Fast and Differentiable MHD Surrogate for Solar Wind using Neural Operator

Abstract

Global magnetohydrodynamic (MHD) simulations provide the complete plasma state of the heliosphere, but their computational cost limits applications requiring repeated model evaluations. We introduce SNOW (Solar Neural Operator for Wind), a fast and differentiable surrogate for global MHD solar wind. SNOW combines a rotational ballistic approximation, a spherical harmonic neural operator, and causal radial context to autoregressively propagate all eight MHD variables from 20 to 240 Rs through con…