Dear Experts,
I would like to ask you what is the difference between Minuit and Minuit2 minimizers.
The reason I’m asking this question is because:
if I run my RooFitter with the option Minimizer(“Minuit”), then the fit does NOT CONVERGE,
instead if I run my RooFitter with the option Minimizer(“Minuit2”), then the fit does CONVERGE.
To be noticed:
(a) I use MINOS in my minimization ( Minos(true) )
(b) By the word “CONVERGE” I mean that the minimizer reached “0” AND the covariance matrix is positively defined
I’m attaching two files with the screen output of the two cases.
If you look correctly at the output log, you see that both work in the same way. The fit for the minimum converged in both case and both they have a problem computing the error for the first parameter (the parameter “AfbS”).
I think the only problem I see is in the returned code propagated to RooFit, which flags as the fit invalid which is not the case. There was only a problem in computing the Minos error for one parameter.
So, you can trust both, but in both case, you should look carefully at the output produced.
From my current experience and from what I have see, there are however some cases where Minuit fails and Minuit2 works, due to some issues which have been fixed only in the new version.
In general if one does not work is good to try with the other one and the ultimate rule, is that the one who has the lowest function minim value is the one to choose for the final result