ESSνSB GNN Flavour ID in Near WC Detectors
A study shows that graph neural network classification can identify electron and muon neutrino interactions in ESSνSB near water Cherenkov detectors even when the detector volume is reduced eight-fold or when photomultiplier coverage is lowered, with only modest loss in efficiency that can be offset by longer exposure.
ESSνSB GNN Flavour ID in Near WC Detectors
The ESSνSB experiment, submitted on 24 Aug 2026 by Kaare Endrup Iversen, seeks to measure CP violation in the leptonic sector with high precision. To achieve this, it relies on accurate reconstruction of neutrino events in water Cherenkov (WC) detectors. The study explores whether a smaller or less instrumented near WC detector can still deliver the required flavour-identification performance using graph neural network (GNN) classifiers.
Detector Design Variations
The researchers used detailed Monte Carlo simulations of charged-current (CC) neutrino interactions to train GNN classifiers that distinguish electron and muon neutrino CC events. They varied key detector parameters, including overall volume and photomultiplier tube (PMT) coverage. The simulations employed fixtures that model the detector geometry and PMT layout. Results indicate that even when the detector volume is reduced by a factor of eight, the GNN remains accurate. The loss in classification efficiency is moderate when background rejection is held fixed. This suggests that a smaller detector can still meet the experiment’s needs, provided the PMT coverage is maintained in the most signal-rich regions, especially near the forward end-cap.
GNN Classification Performance
The study found that reduced PMT coverage in the nominal detector design has a limited impact on classification performance. The GNN can compensate for lower light collection if the remaining PMTs are concentrated in areas of highest signal yield. This is crucial for maintaining high purity in flavour identification. The authors report that the efficiency drop can largely be offset by increasing the exposure time, which would allow the experiment to collect more events without sacrificing classification quality.
Implications for Exposure Time
Because the GNN retains robust performance across a range of detector configurations, the ESSνSB collaboration can consider cost-saving measures such as smaller detector volumes or selective PMT deployment. However, any reduction in detector size or PMT density will require longer data-taking periods to achieve the same statistical power. The study’s findings are documented in the arXiv preprint, where the authors provide detailed stats on classifier performance under each configuration.
The work demonstrates that advanced machine-learning techniques like GNNs can mitigate hardware limitations in neutrino detectors. By carefully designing PMT placement and accepting modest efficiency losses, the ESSνSB experiment can still pursue its goal of measuring CP violation with high precision. The next step will involve validating these simulation results with prototype detector data, a task that the collaboration plans to undertake in the coming months.
