[Coin-ipopt] Release of IPOPT 2.0.1
andreasw at watson.ibm.com
Wed Jul 24 00:58:24 EDT 2002
Not too much has happened on this mailing list ;-). Again, if you have
any feedback (positive or negative) or questions, we would be very happy
to hear about it.
After the initial release of IPOPT 2.0, some gradual changes have been
made to the CVS repository version (which is always reflected in the
dated tarball the day afterwards). Today, I added some more changes,
and released the new version as IPOPT 2.0.1. (A tarball with this
name will be available soon.)
Beside some cosmetic modifications and bug fixes, the following
changes have been made since IPOPT 2.0:
- AMPL_interface: Fixed bugs related to Hessian-vector products and
to using maximization (instead of minimization) in AMPL formulation.
- MC19 (available in Harwell Archive) can be used as alternative to
MC30 (not available in Harwell Archive).
- CUTE interface upgraded to CUTEr (CUTE revisited) interface. The
official CUTEr distribution (see http://hsl.rl.ac.uk/cuter-www/) now
comes with the interface and scripts for IPOPT. The source code for
this interface is also available in the IPOPT distribution, and
modifications to the interface can be performed here.
- Parts of the package have been ported to Cygwin, a UNIX environment
running on Windows, which provides the GNU compilers available in
the standard installation of Cygwin. The IPOPT makefiles are set up
in a way, so that the executables will run on any Windows (command
prompt) environment, i.e. no installation of Cygwin is required to
execute them. For this, MinGW (minimal GNU for Windows) is used.
At this point, nothing related to DLLs has yet been ported. In
particular, the DYNOPC component of the IPOPT package cannot be
compiled. If you would be interested in this part, let us know.
Also, I was only able to test the Cygwin version on Windows 2000 -
any feedback regarding other Windows versions is highly appreciated.
- I just realized that the patch files responsible for the
modification of TRON do not work on our (old) Dec Alpha running
OSF1, and IPOPT crashes if the restoration phase with TRON is
entered. If you would like us to fix this, please send a message
to this mailing list.
- The makefiles have been re-arranged slightly. In particular, the
system dependent definitions are now done in separate include files.
Also, it is now possible for a user to separately specify the
makefile variables FFLAGS, CFLAGS, and CXXFLAGS defining the flags
passed to the Fortran, C, and C++ compilers.
- IMUINIT = 0 allows "warm start", i.e. the user provided estimates
for the Lagrangian multipliers (LAM, V_L, and V_U) are used as
starting points (instead of initializing them within IPOPT). V_L
and V_U might be perturbed by IPOPT in order to make sure they are
sufficiently positive, that is, greater than DBNDPUSH.
- The initialization of the equality constraint multipliers via the
solution of a least square system (if applicable) is skipped, if
the Jacobian of the equality constraints is rank-deficient in the
first iteration. In that case, the multipliers are set to zero.
More information about the Coin-ipopt