Top mass measurement in the all-jets channel for run II. ======================================================== (Brian Connolly, FSU) Abstract: --------- We have applied a neural network analysis to the D0 all-jets data in order to obtain an estimate for the precision with which we can measure the top mass form the all-jets channel in run I, and then from this extrapolate to run II. The analysis is similar to the analysis performed previously for the measurement of the top cross sections from the all-jets channel and the analysis done by Eunil Won for his PhD thesis. Description of analysis: ------------------------ The data, signal, and background discussed here are tagged unless stated otherwise. An event is tagged if at least one of its jets has a muon within a 0.5 cone radius of the jet. The signal sample is produced from t-tbar -> all jets Herwig Monte Carlo ranging from a top mass of 120 to 230 GeV in 10GeV steps. The background is produced in exactly the same fashion as that in the D0 all jets cross section measurement. The tagged and untagged jets from the data are histogrammed separately as functions of ET and eta. With two functions corresponding to the histogrammed tagged and untagged data (T(ET,eta) and U(ET,eta)), the tag rate function is T(ET,eta)/U(ET,eta). The tag rate function is essentially the probability that a jet will be tagged. The tag rate function is then applied to the untagged data to produce an independent background sample. For the Run I calculation of the mass, the signal , background, and data are histogrammed as functions of a kinematic mass fit and neural network discriminant for a given neural network cut. Figure 1 shows a signal sample (180 GeV), Fig. 2 background, and Fig. 3 shows the data plotted vs fitted mass and neural network (These are box diagrams with the size of the box representing the number of events) The signal (for a particular mass) and background are combined and compared to the data to produce a likelihood [] for that particular signal and background combination and the data to come from the same distribution. Table 1 shows the mass, statistical error for the mass, and the expected number of signal and background events for Run I and Run II. Specifically, the numbers are calculated as follows. The -ln(Likelihood) is plotted as a function of fitted mass, and then fitted to a second degree polynomial, assuming the -ln(Liklihood) is a gaussian around the minimum. the minimum is then the mass for which the signal and background combination and the data are most likely to come from the same distribution. Figure 2 shows a sample likelihood as a function of mass for a neural network cut of >0.8. If the minimum is the location of the mass, then the statistical error is calculated by taking the half width between the mass values for which the -ln(Likelihood)is by 1/2 bigger than its minimum value. Before the calculation for the statistical error for Run II mass is discussed, the calculation of signal and background events deserve some explanation. The signal events for Run I were found by making neural network cuts on Monte Carlo. The background for Run I was determined by making the same neural network cuts on data. The Run I signal and background were scaled by a factor of 20 for the increased luminosity, and another factor of 1.4 for the 3 increased cross section from the increased center of mass energy. Of course this is a conservativve estimate for the background. Since the signal is around threshold, and the multijet QCD is not, it is not expected that the background will increase nearly as fast as a function of energy. In addition, the Run I signal receives an extra factor of 3 from the increased b-tagging efficiency. After the signal and background are calculated for both Run I and Run II, the statistical error is scaled by sqrt[(1+2B/S)/S]/sqrt[(1+2b/s)/s], where B(b)=number of background events for RunII (Run I), and S(s)=number of signal events for RunII (Run I). Summary of assumptions for extrapolation to run II: ---------------------------------------------------- * same trigger, selection and reconstruction efficiency as in run I * b-tagging efficiency for top events bigger by factor 3 than in run I * integrated luminosity = 2fb^-1 * top cross section at 2 TeV = 1.4 * cross section at 1.8 TeV * background cross section rises with energy as fast as the top cross section * statistical uncertainty on mass measurement scales as sqrt[(1+2B/S)/S] (S = number of signal events, B = number of background events) (this corresponds to the assumption that the mass error scales with number of signal and background events the same way as the relative error on the top cross section determined from the same data set) (Note that the assumption of same trigger efficiency as in run I may be interpreted as an additional argument in favor of the STT: In Run I, the trigger efficiency for top events in the all-jets channel was high (~100%?). However, with the increased luminosity and trigger rates, this may not necessarily be the case in Run II. The STT may help with rate control at L2 to ensure high trigger efficiency for the top to all-jets signal. Without STT, it may not be possible to achieve the same high trigger efficiency) Results: -------- NN> Mass (GeV) Statistical Error (GeV) (Run I) (Run II) 0.8 182.19 21.89 1.45 0.85 182.17 21.91 2.04 0.9 189.51 25.54 2.11 0.95 194.46 19.92 1.37 NN > Signal Background Signal Background (Run I) (Run I) (Run II) (Run II) 0.8 30 232 2520 6496 0.85 27 182 2268 5852 0.9 22 127 1848 4200 0.95 15 56 1260 1988 Table 1 Uncertainty in the Jet Energy Scale: ------------------------------------ The study of the influence of the jet energy scale uncertainty on the top mass uncertainty is not yet finished. Since the kinematic mass fitting procedure has the the mass of the W's as a constraint, the top mass uncertainty is bound to be smaller than the energy scale uncertainty, and is presumably dominated by the uncertainty on the b - jet energies. The present jet energy scale uncertainty is (2.5% + 1GeV). A conservative estimate of the top mass uncertainty arising from the jet energy calibration is then 5.4 GeV. Assuming the same values for the other systematic errors as for the lepton+jet mass determination, we obtain a total systematic uncertainty of 6.5 GeV. The study of U. Heintz and M. Narain concludes that with the help of the Z to b bbar signal, the uncertainty in the jet energy scale for b-jets is to improve from (2.5% + 1GeV) in Run I to ~1/3% in Run II, while without the Z to b bbar signal it would be 1% at best. A more realistic estimate of what can be achieved is about 1.5%. Assuming that the top mass uncertainty reflects just the b-jet energy uncertainty, we would expect that in run II the top mass uncertainty contribution from the jet energy scale should be about 1.5% without and 1/3% with using the Z to b bbar signal. If we assume the same values for the other systematic errors as was done in the note by Uli and Meena, we find that the total systematic error on the top mass from the all-jets channel is about 2.6 GeV without Z->bb, and about 1 GeV with Z->bb, to be compared with a statistical uncertainty of about 1.7 GeV (the average of the values for the 4 different neural network cuts from Table 1 above).