[Trilinos-Users] AztecOO::Iterate for Multiple RHS

Chetan Jhurani chetan.jhurani at gmail.com
Tue May 21 22:39:09 MDT 2013


I see in trilinos-11.0.3 that although arguments of AztecOO::Iterate
are Epetra_MultiVectors, it solves for only the first column and ignores
the rest.

Is this observation correct?  Also, AztecOO::recursiveIterate does
contain the logic for multiple RHS.  Should that be used always?

I have attached a file, which is a modification of
aztecoo/example/simple_epetra_aztecoo/cxx_main.cpp and prints the
norms of all columns post-solution.  Only the first column has a
non-zero norm.  (Just a quick test rather than computing the residual
AX - B.)  Its output is shown below.  See the last few lines.

Thanks,

Chetan

----------------------------------------------------------
AztecOO in Trilinos 11.0.3

Epetra::MpiComm::Processor 0 of 1 total processors.

                *******************************************************
                ***** Problem: Epetra::CrsMatrix
                ***** Preconditioned GMRES solution
                ***** 1 step block Jacobi
                ***** No scaling
                *******************************************************

                iter:    0           residual = 1.000000e+000
                iter:    1           residual = 5.334313e-001
... skipped
                iter:   19           residual = 7.398873e-003
                iter:   20           residual = 7.217189e-034


                Solution time: 0.004107 (sec.)
                total iterations: 20
Solver performed 20 iterations.
Norm of true residual = 5.10944e-015

xColumnNorms[0] = 38.0818
xColumnNorms[1] = 0 ######################  all others zero
xColumnNorms[2] = 0
xColumnNorms[3] = 0
xColumnNorms[4] = 0
----------------------------------------------------------


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: cxx_main_modified.cpp
Url: https://software.sandia.gov/pipermail/trilinos-users/attachments/20130521/458c2792/attachment.pl 


More information about the Trilinos-Users mailing list