![HEP2026: 42nd Conference on Recent Developments in High Energy Physics and Cosmology [NTUA] Jul 8 – 11, 2026](/event/1654737/logo-1929213524.png){kind=logo}
Speaker
Description
ESSnuSB is a future long base-line (LBL) neutrino oscillation experiment, based on the low energy neutrino super-beam which can be generated using the European Spallation Source (ESS) Linear proton accelerator.The experiment is designed to measure the CP violating phase angle with an accuracy of less than $8^{\circ}$. To achieve such a goal, the very precise knowledge of the neutrino interaction cross section is extremely important.
LEMMOND is a water Cherenkov detector currently under development for measuring low energy neutrino interaction cross sections on water. The detector is planned to operate during the first development phase of the implementation of the ESS$\nu$SB project. This work demonstrates the application of Graph Neural Networks (GNNs) and other Deep Learning (DL) techniques for reconstructing LEMMOND signals, estimating neutrino energies, and extracting neutrino cross sections.
In this study, we use simulations of the Low Energy NuSTORM facility as a source of neutrino and antineutrino fluxes originating from muon decays, together with the GENIE event generator to simulate neutrino interactions within the effective volume of the LEMMOND detector. A modified version of Geant4 is employed to model in detail the propagation of charged particles in water, the production of Cherenkov photons, and the response of the photodetectors. The reconstruction of the charged lepton kinematic variables is based on GNN models trained on Geant4-simulated data and evaluated using statistically independent simulated samples.
Due to uncertainties in the exact neutrino direction and, more importantly, the impact of nuclear effects on the final state kinematics, a probabilistic Machine Learning (ML) approach has been advanced to provide a probability density function describing the experimental knowledge of the neutrino energy on an event by event basis. We demonstrate that this probabilistic approach can successfully reproduce the energy spectrum of the incoming neutrino flux.
Furthermore, the reconstructed information is used to estimate the energy dependence of the neutrino cross section. In addition, we investigate the size of systematic uncertainties arising from the modelling of nuclear effects by training the probabilistic neural network using different nuclear models.