[RooFit] model with constraint, likelihood vs parameters

Roofit and RooStats
1

Hello

I am having difficulties in finding the dependence of likelihood on model parameters.

The PDF has one observable and 2 parameters. One of the parameters is constrained using conditional PDF. The model is fit do data and I would like to study the dependence of -ln(L) on parameters. The fit gives reasonable results however the plots of -ln(L) vs parameters contradict the fit. The plot of various distributions are attached together with essential parts of the code. The last pad shows RooMinuit contour plot, which contradicts -ln(L) vs mu dependence shown on the first pad. The third pad show -ln(L) vs both parameter with no minimum corresponding to contour plot. The likelihood calculated “manually” shows maximum, however createHistogram does not work and contains -NAN. RooFitResult::numInvalidNLL() returns 0. Would be great if an expert could help solving this issues.

Thanks,
Margar
tmp.C (3.53 KB)

canvas1.png1596×1106 61.2 KB

2

Hi Margar,

Can you please post also the data set to be able to look into your problem ? I would like to see also the third plot
made as a contour plot. If you are fitting only two parameters should agree with the contour obtain by Minuit.

Lorenzo

3

Hi Lorenzo

the attached ROOT file contains RooWorkspace and the data has name dataST.

I have to tell that I did some toy MC studies and now have doubts about the PDF itself. It seems the combination of RooExtendPdf and RooProdPdf with conditional normalization parameter does not work as I was expecting.

Thanks,
Margar
model.root (16.7 KB)