[Paper Review] Tau Reconstruction and Identification Performance at ATLAS

B. Gosdzik|DESY Publication Database (PUBDB) (Deutsches Elektronen-Synchrotron)|Sep 30, 2010

Particle physics theoretical and experimental studies1 references3 citations

TL;DR

This paper evaluates the performance of ATLAS tau reconstruction and identification algorithms using 2010 LHC data at √s = 7 TeV. It presents cut-based, boosted decision tree, and likelihood-based identification methods, demonstrating excellent agreement between data and Monte Carlo simulations, with background rejection efficiencies of ~1.6% for tight selection and good separation power for tau leptons from QCD jets.

ABSTRACT

For many signals in the Standard Model including the Higgs boson, and for new physics like Supersymmetry, $τ$ leptons represent an important signature. This work shows the performance of the ATLAS $τ$ reconstruction and identification algorithms. It will present a set of studies based on data taken in 2010 at a center-of-mass energy of $\sqrt{s}$ = 7 TeV. We measured some of the basic input quantities used for these identification methods from selected reconstructed $τ$ candidates and compared the results to the prediction of different Monte Carlo simulation models. For early data taking a cut-based identification method will be used. We also measured the background efficiency for the cut-based $τ$ identification.

Motivation & Objective

• To evaluate the performance of ATLAS tau reconstruction and identification algorithms using early LHC data.
• To assess the effectiveness of cut-based, boosted decision tree (BDT), and projective likelihood (LL) identification methods in distinguishing tau leptons from QCD jets.
• To measure background rejection efficiency for tau identification using data and compare it to Monte Carlo predictions.
• To quantify systematic uncertainties from transverse momentum calibration and pile-up effects on identification performance.
• To validate that key identification variables are well described by Monte Carlo simulations for subsequent physics analyses.

Proposed method

• Tau reconstruction uses track-seeded and calorimeter-seeded candidates with thresholds: pT > 6 GeV, |η| < 2.5, and |d0| < 2 mm for tracks; E_T > 10 GeV and |η| < 2.5 for calorimeter clusters.
• Seven input variables, including EM radius and track radius, are used to distinguish tau leptons from QCD jets, with definitions based on transverse energy and momentum within specific ΔR regions.
• Cut-based identification uses three variables: R_EM, R_track, and f_trk,1, optimized for signal efficiencies of 30% (tight), 50% (medium), and 60% (loose).
• BDT and likelihood-based identification methods are trained and validated using QCD jet Monte Carlo samples generated with the Pythia DW tune (tev4lhc).
• Background efficiency is measured in data and compared to MC predictions, with normalization to observed tau candidate counts.
• Systematic uncertainties are evaluated via comparison of two pT calibration schemes (GCW vs. EM+JES) and dependence on vertex multiplicity (n_vtx) as a proxy for pile-up.

Experimental results

Research questions

• RQ1 How well do the ATLAS tau identification algorithms perform in rejecting QCD jets in early LHC data at √s = 7 TeV?
• RQ2 To what extent do Monte Carlo simulations accurately predict the distributions of key identification variables like EM radius and track radius?
• RQ3 What are the background rejection efficiencies for cut-based tau identification at tight, medium, and loose working points?
• RQ4 How do systematic uncertainties from pT calibration and pile-up affect the performance of tau identification algorithms?
• RQ5 Do the BDT and likelihood-based identification methods show consistent performance between data and Monte Carlo simulations?

Key findings

• The cut-based tau identification achieves a background efficiency of (1.6 ± 0.3) × 10⁻² in data for the tight selection, with MC prediction of 1.9 × 10⁻².
• For the loose selection, the background efficiency is (3.2 ± 0.2) × 10⁻¹ in data, closely matching the MC prediction of 3.4 × 10⁻¹.
• The medium selection yields a background efficiency of (9.5 ± 1.0) × 10⁻² in data, consistent with the MC prediction of 9.9 × 10⁻².
• The BDT and likelihood score distributions in data show very good agreement with QCD jet Monte Carlo predictions, indicating reliable modeling of identification variables.
• Systematic uncertainties from pT calibration are small, with the ratio of background efficiencies between GCW and EM+JES schemes varying by less than 10% across pT ranges.
• Background efficiency increases with vertex multiplicity (n_vtx), indicating a measurable pile-up effect, but the overall performance remains stable across pile-up conditions.