[Trilinos-Users] Amesos and complex numbers

Heroux, Michael A maherou at sandia.gov
Thu Jul 24 08:49:17 MDT 2008


As long as your original matrix is well-conditioned, the answer you get back from the equivalent real form will be unique.

On a related note, I have completed a class that forms an equivalent real system using two Epetra matrices in much the same way that the Komplex package works.  I am testing it now and should have it in Trilinos 9.0.

I don't think we will have time to make Amesos work directly with Tpetra complex matrices before the Trilinos 9.0 release.


On 7/23/08 1:56 PM, "Nikhil Kriplani" <nmkripla at ncsu.edu> wrote:


I am trying to use Amesos LU factorization for solving a matrix that
consists of complex numbers. I've set up 2 real matrices using Epetra,
Areal and Aimag, and built up an augmented system as described on the
komplex web page.
So essentially, I am doing an LU factorization on a real matrix of
twice the dimension.

My question is, after the LU factors are generated and the solution
generated, is this solution unique? I guess this is more a
mathematical question that a software one, but do I have to some
compensation/scaling to the solution with this scheme?

Also, when Tpetra complex number support comes out in Sep, will Amesos
LU solver be compatible with factorizing a complex matrix?


Trilinos-Users mailing list
Trilinos-Users at software.sandia.gov

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://software.sandia.gov/mailman/private/trilinos-users/attachments/20080724/815a69d5/attachment-0001.html 

More information about the Trilinos-Users mailing list