Search for vector-like leptons at CMS and prospects at the HL-LHC

Abstract

The standard model of particle physics (SM) describes fundamental particles and their interactions. The predictions of the SM have been extensively tested in the experiments. However, it does not explain key phenomena such as the presence of dark matter, small non-zero neutrino masses, or gravity, suggesting the existence of beyond the standard model (BSM) physics. Identifying rare BSM signals from the SM backgrounds with the increas- ing complexity arising from high-granularity detector information and the growing volume of data taken or proposed (High luminosity LHC) by the current experiments sets a unique challenge. Advanced techniques such as Machine learning (ML) algorithms are well-suited for this task, as they can efficiently process large datasets and uncover hidden patterns in high-dimensional data. Even if the LHC experiments excluded a large BSM parameter space at the TeV scale, minimal extension to SM, such as vector-like leptons (which appear in SUSY, GUT, etc) at the electroweak scale can evade detection primarily due to the overlapping signatures with the dominant SM processes (control region in usual searches), or the analysis choices that are tuned to probe massive particles. A targeted search for vector-like leptons at the electroweak scale is conducted in the final state with one muon and at least two jets using proton-proton collision data of integrated luminosity of 138 fb −1 collected by the CMS experiment at the LHC from 2016–2018 at center-of-mass energy 13 TeV. Deep neural network (DNN) based classifiers are used in a unique combination to suppress each background. No significant deviations from the SM expectations were observed. While this search did not have enough sensitivity to exclude phase space for the considered models, it exercised the power of good analysis to improve the signal-to-background significantly. The heavy fermion search program will benefit from the detector enhancements and the increased center-of-mass energy ( s = 14 TeV) with an unprecedented integrated luminosity of 3000 fb −1 at HL-LHC. The discovery potential of vector-like leptons coupling to first-, second-, and third-generation SM leptons is reported using multiple leptons in the final state for a high-luminosity LHC scenario (HL-LHC). Detector granularity and increased luminosity expected at HL-LHC require a significantly faster yet accurate event simulation pipeline to be developed to exploit the full potential of the HL-LHC physics reach. Variational Auto Encoder and Generative Adversarial Network-based generative models with dimensionality reduction techniques are discussed to build such a faster simulation chain, and as an aid to understand the performance of deep learning networks.