Energy measurement of hadron showers in the CMS HGCAL using Graph Neural Networks and a new physics search with photons at the CMS experiment

The CMS experiment at the LHC has been rigorously testing the Standard Model inproton-proton collision data for more than a decade now. As the LHC progressesthrough Run-3 at a center-of-mass energy of 13.6 TeV, the CMS collaboration is alsopreparing for the High-Luminosity LHC (HL-LHC) phase to operate with moreintense beams in the upcoming decade. To meet the challenges of high radiationdoses and more particle flux, current CMS endcap calorimeters will be replaced bythe High-Granularity Calorimeter (HGCAL), a state-of-the-art detector that featuressix million read-out channels capable of providing highly detailed 5-dimensionalmeasurements: energy, position, and timing. These capabilities open new frontiers inthe reconstruction of particle showers with unprecedented resolution.In this thesis, a novel dynamic reduction network (DRN) based on graphical neuralnetworks (GNN) is used to significantly improve energy resolution of charged pionsin the HGCAL. The improvements from the algorithm are validated on negativelycharged pion data collected with the HGCAL beam test experiments at the CERNfacilities. Further insight into the additional information learned by the GNN isobtained using GEANT4 based detector simulation. As the HGCAL approaches itspreproduction phase, it is imperative to evaluate the efficacy of the finalized design,with a particular focus on the readout components, through both system tests andbeam tests to guarantee operational integrity and seamless data transmission withoutloss. This thesis presents key results from software based studies of the HGCALreadout electronics and data handling and from the beam test of complete near-finalHGCAL readout system during the August-September 2023 experiments at the CERNH4 beam test facility.I also present a new methodology for searching for new physics produced via strongand electroweak supersymmetry scenarios in Run 2 data using soft photons as aprobe. Lowering the photon momentum threshold increases the background tenfoldand reduces signal sensitivity when using the jet multiplicity and MET as primarycriteria based on previous searches. By reevaluating events with photon momentumand the scalar sum of all jets and photon momentum, signal sensitivity improvesacross strong and electroweakino productions.