Beam-Level Nonlinear Compton Spectra via a Neural Network Surrogate Model

Abstract

Nonlinear Compton scattering enables sources of intense high-energy radiation. However, predicting source-level spectra for realistic electron beams typically requires computationally expensive trajectory-based calculations, whereas existing analytical models for spectral envelopes are limited to a restricted range of laser pulse parameters. Here, we demonstrate that, for electron beams with sufficiently broad phase-space distributions, beam-level nonlinear Compton spectra can be predicted without lengthy numerical calculations. Our approach combines a fast neural-network surrogate for single-electron spectral envelopes with particle-wise Lorentz transformations. Benchmarked against trajectory-based numerical calculations with thousands of macroparticles, the surrogate reproduces the macroscopic spectral-angular structure while reducing the computational cost by orders of magnitude.