[petsc-users] best PETSC solver for long nozzles

[Matthew Knepley]

On Sat, Jan 29, 2011 at 4:26 AM, Stephen Wornom <stephen.wornom at inria.fr>wrote:

> Geometry: Tube with L/D is >> 1
> Characteristics:
> -Unstructured partitioned mesh with 500K vertices globally.
> -The tube wall is heated, thus solution evolves in the x-direction.
> -X is the dominant  flow direction.
> -Seek steady solution using the time advancing scheme with a low Mach
> number preconditionner.
> *Question*: Which *PETSC* solver would give the best convergence? Stephen
> p.s.
>

What are you currently using? I see two main points

  a) Make sure your mesh partitioning is compatible with this dominant flow
direction. It sounds like
      any decent partition should be, but check it.

  b) Probably more important is a good preconditioner for heat transfer
coupled to incompressible flow. Have
      you tried block preconditioning using FieldSplit. Treating the whole
thing as a black box is usually
      pretty terrible. An optimal preconditioner is described in the 4th
talk by Marc Spiegelman here: http://www.bu.edu/pasi/materials/

   Matt


> My present solver  does not  take into consideration that the flow has a
> dominant flow direction (x).
> Thus the convergence is very slow,  30000 time steps using 64-processors
> with a CFLmax= 50 applied in each cell. At 15000 time steps the 0-L/2
> portion of the tube is converged with an additional 15000 time steps needed
> to converge the L/2-L part of the tube.
>
> --
> stephen.wornom at inria.fr
> 2004 route des lucioles - BP93
> Sophia Antipolis
> 06902 CEDEX
>
> Tel: 04 92 38 50 54
> Fax: 04 97 15 53 51
>
>


-- 
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/20110129/0a87e742/attachment-0001.htm>