Problem with convergence if fit is performed in range

[pkalbhor]

Hi experts,
I have a code which performs an extended likelihood fit on a data sample containing 4 yield components. When I perform a fit in the full range of mass observable, I get good convergence as shown here.

But when I perform a ranged fit(hiding signal region) of data, the fitter fails to achieve convergence.
I have attached a code here. Please help me identify the issue.

To perform a ranged fit I am following this tutorial example: rf204b__extendedLikelihood__rangedFit_8C.html

Attachement: RangedFit.cc (8.7 KB)

[pkalbhor]

When fit is performed in range, gives this plot:

But it is failed fit. It looks good just because i have taken initial parameters close to minimum. Please let me know if you require additional info.

Note: Try running attached file in lxplus.

[moneta]

Hi,

You are having some problem with a pdf in your model, I think this is due to your generic pdf, pdf_errf. The computation of its integral results in a NaN.
You should not disable the error messages, because they are often meaningful, and will help you understanding the problem such as why the fit is not converging

Lorenzo

[pkalbhor]

Thank you @moneta for your reply.
I would like to point out that the same PDF works well when RooFit::Range() argument is removed(Changing rangeFlag from line#19 of the attached script).
Do you think of any workaround?

[moneta]

Hi,
This can happen if the integral of the pdf has problems in the restricted range, but works fine in the full range. I would try to understand if that pdf has no particular implementation issues in the restricted range, and if not, you can try one of these:

• use a different numerical integration algorithm. For this see this tutorial: ROOT: tutorials/roofit/rf901_numintconfig.C File Reference
• provide an analytical integral for your pdf

Lorenzo