Abstract: We used the physics-informed neural networks method to numerically solve the problem of discontinuous coefficient light wave diffraction. The results show that approximating the discontinuous coefficient with a smooth function can significantly improve the accuracy of the physics-informed neural network solution. Using physics-informed neural networks to solve the scattered field is better than directly solving the total field. Finally, the correctness of the theoretical results is verified through numerical experiments.

Key words: physics-informed neural network, light wave diffraction, discontinuous coefficient, smooth function