Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 24: Standard Model Physics II

T 24.5: Vortrag

Dienstag, 17. März 2026, 17:15–17:30, KH 00.014

Machine Learning Based Tagging of Vector Boson Polarization — •Lena Alshut, Erik Bachmann, Mareen Hoppe, and Frank Siegert — Institut für Kern- und Teilchenphysik, Technische Universität Dresden, Germany

The longitudinal polarization of massive vector bosons (MVBs) is a direct consequence of electroweak symmetry breaking and serves as a sensitive probe of the Higgs mechanism and potential new physics. Because longitudinally polarized MVBs are rare, previous studies at the LHC have relied on neural-network classifiers trained to distinguish polarization states. However, this formulation depends on an nonphysical quantities with unknown distribution and assumes classes of purely longitudinal or transverse polarized MVBs.

In this work, the alternative strategy of polarization fraction regression is explored. These fractions at phase-space point level are physical observables with well-defined distributions. This enables a direct comparisons of model predictions and MC truth values, providing the opportunity for systematic improvement of polarization tagging.

We develop a Machine Learning-based tagger that predicts event level polarization fractions using the four-momenta of final-state particles. A MLP and a Transformer are trained on Standard Model expectations. In this talk, we present results for Z+jets in the leptonic decay channel at LO, LO+PS, and higher perturbative orders as a exemplary test case. The approach can directly be translated to the analysis of LHC pp colission data, where it allows for a more physical interpretation of polarization measurements.

Keywords: Vector Boson Polarization; Polarization Tagger; Machine Learning