Evidence for Higgs boson decays to a low mass lepton pair and a photon with the ATLAS detector at the LHC

This thesis presents the first evidence of Higgs boson decay to two leptons and a photon. Rare Higgs boson decays predicted in the Standard Model (SM) presently come within experimental reach thanks to the large amount of data collected by the experiments at the Large Hadron Collider (LHC) at CERN. Given a central role played by the Higgs boson in the SM, studying its rare decays opens up a possibility for more stringent tests of the SM. A very rare decay of Higgs boson to a low mass electron or muon pair and a photon is explored in this thesis using $139~\text{fb}^{-1}$ of $pp$ collision data at a center-of-mass energy $\sqrt{s}=13~\text{TeV}$ recorded with the ATLAS detector at the LHC during Run 2 (2015-2018). In the phase space with lepton pair invariant mass $m_{ll} < 30~\text{GeV}$ ($l=e,\mu$) the expected SM branching ratio is $\mathcal{B}(H\rightarrow\ell\ell\gamma)\approx 10^{-4}$. The low branching ratio presents an experimental challenge, with the expected number of background events vastly exceeding the number of signal $H\rightarrow\ell\ell\gamma$ events. Another experimental challenge of the analysis is that in the low lepton pair mass regime the two leptons tend to be highly collimated. For the electrons, their energy deposits in the calorimeter of the ATLAS detector often remain unresolved by the standard reconstruction algorithms. To overcome this, a new identification algorithm is used, based on multivariate discriminant. The new algorithm more than doubles the number of selected $ee\gamma$ events compared to the standard reconstruction algorithms. In addition, collimated leptons often do not satisfy standard criteria on additional activity in their vicinity (isolation), which is imposed to reduce backgrounds. Isolation criteria are therefore corrected taking into account contribution of leptons to each other's isolation. A combined statistical model is constructed using parametric functions describing signal and background $m_{ll\gamma}$ distributions in each analysis category. The observed signal yield is extracted from the fit of this model to data. An excess is observed over the background-only hypothesis with a significance of 3.2 standard deviations. The best-fit ratio of the observed event yield to the SM expectation is $1.5\pm0.5$. The fiducial cross-section times branching ratio in the $m_{ll} < 30~\text{GeV}$ region is measured at $\sigma(H)\times\mathcal{B}(H\rightarrow\ell\ell\gamma)= 8.7^{+2.8}_{-2.7}~\text{fb}$. The analysis is still limited by the small number of expected events and the systematic uncertainties constitute only 35% of the statistical uncertainty. Among systematic uncertainties, the uncertainty associated with a bias in background function choice is dominating. The impact of systematic uncertainties on the results is studied in detail and the results are additionally verified using pseudo-experiments.