ROOT gets stuck during fitting a user defined function

ROOT
1

Hi!

I’ve the problem that ROOT sometimes get’s stuck when fitting with my user defined function which contains a numeric integration and looks like this:

[code]Double_t resonance(Double_t *y,Double_t par)
{
Double_t crosss = 0, lambda;
Double_t vf = 0;
if (par[0] != 0)
{
lambda = (y[0] - 2par[0]) / par[1]; //y[0]=sqrt(s)
vf = sqrt(par[0] * par[1]); //par[0]=mr,par[1]=gamma

  static TF1 *f = new TF1("f","sqrt(([2] - 2 * [0]) / [1] - x) / ( x * x + 1)");
  f->SetParameters(par[0],par[1],y[0]);
  crosss = f->Integral(-1000,lambda);
  crosss = par[2]*crosss * vf / TMath::Pi()+par[3];
  //cout <<lambda<<endl;
}

return crosss;
}[/code]

Is there a way to tell ROOT to interrupt the fit procedure after a particular number of iterations or something like that?
I know this might not be the smartest way to implement the function… does somebody have a better idea instead of using a TF1 for integration? In principle it even gets worse, cause I need this function convoluted with a Gaussian in addition. That would add another integration… and then fitting really gets slow!

Can somebody help me on this?

Regards and thanks,
Klaus

2

Could you send the shortest possible RUNNING script reproducing this problem?

Rene

3

Hi Rene!

I attached the script which I run with ROOT 5.12/00. Whether the fit gets stuck or not strongly depends on the parameter values N and B. Usually it stucks more often when B>N. Actually I don’t know whether there really is an endless loop or I would have to wait much longer.

The last printout on screen is something like:

PARAMETER DEFINITIONS:
NO. NAME VALUE STEP SIZE LIMITS
1 m_{R} 2.31814e+03 6.95443e+02 no limits
2 #Gamma_{R} 6.33272e-01 1.89982e-01 no limits
3 A 2.12923e+00 6.38769e-01 no limits
4 BG 9.19519e+02 2.75856e+02 no limits

** 5 **SET PRINT 0

** 6 **MIGRAD 5000 0.011

and this takes forever.

Cheers and thanks,
Klaus
testcross.C (1.26 KB)