Using parfor with fminsearch

Hi everybody. I would like to generate a distribution of parameter vectors using different subsamples. The estimates are produced using my self-defined function criterion. I have B samples, so Bsamples{b} picks out the b-th saple.

 parfor b=1:B
    Bresults(b,:)=fminsearch(@criterion,x,options,nsimu,Bsamples{b},0,ss);
         % find estimator, x=starting parameter vector    
 end

While this works fine with a "for" loop, when using "parfor" it always returns the starting parameter vector as result. It seems not to recognize the different inputs Bsamples{b}. Is it in general not possible to parfor with optimization routines? Other operations, like mean(Bsample{b}) in the loop work fine.

Many thanks! Philipp

[Sean de Wolski]

Can you provide some values? I see no reason why what you have above would not work with a parfor() loop.

Also, I would recommend using an anonymous function do pass the extra parameters:

http://www.mathworks.com/help/releases/R2012b/optim/ug/passing-extra-parameters.html