Reconstruction, Identification, and Calibration of hadronically decaying tau leptons with the ATLAS detector for the LHC Run 3 and reprocessed Run 2 data

This document describes the reconstruction algorithms employed by the ATLAS experiment for Run 3 of the LHC to reconstruct, identify, and calibrate hadronically decaying $\tau$ leptons. The same algorithms are also used for the reprocessed data from Run 2 of the LHC. All algorithms have been improved and optimised with respect to the ones employed during Run 2 data-taking. The energy calibration relies on Monte Carlo simulation samples with hadronic tau lepton decays, and applies a Boosted Regression Tree based on calorimetric energy and tracking momentum measurements. The tau reconstruction and identification algorithms, as well as the electron rejection algorithm, employ Recurrent Neural Networks. The decay mode classification algorithm, based on DeepSet Neural Networks, is used to discriminate among the five main decay topologies. The performance of the deployed algorithms is assessed using Monte Carlo simulations.