Comment by improbable22

Is there a good reason not to regard this as a standard few-parameter no-gradient optimisation problem, and use something like Nelder-Mead on it?