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