Selection of $t$-channel Single Top Quark Events Using a Bayesian Neural Network to Explore the Wtb Vertex with the ATLAS Experiment

The use of a multivariate method based on a Bayesian neural network (NeuroBayes) was explored for the first time to provide a selection of t-channel single top quark events with the purpose to explore anomalous couplings in the Wtb vertex. The training and optimisation of this selection have been performed with simulated signal and background samples of Standard Model processes produced in proton-proton collisions at both centre of mass energies of sqrt(s) = 8 and 13 TeV provided by the ATLAS experiment at the LHC. After the fine-tuning of the neural network’s parameters and a careful choice of kinematic training variables, a cut in a single output discriminant, ONN, was applied to optimise the samples. For a fixed working significance of approximately 80 at $\sqrt{s}$ = 13 TeV, the neural network based method outperformed the previous selections based on sequential optimised cuts, yielding a much larger final value of 1.25 for the ratio of signal to background in comparison to the 0.60 from before. The Wtb vertex of in t-channel single top quark events, probed in both top quark production and decay, is described by a Lagrangian which parametrises the couplings present in the Standard Model, including the possibility of non Standard Model anomalous couplings. Measurements of these couplings can be extracted from calculations of the forward-backward asymmetries of the relevant angular distributions. To determine whether the new neural network based optimisation will bias the future measurements of potential anomalous couplings, an unfolding procedure to parton level was performed to simulated samples with anomalous couplings for the angular distributions cos(${\theta}_{\rm \ell}$) and cos(${\theta}_{\rm N}$) and their unfolded asymmetries calculated. It was found that the analysis maintained sensitivity to the presence of the anomalous couplings, as quantified by small statistical uncertainties in unfolded asymmetries, allowing us to make a consistency check of the Standard Model with these observables. However, the analysis exhibited deviations from the linearity of the response to their measurement. The deviations should be corrected for or prevented in future work prior to the application of the optimisation to real data in the search for anomalous couplings. Prospects for that purpose are discussed.