|
|
|
Work on most final systematic prescriptions is still ongoing. We need
the support and active involvement of all top mass analysis people to
make the prescriptions final. The sooner you volunteer to participate
in pinning down your favorite systematic, the sooner you can go ahead
and publish your analysis. Waiting on the sideline is not an option!
|
To do items for each systematic source with rough time scale estimation
(1) Jet Energy Scale
Most likely we will move to a L7 systematic that is taken as the difference between data and PYTHIA MC of the pt balance (as seen in gamma+jet and Z+jet events) as function of the jet pt, while before the maxium of the (data-PYTHIA, data-HERWIG) difference was taken. Herwig performs worse than PYTHIA in the description of the underlying event (even with Jimmy), and has large discrepancies with gamma+jet and Z+jet data in the description of the 2nd jet. This gives us enough reasons not to use HERWIG here as MC model to compare the pt balance with. Note that we have an additional generator systematic where we do difference between PYTHIA and HERWIG into account. We would like to supplement the picture with Z->mu mu data to increase the statistics. Volunteer if you have samples or are confident about Z->mu mu selections. We would also like to include ALPGEN as an additional generator to compare our gamma+jet and Z+jet data with. (2) Residual Jet Energy Scale
(3) PDF
Our current default mass samples use the CTEQ5L PDF (LO with LO form of alpha_s). We currently have the following PDF alternatives:
Currently the PDF systematics cover the uncertainties on one NLO PDF (CTEQ6M, the difference between LO and NLO with the same alpha_s and the difference between two diffrent alpha_s values at LO. In principle both CTEQ and MRST PDF's should be used with NLO generators, but our default PYTHIA is LO. They can be used in conjunction, but for consistency we use as default the LO PYTHIA with LO CTEQ5L PDF. Therefore the reweighting of the CTEQ6M 40 eigenvectors with this default PYTHIA sample as basis is a bit awkward. One could set up a new MC set That is based on either: (1)LO Pythia with CTEQ6M default PDF around which we reweight. OR (2) MC@NLO generator with CTEQ6M and use that PDF's eigenvectors. Point (1) requires probably the least amount of work. (4) ISR/FSR
We decided to follow the old prescription based on Drell-Yan events, but want to make an attempt to reduce the uncertainties on the Q^2 versus <pt_lep> plot mostly by inlcuding higher mass (Q^2)points. We will in any case switch to two alternative samples that will replace the old ISR/FSR samples. The new samples will have both ISR and FSR simultaneously increased or decreased. (This samples are being generated right now, Jan 30 2008.) This is more consistent from a physics point of view and is also what other analysis teams in CDF have been doing for a long time. We will not adopt the R-ratio method as is used by D0, because we are convinced that this has several flaws: (1) It accounts only for hard radiation , which we treat as NLO effects, (2) It is a statistically limited quantity and (3) It is analysis specific and sensitive to backgrounds. We need a volunteer to make a reduced ntuple of dimuon events in data which can be analysed by Un-Ki. Un-Ki might release a new set of ISR/FSR parameters after this study is finalized. (5) MC Generator
The new HERWIG samples are currently under production. The HERWIG+Jimmy sample is to be used as HERWIG default, since it describes the data better than HERWIG alone. But, always compare your result as a crosscheck with the HERWIG (no Jimmy) sample. Someone should check and validate the new HERWIG samples as soon as they become available. We need a volunteer to have a look into SHERPA. Generate some test sample and compare with the existing generators and data. (6) Background fraction and shape
Both components decribed above make our current background shape imperfect. Either we should improve our understanding of both components and fix the problem or use a reweighting scheme to cover the discrepancies between data and background in a background dominated region (W + 2 jet bin). The variable used for reweighting can not be very correlated with the top mass itself. We need essentially new ideas on a unbiased and stable reweighting procedure. Issues are: (1) Which component to reweigt andby how much, (2) Some idea about extrapolation uncertainty from the W+n-1 jet bin to the signal region, (3) Verify weights obtained from pretag or tagged events. (7) b-tagging efficiency This systematic is obsolete (8) b-jet energy scale
The semileptonic branching uncertainties are very well known from the LEP experiments, a prescription on how to weigh events according to these uncertainties exists and should be tested in each analysis. The B-meson fragmentation uncertainty is less well understood, however there are parameterisations of the uncertainty as function of the z-variable (longitudinal momentum fraction of the B-meson wrt the original b-quark). Top ntuples do not contain this information but a hack that approximates this information from hepg bank info can be made. Finally, there can be differences in the E/P response of the calorimeter due to different charge fraction in B-jets. This effect is curretly estimated to be small. We need to implement and test the 'hack' of the z-fraction weights. (9) Lepton PT
Summary of Victoria's Method: The pT of leptons can be calibrated accurately against the mass of teh Z boson using dilepton events. Assuming that the deviation from the world average Z mass is only due to teh lepton Pt (and not the rapidity or azimuth) one can relate the deviation from the Z mass to relative uncertainty in lepton pt by: delta(pt)/pt=delta(mz)/mz. Currently there are 4 different relative uncertainties (2 for lepton pt in ttbar signal events in MC and data and 2 for dilepton background events in Mc and data). Both signal and background uncertainties are propageted into the analysis and the shifts in Mtop seen in MC and in data are added in quadrature. We need to calculate similar numbers for the l+jets analyses and decide what to do with MC and data, since both have different relative uncertainties on the lepton Pt. (10) Multiple Interactions
All of our default Monte Carlo's have an average number of minimum bias evenst added on top of the hard scatter to account for the extra energy seen in our detector due to multiple interactions. It is clear that rather than having this constant, it should evolve with instantaneous luminosity. The amount of multiple interactions can be related to the number of reconstricted vertices in the event. Studies have shown that for each additional reconstructed vertex we observe an increase in jet energy between 142 and 487 GeV (depending on whether it's a b-jet or a light quark jet and whether one uses PYTHIA or HERWIG). The main question is whether this behavior also occurs in data. Studies by Un-Ki have show that the number of reconstructed vertices is quite different between PYTHIA and HERWIG. In principle each analysis should check whether the Nvtx distribution of their data is well modeled. Different conclusions can be expected depending on the number of b-tags each analysis uses. Secondly, it should be verified whether the top mass is very dependent on the number of reconstructed vertices and by how muc this is mismodeled in MC. (11) NLO effects
Some people have indicated that when comparing different NLO generators with our current LO MC, big shifts in the reconstructed top mass can be observed. We also know that some of our FSR systematic only covers softer radiation and cannot account for certain NLO diagrams which D0 is using for their FSR systematics. Our current MC@NLO and Alpgen tt+0p/1p samples have bene rarely used and verified by top mass analyses. The latest TMT l+jet analysis sees shifts of order 1 GeV when they compare these samples with our default PYTHIA. This should be further investigated. In addition we could compare with additional Madgraph samples of which even larger shifts have been reported from teh LHC community. It is also unclear which PDF's to use with those new generators. (12) Color Reconnection
The issue here is to have a reliable tuning in combination with Skands' model. The color reconnection parameterization is very interwoven with the underlying event and as such we cannot use it in combiination with our TuneA of Pythia. At the moment we only know that the pt of the ttbar system is affected the most (although not all distributions have been verified). Preliminary studies of teh lepton+track analysis have shown that by reweighting the PYTHIA ttbar pt distribution to the one given by Skands, result sin a O(500 MeV) shift, but with an uncertainty of teh same magnitude. A more precise top mass analysis should repeat this exercise and a fully simulated MC sample with color reconnecetion needs to be generated as soon as we are confident that the underlying event is reasonably well tuned. |