[Ipopt] IPOPT with CPLEX?

[Mikhail Nediak]

Well, hypothetically, could not one try to find a feasible solution to
min   0^T x
           A x = b

using CPLEX? Of course, CPLEX was NOT specifically designed to do this...

However, I wonder if linear algebra routines in CPLEX would beat MUMPS 
when accessed like that.

Mikhail




From:
Andreas Waechter <andreasw at watson.ibm.com>
To:
Vatsal Bhatt <vatsalbhatt at gmail.com>
Cc:
ipopt at list.coin-or.org
Date:
23/10/2008 02:38 PM
Subject:
Re: [Ipopt] IPOPT with CPLEX?



Hi Vatsal,

MUMPS and Pardiso are solving linear systems, i.e., for a given matrix A 
and a vector b, it gets a vector x that satisfies

Ax = b

Here, A is a symmetric indefinite matrix.

CLPEX is a Linear Programming solver.  This is something entirely 
different.  It solves LPs like

min  c^T x
      A x >= b

CLPEX has completely different functionality from MUMPS and cannot replace 

it.

Andreas

On Thu, 23 Oct 2008, Vatsal Bhatt wrote:

> Many thanks to the developers of IPOPT for large scale NLPs and making 
it
> opensource.
>
> Since our MARKAL-MACRO energy systems analysis model had a problem 
running
> with other solvers, GAMS suggested us to use IPOPT. It handles our model
> well, we are happy that atleast it runs...
>
> Since our model has several thousand equations with several hundred 
thousand
> non-zeros, it takes nearly an hour for a run. It uses MUMPS as a linear
> solver and the documentation has suggestions on using PARDISO.
>
> We use CPLEX for our LP based MARKAL, which solves the same system in LP
> within 5 minutes (without 5 macro economic equations).
>
> Would it be possible for us to use CPLEX instead of MUMPS within IPOPT?
>
> Thanks in advance for your help.
>
> Vatsal
>
_______________________________________________
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/20081023/3f5eb223/attachment.html