AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
Properties and challenges of machine learning closures for ten-moment fluid models in the case of Landau damping
Two-fluid ten-moment plasma models need a heat flux closure, and this study tests machine learning closures in place of the usual Hammett-Perkins-type analytical closures. On Landau damping, the ML closures reproduced the damping frequency and nonlinear phase better than Hammett-Perkins closures within and between the training parameter ranges. Performance degraded when extrapolating beyond the training data, and offline metrics did not fully predict online performance, so the authors propose the temporal evolution of entropy as an alternative evaluation metric.