Analysis of the Boosted Higgs Boson in the bb Decay Channel Using the ATLAS Detector at CERN

Abstract

The study of the Higgs boson at high transverse momentum (pT) offers a powerful probe for the exploration of potential new physics beyond the Standard Model. This thesis contributes to an inclusive, boosted H->bb analysis conducted by the ATLAS Collaboration at CERN, using data from Run 2 and part of Run 3, corresponding to an integrated luminosity of 300 fb^(-1). The objective is to measure the signal strength in inclusive, fiducial, and differential pT regions, aiming to identify possible deviations from the predicted Standard Model values. Improving the uncertainties of the results in comparison to the previous published analysis using only Run 2 data, is of great importance. The work presented in this thesis, mainly focuses on four tasks. First, the b-tagging efficiency is investigated using Monte Carlo samples. The optimal tagging algorithm, which is based on transformer neural networks, is selected. Then, a method using the relative transverse momentum of muons to the jet axis, as a potential discriminant quantity in order to evaluate the bb content of the QCD background before and after flavour tagging, is developed. The results prove that indeed this quantity (muon pTrel), is appropriate for our goal. Furthermore, crucial for the signal extraction strengths, is the resolution of the mass distribution plots on Monte Carlo samples. Hence a detailed study of the resolution across different Working Points of b tagging and across two different methods of object reconstruction (PFlow and mass after regression -bJR), is performed. One important factor that deteriorates the resolution is the energy lost by muons produced in semileptonic b decays. To mitigate this effect, the Muon-In-Jet correction is developed and applied. Two methods were considered and the optimum using Variable Radius (VR) jets is selected. An optimization, regarding the muon pT cut was conducted and the correction rates between data and Monte Carlo were tested. Based on the results, the correction is applied only in the subleading signal region. Finally, statistical fits were performed on 10 fb^(-1) of 2022 data, before and after the correction. The μZ signal strength was extracted, and a 15% reduction in its statistical uncertainty was observed. Similar fits are performed to different pT bins and to additional datasets from other years.