About: Abnormal termination of minimization

ROOT
,
1

Hi, Rooters,

My code is about the two-dimension fit, where use my custom function fun2 to fit TH2F *h2

And there are four parameters. I don’t know why the fit result is always status:FAILED.
Can you help me to find the wrong of my code?

My code is here:

#include "TF2.h"
#include "TH2.h"
#include "TMath.h"

         
Double_t fun2(Double_t *x, Double_t *par) {
    //par[0]=N    kappa = par[1]    longp = par[2]    trans = par[3]
 double re = 0.665/(8*TMath::Pi()/3);
 double dsig0 = re/(par[1]*par[1]*pow((1+x[0]),3))*(par[1]*(1+pow((1+x[0]),2))-4*x[0]/par[1]*(1+x[0])*(par[1]-x[0]));
 double dlongi = re*par[2]/(par[1]*par[1]*pow((1+x[0]),3))*x[0]*(x[0]+2)*(par[1]-2*x[0]);
 double dtrans = -par[3]*re/(par[1]*par[1]*pow((1+x[0]),3))*2*x[0]*sqrt(x[0]*(par[1]-x[0]))*TMath::Sin(x[1]);

 return  par[0]*(dsig0+dlongi+dtrans); 
   //  return par[0]*dsig0;
}   


void fit2() {
   //gStyle->SetOptFit("1111");

   const Int_t npar = 4;
   Double_t f2params[npar] = {26330,2.2794,0.05,0.1};

   TF2 *f2 = new TF2("f2",fun2,0,2.2794,0,2*TMath::Pi(), npar);
   f2->SetParameters(f2params);
   f2->SetParLimits(0,25000,28000);
   f2->SetParLimits(1,2.268,2.29);
   f2->SetParLimits(2,0.045,0.055);
   f2->SetParLimits(3,0.095,0.103);

   //Create an histogram and fill it randomly with f2
   TH2F *h2 = new TH2F("h2","from f2",200,0,2.2794,100,0,2*TMath::Pi());
   Int_t nentries = 1E7;
   h2->FillRandom("f2",nentries);
	
  h2->GetXaxis()->SetTitle("X Axis");
  h2->GetYaxis()->SetTitle("Y Axis");
  //ROOT::Math::MinimizerOptions::SetDefaultMaxFunctionCalls(1000000);
  //ROOT::Math::MinimizerOptions::SetDefaultTolerance(1);
   h2->Fit("f2","V");
   h2->Draw("SURF2");
   //f2->Draw("SURF same");
   cout<<"Inegral of f2: "<<f2->Integral(0,2.2794,0,2*TMath::Pi())<<endl; 
   
}

ROOT Version: 6.20/02

3
3763×154 41.3 KB

Best Regards,
hurricane.

2

If you use a smaller initial value for the first parameter and remove the ParLimits, the fit converges, e.g. doing this:

   Double_t f2params[npar] = {2633,2.2794,0.05,0.1};  // up to par[0] ~ 12900 seems to still work
   //...
/*
   f2->SetParLimits(0,25000,28000);
   f2->SetParLimits(1,2.268,2.29);
   f2->SetParLimits(2,0.045,0.055);
   f2->SetParLimits(3,0.095,0.103);
*/
  //...

I’ll let you check if the fit is good, and also leave the explanation to the experts (I can be very wrong here but it seems like the problem could be a combination of: your initial values being too close to the real ones, and the large difference in scale between par[0] and the others).

1 Like
3

OK, thank you for your reply.
And I check the fit result-------if the data is the sample by:

Int_t nentries = 1E7;
h2->FillRandom(“f2”,nentries);

the fit result is very well.

But if the sample is set by acceptance-rejection sampling, the fit value has a large deviation from true value when one don’t set the Parameters Limits.
So it still puzzles me…

4

What do you mean exactly with

are you sure you are sampling your distribution correctly ?

Lorenzo

5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.