A new neural network to select the b-jet pair from H -> bb in HH -> bbtautau searches at CMS

This thesis presents the search for non-resonant double Higgs boson (HH) production via gluon-gluon fusion (GGF) and vector boson fusion (VBF) mechanisms. The HH production gives access to two fundamental parameters of the SM: the Higgs boson trilinear self-coupling (λHHH), and the coupling between two Higgs bosons and two vector bosons (λ2V), via the measurement of the GGF and VBF cross sections. These parameters are sensitive to the presence of physics beyond the Standard Model. The investigated final state has one of the Higgs bosons decaying into two b-quarks and the other decaying into two τ leptons (HH → b¯bτ+τ−). This process is studied through the examination of the three decay modes of the τ+τ− system: τhτh, where τh means a τ which decays into hadrons plus a ντ , and τℓτℓ (with ℓ denoting either an electron or a muon). The search uses data from proton-proton collisions at a centerof- mass energy of √s = 13TeV recorded with the Compact Muon Solenoid (CMS) detector at the LHC, corresponding to an integrated luminosity of 137.1 fb−1. The new techniques employed in the analysis allowed for a major improvement of the signal sensitivity compared to the previously published CMS results. Among these new techniques, a novel neural network, HH-Btag, is used to improve the selection of the b-jet pair forming the HH→b¯b with respect to the previous methods. HH-Btag has been designed and optimized for the searches of resonant and non-resonant production of the HH in the aforementioned final state. The performance of the algorithm with respect to their predecessors are evaluated under different hypotheses of mass and of spin in the case of resonant production, and as a function of couplings modifiers in the case of non-resonant production, in order to cover a large phase space of the possible scenario. The internal CMS collaboration review of the analysis started in April 2021 and is still ongoing, thus the statistical analysis of the results is carried out without the inclusion of the observed data (blind analysis), and only expected results from standard model (SM) predictions are reported. The 95% Confidence Level upper limit on the total (GGF + VBF) cross section times branching fraction expected from SM is 10.88 fb (equivalent to 4.55 times the SM value). The study of the VBF process, done for the first time in this analysis, resulted in an exclusion limit on the production cross section expected via VBF of 237.3 fb (137.5 times the SM value).