Improving electromagnetic shower reconstruction in LArTPC-based neutrino detectors: ICARUS and ArCS

Electromagnetic shower reconstruction represents one of the most challenging aspects of event reconstruction in neutrino liquid argon time projection chambers (LArTPCs). Shower fragmentation, overlap with hadronic activity, and ambiguities between electrons, photons, and secondary electromagnetic components directly affect neutrino physics measurements, including charged-current electron neutrino interactions and appearance searches. Improving electromagnetic shower reconstruction is therefore a central requirement for fully exploiting the physics potential of modern neutrino LArTPC experiments. This thesis addresses this problem through two complementary approaches. The first acts at the detector-design level and is developed within the ArCS (Argon detector with Charge Separation) R&D program, which aims at studying the operation of a magnetized LArTPC in a test-beam environment. The presence of a magnetic field enables charge-sign identification and curvature-based momentum reconstruction, thus yielding additional handles for the reconstruction of electromagnetic topologies. The work presented here focuses on DAQ-related tests and validation of the inherited LArIAT acquisition chain in preparation for future ArCS data-taking operations. The second approach acts at the pattern-recognition level within the ICARUS experiment, which is a large-scale LArTPC currently operating at Fermilab within the broader Short-Baseline Neutrino (SBN) program. In particular, this thesis presents the retraining and validation of NuGraph2, a graph neural network (GNN) algorithm performing hit-level background filtering and semantic classification. Finally, preliminary studies are presented on the impact of NuGraph2 outputs at the end of the reconstruction chain, within electron neutrino charged-current interactions, and, more specifically, compared to the standard reconstruction.