Measurement of the ttbar Cross Section in the Lepton + Jets Channel 

Using Neural Network in 2.8 fb-1 of CDF Data

Including Ratio Over the Z Cross Section

J. Conway, R. Erbacher, J. Johnson, A. Lister, T. Schwaz,
UC Davis

K. Lannon
University of Notre Dame

R. Hughes, B. Winer
The Ohio State University

CDF public conference note (updated since ICHEP)

Link to public web-page for result shown at ICHEP 2008

NN Fit

ttbar Cross Section Results for 2.8 fb-1

Note: All results assume Mttbar = 175 GeV/c2

Using Event Kinematics and Ratio over the Z cross section

W + >= 3 jets
σ ttbar 6.89 ± 0.41(stat) +0.41-0.37(sys) ± 0.14 (theory) pb

Previous Result: Using Event Kinematics Only
W + >= 3 jets
σttbar = 7.08 ± 0.38 (stat) ± 0.36 (syst) ± 0.41 (lumi) pb

For information: Result shown at ICHEP and used in the CDF combination
σttbar =  6.80 ± 0.38 (stat) ± 0.61 (syst) ± 0.39 (lumi) pb

W + >= 4 jets
σttbar = 6.81 ± 0.39 (stat) ± 1.01 (syst) ± 0.39 (lumi) pb

For information: Result shown at ICHEP
σttbar =  6.50 ± 0.39 (stat) ± 1.02 (syst) ± 0.38 (lumi) pb


We present a measurement of the top pair production cross section  in ppbar collisions at 1.96 TeV, with an integrated luminosity of  2.8 fb-1 at the CDF experiment on the Fermilab Tevatron. We use a neural network technique to discriminate between top pair production and background processes in a sample of events with an isolated, energetic lepton, large missing transverse energy and three or more energetic jets. We measure a top pair production cross section of 
σttbar =  7.08 ± 0.38 (stat) ± 0.36 (sys) ± 0.41 (lumi) pb for a top mass of 175 GeV/c2
We then significantly reduce the dependence on the luminosity measurement and its associated large systematic uncertainty. We compute the ratio of the ttbar to Z cross section, measured using the same triggers and dataset, and then multiplying this ratio by the theoretical Z cross section. The final ttbar cross section, assuming a top mass of 175 GeV/c2,  is measured to be
σttbar = 6.89 ± 0.41(stat) +0.41-0.37(sys) ± 0.14 (theory) pb.
The total uncertainty is 8.2%, greatly surpassing the Tevatron Run II goal of 10%, and now as precise as the best theoretical calculations.
For this analysis the signal to background ratio is about 1:4.5 after final event selection.

Event Selection

We use the standard CDF lepton + jets selection
  • Central Lepton (electron or muon) pT >= 20 GeV/c;
  • Missing transeverse energy >= 20 GeV;
  • At least 3 jets with ET >= 20 GeV. 
 We apply additional cuts
  • Tigher missing transeverse energy cut at 35 GeV;
  • Tighter leading jet ET cut at 35 GeV.
The tighter cuts were optimised to remove the most QCD background while mainting reasonable efficiency for the ttbar signal. As the statistical uncertainty is not our dominant uncertainty, we can affort to cut quite hard. These cuts remove ~85% of the QCD, ~40% of the W+jets while maintining ~80% efficiency for the ttbar signal.

Both central electrons and muons have been used for the final fit.
Note: We do not require any b-tagging.

Signal and Background Modeling

  • The ttbar signal is modeled from Pythia Monte Carlo with an assumed mass of 175 GeV/c2.
  • The W+jets background is used to model all EWK backgrounds, as the previous analysis showed that the effect of including all backgrounds was very small and such a difference is included as a source of systematic uncertainty.
  • The W+jets background us modeled from ALPGEN+Pythia MC. The W+jets sample is obtained from a combination of W+0p, W+1p, W+2p, W+3p exclusive as well as W+4p inclusive
  • The QCD shape is modeled from a sample of dijet events which pass our event selection criteria. These events are dominated by jets that fake electrons. The QCD contamination in the electron sample is significantly larger than in the muon sample

Neural Network Input Variables

The Neural Network uses 7 kinematic distributions as inputs, with 1 hidden node.
The NN is trained to separate W+4p from ttbar Monte Carlo.
The input variables are
Global event variables
  • Σjets ET of all jets excluding two leading jets
  • HT of the event (sum of transverse energy of all reconstucted objects)
  • Aplanarity of the event
Variables related to the 3 leading jets
  • Σjets pz / Σjets ET
  • Mininum dijet mass
  • Mininum dijet separation
  • Maximum jet ET

Ratio over the Z Cross Section

The dominant systematic uncertainty, 5.8%, is is the luminosity measurement due to the detector used to measure the inelastic ppbar cross section.
We can almost entirely cancel this uncertainty by considering the ratio of the ttbar to the Z cross sections.
The Z cross section is measured using the central electron and muon triggers and the same data sample as the Z cross section.
For this measurement, the signal MC for ttbar and Z is re-weighted to the CTEQ6.6 PDF (with its associated uncertainties).
The meausred ttbar cross section is found to be
σttbar = 6.97 +0.42-0.41 (stat) +0.40-0.42 (sys) ± 0.40 (lumi) pb.
The Z cross section is measured to be
σz =  253.27 ± style="font-weight: bold;"> 1.01(stat) +4.4-4.6 (sys) +16.63-13.71 (lumi) pb.
The ratio of the ttbar to Z cross section is computed, taking into account the correlations between the systematics
1/R = σZttbar =  36.47 +2.06-2.29 (stat) +1.88-1.96(sys).
Multiplying the ratio R by the theoretical Z cross section
σZtheory = 251.3 ± 5.0 (sys) pb,
We get a final result for the measured top pair production cross section of
σttbar = 6.89 ± 0.41 (stat)+0.41-0.37 (sys)± 0.14 (theory) pb.
The total uncertainty is 8.2%, a significant reduction on the 9.2% obtained using the kinematic fit only.
The total uncertainty is decreased by 10% by taking the ratio of the top pair to the Z cross sections!!!

Final Fits to NN Output Distribution

Fits for the 2.8 fb-1 data sample follow bellow.

The input QCD normalisation is obtained from a fit to the missing ET distribution before the missing ET cut is applied.
The uncertainty on the constraint is set to 50% in the final fit.
The ttbar and W+jets shapes float freely in the fit.  


NN Output distribution showing the signal and background contributions obtained from the fit to data.
The fit returns 1273 reconstructed ttbar event (after selection)

NNfit >=4 jets
NN Output distribution for the W+>= 4 jets case.

Plots of NN Input Variables

These plots are shown using the signal and background normalisations obtained from the final fit.
Sum Et jets 3,4,5
Σ ET of jets excluding 2 leading jets

Ht HT of the event

Sum Pz / Sum ET Σ pz / Σ ET of 3 leading jets

Aplanarity Aplanarity of the Event

minimum dijet separation Minimum dijet separation of 3 leading jets

minimum dijet mass Minimum dijet mass of 3 leading jets

Maximum jet eta Maximum jet eta of 3 leading jets

Last modified on 10/21/08 by Alison Lister