Multistart apparently does not respect the supplied initial points

Hello,

I have a simple, but annoying, problem about multistart in MATLAB. Consider the following example from MATLAB description of multistart:

fun = @(x) x.^2 + 4*sin(5*x);
opts = optimoptions(@fmincon,'Algorithm','sqp');
problem = createOptimProblem('fmincon','objective',...
    fun,'x0',2.5,'lb',-5,'ub',5,'options',opts);
ms = MultiStart;
[x,f] = run(ms,problem,1)

This gives me the following answer

The local solver ran once and converged with a positive local solver exit flag.

x =

-2.7713

f =

3.8366

But, I expected to roughly get x=2.159 based on the initial point x0=2.5 I supplied. I attach a figure which shows my initial point (open circle) and my expection of local minimum (red star).

Any idea?

Thanks in advance!

Babak

Answer 1

[Alan Weiss]

The only error here is your expectation that fmincon always converges to the closest local minimum. It does not.

fun = @(x) x.^2 + 4*sin(5*x);
opts = optimoptions(@fmincon,'Algorithm','sqp');
problem = createOptimProblem('fmincon','objective',...
    fun,'x0',2.5,'lb',-5,'ub',5,'options',opts);
ms = MultiStart;
[x,f] = run(ms,problem,1)

MultiStart completed the runs from all start points.

The local solver ran once and converged with a positive local solver exit flag.

x = -2.7713

f = 3.8366

% Now let's see where fmincon goes
[x2,f2] = fmincon(problem)

Local minimum found that satisfies the constraints.

Optimization completed because the objective function is non-decreasing in 
feasible directions, to within the value of the optimality tolerance,
and constraints are satisfied to within the value of the constraint tolerance.

x2 = -2.7713

f2 = 3.8366

Understand? MultiStart works as documented. When you call it with 1 run, it takes the supplied start point and proceeds to call fmincon. The error was expecting fmincon to converge to the closest local minimum.

Alan Weiss

MATLAB mathematical toolbox documentation

[Mohammad Shojaei Arani]

Hi Alan,

Thanks for your response.

If you look at my figure you will see that my starting point is very far from the solution being found. But, as you mentioned in your response tro my other question the use of CustomStartPointSet should fix my problem. I hope it is not so complicated to do it (I am a mathematician but unfortunately am not a good programmer :-) ).

Thank!