Fiducial XS measurements in HZZ4L

Standalone framework for fiducial differential cross section measurements using CJLST TTrees for Run 2 data.

A CMSSW working area with the latest version of combine is required:

# cmssw-el7 (maybe needed)
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 fetch origin
git checkout v8.2.0
scramv1 b clean; scramv1 b
cd ../..

Once that is done, combine is properly installed. The fiducial framework should be properly set up:

git clone https://github.com/AlessandroTarabini/FiducialXSFWK.git
cd FiducialXSFWK
source ./env

The PhysicsModel(s) used in this analysis (cf. fit/createDatacard.py) can be found in the models folder and are copied automatically when source ./env.sh is called to $CMSSW_VERSION/src/HiggsAnalysis/CombinedLimit/python. If when running fit/RunFiducialXS.py the fit crashes because the physics model is not found, you should check that the HZZ4L_Fiducial*.py files are present in your $CMSSW_VERSION/src/HiggsAnalysis/CombinedLimit/python folder.

Set Up

For all the imports in the various scripts to work properly, set up the working environment with:

cmsenv
export PYTHONPATH="${PYTHONPATH}:<PATH_TO>/FiducialXSFWK/inputs"
export PYTHONPATH="${PYTHONPATH}:<PATH_TO>/FiducialXSFWK/helperstuff"

Workflow

A schematic representation of the framework's workflow is given in the two following sketches:

The input files of the analysis workflow are the HZZ4L ntuples generated with the CJLST framework. This framework starts from those files and:

1. config: Starting from CJLST TTrees, only relevant branches are selected and stored by skim_MC_tree.C and skim_data_tree.C macros.

Having created these skimmed TTrees, the next steps of the analysis involve the caluclation of the different coefficients needed for the pdf parameterisations and unfolding, as well as the creation of background templates. To do so:

2. templates: Templates and normalization coefficients for the backgrounds' pdf are extracted from MC (ggZZ and qqZZ) and data (ZX) using RunTemplates.py
3. coefficients: All the coefficients of the signal parameterization are calculated with RunCoefficients.py and stored in inputs folder.
4. fit: The maximum likelihood fit is performed. This step relies on the RunFiducialXS.py script and it can be run either as part of the entire framework, creating the datacards and workspaces from scratch, or using pre-existing datacars as input. Datacards are produced and stored in a datacard directory, while fit results (combine .root files) are stored in combine_files folder.

Additional scripts are provided to plot negative log-likelihood scans and to produce the usual differential xsec plots:

5. LHScans: Likelihood scans are plotted, best-fit values and the corresponding uncertainties are calculated using plotLHScans_compare.py.
6. producePlots.py: Plot of unfolded differential xsec distributions.

Commands: a gentle introduction

The frameworks picks up the variables to be used in a measurement and the binning to be used in the fit thanks to the dictionaries available in helperstuff. More in detail, to define a new measurement (or to understand what is used in a current measurement):

• Define the variable name and the reco- and gen-level observables modifying the observables dict in helperstuff/observables.py. The syntax is:

   observables = {NAME: {"obs_reco": TBranch name, "obs_gen": TBranch name}}
  

• Define the binning in helperstuff/binning.py following the conventions defined here for 2D measurements (for 1D measurements the binning definition is intuitive).

1. Skimming of the TTrees: TODO. Usually provided centrally

2. Creation of the templates: in templates run (the framework detects automatically the binning and the observables to use):

   python RunTemplates.py --obsName NAME (str) --year YEAR (str)
   

3. Computation of the coefficients: in coefficients run (the framework detects automatically the binning and the observables to use):

   python RunCoefficients.py --obsName NAME (str) --year YEAR (str)
   

4. Run the fits: in fit run:

   python RunFiducialXS.py --obsName NAME (str) --year YEAR (str) 
   

5. Plot the NLL scans: in LHScans:

   python plot_LLScan.py --obsName NAME (str) --year YEAR (str)