Stopping fminsearch before convergence

I am estimating 12 parameters in fminsearch.

For most starting values, after about 200 iterations and 200 function evaluations the fval becomes -Inf.

Is it ok to just keep the 200 iteration and 200 function evaluations and not let fminsearch converge--- which gives me an fval = 600-- or is it too "low"?

Answer 1

[John D'Errico]

I think this is entirely up to you. If that point is adequate for your purposes, then yes. Only you know if something is ok.

The reason it went to -inf is probably that fminsearch went into an area where the function was unbounded. But if you are minimizing something, then whats the problem? It went as small as possible! -inf is surely smaller than any other number.

Really, this points out the problem you have here. Your objective is unbounded. Minimizing it will give you -inf. This means you have a problem with the objective function. Asking if you can accept any intermediate value is silly IMHO. The function has no minimum. It is unbounded.