[Ipopt] IPOPT uses incorrect number of variables

[Seth Watts]

Hi Stefan -

>> By default, all fixed variables (lb=ub) are taken out there.

Thanks, with that clue I was able to find a bad loop index in my bounds
assignment that left all but one of the parameters with lb = ub.

Thanks for the help!

- Seth


On Mon, Jun 18, 2012 at 1:26 PM, Stefan Vigerske
<stefan at math.hu-berlin.de>wrote:

> Hi,
>
> the output
>
> Total number of variables............................:        1
>
> refers to the internal problem representation of Ipopt.
> By default, all fixed variables (lb=ub) are taken out there.
> Can you make sure that you do not have almost all variables fixed?
>
> You can change what Ipopt is doing with fixed variables via the
> fixed_variable_treatment option.
>
> When asking for function values, Ipopt transforms the internal point
> back to the users formulation. That's why you see again 640 variables in
> the eval methods.
>
> Hope this helps,
> Stefan
>
>
> On 06/18/2012 08:11 PM, Seth Watts wrote:
> > Hello all -
> >
> > I have written a NLP class that uses code I have written using C++ and
> the
> > Petsc library to evaluate the objective and constraint functions, their
> > derivatives, etc. The number of parameters is set at run-time by reading
> in
> > a value from an input file. Although I have confirmed that this value
> (128)
> > is read in correctly and passed to my class file correctly, IPOPT appears
> > to act as though there is only one variable. That is the number displayed
> > in the summary table (Total number of
> > variables............................:        1) and when I have the
> > parameters sent to the screen, only the first value changes - the others
> > remain zero (or at least on the order of 1e-310).  Additionally, despite
> > setting number of constraint functions to 5 and number of nonzeros in
> jac_g
> > as 5*n = 640, the summary table shows only 5 nonzeros in jac_g. However,
> > the value of 'n' in each of the class methods is the correct value of
> 128.
> >
> > I can't reproduce this problem with simple toy problems; that is, if I
> make
> > a simple MyNLP class file for which n is set to 128, m to 5, and
> nnz_jac_g
> > to 640, IPOPT runs correctly, the right values appear in the summary
> table,
> > and so on. This leads me to suspect the problem is in something I have
> done
> > in my class definition, but I don't know what it could be. In the
> > get_starting_point method, the values returned to x are as I intend, but
> > the next time a new_x flag is true and I print out the parameters, all
> but
> > the first value have been set to near-zero values.
> >
> > I am running IPOPT version 3.10, checked out via subversion last week. I
> am
> > using MUMPS as my linear solver. I am using the quasi-Newton option so
> that
> > I do not supply the Hessian matrix (as it is both dense and expensive to
> > compute). I am running Ubuntu 12.04 64-bit.
> >
> > Thanks in advance for any suggestions you might have.
> >
> > - Seth
> >
> >
> >
> >
> > _______________________________________________
> > Ipopt mailing list
> > Ipopt at list.coin-or.org
> > http://list.coin-or.org/mailman/listinfo/ipopt
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20120618/fdb9ad04/attachment.html>