[Coin-ipopt] New C++ Version

Andreas Waechter andreasw at watson.ibm.com
Tue Dec 14 23:03:51 EST 2004


Hi Javier,

It's nice to hear that you are interested in helping us with the
interfaces to other linear solvers!  (The only thing is that we probably
can't start right at this moment, since I'm going on vacation tomorrow and
won't be back until beginning of January - I hope you are not in a
terrible hurry to get this done...)

As for the linear solvers, one important item here is that we usually need
a linear solver that can factorize symmetric indefinite systems and tell
us how many negative eigenvalues the factorized matrix has (except for the
special case where the problem is convex, or at least it is clear that the
reduced Hessian is always positive definite - in that case any linear
solver could be used).

Pardiso is definitely an option, but as far as I know, only the most
recent version (Aug 2004) has the capability to compute the number of
negative eigenvalues (inertia) of the factorized system; as far as I
understand, the version in the MKL library does not provide that.

As for TAUCS, I have heard that it is now able to compute the inertia, but
I haven't followed up on that yet - it has been on my to do list for a
while but I keep being distracted.  I think it would be great to try it
and see how well Ipopt and TAUCS would work together.

I suggest that we correspond in January about the interface to the linear
solvers (we still have to come up with a clear and general definition of
the interface).  The goal is that we come up with something that will
allow us to easily hook up any of the linear solvers that could do the
job, and provide a variety of interfaces so that Ipopt users can choose
whatever fits their needs.

Best regards,

Andreas



On Tue, 14 Dec 2004, Javier Pedrido wrote:

> Hi Andreas,
>
> First of all, thank you very much for your reply.
> I'm interested in writting/trying to implement the interfaces with other
> sparse linear solvers. Which of the solver you've mentioned would be the
> best replacement for the harwell libraries?. There was a discussion in
> Sep-2003 about linking ipopt with TAUCS in the mail list, but TAUCS wasnt
> ready for the task back then. How about the pardiso solver included in the
> intel MKL library? It is not free, but at least it does not have a per user
> license.
>
> Best regards,
> Javier
>
> _________________________________________________________________
> Comunica le tue emozioni in tempo reale con MSN Messenger!
> http://www.msn.it/downloadmessenger/
>
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-ipopt
>




More information about the Coin-ipopt mailing list