Abstract
Super-Kamiokande (Super-K) is a water Cherenkov detector nestled in the mountains of Gifu Prefecture in Japan. With over 27 years of history, it has studied neutrinos from diverse origins. Building on its legacy, the construction of Hyper-Kamiokande (Hyper-K) commenced in 2020. A significantly larger detector, Hyper-K aims to improve upon the Super-K results.This thesis is divided into two sections. The first focuses on the calibration of the
Super-K Outer Detector (OD) Photomultiplier Tubes (PMTs). It reports the OD PMT constants calculation techniques, results, and the impact of geomagnetic compensation coil failures in 2023 on those constants. Furthermore, it provides a detailed account of the current OD laser light injection system. The analysis of SK-VII laser data highlights a significant drop in laser light intensity, emphasising the need for an immediate laser replacement. It concludes with an evaluation of alternative light sources and the photon output required for an effective replacement.
The second section explores particle classification (distinguishing between e- and μ-/π0/γ) and kinematics reconstruction (momentum and vertex positions for e- and μ-) within Hyper-K, utilising PointNet, a machine learning model. The study analyses data of 3 million events per particle type, simulated with Hyper-K’s event simulation software, WCSim, which uses a tank geometry that consists of both 20-inch diameter PMTs and multi-PMT (mPMT) modules.
In comparison with FiTQun, a statistical reconstruction tool adapted from Super-K, PointNet demonstrates comparable efficacy in particle classification. Specifically, PointNet and FiTQun achieve similar performance in e-/μ- (99.7% and 99.8%) and e-/π0 (94.4% and 92.5%) discriminations, though both struggle to differentiate e- from γ. However, PointNet significantly outpaces FiTQun, being at least seven orders of magnitude faster in processing time. While PointNet surpasses FiTQun in momentum reconstruction for energies above 100 MeV, it falls short in accurately reconstructing vertex positions. Notably, the simulation geometry led to an overly optimistic position resolution with FiTQun, highlighting an area for future research improvement.
| Date of Award | 1 Nov 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Teppei Katori (Supervisor) & Jeanne Wilson (Supervisor) |