Suppressing Beam Background and Fake Photons at Belle II using Machine Learning

Abstract

The Belle II experiment situated at the SuperKEKB energyasymmetric e+ e− collider began operation in 2019. It has since recorded half of the data collected by its predecessor, and reached a world record instantaneous luminosity of 4.7 × 1034 cm-2s-1. For distinguishing decays with missing energy from background events at Belle II, the residual calorimeter energy measured by the electromagnetic calorimeter is an important quantity. Ideally, calorimeter clusters due to beam backgrounds and fake photons should be excluded when the residual calorimeter energy is calculated, so identifying them during the analysis process is key. We present two new boosted decision tree classifiers that have been trained to identify such clusters at Belle II and distinguish them from real photons originating from collision events at the interaction point. We provide results from their application to the B → D*ℓν decay mode, and show that the distribution of residual calorimeter energy for signal events is significantly improved.