[Trilinos-Users] AztecOO conjugate gradient

Bart Janssens bart at bartjanssens.org
Thu May 9 10:17:00 MDT 2013


Hello,

I'm trying to use the AztecOO CG method to solve a 3D Poisson problem,
resulting from a finite element discretization. Unfortunately, I always get
the following error, and the solution stops on the first iteration:
                *******************************************************
                ***** Problem: Epetra::CrsMatrix
                ***** Preconditioned CG solution
                ***** Ifpack_AdditiveSchwarz, ov = 0, local solver =
                ***** `IFPACK IC (fill=0, drop=0.000000)'
                ***** Condition number estimate = 1.31748
                ***** No scaling
                *******************************************************

                 iter:    0           residual = 1.000000e+00


        ***************************************************************

        Warning: direction vector is no longer A-conjugate
        to previous vector but solution has not converged.


The AztecOO parameters are as follows:

 AztecOO ->
  Output Every RHS = 0   [default]
  Forward Solve ->
   Max Iterations = 400
   Tolerance = 1e-12
   AztecOO Settings ->
    Aztec Solver = CG
    Output Frequency = 1
  Adjoint Solve ->
   Max Iterations = 400   [default]
   Tolerance = 1e-06   [default]
  VerboseObject ->
   Output File = none
   Verbosity Level = high

Using Belos BlockCG I seem to get a solution, but there it seems fragile
when using ML preconditioning, i.e. depending on ML parameters I get this
error (only in parallel):
src/build/aur/trilinos/src/trilinos-11.0.3-Source/packages/belos/src/BelosCGIter.hpp:419:

Throw number = 2

Throw test that evaluated to true: SCT::real(alpha(0,0)) <= zero

Belos::CGIter::iterate(): non-positive value for p^H*A*p encountered!
src/build/aur/trilinos/src/trilinos-11.0.3-Source/packages/belos/src/BelosCGIter.hpp:419:

I use trilinos 11.0.3. I have already verified that the matrix is exactly
symmetric. Any ideas what could be wrong?

Kind regards,

Bart
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20130509/065642a5/attachment.html 


More information about the Trilinos-Users mailing list