Abstract: This dissertation presents a search for massive, narrow-width resonances decaying to pairs of Higgs bosons in the bb̄ττ final state, where one Higgs boson decays into a pair of bottom quarks and the other into a pair of tau leptons (X → HH → bb̄ττ). Such resonances are predicted by beyond-the-standard-model theories, which aim to address the shortcomings of our current understanding of fundamental particles and their interactions. The search uses proton–proton collision data at a center-of-mass energy of 13 TeV recorded by the Compact Muon Solenoid (CMS) experiment during 2016–2018, corresponding to an integrated luminosity of 138 fb⁻¹, and targets resonances in the mass range of 1–4.5 TeV. The analysis uses a single large jet to reconstruct the H → bb̄ decay, while the H → ττ decay products can either be contained within a single large jet or appear as two isolated tau leptons. The reconstruction and identification of physics objects are enhanced using advanced machine learning techniques, including a graph convolutional neural network for merged bb̄ jets and a convolutional neural network specifically designed for this search to identify merged ττ decays. Upper limits at the 95% confidence level are set on the production cross section for resonant HH production in the mass range considered, with this analysis providing the most sensitive limits to date on X → HH → bb̄ττ decays for masses above 1.4 TeV.
The second component of the thesis describes the commissioning and validation of graphics processing unit (GPU)-based reconstruction at the CMS high-level trigger for Run 3 data-taking. To address increasing computational demands arising from higher instantaneous luminosity and greater event complexity, reconstruction algorithms for the hadron calorimeter, electromagnetic calorimeter, and pixel tracker were offloaded to GPUs to take advantage of parallel processing wherever feasible. Dedicated physics validation was required to ensure that the GPU-offloaded algorithms produce physics results consistent with the central processing unit (CPU)-based reconstruction. The final trigger configuration seamlessly utilizes GPU hardware when available while maintaining backward compatibility with CPU-only configuration, establishing a foundation for meeting the computational challenges of the high-luminosity LHC era.