Hello.
Solely for testing purposes I constructed two vectors of points X and Y such that X = Y.
I try to fit Y = a*X + b using chi^2 minimization (error on each chi^2 element is VAR(X) + VAR(Y)).
I would expect a = 1 and b = 0, since X = Y. This is what I get from GSL minimizers (although they sometimes complain about “not making progress towards solution”). But when I try with Minuit2 it sometimes fails. I mean chi^2 is still very small but it is not entirely what I would expect, definitely not in comparison with GSL minimizers.
I understand that this is sort of extreme, but is there a reason why should GSL algorithms perform better than Migrad and Simplex in this particular case?
Can you please post the code you have been using and the data where Minuit2 fails ? I need to look at your example in detail to understand this problem.
Can you please post the code you have been using and the data where Minuit2 fails ? I need to look at your example in detail to understand this problem.
Best Regards
Lorenzo[/quote]
Hi!
Thanks for your quick response. I’ve discovered the problem. It was in my variance. The error was vastly exaggerated. Obviously, chi^2 is sensitive to this and I couldn’t expect correct results. Interestingly, GSL algorithms are probably less sensitive to this than the Minuit’s algorithms…
