A Deep Neural Network capable of discriminating between jets coming from the decay of longitudinally and transversely polarized W or Z bosons with a large Lorentz boost

The amount of data that will come available from the future High Luminosity Large Hadron Collider enables precision studies to look for deviations from Standard Model expectations. New Physics mainly couples to longitudinally polarized vector bosons, leading to energy enhanced New Physics effects in the longitudinal channel. As the transverse channel dominates the Standard Model cross section, it becomes an irreducible background. For future studies therefore, the discrimination power between transversely and longitudinally polarized weak gauge bosons will play an important role. In this project note a study of separation power between the polarization states is presented. Two deep neural networks have been used to create a polarization tag by using jet information coming from hadronic decays of weak gauge bosons with a large lorentz boost. This has been done in a pT independent manner and such that it covers weak gauge bosons originating from decays of new heavy resonances with masses up to 4 TeV. The performance of this tag is quantified using the area under the curve of the receiver operating characteristic curve. Using only the angular emission distribution in the rest frame, a value of AUC = 0.64 can be considered as base value. On Monte Carlo for reconstructed level, the highest obtained AUC value was found to be 0.70.