Toy studies problem (large likelihood value)

[Jacopo]

Hi all,

I am doing toy studies on a binned ML fit using the RooMCStudy class and when calculating the pull on my parameter of interest I get many warning messages:

[#1] INFO:NumericIntegration -- RooRealIntegral::init(pullGauss_Int[Signalpull]) using numeric integrator RooIntegrator1D to calculate Int(Signalpull)
[#1] INFO:Minization -- RooMinimizer::optimizeConst: activating const optimization
[#0] WARNING:Eval -- RooAbsPdf::getLogVal(pullGauss) WARNING: large likelihood value: 2.994e+07
[#0] WARNING:Eval -- RooAbsPdf::getLogVal(pullGauss) WARNING: large likelihood value: 2.96e+07
[#0] WARNING:Eval -- RooAbsPdf::getLogVal(pullGauss) WARNING: large likelihood value: 2.712e+07
[#0] WARNING:Eval -- RooAbsPdf::getLogVal(pullGauss) WARNING: large likelihood value: 2.735e+07
…

Moreover the plot looks quite bizarre:
Toy_3pi3pi_SecondStrategy_1000_Normal_sigShape_0Signal_0.7max_0.0min-3.pdf (21.3 KB)

Do you have any idea why?

Cheers,
Jacopo

[couet]

I guess @moneta can help you.

[moneta]

Hi,

The reason of the large likelihood value is probably because the pdf gets evaluated in bad parameter space regions. This might cause bad fits which give you crazy parameter values.
I would try constraints your apply stringent and stronger fit parameter constraints (limits in the RooRealVar objects ) that could help the toy fits to work and converge

Cheers

Lorenzo