MATLAB: Setting optimality tolerance for fminunc

fminuncoptimality tolerance

I want to set the optimality tolerance for the fminunc function using optimoptions. Based on the documentation (<http://uk.mathworks.com/help/optim/ug/optimoptions.html)>, I have tried this:

     options = optimoptions(@fminunc, 'OptimalityTolerance', 1e-10);

However, this gives me the following error:

     Error using optimoptions (line 105)
     'OptimalityTolerance' is not an option for FMINUNC.

Furthermore, when I check the default option values for fminunc using

     options = optimoptions('fminunc')

then optimality tolerance is not listed at all.

Does anyone have an idea why this might be? I should note that I am running Matlab 2015b – perhaps this has to do with the version I am using?

Best Answer

Yes, indeed. R2016a is where OptimalityTolerance was introduced. More information is in the Current and Legacy Option Name Tables.

Alan Weiss

MATLAB mathematical toolbox documentation

Related Solutions

MATLAB: ‘fminunc’ function

See the documentation on fminunc (link). It always has been part of the Optimization Toolbox.

Depending on the problem you are solving, you might be able to use the fminsearch (link) function. You may have to use it with the norm (link) function to get the same results

For example:

x = rand(5,1);
YourFunction = @(p) p(1).^2 + p(2).*x - 10;
P = fminsearch(@(p) norm(YourFunction(p)), [1; 1])

MATLAB: Problem using FSOLVE ‘OptimalityTolerance’ is not an option

I am surprised that you got that error, but here is what the fsolve documentation has to say about that option:

Internally, the 'levenberg-marquardt' algorithm uses an optimality tolerance (stopping criterion) of 1e-4 times FunctionTolerance and does not use OptimalityTolerance.

I would also like to point out that many of your chosen options and tolerances do not make sense. To quote from that link: Generally set tolerances such as OptimalityTolerance and StepTolerance to be well above eps, and usually above 1e-14. Setting small tolerances does not always result in accurate results. Instead, a solver can fail to recognize when it has converged, and can continue futile iterations. A tolerance value smaller than eps effectively disables that stopping condition.