Comprehensive magnetized collisionless shock observations with deep learning aided diagnostic analysis

Collisionless shocks are of great interest to the astrophysics community due to their prevalence throughout astrophysics phenomena, including supernova remnants and planetary bow shocks, and candidacy for particle accelerators of cosmic rays. Despite their ubiquity, key formation and sustentation processes of collisionless shocks are still not fully understood. Laboratory experiments have become a valuable tool to further study these systems in controllable and repeatable conditions. Furthermore, the improvements to and growing accessibility of high repetition rate experiment capabilities potentially offer unparalleled insight into system evolution, but the dynamicity and sheer size of these datasets could render conventional analysis challenging. In this work we demonstrate how deep learning models could serve as potential candidates to help meet the needs of future experiments.

We present results from experimental campaigns at the OMEGA laser facility to study quasi-perpendicular (Tubman et al., in preparation) and quasi-parallel shock formation and highlight how optical Thomson scattering (OTS) is used to measure key plasma parameters to better understand and detect shock formation. Results are compared with particle-in-cell simulations. We then demonstrate how deep learning models can be trained to accelerate and aid OTS analysis by predicting plasma conditions directly from OTS spectra (Pokornik et al., Phys. Plasmas 31, 7 2024) collected from both experimental campaigns and theoretical spectra forward modeled from simulations. We discuss the advantages and limitations of the models and present preliminary work on predicting fully arbitrary particle velocity distributions.