The BEST Thesis: The Boosted Event Shape Tagger, A Search for Vector-like Quarks, and A Real GEM in CMS

This thesis is a collection of three topics that take place at the Compact Muon Solenoid experiment---a particle detector which observes proton-proton collisions at CERN's Large Hadron Collider. These topics are the installation of GEMs into the CMS experiment, the development of the Boosted Event Shape Tagger, and a search for vector-like quarks. The LHC is undergoing upgrades which will increase the instantaneous luminosity to $5\times10^{34}$cm$^{-2}$s$^{-1}$, a factor of 2.5 higher than the current maximum value. Therefore, the experiments are implementing upgrades to cope with the augmented particle rates. In the muon system of the Compact Muon Solenoid (CMS) experiment, Gas Electron Multipliers (GEMs) are being installed to complement the existing Cathode Strip Chambers (CSCs). This will provide a more precise measurement of the muon bending angle and thus improve the muon trigger capabilities. GEMs are micro-pattern gaseous detectors with high rate capabilities–ideal for the forward regions of the CMS muon system. In preparation for the LHC Run 3, 144 GEM chambers were installed in the first muon station and are now operational in Run 3. This thesis introduces the GEM technology and discusses the production, installation, commissioning, and operation of the new GEM muon detectors at CMS. The first GEMs in CMS will improve the identification of muons, but proper identification of hadronic decays requires the development of new analysis tools. Jets from heavy particles (top, Higgs, $W$, $Z$) have characteristic patterns that can be identified by Lorentz boosting the jet to various hypothetical rest frames. The Boosted Event Shape Tagger (BEST) is a deep neural network that utilizes this technique to classify heavy particles from QCD background. A version of BEST was previously used for 2016 collision data. This version was improved on for the full Run 2 dataset. In the effort to improve BEST resulted in a method for creating images of jets in rest frames. These images were passed to a convolutional neural network for classification. This thesis discusses this method and the other improvements to BEST in detail. The improved version of BEST was used to search for a pair of vector-like quarks in an all hadronic final state in LHC Run 2 data. Vector-like quarks arise in extensions to the Standard Model which aim to solve the gauge hierarchy problem. This search uses BEST to classify collision events into 126 orthogonal regions. The $H_{T}$ distributions are tested in each region for the presence of signal and exclusion limits are set for $T'$ and $B'$ masses. This search is currently being approved by the CMS experiment, so only expected limits are presented---the expected sensitivity of the search if no signal is present. The process for setting expected limits is completed using Monte Carlo simulated data and data driven estimates, so no collision data from the signal region are included.