Reconstruction of muon trajectories in the 1-ton water-based liquid scintillator detector using machine learning

The Standard Model (SM) is incomplete—there exist many unanswered questions and instances where theory does not corroborate experiment. For example, neutrinos are predicted by the SM to have zero mass; however, flavor oscillations detected at Super-Kamiokande experimentally contradict this. To investigate this question, the Deep Underground Neutrino Experiment (DUNE) is being constructed in the United States with multiple detectors that employ uniquely novel methods to detect neutrinos. This project is part of one of these detector modules, Theia, which will use Water-based Liquid Scintillation (WbLS) to produce Cherenkov and scintillation light. As part of research and development for the future 100 kiloton Theia, Brookhaven National Lab has constructed 1-ton and 30-ton detectors to demonstrate the effectiveness of WbLS technology. This project is specifically focused on using machine learning to reconstruct the trajectories of muons passing through the 1-ton detector. This involves first applying timing corrections to synchronize the data, extract the PMT hit times, and calculate the charges at each PMT. Afterwards, using this observable input, event reconstruction is performed using convolutional neural networks, which use an extended maximum likelihood technique to evaluate parameters of interest. Despite the low photocathode coverage, I found that machine learning is a viable approach in both water and a solution with 1% WbLS.