Using Machine Learning to Hunt for Simulated WIMPs in the NOvA Near Detector

Abstract

A neural network was trained on simulated data that included events in which electrons were scattered by hypothetical Dark Matter particles (χ) of mass mχ = 30 MeV assuming a dark vector portal mechanism of a dark photon (A') with mass mA' = 90 MeV, a gauge coupling parameter αD = 1/2, and kinetic mixing parameter e = 2 × 10 -5. The NOvA Near Detector’s response to these events was then simulated, and the pixelmaps (images) of these events occurring within the NOvA Near Detector were then used to train a machine learning algorithm designed to differentiate between the each of the ordinary observed event types that involve an electron scattered by a neutrino and hypothetical events in which an electron was scattered by a dark matter particle.