[petsc-users] Pulling submeshes out of a DMPlex object

[Matthew Knepley]

On Wed, Sep 6, 2017 at 2:29 PM, Lawrence Mitchell <
lawrence.mitchell at imperial.ac.uk> wrote:

> Dear all,
>
> for various reasons I'd like, given an existing DMPlex mesh, to be
> able to pull submeshes out.  These might take various forms:
>
> 1. The locally visible part of a distributed mesh (so that I can
> discretise and solve independent problems on each local part)
>

This is already a standalone Plex. If you delete the PetscSF, it never
knows anyone else is out there :)


> 2. Some subset of entities of a given codimension (for example, the
> boundary of a domain, or the skeleton of the mesh).
>

There is a "depth" label in a Plex (DMPlexGetDepthLabel) that does this.


> So I think conceptually what I want is:
>
> Create a DMLabel that marks entities of a given codimension.
>
> Say
>
> DMPlexCreateSubmesh(dm, label, value, height, comm, &subdm)
>
> This should walk all entities of the specified height and gather them
> if they are labelled with the provided value, then the subdm is
> created by treating those entities as the cells of a new mesh and
> adding their transitive closure.
>

Yes, this makes sense. I also tend to include a map from the submesh back
to the original.


> There appears to be some submesh functionality in plex right now, but
> it doesn't appear to do what I want:
>
>   DMPlexCreateSubmesh - Extract a hypersurface from the mesh using
> vertices defined by a label
>   DMPlexFilter - Extract a subset of mesh cells defined by a label as
> a separate mesh
>
> The latter looks close to what I want for 1, except the change of
> communicator.
>

Can you explain this problem? I think this does what you want.


> The former seems somewhat odd.  I would have thought the way to
> extract a hypersurface would be to mark the faces of the hypersurface
> and build a mesh as a I described above.
>

It just shows you that everyone wants something different. Engineers and
scientists tend to
only know about cells and vertices, so they always want to the interface to
use those.

Earthquake people want to extract faults, but they also want to cells
attached to either side
to make things understandable.


> It looks like it should be possible to reuse some of the existing
> submesh routines to do what I want to do, but I'm a little lost.
>
> What I tried was just adding an option to
> DMPlexCreateSubmeshGeneric_Interpolated to mark a submesh by
> traversing the closure of points marked with a given value.
>

It should do that automatically. Did you try it as is? Maybe you could give
me a small
Plex and show me why its giving something wrong.

  Thanks,

     Matt


> But that swiftly ran into problems because the resulting submesh
> didn't have depth strata as expected, and looking at the existing code
> I'm quite confused.  The subpointMap that a submesh DM has labels the
> points of the parents with their depth (in the original mesh?).  But
> this seems to be used to determine the depth of the points in the
> submesh.  Which seems odd to me.
>
> I realise this is all rather woolly right now, perhaps in attempting
> to clarify any queries I will explain better what I want to do.
>
> Cheers,
>
> Lawrence
>



-- 
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

http://www.caam.rice.edu/~mk51/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20170906/2b0b93e0/attachment.html>