Photon Energy Reconstruction using Machine Learning at the Pierre Auger Observatory

An energy reconstruction for photon-induced air showers at ultra-high energies ( ≥ 1018 eV) is presented for the surface detector of the Pierre Auger Observatory. Photon showers have a signature that differs from that of hadron-induced showers: the photon shower composition is almost exclusively electromagnetic and they show a steeper lateral distribution function as well as a larger depth of the shower maximum. In order to improve the resolution of the energy prediction, a reconstruction method based on ML is taken into consideration and compared to the classical hadron shower reconstruction applied to photon-induced extensive air showers. Due to the high success rate in other areas of machine learning, the encoder stack of the so-called transformer architecture is explored as an alternative to the more traditional approach of convolutional networks. So far, no photon events in the Pierre Auger dataset have been unequivocally identified as photons, but the advances in ML could play a key role in detecting them in the future.