1
$\begingroup$

I'm working on a problem that gives a matrix system of the form D + epsilon*S, where S is a skew-symmetric matrix. I'm interested in finding if any work has been done to develop a conjugate gradient method to exploit this structure (ideally for any epsilon, but possibly just for epsilon << 1), or if it's necessary to use biconjugate gradient or something similar.

I found the paper "ITERATIVE SOLUTION OF SKEW-SYMMETRIC LINEAR SYSTEMS" by CHEN GREIF AND JAMES M. VARAH which outlines what I have in mind (for epsilon = 0). After naively coding it up and seeing what happens when there is a non-zero diagonal I've found that their method will only converge for large epsilon.

Improve this question
$\endgroup$
2
  • $\begingroup$ How large is your system? Where does it come from? I don't understand your second paragraph; if $ \epsilon = 0 $ don't you just have a diagonal matrix? Since $ D + \epsilon S $ is a small perturbation of a diagonal matrix, I would expect a very simple scheme like Jacobi or Gauss-Seidel to converge fairly quickly. To be a little more sophisticated you could try diagonally preconditioned GMRES. Finally, you might get more answers at scicomp.stackexchange.com $\endgroup$
    – Andrew T. Barker
    Commented Sep 21, 2012 at 19:09
  • $\begingroup$ I recommend asking this question on scicomp.stackexchange.com. $\endgroup$
    – David Ketcheson
    Commented Oct 10, 2012 at 18:19

0

Reset to default

You must log in to answer this question.

Browse other questions tagged .