During my use of Roofit to fit my workspace, I found the result did not correspond to the NLL minimum in the fit range. To put it simply, there is a well-defined NLL minimum in the range of [0, pi] but the fit finds the best-fit value to be 0 if starting from 1.5, and for some unknown reason if we start from 2 the correct best-fit value can be found.
The root version I’m using is 6.28.00. For more details about the fit model, dataset and the observed fit issue, I include them in some slides and you may find them in the attached pdf.
roofit_issues.pdf (341.6 KB)
There is also a tarball attached which contains the input workspace file and an example macro to reproduce the fit issue. RooFitExtensions are needed for the fit.
roofit_issue.tar.gz (1.0 MB)
It would be great if there is any suggestion or solution for this issue. Many thanks in advance!