Physics-Informed Neural Modeling of 2D Transient Electromagnetic Fields

Electromagnetic wave propagation in complex environments demands accurate yet efficient modeling techniques. This study introduces a physics-informed neural network (PINN) framework for two-dimensional transient electromagnetic analysis, where Helmholtz equations are directly incorporated into the loss function. The model learns spatiotemporal field evolution without relying on spatial discretization or labeled data. Various excitation and material conditions are examined, including single and dual Gaussian sources in both free space and inhomogeneous regions with dielectric and conducting inclusions. Through this formulation, the network captures key wave phenomena such as propagation, reflection, and scattering with high precision. Validations against finite-difference time-domain (FDTD) simulations confirm strong agreement in both temporal and spatial field distributions. The results demonstrate that the proposed PINN provides an effective, mesh-free alternative for modeling electromagnetic wave dynamics, offering scalability for complex and data-sparse scenarios.