Speaker
Description
Jet tagging is an essential tool for physics analyses involving heavy particles decaying into hadronic final states. In the boosted regime, where bosons have high transverse momentum, their hadronic decay products are reconstructed as a single large-R jet, and identified using jet substructure and constituent-based taggers. Solutions to the classification of hadronic decays of W and Z are well known and, historically, exploited primarily in searches for beyond the Standard Model physics, while it has recently been employed in Standard Model measurements. However, current approaches generally do not exploit polarisation information. In this regime, the angular separation of the decay quarks, which encodes polarisation, is not directly accessible due to the collimated topology. This study explores the development of a boson polarisation-aware tagger using UFO large-R jets reconstructed with the ATLAS detector. A solution is proposed by exploiting jet constituents as inputs to novel machine learning methods, such as Particle Transformers. First, the longitudinal-versus-transverse (L vs T) classification task is investigated; a comparison of the performance at detector level and particle level is presented. A regression approach is then introduced to overcome current experimental limitations in polarisation studies at reconstruction level in the boosted regime and provides, for the first time, experimental observables directly sensitive to boson polarisation in hadronic decay channels. The tagger is trained on a large simulated sample of W-, Z-, and quark/gluon-initiated jets produced with Pythia and reconstructed with the ATLAS detector simulation. Performance studies for both the classification and regression tasks are presented.