Physics-Informed Neural Operator Surrogate for 2D Magnetohydrodynamic Reconnection

The paper targets magnetic reconnection in two-dimensional resistive magnetohydrodynamics, where direct numerical simulation becomes prohibitively expensive at large Lundquist numbers because the current layer becomes thinner and requires fine grids and short time steps. It explores a physics-informed neural operator as an alternative that learns a mapping between function spaces, allowing one trained model to return reconnection states for different parameters and times at inference.