Using Geometric Deep Learning for the Detection of the Diffuse Supernova Neutrino Background in Water-Cherenkov Detectors

Abstract

The study of the Diffuse Supernova Neutrino Background (DSNB) is an essential approach to improve our understanding of the formation of Core-Collapse Supernovae, as well as investigating fundamental neutrino properties. The detection of the DSNB flux however is a difficult task, as it requires the disentanglement of its signal with many competitive backgrounds in the low energy range. The main DSNB detection channel is the Inverse Beta Decay (IBD), which is characterised by a distinct, time-separated two-signal signature. This work introduces the use of Graph Neural Networks (GNNs) to handle critical detection and reconstruction tasks for the DSNB through the IBD channel. In particular, we present a model to recover the energy of the first signal from a low energy positron. Additionally, we develop a signal-to-background classifier optimised for the secondary signal, which corresponds to the characteristic photon emission following a neutron capture event. We test our models on two simulations in a Hyper-Kamiokande-like water Cherenkov detector geometry, and show that we obtain performances that are comparable to the state of the art.