Abtin Narimani Charan ; Torben Ferber

2024
Deutsches Elektronen-Synchrotron DESY Hamburg

Abstract: This dissertation revolves around the utilization of Convolutional Neural Networks (CNNs) to advance Particle Identification (PID) within the Belle II Electromagnetic Calorimeter (ECL). The core goal of the research is to refine the differentiation process between low-momentum muons and charged pions. The ECL plays a significant role in the PID system as it is engineered to measure the energy deposition by both charged and neutral particles. The task of identifying low-momentum muons and charged pions within the ECL becomes particularly vital when they fail to reach the outer muon detector. In order to provide optimal data, the study employs track-seeded cluster energy images. The energy deposition patterns for muons and charged pions, as detected within crystals surrounding an extrapolated track at the ECL’s entry point, are integrated with crystal positions in the θ − φ plane along with the track’s transverse momentum. This amalgamation of information is then utilized to train the CNN, capitalizing on the distinctiveness between the dispersed energy depositions of pion hadronic interactions and the more localized muon electromagnetic interactions. The study includes a comparison of the CNN algorithm’s performance with other PID methods currently in use at Belle II, which predominantly rely on track-matched clustering information. The findings imply that the CNN PID method improves the separation between muons and charged pions in low-momentum regions. The research includes samples with varying beam backgrounds, including no beam background. The effectiveness of the CNN method has been assessed with different energy thresholds for ECL crystals, utilizing 21.5 fb−1 data from 2020 and 2021 and Monte Carlo (MC) samples. To substantiate the CNN method with real data, clean samples of muons and charged pions have been singled out using e+e− → μ+μ−γ and D∗+ → D0(→ K−π+)π+, respectively. Finally, recognizing that the CNN is sensitive to tracks in close proximity within a single event, additional research was conducted to evaluate the CNN’s performance with isolated and non-isolated tracks within the ECL.

Note: Presented on 09 02 2024
Note: PhD