[Ipopt] Infeasible Problem Detected when the problem is actually feasible

[Filip Jorissen]

Dear Maxime,

Changing the linear solver should not make any difference as far as I know. I suggest to read up a it on the theory of interior point solvers and derivative-based optimization. Then you’ll be able to understand better why it does not work and what you can/should do to fix it :)

Kind regards!

Filip

Op 6 apr. 2018, om 16:25 heeft Maxime Boulay <mboulay at flogen.com<mailto:mboulay at flogen.com>> het volgende geschreven:

Hi Filip,

Thank you for your help. Is it possible that using a different linear solver may help to avoid those local points of infeasibility on top of having  a good initial guess?
I am presently using mumps.

Maxime

From: Filip Jorissen [mailto:filip.jorissen at kuleuven.be]
Sent: Friday, April 06, 2018 3:13 AM
To: Maxime Boulay
Cc: Ipopt at list.coin-or.org<mailto:Ipopt at list.coin-or.org>
Subject: Re: [Ipopt] Infeasible Problem Detected when the problem is actually feasible

Hi,

It sounds like IPOPT converges to a local point of infeasibility? I.e. your equations are non-convex and you need a good/feasible initial guess to be able to converge to a feasible (local) optimum.

Filip


Op 5 apr. 2018, om 17:27 heeft Maxime Boulay <mboulay at flogen.com<mailto:mboulay at flogen.com>> het volgende geschreven:

Hello,

I am using IpOpt to solve a problem with about 20 variables and 20 constraints. I know the solution to this problem, and I have verified that my constraints are respected at the solution, there is also no error in the Derivative Checker. However, IpOpt still ends up giving me the Infeasible Problem Detected message when I run it. If I keep running it again with the "warm_start_init_point", it eventually does find the solution. This happens for many different entry values to this problem. Can anyone help me find a clue as to why this would happen?

Thank you.

Maxime

_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org<mailto:Ipopt at list.coin-or.org>
https://list.coin-or.org/mailman/listinfo/ipopt

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20180407/39024253/attachment.html>