Enhancing the Resolution of the MUSE Beamline Calorimeter Using Machine Learning

2010 studies of μ spectroscopy found the radius of the proton to be ∼ 0.842 ± 0.001 fm, a deviation of more than 5σ from electron-proton scattering and atomic hydrogen spectroscopy. This discrepancy, dubbed the "Proton Radius Puzzle," has found several possible explanations concerning lepton universality, differences in the handling of radiative corrections, and systematic uncertainties in ep proton form factors. The MUon Scattering Experiment (MUSE) aims to address the Proton Radius Puzzle with simultaneous ep and μp scattering at both positive and negative polarities. MUSE, located at the Paul Scherrer Institute, consists of a mixed beam of e, μ, and π incident on a liquid hydrogen target with downstream scattering detectors. A leading order radiative correction to the ℓp cross-section is Bremsstrahlung radiation. To suppress these effects, MUSE introduces an electromagnetic calorimeter downstream of the target to detect any hard initial state photon radiation. We present a convolutional neural network framework designed to improve the reconstruction of photon energies, enhancing the calorimeter's resolution and its effectiveness in identifying hard initial-state radiation