Optimization problem not finding the actual optimal value

Hello!

I am trying to solve an optimization problem in the Problem-Based mode and I am having problems finding the local optimum. The function I am looking to optimize is located in a .m file developed by me, in which a single output Y value is calculated, which depends on the simulation and run of a computational model of my own. I am defining the objective of my problem with fcn2optimexpr and the optimization method I am using is fmincon. No matter what initial value X (a single variable) I assign to the problem, the program always tells me that it has already found the local optimum at the initial point assigned to X, without further iterations. I have already tried to modify the program tolerances (step size, objetive and contraint) and it keeps throwing the same result. Also, I tried changing the searching algoritm and nothing has worked out. Any idea what I can be doing wrong? Thank you very much. Below is the code used to define the problem and the text obtained in response.

prob = optimproblem("Description", "Optimization SUDS", "ObjectiveSense", "maximize");

x = optimvar('x');
x.LowerBound = 0;
x.UpperBound = 1;

obj = fcn2optimexpr(@opti, x);
prob.Objective = obj;

initialGuess.x = 0.1;


options = optimoptions("fmincon","Display","iter", "PlotFcn",["optimplotx",...
    "optimplotfval","optimplotstepsize"]);


[sol, fval, exitflag, output] = solve(prob, initialGuess, "Options",options);

Initial Point is a local minimum that satisfies the constraints.

Optimizacion completed because at the initial point, the objetive function is non-decreasing in feasible directions to within the value of the optimality tolerance, and contraints are satisfied to within the value of the contraint tolerance

[Torsten]

What do you get back if you call your objective with the initial guess for x ?

[Pascual Ferrans Ramírez]

Hi Torsten, thank you for your message. i get the actual value that Im looking for, 7.5

Answer 1

[Matt J]

Making sure you don't have any round(), ceil(), floor() or other discretization operations in your simulation code. Also, make sure you have read the guidelines on optimizing a simulation here,

https://www.mathworks.com/help/optim/ug/optimizing-a-simulation-or-ordinary-differential-equation.html

Finally, since your problem only has a single unknown, I would use fminbnd().

[Walter Roberson]

Exactly. round or floor or ceil will cause this behavior. fmincon is not suitable for any quantized formula.

[Pascual Ferrans Ramírez]

Thank you both for your messages. That is exactly what was happening. fminbnd() is working fine. Thanks.

[Matt J]

Glad it's working, but then please Accept-click the answer to indicate so.