Winter 2004 Preblessing Questions

Winter 2004 Preblessing Questions, 12/18/03

  • Q from Patrizia Azzi: Make sure we all agree upon the good run list version we use.

    A: We are now using good run list v4.0 from the DQM pages, with run 164844 commented out. Luminosities are 193.5 +/- 11.4 pb-1 for CEM/CMUP and 175.3 +/- 10.4 pb-1 for CMX.


  • Q from Patrizia Azzi: We need to come up with a plan for the lepton ID efficiency scale factors.

    A: We have come to an agreement in the lepton plus jets group to use the scale factor values obtained from the ETF and muon groups, but to apply a 5% error to these numbers. See CDF Note 6858 for studies leading to this conclusion.


  • Q from Young-Kee Kim: Given the poor fits in the 1-jet bin, are we fooling ourselves in the 3 jet bin by getting good agreement in data v. MC via the KS test simply due to poor statistics?

    A: We think we understand the discrepancies in the 1-jet bin. Our suspicion is not that our MC model (jet energy scale) is bad, nor that the higher statistics in the 1 and 2 jet bin make the discrepancies more visible, but that the "feed down" from the higher multiplicity W+n parton events alters the kinematics in the inclusive 1 jet bin which is not reflected in the W+1parton MC samples only. Indeed we have shown that other samples (Mrenna's CKKW samples) that combine all jet multiplicities into a single MC sample describe the data well. See Evelyn's talk at the 12/19/04 lepton + jets meeting for plots. Michelangelo's matching scheme should also give better agreement, and this is under study.


  • Q from Young-Kee Kim: Wants to be reminded as to how the fits changed in the summer when we didn't have the QCD deltaPhi cut in there. (Curious because of George and Guram's analysis.)

    A: We had the fits in summer with no deltaPhi cut for our initial preblessing, but at that point we were not considering the 4 or more jet bin. We have now revisited this to answer your question, but with the three component only fit to W+jets, ttbar, and QCD for simplicity (all the small backgrounds not included).

    Cuts

    W + >= 3 Jets

    W + >= 4 Jets

    With QCD dPhi Cut

    ttbar: 13.2 +/- 4.2%, or 4.9 +/- 1.5 pb

    ttbar: 48.9 +/- 13.4%, or 7.6 +/- 2.0 pb

    Without QCD dPhi Cut

    ttbar: 11.7 +/- 3.8%, or 5.0 +/- 1.6 pb

    ttbar: 45.3 +/- 12.8%, or 7.6 +/- 2.2 pb


  • Q from Weiming Yao: Note which systematics in the table are uncorrelated and which are correlated. (He noticed that some were added in quadrature and some were not.)

    A: We will specify this in note. Will note this in blessing talk in the systematic error table.


  • Q from Patrizia Azzi: Can we remind her of the efficiency of the deltaPhi cut per Njet bin (for W+jets)?

    A: We have done this for CEM and CMUP for the different W+jet samples. We use AlpGen W+np for the inclusive n jet bin, and W+4p for the >=4 jet been. Here are our results: 

    CEM
W+1     15408           14413           93.5 +- 0.2
W+2      5267            4918           93.4 +- 0.3
W+3      3789            3528           93.1 +- 0.4
W+>=4    2325            2182           93.8 +- 0.5
Average                                 93.5 +- 0.2

CMUP
W+1      6132            5668           92.4 +- 0.4
W+2      4069            3723           91.5 +- 0.4
W+3      2461            2249           91.4 +- 0.6
W+>=4    1522            1403           92.2 +- 0.7
Average                                 91.9 +- 0.3

    The CEM are all quite consistent with 93.5% across all bins The CMUP are also consistent across all bins but with slightly lower efficiency (1.6 +- 0.4 % lower).


  • Q from Tony Liss: Would like us to do a cross section versus mass plot with skewed errors and the measured mass for public presentations.

    A: We are working on making this plot.


  • Q from Jeremy Lys: It is not clear to me why both .ge.3 jets and .ge. 4 jets results are being blessed. Clearly one cannot combine the two results. Since the .ge. 3 jets sample includes the .ge. 4 jets, and also gives smaller uncertainties, is the .ge. 3 jets result the "official" result? (Assuming that the present data is "typical" uncertainty-wise). Then the point of .ge. 4 jet result must be as a check? But how does one check, given the correlation?

    A: Our a priori choice for selection was the .ge. 3 jet sample since we determined it would give us the best sensitivity (early studies using pseudo-experiments). The .ge. 4 jet sample is meant to be a cross check, useful since the S/B is about 1-1. We can calculate exactly how correlated the samples are using pseudo-experiments (will do!) but we expect to use the .ge. 3 jet sample for any combination with other cross section measurements at this point.


  • Q from Jeremy Lys: I can play with cdf6802 Table 13 numbers. Find that 4.7 ttbar events are found in .eq. 3 jet events, compared to an expected 50.8 based on the .ge. 4 jets. Or other things. Then need uncertainties. But at a naive level there seems to be disagreement between the two results.

    A: From Table 13 of v1.1 of CDF 6802, we find a cross section of 4.7 pb, rather than 4.7 events. Perhaps this is part of the confusion. If you take 12.5% of 525 events, you get that you expect about 66 events in the three or more jet sample from the fit results in the .ge. 3 jet bin. For the .ge. 4 jet sample you have 51.2% of 119 events, or 61 events. While there are no errors included, Table 3 gives an indication from monte carlo that we could expect about 98 events in the .ge. 3 jet sample assuming a 7 pb cross section, and about 53 events for the 4 or more jet sample. Our final cross section results are consistent with these lower/higher numbers.


  • Q from Jeremy Lys: On systematic uncertainties. As for example in sect 13.1.2. Why are fits being made with the shifted templates? Seems to me that there is a fixed procedure, that is used on the data, and one is asking about properties of that fixed procedure. One inputs events according to various hypotheses (e.g., that the energy scale is wrong by x %) and sees what result the procedure yields? That is, what result compared to the default hypothesis (e.g., that the energy scale is correct)? Besides the principle, wouldn't life be easier if the fits always used the same templates?

    A: This has been clarified with Jeremy and in the text of the note. For the systematic errors from, for example, the absolute jet energy scale, the procedure the following: Suppose we don't know that correction within 1 sigma. We then shift the corrections by 1 sigma, then generate a fake dataset using 1k pseudo-experiments. We then fit the fake dataset to the same (nominal) templates, and see what the difference is. (From the jet corrections group, the procedure is to take the difference between the results obtained with +/- 1 sigma shifts and divide by two.) In other words, we do use the same templates for the fits.


  • Q from Jeremy Lys: I see nowhere stated that pseudoexperiments with default input give (on average) the correct answer? That is, one can take random samples of A W+jet bkg MC events and B QCD bkg events and x ttbar MC events, do the standard fit and look at the returned number of ttbar events. With A and B fixed, and many pseudoexpts at each of several x values that correspond to a reasonable cross section range (0 to 15 pb ?).

    A: We have demonstrated the performance our fitters in past meetings, and have shown pull distributions, but we are happy to make these plots again and perform the test you suggest.

    Here are the results of 10k pseudoexperiments for sigma(ttbar) = 4, 7, 10 pb. Clearly for the 4 pb case there are a small number of pseudoexperiments where the fit fails (gives a non-physical result). But in each case we fit to the central part of the distribution, and print the gaussian parameters. The mean tends to be a bit high, a result of the poisson asymmetry. This is to be expected, and the effect is much smaller than the statistical error on the individual results.


  • Q from Jeremy Lys: Associated with the preceding point, on Fig. 7. It would be good to know whether the uncertainty returned by the fit agrees with the spread of returned fit values. As a function of the returned uncertainty.

    A: We have generated pseudoexperiments for both the 3 or more jet sample and the 4 or more jet sample assuming a theoretical top cross section of 7 pb and using the normalization of the W+jets shape from our fit (since we have no a priori guess at this cross section within 20-25% according to Michelangelo, and normally let it float in the fit). Our fit results from our particular experiment are shown with the cross hairs. For the 3 or more jet case, we are on the edge of the spread, but one might expect this because our measured cross section in this sample is quite a bit lower than 7 pb. For the 4 or more jet case we are somewhere in the middle.


  • Q from Jeremy Lys: A quibble about Table 4 - A test of the delta-phi cut. I don't know how to compare the numbers given. For example, the mu + 3 jets case: the observed number of events in the cut region is 31 (183 - 152), and the predicted number is 16 (183 - 167). The 16 and 31 do not agree very well. But there is some uncertainty on the prediction and I have no idea what it is.

    A: We don't have a good handle on this due to the limited statistics in the data. For example, in the 3 or more jet sample, we have numbers like 183 +/- 14 before the dPhi cut. So, if you say the predicted number is (183 +/- 14) - (152 +/- 12) = 31 +/- 18, you are within error of the predicted number of 16. Nevertheless, the predicted numbers should also have errors associated according to the equation "n(predicted)" in section 5.2, page 6. We can try to get them.


  • Q from Jeremy Lys: On the systematic uncertainty from QCD bkg shape. Is the 16% the difference between using the default shape and the conversion- electron shape? - or half that? or ...?

    A: The uncertainty is one-half the difference in the mean of fit results to 1k fake datasets obtained from pseudo-experiments based on the shifted (conversion) QCD shape. We believe this might be an over-estimate, because while the conversion sample might be a good alternate shape for CEM fakes, there is no reason it should model muon fakes. We plan to investigate using the low MET sample instead for the future alternate shape for systematics.


  • Q from Jeremy Lys: On the cross section versus top mass. The old systematic of 13% (cdf6802) implies that a 5 GeV mass shift would change the cross section (.ge. 3 jets) by 0.6 pb. But the preblessing talk says 0.3 pb. Are those two compatible (they may well be)?

    A: Note that in the old systematic (which was included in our table for Lepton Photon) we used Pythia at different masses, while in the current presentation we are using Herwig. Also, we are doing something fairly different right now. The old systematic is from pseudo-experiments, and we get 0.6 pb (13%), whereas the 0.3 pb in the preblessing talk is what we measure with data using a MC sample where the top mass is different. This is so we can make a plot of cross section versus top mass instead of including top mass as a simple systematic. Still, to check the compatibility of these numbers, ideally we would use the same generator and do pseudo-experiments so we can fairly compare. Unfortunately, right now we have only the Herwig samples used for the pre-blessing answer in hand.