analysis

Physics analysis

In this component the physics analysis is performed, starting from files with plain ROOT trees produced in events. Various plots with comparisons between data and simulation are produced, and input files for the global fit are prepared.

Environment

Dependencies:

• Python 3.6
• NumPy 1.14
• Matplotlib 2.2
• ROOT 6.16

As in component events, it is recommended to use LCG_95apython3 environment. From /cvmfs/, it can be set up with the following command (change the architecture if needed):

. /cvmfs/sft.cern.ch/lcg/views/setupViews.sh LCG_95apython3 x86_64-slc6-gcc8-opt

On top of this, file env.sh needs to be sourced.

Comparison between data and simulation

Most important p_T distributions and mean values of the balance observables as a function of p_T of the leading jet are plotted with script plot_data_sim.py. It also produces control plots with distributions of balance observables in bins of p_T of the leading jet. These are useful to check for possible outliers in the tails of the distributions.

Here is an example command to produce the plots:

plot_data_sim.py --data JetHT-Run2016*.root --sim QCD-Ht-*.root \
    --config "$config_dir/plot_config.json" --era 2016All -o fig/2016All

A number of settings, including the binnings, are read from the configuration file.

Inputs for the fit of L3Res corrections

Input files for the fit of L3Res jet corrections are constructed in two steps. First, the nominal inputs are created with script build_fit_inputs.py, as in the example below:

build_fit_inputs.py --data 2016BCD.root --sim simulation.root \
   --binning "$config_dir/binning.json" --era 2016BCD --plots fig/2016All \
   -o multijet.root

The produced ROOT file contains the following entries:

• Binning: Binning in p_T of the leading jet (hereafter denoted with τ₁) to be used for the χ² in the fit.
• PtBalProfile, MPFProfile: Profiles of the two balance observables as a function of τ₁.
• PtBalThreshold, MPFThreshold: Reference values that define smooth jet p_T thresholds used for the two balance observables.
• PtLead, PtLeadProfile: Distribution of τ₁ and profile of τ₁ versus τ₁.
• RelPtJetSumProj: Matrix S used at slide 8 here.

All the histograms and profiles above are filled with data. Information about simulation is provided in in-file directories PFJet*. Each of them corresponds to one trigger bin and contains the following entries:

• Range: Range in τ₁ for that trigger bin.
• SimPtBal, SimMPF: Mean values of the two balance observables in simulation versus τ₁. They are represented with splines.

In the current version of the code, the splines describing ⟨B⟩(τ₁) in simulation are identical between all trigger bins.

Script build_fit_inputs.py reports in the standard output under-populated bins. Typically, bins with τ₁ > 1.6 TeV are affected. These must not be considered in the fit because uncertainties become not reliable for them. The script also produces diagnostic plots that illustrate the quality of the fit of ⟨B⟩(τ₁) in simulation with splines.

Variations due to systematic uncertainties are constructed with script build_syst_vars.py as in

build_syst_vars.py $config_dir/syst_config.json \
    --binning $config_dir/binning.json --plots fig/2016All -o syst.root

The variations to be produced are described in the dedicated configuration file (example). They can be represented with a dedicated set of input files (created by supplying appropriate flags to the program multijet from component events) or with a set of additional per-event weights. Uncertainties that affect data or simulation are stored in the output file syst.root in different ways. In case of data, ROOT histograms representing up and down relative deviations from nominal ⟨B⟩ are stored in the root of the output file. In simulation, similar relative deviations are described with splines and saved in in-file directories for different trigger bins. If some uncertainty affects both data and simulation, both types of entries are saved.

Saving systematic variations in the form of relative deviations in a separate file allows to include uncertainties from an older version of the analysis in case up-to-date uncertainties are not available for some reason. Of course, this is only an approximation, and the uncertainties need to be recomputed to obtain reliable results.

In addition to storing the (smoothed) relative variations in a ROOT file, script build_syst_vars.py also plots them, together with the inputs that have been fitted to produce the variations.

Finally, the files produced by the two scripts above are combined using hadd. The resulting file is used in the fit of the L3Res correction, see the dedicated repository.