I am trying to use UnbinnedFit, but I get zero for MINOS errors unless I set UP=0.5 (i.e. 68% CL intervals).
Here (users.df.uba.ar/aia/MinosErrors.C) a simplified script to illustrate the problem, it just fits 5 events to an exponential.
With UP=0.5 all is OK: the parabolic error equals the Migrad Error and MINOS errors are reasonable.
With UP=2.0 the parabolic error is OK (twice the Migrad error) but MINOS errors yield zero.
Same behavior for all UP!=0.5
Thanks for your help
ROOT Version: 6.14/04 Platform: Linux Mint 19 Compiler: gcc 7.5.0
Apologies for the late reply.
When changing the Error definition (UP) you would need to run MINOS again. You could do this directly using gMinuit or the Fitter class, if fitting by using this interface, or easier for your case, just call TTree::UnbinnedFit again.
Attached is your corrected example working.
Note in case of fitting an histogram you can return a FitResult object and you can get Minos error from there. See FitResult::LowerError/UpperError.