(1) Interpretation of shifts

The interpretation of all systematics with +/- 1 sigma shifts should now go as follows. The systematic is half the difference between positive and negative mass results, unless both positive and negative go in the same direction from nominal, in which case the systematic is half the largest difference from nominal. This change applies to: IFSR, the semi-leptonic decay component of bJES, and lepton pt systematics. (JES and PDF systematics were already using the half-difference approach.)

(2) MC Generator

• For the generator systematic, we quote the difference between the mass values obtained in PEs with Herwig and Pythia Monte Carlo generators. Other generators are also under investigation and should be analyzed as cross-checks.
• As for HERWIG:
  • The old htop75 sample had incorrect JIMMY settings.
  • The sample otop01 should be the same as otop1s, but otop01 has 0.5M events, while otop1s has 4M. We require to use otop1s for HERWIG.
• As for ALPGEN:
  • For the time being there are three ALPGEN samples: otop1q which is ttbar inclusive and atop0p+atop1p which are ALPGEN tt+0p and tt+1p respectively. The difference between these two ALPGEN models probes the importance of the higher-order ME.
  • ALPGEN samples of tt+0p and tt+1p should be combined according to the cross-sections listed on the the top group MC page.
  • It seems as though our ALPGEN samples have been generated with zero (or very small) top width. Besides being plain wrong, this causes a bias in the ALPGEN samples. The top quark Breit-Wigner interacts with the PDFs such that a significantly lower width results in a larger generated top mass on average.
  • Because of the problem with the top width, ALPGEN is not part of the generator systematic for 2 fb^-1. No need to run on these samples until the problem is resolved.
• As for MC@NLO:
  • For the blessing, we require to cross-check your result with MC@NLO(mtop75, ctop75) samples.
  • Note that MC@NLO samples have ~10% of events with negative weight. This should be taken into account when running PEs!
    • The weight is not directly stored in TopNtuple. Instead, as a workaround, it is encoded in the hepg energy of the proton vs anti-proton.
    • In practice: hepg.E[0]=980 for W=+1 and hepg.E[0]=980.09997 for W=-1.
  • mtop75 and ctop75 differ in the PDFs used. See the top group MC page for details.
  • Finally, there are random number issues with these MC@NLO samples. Events are duplicated many times within the sample at the hard-scatter level (though the showering/hadronization is then not identical). So the statistical uncertainties on quantities derived from these samples are not understood!

(3) b-jet energy scale

• Take into account the b-jet energy scale uncertainty. This systematic comes from three main components:
  • semi-leptonic branching ratios - move b and c BRs together by +-1 sigma
  • fragmentation - reweight to LEP/SLD Bowler parameters
    • To compute the two above effects, Erik and Jahred prepared a code that can be found at the following webpage
  • calorimeter response effect:
    • run S+B pseudo-experiments where the B-jets have their input energies shifted by +/-1%
    • the B-jets are identified using hepg information; other groups only identified signal B-jets (ie. ignored real Bjets in Bgd)
    • the energies are shifted so that their inputs to the kinematic fit (or whatever the equivalent is for a given analysis method) are +/-1% higher/lower than when using the default jet energy corrections
    • in these studies we've _ignored_ changed in the acceptance on account of these shifts and _ignored_ changes in the ratio of single-to-double tag events (Btag efficiency doesn't have that steep of a dependence on Bjet energy that 1% shifts should dramatically change any of this relative to dflt)
    • compare the mean fitted M_top for these two shifted samples with the mean fitted M_top using default jet energy corrections; from this you can estimate that a 1% shift in Bjet energy corresponds to a X.X GeV shift in top mass
    • your final systematic needs to be scaled to 0.2%, which is our best estimate of how 1-sigma changes to calorimeter response affect the B-jet energy scale. Thus this specific systematic will be 0.2*X.X = Y.Y GeV

(4) Pile-up

• There is a change in terminology here: for public presentation instead of "multiple interaction" use "pileup": from Lina Galtieri's e-mail:"in talking to Seymour, Torbjorn Sjostrand and others, I realized that our systematics from Multiple Interactions is taken to mean systematics from PARTON MULTIPLE INTERACTION, not from MULTIPLE P-P COLLISIONS". Note: At LHC pileup can also include information from more then one event due to slow recording of event information. But that's not the case at Tevatron.
• To estimate the pileup systematic there are two pieces.
  • Known mismodeling
    • The first piece is related to a known midmodeling: our Monte Carlo does not have the same luminosity profile as the data we compare to it. We use MC datasets with a luminosity profile corresponding to just the first 1.2 fb^-1 of data. Additionally, the conversion from luminosity to number of minbias events was wrong by ~20% in datasets of that vintage.
    • Calculate the importance of this mismodeling on one sample (presumably 175 GeV). Possible approaches:
      • Run PEs using reweighting, where the weights depend on the number of interactions in the event, and the reweighting reproduces the Nint profile of the data. Find the difference (and its uncertainty) between these PEs and the nominal PEs.
      • Run batches of PEs where the input events are segregated by number of interactions. Plot the average fitted mass vs Nint. Fit a pol1. Hopefully the slope is within say two sigma of zero. Multiply the slope, or its uncertainty, by the difference between the average number of interactions in data and MC.
      • Something else convincing.
  • Unknown mismodeling
    • The second piece is related to possible mismodeling of the minbias events, which would induce a systematic shift even if the correct luminosity profile were used in the MC. We expect the first piece to dominate, but want to check what could be the magnitude of the unknown mismodeling.
    • MC-only studies on a ttbar sample show that the jet response vs number of vertices in an event does not match our expectations from minbias data, and the difference is significantly larger than the L4 systematic for the effect of pileup on jet energies. This is not necessarily a problem, because the data could have the same behavior in ttbar events. (There is not enough statistical power to check in the data.) But the observed effect in MC is a good probe of the order of possible mismodeling.
    • The observed jet response slope after jet corrections is ~250 MeV/jet/vertex. The nominal L4 systematic is 107 MeV/jet/vertex. So we can estimate the importance of this order of mismodeling by scaling up the L4 systematic by a factor of 2.3.
    • An additional complication is that the L4 systematic is usually evaluated on e.g. ttop75, which as we know has the wrong luminosity profile. The L4 correction/systematic goes as the number of vertices minus one, so the scaling factor becomes 2.3 * (<Nint>(data)-1) / (<Nint>(MC)-1).
• The larger of the two pieces (known mismodeling, unknown mismodeling) is quoted as the pileup systematic.