Modeling Particle Acceleration in Magnetic Reconnection with Machine Learning-Driven Sub-grid Physics

Magnetic reconnection is a ubiquitous mechanism for particle acceleration in astrophysical environments, from solar flares to the turbulent accretion flows around black holes. While particle-in-cell (PIC) simulations have provided detailed insights into the microphysics of reconnection and associated acceleration processes, the extreme disparity between kinetic and global scales makes it infeasible to resolve reconnection-driven acceleration in global simulations. In this work, we leverage modern machine learning algorithms applied to extensive suites of PIC simulations to discover dynamical models for the evolution of particle energy distributions during reconnection events. These data-driven models are designed as sub-grid prescriptions for global simulations, enabling the study of nonthermal particle populations in complex astrophysical systems.