Physics-Informed Neural Operator Surrogate for 2D Magnetohydrodynamic Reconnection

This paper presents a physics-informed neural operator surrogate for 2D resistive magnetohydrodynamic (MHD) reconnection, addressing the computational challenges posed by scale separation at large Lundquist numbers. The neural operator learns the mapping between function spaces rather than individual solutions, enabling a single trained model to predict system states at any parameter and time. This approach offers an efficient alternative to direct numerical simulation by avoiding the need for fine grids and short time steps as the current layer thins.