Authors: Nathaniel H. Mathews (NASA GSFC), Barbara J. Thompson (NASA GSFC)

Predicting the current or future state of the coronal magnetic field requires high-resolution models with accurate physics. Such models are typically too expensive to be used in an inversion framework, which is a kind of nonconvex optimization scheme that must run the forward model many times. We present ongoing work on emulating expensive physical models with a lightweight machine learning technique combining Fourier network layers and Physics-Informed Neural Network algorithms. By training the technique to emulate a specific numerical model, we aim to produce easily-interpretable results that can be refined by full numerical simulations.