RooFFTConvPDF constant to zero

Dear Expert,

I am trying to perform a fit using as model an exponential PDF convoluted with a double gaussian. Here is my code

 RooRealVar t_val("t","t",22.15,22.65);
  RooRealVar sigma("#sigma","#sigma core",0.015,0.,0.03);
  RooRealVar sigma1("#sigma1","#sigma tail",0.2,0.,0.4);
  RooRealVar mean("#mu","#mu",22.36,22.35,23.38);
  RooRealVar alpha("#alpha","#alpha",-25,-30.,0.);
  RooRealVar frac1("fraction1","fraction",0.8,0,1);
  RooExponential Expo("Expo","laser flash",t_val,alpha);
  RooGaussian Gauss("Gauss","ph distribution",t_val,mean,sigma);
  RooGaussian Gauss1("Gauss1","ph distribution",t_val,mean,sigma1);
  RooAddPdf res("resolution","resolution function",RooArgList(Gauss,Gauss1),RooArgList(frac1));
  RooFFTConvPdf conv("conv","convolution",t_val,Expo,Gauss);
  RooDataHist jitter_data("data","dataset",t_val,Import(*hdata));
  RooPlot* frame=t_val.frame();

 // sigmaG.setConstant();
 //RooFitResult *r=conv.fitTo(jitter_data);


the problem is not the fit itself, but the fact that when I try to plot just the convolution function, even without fitting, I get a constant PDF with 0 value. I tried to do the convolution with one gaussian and between the two gaussian, but the result is the same. I think I am doing a systematic error I can’t figure out. On the other hand I tested the model with this code

RooRealVar dt("dt","dt",22.15,22.65) ;
  RooRealVar tau("tau","tau",0.04) ;

  //Gaussian model
  RooRealVar mean("mean","mean",22.36) ;
  RooRealVar sigma("sigma","sigma",0.015) ;
  RooRealVar sigma1("sigma1","sigma",0.2) ;
  RooRealVar gm1frac("gm1frac","fraction of gm1",0.8);
  RooGaussModel gaussm("gaussm","gauss model",dt,mean,sigma) ;
  RooGaussModel gaussm1("gaussm1","gauss model",dt,mean,sigma1) ;

  RooAddModel gmsum("gmsum","sum of gm1 and gm2",RooArgList(gaussm,gaussm1),gm1frac);
  //Create basis function
  RooFormulaVar basis("basis","exp(-@0/@1)",RooArgList(dt,tau));
  // Construct decay(t) (x) gauss1(t)
  RooAbsReal* decayConvGauss=gmsum.convolution(&basis,&basis);
  RooPlot* frame1 = dt.frame(Title("Bdecay (x) resolution"));
  // Plot p.d.f. 
  TCanvas *c2=new TCanvas();

and I saw that the curve I want to see is corrected described.

Thanks for helping.



Note: I enclosed the “code parts” of your post with triple back quote to make it more readable.

Hi @emarip - we will need our RooFit expert @StephanH to be back from travel to have a good answer! He should be back in a week. Please ping us should you not have received a reply by end of next week. Thanks!

Hi @emarip,

can you plot both the addition of the gaussians and the exponential before the convolution? Maybe the parameters are not chosen correctly.

yes I can plot them separately. I fear that the problem is due to the fact that the exponential is practically 0 at the gaussian mean. In this case is it possible to shift the exponential pdf?



You can shift by not giving x as on observable to exp, but giving x - a, where a is a constant parameter that denotes the shift strength.


thanks. You mean by using a rooformulavar?




A RooAddition("name", "title", RooArgList(x, a)) should also work.

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