Measurement of the ttbar production cross section
in the MET+jets channel in 2.2 fb-1

Gabriele Compostella, INFN-CNAF and University of Padova
Silvia Amerio, Donatella Lucchesi, University of Padova
[Contact Authors]


Abstract

We study 2.2 fb-1 of data events collected by CDF with a multjet trigger that contain jets identified in the Silicon VerteX detector as originating from b-quarks.
Rather than relying on lepton identification requirements, we use significant missing transverse energy and a neural network with input variables related to event and jets energy and topology to discriminate the top quark pairs production from other background processes. The overall amount of b-jets produced by background processes in the selected sample is estimated using a parameterization of b-jet identification rates measured directly from data.
Assuming a top quark mass of 172.5 GeV/c2 we obtain a top pair production cross section measurement of

σttbar = 7.99 ± 0.55 (stat) ± 0.76 (syst) ± 0.46 (lumi) pb

Analysis Description

The data sample used in this analysis was collected with an optimized multijet trigger. The event cleanup selection requires significant MET in the event (METSigf ≥ 3 GeV1/2) and vetoes events with well identified electrons or muons.
Events passing prerequisites and having at least 4 jets are then used to train a Neural Network. Since data events passing cleanup cuts have a negligible contamination from top signal of about 3.5%, they are used as a representation of background during the training process. Pythia Monte Carlo events generated with a top quark mass of 172.5 GeV/c2 and passing prerequisites are used as a representation of top signal during Neural Network training. As inputs the Neural Network uses variables related to event and jets energy and topology. The result of the training is shown in the following plot:

Output of the Neural Network training

Background prediction method

The analysis uses a data driven background prediction. The main idea is that the probability for a jet to be identified as a b-jet is different for the ttbar pairs (our signal) and for the concurring background processes. This difference can be used to distinguish the two components in the collected data.
The secondary vertex tagging probability for background processes is calculated using data events depleted of signal (i.e. data events passing analysis prerequistes and having 3 jets, with top contamination lower than 0.1%) and parameterized as a function of the jet transverse energy, jet number of tracks and the missing transverse energy projection along the jet direction.

b-jet tagging rate as a
function of Jet Et
b-jet tagging rate as a
function of Jet number of tracks
b-jet tagging rate as a
function of missing energy projection along the jet direction

We can then apply the tagging rate parametrization in the form of a 3-dimensional matrix, and use it to estimate the number of positive tags in events with 3 and more jets. As a check of the method, we can compare the number of observed positive tags with the matrix predicted one as a function of the number of jets in the event and of the Neural Network output. The following plots show the number of observed and matrix predicted number of positive tagged jets along with the expected contribution coming from inclusive top signal normalized to our measured cross section of 7.99 pb.

Observed and matrix predicted number of positive tagged
jets by jet multiplicity in the multijet data after prerequisites
Observed and matrix predicted number of positive tagged
jets by Neural Network output in the multijet data
with at least 4 jets after prerequisites

Neural Network selection optimization

The cut on the output of the Neural Network used to isolate top decays from the multijet background has been optimized by accounting for the expected positive tagged jets from Monte Carlo and the number of background tags provided by the matrix application to data. We scanned all cuts in the range 0.6-1.0 and by minimizing the expected relative statistical error on the cross section measurement we choose the cut for Neural Network output values grater than 0.8. The effect of the cut on the amount of observed and matrix predicted tags is shown in the figure along with the expected contribution from inclusive top signal, normalized to our measured cross section of 7.99 pb.

Observed and matrix predicted number of positive tagged
jets by jet multiplicity in the multijet data after cut on Neural Network output greater than 0.8

Systematic Uncertainties

The summary of all the systematics sources of uncertainty is listed in the following table:

Summary of systematic uncertainties

Sample Composition

The following table shows the sample composition after cut on Neural Network output greater than 0.8

Sample composition after cut on Neural Network output greater than 0.8

Additional blessed material

The following plots show the observed and matrix predicted number of positive tagged jets by Neural Network output in the multijet data after prerequisites for events with 3 jets, 4 jets and 5 jets.

Observed and matrix predicted number of positive tagged jets
by Neural Network output for events after prerequisites with 3 jets
Observed and matrix predicted number of positive tagged jets
by Neural Network output for events after prerequisites with 4 jets
Observed and matrix predicted number of positive tagged jets
by Neural Network output for events after prerequisites with 5 jets

The following table shows the efficiency of the different prerequisites on multijet data and on Monte Carlo simulated inclusive top signal:

Prerequisites cuts

The following table shows the top contamination in data after prerequisites by event jet multiplicity:

Top contamination after prerequisites

The following table shows the input values for the cross section determination:

Input values for the cross section measurement

This measurement became a public CDF result on September 24, 2009.
Page last modified: Oct 6 2009