ModeCast: One-Shot Prediction of Linear MHD Eigenmodes and Growth Rates in Tokamaks

Large-scale magnetohydrodynamic instabilities break plasma confinement and set hard limits on fusion reactor design; certifying one configuration against them takes hours with a reference solver such as JOREK, and a design study needs thousands. ModeCast is the first learned surrogate to return the complete linear stability assessment across instability families: trained on 46,034 JOREK simulations over 12,990 equilibria, it screens stability and predicts the unstable mode's full spatial structure and growth rate directly from the equilibrium, roughly $10^5\times$ faster than the solver. The gain comes from building the linear-phase physics into the formulation, the objective, and the inputs, worth up to six times more training data. It misses none of 1,258 unstable held-out cases, and its predicted fields reproduce the simulation's measured physics, from magnetic island topology to resistivity trends. Differentiable and full-field, ModeCast can assimilate sparse experimental measurements: a bridge toward real devices and an accelerant toward commercial fusion.