Dear Johan,

> i need to perform a modal analysis on my solver, would like to avoid
> poweriterate and use levelshift. Anyone implemented an efficient jacobi
> solver?

I don't know about an efficient implementation of a Jacobi eigensolver
- you want to get the matrix eigenvalues and eigenvectors, don't you?
Or is it enough to repeatedly use a Jacobi preconditioner?

Regarding eigenvalues:
One thing you could try instead (and that is probably even more
efficient than the Jacobi transformation method) is to find a viable
method for your problem in the eigensolver package Anasazi of the
software Trilinos. You can find a tutorial of it at 
http://trilinos.sandia.gov/Trilinos9.0Tutorial.pdf
and the doxygen documentation here:
http://trilinos.sandia.gov/packages/docs/r9.0/packages/anasazi/doc/html/group__anasazi__solver__framework.html).
 
A more general paper that discusses software for eigenanalysis can be
found at
http://trilinos.sandia.gov/packages/anasazi/anasazi-toms.pdf
Anasazi implements a few efficient (and even parallel) algorithms for
matrix eigenvalues. So you could use the Trilinos wrappers in the latest
subversion of deal.II to generate your matrix (or just copy over a
deal.II matrix into that format) and then call the Anasazi algorithm of
your choice...

And probably PETSc also has some eigenvalue algorithms...

Best,
Martin


_______________________________________________

Reply via email to