[Trilinos-Users] anasazi and eigs with singular matrix
Simone Deparis
deparis at MIT.EDU
Thu Dec 1 18:57:16 MST 2005
Hi Heidi,
Thornquist, Heidi K wrote:
> Hi Simone,
>
> Yes, we have experience with this type of problem. In fact, an example
> for solving something similar
> to this can be found in BlockKrylovSchurEpetraExGen.cpp.
thank you, I look through this.
> What specific eigenvalues are you looking to
> compute for this generalized problem (Ax = \lambda Bx), i.e. largest
> magnitude, smallest magnitude, rightmost?
Actually I'll work with the symmetrized version of A: (A+A^T)/2 and B is
already simmetric.
I look for the larger and the two smallest (in magnitude) eigenvalue +
vector.
> I would be glad to help you set this up. If you want to take a look at
> the example, we can start there.
Thank you. I have to finish to set up my matrices and run some tests and
then I'll start with the eigsproblem.
Thank you
Simone
PS: I add here some answers to David:
> Another question:
> How accurate do the eigenvectors need to be?
I need an accurate upper bound for the largest and accurate lower bounds
for the smallest ones. But since I am realistic, I take what I get :-)
> Are you checking that the inf-sup bound?
Do you mean the inf-sup lower bound? Among others.
I have a set of matrices A, each independend one from an other and they
are undefinite. The only helping bit is that since I simmetrize, the
eigs will be real.
> Hello,
>
> I have an eigenvalue problem of the form Ax = \lambda B x
>
> where B is a singular matrix (positive semidefinite).
> I am interested in modes orthogonal to the zero mode and I whish to use
> the Anasazi::BlockKrylovSchur method.
>
> Does somebody have an experience with something similar?
>
> Thank you
> Simone
>
-------------- next part --------------
An embedded message was scrubbed...
From: "David M. Day" <dmday at sandia.gov>
Subject: Re: [Trilinos-Users] anasazi and eigs with singular matrix
Date: Thu, 01 Dec 2005 16:39:01 -0700
Size: 3548
Url: http://software.sandia.gov/mailman/private/trilinos-users/attachments/20051201/470d4ace/AttachedMessage.mht
More information about the Trilinos-Users
mailing list