[Ipopt] Using IPOPT as a Linear (Sub-)Solver for LP and QP Problems.

[Jeffrey Dean Kelly]

Hi Jonathan;

Yes, I also use active-set solvers in my SLP such as LPSOLVE, GLPK, COINMP
(CBC/CLP), GUROBI, XPRESS and CPLEX where it is known that solving with an
IPM first for 1 to 5 major iterations is useful then to switch-over to an
active-set solver as well as switching between primal and dual methods as
convergence gets close is also beneficial.

Hence, I am trying to use IPOPT as a preliminary IPM solver for the first
few major iterations to see if that is effective but I am finding that for
pure LP's and QP's IPOPT's performance is less than expected unless I can
find some better settings/options.

Thanks, Jeff


On Sun, Sep 1, 2013 at 3:25 PM, <jonathan.hogg at stfc.ac.uk> wrote:

> If you're using SLP/SQP, the solution for one iteration is probably
> related to the iteration for the next. If you use a simplex/active set
> based solver such as CLP instead of an interior point one, you'll be able
> to warmstart which is harder to do with an IPM algorithm like Ipopt (if you
> want to get complicated you can look at crossover on the first iterate).
>
> Jonathan.
> ________________________________________
> From: ipopt-bounces at list.coin-or.org [ipopt-bounces at list.coin-or.org] on
> behalf of Stefan Vigerske [stefan at math.hu-berlin.de]
> Sent: 01 September 2013 14:56
> To: Jeffrey Dean Kelly
> Cc: ipopt at list.coin-or.org
> Subject: Re: [Ipopt] Using IPOPT as a Linear (Sub-)Solver for LP and QP
> Problems.
>
> Hi,
>
> I heard that enabling the mehrotra_algorithm can be useful when solving
> LPs or QPs:
> http://www.coin-or.org/Ipopt/documentation/node45.html
>
> However, using a solver like Clp, which has been made especially for
> LP/QP, could also give good performance.
>
> Stefan
>
> On 08/31/2013 02:10 PM, Jeffrey Dean Kelly wrote:
> > Hi All;
> >
> > I am wondering what would be the best settings for the IPOPT options if
> > IPOPT is being used to solve a pure LP or QP problem beyond the obvious
> > ones of jac_c/d_constant = yes and hessian_constant = yes, etc.?
> >
> > I am embedding IPOPT has a sub-solver in a SLP framework whereby at each
> > major iteration the Jacobian, etc. are re-estimated and IPOPT is called
> as
> > a pure LP or QP.  However, it takes alot of minor iterations for IPOPT to
> > converge the successive LP and QP sub-problems and hence the question.
> >
> > Looking forward to a response - Jeff
> >
> >
> >
> > _______________________________________________
> > Ipopt mailing list
> > Ipopt at list.coin-or.org
> > http://list.coin-or.org/mailman/listinfo/ipopt
> >
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
> --
> Scanned by iCritical.
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>



-- 
**********************************************************
Better Data + Better Decisions = Better Business
Jeffrey D. Kelly
Industrial Algorithms LLC.
Email: jdkelly at industrialgorithms. <alkis at industrialgorithms.com>ca
http://www.industrialgorithms.com/
Skype: jeffreydeankelly
**********************************************************
This email and any files transmitted with it are confidential, proprietary
and intended solely for the individual or entity to whom they are addressed.
If you have received this email in error please delete it immediately.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130902/40ffb467/attachment.html>