AI-accelerated solution of Maxwell's equations for the simulation of high-temperature superconductors dedicated to nuclear fusion

High-temperature superconductors (HTS) are a key enabler for the next generation of nuclear fusion reactors, thanks to their ability to sustain high magnetic fields while reducing power consumption. However, accurately modeling their electromagnetic behavior requires solving Maxwell’s equations through computationally expensive numerical simulations. In this work, we propose an AI-accelerated approach to solving Maxwell’s equations for HTS materials, leveraging state-of-the-art deep learning techniques. Specifically, we train an Adaptive Fourier Neural Operator (AFNO) and a Fourier Neural Operator (FNO) on a dataset generated using MAGNET, a finite-element solver developed at the Barcelona Supercomputing Center. We obtain an AI model that can predict the magnetic field evolution on a coarse time grid, in a completely unsupervised way. This approach opens new possibilities for integrating AI-based surrogate models into large-scale fusion simulations, potentially enabling real-time digital twins for superconducting magnet design.