[Ipopt] solving maxflow as LP

[Tamas Vinko]

Thanks for the replies.

I do not think that AMPL spends long time on the processing and
passing the problem to ipopt; after starting the optimisation, I
checked the processes (unix) and saw that AMPL did take some 10
minutes and then ipopt got started to run (for 24 hours or so). After
reading your replies, my guess is that mumps takes that long time at
the beginning.

I made the change suggested by Jeff (see his 3rd point) in the AMPL
model, it helps, but only on the AMPL side, now it takes only 2
minutes to get ipopt started... However, it seems that this trick does
not help to ipopt, it is still running for long time.

I guess changing the mumps to pardiso and using an optimised blas
would help, these will be the next steps.

As a matter of fact, I would still like to use AMPL+ipopt, because
(among other things) I wanted to see if I can compare the running time
of it vs. a Ford-Fulkerson implementation in java. This is why I am
not using ipopt with c++ code.

Thanks again,
Tamas