Tau Reconstruction Meeting

Friday 6 Nov 2026, 12:00 → 13:00 Europe/Zurich

Laurits Tani (National Institute of Chemical Physics and Biophysics (EE)), Michel Hernandez Villanueva (Brookhaven National Laboratory (US))

FCC Tau Reconstruction - November 6, 2026

Present: Laurits Tani, Norman Seeba, Federico Melis, Alberto Lusiani, Attilio Andreazza, Federico Melis, Karl Ehataht, Mohadeseh Dadashzadeh, Ignacio Diaz, Olmo Arquero, Maria Cepeda, Michel Villanueva

ML based Z->tautau event tagger

• From DELPHI, di-fermion tagging
  • Use hemisphere level selection cuts for BR measurements
• Starting with an early DELPHI tau polarization paper
  • At DELPHI, acceptance limited to barrel; explored full FCC acceptance (IDEA, CLD)
  • Efficiency in acceptance region of 87%
  • Background 0.8% (qq), 0.4% (two-photon)
• Used Winter 2023, IDEA fast simulation, CLD full simulation
• No generated samples available for two-photon background events (ee --> ee ff)
• Using the base selection from DELPHI, using two leading tracks divided by hemisphere
• Distributions of charged track multiplicity, thrust, visible energy, normalized visible energy, acollinearity and isolation angle
  • 2D distributions of radiated momentum (definition on slides)
• Comparing cut-based di-tau selection vs Multi-layer perceptron network vs Particle transformer starting from the list of particles in the event
• MLP shows 99.9% efficiency vs 88.45% cut-based in fast sim; similar performance in full simulation
• Degradation of transformer in full simulation vs fast sim

Questions/comments

• We should investigate which generators are available for two photon, and put the request for MC production. Some options to explore:
  • TREPS - https://arxiv.org/abs/1310.0157
  • AAFH - https://doi.org/10.1016/0550-3213(85)90541-3
  • The blocker may be the interface with k4Gen
• What happen when you add Bhabha? - No centralized samples, but Olmo has a small sample
• What is the ranking of the variables and the correlation?
• What is the 99.9% efficiency respect to what?
  • Within the acceptance
• P_rad definition may be incorrect
• Plot the ROC curve for MLP and compare with the point of cut-based
• Have you considered looking at the LEP data with key4hep?
  • If time is available, yes
• Is ISR included in the simulation
  • We have to look at the Pythia configuration, but probably no
  • KKMC samples were requested, they should be used for tau studies

12:00 → 12:10 Introduction

Speakers: Laurits Tani (National Institute of Chemical Physics and Biophysics (EE)), Michel Hernandez Villanueva (Brookhaven National Laboratory (US))

12:10 → 12:30 ML based Z->tautau event tagger

Speakers: Prof. Attilio Andreazza (Università degli Studi e INFN Milano (IT)), Federico Melis (Università degli Studi e INFN Milano (IT))

20261106_DiFermionTagger_v2.pdf

20261106_DiFermionTagger_v2.pptx

12:30 → 13:00 Round table