How to manually change MaxFunctionEvaluations in fmincon

I would like to manually change the number of evaluations from 3000 to something bigger. Is it possible that with more evaluations I could increase the chance of finding global minimum of given function fun? How can I change this in script?

Thank you for your help:)

Answer 1

[Alan Weiss]

I think that the short answer to your question is to set the MaxFunctionEvaluations option to a higher-than-default value. See Set and Change Options. You might also want to increase the MaxIterations option.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

Answer 2

[John D'Errico]

No. Increasing the maximum number of function evaluations is NOT how you will ensure finding the global minimum. In fact, if the optimization has terminated happily in the first instance, then changing that limit will not do a thing.

Fmincon is not a global optimizer anyway. It will find a solution that is as best as it can, but different start points can yield different solutions. An increase in the function evals allowed will not change a thing.

You may want to try a global optimizer. There is a toolbox for that. Or you may want to try a stochastic optimizer such as a genetic algorithm. Again, a toolbox for that. Or simulated annealing, or particle swarm tools, etc. Or just better starting values.

[sycXZOR]

I am not so sure in your answer. For instance if I put in random x0 vector as starting point for fmincon, I get, afet 3000 evaluations the vector x that I was searching and fval, which is current value of function:

[x,fval] = fmincon(fun,x0,...);

However, if I use the calculated x vector as new starting point:

x0 = x;

and repeat the algorithm for new starting point I get improved new fval and therefore improved new vector x and improved local minimum of the function. If I loop this until the absolute value of difference between new and old fval is let's say bigger than 0.0001 I get into ACTUAL local minimum. Not global minimum probably, but waaay improved local minimum. So yes, the number of evaluations should play some kind of a role or is my theory wrong?

Answer 3

[Walter Roberson]

Changing the maximum function evaluations can help find the local minima, in some cases where the function is complicated enough mathematically that a lot of evaluations are needed to predict the right directions to search, or if you want the position of the minima to high tolerance.

There are functions where relative movements on the order of 1E-10 or less make a big difference in the minima, when the local minima sits in a very narrow slot within a larger valley, so sometimes increasing the function evaluations can greatly affect the outcome.

Every once in a while the local minima turns out to be the global minima, but if you are searching in the wrong local minima, no matter how many function evaluations you permit, you are not going to find the global minima with fmincon. fmincon is mostly unable to escape from sufficiently deep local minima.

[Walter Roberson]

options = struct('MaxFunctionEvaluations', 10000);
[x, fval] = fmincon(objective_function, A, b, Aeq, beq, lb, ub, nonlcon, options);

If you have no value for any of the above such as Aeq, beq, then put [] in their place.

[Fares Bettahar]

hello sir ,can you help me please

I have this problem in Matlab

Solver stopped prematurely.

fsolve stopped because it exceeded the function evaluation limit,

options.MaxFunctionEvaluations = 5.000000e+02.

X =

18.7410 6.6035 138.6942 16.4586 9.6161

FVAL =

16.1142 6.6468 5.1229 22.6175 -21.6852

also When used

f

un = @tp1000;
X0 = [11 5 135 13 10];
% MaxFunEvals=1000;0
%   options = optimoptions('levenberg-marquardt');
% oldoptions = optimoptions(@lsqnonlin,'Algorithm','levenberg-marquardt',...
%     'MaxFunctionEvaluations',1000)
options = struct('MaxFunctionEvaluations',1000)
 [X,FVAL]= fsolve(fun,X0)

i dont know how do "MaxFunctionEvaluations" =1000

[Alan Weiss]

I suggest that you ask your question in a new question, not attached to one already answered.

But if you want to see how to create and use options, see Set Options.

Alan Weiss

MATLAB mathematical toolbox documentation