Fmincon tries the same values for manipulated variables for three iterations

Hello, I have a simulink model that I control and minimize one of its outputs with fmincon. My problem is that fmincon tries two values for the two manipulated variables for three iterations before changing to another guess. The optimization in other words is three times slower. I've tried to change the Algorithm with no result. How can I fix this?

Here is my code:

options=optimset('Algorithm','sqp','TolX',0.01,'TolFun',0.001);

x=fmincon(@obj,xo,A,B,[],[],xmin,xmax,[],options);

function q = obj(z)

        cr=z(1);
        tr=z(2);

        sopt = simset('solver','ode3','SrcWorkspace','base','DstWorkspace','base','SaveFormat','Array');
        [~,~,yo]= sim('sgs2.slx',[0 1], sopt);

        q=abs(yo(end,1)-yo(end,2));

    end

Answer 1

[John D'Errico]

Are they really IDENTICAL? Somehow I doubt it. Look more carefully. Perhaps you need to write out the parameters with more precision.

Fmincon must compute a gradient. So this means it must make a tiny adjustment to each parameter, to then compute an approximate derivative. Once it gets an estimate of the local gradient, it can then take an actual step. At some point, it will then decide it need to know the gradient again. So it does so.

You cannot magically make it run more quickly. Knowledge of the gradient allows it to make an INTELLIGENT choice for the next place to look. For example, you might use fminsearch instead, a solver that does not need to compute a gradient. But that lack of information costs you greatly, so it will converge more slowly and less reliably.

Answer 2

[Stefanos Kalandaridis]

They are identical. And this is a problem, not only for the speed. I cannot reduce the "TolFun" option because if I set it to 1 (or something bigger than 0.0001), an error occurs that says "the difference between the values of the iteration is smaller than TolFun's value so the optimization will end".