Fit return warning message: large likelihood value

[Riffard]

Hello,

I’m using the PLR procedure with RooStat and RooFit and I have a very concerning issue.

I’m defining a likelihood with external constrains on some nuisance parameter and during the first fit of the PLR, I receive this message:

[#0] WARNING:Eval -- RooAbsPdf::getLogVal(Constraint_Neutrino) WARNING: large likelihood value: 2.5168e+07

I’m defining the external constrain in this way:

RooRealVar N(Form("N_%s", contrib_name), Form("Number of %s bkg events", contrib_name), Ncount, 0, Ncount*10);  
N.setVal(Ncount);                                                                                               

RooRealVar N0(Form("N0_%s", contrib_name),Form("Estimated number of %s bkg events from aux measurement", contrib_name), Ncount);         
N0.setConstant(1);                                                                                                                       
                                                                                                                                         
RooRealVar sigmaN0(Form("sigmaN0_%s", contrib_name), Form("Absolute error on expected %s BG", contrib_name), activitySigma * Ncount);    
sigmaN0.setConstant(1);                                                                                                                  
    
fw->import(N0);            
fw->import(sigmaN0);       

fw->factory(Form("Gaussian::Constraint_%s(N_%s, N0_%s, sigmaN0_%s)", contrib_name, contrib_name, contrib_name, contrib_name)); 
                                                                                                                                     

The value of Ncount is ~10-7 and sigmaN0 = 0.2* Ncount

Any idea where this could come from ?

[moneta]

Hi,

Probably you are evaluating the constraints many sigma away and it is normal that the resulting log-likelihood is very large since the pdf (the gaussian constraint) is almost zero.
The message is not normally a problem if it is coming during the fit, since this can be normal that during the minimisation procedure such points are tested

Lorenzo

[Riffard]

The warning message is not giving enough information about the value that is tested. Is there a way to check that ?

[StephanH]

You can try RooFit::Verbose(true) as an argument to fitTo(). If I remember correctly, it will print the current function value and the parameters that have been updated in each minimisation step.