I received an unexpected wrong result that contradict the constraint condition when using optimproblem

function sol = optimize_crop_allocation(A_init, P, Q, C, D, U, V, M, N, S, B_1, B_2)

    % P,Q 1x41
    % C,D 1x41
    % U,V 54x41
    % M,N 54x41
    % S 1x54
    % B_1,B_2 54x41


   
    prob = optimproblem('ObjectiveSense','maximize');
    
    A = optimvar('A', 2, 54, 41, 'Type', 'continuous', 'LowerBound', double(0));
    
    X=A(1);
    Y=A(2);

    phi_f=@(X) double(0.5*min(sum(X .* M)+sum(X .* M),sum(X .* M)+C)*P'-sum(sum(X.*U)));  
    f_expr=fcn2optimexpr(phi_f,X);
    phi_g=@(Y) double(0.5*min(sum(Y .* N)+sum(Y .* N),D+sum(Y .* N))*Q'-sum(sum(Y.*V)));  
    g_expr=fcn2optimexpr(phi_g,Y);

    prob.Objective = f_expr+g_expr;

    prob.Constraints.cons1 = sum(A(1), 2) <= S'; % the output fails to satisfy this condition
    prob.Constraints.cons2 = sum(A(2), 2) <= S'; % and this
    
    A0.A=double(A_init);

    sol = solve(prob,A0);

    disp(sol);
end

thanks for your help

[John D'Errico]

You don't provide the data that went into it, so there is nothing we can test.

You don't show how far the constraints failed by. Is it less than the constraint tolerance? If so, then it did not actually fail. That sort of thing is to be expected.

We cannot see inside your computer, or see into your mind. If you want help, then you need to make it possible for us to help you.

[昀泽]

Now this txt file below contains all the data required.

[idris]

Your error's description is probably as a result of how the summation and dimensionality of your variables are set in relation to the constraints. Can you provide more facts?

[昀泽]

this is the output

[John D'Errico]

Still missing information to recreate what you did.

What was A_init?

As my first comment without even trying yet to run your code, I see that your objective function has calls to min in it. That makes your objective not a differentiable one. And since the solver will probably decide to use fmincon, the call to min can often make fmincon fail to properly converge.

[昀泽]

sorry for that, i've missed this constant... A_init was the initial state of A.

[昀泽]

That seems reasonable! But is there any alternative methods that I could give a try)

[Torsten]

As @idris said: you should experiment first with numerical matrices what you get by your summation and indexing operations. E.g. I can't believe that you want to set these constraints:

prob.Constraints.cons1 = sum(A(1), 2) <= S'; % the output fails to satisfy this condition
prob.Constraints.cons2 = sum(A(2), 2) <= S'; % and this

A(1) and A(2) are scalar values like 4 and 5. So why would it be necessary to sum over them ?

Also setting

X=A(1);
Y=A(2);

is obscure. Maybe you mean

X=A(1,:,:);
Y=A(2,:,:);

I don't know.

[昀泽]

Now it’s fine) the optimizer works not so well, I think, due to the non-differentiable object function min) I’ve tried other methods)