[Libmesh-devel] idea & plans digest

[Daniel D. <d.d...@tu...>]

libMesh is cool: stable and offering truly _impressive_
functionality.  It's fun coding!


For now, i chose to babble and took the liberty to summarize
the ideas currently spread across the 1e6 discussion and the good ol'
roadmaps.  Note that there is no order or something in this list,
and that some issues may already be outdated?







* more examples
  - one that shows the differences between the partitioners,
    and perhaps simply outputs each partitioning as a GMV.



* parallelizing the mesh
  - with AMR not in the near future
  - with disabled AMR rather straightforward, possibly
    already in progress?



* moving new points to a user-supplied geometry
  - this would require some more Surface children,
    from which also something like a MultiSurface
    could be derived that can handle multiple surfaces
  - the actual moving: straightforward



* robust smoothers




* improve web page
  - libmesh-users@: done
  - overview of examples available
  - documentation for the src/apps things?
  - eye candy:
    - either as links to some other web pages,
    - or included in libMesh's page,
    - or combinations thereof.
    - some exterior acoustics things will come
      from my side, 11-12/03, perhaps earlier




* more shape functions
  - in progress: Steffen & Hendrik



* interface to tetgen
  - i really want to volunteer for that, provided
    you accept that it will take quite some time
    to come...




* FE for dim-1 elements
  - the inverse map (from x,y,z to xi, eta, zeta)
    works for dim-elements living in any-dimensional
    space (where any form 1,2,3).  So this is fine.
  - However, the Jacobian is computed assuming
    that 2D faces _do_ live in 2D space.  So we always
    assume volume-like elements



* solve eigenproblems
  - interface to SLEPc: already done once, should
    be straightforward.  I think Steffen already
    volunteered, but he has enough other things to
    do right now (see shapes)
  - is the interface to SLEPc the only one to offer?
    May only be said once this is tested some more




* integrating s3/simulator class?
  - integrating s3 in libmesh would be nice.  However
    to be honest, s3 is quite complicated, not only for a
    newbie.
  - i have coded some simulator -- however,
    it is custom-tailored to acoustics things.
  - Is the whole simulation thing (at least currently)
    not too problem-specific, so that it is hard to
    integrate it into one uniform design?
  - How do other packages resolve this issue?
    --- To consider the giant, PETSc seems to go
    towards being multilingual.  Is this  the way libMesh
    wants to go?




* regression tests
  - i have been thinking a bit about this.  I would
    like to volunteer for this, provided that you
    accept that
    (a) it is more or less make- and csh-script based,
    (b) will take time to come.
    First, i think it makes sense to test the output
    of the examples.  Then, there may be additional
    executables in the test directory.  Haven't thought
    about the latter much yet.




* and now the ideas that are probably way too far off:
  - matrix-free: actually not really necessary.  From
    my experience, the real memory hog is the mesh,
  - incompatible meshes/interaction/mortar...