Further information about the study may be found in AN-2016/350:
http://cms.cern.ch/iCMS/analysisadmin/cadilines?line=SUS-16-033
This makes use of the madMinPhotonDeltaR variable, defined in the TreeMaker module here:
https://github.com/TreeMaker/TreeMaker/blob/Run2_2017/Utils/src/MinDeltaRDouble.cc
A fragmentation photon is one defined such that dr<0.4. The fragmentation fraction is defined as the the number of photons with dr>0.4 divided by the total number of photons.
The high statistics GJets_0p4_HT sample used in the Z->nunu estimation has a hard cutoff requiring dr>0.4. One needs to account for these events in the MC as any cutoff is artificial and does not match data. More practically, this feature is not present in the ZJetsToNuNu_HT MC due to the non-zero Z mass.
We use the sum of the low statistics GJets_HT, with the hard cutoff requiring dr>0.05, and QCD_HT samples to get the number of photons with dr<0.4. In order to not overcount, a stitch point is defined where QCD_HT is used for values less then the stitch point and GJets_HT is used for values greater than the stitch point. The stitch point value is nominally set to 0.4, but values 0.05 -> 0.4 make perfect sense and can be used for an estimation of the systematic errors associated with this choice.
To see what is going on with these distributions, plot the madMinPhotonDeltaR for the MC samples:
root madMinPhotonDeltaR.c+
This produces Figure 34 of AN-2016/350.
To produce a plot with a wider range, comment/uncomment lines 44-50 appropriately.
Make the root file containing the fragmentation histograms:
root fragmentation.c+
fragmentation f
f.run()
Three files need to made for stitch points of 0.2, 0.3, and 0.4 to be used for systematic errors. To do so, comment/uncomment lines 127-129 appropriately.
This code is also able to calculate the fraction the low-deltaphi region used for the QCD estimate. To do so, comment/uncomment lines 131-137 appropriately.
Plot the fraction for all the analysis bins:
root fragPlot.c+
This produces plots of the fragmentation binned in HT, MHT, NJets, and the 46-bin analysis plane.
This produces Figure 35 of AN-2016/350.
For a systematic error, see how the fraction changes when using a stitch point of 0.2 or 0.3:
root dfover1mf.c+
If you want a simpler value to assign for the systematic error, and not bin by bin, we can look at the scatter of the central value of the differences (ignoring the statistical error seen in the bin-by-bin plot):
root scatter.c+