Data-Driven Recovery of Hammett-Perkins Closure from Particle Data

Finding fluid approximations for particle behavior is useful for both simplifying simulation complexity and finding interpretable ways to describe behavior. Data-driven model identification is similarly useful as it can identify fluid equations from particle simulations where otherwise the best fluid approximation may be ambiguous. When moving from particle models to fluid approximations there is also a question of what behavior is significant at the spatial and temporal scales of concern. In this work we examine using one such model identification method, WSINDy, to learn the Hammett-Perkins closure from particle simulations of landau damping. We examine how the simulation parameters and test function hyperparameters affect the resultant identified model as well as modifications to the WSINDy algorithm to account for terms that do not fit into the weak form.