Large, negative values of FCN when fitting RooAddPdf


I’m trying to fit a 3rd order polynomial to a very simple-looking (though high-statistics and finely-binned) data set. I construct an extended likelihood model with a signal pdf represented by a Gaussian and a background shape given by a polynomial. The fits look fine and MINUIT returns 0 and 3 for status and covariance matrix quality, respectively, but the value of FCN is very large and negative; please see the attached code. As you can see, this happens whether I use a regular, Bernstein, or Chebychev polynomial. It also occurs whether or not I use a binned or an unbinned data set.

Does anyone have any idea why this is so? Am I setting something incorrectly so that MINUIT chooses something other than -log(L) for FCN?

Thanks in advance for any help you can offer.

simpleWindow_1.C (1.84 KB)