Exploring Physics Informed Deep Learning for Resolving Subgrid-Scale Effects in Binary Neutron Star Simulations

We explore the promise of physics informed deep learning to capture the physics of subgrid-scale magnetohydrodynamic turbulence in simulations of the magnetized Kelvin-Helmholtz instability (KHI). The KHI creates general relativistic magnetohydrodynamic turbulence that amplifies the magnetic field at a smaller scale than binary neutron star simulations are capable of resolving. We develop physics informed artificial neural network models and evaluate their ability to resolve similar subgrid effects and other relevant phenomena. Specifically, we use models to reproduce the results of various simulated test problems involving a variety of initial conditions, shocks, and coupled fields. Finally, we discuss the feasibility of using such methods to capturing subgrid-scale general relativistic magnetohydrodynamic turbulence in numerical relativity simulations of binary neutron star mergers.