GNN and Optimal Transport for Neutrino LArTPC Event Reconstruction

Liquid Argon Time Projection Chamber (LArTPC) detectors provide high-resolution images of neutrino interactions. Event reconstruction and particle identification of these images is crucial for neutrino oscillation measurements and searches for Beyond Standard Model (BSM) physics. While significant progress has been made in the past decade, accurately reconstructing electron and photon showers remains a challenge. We will be presenting novel reconstruction methods and preliminary results on their performance.

NuGraph2, a graph neural network developed for neutrino event reconstruction, is capable of performing accurate particle identification at a level that exceeds that of traditional reconstruction methods. When used in conjunction with existing reconstruction algorithms, NuGraph2 leads to improved reconstruction efficiency for shower events.

Classification of different types of showers produced by electrons and π⁰ decaying to two photons is also vital for LArTPC experiments. We employ Optimal Transport (OT) as a novel way for e/π⁰ separation in LArTPCs, leveraging this methods' ability to interpret and classify the topological differences in the charge deposition. Preliminary results on e/π⁰ separation using OT with the MicroBooNE public dataset are presented.