1. Prospects on Lonely Top Quarks searches in ATLAS
Benoit Clément
LPSC – CNRS/IN2P3 – UJF Grenoble – Grenoble INP
Physics at LHC 2008 – Split, Croatia.

2. THIS IS OUR BEST GUESS AT THE MOMENT
Analysis strategy -
Goal
Determine what will be the sensitivity of the ATLAS experiment to single top production for a given luminosity, taking into account :
- Expected event statistics (S/B, S/√B …)
- Expected systematics.
Identify the critical points of the analysis and develop method to optimize the analysis strategy.
Inputs
Our current knowledge of what ATLAS data will look like
- theoretical cross sections
- detector full simulation
- trigger and reconstruction algorithm efficiencies
THIS IS OUR BEST GUESS AT THE MOMENT

3. Top quark electroweak production
Top quark production in pp
Strong interaction
pp -> gluon  top pairs (833 pb)
dominant mode.
Electroweak interaction
Wtb vertex (single top)
3 diagrams : s-channel, t-channel
And associated production Wt.
Cross section ~ |Vtb|²
Sensitivity to new physics :
H+, W’, 4th generation quarks.
Generic signature:
1 central, high-pT b-jet (top decay)
1 leptonic W (high-pT lepton + MET)
1 or 2 extra jets (including b-jets)
t-channel Wt
s-channel
σNLO = 246 ± 12 pb
σNLO = 11 ± 1 pb
σNLO = 66 ± 2 pb

4. Larger cross-section at LHC
From Tevatron to LHC
t-channel
Larger cross-section at LHC
t-channel : 2 pb -> pb (x120)
s-channel : 1 pb -> pb (x11)
Wt : 0.1 pb -> pb (x660)
Top pair : 7 pb -> pb (x120)
W+jets : ~2 nb -> ~20 nb (x10) 
s-channel
Tevatron claimed a sigma observation of (s+t channels) in 2007.
Complex multivariate analyses (NN, BDT, ME)
No discovery yet !!!  Wt 
S/B ratio similar for t-channel
S/B worse by a factor 10 for s-channel
S/B improved by a factor 6 for Wt
BUT HIGHER STATISTICS

5. Associated Wt production
Single top signatures
In all channels : 1 high pT (~70 GeV) b-jet
1 W boson
from the top quark decay
t-channel single top
1 or 2 extra forward jet (b-jet often very forward)
Leptonic decay of W only : BR(W->(τ->)e/μ) = 25.4%
-> 1 high-pT e or μ + missing ET
s-channel single top
1 second central b-jet
Leptonic decay of W only : high-pT e or μ + missing ET
Associated Wt production
1 second W boson
- lepton + jet channel (1 had W, 1 lep W) : BR = 34%
2 jets, 1 lepton, missing ET
- di-lepton channel (2 leptonic Ws) : BR = 6.4%
2 leptons, some missing ET, no extra jets
not considered yet

6. Main backgrounds
Top pairs (σ≈800pb ) in both lepton+jets and di-lepton decay.
-> Dominant background to all analyses
W + jets, associated production
W+light jets(σ≈O(100) nb)
W+bb+jets (σ≈O(0.1) nb)
Multijet events with a fake lepton
- jet identifed as an electron
- muon from heavy flavour or pion decay, reconstructed outside the jet cone. -
tt -> l + jets
Wbb

7. Event Selection Common preselection : Selecting top-like events
Single electron or muon (20 GeV threshold) trigger
1 isolated electron or muon, pT>30 GeV, |η|<2.5
no extra lepton with pT>10 GeV, |η|<2.5
missing ET > 20 GeV
1 b-tagged jet, pT > 30 GeV, |η|<2.5 (b-tag eff ~60%, rej ~ 100)
at least 1 extra jet, pT > 30 GeV, |η|<5
Kills most of the Multijet and W+jet background
Leptonic W
Top quark
The issue of multijet events
No credible simulation of mutlijet with fake isolated lepton -> need real data.
Check that Tevatron’s “triangular cut” should do the job
- cut in MET vs Δφ(Lepton, MET) plane
- kills events with fake MET from mis-id
- only suppress a few % of the signal

8. Analysis specific selection
t-channel
Harder b-jet cut : pT > 50 GeV -> against Wbb
Forward light jet (cut based analysis only): hardest untagged jet with |η|>2.5 -> against tt
Second b-tagged jet with pT > 30 GeV
Veto on extra jet with pT > 15 GeV
Topological cut (cut based analysis only)
0.5<ΔR(b1,b2)<4
80 < HT(jets) < 220 GeV (scalar sum of pTs)
60 <lepton pT < 130 GeV
s-channel Wt
Harder b-jet cut : pT > 50 GeV
Veto extra b-jet, using looser b-tag (eff~90%, rej~2) > against tt (improve S/B by a factor 3)
2 to 4 jets (incl b’s)
Hadronic W mass cut (cut based analysis only): if at least 2 light jets : <Mjj< 125 GeV

9. Number of Selected Events : 1fb-1
Signal
1460
24.8
639
Other ST
148
39.5
1418
Top pairs
2816
145.1
3022
W+jets
942
66.4
3384
Total Bkg
3906
251
7824
S/B
0.37
0.1
0.08
S/√B
23.4
1.6
7.2
t-channel
s-channel Wt
Statistical significance ok for t-channel and Wt
Poor S/B separation for s-channel and Wt
Everything gets far worse once systematics enter !!!

10. Multivariate analysis
Combining information of several (poorly) discriminating variables to improve separation
Cut on discriminant = surface in var space
Many methods exist : Likelihood, ANN, DT, Fisher…
Boosted decision trees (t-channel and Wt)
Decision Tree : At each node find the variable and the cut that gives the best separation and split the sample into 2 new nodes.
Boosting : Average over several trees obtained by reweighting wrongly separated events.
Use TMVA implementation
Likelihood ratio (s-channel)
Compute the likelihood ratio S/(S+B) assuming uncorrlated variables.
L= Π Psignal(xi) / (Π Psignal(xi) + Π Pbruit(xi) )
Pdf are derived from MC histograms

11. Multivariate analysis (2)
For each analysis, several discriminants are trained
- 1 discriminant per principal background
- different sets of discriminating variables
- split analyses in separate channels (electron/muon, jet multiplicity)
t-channel : 1 BDT against top pairs
s-channel : 5 LH against : tt->l+j, tt->tau+j, tt->ll, W+jets, t-channel
Wt : 4 BDT against : tt->lj, tt->ll, W+jets, t-channel
LH s-ch vs tt->tau+j
BDT Wt vs tt->l+j

12. Cross section error estimation
Poisson likelihood
- Bi : # of background events
- αi : the signal acceptance.
- L : integrated luminosity.
- Di : Expected data = Bi+ αi L σth
- σ : cross section to fit
Random shift of Bi, αi L : gaussian errors
Random Di following a Poisson distribution
Errors from a same source are 100% correlated between MC samples and channels
Errors from different sources remains uncorrelated
Error source D B α L
Data statistics 
MC Statistics
Luminosity
Backgrounds XS
B-tagging, JES…
Minimisation obviously gives the input cross section.
Use MC approach to estimate cross section uncertainty.
Estimates cross-section pdf -> Variance gives the error

13. XS Errors and sensitivities
Stat error
5.7%
64%
20.6%
MC statistics
7.9%
29%
15.6%
Luminosity
8.8%
31%
20%
B-tag efficiency
6.6%
44%
16%
Jet energy scale
9.9%
25%
11%
Lepton ID
0.7% 6%
2.6%
Trigger
1.7% 2%
Theory (xs, pdf,ISR/FSR…)
13.5%
74%
35%
Total 1 (10)fb-1
22% (10%)
95% (48%)
48% (19%)
t-channel
s-channel Wt
For 1 fb-1

14. Summary Single top studies from the CSC book focused on
- Separate studies of each 3 channels : s, t and associated production.
- Establishing and optimizing selections for single-top events in early data (<10fb-1), including multivariate analyses.
- Estimating the systematics uncertainties one can expect and the impact on the mesured cross-section.
5 sigma incompability with the null hypothesis (ie discovery) should be attained for:
- t-channel production with 1fb-1 (Δσ/σ = 22%)
-> leads to a mesurement of |Vtb| with Δ |Vtb|/ |Vtb| = 12%
- Wt production with 10fb-1 (Δσ/σ = 19%)
s-channel discovery will require more statistics (>30 fb-1) but some systematics will have been reduced when this luminosity is reached .
Once single top signal is established :
- Top properties in single top channels (in paticular spin correlation)
- Extension to new physiscs search (deviation in cross-section, modified kinematics) -> Charged Higgses, W’…