[Coin-ipopt] Small problem to solver the big one...

[Andreas Waechter]

Hi Karsten and Arvind,

I found out what the problem with setting IPOPT options from within AMPL
is:

The solver executable thinks it's name is 'ipopt' (not 'ipoptAMPL') so
that it is looking for

option ipopt 'imaxiter = 10000';

and not option ipoptAMPL 'imaxiter = 10000';

So, in order to set IPOPT options from within AMPL, please use the solver
keyword 'ipopt_options' instead of 'ipoptAMPL_options'. However, you will
still need to specify 'option solver ipoptAMPL', unless you rename your
solver executable to `ipopt'.

In the next release I will change the name of the AMPL solver executable
to `ipopt' instead of `ipoptAMPL' (which is a nicer name anyway).
However, this will conflict with the CUTEr script `ipopt' which we might
want to rename then.  Does anyone have objections to that, or a better
idea?

Karsten, I also had a look at the example model that you set to Arvind.
My version of IPOPT was able to solve it in 1 iteration.  Let's try to get
this part sorted out off-line - and possible post a resolution of our
discussion on the mailing list afterwards.

Thanks again for bringing the problem with setting IPOPT options from AMPL
to our attention!

Best,

Andreas


On Fri, 16 Apr 2004, Arvind Uppili Raghunathan wrote:

> Hi Karsten,
>
>   I have some prior experience in formulating optimal control problems in
> AMPL and solving it using IPOPT. Usually things have worked or atleast I
> have made them work. Let me know if I can help you in any way. You can
> send me the mod files if you wish and I can get back to you with my
> understanding.
>
>   About AMPL not accepting the ipoptAMPL_options flag - I have had trouble
> too. It seems like AMPL is not passing the flags correctly when invoking
> the solver. You can also try the following,
>
>  i) create a .nl file by issuing
>     write gstub ;
>  after you have loaded the model in AMPL.
>  ii) run the model from command prompt as
>      ipoptAMPL stub "imaxiter = 10000"
>  This should work.
>
>   Please let me know if I can be of any help.
>
> Cheers
> Arvind
> >
> > On Fri, 16 Apr 2004, Karsten Theissen wrote:
> >
> > > Hello again,
> > >
> > > I tested our ipopt-version with the programs Andreas told me. "of course it
> > > worked...." So it seemed that something has to be wrong with my models (by
> > > the way : AMPL and LOQO solved them without problems). So I write down the
> > > easiest model (for solving optimal control problems with ODE's) that I can
> > > imagine and  "of course it doesn't work". He told me that he reached the
> > > iteration limit. So, now comes the small problem:
> > >
> > > Telling him:
> > >
> > > option ipoptAMPL_options "imaxiter = 10000";
> > >
> > > doesn't work. He ignores it.
> > >
> > > Can anybody tell me how I can write ipoptAMPL_options so that ipopt get it??
> > > I only find the names of the options in the "Help-Readme-File" but not the
> > > exact syntax (but maybe I'm blind).
> > >
> > > Thank you very much!
> > >
> > > Karsten
> > >
> > > P.S.: If anybody has solved discrete optimal control problems with ODE's or
> > > PDE's with ipopt it would be nice if he can exchange some experience about
> > > the ipoptAMPL options.
> > >
> > > _______________________________________________
> > > Coin-ipopt mailing list
> > > Coin-ipopt at www-124.ibm.com
> > > http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-ipopt
> > >
> > >
> >
> >
> >
>
>
>
>
>
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at www-124.ibm.com
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-ipopt
>