The identification of hadronically decaying W and Z bosons is based on algorithms that use machine learning and analyse the substructure of jets (distribution of energy within the jet). For training, Monte Carlo simulations are used that contain mainly longitudinally polarised vector bosons as signal and so-called working points with certain signal efficiencies, e.g. the algorithm correctly identifies 80% of the signal, are defined. We know that in Standard Model processes not only longitudinally polarised vector bosons occur but also a high proportion of transversely polarised vector bosons, the precise fraction depends on the process. The substructure of jets depends on the polarisation of the particles, since the decay products are emitted either along/opposite the trajectory of the decaying particle or perpendicular to it. This means that the algorithms developed so far have not been well optimised for transversely polarised vector bosons and thus the sensitivity in precision measurements decreases.

In this project, Monte Carlo simulations of transversely and longitudinally polarised vector bosons are analysed and the most important substructure variables are compared. Machine learning can then be used to train new algorithms that identify transversely polarised vector bosons. Ideally, it would also be possible to study how efficiently an algorithm could distinguish between the different types of polarisation.

Back