Iterative solver (again) [Was: Re: [ODE] Russ' plans forODE?]

[Alen Ladavac]

> however as we well know from the theory of iterative linear system
solvers,
> for large systems a method based around the biconjugate gradient
> method(Mendoza,Laugier, etc..) will
> be faster. At least theoretically, given that simpler methods usually lead
> to a more efficient implementation.

All correct. That's why we'd like to first be sure that Gauss-Seidel doesn't
converge fast enough before delving into BCG method. Because GS is
_extremely_ simple. :p

Alen

----- Original Message -----
From: <Antonio_Martini at scee.net>
To: "Alen Ladavac" <alenl-ml at croteam.com>
Cc: <ODE at q12.org>
Sent: Tuesday, April 27, 2004 13:28
Subject: Re: Re[2]: Iterative solver (again) [Was: Re: [ODE] Russ' plans
forODE?]


>
>
>
>
> >Yes, I'm talking about something similar to that. Though if you are not
> >running massively parallel (as GPU has to), you can just use
Gauss-Seidel.
> >But those are all just fancy-words. It is same as Jacobi, just meaning
> that
> >you immediatelly use available results.
> methods can only be coupled, like a mixed Jacobi/Gauss-Seidel in that case
> you may trade-off between parallelism and speed of convergence.
>
> however as we well know from the theory of iterative linear system
solvers,
> for large systems a method based around the biconjugate gradient
> method(Mendoza,Laugier, etc..) will
> be faster. At least theoretically, given that simpler methods usually lead
> to a more efficient implementation.
>
>
>
>
>
>
>
> "Alen Ladavac" <alenl-ml at croteam.com>@q12.org on 27/04/2004 14:07:27
>
> Sent by:    ode-bounces at q12.org
>
>
> To:    "Nguyen Binh" <ngbinh at glassegg.com>, <Antonio_Martini at scee.net>
> cc:    ODE at q12.org
> Subject:    Re: Re[2]: Iterative solver (again) [Was: Re: [ODE] Russ'
plans
>        forODE?]
>
>
> > you may me interested in:
> >
> > http://www.shaderx2.com/shaderx.PDF
> >
> > where a projected Jacobi method is mentioned.
>
> Yes, I'm talking about something similar to that. Though if you are not
> running massively parallel (as GPU has to), you can just use Gauss-Seidel.
> But those are all just fancy-words. It is same as Jacobi, just meaning
that
> you immediatelly use available results.
>
> > http://vcg.isti.cnr.it/people/vcgpeople/mendoza/jnrr.pdf
> >
> > if you look carefully the first jacobi iteration started with force=0 is
> the same as step fast where off diagonal elements disappear. so it looks
> > like that step fast is like running the first iteration, setting the
> force
> to zero and running the first iteration again.
>
> I believe that Mendoza uses biconjugate gradient method, but other than
> that, yes, it is very similar. And it's exactly what my recent objection
to
> stepfast was - it does only the first iteration.
>
> Alen
>
> _______________________________________________
> ODE mailing list
> ODE at q12.org
>  http://q12.org/mailman/listinfo/ode
>
>
>
>
>
>
>
>
> **********************************************************************
> This email and any files transmitted with it are confidential and
> intended solely for the use of the individual or entity to whom they
> are addressed. If you have received this email in error please notify
> postmaster at scee.net
>
> This footnote also confirms that this email message has been checked
> for all known viruses.
>
> **********************************************************************
>  SCEE 2004
>
>