fminsearch vs. fminunc

[Rob Lynch]

I cannot find a comparison of these two minimization functions. In
using them I find that they do not perform the same (specifically,
fminsearch finds a good solution where fminunc gives up). What I
really want is a guideline on the conditions under which each program
could be expected to produce good results.

[Claudio]

Rob,

doc fminsearch
doc fminunc

Read the algorithm part. There are some comments on when fminsearch
performs better than fminunc.

Claudio

[John D'Errico]

I do have a short piece that describes fminsearch, and contrasts
it to a more traditional line search based minimizer. I think it
was in section 21 where I discuss fminsearch, although this was
one of the parts I wanted to expand.

http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=8553&objectType=FILE

However, I think its virtually impossible to write down hard rules
that define which problems one optimizer will work better on than
the other. Fminsearch tends to be a bit slow, especially for larger
problems. I'd never use it for more than 5 or 6 variables. It may
be more robust to some things that will cause problems for fminunc.
Derivative discontinuities are a good example. Fminsearch MAY
survive their presence, although there is no assurance.

The best guide would come from learning about these methods. Once
you understand the way they work, you will also understand when
to use one or the other on a specific problem. You will also
understand what it is about your problem that may cause one to
fail.

HTH,
John D'Errico

--
The best material model of a cat is another, or preferably the same, cat.
A. Rosenblueth, Philosophy of Science, 1945

Those who can't laugh at themselves leave the job to others.
Anonymous

[Myrtle42]

If anyone else stumbles over this question years later, as I did -- I found this page useful as a broad overview of the many different optimization functions in matlab:
https://www.mathworks.com/help/optim/ug/optimization-decision-table.html

(not equivalent to understanding the algorthims, but a place to start)