General Search for beyond the Standard Model physics in 13 TeV $pp$ collisions with the ATLAS detector

The high-energy collision data gathered by the LHC experiments allow many hypothetical models of beyond the Standard Model physics to be tested. Signals of beyond the Standard Model physics might, however, remain hidden from direct searches motivated by theoretical models. This thesis describes a strategy employed by the ATLAS Collaboration to systematically search for deviations of the collider data from the Standard Model prediction. The strategy used is one of sequential hypothesis testing on two independent datasets. Initially, many background-only tests are performed in orthogonal data selections using Monte Carlo simulated background estimates. A significant deviation, in a given data selection, between the event counts observed in the experiment and expected from the simulated background declares that data selection a \emph{data-derived signal region}. A data-derived signal region is then analysed in a dedicated analysis for which the background model can be optimised further, e.g. by using data-driven techniques. This strategy is first applied to 3.2 fb$^{-1}$ of proton--proton collision data at a centre-of-mass energy of 13 TeV collected with the ATLAS detector at the LHC in 2015. Collision events are classified according to their final state into 750 mutually exclusive event classes. For each event class, an automated search algorithm tests whether the data are compatible with the Monte Carlo simulated expectation in several distributions sensitive to the effects of new physics. The significance of a deviation is quantified using pseudo-experiments. Nearly $10^6$ regions have been analysed, however, no significant deviations have been found, and consequently, no data-derived signal regions for a follow-up analysis are defined. Following this, the strategy is applied to a larger dataset of 36.2 fb$^{-1}$ collected in 2015 and 2016. Several distributions of more than 1,000 mutually exclusive event classes are tested in a preliminary version of the analysis. Additionally, the data are tested using fewer but more inclusive event classes enhancing the sensitivity to possible signals that might otherwise be distributed over many classes. This is supplemented with the reconstruction of $Z$ and Higgs bosons in typical decay channels. Due to the incomplete background model used for this preliminary analysis, regions can only be eliminated from data-derived-signal-region candidacy. The majority of regions are ruled out for candidacy, however, several regions cannot be ruled out. These regions will need to be re-examined with a complete background model for those regions to either rule them out from candidacy or confirm the region as a data-derived signal region.