Correlations between NormSys modifiers

[marfari]

Hi!

Given two samples (signal and BDDKp) I have added a NormSys modifier to both samples in this way:

sig_modifiers.append({"data": {"hi": upward_eps_ratio_sig, "lo": downward_eps_ratio_sig}, "name": "alpha_eff_ratio", "type": "normsys"})
BDDKp_modifiers.append({"data": {"hi": upward_eps_ratio_BDDKp, "lo": downward_eps_ratio_BDDKp}, "name": "alpha_eff_ratio", "type": "normsys"})

The uncertainty that I am trying to describe is correlated between the signal and B->DDK+ samples, but the value of the uncertainty is not the same for the two samples

I have given the same name to both modifiers with the intent to share it between the 2 samples (100% correlation) but I provide different errors for the two samples (upward_eps_ratio_sig / upward_eps_ratio_BDDKp)

Will pyhf take into account the two errors? Or will it ignore one of them?

[kratsg]

This has nothing to do with pyhf, it's about histfactory. The correlation only applies to the parameter controlling that modifier. In this case, the impact of the parameter will be different for each value controlled by the configuration data in the modifier, and pyhf won't ignore that. If one has a hi of 2 and the other has a hi of 2.5, it will take both into account for the corresponding sample.

[alexander-held]

To be very explicit: yes, pyhf does what you intend here and one shared parameter controls fully correlated variations of both samples.

[marfari]

@kratsg @alexander-held thank you a lot for confirming!