Re: initial guess for vector x in bicgstab on ITL 4.0

Hi, Rich Lee.
Sorry for the delay, I was making some additional tests.
I'm using Kdevelop 3.1.0 on KDE 3.3.0, and gcc version 3.3.4 (pre
3.3.5 20040809) and SuSE 9.2.
Yes, the same thing occurs with gmres and qmr, but when I use gmres, I
have to do vX = vB2, where vB2 is the preconditioned vector b, as in
the example gmres.cpp:
/*
[line 132]  itl::solve(p, b, b2); //gmres needs the preconditioned b
to pass into iter object*/

In bicgstab, vX0 = vB means that the initial search vector is null,
right? Does this problem has something to do with it? Or the problem
is with my matrices?

Thanks again,

Márcio Ricardo

On Wed, 23 Mar 2005 08:08:33 -0800, Lie-Quan Lee
<[email protected]> wrote:
> What happens if you use gmres or qmr instead of bicgstab? What is the
> preconditioner you used?
> 
> --Rich Lee
> 
> On Mar 23, 2005, at 5:23 AM, Márcio Ricardo Pivello wrote:
> 
> > Hi,I'm using ITL 4.0.0-1 with MTL 2.1.2-21, and bicgstab only
> > converges when I use the right-hand-side vector as the initial guess
> > for the solution vector. Otherwise, a dot(r,rtilde)=0 always happens
> > in iteration=2. Is that ok or maybe my linear system has something
> > wrong?
> >
> > Thanks in advance
> >
> > Márcio Ricardo
> >
> > _______________________________________________
> > This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/
> 
>

_______________________________________________
This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/