Speaker
Description
Precise modelling of hadronically decaying tau leptons is critical for many ATLAS measurements and searches, particularly in the top quark, electroweak, and Higgs sectors. The higher centre of mass energy and increased pile up conditions of Run~3 require updated, data driven calibrations of tau reconstruction and identification performance over an extended kinematic regime.
This talk presents a measurement of tau identification performance and data to simulation scale factors using proton proton collisions at $\sqrt{s}=13.6$ TeV recorded by ATLAS during 2022 to 2024. The study targets the GNtau algorithm, a graph neural network based identifier that exploits tracking, calorimeter, and substructure information to enhance discrimination between genuine hadronic taus and jets.
The measurement is performed in $t\bar{t}$ events with opposite sign $\tau_{\mathrm{lep}}\tau_{\mathrm{had}}$ final states, providing a high purity tau sample and coverage across transverse momenta from $20$ to $220$ GeV. Identification efficiencies are extracted using a simultaneous template likelihood fit in pass and fail regions, constraining signal and jet backgrounds in situ. Scale factors are derived as functions of $p_{\mathrm{T}}$, decay mode (1 prong and 3 prong), and identification working point within a unified fit combining the full 2022 to 2024 dataset to maximise statistical precision and ensure stability across detector and running conditions.
These results constitute a key component of the ATLAS Run-3 tau calibration programme and provide essential inputs for analyses with hadronic taus, while illustrating the performance gains enabled by modern graph neural network techniques.