Bayesian Beam Optics Measurement and Correction

Particle accelerators require extensive optics measurement and correction. Due to the complexity of a full analytic treatment, standard methods rely on multiple iterations of simplified linear algorithms under optimistic assumptions as to the noise distributions and systematic errors. We explore a reformulation of beam optics measurements as a Bayesian inference problem. This provides several advantages - posterior distributions of the fitted parameters, faster optics correction convergence, better interpretability, and the ability to take into account various nonlinear elements and effects. Algorithm is implemented by converting beamline elements into PyTorch framework so as to make use of efficient automatic differentiation, and Pyro probabilistic programming language is then used for inference. We demonstrate correct results of our algorithm on both simulated and experimental data measurements, as compared with LOCO algorithm. We also showcase advantages of our method in the presence of non-Gaussian noise and non-standard magnetic elements.