Date Thesis Awarded

4-2020

Access Type

Honors Thesis -- Access Restricted On-Campus Only

Degree Name

Bachelors of Science (BS)

Department

Physics

Advisor

Justin Stevens

Committee Members

James Kaste

Keith Griffioen

Abstract

Lattice Quantum Chromodynamics (QCD) calculations predict a whole spectrum of exotic hybrid mesons thought to arise from gluonic excitation between a quark and antiquark. A systematic survey of strange and non-strange decay modes would be necessary to confirm this, and in particular studying decays to kaonic final states could confirm initial lattice QCD predictions. The GlueX experiment at Jefferson Lab studies the non-perturbative regime of the strong interaction and searches for hybrid mesons through light-quark meson spectroscopy; thus, kaon identification is critical for the GlueX experiment. Currently, the GlueX DIRC (Detection of Internally Reflected Cherenkov light) detectors provide clear kaon/pion separation up to momenta of 4GeV; however, at high momenta the hit patterns become nearly indistinguishable as the Cherenkov angles are quite similar. We have developed Convolutional Neural Networks (CNNs) with supervised learning to distinguish high momentum kaon and pion hit patterns on the GlueX DIRC detectors. Training on simulation data, we achieved fairly high test accuracy up to 6GeV momenta; however, further work is needed to improve CNN performance at higher momenta and across a range of different particle trajectories. If successful, this method could significantly expand the kinematic coverage of the GlueX detector and could be applied to analyses of further reactions.

Recommended Citation

McEneaney, Matthew, "Development of Convolutional Neural Nets for Κ/π Differentiation at the GlueX Experiment" (2020). Undergraduate Honors Theses. William & Mary. Paper 1477.
https://scholarworks.wm.edu/honorstheses/1477