Author ORCID Identifier
Document Type
Dissertation
Date of Award
2024
Degree Name
Doctor of Philosophy (PhD)
Department
Physics
First Advisor
Wenqin Xu, Jing Liu
Abstract
Neutrinoless Double-Beta Decay (0$\nu \beta \beta $) is a presumed rare nuclear decay process and is considered the most promising way to prove the Majorana nature of neutrinos, that is, neutrinos are their own antiparticles. The discovery of \zero decay would also enhance our understanding of nuclear physics, astrophysical observations, and physical processes in the early Universe. The Large Enriched Germanium Experiment for Neutrinoless Double-Beta Decay (LEGEND) is a phased $^{\mathrm{76}}$Ge-based 0$\nu \beta \beta $ decay experimental program, aiming for the discovery potential of a half-life of 0$\nu \beta \beta $ beyond 10$^{\mathrm{28}}$ years. The first phase, LEGEND-200, comprises 200 kg of $^{\mathrm{76}}$Ge-enriched detectors and has started its physics run with an initial 142 kg of germanium detectors. The second phase of the experiment, LEGEND-1000, has a background goal of 1 $\times 10^{-5}$ cts/(keV kg yr) at the Q-value (2039-keV) of 0$\nu \beta \beta $. Mitigating sources of background and rejecting background events through optimized detector design and analysis techniques are crucial to detect rare events such as 0$\nu \beta \beta $. In this work, detailed studies are done to understand and estimate backgrounds from ($\alpha, n$) neutrons, gammas, and decays from radioactive chlorine isotopes produced in the argon shield. Estimates are provided for the production rates of $^{40}$Cl, $^{39}$Cl, and $^{38}$Cl using nuclear and particle transport and simulation tools. In addition, the analysis of the LEGEND-200 data to identify the signatures of decays of these isotopes is presented. Apart from using conventional physics and statistical techniques, this work also uses machine-learning techniques to perform pulse-shape analysis. The performance of the recurrent neural network (RNN) model is evaluated based on classification and regression analysis of single and pileup events. This study shows that the RNN model can be used to discriminate single and pileup waveforms, single-sited and multi-sited events to improve the sensitivity of experiments of rare-event searches.
Subject Categories
Elementary Particles and Fields and String Theory | Nuclear
Keywords
(alpha n) neutrons, Artificial Intelligence (AI), Backgrounds, Isotopes production, Neutrinoless double-beta decay, Recurrent Neural Network
Number of Pages
156
Publisher
University of South Dakota
Recommended Citation
Paudel, Laxman, "BACKGROUNDS STUDY AND AI-POWERED WAVEFORM PROCESSING FOR LEGEND EXPERIMENT" (2024). Dissertations and Theses. 225.
https://red.library.usd.edu/diss-thesis/225