Comparison of physics-informed neural networks in solving electromagnetic interior scattering problems including a relativistic beam current

Abstract

The strength of electromagnetic interaction between a relativistic beam and its surrounding environment in a particle accelerator can be characterized by the coupling impedance. Recently, the physics-informed neural networks (PINN) have been introduced into the impedance modeling in accelerator physics. Total-field (TF) and scattered-field (SF) formulations are available in calculating the coupling impedance with PINN. In this paper, direct comparison of the two PINNs based on the TF and SF formulations is presented with application to an elliptical vacuum chamber with practical geometry parameters. The numbers of iterations for the training processes and the accuracy of indirect space charge impedance are assessed for the two different PINNs in this comparison.