Minuit2 - Simplex - Errors displayed?

Hi,

I would like to ask about the errors displayed in the outputs of Minuit2. I understand the situation when gradients are involved and the inverse Hessian is available. However when running the Nelder-Mead method in Minuit2 there are still errors displayed in the output. In the documentation that the simplex method says does not give reliable information on parameter error, correlation etc… (mnusersguide.pdf: page 37) .

How are those numbers determined and calculated in the output?
I have run some Monte Carlo error estimation and loads of times those numbers are actually very close to the "synthetic data-set " errors.

Thanks,
Zolo

Hi,

The displayed errors are actually the steps sizes used in Simplex. It is likely they are of the same order of magnitude as the errors, however are not precise. I would reccomend you to run Hesse or Migrad in any case after Simplex to estimate those errors or if you cannot do it, perform a scan of the likelihood function

Best Regards

Lorenzo

Thanks,

This is very useful to know and I will try to run the Hesse or Migrad but the migrad gave me a seg-fault for long time now (maybe weird data or slightly buggy FCN function which actually returns a vector ) but I will give it another try. The scanning of the likelihood function I have not tried. That sounds excitation and a must do.

Thanks again,
Zolo