[Ipopt] Any method to speed up the NLP solution via IPOPT

[Jonathan Currie]

Hi,

OPTI Toolbox supplies the MKL PARDISO version, thus I’ve added the option to the ipoptset() interface so it can be used. This is in the OPTI Toolbox development branch on Github, or simply replace your ipoptset.m with the file below:
https://github.com/jonathancurrie/OPTI/blob/Development/Utilities/opti/ipoptset.m

Use via:
ipoptOpts = ipoptset('pardiso_redo_symbolic_fact_only_if_inertia_wrong','yes');
opts = optiset(‘solverOpts’,ipoptOpts);
%etc…

Cheers,
Jonathan

From: Ipopt [mailto:ipopt-bounces at coin-or.org] On Behalf Of Damien
Sent: Tuesday, 10 January 2017 4:11 p.m.
To: ipopt at list.coin-or.org
Subject: Re: [Ipopt] Any method to speed up the NLP solution via IPOPT


Fan,

If you're using MKL Pardiso, one of the things we found that really helps speed on our optimal control problems is setting "pardiso_redo_symbolic_fact_only_if_inertia_wrong" to "yes".  On large problems the symbolic factorisation can take a long time.
Damien
On 1/9/2017 8:06 PM, zhadamfan wrote:
To whom it may concern,

I am trying to solve the nonlinear model predictive control problem. In each loop, a large-scale NLP is required to be sovled.
To obtain a satisfying dynamic performance, a large predictive horizon is chosen. As a result, it takes a considerably long time to get the result.

Is there any method to speed up the compuation? For example, parallel computing or some other technique?

I used the OPTI toolbox in Matlab 2016a on Windows 7, 64bit. I checked the instruction on OPTI wiki. It is said that PARDISO is helpful. However, the effectiveness is limited.


Thank you for any helpful suggestions sincerely. Looking forward to your reply.


Best regards
Fan




_______________________________________________

Ipopt mailing list

Ipopt at list.coin-or.org<mailto: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/20170110/837a55e2/attachment-0001.html>