Searching for Dark Matter in the Light of Dark-Higgs Strahlung: A first search for a resonant di-Higgs + $E_T^{\text{miss}}$ signature with the ATLAS experiment

This thesis summarises the research conducted by the author on the first search for a new particle decaying in a di-Higgs signature with a significant missing momentum component. The work presented utilises the 140$\text{fb}^{-1}$, $\sqrt{s}=13$TeV proton-proton collision dataset collected with the ATLAS detector. The analysis targets new physics in the di-Higgs decay into four-b channel and exploits the unique kinematic features of this resonant decay topology. In particular, the search is optimised for a scenario in which a hypothetical heavy $Z^{\prime}$ mediator is produced, radiating a dark-Higgs boson before decaying invisibly into dark matter particles. The dark-Higgs boson then decays into two Standard Model Higgs bosons: $pp \rightarrow Z^{\prime *} \rightarrow Z^{\prime} (\rightarrow \chi\chi) s (→hh (\rightarrow b\bar{b}b\bar{b}$)). The corresponding model is the Two-Mediator Dark-Higgs Model. Optimised kinematic selections, together with the innovative use of machine learning algorithms, are employed to achieve high sensitivities. Measures are explored to ensure that the machine learning classifier does not distort the reconstructed di-Higgs mass distribution, thus preserving its physical interpretability in the fit. As no significant excesses are observed, limits on the dark-Higgs model are presented in the dark-Higgs and $Z^{\prime}$ mass plane. Dark-Higgs masses are excluded between 250 and 405GeV for $Z^{\prime}$ masses up to 2400GeV in an benchmark model with a dark matter mass of 200GeV. These derived limits significantly extend the current best constraints on this model, particularly in the high dark-Higgs and high $Z^{\prime}$ mass regions. Additionally, this thesis presents a calibration analysis for a boosted double-b tagger, developed to identify two collimated b-jets originating from the decay of a heavy resonance. Scale factors are derived to correct for any differences in the mis-tagging efficiencies between simulation and data, using simulated dijet events enriched in $g \rightarrow b\bar{b}$ splittings.