Wrong Yield from extended maximum likelihood fit in RooFit

[amroo]

Hi all,

I am trying to do an extended maximum likelihood fit in Roofit. I am not able to make sense of my results. I am fitting a dataset with some pdf that shouldn’t be able to fit it. peaking distribution (dataset) and 2nd order Chebyshev polynomial is pdf used. I don’t expect the fit to converge and also expect the yield returned from fitter to be completely different from the observed number of events (in this case size of the dataset). But this is what I get:
1> the fit is very bad as expected but MIGRAD did converge. Why ?
2> the yield returned from extended maximum likelihood fitter is exactly equal to the number of events read from dataset. How??

If anybody knows what’s happening here please help? I am attaching the dataset and also the fitting script usedsignal_new.txt (321.7 KB)
combinatoric_mdz.C (4.5 KB)

Thanks
Aman

[moneta]

Hi,

1. The fact that MIGRAD converges does not mean that your fit is good. It could be that this is the best possible solution that is found. However, you see from Migrad the message that the first two parameters are at the limit. Having a solution at the boundary is not normally good and this could also have confused Migrad.

2. The third parameter ((yield) did not change from the initial given value. This can be explains from the fact that the derivative for that parameter is very small compared to the others. So if you probably input a different initial value you will get that value.

Lorenzo