Using a GNN to Determine the Primary Cosmic Ray Energy in IceCube

The IceCube Neutrino Observatory consists of an array of optical sensors a mile deep in the glacial ice near the South Pole station and an array of ice Cherenkov detectors on the surface. IceCube detects cosmic messengers like neutrinos, high energy gamma rays and cosmic rays. In the analysis of cosmic ray events in IceCube, one of the goals is to reconstruct the energy of primary cosmic ray particles that have struck the atmosphere and generated cascades of particles known as air showers. To do this, a lateral distribution function is often fitted to the charge distribution measured by the surface detectors. The true energy of the cosmic ray particle can then be determined by a lateral distribution parameter known as S125. An alternative measurement of the primary energy is through machine learning. This work investigates if using a Graph Neural Network (GNN) is more efficient, flexible or precise than the standard S125 fit.