Search for long-lived heavy neutral leptons with the CMS experiment using machine learning techniques

This Thesis presents the analysis strategies and outcomes of a search looking for long-lived Heavy Neutral Leptons (HNLs) in proton-proton collision data corresponding to 137fb$^{-1}$ integrated luminosity, collected by the CMS experiment at the Large Hadron Collider with $\sqrt{s}$=13 TeV. The HNL particles are searched for via charged-current production and semi-leptonic decay mode into a lepton and jet that can be displaced from the primary collision point. A fine-grained event categorisation scheme ensures enhanced sensitivity to a number of HNL benchmark models including both Dirac- and Majorana-type particles. The search employs a novel deep neural network-based displaced jet tagger and a complementary event-level boosted decision tree algorithm with the aim of separating HNL signal events from Standard Model (SM) background processes. The blinded analysis uses a data-driven background estimation method to evaluate the expected yields, the method being validated in two control regions. After unblinding, no excess of events was found and upper limits on the HNL production cross-section were set for a broad range of HNL mass, lifetime and coupling scenarios, including universal mixing with all SM lepton generations. A particular focus of this Thesis was the evaluation of displaced lepton reconstruction and identification of scale factors, which were found to be one of the dominant systematic uncertainties associated with the signal interpretation. The results presented in this Thesis are competitive with previously published similar searches in high energy physics and are the first in the CMS collaboration to consider arbitrary coupling combinations involving all three SM lepton generations.