To address this, reduced order models such as proper orthogonal decomposition (POD) [1] and dynamic mode decomposition (DMD) have proved especially useful [2].