[Trilinos-Users] Applying Anasazi::BlockKrylovSchurSolMgr

[Alicia Klinvex]

Hello Joseph,

BKS tries to build a subspace of dimension block_size * num_blocks.  If the
dimension of your problem n is less than that subspace dimension, it's
impossible to orthogonalize the basis vectors (since you can't generate 10
orthogonal vectors within a subspace of dimension 4).  Perhaps we should
modify the error message to make this clearer.

Can you try a slightly larger example, perhaps 10x10, with block_size 1 and
num_blocks < 10?

Best wishes,
Alicia

On Sat, Dec 12, 2015 at 1:57 PM, Rutherford, Joseph M <jmruther at illinois.edu
> wrote:

> I’m having some difficulty using Anasazi to compute the smallest
> eigenvalues of a (non-Hermitian) complex-valued linear system.  Before
> addressing a general case,  I have written a simple test using two
> Tpetra::CrsMatrix<std::complex<float>,int,int> instances with reference
> data – this is a very small 4x4 dense case with reference data available.
> (I can share the example code if needed.)
>
>
>
> Unfortunately, I am having a hard time getting the SolMgr to drive the
> solution.  As I’m solving a reference 4x4 system with 4 nondegenerate
> eigenvalues, I’m fairly confident I don’t have a pathological case,
> although it is much smaller than the usual Anasazi use case.  Could you
> please look at the error message reported to my console (below)?  I’ve not
> found a clearly-aligned example in the Anasazi distribution, so I’m hopeful
> this is just a trivial case that is easily fixed.
>
>
>
> Sincerely yours,
>
>
>
> Joe Rutherford
>
>
>
> <console>
>
> $ ./bin/test_eigensolver.exe
>
> Running 1 test case...
>
> unknown location(0): fatal error: in "eigensolver_dense_4x4_complex":
> std::invalid_argument:
> /home/jmruther/branches/bean/externals/trilinos/include/AnasaziBlockKrylovSchurSolMgr.hpp:344:
>
>
>
> Throw number = 2
>
>
>
> Throw test that evaluated to true:
> static_cast<ptrdiff_t>(_numBlocks)*_blockSize >
> MVT::GetGlobalLength(*_problem->getInitVec())
>
>
>
> Anasazi::BlockKrylovSchurSolMgr: Potentially impossible orthogonality
> requests. Reduce basis size.
>
> /home/jmruther/branches/bean/source/tests/eigensolver.cpp(57): last
> checkpoint: "eigensolver_dense_4x4_complex" entry.
>
>
>
> *** 1 failure is detected in the test module "Eigensolver"
>
> $
>
> </console>
>
>
>
>
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at trilinos.org
> https://trilinos.org/mailman/listinfo/trilinos-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://trilinos.org/pipermail/trilinos-users/attachments/20151212/7089f048/attachment-0001.html>