ROOFIT low statistics fits with potential negative PDF value

Dear All,

I’m am attempting to understand how ROOFIT will behave in a particular low stats situation and would appreciate some advice. As you will see from the attached plot (in particular the red curve) I have situations where the functional form I am fitting can technically be negative. However, this is unphysical given we can’t have a negative number of events. With more statistics the shape could cover the entire range without these gaps in the data.

Could someone please advise on how ROOFIT deals with such a case internally? i.e. does the pdf evaluate to zero in these cases or is it allowed to extrapolate into the negative region even if when plotted this is not visible?

testplot.eps (19.1 KB)

Many thanks