A Search for Stealth and R-Parity Violating SUSY using a Neural Network Based ABCD Method

The Compact Muon Solenoid (CMS) collaboration is responsible for many searches for supersymmetry using collision data from the Large Hadron Collider (LHC). Most of these searches assume that the lightest supersymmetric particle (LSP) is stable and does not leave a direct signature in our detector. However, some theories predict that supersymmetry could be "stealthy" and evade previous search techniques by decaying to SM particles. Two such candidate theories are the R-Parity violating (RPV) and Stealth SYY SUSY models which result in supersymmetric quark decays to SM particles with little to no missing energy. We have performed a search for RPV and Stealth SYY-like decays of top squarks (stops) in LHC collision data using the full LHC Run 2 dataset (137 /fb) with center of mass energy √s=13 TeV. This analysis presents a particularly challenging problem in that these decays are almost indistinguishable from those of top-antitop production, a common LHC background. Through the use of a neural network (NN) in conjunction with the ABCD background estimation method, following the "Double DisCo" approach extended by an additional NN loss term minimizing the ABCD non-closure directly, we are able to establish a data-driven background estimation in the search for these supersymmetric decays. This novel approach can be applied to a wide range of particle physics analyses.