I’ve improved the fits for the SE parameterization.  Here are the plots.

———————————————————————————

-Scott

To parameterize the new likelihood, I tried to take the correlations into account more carefully than last time.  I made three bins each of Pt, Eta, and Isolation, then fit the fake rate dependence on each of those variables in each of the 27 bins.  This parameterization models the fake rate in the muon-triggered sample quite well:

However, the Jet50 sample looks very different:

I’ll examine which variables are causing this difference, and have something by the Monday meeting.

-Scott

Edit:  the signal-only likelihood also has this problem, but it’s less dramatic in a way.  The fake rate is still off by a factor of two at higher pT, but it’s a factor of two of a smaller number.  (In the signal-only likelihood, the fake rate is higher at low pT and lower at high pT.)

I’ve finished retraining the likelihood with the new CP2 radius and more of the Eta correlations added in.  Here are some plots.

First, here’s the efficiency and fake rate as a function of pT, eta, and isolation for a likelihood cut of 0.9:

The strange eta effect is gone, and the efficiency is extremely high for high-pT candidates. (In this case, high-pT means “above about 4 GeV”.) This high efficiency in the high-pT case remains even at stricter likelihood cuts.

Here’s a plot of the low-pT (below 1.25 GeV) efficiency and fake rate as a function of the likelihood cut:

Also, here’s a comparison of the efficiency vs. fake rate for the “signal vs. background” likelihood (smooth lines) and the “signal” likelihood (many points).  The red is the 1 < pT < 1.25 range, green is 2 < pT < 2.5, blue is 4 < pT < 5, and purple is pT > 8.  As expected, the “signal vs. background” likelihood performs better.

-Scott

I figured out why the CPR response varies so much with Eta. See the recent posts about the Eta effect in the electrons, and see this plot of Conversions/Generic tracks in E_{CPR} vs. Eta:

Here’s a high-resolution plot of CPR response vs. the extrapolated Z position of the track.  This plot includes all tracks (after quality cuts) so we expect to see a MIP peak in the response:

The MIP peak is clearly visible near the center of the detector, but is only visible along one edge of each pad near Z=100 cm. (The pads are 12.5 cm wide, as you can see in this plot.) It’s clear that the actual maximum of the response in the detector is at a higher Z than where the extrapolator puts the track.  By Z=150 cm, we’re completely missing the energy in the detector. This Z range corresponds very closely to the Eta range in which we see the strange electron effect.

It should be relatively easy to fix the extrapolator to correctly find the CPRZ of the electron.  A simple increase in the radius of the CPR should do it.  This would also affect the E_CPR vs. X_CPR distribution, but it’s difficult to tell if the same problem exists there, the the CPR is only three pads wide in X:

Edit:

I changed the CP2 radius from 170.47 cm to 172.47 cm, and the effect goes away.  I’ll rerun my soft electron stuff now, and we’re going to have to reprocess the ntuples to get the correct CP2 data in there.

-Scott

As we’ve talked about, there’s a strange effect in the soft electron eta distribution, with edges at about +/- 0.5 that make it look like a muon distribution:

I’ve found out what the problem is: the CES response is different at different points in CesZ.  Pasha’s 2D CES calibrations normalized the E_CES / P response to real electrons with respect to position, but the fakes have a different profile:

I neglected this correlation between E_CES and Z_CES in the likelihood, so we end up with more mistags (because more pions have a high E_CES) at low eta and fewer mistags at high eta, which is consistent with the distribution that we saw.  I could retrain the likelihood to take this correlation into account, or just modify my MC parameterizations to try to make them agree with the data.  The latter would take less time, so I’ll work on that now, but we really should actually fix it (and retrain the likelihood again) someday.

Edit: (7/7/10)

There’s another effect which also contributes:  the fakes have a larger spread of their CES Delta Z when the tracks are at a higher eta:

Again, the CES calibrations normalized this for real electrons, not for fakes.  This is another effect that gives fewer mistags at high eta.

-Scott

The heavy-flavor contribution is found by fitting the pTrel and signed d0 of soft electrons.  The shapes of the various templates are too similar to each other, so I combined the bottom and charm templates into a single “heavy” template and fit a*(light) + b*(heavy).  The electron-triggered and muon-triggered data give the same result (heavy fraction is 0.22.)  The soft muon fits don’t work with the low-lumi template. (every template happens to have a 0 in a particular pTrel bin where we do have a data point.  This problem will go away when we use the high-lumi templates.)

TCE trigger:

CMUP/CMX trigger:

Here’s the table of predicted and observed values:

I think the next step is to improve the parameterizations of efficiency and fake rate for soft leptons (of both types) in the MC.  neither one is looking great right now.  Below the fold, see pT and eta plots of soft electrons and soft muons: Continue reading new expected/observed soft lepton numbers →

With all that we’ve been talking about the soft electrons outside of jets recently, I decided to see if looking at those (which are almost certainly all conversions) could give us a better handle on the conversions inside jets.  I took the d0 distribution of soft electrons in jets and fit it to a sum of a “conversion” template (found from data by looking at soft electrons outside of jets) and a “non-conversion” template (found by looking at the MC and only taking electron that matched to a non-photon) Here’s the fit (currently it only includes the low-lumi electron-triggered data; I’ll integrate it into the rest of our fitting programs and make it include everything soon.

I expect that this result will greatly improve the heavy flavor fits, which still have the excess at high d0, presumably due to conversions:

-Scott

Previously, there were a large number of soft electrons that had no jet associated with them, partially because the jet they were in was removed from the jet block because it overlapped with the trigger electron.  Here are plots showing the Delta R between the trigger and the soft electron for those that do have a jet associated with them and those that do not:

Those that do have a jet:

Those soft electrons that do not have a jet:

Obviously, the problem is not completely due to being too close to the trigger electron.

Bonus Section! Continue reading Removing electrons that have no jet associated with them →

After our talk on Friday, I put a ttbar selection into the code.  It seems to be working well: the ttbar MC accounts for a large portion of the observed events, as seen below.  The selection is:

• Trigger lepton
• Met > 30
• hT > 250
• nJet >= 4
• dPhi(Met, Lep) > 0.5

From the plots below, it looks like the hT cut isn’t doing anything.  Also, I’m going to have to make a new selection with the Met cut reversed in order to fit the QCD.

Bonus section! Continue reading ttbar Check →

We’ve finally run the analysis all the way through, although nearly every element could use some improving.  Here are the results from the first femtobarn of data:

TCE trigger:

CMUP/CMX trigger:

We should talk about these numbers at some point, of course.  Until then, I’m going to see what’s wrong with the soft electron efficiency and fake rate parameterizations.  They agreed well with my conversion and generic track test samples, but there’s something wrong with at least the pT part when applied to our W sample:

This problem went away in the past, but I think it came back when I recently updated the matching to find conversions. I’m going to check that now.

-Scott