Thread Subject: Problem with fmincon

Hi,

I have a problem with a program using fmincon: after several iterations the real-valued x vector (seeking the minimum of the obj function) becomes complex which makes the program crash at the next iteration.
How is that possible and how could I correct this???

NB:
My obj function does work with complex numbers but the final result R must mathematically be real. I've checked that this might not be exactly the case, probably because of numerical aberrations: I typically get |Im R / Re R|~10^-15. Hence at the end of my function I force R=real(R) and check that R is then real each time my obj function is called (including the call just before x becomes complex).

Any advice?

Thanks!

Problem solved:
I forgot to clean as well the numerical aberrations in my nonlinear constraints.

i have the same problem. i wanna find the optimum of complex vector but it seems that fmincon work only with real parameters?
How can I manage this problem ?

"anis riahi" <riahianis1@gmail.com> wrote in message
news:hg7jbb$k7a$1@fred.mathworks.com...
>i have the same problem. i wanna find the optimum of complex vector but it
>seems that fmincon work only with real parameters?
> How can I manage this problem ?

Convert your problem from one in N complex parameters into one in 2*N real
parameters; inside your objective function, convert the 2*N-element real
vector into an N-element complex vector and use the complex vector to
evaluate the objective.

--
Steve Lord
slord@mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ

"Steven Lord" <slord@mathworks.com> wrote in message <hg86nc$ka2$1@fred.mathworks.com>...
>
> "anis riahi" <riahianis1@gmail.com> wrote in message
> news:hg7jbb$k7a$1@fred.mathworks.com...
> >i have the same problem. i wanna find the optimum of complex vector but it
> >seems that fmincon work only with real parameters?
> > How can I manage this problem ?
>
> Convert your problem from one in N complex parameters into one in 2*N real
> parameters; inside your objective function, convert the 2*N-element real
> vector into an N-element complex vector and use the complex vector to
> evaluate the objective.

Just remember as well to make sure that the objective returns a purely real-valued result, i.e., if you are calculating the value of the objective (or its gradient etc...) using complex arithmetic operations, be sure to strip away any residual complex part of the result arising from round-off errors and so forth.