1. Tau-pair analysis for LoI+
Taikan Suehara
ICEPP, The Univ. of Tokyo

2. Tau-pair process [Observables] P(e-)=80%, P(e+)=30%, 500 fb-1
Difficulty on decay analysis
σ=2600 fb-1 (e-Le+R)　σ=2000 fb-1 (e-Re+L)
radiative events: ~70%
[Observables] P(e-)=80%, P(e+)=30%, 500 fb-1
σ, AFB　(bg suppression)
Polarization P(t) ↑Decay angle determination

3. Progress
Looser tau-selection cuts – to improve statistical error (compatible with SiD).
More background of Bhabha and gg-tt
Better decay-mode selection by a neural network
‘Optimal observable’ for polarization measurement

4. Background events SM background of the mass production
2-photon and Bhabha have low statistics.
Bhabha – re-preselection
Compatible with looser cut |cos(q)| < 0.95
|cos(q)|<0.96, opening angle < 15 deg
~200k events for 1 fb-1
2 photons – tautau
Preselection cuts: Opening angle < 10 deg, Evis > 30 GeV
~150k events for about 10 fb-1

5. Tau selection cuts Signal increase: ~20%
Evis cut changed: 40 to 70 GeV
Background level: almost the same
Results are still worse than SiD about 20%...
Might be difference on tau-clustering: they accept neutral clusters

6. Mode separation – 1 prong
Need to separate leptonic, pinu, rhonu and a1nu
Neural net tried
Variables (9 params)
Ecalo/Etrack (muon ID)
EECAL/(EECAL + EHCAL) (electron ID)
Echarged , Eneutral ,En3 (Third-largest photon energy),Nn
Mall, Mn w/neutral hadrons, Mn wo/neutral hadrons
18-10 hidden neurons, 5 output neurons
Selected among 16-10, 16, 10-5, 10 (before adding Mn w/n)
Double layers give much better results
1000 epochs, a half of tau-pair ( events): ~5-10 hours

7. Result of mode separation – 1p
Better than SiD!
ILD
SiD

8. Mode separation – 3 prong
Need to separate a1nu. Neural net tried
Variables (8 params)
Ecalo/Etrack (muon ID)
EECAL/(EECAL + EHCAL) (electron ID)
Echarged , Eneutral
Number of neutral particles
Invariant mass of all visible decay daughters
Invariant mass of charged particles
Invariant mass of neutral particles
10 hidden neurons, 1 output neurons, single layer
Not optimized…

9. Result of mode separation – 3p
ILD
SiD
Need to improve?

10. Rho optimal observable2 2

11. Omega distributions
ω = (PeL(ω) - PeR(ω)) / (PeL(ω) + PeR(ω)) No P dependence at ω=0, L(R) only at ω=±1
Electron channel
Muon channel
Pion channel
Rhonu channel

12. Polarization by ω
P(eL) = ± P(eR) = ± (a1 not included)

13. Issues & prospects Tau selection – slightly worse than SiD
NN tuning for 3-prong events
a1 (very complicated formula)
Tau direction can be used for a1
But need to calculate ω by ourselves
Finalizing analysis and write a paper
Analysis note updated: