[Ipopt] Many line searches are required near the solution when L-BFGS is used.

[Youngsoo Choi]

Dear all,

I am using IPOPT in order to solve a simple shape optimization problem in
which I know what the optimal shape should be. Because it is hard to get an
exact Hessian analytically, I am using L-BFGS instead. I notice that IPOPT
converges to the optimal solution, but near the solution it tends to do a
lot of line search which slows down tremendously. I cound increase the
convergence tolerance a bit in order to terminate IPOPT in a reasonal time
and with a good enough solution.

When BFGS-type algorithm is used, I expect to see that a unit step to be
normally satisfied by whatever the linesearch criteria is used. However, I
am not seeing that from IPOPT.

Any feedback on this issue is welcome.

Thanks.

-- 
Jesus Loves you,
Young (Youngsoo).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130314/59c2b5ed/attachment.html>