[petsc-users] DMComposite and ksp

[Barry Smith]

On Mar 3, 2011, at 6:52 AM, gouarin wrote:

> Hi,
> 
> I'm still working on my stokes problem with 2 DM (one for the velocities and an other for the pressure). I do the best preallocation for my matrix.
> After the construction of my matrix, I do a matrix vector operation and I compare the result for one process and 2 processes. If I pay attention to the numbering of my fields between these 2 problems, I have the same result.
> 
> But when I do just 2 iterations of my ksp solver, I don't have the same result at the end and of course the residual is not the same.

  What preconditioner are you using? Unless you use Jacobi you will always get a different set of iterations. (And even with Jacobi due to different order of operations you will eventually see a different convergence route between 1 and 2 processes.  You cannot be concerned about this, the things to be concerned about are:

1) Do 1 and 2 processes eventually converge to the same answer (bad news if they do not)

2) Does 2 process converge almost as quickly to the answer as one process (bad news if adding more processes slows the convergence greatly so the solution time ends up being longer).

  Barry


  
> 
> Do you know what I do wrong ?
> 
> Thanks,
> Loic
> 
> -- 
> Loic Gouarin
> Laboratoire de Mathématiques
> Université Paris-Sud
> Bâtiment 425
> 91405 Orsay Cedex
> France
> Tel: (+33) 1 69 15 60 14
> Fax: (+33) 1 69 15 67 18
>