[Ipopt] asymmetric Hessian matrix

[Stefan Vigerske]

Hi,

the only thing that comes into my mind is that if you use Mac OS X Lion, 
then there is a problem when using the optimized blas routines that come 
with the system, for details see
  https://projects.coin-or.org/Ipopt/ticket/181

This issue has not been fixed in an Ipopt release yet (only in trunk and 
latest stable branch).
One workaround is to use netlib's blas/lapack implementation (do 
get.Blas and get.Lapack in ThirdParty/Blas and ThirdParty/Lapack, resp.) 
and building Ipopt with the configure options --with-blas=BUILD 
--with-lapack=BUILD .

Stefan

> Hi,
>
> First when I use IPOPT for my problem, I see the bound constraints for
> parameters are violated. The library throws an exception, and the program
> terminated. To be specific the contraint on that parameter is [0.005,
> 0.006], however I see the library is trying to use 0 for that parameter.
>
> Later, I enabled second order deriviative check, however the numerical
> calculation gives asymmetric Hessian matrix for contraints. The values are
> very different one is about 3000, the other is about 3.0e13. I spend some
> time on the library source code. I found the following line changes the
> parameter magically:
>
>   for (Index i = 0; i<  nx; ++i) {
> std::cout<<  ivar<<  "  2.3     "<<  xref[i]<<  "  "<<  xpert[i]<<
> std::endl;
> }
>
> IpBlasDcopy(nx,&zero, 0, gradref, 1);
>
> for (Index i = 0; i<  nx; ++i) {
> std::cout<<  ivar<<  "  2.5     "<<  xref[i]<<  "  "<<  xpert[i]<<
> std::endl;
> }
>
> The output is:
> 1  2.3     0.0481804  0.0481804
> 1  2.3     700.976  700.976
> 1  2.3     0.0165778  0.0165778
> 1  2.3     2.54335  2.54335
> 1  2.3     0.27909  0.27909
>
> 1  2.5     0.0481804  0
> 1  2.5     700.976  700.976
> 1  2.5     0.0165778  0.0165778
> 1  2.5     2.54335  2.54335
> 1  2.5     0.27909  0.27909
>
> IpBlasDcopy is a Fortran function, is that the problem?
>
>
>
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt


-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan