[Trilinos-Users] Constrained Eigenproblem Any Experience ?

Davood Ansari david.ansari at gmail.com
Sat Jul 5 22:23:44 MDT 2008


Thanks to the help from Chris Backer I managed to do what I had in mind with
a projection operator for solving my constrained generalized eigen problem A
x = lambda B x.
I am using a projection operator O that operates on any vector which needs
to be exposed to the A^-1 B operation in the BKS solver.
Basically O's purpose is to keep the vector out A's null-space.

The issue is that the projection operator is not so sparse and occupies lots
of memory. I started wondering that the C (as in Cx=0)
from which I construct my projection operator O can be manipulated
arbitrarily by adding and subtracting linear combinations of its rows (it is
a full rank matrix).
I guess then I can make C a lot more sparser but the new C cannot be used to
build the projection O (though it still satisfies Cx=0).
Thus I am thinking of alternative methods including [A C' ; C 0] x = lambda
[B 0 ; 0 0 ] and more particularly CLOPS from the CLAPS package .
Do you think it will work better ?

I wonder if CLOP directly uses the provided C or goes to further manipulate
it which then will impose unpredicted computational and memory costs?

BTW is there any way I can get the CLAPS package.

Kindly Comment

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://software.sandia.gov/mailman/private/trilinos-users/attachments/20080706/1be1f0a7/attachment.html 

More information about the Trilinos-Users mailing list