Probing the B-Meson Anomalies Using ttbar+Z' Processes and Machine Learning

We perform a feasibility study to search for $Z'$ particles by probing their production in association with boosted top quarks at the LHC. In particular, we focus on the $Z' \to b\overline{b}$ decay mode and a semileptonic, semihadronic final state. Motivated by the lepton non-universality suggested by recent experimental findings in $b$ physics and muon $g - 2$ studies, we focus on a $Z'$ which couples strongest to the third generation fermions, substantiating our selected signal process. Assuming proton-proton collisions at $\sqrt{s} = 13$ TeV, we present the total $pp \to t\overline{t}Z'$ production cross sections and relevant kinematic distributions as a function of $m_{Z'}$ and the coupling product $g_{t\overline{t}Z'}g_{b\overline{b}Z'}$. We employ contemporary machine learning binary classification models---in particular, decision trees with gradient boosting---to discriminate between our signal and Standard Model backgrounds, resulting in increased discriminating power and a $5\sigma$ discovery reach that exceeds existing $m_{Z'}$ limits.