MINUIT problems with discontinuos functions

ROOT
1

Hello all,
I use TMinuit (and MIGRAD) for doing some fitting. Now, if the fitting function has a derivative which is discontinuos then at least the errors which are reported are incorrect (way too small). While the minimum looks real and sensible I fear that in some cases it might be wrong.

A typical function would be (but also see broken power-laws etc):
if(x<par[2]):
return par[0]
else:
return pow(1+x)^par[1]

Are you aware of any solution for that ? Is there are a better minimizer ? How should I compute the errors ? Any help is greatly appreciated!
thanks
marco

2

Hi,

a potential problem is in the discontinuity of the derivatives with respect to the parameters of your chi2 or likelihood function.
It depends where the discontinuity is. If it is close to the minimum, there could be a problem, otherwise if it is far away it is fine.
I would make a scan of the likelihood around the minimum and eventually try to run MINOS.

Best Regards

Lorenzo