Minuit fit problem with large statistics

Roofit and RooStats
17

Hi @moneta,

I managed to setup a test code that reproduces what I am doing.

Within fitTest, the fN parameter allows to set the normalization coefficient to reproduce the expected statistics for data. There are two values, one for low statistics, where the fit is working (the likelihood profiles are as I would expect them), and one for high statistics, where the fit is showing the behaviour I discussed.

In Chi2FCN there is the possibility to use a simple chi2 or the mu -n + n*log(n/m) (did not manage to make mu - n*log(m) converge).

Thanks,
Antonio

test_code.tar.gz (8.3 KB)