[Trilinos-Users] Tpetra::Vbr results for 1x1 block not matching Tpetra::Crs

Hoemmen, Mark mhoemme at sandia.gov
Mon Oct 28 14:04:33 MDT 2013


On Oct 28, 2013, at 12:00 PM, <trilinos-users-request at software.sandia.gov>
 <trilinos-users-request at software.sandia.gov> wrote:
> Message: 2
> Date: Sun, 27 Oct 2013 20:03:34 -0500
> From: "C.S. Natarajan" <csnataraj at gmail.com>
> Subject: [Trilinos-Users] Tpetra::Vbr results for 1x1 block not
> 	matching	Tpetra::Crs
> To: "trilinos-users at software.sandia.gov"
> 	<trilinos-users at software.sandia.gov>
> Message-ID:
> 	<CAO_E+w6AwyDVwWu6kmh+-LBtsUUU8W0-8MYNw4LsdwxcWLyuVg at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
> 
> Hello All,
>        I have a question concerning the Tpetra::VbrMatrix interface. For
> 1x1 blocks I would think that the Tpetra/Epetra Crs interface and the
> Tpetra Vbr interface should produce identical results. However, I am seeing
> that this is not the case. I am wondering if I am abusing the interface or
> just doing something blatantly wrong. I am attaching a code below that
> generates the FEM system ( with Q1 elements) for 2D (vanilla-) Poisson in
> parallel. The Tpetra::Vbr interface doesn't produce the same results (as in
> correct) as the crs interface in epetra/tpetra.  (The stencil should be 9
> nnz's per interior point--not adjacent to boundary; with 8/3 diagonal and
> -1/3 off diagonal). The correct values are being passed for insertion,
> however, it looks like there is some issue during the summation from
> off-processor elements. The epetra/Tpetra crs matrices and the Tpetra Vbr
> Matrix all have their values inserted into identical row and column indices.

Thanks for noticing the issue and submitting a test case!  Would you mind filing a bug using our bug tracking system?

https://software.sandia.gov/bugzilla/enter_bug.cgi?product=Tpetra

You might have to create an account if you don't have one already.  You can submit test cases more easily through this interface, and track progress until the issue is resolved.  This is a particularly busy time of year for Trilinos developers, due to the upcoming Trilinos Users' Group meeting.  Thus, it could take us a while to address the issue.  Filing a bug report will help insure that we don't lose track of it.

Thanks!
mfh


More information about the Trilinos-Users mailing list