o a maximum rather than a minimum of maxiter starting values are
generated - todo
o if start is a data frame with n+1 rows where n is the number of
parameters to optimize then Nelder Mead is used first - todo
o add minpack.lm interface - todo