Measurement of the top quark mass in the all hadronic channel using an in-situ calibration of the dijet invariant mass with 943pb^-1

We present a preliminary measurement of the top quark mass in the all-hadronic channel, using 943 pb^-1 of ppbar collisions at sqrt(s) = 1.96 TeV collected at the Collider Detector at Fermilab. Assuming standard model ttbar production and using the matrix element as a weight, an event probability is calculated. The top quark mass is reconstructed for each event by maximizing the event probability, and so Monte Carlo templates are produced, dependent of the true top quark mass and the jet energy scale. The most likely top mass is extracted by fitting the data to the templates distributions. From 72 events observed with 943 pb^-1 we measure a top quark mass of 171.1 ± 3.7 (stat.+JES) ± 2.1 (syst.) GeV/c².

The final state of the ttbar all hadronic channel will be observed in the detector as 6 jets. However, the same 6 jet signature is produced via other sources, constituting the background events (mostly QCD multijets). In fact the background events are produced in much larger amounts than the ttbar events, and therefore we need to separate the two samples as best as possible. The following set of cuts are designed to enhance the ttbar content of the multijet dataset:

• ask for at least 6 jets with E_T > 15GeV & |eta| < 2
• Aplanarity + 0.005 * SumE_T3 > 0.96
• Centrality > 0.78
• SumE_T > 280 GeV
• minLKL < 10

The last cut provides the most drastic background reduction and it is essentially asking for the event to have a high probability of being a ttbar event. This probability uses the ttbar matrix element to weigh the observed jet configuration.

We use the matrix element to build the top mass templates. The same event probability mentioned for the event selection is used again, more exactly, its dependence on the top mass is used. This probability will have a minimum in negative logarithmic scale for a certain value of top mass, and this value we'll use to form the top templates. For ttbar events, the shape of these templates depends on the input top mass [eps] [gif] and JES [eps] [gif]. These dependences are shown below.

The dijet mass templates are formed by considering the invariant mass of all possible pairs of untagged jets in the sample. Their shapes are build for a variety of top mass samples  each having several possible JES values. The invariant mass of an untagged pair of jets shouldn't depend on the top mass [eps] [gif], but should be sensitive to change in the jet energy scale [eps] [gif]. The shapes for various ttbar samples are shown below.

We use a data-driven background model by extrapolating the heavy flavor rates from background dominated sample into our signal region. This is also known as "Method 1" background model, and it is using the tag rates as determined in the ttbar cross-section analysis described here.
The background templates are independent of the top mass or JES, and they are shown below. First row are the event top mass templates for single tagged events [eps] [gif] and for double tagged events [eps] [gif]:

  Here are the dijet mass templates for background single tagged events [eps] [gif] and double tagged events [eps] [gif]:

We need to make sure that our estimator of the top mass and JES is unbiased and that the statistical uncertainties are well understood. The reconstructed top mass and JES is represented by the red squares and they are plotted in the top mass-JES plane of true values, represented by the grid of dashed lines [eps] [gif]. The pull means for the mass reconstruction is consistent with an unbiased estimator [eps] [gif]. The pull width for the mass reconstruction as a function of the JES is 1.13, which means we need to inflate the uncertainty on the top mass by 13% [eps] [gif].

The largest sources of systematic uncertainty are summarized below. The jet energy scale is treated as a nuisance parameter in this analysis, and is constrained both by the a priori calorimeter calibration and the statistical information from the W mass through the dijet mass of the event. It is included in the uncertainty arising from the likelihood fit, thus not considered a traditional source of systematic uncertainty.

We perform a simultaneous unbinned likelihood fit of the data to the dijet mass and reconstructed top mass templates. The number of signal events in each subsample is constrained to its expectation value following a Poisson distribution. The overall JES is constrained to a Gaussian centered around 0.0 and a width of 1.0 (sigma). Here are shown the event-by-event reconstructed top mass distribution (black points) for data events in the case of the single tagged events [eps] [gif] and of the double tagged events [eps] [gif]. The  combination of signal and background top mass templates that best fitted the data points is represented by the orange histogram. The background template is given by the blue histogram. The reconstructed values of top mass and JES are shown (cross) in a two dimensional plot where the delta ln L contours are roughly corresponding to 1, 2, and 3-sigma levels [eps] [gif]: