Dark Matter and How To Find It: A search for low-mass leptophobic Dark Matter mediators and the development of mass-decorrelated jet taggers with the ATLAS experiment

A search for low-mass leptophobic Dark Matter (DM) mediator particles in $36~\mathrm{fb}^{-1}$ of $pp$ collision data at $\sqrt{s} = 13~\mathrm{TeV}$ collected by the ATLAS experiment is presented. The search is performed in final states where the mediator decay into a quark pair is reconstructed as a single, large-radius jet produced in association with a photon or a jet. No deviations from the Standard Model expectation are observed, and limits are placed on the production cross-section of leptophobic mediator particles and their coupling to quarks for mediator masses between $100$ and $220~\mathrm{GeV}$. At the time of publication, this result constituted the lowest limits on leptophobic DM mediator masses for high-mass DM particles reported by ATLAS. Adversarial neural networks (ANN) are presented as a way to train jet taggers which decorrelates them from the invariant mass of the jet. An extensive study of five different approaches to constructing mass-decorrelated jet taggers is presented. The ANN tagger is found to provide the largest QCD multijet rejection at similar levels of mass-decorrelation.