We present the first observation in hadronic collisions of the electroweak production of vector boson pairs (VV, V=W,Z) where one boson decays to a dijet final state . The data correspond to 3.5 inverse femtobarns of integrated luminosity of pppbar collisions at sqrt(s)=1.96 TeV collected by the CDFII detector at the Fermilab Tevatron.

We observe  1516±239(stat)±144(syst) diboson candidate events and measure a cross section σ(ppbar→VV+X) of 18±2.8(stat)±2.4(syst)±1.1(lumi) pb, in agreement with expectations from the standard model

Trigger and dataset

The VV production and decay into hadronic final states are topologically similar to the VH production and decay which is the most promising Higgs discovery channel at low Higgs mass. Also, study of the diboson production is sensitive to extra gauge couplings not present in the Standard Model.

We start with a suite of MET based triggers and measure the trigger efficiency based on Z→μμ data. The latter is a good standard candle for MET processes because muons are minimum ionizing particles that deposit little energy in the calorimeter so from the calorimeter standpoint it looks like real MET is present in the event. Of course muonic decays of Z are easily identifiable with the muon chambers. We will do a dijet mass fit so we want to ensure the trigger has little turnon effect above our low mass boundary which is 40GeV.


There are two main classes of backgrounds we consider:

  1. 1. Electroweak (EWK): these are V+jet processes and top production processes.

  2. 2. Multijet: these are generic QCD jet production which can result in MET through mismeasurements of jets

The first one we derive from MC (combination of pythia and alpgen). The different contributions are shown here:

Although the normalization of this background is allowed to completely float in the final fit, there is still the question of how well we understand the shape.

For this we use γ+jets data since there are similarities in the kinematics of this process and the V+jets process. In order to account for any differences in kinematics between γ+jets and V+jets we correct the γ+jets data based on the ratio of V+jets MC and γ+jets MC invariant mass distributions. This way, any production difference is taken into account, however, detector effects, PDF uncertainties, ISR/FSR, etc. cancel when using γ+jets data. After we apply this correction to the γ+jets data there is little difference between this and our V+jets MC and this difference determines our systematic uncertainty on the shape of the V+jets background shape.

For the multijet background, we first try to reject as much of the background as possible and then we use a data driven approach for the remaining contribution. The rejection is done mainly with the MET significance and Δφ (the angle between MET and the closest jet above 5GeV). These two quantities in a control W+jet sample with identified electrons is shown below. One can see the multijet contribution at low values of MET significance and Δφ:

Once we cut away most of the multijet background we define another variable sensitive to this, the angle between the MET and trkMET (the missing track PT as measured in the tracking chamber). For real MET, like MET from neutrinos the two vectors will be aligned. However, for fake MET, the trkMET will not necessarily point to the same direction as MET. We can define our multijet enhanced region by selecting events for which the angle between MET and trkMET is greater than 1. The EWK MC is normalized to the peak region and corrected based on Z→μμ data.

Subtracting the EWK MC from the data in the >1 region gives us an estimate of the remaining multijet background and the shape of this background in any variable of interest, like dijet mass which we will then use in the fit. Of course, one has to account for any multijet contribution in the peak and any potential difference between the >1 and <1 regions in shape. The multijet background shape is then fit with an exponential and we assigned 20% uncertainty in our total estimate of this background and 20% uncertainty on its shape (the slope of the exponential). Both these uncertainties stem from our poor knowledge of the multijet contribution in the  <1 region and they are derived by looking at the differences between <1 and >1 region in a dijet MC sample.

The last template in the fit is the signal. This is a combination of WW, WZ and ZZ from MC weighted by the appropriate SM cross sections. The dijet mass is fit with a gaussian and a polynomial and the mean and the width of this gaussian are dependent on the jet energy scale. The latter is constrained to be in the range allowed by the external measurements.


After all cuts are applied we find 44910 diboson candidate events. The final dijet mass fit is an unbinned extended maximum likelihood with JES, and the slope and the normalization of the multijet background treated as nuisance parameters and allowed to float in the fit within their predetermined uncertainties. The EWK normalization is also freely floating in the fit with no constraints. The following table shows the results of this fit:

The 1516 events represent a 5.3 σ observation of the diboson production in the hadronic channel.

We can now look at variables that are sensitive to the shape and amount of multijet background and see how well we reproduce them with our model detailed above.

The table of uncertainties is listed below:

The EWK shape is determined from the alternative γ+jets data and the resolution uncertainty determined by smearing each jet by the expected uncertainty on the jet energy resolution.

The acceptance for WW, WZ and ZZ is 2.5%, 2.6% and 2.9% respectively. Assuming the ratio of cross sections predicted by the SM the only thing remaining to calculate the combined cross section is the luminosity of the sample. Since we use a large number of trigger paths this is not a trivial question.

We determined the luminosity of the sample by counting the number of Z decays to muons that pass all our analysis cuts in our MET sample and the same number in the well understood muon triggered data. We find 3.5 fb-1 as the total effective luminosity.

The measured cross section is: 18±2.8(stat)±2.4(syst)±1.1(lumi) pb in agrement with the SM calculations of 16.8±0.5 pb.



Wine and Cheese talk