[Trilinos-Users] Problem with Moocho...

[Salinger, Andrew]

Hi Luca,

The developer of MOOCHO is unavailable 
for a few weeks. 

One thing that sounded suspicious is that you
set xL = xU for boundary conditions. BCs should
be handled easily enough by the c(x) and not
also need to be set by the bounds.

Since this is a modification of a working 
problem, it appears that the problem has
something with the formulation. In general,
I would try removing or loosening bounds,
since these are complications on the core
algorithm. The "Throw" shows a null pointer
to a matrix operator of a sub-matrix. I could
imagine this happening if the current guess
for x violates all bounds and the remaining
space is null. 

Sorry I can't be of more help.
Andy

-----Original Message-----
From: trilinos-users-bounces at software.sandia.gov [mailto:trilinos-users-bounces at software.sandia.gov] On Behalf Of Luca Heltai
Sent: Thursday, October 08, 2009 4:49 AM
To: trilinos-users at software.sandia.gov
Subject: [Trilinos-Users] Problem with Moocho...

Dear Trilinos users,

I've been using Moocho for a while now (ver. 9.0.3), and I managed to solve a few interesting problems with it. Lately I modified a piece of my nlp specification to generalize a work i did in the past. Just to give you some insights:

we're talking of

min f(x)

subject to c(x) = 0

xL< x < xU

x is a n vector
c(x) is a m dimensional non linear function, f(x) is a (highly) non linear function and some of the xLs and xUs are the same (boundary conditions).

The difference between now and before is that the m dimension grew a lot. I don't use the ThyraEpetraModel, because i do not have a natural "a priori" separation of the variables in dependent and independent.

I get the following warning in the MoochoJournal:


***********************************************************
*** DecompositionSystemVarReductImp::update_decomp(...) ***
************************************************************

Warning!!! mat_rel != MATRICES_INDEP_IMPS; The decompsition matrix objects may not be independent of each other!

I was getting the same warning before, but then everything always worked out. I checked the matrices by hand, and they seem full rank to me, so this should not be a problem...

The nlp test is succesfull, however Moocho throws an exception which I don't know if it's my fault or not:

MoochoSolver: Caught an std::exception of type std::logic_error described as : ../../../../package s/moocho/src/AbstractLinAlgPack/src/abstract/interfaces/
AbstractLinAlgPack_MatrixOpSubView.cpp:427
:

Throw number = 1

Throw test that evaluated to true: M_full_.get() == NULL

Error, the MatrixOpSubView object has not been initialize!


Any idea of what can be wrong? My code is a simple modification of the WB counterexample (different xL, xU, x0, m_orig_, n_orig_ and definitions of c(x) and f(x), but the rest is pretty much identical).

Any help is really appreciated.

Thanks,

Luca.

--
Luca Heltai <luca.heltai at gmail.com>
http://people.sissa.it/~heltai/
Scuola Internazionale Superiore di Studi Avanzati
Phone:  +39 040 3787 449, Office: 255
--
There are no answers, only cross references


_______________________________________________
Trilinos-Users mailing list
Trilinos-Users at software.sandia.gov
http://software.sandia.gov/mailman/listinfo/trilinos-users