Event Reconstruction in JUNO-TAO using Graph Convolutional Networks & Optimization of Horn Geometry in ESSnuSB+ using Genetic Algorithm

Abstract

The primary goal of JUNO is to resolve the neutrino mass hierarchy using precision spectral measurements of reactor antineutrino oscillations. To achieve this goal a precise knowledge of the reactor spectrum is required. Since the existing reference spectra show a deficit in measured reactor fluxes, TAO, a ton-level, liquid scintillator detector with a baseline of 44 m, is set up as a reference detector to JUNO. With a set of 4024 Silicon Photomultipliers (SiPM) and an operating temperature at -50∘C, TAO is expected to record about 2000 antineutrino events per day and aims to achieve a resolution of less than 2% at 1 MeV. For that, a precise reconstruction of the reactor antineutrino events is necessary. These events occur through the Inverse Beta Decay (IBD), producing a prompt positron and delayed neutron signal. Since positron events carry most of the energy, this thesis focuses on the vertex and energy reconstruction of positron events generated by the official TAO offline software. The reconstruction was carried out through Graph Convolutional Networks (GCNs). A graph, resembling the detector with 4024 nodes representing SiPMs with features, first hit time and hit counts was modeled. The model was trained and validated on 5 million events covering energies from 1-10 MeV. The final evaluation on the 1 MeV subset resulted in a vertex resolution of 8 mm and energy resolution of 1.8 %. Notably, both vertex and energy resolutions even increased for higher energies. The ESSnuSB+ aims to precisely measure neutrino interaction cross sections below 600 MeV. The 2 GeV proton beam from the ESS hits the titanium target, resulting in a secondary hadron beam predominantly consisting of pions. The focusing of charged pions is done by the magnetic horns and is critical for generating intense neutrino beams. In this study, the horn is simulated using FLUKA and its configuration is optimized utilizing a Genetic Algorithm (GA). Dimensions of the horn like the lengths, radii, heights, and current were optimized for 50 generations, after which no significant improvement was observed. The fitness score, a measure of detection efficiency, improved from 0.725 to 0.860, resulting in a 20% increase in pion concentration, with the optimized horn configuration. This enhanced focusing will improve neutrino flux and precision in measurements of neutrino cross sections.