[Ipopt] Best practice (in terms of stopping criteria) when using "limited-memory"

[Mehmet Ersin YUMER]

Hi,

I am using Ipopt to solve a very large problem. Although I am able to
supply the exact hessian, I cannot solve for the exact problem because of
memory constraints. (Quarter the size of my problem requires ~12GB memory
when solving with ipopt-mumps).

Therefore, I am trying to use the limited-memory option. For some test
cases of my problem, I observed that although ipopt issues a fail in the
limited memory case, the solution is very close to the exact solution.

Based on my search in Ipopt mailing list archive, I have seen that
"limited-memory" convergence criteria is somehow not that reliable. I will
appreciate if you can share, advise on best practices for convergence
criteria parameters when using ipopt in limited-memory mode.

Thanks.

-- 
M. Ersin Yumer
http://www.meyumer.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130222/18d81929/attachment.html>