Search for the Standard Model Z Boson Production in Association with W Boson using 4.3 fb-1 (CDFnote 10044)

Justin Keung, Evelyn Thomson
University of Pennsylvania
Chris Neu
University of Virginia
Tom Schwarz
University of California - Davis

Documentation

Public Conference Note


Send mail to authors

RomaNN Efficiency and Mistag Rate Measurements

Abstract

We present a search for Standard Model (SM) Z boson produced in association with a W boson, where the Z boson decays to a pair of b quarks and the W boson decays into a charged lepton and a neutrino. This search uses data through period 23, corresponding to an integrated luminosity of 4.3 fb-1. We select WZ->lvbb candidate events with two jets, MET, and exactly one CEM, CMUP, or CMX lepton candidate. The two jets are identfied to be coming from bottom quarks using the RomaNN algorithm. Discrimination between the WZ signal and the comparatively large backgrounds in the W + 2 jet bin is improved through the use of a neural network. We see no evidence for a WZ->lvbb signal, and set a 95% confidence level upper limit on the WZ cross section times the branching ratio of the Z->bb and W->lv. This is achieved by performing a fit of the neural network output distributions for signal and backgrounds. The resulting 95% C.L. limit is 3.9xSM (3.9xSM expected).



Event Selection
Lepton Central isolated electron or muon, > 20 GeV, |η| < 1.1
Missing Et > 20 GeV
Two Jets > 20 GeV, |η| < 2.0, both UltraLoose RomaNN tagged


Results

(click to enlarge)


Sensitivity improvements in WZ over WH(120GeV) limit, showing improvements expected and presently obtained.
  • Acceptance: the average jet ET is lower for WZ than for WH, resulting in fewer jets making the > 20 GeV requirement
  • b-tagging: b-tag efficiency increases with jet ET, the net result is a lower b-tag efficiency for WZ than for WH
  • Dijet mass background: the dijet mass is the single strongest discriminating variable, accounting for more than 90% of the sensitivity. The lower mass of the Z also places its reconstructed mass in a kinematic region where there are more backgrounds
  • Dijet mass resolution: the CDF detector is able to reconstruct the mass of the parent particle with a resolution error proportional to the reconstructed mass itself, the lower mass of the Z makes the dijet mass resolution better
  • WZ->lvcc: Z decays into charm quark pairs (BF=12%) almost as frequently as bottom quark pairs (BF=15%), but the tagging efficiency for charm quarks is an order of magnitude lower than bottom quarks. The Higgs however decays into charm quark pairs much less frequent (BF=0.3%) than bottom quark pairs (BF=70%)
  • Advanced techniques not included: a similar search for WH also using Neural Networks as the final discriminant, has additional improvements, such as loose leptons and forward leptons giving 45% more signal acceptance, and neural network b-jet specific energy correction improving dijet mass resolution by 35%

Table of the expected and observed number of events as a function of jet multiplicity for events with at least two RomaNN "tag", with no Ht requirements and MET > 20GeV, using the ttbar calibrated scalefactor.

Systematic uncertainty on the signal acceptance from several different sources.

Output distribution of the neural network discriminant for data measurement and of the full sample composition prediction. WZ and WW/ZZ are added last to the stacked prediction.

Output distribution of the neural network discriminant for data measurement and of the full sample composition prediction in log scale. WZ and WW/ZZ are added first to the stacked prediction.

Expected and observed 95% C.L. limit obtained using the neural network discriminant.

The distributions of -2lnQ in simulated S+B and B-only pseudo-experiments, assuming a Standard Model WZ signal. The value of -2lnQ observed in the data is -0.43, indicated with a black arrow, corresponds to a P-Value of 0.40.


Neural Networks Validation

(click to enlarge)


Output distribution for test output samples of MC events, of signal WZ and each of the background component, normalized to unit area.

Output distribution of the neural network discriminant using single tag control sample for data measurement and of the full sample composition prediction. WZ and WW/ZZ are added last to the stacked prediction.

Output distribution of the neural network discriminant using single tag control sample for data measurement and of the full sample composition prediction in log scale. WZ and WW/ZZ are added first to the stacked prediction.

Distribution of the Neural Network Input: Dijet invariant mass. This is the invariant mass reconstructed from the two jets.

Distribution of the Neural Network Input: MET. This is the magnitude of the missing transverse energy.

Distribution of the Neural Network Input: PTImbalance. The scalar sum of the lepton and jet transverse momenta minus the MET.

Distribution of the Neural Network Input using single tag control sample: Dijet invariant mass. This is the invariant mass reconstructed from the two jets.

Distribution of the Neural Network Input using single tag control sample: MET. This is the magnitude of the missing transverse energy.

Distribution of the Neural Network Input using single tag control sample: PTImbalance. The scalar sum of the lepton and jet transverse momenta minus the MET.


b-jet Identification Efficiency Calibration using Top Quark Pair Production Cross Section

(click to enlarge)


Expected and observed number of events as a function of jet multiplicity for events with at least one RomaNN "tag", with the requirements that the events Ht > 250 GeV, and MET > 25 GeV, using the b-tag scalefactors obtained from dijet events using PTrel method.

Expected and observed number of events as a function of jet multiplicity for events with at least one RomaNN "tag", with the requirements that the events Ht > 250 GeV, and MET > 25 GeV, with b-tag scalefactors obtained from dijet events using PTrel method multiplied by 1.155 +- 0.065, such that the top quark pair production cross section is measured to be 7.5 +- 0.4pb using the 3/4/>=5 jet bins. The b-tag scalefactors obtained from dijet events using PTrel method multiplied by 1.155 +- 0.065 is called the ttbar calibrated scalefactor.

Expected and observed number of events as a function of jet multiplicity for events with at least two RomaNN "tag", with no Ht requirements and MET > 20GeV, using the ttbar calibrated scalefactor. The top quark pair production cross section is measured here to be 7.6 +- 0.4pb using the 3/4/>=5 jet bins.
Last Update on June 15th, 2010 keungj@fnal.gov