Well, I’d like to propose that the description of how the “empty bins” are treated in the fit is changed.
For example, in http://root.cern.ch/root/html/TH1.html#TH1:Fit@1 I find:
[quote]The likelihood method has the advantage of treating correctly the empty bins and use them in the fit procedure.
In the chi2 method the empty bins are skipped and not considered in the fit.[/quote]
Well, according to the discussion here, that’s wrong.
Actually it should talk about bins with zero errors, as even if the bin is “empty” but it has a non-zero error, it will be considered in the fit. Moreover, if the bin is non-empty but it’s error is zero, it will be skipped and not considered in the fit.
That concerns the “chi2 method”. The “likelihood method” doesn’t seem to care whether the bin errors are zero or non-zero (it considers all available bins in the fit).
I don’t think I have ever found these informations anywhere in ROOT’s docs.
Of course, if “Sumw2()” has not been called, errors are sqrt(N) and so empty bins automatically have zero errors. However, if “Sumw2()” has been called, then the bin contents and bin errors are “unrelated”.
P.S. I know TH1::Sumw() doesn’t exist. I hoped you could have any idea what he meant by “Sumw()”.