I was trying to find out whether RooAbsPdf::fitTo(…) uses analytic derivatives wherever possible or not.
As far as I can see does RooAbsPdf::derivative(…) return an object which seems to do numerical derivatives only. Also, I suspect that during fitting, numerical derivatives of the goal function (e.g. log likelihood) are calculated numerically by the fitting code. Is that correct ?
Analytical derivatives of the log-likelihood function are not used when fitting in RooFit. If you want to use it you need to implement them and use the ROOT::Math::Minimizer interface for minimizing the log-likelihood function