Re: Problem with MINOS when the derivative around the minimum is discontinuous (TMinuit)

Hi Pierre,

  I don;t think there is much one can do. I would not trust the result of Migrad very much if the derivative are not continuos at the minimum. Also the derivative should approach zero, if they have a value of 1e4 is not a good sign. I would try to perform a scan around the minimum to understand better the shape of the likelihood function.

 Best Regards

 Lorenzo
On Jul 17, 2010, at 6:58 AM, Pierre-Luc Drouin wrote:

> Hi,
>
> I am having troubles with MINOS when I am minimising a likelihood which has a discontinuous derivative at the minimum for a given parameter. I call MIGRAD twice followed by MINOS. Strangely, Minuit does not generate any warning and MIGRAD convergence is successful (although the step size goes down to 7e-14 for that parameter and the first derivative to 1e4 while the fitted value is 2.5e-7), but MINOS returns a "PROBLEMS" status and returns only the parabolic error for that parameter.
>
> I have tried the strategies 0, 1 and 2 but they all produce the same status at the end. Is there some kind of trick with Minuit to avoid that kind of problem?
>
> Thanks!
>
Received on Tue Jul 20 2010 - 10:21:41 CEST