Comparison of deep neural networks and variational quantum circuits for quark-gluon jet classification

Abstract

Jets, the hadronization form of quarks and gluons produced in high energy collisions, is one of the significant probes to investigate the quark-gluon plasma, a state of matter that existed shortly after the Big Bang. The origin of these collimated sprays of particles is identified by a jet discriminator providing information about the dynamics of quarks and gluons for a deeper understanding of quantum chromodynamics. Machine learning (ML) techniques, particularly deep neural networks (DNNs), have been widely used in high energy physics analysis to enhance the interpretation of datasets at the large hadron collider (LHC). Recently, the quantum ML (QML) approach with variational quantum circuits (VQC) offers potential advantages in processing high dimensional data with fewer resources. Therefore, in this study DNNs and VQC were implemented for the identification of jet origin if it is from light quarks [up (u), down (d), and strange (s)] or gluons (g) from simulated proton-proton collisions at the compact muon solenoid (CMS) detector with a center-of-mass energy of 13 TeV generated with Pythia 8. The results were compared with the jet likelihood discriminator tool used at the CMS to evaluate the performance of ML and QML for jet origin determination in large datasets produced at the LHC. While DNNs achieve superior performance across precision, recall, and F1 metrics, VQC shows potential despite optimization and data size limitations. This study highlights the strengths and challenges of classical data analysis with classical and quantum computing approaches, offering valuable insights into their applicability to particle physics.