[Ipopt-tickets] [Ipopt] #201: Inexact Algorithm

[Ipopt]

#201: Inexact Algorithm
----------------------------+------------------------
  Reporter:  ccoptimizer    |      Owner:  ipopt-team
      Type:  clarification  |     Status:  closed
  Priority:  normal         |  Component:  Ipopt
   Version:  3.11           |   Severity:  normal
Resolution:  fixed          |   Keywords:  Inexact
----------------------------+------------------------

Comment (by ccoptimizer):

 Thanks for shedding some light on it, kelman.

 Could I ask the optimization problem type (e.g. linear, quadratic, conic
 or nonconvex) you were solving with inexact algorithm and Pardiso
 iterative solver to find inefficiency? And I guess you tested it with very
 large-scale problem?

 I agree that it only helps for some types of problem – that's why I try to
 understand the source codes and settings there but somehow get lost due to
 the lack of documentation.


 Replying to [comment:1 kelman]:
 > It's not well-documented, as I believe it's still considered
 experimental. You can enable it by configuring Ipopt with --enable-
 inexact-solver and setting the option inexact_algorithm to yes. It also
 requires you to build Ipopt with Pardiso, though I believe some of the
 other linear solvers might also be supported with the same algorithm.
 >
 > Don't be surprised if it doesn't work as well as you hope it might, I
 was disappointed with its performance on the optimization problems I was
 solving. It seems it only helps for some types of problem structures.

-- 
Ticket URL: <https://projects.coin-or.org/ticket/201#comment:3>
Ipopt <http://projects.coin-or.org/Ipopt>
Interior-point optimizer for nonlinear programs.