Toward the Fast Electromagnetic Solver with Physics-Informed Fourier Neural Operator

In this study, a Physics-Informed Fourier Neural Operator (PI-FNO) is proposed for solving electromagnetic problems. The imbalance among different loss terms in conventional Physics-Informed Neural Networks (PINNs) is avoided by directly approximating the integral physical laws. By leveraging the Green's function and neural operators, the PI-FNO exhibits strong adaptability to varying mesh densities with a single training run. Its effectiveness is validated through several electromagnetic scattering examples, including strong field-scattering interactions. Results indicate that proposed method delivers high accuracy and robustness while significantly reducing computational cost, providing a resolution-insensitive neural framework for fast electromagnetic modeling and solving.