GammaLib

Note that the implementation should be done in the GOptimizerPar class, not the GModelPar class.

We have to decide how to name the asymmetric errors in the XML file. We also need to decide how to name the GOptimizerPar methods to access the asymmetric errors.

I see basically two choices.

1) add an asymmetry parameter given for example by and define the mean error as . Then and . This reutilizes the existing error attribute of the class. For symmetric errors we simply would have , which would be the default in case that no alpha is provided in the XML file (we may think about a better name for the parameter, e.g. asymmtry or error_asymmetry or something like that).

2) introduce two new parameters, something like error_pos and error_neg. This would however make the existing error attribute obsolete in case that an asymmetric error exists.

The only advantage that I see in 2) over 1) is that the error are more “readable” (you don’t need to apply the above formulae). However, solution 1) would be a more natural expansion of symmetric errors.