Analytic derivatives in fit

ROOT
1

Hi All!

I’ve got a question about fitting.
Is there any way of setting analytically calculated derivatives, instead of calculate them each time numerically?

My thanks for your answers and corrections!

2

Hi,

Yes you can tell Minuit to use the analytical derivatives. Which interfaces are you using for fitting, TH1::Fit or the Fitter class or directly Minuit ?

Cheers

Lorenzo

3

The first one. TGraph::Fit.

4

Hi,

By using the “G” option when calling TH1::Fit, Minuit will not compute internally the derivatives. However the TF1 does not implement by default the analytical derivatives. You would need to re-implement yourself a derived class of TF1 wich re-impelment the TF1::GradientPar method using analytical derivatives.

Best Regards

Lorenzo

5

Thank you, but i still can’t understand how to re-implement it?
Something like that?:

class my_TF1: public TF1
{
public:
     Double_t Gradient_par(Int_t ipar, const Double_t* x, Double_t eps = 0.01)
     {
     	....
     }
}
{
...
my_TF1* myf = new TF1("f1","[0]+x*[1]",0,10);
th1->Fit(myf);
...
}