TRACKING THE INVISIBLE. A SEARCH FOR SUPERSYMMETRIC HIGGSINOS WITH THE ATLAS DETECTOR.

This thesis presents two complementary searches for supersymmetric particles with compressed mass spectra using proton-proton collision data collected by the ATLAS detector during the LHC Run 2 and Run 3. The analyses target the production of charginos and neutralinos near the electroweak scale and with mass splittings in the range $0.3-1.5$ GeV. Owing to the small mass differences, the heavier neutralino and chargino are relatively long-lived and decay into low-momentum charged pions, which are reconstructed as soft tracks displaced by a few millimeters from the proton-proton interaction point and characterised by unusually large impact parameters. The resulting final-state signature is defined by the presence of an energetic jet, missing transverse momentum, and one or more than one soft, mildly displaced tracks. The first search uses data collected at a center-of-mass energy of $\sqrt{s} = 13$ TeV, corresponding to a total integrated luminosity of $140 \ \text{fb}^{-1}$. Dedicated event-level and track-level selections are applied to define two Signal Regions maximising the overall signal-to-background ratio. No significant data excess above the Standard Model expectations is observed, and 95\% confidence level exclusion limits are set on the chargino and neutralino masses as a function of their mass splitting. In the mass splitting range $0.35-1$ GeV, chargino masses are excluded up to about 170 GeV, significantly extending the reach beyond the legacy LEP results for the first time. The second search uses data collected at center-of-mass energies of $\sqrt{s} = 13$ and $\sqrt{s} = 13.6$ TeV, corresponding to a total integrated luminosity of $299 \ \text{fb}^{-1}$. In this second analysis, an event-level binary neural network and a track-level parameterised multi-class neural network are developed and employed in place of the cut-based approach used in the first analysis. The enhanced discriminating power provided by the machine learning techniques, combined with the increased statistics of the Run 3 data set, is expected to further extend the LHC sensitivity into regions of parameter space that are anticipated to remain inaccessible to future direct detection Dark Matter experiments.