[ODE] biconjugate gradient

[Ignacio Castaño]

On Friday 21 May 2004 00:50, Russ Smith wrote:
> i was reading the paper by Mendoza, Laugier, and Faure ... they appear
> to make the claim that the biconjugate gradient method is better than
> plain 'ol conjugate gradient for near singular rigid body systems.
> i quote:
>
>   "matrix sparsity allows the use of a biconjugate gradient algorithm
>    [16], which iteratively refines a global solution even with singular
>    matrices"
>
> however: aren't the two methods equivalent when the matrix is symmetric?
> (as J*inv(M)*J' definitely is). why is mendoza using BiCG when it has a
> speed penalty but no advantage?

Hmm... that's strange, in my experience BiCG and other variants had worse 
convergence than standard CG and I had to rely on GMRES or hierarchical 
preconditioners when the matrix was near singular. In any case I'm not an 
expert in numerical methods and that was on a completely different problem...

-- 
Ignacio Castaño
castanyo at yahoo.es