failure in initial objective function evaluation

Hello,

I want to solve an optimization problem by Matlab fmincon function. I defined my objective function as:

    function f = objfun(x,p1,p2)
    x1=x(1:24);
    y1=x(25:48);
    z1=x(48:end);

    f=-mean(x1*(p1)'+y1*(p2)')+mean(z1*(p1)')
    end

p1,p2 are 1x24 vectors which are calculated in the main file. I need to find the optimum values for x1, y1 and z1 which are 1x24 vectors (total 72 variables).

When I run the main code I get an error message about the objective function and then for fmincon by failure in initial objective unction evaluation. caused

Would you please advise how can I solve the problem? Thanks a lot!

[Alan Weiss]

Please show us your fmincon call, the complete copy-pasted error message, and the value of x0. The problem could be as simple as you having x0 as a row vector instead of a column vector, or p1 a row vector, but without seeing the inputs to the problem and the complete error message, we cannot know.

Alan Weiss

MATLAB mathematical toolbox documentation

[Sara]

Thanks for your reply. I use below in my main code: fun=@(x)objfun(x,p1,p2) x0 is a 1x72 vector. p1 and p2 are 1x24 vectors. The error is as below: Error using * Inner matrix dimensions must agree.

Error in objfunFCR (line 6) f = -mean(x1*(p_FCRN/100)'+y1*(p_FCRD/100)')+mean(z1*(p_FCRN/100)');

Error in FCR_N_D>@(x)objfunFCR(x,p_FCRN,p_FCRD)

Error in fmincon (line 536) initVals.f = feval(funfcn{3},X,varargin{:});

Error in FCR_N_D (line 398) [x,fval] = fmincon(fun,x0,A,b,Aeq,beq,lb,ub,@chg_dischg,options)

Caused by: Failure in initial objective function evaluation. FMINCON cannot continue.

p.s. I made a bit changes to the names of inputs to simplify my question here.

I appreciate your comments!

Answer 1

[Alan Weiss]

Thank you for the complete error message. z1 should be defined as

z1 = x(49:end);

I believe that, as you have defined it, z1 has 25 elements.

I am not really sure that you gave us the code of your function. The mean statements do not do anything, as all of the entries x1*p1' and the others are scalars, and mean leaves scalars alone.

Alan Weiss

MATLAB mathematical toolbox documentation

[Sara]

Thank you very much for your reply. It was exactly my mistake in z definition. I used 'mean' because I need to calculate expectation values of x1*p1, x2*p2, .... . Is there any other way in Matlab to define expectation value?

[Alan Weiss]

The mean statement is superfluous. When you multiply, the matrix definition of multiply gives

x1*(p1') = x1(1)*p1(1) + x1(2)*p1(2) + ... + x1(24)*p1(24)

This is a scalar. To get the mean, you could divide by 24. But to optimize, the factor of 24 does not matter.

Alan Weiss

MATLAB mathematical toolbox documentation