[Ipopt] MPI error using Ipopt‏

[Seth Watts]

It has been something like 3-4 years years since I had this problem, so my
memory is a bit hazy. Perhaps someone with more recent knowlege can chime
in, but if not, I can go dig through my notes to find out the details. With
that caveat:

ipopt.opt is for specifying runtime options, not compile-time settings. It
allows you to, e.g. change the maximum number of iterations at runtime. You
specify compile-time settings via the invocation to the compiler. To use
the ma57 solver, I recall needing to download the source files from the
HSL, which requires you to certify non-commerical (and perhaps only
academic) use. Once you have the source file, you place it in the
appropriate ThirdParty directory within the ipopt directory and compile
with the appropriate flags as specified in the ipopt documentation.

- Seth

On Tue, Oct 20, 2015 at 11:36 AM, max leo correa cordova <
max-leo-19 at hotmail.com> wrote:

> Thanks for the hints. It seems that compiling with -lcoinmumps makes that
> error even when deleting the dummy mpi.h. I have only compiled with
>
> mpic++ -o prueba MyNLP.cpp prueba.cpp
> -I/home/max/CoinIpopt/build/include/coin -L/home/max/CoinIpopt/build/lib
> -lipopt -llapack -ldl -lblas -lgfortran -lm -lquadmath -lcoinhsl -lcoinmeti
> s
>
>  and worked, but the solution time has increased (6x). I have tried to use
> the ipopt.opt file to use ma57 but the solver is still using ma27.  Is
> there any chance to use mumps or other solver to get a better performance?
> If so, which is the most recommendable?? How to use the ipopt.opt file (is
> that included in the compilation)?? I will appreciate your help.
> Thanks in advance
> Max
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20151020/b2ab2f2d/attachment-0001.html>