Search for Heavy $ZZ$ Resonances in the 4$\ell$ Final State with the ATLAS Detector and Identifying Muons in the ATLAS Calorimeter using Machine Learning

This doctoral thesis explores two lines of work: (i) a search for heavy Higgs-like resonances decaying via two $Z$ bosons to four leptons, and (ii) the development of a novel algorithm for tagging muons in the ATLAS calorimeters. The search concerns signatures of an extended Higgs sector motivated by extensions to the Standard Model of particle physics that address the large discrepancy between the gravitational and the electroweak energy scales known as the hierarchy problem. A search for heavy $ZZ$ resonances decaying to $\ell^+\ell^-\ell^{\prime +}\ell^{\prime -}$, where $\ell$ stands for either an electron or a muon, is conducted in the mass range between 200 and 2000 GeV using 139 fb$^{-1}$ of proton-proton collision data collected at a centre-of-mass energy of $\sqrt{s}=13$ TeV with the ATLAS detector during $2015-2018$ data-taking at the LHC. The search is performed separately in two analysis categories corresponding to the resonance production mechanisms presumed to be dominant: the fusion of two gluons (ggF) and the fusion of two vector bosons (VBF). A novel method for categorising candidate events according to their production mechanism based on deep learning is developed. It is validated in the context of the search in the $\ell^+\ell^-\ell^{\prime +}\ell^{\prime -}$ final state and is sensitive to high-mass signals with up to $40\%$ higher discovery significance compared to a previously employed, simpler categorisation scheme. No significant excess in the four-lepton invariant mass distribution is found, and upper limits on the production cross-section times branching fraction of a heavy scalar are set. Using a combination of results with the $\ell^+\ell^-\nu\bar{\nu}$ analysis channel, where $\nu$ stands for a neutrino, the observed limits on the production cross-section times branching fraction are between 215 fb at 240 GeV and 2.0 fb at 1900 GeV in the ggF production mode, and between 87 fb at 255 GeV and 1.5 fb at 1800 GeV in the VBF production mode. Upper limits on the production cross-section times branching fraction of a spin-2 Randall-Sundrum graviton are set. Randall-Sundrum gravitons are excluded up to a mass of 1830 GeV at 95% confidence level. The efficient identification of muons is central to the search in the $\ell^+\ell^-\ell^{\prime +}\ell^{\prime -}$ decay channel and many other physics analyses. The ATLAS detector has a dedicated instrument for muon identification: the muon spectrometer. There are, however, regions of the detector where the muon spectrometer is only partially instrumented, and therefore cannot be used to identify muons. A novel approach to tagging muons in the ATLAS calorimeter called CaloMuonScore is developed based on convolutional neural networks trained on calorimeter energy deposits. The method is validated on simulation and its performance is assessed on data in a tag-and-probe analysis. Compared to the previous approach to calorimeter-based muon identification, CaloMuonScore identifies muons with up to 9% higher efficiency while rejecting background coming predominantly from pi mesons up to 20 times more effectively. The method is now commissioned within the central ATLAS software repository for use in future physics analyses.