Argh!!!!!!!!!!
This function is only well-behaved and smooth with in a narrow region around the optimal parameter value
Outside this area all conventional optimisers will fail!!
- to the left there are underflow problems
- to the right there are overflow then NaN problems
the IEEE arithmetic creates false local optima before becoming a NaN black hole that will totally wreck the nonlinear optimiser although it will not tell you that this has happened!!