# What is the right way to treate "Optimization terminated: no feasible point found." ???

31 views (last 30 days)
Tarek Omar on 12 Nov 2017
Commented: Walter Roberson on 29 Sep 2018
What is the right way to treate "Optimization terminated: no feasible point found." ???
##### 2 CommentsShowHide 1 older comment
Walter Roberson on 12 Nov 2017
Check that your constraints can be met; if the optimizer permits an initial value then try to give it one that is within constraints.

John D'Errico on 13 Nov 2017
A feasible solution is some point, ANY point, that satisfies all of the constraints. There is no assurance that a feasible point exists at all for some unknown problem. However, the odds are decent that no such solution exists if MATLAB returns that statement. But we can never know that, given no information at all.
So it depends. Are your constraints highly nonlinear? Just how nasty are they? Have you even implemented them in MATLAB properly? We cannot know this, given no code to see at all, and no information as to the problem you want to solve.
So what should you do?
1. Verify the implementation. CAREFULLY. Do this twice. Then do it a third time. Get a colleague to check your implementation.
2. Try some random points. Do the constraints make sense in what they return? APPLY COMMON SENSE. Do any of them satisfy all of the constraints? MATLAB starts looking for a feasible point based on your initial values. So a different start point might allow the solver to find a feasible point.
If it appears that no feasible point seems to be found, then consider if there is a good reason why no feasible point was found.
Of course, if you want better help, you would have more of a chance if you explain what you are trying to solve, and showed what you did. Otherwise, it is difficult to utter more than the vague platitudes that I did here.
Walter Roberson on 13 Nov 2017
Note that especially in the case of highly non-linear functions, sometimes MATLAB is able to detect that a zero crossing exists but is unable to determine where the crossing is, because the values for the last representable number on each side of the crossing might be quite distant from 0. For example even something as simple as 1E100*x cannot have its zero crossing determined with enough precision for some of the routines to satisfy their tolerances.

Alan Weiss on 13 Nov 2017
Good luck,
Alan Weiss
MATLAB mathematical toolbox documentation

tahseen alshmary on 29 Sep 2018
Edited: Walter Roberson on 29 Sep 2018
i have this problem appear with genetic algorithm ,what is the solution
Walter Roberson on 29 Sep 2018
Mathworks provides ga() and gamultiobj() in the Global Optimization Toolbox.
There are over 50 genetic algorithm implementations in the File Exchange: https://www.mathworks.com/matlabcentral/fileexchange/?utf8=%E2%9C%93&term=tag%3Aga