[Trilinos-Users] Behavior of Epetra_VbrMatrix and Epetra_MultiVector

Heroux, Michael A maherou at sandia.gov
Thu Jan 3 15:41:21 MST 2013


Bill,

This is in principle how the VBR format is supposed to work.  Here are a
few suggestions:

- Check return codes on the Epetra functions you call.  You could be
running into an undetected error.
- Make sure the Epetra_BlockMap you use for your vectors is compatible
with the matrix maps.  When solving Ax=b, typically, you should use the
same map for x, b, and for the row map of A (which is passed in to the
constructor).
- You typically do not specify the column map.  It gets built for you.
Also, the domain and range maps are typically defined for you (usually
just a reference counted copy of the row map).

If you continue to have problems, send me a code snippet or small test
program.

Mike  

On 12/28/12 9:24 AM, "Bill McGrory" <mcgrory at aerosoftinc.com> wrote:

>I'm confused about the use/behavior of Epetra_VbrMatrix and
>Epetra_MultiVector
>
>I am trying to port a legacy CFD software package to use Trilinos. I
>have internal iterative sparse solvers which I can compare against.
>
>I am solving a coupled system of equations (let's just say 5 for now,
>(mass, 3x momentum and energy)
>So I thought I would put fill a VbrMatrix where each block was the
>Jacobian of the local system (dF/dQ_i) 5x5 blocks
>and I filled my MultiVector over the number of control volumes, x 5
>where the 5 was each equation I am solving
>
>I'm Implicit Newton iterations for time integration, so I need to solve
>an A dQ = -R type system, and I'm using Belos.
>
>This fails to reproduce my behavior of the legacy code, so I step back
>and tried to fill my A with  diagonal dense blocks, and 0 off diagonals,
>and when I solve this embarrassingly simple problem I get a solution
>that is only dependent on the 1,1 entry of the dense block, which tells
>me I'm not interpreting the use of VbrMatrix and MultiVector correctly?
>
>What am I doing wrong?
>
>I'd hate to think that I have to fill each term in a CrsMatrix
>
>TIA
>Bill
>
>-- 
>Dr. William D. McGrory    AeroSoft, Inc.
>mcgrory at aerosoftinc.com   Suite 1400
>(540) 557-1904            2000 Kraft Drive
>(540) 557-1919 (FAX)      Blacksburg, VA 24060
>
>




More information about the Trilinos-Users mailing list