Pseudo-observables and Deep Neural Network for mixed CP -- H to tau tau decays at LHC

TL;DR

This paper explores the use of deep neural networks and pseudo-observables to analyze Higgs to tau tau decays at the LHC, aiming to improve CP property measurements through machine learning techniques.

Contribution

It introduces a machine learning framework to predict CP-sensitive pseudo-observables and the CP mixing angle from decay product kinematics, enhancing analysis methods for Higgs CP properties.

Findings

ML models can predict CP-sensitive parameters from tau decay data

Predicted distributions can be used to measure Higgs CP mixing angle

Method extends previous studies with detailed feature analysis

Abstract

The consecutive steps of H to tau tau cascade can be useful for the measurement of Higgs couplings and parity. The analysis methos of ATLAS and CMS Collaborations was to fit a one-dimensional distribution of the phi* angle, phi*, which is sensitive to transverse spin correlations and, hence, to the CP mixing angle, phi^CP. Machine Learning techniques (ML) offer opportunities to manage complex multidimensional signatures. The 4-momenta of the tau decay products can be used as input to the machine learning and to predict the CP-sensitive pseudo-observables and/or provide discrimination between different CP hypotheses. We show that the classification or regression methods can be used to train an ML model to predict the spin weight sensitive to the CP state of the decaying Higgs boson, parameters of the functional form of the spin weight, or the most preferred CP mixing angle of the analysed sample. The one-dimensional distribution of the predicted spin weight or the most preferred CP mixing angle of the experimental data can be examined further, with the statistical methods, to derive the measurement of the CP mixing state of the Higgs signal events. This paper extends our previous studies, with more details for the features and how proposed pseudo-observables can be used in the measurement.