Generically Orthogonal Decompositions of Collision Events and Measurement Combinations in Standard Model $VH\left(b\bar{b}\right)$ Searches with the ATLAS Detector

Abstract

This thesis describes variations on the two lepton channel of the Run 2 search for the SM Higgs boson produced in association with a vector boson using different variable sets for multivariate analysis (MVA) training. The three variable sets in question are the set of variables from the fiducial analysis, a set based on the Lorentz Invariants (LI) concept, and a set based on a combination of masses and decay angles derived using the RestFrames (RF) package. Aside from the variable sets used for MVA training and discirminant distributions, the analysis is otherwise identical to the fiducial analysis. Both the LI and RF sets perform competitively on the basis of significances, with the RF set showing a ~3.5% improvement in expected fits to Asimov and data, though neither set boosts observed significance. Both sets also reduce the observed error on the observed signal strength, with the LI set reducing the error due to systematics by 7.5% and the RF set doing so by 16%. The issue of combining multiple results from different channels and datasets is also examined through the combination of the fiducial Run 1 and Run 2 ATLAS $VH\left(b\bar{b}\right)$ results, which results in an observed signal strength of $0.90^{+0.18}{-0.18}(\textrm{stat.})^{+0.21}{-0.19}(\textrm{syst.})$ and an observed (expected) significance of 3.6 (4.0) standard deviations, the first ever evidence of this process.