[Ipopt] "Slow convergence on quadratic problem"

[tony at kelman.net]

Which linear solver were you using? I'd usually experiment with problem scaling both at the modeling level and at the linear solver level. Were you seeing any warnings? Can you share some representative logs at the default print level, to see how the primal vs dual infeasibilities, barrier parameter, and any regularization were behaving? I also assume you've enabled the constant jacobian and hessian options for a qp, which should at least make each iteration cheaper though should not change convergence behavior.




On Tue, Oct 13, 2015 at 11:09 PM -0700, <jmogali at andrew.cmu.edu> wrote:
Hi all,
        For my problem (~10000 constraints, ~12000 variables), IPOPT fails
to converge on Convex Quadratic  problem (positive semi definite
though) within 3000 iterations. I am aware of some earlier posts
on this issue which stated enabling Mehrotra's algorithm. Enabling
that still did not give me satisfactory timings. I was wondering
if there were any recent additions to the solver, methods such as
block coordinate descent or BSUM or any other fancy method which
although may not fall into an interior point framework. Anyways, I
would still like to hear any suggestions that has helped people in
particular with convex quadratic problems using IPOPT or any other
solver which is very effective on these sort of problems.

Thanks,
Jayanth

_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20151014/30e11571/attachment-0001.html>