RooFit - unbiased limits for fit parameters?

Hello roofitters,

I’m using RooFit to fit a voigtian signal + exponential background to an unbinned data set.

There are 6 parameters: nsig, nbkg, exp, mean, width and sigma(gaussian resolution).

Is there a way to give RooFit only the initial guess without constraining the parameters, such that the fit will be efficient and will return a sane result ?

I tried removing the limits from the RooRealVar parameter definition, leaving just the initial guess, but then the fit simply gave me my initial guess with errors=0 for all 6 parameters.

The problem is that to get a good fit result with the most loose parameters-limits, I’, still feeling very biased since I happen to know the “correct” result.

btw, a simple chi2 (binned) fit to the same distribution, using ROOT, gives a perfect result, when supplied only with the initial guess without any limitations on the parameters and without defining any step sizes.


It is usually helpful for RooFit if you also define the initial errors as well as the values for parameters. Alternately you can run HESSE before MIGRAD and that will give you an estimate given your initial parameters.