Re: [Help-gsl] nmsimplex with constraints

[Peter Melchior]

On 11/1/06, Ines Hellmann <address@hidden> wrote:

> I am using the gsl_multimin_fminimizer with the Nelder Mead ("ambeba")
> algorithm and want to
> put some constraints on the parameters that are optimized. Like I
don't
> want them to become negative.
> Is there a way to do this and if so, what is it?

A regularization does the job.
You can modify the function you are minimizing to penalize negative
parameters. If your function is f(x,params), replace it by

f'(x,params) = f(x,params) + penalty(params)

where penalty is only unequal to 0 when the parameters are negative.

Given a sufficiently high penalty, this prevents the simplex to move in
this area of parameter space.


Best regards,

Peter Melchior