Reconstructing High-Energy Quantum Entanglement with Equivariant Machine Learning

This thesis presents a phenomenological analysis of top-quark pair entanglement in the boosted regime utilizing the recent Lorentz and permutation equivariant machine learning architecture, PELICAN, with 140 fb^-1 of simulated Monte Carlo proton-proton collisions at √s = 13 TeV where one top-quark decays semi-leptonically and the other fully-hadronically. Measuring entanglement in such an environment is a recent and novel area of research where energy scales are ten to twenty orders of magnitude higher than in typical entanglement experiments. Boosted hadronically-decaying top-quarks and their direct decay products are difficult to reconstruct with current popular methods. PELICAN directly regresses on particle four-momenta and this performance is compared to typical non-machine learning methods. We show that the impact of deploying this advanced symmetry-group-based machine learning architecture is robust against detector effects and systematic uncertainties. It is shown that PELICAN achieves entanglement detection with less physical bias and more accurate momentum reconstruction than the compared method. This unique use of a machine learning model to directly regress on four-momenta represents a significant contribution to the community's tools for understanding and using particle physics data.