TopJetCombination

Top Jet Combination

This study is actually done with sisCone05 jets calibrated at L2L3 level.
The selection of jets happens in different steps:

0. Jet cleaning

electron removal done in PAT-L0

1. Jet rejection

use of the JetRejectorTool (ported in CMSSW_2_2_3 by Ilaria)

2. Selection of a collection of 4 jets

tools & study performed by Gregory

3. Selection of a combination in a 4 jets collection: MVA package

Tuning of variables for MVA package (Joris)

• A list of >80 variables is implemented and can be found in (code is compatible with TQAF tag V04-07-01)

/user/jmmaes/CMSSW/CMSSW_2_2_4_v1/CMSSW_2_2_4/src/TopQuarkAnalysis/TopTools/interface/TtSemiLepJetComb.h

Due to Recent changes in TQAF this file became incompatible

• For this list the MVA trainer is used to make the distributions for signal and background in two cases
  • case1: The 4 top jet candidates are selected from the 6 jets with highest pt. The 4 jets are chosen according the chi square method in the jet selection tool of gregory (see previous section)
  • case2: The 4 top jet candidated are the 4 highest pt jets

in the following locations you can find the complete cfg file used to produce the root files containing signal and background plots

/user/jmmaes/CMSSW/CMSSW_2_2_4_v1/CMSSW_2_2_4/src/TopQuarkAnalysis/TopJetCombination/test/caseX/ttSemiLepJetCombMVATrainer_muons_cfg.py

• The root files created are in

/msa1/TopGroup/Results/TopJetCombination/SValues/root/train_monitoring_caseX.root

• For the list of all variables a separation power is calculated, two definitions are used to calculated S:
  • The overlapping area of the (normalized) signal and background histograms (surface)
  • The maximum value of the scan of the multiplication of efficiency for signal times 1-efficiency for background (pte)

The code for calculating S can be found in

/user/jmmaes/CMSSW/CMSSW_2_2_4_v1/CMSSW_2_2_4/src/Svalue/DistributionAnalyzer.C

The list of S values for the 2 cases X times the 2 definitions Y is located in

/msa1/TopGroup/Results/TopJetCombination/SValues/table/caseX_Y.txt

The main conclusion is that the ordering of the variables doesn't change significantly when using another definition. Also when selecting the 4 jets from the 6 highest pt jets the relative ranking of the variables w.r.t. separation power stays the same

• To make a selection of O(5) variables we have to take into account the correlation between them. This list is also in the former text file.
• The list of 5 variables looks like this (numbering comes from the 81 variables)
  • 1. mass of hadronic W (massHadW)
  • 80. sum of b-tag discriminant value of leptonic b and hadronic b (sumBTag1HadBLepB)
  • 78. relative pt of the hadronic top (relPtHadTop)
  • 53. delta theta between leptonic b and hadronic top (deltaThetaLepBHadTop)
  • 50. delta theta between leptonic b and muon (deltaThetaLepBLepton)
• The plots of the variables Z can be found in

/msa1/TopGroup/Results/TopJetCombination/SValues/image/varZ_caseX.png

• The correlation between these 5 variables is made from the train_monitoring.root files of petra. The numbering of the variables is different from the previous section. The plots for both cases X can be found in

/msa1/TopGroup/Results/TopJetCombination/SValues/image/ProcMatrix_rot_caseX.png

• The numbering of the variables is (1:"massHadW",2:"deltaThetaLepBLepton",3:"deltaThetaLepBHadTop",4:"relPtHadTop",5:"sumBTag1HadBLepB")
• in the txt files you can find explicitly the value of the correlation

/msa1/TopGroup/Results/TopJetCombination/SValues/table/caseX_corr.txt

Study of MVA tool (Petra)

• MVA tools: NN & LR are performed by Petra