raspereza/bbhrun2legacy

Run2 Legacy bbH analysis

Instructions

Setting up the area:

export SCRAM_ARCH=slc7_amd64_gcc700
cmsrel CMSSW_10_2_13
cd CMSSW_10_2_13/src
cmsenv
git clone https://github.com/cms-analysis/HiggsAnalysis-CombinedLimit.git HiggsAnalysis/CombinedLimit
cd HiggsAnalysis/CombinedLimit
git checkout v8.2.0
cd ../..
git clone https://github.com/cms-analysis/CombineHarvester.git CombineHarvester -b 102x
cd CombineHarvester
git clone https://github.com/raspereza/bbHRun2Legacy.git bbHRun2Legacy
cd bbHRun2Legacy
git clone https://gitlab.cern.ch/adewit/bbhshapes.git shapes
cd ../..
scram b

The input shapes should be put in a directory bbHRun2Legacy/shapes and should be named htt_CHANNEL_bbH.RunYEAR.root, e.g. htt_mt_bbH.Run2018.root. The scripts currently look for the individual process shapes in the main directory (tt, em) or in the BDToutput subdirectory (et, mt). Make sure the latest shapes for each channel are always in the bbhshapes repository so that everyone has access to them.

TODO: we should make sure the latest input shapes are always available to everyone, either by mirroring them from a common afs/eos directory or by sharing them in a separate repository.

The models should be copied to HiggsAnalysis/CombinedLimit/python/

Setting up datacards

python scripts/setupCards.py [options]

The default channel is tt (set e.g. --channel tt,mt to run the tt and mt channels, or --channel all to run all 4 channels.

The default year is 2018, can switch this to other years but only one year will run at a time for the moment. If you run the script multiple times in succession, switching years in between but not changing the output directory, the datacards for each year will all end up in the correct place (/places).

Make the workspace

combineTool.py -M T2W -i output/cards/* -P HiggsAnalysis.CombinedLimit.bbHModel:bbhModel -o ws.root - this will make the workspace for all subdirectories. When running only one channel 'cmb' should equal that channel, otherwise there can be differences.

It is also possible to just run

combineTool.py -M T2W -i output/cards/cmb -P HiggsAnalysis.CombinedLimit.bbHModel:bbhModel -o ws.root to produce the workspace only for the combined directory.

This will create workspaces with the default physics model, which creates one POI (r) which will scale both signal processes (ggH+2b and bbH). The interference term is scaled with -r.

For the kappa model run:

combineTool.py -M T2W -i output/cards/* -P HiggsAnalysis.CombinedLimit.LHCHCGModels:K1Int -m 125.38 -o ws_kappas.root

Calculate limits

combineTool.py -M AsymptoticLimits -d output/cards/cmb/ws.root --there --run blind

Note the --run blind is important - we don't want to accidentally unblind the analysis.

Impacts

To run the impacts one needs to run first an initial fit: combineTool.py -M Impacts -d output/cards/mt2018/ws.root -m 125.38 --robustFit 1 -t -1 --expectSignal 1 --doInitialFit

Next the scan is done for each nuisance parameter: combineTool.py -M Impacts -d output/cards/mt2018/ws.root -m 125.38 --robustFit 1 -t -1 --expectSignal 1 --doFits

Once the jobs are completed the output can be collected with: combineTool.py -M Impacts -d output/cards/mt2018/ws.root -m 125.38 -o impacts.json

and plotted with

plotImpacts.py -i impacts.json -o impacts

MultiDimFit

The MultiDimFit for kappa_b and kappa_t is run with:

combineTool.py -M MultiDimFit -d output/cards/cmb/ws_kappas.root --there --redefineSignalPOIs kappa_b,kappa_t --setParameters kappa_b=1,kappa_t=1 -m 125.38 --algo grid --points 300 -t -1

The 2D plot for kappa_b and kappa_t can be produced with:

python scripts/plotMultiDimFit.py output/cards/cmb/higgsCombine.Test.MultiDimFit.mH125.38.root

Useful for debugging - just run a single fit (turn up the verbosity with -v 3 to get more detailed fit logs from minuit).

combineTool.py -M MultiDimFit -d output/cards/cmb/ws.root --there -t -1 --expectSignal 1 (or --expectSignal 0)

NB don't forget about the -t -1 so as not to unblind accidentally

Running FitDiagnostics

Useful for making post-fit plots of the signal region later / inspecting NP correlations&constraints without running the full impacts.

combineTool.py -M FitDiagnostics -d output/cards/cmb/ws.root --there -t -1 --expectSignal 1 (or --expectSignal 0)

NB don't forget about the -t -1 so as not to unblind accidentally