Electron id study scratch book
Dave Study
Summary
Quick and simple test just now. I will not plot everything out in the categories, I do not have the time to waste. Just get a feel for what is going on.
Here is the "executive summary" of what I think we should do for now.
- H/E variable was screwed up for no good reason and needs to be tightened. Probably performance is not as good as before.
- SigmaEtaEta was changed for no good reason and performance is probably about the same as before
- dPhiIn was more or less unchanged as expected. Could not check dEtaIn but no reason to expect major change.
- Matteo has re-optimised all the variables in his style ID in 2_2. We should take his replacement thresholds for the SigmaEtaEta and H/E cuts. This is as defensible as anything else. The performance will probably be a little worse than before due to H/E, but maybe only by a few percent.
Old tight ID in 2_2 gives 0.73% passing in the ZJets sample that I re-ntupled myself (1000 events). The tightIdMinMatteo gives 0.70. Similar performance, hypothesize difference due to H/E being less good than before, despite cutting tighter.
The Cuts...
For reference, the old cuts for H/E and
SigmaEtaEta . First row is barrel second is endcap. Each entry is for a different "class". Summary: H/E is tightened, sigmaEtaEta is not changed much.
float cuthoe[] = {0.05, 0.042, 0.045, 0.,
0.055, 0.037, 0.050, 0.};
float cutsee[] = {0.0125, 0.011, 0.01, 0.,
0.0265, 0.0252, 0.026, 0.};
and the new...
float cuthoe[] = {0.041, 0.028, 0.025, 0.,
0.034, 0.012, 0.016, 0.};
float cutsee[] = {0.0118, 0.0112, 0.0104, 0.,
0.0271, 0.0263, 0.0256, 0.};
Data used
Quick and simple just now test. Used the following data
"SIGNAL"
TFile f_dy22X("/data/tmp/cms2-V01-02-01/ZJets-madgraph/merged_ntuple_1.root");
TFile f_dy16X("/data/tmp/cms2-V00-04-01/merge_DY.root");
"BAKGROUND"
// TFile f_dy22X("/data/tmp/dmytro/cms2-V01-02-01/WJets-madgraph/merged_ntuple_1.root");
// TFile f_dy16X("/data/tmp/cms2-V00-04-01/merge_Wjet.root");
- Looked at SigmaEtaEta , H/E, dPhiIn (dEtaIn was missing from the ntuples)
- Didn't know where the QCD was or if it was suitable so used WJets as background. Just ofr illustration only.
- Did a cheap N-1 with H/E and SigmaEtaEta as appropriate. This means sigmaEtaEta has H/E > 0.05, H/E has SigmaEtaEta < 0.0125 for signal samples. For background, cuts are inverted to select the "jets".
- Looked at Pt > 20 GeV
1_6 is in BLUE and 2_2 is in RED
What has changed: The weighting factor changed from 4.2 to 4.7 or the other way round, I forget which. The reason I have no idea. I hear it on good authority it is basically made up on the spot for no good reason.
BCKGROUND
SIGNAL
H/E
What has changed: H/E used to be calculated as the ET of the HCAL rechits in a cone of 0.3 about the super cluster to the ET of the super cluster. In 2_2 the "H" changed to ET of the calo tower than has highest ET at depth 1 that is within some distance of the electron, either that or just the closest calo tower. Doesn't matter which it was, because both are the wrong thing to do and degrade the performance of the variable. The optimum was found to be the sum of the hcal energy in a cone of 0.15. This will be done in 3_X.
BACKGROUND
SIGNAL
dPhiIn
Here just a quick check. The "N-1" here is with
SigmaEtaEta . "Signal" sample only plotted. Nothing should have changed much and indeed that seems to be the case.
SIGNAL
Oli: Electron selection variable comparison between 1_6 and 2_2
-
left
: 1_6, right
: 2_2
- all for:
- pTelectron > 15 GeV
- | ηelectron | < 1.
- histograms not normalized, plainly drawn:
- 1_6: use weights for everything except QCD samples
- 2_2: use weights for all
DYee
DYmm
DYtautau
QCD_Pt_15_20 / QCD_BCtoE_20_30
QCD_Pt_170_230 / QCD_BCtoE_30_80
QCD_Pt_20_30 / QCD_BCtoE_80_170
QCD_Pt_230_300 / QCD_EMenriched_20_30
QCD_Pt_300_380 / QCD_EMenriched_30_80
QCD_Pt_30_50 / QCD_EMenriched_80_170
QCD_Pt_380_470 /
QCD_Pt_470_600 /
QCD_Pt_50_80 /
QCD_Pt_80_120 /
ttbar
tw
wjets
ww
wz
zz