anyone who tried or made BSOR ( block Succesive over relexation method)?

[genesis]

anyone who tried or made BSOR ( block Succesive over relexation method) Cn version?

In my thought, BSOR is a possible algorithm with PE core ( it is a kind of parallel version of SOR)

So many CAE(Computer Aided Engineering) related softeware is made by SOR.
but applying Explicite Finite Difference algorithm or Method (EFDM) to the board is not easy and not efficient.

(of course, using swazzle function, poly version EFDM was already provided as finance technical example but its efficiency is relatively low )

Anyone who have strategy of efficient acceleration of EFDM or BSOR ?

[julien.langou@cudenver.ed]

Hello,

Interesting post. If you are using SOR or BSOR, you need to know the optimal "omega" or at least you need to
be able to approximate its value failry well. For which kind of matrices do people in CAE use SOR (or BSOR)?
Do you have any reference? In any case, you are right. It should be possible to parallelize BSOR fairly well on a
board. (Whether it converges or not is another business).

In the case of the 2D Laplacian problem using Finite Difference Method, you can use SOR and get a
convergence rate that depends on O(h) so that's a big improvements compared to Jacobi or Gauss-Seidel
(which depends on O(h^2)) but you need to know this optimal omega ... In general, people prefer to use CG
which does not require any "optimal" parameter to be set while providing a convergence rate which depends
on O(h).

Julien.