[petsc-users] Two PCs on same residual in multiplicative fashion !

[Matthew Knepley]

On Fri, Oct 26, 2012 at 1:05 PM, Abdul Hanan Sheikh
<hanangul12 at yahoo.co.uk>wrote:

>
> Yes, it sounds crazy when we look exclusively at this. But I do not do
> this exclusively.
> Basically,  this happening [ P1^{-1} P2^{-1} A x = P1^{-1} P2^{-1} b ]
> occurs as a part of my two-level solver.
> Any how I have to do this.
>
> I had a realization, PCCOMPOSITE does not do exactly what I want. I read
> about PCCOMPOSITE in manual. :(
>

Just use PCSHELL, and in the PCApply() function do two KSPSolve()s, one
with each matrix.

   Matt


> Thanks,
> Abdul
>
>
>   ------------------------------
> *From:* Matthew Knepley <knepley at gmail.com>
> *To:* Abdul Hanan Sheikh <hanangul12 at yahoo.co.uk>; PETSc users list <
> petsc-users at mcs.anl.gov>
> *Sent:* Friday, 26 October 2012, 18:56
> *Subject:* Re: [petsc-users] Two PCs on same residual in multiplicative
> fashion !
>
> On Fri, Oct 26, 2012 at 12:45 PM, Abdul Hanan Sheikh <
> hanangul12 at yahoo.co.uk> wrote:
>
> Thank you Matt, your assumption is alright. It was typo.
>
> I got this idea , yet if I want to invert P2 exactly by PCLU then how
> should I do ??
>
>
> Read this:
> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/PC/PCKSP.html
>
> This makes things more complicated. You can use PCOMPOSITE with
> multiplicative, but its not
> exactly what you have written. This is never used because its not really
> effective, and doesn't make
> much sense from a theory perspective. Are you sure you want to do this?
>
>     Matt
>
>  This is indeed, but I avoid to merge(matrix product) for some reasons.
>
> Thanks again.
> Abdul
>
>   ------------------------------
> *From:* Matthew Knepley <knepley at gmail.com>
> *To:* Abdul Hanan Sheikh <hanangul12 at yahoo.co.uk>; PETSc users list <
> petsc-users at mcs.anl.gov>
> *Sent:* Friday, 26 October 2012, 18:22
> *Subject:* Re: [petsc-users] Two PCs on same residual in multiplicative
> fashion !
>
> On Fri, Oct 26, 2012 at 12:04 PM, Abdul Hanan Sheikh <
> hanangul12 at yahoo.co.uk> wrote:
>
>
> Dear Developers and members,
>
> I intend to solve system Ax=b with two preconditioners as follows:
>
> P1^{-1} P2^{-1} A x = P^{-1} P^{-1} b
>
>
> I assume you mean P1^{-1} P2^{-1} A x = P1^{-1} P2^{-1} b
>
>
>
> With this setup, I think its easiest to use PCKSP with the matrix for
> the KSP being P2. That would give you
>
>   P2^{-1} A x = P2^{-1} b
>
> Now you can precondition that solve with another PCKSP with the matrix
> for that KSP being P1, and get what you want.
>
> This seems like a strange thing since you could just merge those matrices.
>
>    Matt
>
>
>  where P1 and P2 are exclusively available as matrices.
> This might be crazy to many.
>
> I thought to start with PCCOMPOSITE but i do not think it does exactly
> what I intend.
>
> Any Idea would be appreciated.
>
> A beginner!!!
> Abdul
>
>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
>
>
>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
>
>


-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20121026/6b618b96/attachment-0001.html>