Improving combinatorial ambiguities of t(t) over bar events using neural networks

Abstract

We present a method for resolving the combinatorial issues in the t(t) over bar lepton + jets events occurring at the Tevatron collider. By incorporating multiple information into an artificial neural network, we introduce a novel event reconstruction method for such events. We find that this method significantly reduces the number of combinatorial ambiguities. Compared to the classical reconstruction method, our method provides significantly higher purity with the same efficiency. We illustrate the reconstructed observables for the realistic top-quark mass and the forward-backward asymmetry measurements. A Monte Carlo study shows that our method provides meaningful improvements in the top-quark measurements using the same amount of data as other methods.