TOP/ANTI-TOP CROSS SECTION MEASUREMENT IN THE ALL HADRONIC CHANNEL
---- with Neural Net Selection ----
***** 1020 pb-1 *****

Authors (cdf-allhad6@fnal.gov)
L. Brigliadori, A. Castro, F. Margaroli (Univ. of Bologna)
P. Azzi (Univ. of Padova)
G. Lungu, A. Sukhanov (Univ. of Florida)
A. Gresele, I. Lazzizzera (Univ. of Trento)


Dataset
    We are using for the data the multi-jet sample (1020pb-1) which is divided into 3 time portions corresponding to about 350, 410 and 260 pb-1 respectively, during which the detector and accelerator conditions are somehow different. For the signal we are using mainly Monte Carlo tt events generated with Pythia and with the top mass at 175 GeV. Other Monte Carlo samples generated with Pythia and Herwig under different conditions are used to study the systematic uncertainties on the efficiency.


Pre-requisites
    For a minimal clean-up of the samples, we apply the follow pre-requisites:
    A) number of tight lepton = 0 (to guarantee the orthogonality with lepton+jets analysis);
    B) |Zjvert| < 60 cm;
    C) |Zjvert - Zpvert| < 5 cm;
    D) N(vert_class12) >= 1
    E) MET_significance < 3 (to guarantee the orthogonality with tau+jets analysis).


Topology
    F) 6 <= Njet <= 8;
    G) DeltaRmin >=0.5;

Neural Net Kinematical selection
11 Input variables+ 2 hidden nodes (20 and 10 nodes):
    1) SumEt; 2) SumEt3; 3) Aplanarity 4) Centrality; 5) M2jMin; 6) M2jMax; 7) M3jMin; 8) M3jMax; 9) Et1Star(=Et*sin^2\theta^*); 10) Et2Star; 11) <EtStar>3N(geometric average over the 3rd-4th...Nth jets);



Neural Net Plots
NN output test samples (380k events) (eps)
NN output whole samples (sigma_tt=6.7 pb) (eps)
NN output variables 1-4 (sigma_tt=6.7 pb) (eps)
NN output variables 5-8 (sigma_tt=6.7 pb) (eps)
NN output variables 9-11 (sigma_tt=6.7 pb) (eps)


Neural Net selection efficiency
    Requiring NNoutput>=0.94 (see optimization later), we estimate the selection efficiency (mt=175):
    epsilon_NN = 4.8 +- 0.8 %;

Table 1, Systematics on NN selection
Source relative Systematic (%).
Generation/Fragmentation (PYT vs HRW) 1.1
JES (PYT +1sigma vs PYT-1sigma) 16.3
ISR/FSR modeling (more vs less ISR/FSR) 2.9
Multiple Interactions (efficiency vs Nvertices) 2.5
PDF 1.4
NN training (efficiency on not-test sample) 0.1
TOTAL 17.0

Average number of tagged jets
    Degrading tags (scale factor SF=0.89) to reproduce the efficiency seen in data, we estimate the average number of tagged jets:
    N_ave_tag = 0.95 +- 0.07;


Data-driven background with a 3-dim Matrix
    This matrix is made starting from a sample with only 4 jets and using:
    1) 6 bins in Number of primary vertices
    2) 13 bins in Number of good tracks inside jets (with silicon hits)
    3) 5 bins in Et
    The agreement is good when we apply this matrix on the sample before the neural net selection.
NN output 4-jet events (eps)
NN output 5-jet events (eps)
NN output 6to8-jet events (eps)


Systematics on data-driven background
    Two sources of systematics:
    1) apply the 4-jet matrix on events with more jets,
Table 2, background systematics for different jet multiplicity
dataset 1st portion 2nd portion 3rd portion
(Nexp+tt-Nobs)/Nexp % +0.8+-0.9 -0.6+-0.8 -0.8+-1.0
    2) apply the 4-jet matrix on events with large values of NNoutput, like in a region NNoutput>~0.94:
      (Nexp - Nobs)/Nexp % - first portion of data (eps)
(Nexp - Nobs)/Nexp % - second portion of data (eps)
(Nexp - Nobs)/Nexp % - third portion of data (eps)

Table 3, background systematics for NN>~0.94
dataset 1st portion 2nd portion 3rd portion
(Nexp+tt-Nobs)/Nexp % 2.1 2.3 3.2
    bck systematics = 2.5% (summing both sources in quadrature and taking the integrated luminosity-weighted average;

Neural net selection optimization
    We search the NN cut which provides the smallest (relative) total uncertainty on the signal, i.e. we account for both statistical and systematic uncertainties on signal and background and look for the maximum of S/Delta(S+B): S/Delta(S+B) vs NN output cut (eps)
The maximum is reached for NNoutput>=0.94.

Cross section Inputs
    The inputs for the cross section are:

Table 4
Input Value
Luminosity (1020 +- 60) pb-1
Neural net efficiency (6<= nJ<=8) (4.8 +- 0.8) %
Average number of tags (6<= nJ<=8) (0.95 +- 0.07)
Pre-tag Events (6<= nJ<=8) 4205
Tagged Jets (6<= nJ<=8) 1233
Background (6<= nJ<=8) (937 +- 30)
Background corrected (846 +- 37)

The 1.02 fb-1 cross section measurement
    The 1.02 fb-1 ttbar cross section measured in the all hadronic channel is:
    sigma_tt = 8.3 +- 1.0 (stat) +- 0.5 (lum) +2.0 - 1.5 (syst) = 8.3 +2.3 -1.9 pb


Tags vs NNout
Tags vs NNout (eps)
Tags vs NNout (eps)


Tags vs Jets (NNout>=0.94)
Tags vs Njet (eps)
Tags vs Njet (eps)
Table 5
Njet 4 5 6 7 8 6-8
Nevts 118657 16157 2575 1069 561 4205
NEXP 16060+-575 2750+-92 536+-17 255+-8 146+-5 937+-30
NEXPcorr 15961+-677 2653+-112 481+-20 223+-10 142+-7 846+-37
Ntt(8.3pb) 120+-20 266+-45 242+-41 101+-17 38+-7 381+-65
NEXPcorr+Ntt(8.3pb) 16081+-677 2919+-121 723+-46 324+-20 180+-10 1227+-76
NOBS 16555 3139 725 349 159 1233
Note: the uncertainties are the quadrature sum of statistical and systematic uncertainties; the statistical uncertainties on the background for different jet multiplicities are correlated so they are summed linearly to be conservative.