I've been reading up on Downhill Simplex (Nelder-Mead) optimization, but what I was missing were good proposals on what to do when the parameters / coordinates are bound to a fixed interval. What is the best way to handle the case that one parameter goes to the limit of the intervals, especially avoiding that it gets "stuck" there? Let's say I optimize a function of 10-20 parameters which are each limited to a finite interval, say, [0 ; 100]. What is the right course of action if the algorithm would push one or several of the parameters over the limits (<0 or >100)?
Thanks,
Martin
Reading through multiple descriptions and publications on numerical optimization using downhill simplex
