Exploring ESS$ν$SB Near Water Cherenkov Detector Designs Through Graph Neural Network Flavour Identification

Abstract

The ESS$ν$SB experiment aims to measure CP violation in the leptonic sector with high precision, necessitating robust reconstruction of neutrino events in the water Cherenkov (WC) detectors. In this work, we investigate the flavour identification potential of the proposed near WC detector using graph neural network (GNN)-based classification, with a focus on variations of key detector design parameters. In particular, we study whether a smaller and/or less instrumented detector can achieve the required classification performance. Using detailed Monte Carlo simulations of charged-current (CC) neutrino interactions, we train GNN classifiers to distinguish electron and muon neutrino CC events. We find that GNN-based classification remains accurate even for detector configurations with volumes up to a factor of eight smaller than the nominal design, with only moderate degradation in classification efficiency at fixed background rejection. The resulting loss in efficiency can largely be compensated by increased exposure time. Furthermore, we demonstrate that reduced photomultiplier tube (PMT) coverage in the nominal detector has a limited impact on classification performance, provided that coverage is maintained in regions of highest signal yield, in particular near the forward end-cap.