CalculateBreakdown

This action performs a breakdown calculation, computing a breakdown of systematical and statistical uncertainties on your POI by source of the uncertainty.

There are two variations of the computation, the "normal" and the "inverted" one.

• In the "normal" way, two fits are run for each systematic uncertainty: a fully unconditional one with all NPs floating, and one that is partially conditional on this uncertainty (with the respective NPs fixed to their nominal values). Then, a square-difference between the two fits of the uncertainty on your POI is computed.
• The "inverted" calculation works the same way, but compares a fully conditional fit (with all NPs fixed to their nominal values) with a partially conditional one (with the NP in question floating and all other NPs fixed to their nominal values).

Usage:

+CalculateBreakdown {
 +Morphing {
    # compute breakdown for the fit to asimov
    +asimov { 
      <datasetName="asimovData_SM">
      <invert=false>
    }   
    +asimovinvert {
      <datasetName="asimovData_SM">
      <invert=true>
    }   
    @ ? {
      # the "singles" will compute the breakdown for each individual NP matching the filter
      <singles.SYS = "alpha_ATLAS_*">
      <singles.THEO = "alpha_theo*">                                                                                                            
      <singles.MCSTAT = "gamma_*">
      <singles.NORM = "ATLAS_norm*">
      <singles.FAKE = "HWW_FakeFactor_*"> 
      # the "group" will compute the breakdown for the entire group of NPs matching the filter
      <group.MCSTAT = "gamma_*"> 
      <group.SYS = "alpha_ATLAS_*", config.group.SYS.except = "ATLAS_norm*">    
      # since statistical uncertainties are not modelled explicitly, we compute them as the complement of all systematic unceratinies
      <group.STAT = "*", config.group.STAT.invert = true>                                                                                                 
    }
    <fit.startingStrategy=1,fit.printLevel=1,fit.numRetry=3, fit.offset=true, fit.verbose=1> @ ?;
  }
}

As this performs a series of fits, it can take quite long to run. Hence, there is a batch-parallelized version of this action, WriteCalculateBreakdown in ActionBatchCalculate.cxx. A different method of assessing the effects of different systematical uncertainties also exists, see CalculateImpacts.