Search for WW/WZ&rarr l&nu jj at CDF using 1.2 fb-1 of data

Anna Sfyrla
University of Geneva

CDF/PUB/ELECTROWEAK/PUBLIC/9216


1. Introduction

A search for WW and WZ production in lepton-neutrino plus dijet final state is presented here. The contributing Feynman diagrams are shown in the following figure:



The WW and WZ production has been observed so far in the fully leptonic channel at the Tevatron. The aim of this analysis is the observation in the semi-leptonic decay channel.
The semileptonic decay channel is, from all the diboson decay channels, the favorable for Trilinear Gauge Coupling (TGC) measurements, due to the big branching ratio with respect to the leptonic channel.

Signal Selection

The event selection associated with a leptonic W consists of exactly one tight lepton with transverse momentum > 20 GeV, missing transverse energy > 25 GeV and lepton-missing energy transverse mass > 30 GeV. The hadronically decaying W is reconstructed by the two leading jets of events with two or more jets. The dijet invariant mass window considered in the analysis is [45,160] GeV.
The theoretical cross sections for this decay mode are
&sigmaWW × Br(W &rarr l &nu, W &rarr jj) = 12.4 pb × 0.146 = 1.81 pb
&sigmaWZ × Br(W &rarr l &nu, Z &rarr jj) = 3.96 pb × 0.07 = 0.28 pb


Backgrounds

There are several processes that result in the same final state topology (1 lepton, electron or muon, missing transverse energy and jets) as the diboson production, and thus are backgrounds to this search. The background processes that are taken into account are:
The largest of these backgrounds is the W(&rarr l &nu)+jets. The high theoretical cross section of this process (&sim 300 pb) leads by itself, but even more with the rest of the backgrounds, to a very poor Signal over Background ratio (S/B < 1% initially).



2. Improved Event Selection

Neural Network

In order to improve the initially very poor S/B, a tool with big discriminative power is needed, and as such a Neural Network (NN) has been used, and more precisely the ROOT interface for JetNet.
The NN has been trained with 6 input variables. The input variables, that are listed in the following picture (red curves for signal and black for background), are angle- or event shape- related variables. That way, the NN output is kept as uncorrelated as possible with the dijet invariant mass, that is the variable on which the analysis is based, since it provides a good description of the signal. The NN is trained using MC events for the signal and background description and uses events in the signal region only.



The NN output (picture below) is used to perform a cut at the point of the maximum significance gain. In this case, the significance is defined as
Significance = S/&radic(S+B)


The maximun significance gain point, is the point where a cut would maximize the significance. This is at approximately 0.46 at the NN output.



Taking into account the events that correspond in the signal region of the NN output (with the cut as defined with the significance curve), we get an overall significance gain of &sim 10%. The gain in the signal fraction is &sim 50%. Plotting the invariant mass distribution we get good agreement between the MC and the data in the sidebands. Good agreement between data and MC comes also in the NN output, using information from the sidebands (following pictures).





3. Signal and Background Parameterizations

The parametrization of the signal and background shapes after the NN cut will be used in a likelihood fit that will be applied on data to estimate the signal fraction. The signal shape is fixed using MC, and the background parameterization is motivated by MC. Both are plugged into a likelihood fitter, that has three free parameters, two related to the background parameterization and the third is the signal fraction. The background parameters and the signal fraction are given by a likelihood fit to data.
The likelihood fit to data is shown in the following plot. The red curve shows the signal plus background description, while the blue curve shows the background only.



The background subtracted from the data is shown in the following plot. The red curve shows the signal shape for the signal fraction measured on the data (using the fitter) and the red band the statistical error of the signal fraction.






4. Results and Summary

We presented a search for WW plus WZ production in lepton-neutrino plus dijet final state, using 1.2 fb-1 of data and central tight electrons and muons in the lepton selection. A NN has been used for the significance optimization.

The theoretical prediction for the cross section is

&sigma × Br = 2.09 ± 0.14 pb

We measured

NSignal = 410 ± 212(stat) ± 107(sys) signal events

that correspond to a cross section

&sigma × Br = 1.47 ± 0.77(stat) ± 0.38(sys) pb

The 95%CL upper limit to the cross section is estimated to be

&sigma × Br < 2.88 pb









Anna Sfyrla
December 2007.