[Ipopt] Ipopt on Mac

[Andreas Waechter]

Hi Christoph,

The Java interfaces was contributed by Rafael de Pelegrini Soares of 
VRTech Industrial Technologies.  I myself don't know the interface, nor 
Java, and neither do I have a Mac :-)

But maybe I can still say something useful:

It seems that the solution you found via the Java interface is actually 
also a local solution, or at least a stationary point.  If I give that 
point to Ipopt as a starting point (via an AMPL model) and I set the Ipopt 
options so that the initial point is not perturbed much:

mu_init 1e-6
bound_push 1e-6
bound_frac 1e-6
bound_mult_init_method mu-based

then Ipopt takes one iteration and stops, with all termanination criteria 
being satisfied.  This means that your Java interface seems to have done 
its job properly.

Regards,

Andreas

On Thu, 5 Nov 2009, von Tycowicz, Christoph wrote:

> Hello Ipopt-Team,
>
> I am a research assistant at the mathematical geometry processing group of the Freie Universität Berlin.
>
> My colleagues and I would like to use Ipopt via its JNI interface. We 
> already tried the complied binaries for Windows and it worked fine.
> Since I am working on a MacBook (/w Snow Leopard) I tried to compile the 
> latest library (3.8.0) incl. the JNI interface myself.
> Except for the compilation of the tests everything worked fine.
> Running the HS071.java shipped with the library I got the following
>
> Optimal Solution:
> 4.567633003581938
> 1.6613736769136564
> 1.761204198266994
> 3.643449741969065
>
> The documentation of the java-file however states:
> * The optimal solution is
> * x = (1.00000000, 4.74299963, 3.82114998, 1.37940829).
>
> Thats why I am puzzled!? Since there is no error message I don't know 
> where to search for hints.
>
> Additionally I realized that the JNI interface changed since 3.3.2; so I 
> tried to compile and test this old version as well. This however didn't 
> even compile:
>
> ld: duplicate symbol _MPI_COMM_WORLD in 
> .libs/libipopt.lax/liblinsolvers.a/IpMumpsSolverInterface.o and 
> .libs/libipopt.lax/libcoinmumps.a/mpic.o
>
> My colleagues and I would be really grateful if you could help us, so 
> that we can use Ipopt on booth operating systems with a common (java) 
> interface.
>
> Best regards
> Christoph
>
> --
> Christoph von Tycowicz
>
> Freie Universität Berlin
> AG Mathematical Geometry Processing
> Arnimallee 6 — Room 110
> D-14195 Berlin, Germany
>
> Tel.: +49(30)83875860
> Fax: +49(30)83875869
>
> Christoph.vonTycowicz at fu-berlin.de<mailto:Christoph.vonTycowicz at fu-berlin.de>
>
>