Roofit: convergence depending on number of events

Dear experts,

I’m trying a 4D Unbinned Extended Maximum Likelihood Fit using a toy sample. What I see is that the convergence of the fit changes exactly at 100 events. If my toy ntuple contains 99 events, the fit won’t converge, and I get “MINUIT WARNING IN HESSE ============== Matrix inversion fails.”. If I add one more event, then it works and gives a reasonable result. I would say there’s some parameter causing this behaviour, any idea what may be going on? I could provide the script and a couple of samples if this helps.
Thanks a lot in advance,

Xabier

Hi,

It is possible that adding one extra event make the convergence in the minimisation more stable. I would suggest to start maybe with better initial values for your fit

Best Regards

Lorenzo

Hi Lorenzo,

thanks, indeed, you were right. The ‘weird’ 100 effect was caused by the initial value I was setting (exactly at 100). Sorry for the noise