Göttingen 2025 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 83: Methods in Particle Physics IV (Lepton Reconstruction)

T 83.7: Vortrag

Donnerstag, 3. April 2025, 17:45–18:00, VG 4.101

Inference of the Neutral Four-Momentum of Hadronic τ-Leptons using Neural Networks in ATLAS — •Simon Thiele¹, Lukas Cieslik¹, Christian Grefe¹, Alessandra Betti², Philip Bechtle¹, and Klaus Desch¹ — ¹Rheinische Friedrich-Wilhelms Universität Bonn — ²Sapienza Università di Roma

Reconstructing the four-momenta of neutral decay products of hadronically decaying τ-leptons, which are almost exclusively π⁰’s, allows to infer the spin of the τ. This allows for example to measure the CP of the Higgs boson. Therefore it is desirable to reconstruct this momentum as accurately as possible, which is challenging since the photons from the π⁰ decays are only measured in the electromagnetic calorimeter.

Currently these neutral decay products are reconstructed in ATLAS using the Tau-Particle-Flow algorithm, which also performs a decay mode identification, classifying the tau jets by the number of charged and neutral hadrons they contain. In recent years a new neural network based decay mode classifier has been developed. This new classifier has a higher efficiency than the current algorithm. But since it is only a classifier without a reconstruction of the neutral four-momentum, this gain in efficiency is not accessible to these Higgs CP studies. Therefore we are currently working on developing a neural network based solution for that also provides inference of the neutral four-momentum.

In this talk I will first go over this motivation and the current state of the art algorithms and then discuss the performance of the new neural network solution.

Keywords: Neural Network; Machine Learning; Tau Reconstruction; ATLAS