High-dimensional maximum-entropy phase space tomography using normalizing flows

Particle accelerators generate charged particle beams with tailored distributions in six-dimensional (6D) position-momentum space (phase space). Knowledge of the phase space distribution enables model-based beam optimization and control. In the absence of direct measurements, the distribution must be tomographically reconstructed from its projections. In this paper, we highlight that such problems can be severely underdetermined and that entropy maximization is the most conservative solution strategy. We leverage \textit{normalizing flows} -- invertible generative models -- to extend maximum-entropy tomography to 6D phase space and perform numerical experiments to validate the model performance. Our numerical experiments demonstrate that flow-based entropy estimates are consistent with 2D maximum-entropy solutions and that normalizing flows can fit complex 6D phase space distributions to large measurement sets in reasonable time.