FCN of maximum likelihood fit in roofit

Roofit and RooStats
1

Dear All,

In roofit, for the maximum likelihood fit, the FCN should be -log(L) < 0, right?

But for the fit of the non-extended model, there is FCN>0 (as the roofit example: rf101_basics.C ), what’s the problem ?

For the extended model, there is FCN<0 (as the roofit example: rf202_extendedmlfit.C), it is correct. Thanks!

2

Hi,

The overall likelihood value is not relevant, there could be some constant term added or removed from the likelihood. Also in case of pdf which give very small value for some x, could make the overall negative log-likelihood to be positive.
So I would not worry at the overall FCN value. What counts is the difference w.r.t something (e.g. the minimum).

Lorenzo

3

Thank you very much. I understand.

[quote=“moneta”]Hi,

The overall likelihood value is not relevant, there could be some constant term added or removed from the likelihood. Also in case of pdf which give very small value for some x, could make the overall negative log-likelihood to be positive.
So I would not worry at the overall FCN value. What counts is the difference w.r.t something (e.g. the minimum).

Lorenzo[/quote]