Hello.
I read on the documentation that option ‘G’ for a TProfile the following:

‘G’ Errors are 1./SQRT(Sum(1/sigma**2))
For example, this would be useful when all Y’s are experimental quantities
measured with different precision Sigma_Y.

say i want to evaluate the resolution of a detector, and its dependence on a quantity, say “(Pt_reco - Pt_true)/Pt_true” (and atfter that plot it against Pt, eta or phi),
my question is:
is ‘G’ the right quantity to fit a resolution ? it seems not to depend explicitly on the number of entries used …

if you are having 2D data of the spread (Pt-Pt_true) vs pt (or any other quantities) then you can profile them plotting the mean of the spread vs pt using a TProfile with the default error option. The default error option uses the error on the mean ( error = sigma/sqrt(N) )

hi,
thanks for your reply!!!
i think that the error on the mean is not exactly what i am interested in.
i mean: the spread i wish to be computed should be the detector resolution (depending on pt or whatever else) and it cannot depend on the number of entries used to evaluate it.
in principle i should find a quantity that is in agreement with the atlas tdr.
so that if after f i use as sigma parameter with a PULL distribution i should find (for the right measurements) mean=0 and sigma =1.
so i guess that the option which i’m interested in should give the gaussian variance (assuming the distr is gaussian of course
or i’m lacking something ?
thank you again