Equality contrains ignored by fmincon

I use fmincon to minize a function with 250 variables. Some variables are dependent, but the degree of freedom is still very high. I put equality constrains to fix some varibles to constants. When I run fmincon, I received the following error from Matlab

"Solver stopped prematurely.
fmincon stopped because it exceeded the function evaluation limit,
options.MaxFunctionEvaluations = 3.000000e+03."

I checked the premature x variable. It did not meet the equality constrains at all. The elements that were set to zeros all are non-zero number. I am wondering what would be possible causes? The equality constrains seem so straight forward, but still ignored by fmincon. Could someone advise me please?

[darova]

Can you equations?

[John Hoop]

What do you mean by equations?

[darova]

I mean can you show? THe equations?

[BALAJI KARTHEEK]

I hope u can use this code, i too have the same problem while solving the problem with 130 variables and nearly 200 constraints, but when i used this it worked fine with me... if u want further details, u cann contact me through watsapp [phone] or try it by urself. It will work for sure

opts = optimoptions('fmincon','Display','iter','MaxFunctionEvaluations',10e+10);
problem =         createOptimProblem('fmincon','objective',obj2,'x0',your_initial_point,'lb',LB,'ub',UB,'nonlcon',constraints1,'options',opts);
ms = GlobalSearch;
[x,f] = run(ms,problem);

[Walter Roberson]

if you have elements constrained to be a constant then you should rewrite the system to remove them as variables and use the constants. This is easier with Problem Based approach.

[John Hoop]

Hi Darova, the function I tried to minimize is a matlab m-function with many calculations inside, but with a scalar output. Below an example of the equality constrains where the first 100 elements of x are set to zero. So I would expect that fmincon will set these to be constants first, and then only optimize the remaining variables x101,...,x250. It seems that fmincon does not work like this way!

%x=(x1,x2,....,x250)

Aeq=zeros(250,250);beq=zeros(250,1);

% to set x1,...,x100 to zero

for i=1:100

Aeq(i,i)=1;

end

[John Hoop]

Hi Walter, Thank you very much for your suggestion. I thought about this. But, I would like to keep these variables in the system, because I eventually want to use all as optimization variables. For now, I fix these only for testing purposes.

[darova]

This looks strange. Just 100 zero variables? What is the point?

for i=1:100
    Aeq(i,i)=1;
end

Answer 1

[BALAJI KARTHEEK]

I hope u can use this code, i too have the same problem while solving the problem with 130 variables and nearly 200 constraints, but when i used this it worked fine with me... if u want further details, u cann contact me through watsapp [phone] or try it by urself. It will work for sure

opts = optimoptions('fmincon','Display','iter','MaxFunctionEvaluations',10e+10);

problem = createOptimProblem('fmincon','objective',obj2,'x0',your_initial_point,'lb',LB,'ub',UB,'nonlcon',constraints1,'options',opts);

ms = GlobalSearch;

[x,f] = run(ms,problem);

[John Hoop]

Hi Balaji, thank you very much for suggestions. I am going to have some tries and may contact you via whatapp. Cheers, Jun

Answer 2

[Matt J]

You cannot use the Aeq, beq constraint parameters to force exact equality. You could instead try setting the upper and lower bounds,

lb(1:100)=0;
ub(1:100)=0;

but if you are reaching the function evaluation limit, it may be that a feasible solution with the parameters you've set to zero simply does not exist.

[John Hoop]

Hi Matt, Thank you very much. This is very good to know. Equality constrains are a bit misleading.

[Matt J]

You should be mindful that all of the constraints (except usually the bounds) are subject to the ConstraintTolerance parameter listed here,

https://www.mathworks.com/help/optim/ug/optimization-options-reference.html