Measured WZ Diboson Cross Section with 7.1/fb at CDF

Author: Jason Nett
Blessed: 9 June 2011
Data: CDF Run II 7.1/fb in events with three high-pt leptons

Abstract
Event Count
Summary of Signal Region

Basic Kinematics
Neural Network Input Variables
Neural Network Score

Summary of the Low Missing E_T Control Region

Basic Kinematics
Neural Network Input Variables
Neural Network Score

Summary of the Z-Removed Control Region

Basic Kinematics
Neural Network Input Variables
Neural Network Score

Abstract

WZ production is a crucial background to the H→WW search in the associated production mode decaying to a three-lepton signature plus missing E_T signature. We are updating the Standard Model WZ cross section measurement to that found with 7.1 fb^-1 from the previous measurement using 7.1 fb^-1 in the three-lepton plus missing E_T signature. A NeuroBayes neural network is used to distinguish WZ signal from background processes in the final selection region. The WZ cross section is then extracted using a maximum likelihood method which best fits the neural network score distribution to the data. This analysis measures the WZ cross section to be 3.9+0.8-0.7 picobarns (or 3.9+0.6-0.5 (stat.) +0.6-0.4 (syst.)) which is in agreement with the 3.46±0.21 picobarns predicted by theory. For details, see the public note for this measurement.

Event Count

Event count for the signal and control regions after all cuts are made.

Requirements For All Regions

Exactly three leptons (i.e. electrons or muons, or if one of the three leptons is a tau and the tau then decays leptonically
Leading lepton p_T>20.0 GeV
Second and third lepton p_T>10.0 GeV
Any number of jets are acceptable
Trilepton + track rejection to remove events with a 2nd Z-boson.

The ZZ background contamination in the WZ trilepton signal region is the most pernicious, so in the current iteration of this analysis we employ a new cut. Much of the ZZ to lll+MET signature is actually physical ZZ to llll where one of the leptons fails to pass lepton identification, and so that lost lepton appears as missing energy in the final reconstruction. Therefore, we now search for and reject events that have a trilepton + track signature, provided that extra track has Pt > 8.0 GeV. The `DimassLepTrkForZZVeto' figure below illustrates which events are vetoed from the signal region by this new cut. Forming the dimass of the lepton that is not part of the original Z pair and the extra track illustrates how this signature is dominated by ZZ and not WZ. Two of the three data events in the trilepton+track signature clearly fall on the Z-mass, suggesting a second Z-boson for those events. Meanwhile, the Monte Carlo contributions illustrate that the trilepton + track signature is dominated by ZZ with a second Z-mass peak. The signal region is the bin at -1 in the `DimassLepTrkForZZVeto' figure below; everything greater than zero has a 4th track and is cut out. This results in a 36% reduction of the ZZ background while leaving the WZ signal untouched (only 0.2 events lost out of ~47).

Additional Requirements For Signal Region

Missing E_T>25.0 GeV
Z-Selection: Each event has three leptons and so three possible pairings of leptons. If an event has at least one lepton pairing passing the following criteria, then it is accepted into the signal region

opposite electric charge
same lepton flavor
a dilepton invariant mass within [ 76.0, 106.0] GeV

Additional Requirements For Low Missing E_T Control Region

Missing E_T<20.0 GeV
Z-Selection: Each event has three leptons and so three possible pairings of leptons. If an event has at least one lepton pairing passing the following criteria, then it is accepted into the low missing E_T control region

opposite electric charge
same lepton flavor
a dilepton invariant mass within [ 76.0, 106.0] GeV

Additional Requirements For Z-Removed Control Region

No E_T cut
Z-Rejection: Each event has three leptons and so three possible pairings of leptons. If an event has at least one lepton pairing passing the following criteria, then it is accepted into the Z-peak removed control region

opposite electric charge
same lepton flavor
a dilepton invariant mass NOT within [ 66.0, 116.0] GeV

Summary of Signal Region

Basic Kinematic Distributions

Having a proton-antiproton interaction produce three high p_T leptons is a very rare occurence compared the total number of events produced. As such, the statistical sample we have to draw upon after all of the analysis cuts defined above have been implemented is very small. The basic kinematic distributions illustrate good agreement between data and monte carlo predictions, nevertheless.

Leading lepton p_T

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

2nd lepton p_T

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

3rd lepton p_T

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Missing E_T

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Number of Jets

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Reconstructed Z-boson Mass

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Reconstructed Transverse Mass of the W-boson

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Lepton Flavor Combinations of the three leptons

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Neural Network Input Distributions

Once we have a set of data (and monte carlo) passing all analysis cuts, we could simply calculate the measured cross section based on the total count of the data minus the total count of the expected background (according to monte carlo simulation), but several techniques have been developed to more optimally distinguish the signal from background and achieve a better result from the same amount of collected data.

One such technique is a neural network, of which this analysis employs the NeuroBayes incarnation. A neural network is a computational tool that inputs several distributions that show separated, though usually overlapping, signal and background curves. It then outputs a single distribution that ideally will exhibit a larger separation between the signal and background curves than any one of the inputs does.

Some neural net input distributions are more effective at discriminating between the signal and background curves than other. The following is a list of the input variables used in this analysis, ranked from most to least significant as reported by NeuroBayes:

Δφ(W-Lepton, Missing E_T) -- The azimuthal angle between the W-Lepton and Missing E_T
m_T (jets) -- The transverse mass of the jets. Events that have zero jets are assigned a value of zero.
Lepton Type Combinations -- Discriminate among events that have eee (all three leptons are electrons), eeμ, eμμ, etc.
Missing E_T
m_T(W-Lepton, Missing E_T)
Δφ(2nd lepton,Missing E_T)
H_T--The sum of transverse components of all leptons, jets, and missing E_T
m_T(all leptons, jets, and missing E_T)
m_T(3rd lepton,missing E_T)
Δφ(vector sum of the three leptons,missing E_T)
m_T(the three leptons)
Number of jets

Some of these were already shown in the basic kinematics section above. Here are the rest:

Δφ(W-Lepton, Missing E_T)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

m_T (jets)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Δφ(2nd lepton,Missing E_T)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

H_T

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

m_T(all leptons, jets, and missing E_T)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

m_T(3rd lepton,missing E_T)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Δφ(vector sum of the three leptons,missing E_T)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

m_T(the three leptons)

[*.gif] [*Log.gif] [*.eps] [*Log.eps]

Neural Network Score

Here is the neural network score--the output of the neural network from which a likelihood function fit method is used to extract the WZ measured cross section.