[vtkusers] Polyhedral Cells

[Jeff Lee]

Kent Eschenberg wrote:

> Hi Jeff,
>
> --On March 06, 2005 10:41:18 AM <jeff at cdnorthamerica.com> wrote:
>
>> Is there a face-based format for the cell description?  If the cells are
>> all nonconvex, they can be tetrahedralized (tessellated) and vtk can 
>> deal
>> with the tets.  If there are any convex cells, then you'll need to diy.
>> Right now vtk doesn't deal with face-based cells, but I could see it
>> being done.
>
>
> We can convert the polyhedrals to tetrahedrals but would prefer to 
> avoid that. VTK can be extended to new cell types but its a lot of work.

Yes, alot of work, but will probably be your only option.  I believe 
that tetrahedralization only works reliably if the cells are convex.  
vtk has a cell type called vtkConvexPointSet which would work if you 
could ensure that the cells are actually convex.  It actually does a 
delaunay tetrahedralization on the cell points (it holds a collection of 
tets for you).  If you have any concave cells, then this will fail to 
recover the actual polyhedron boundaries.  I think that it is next to 
impossible to guarantee nonconvex cells when doing cfd, therefore a 
face-based approach is the right one.  this means alot of work - first 
defining a cell type, second making sure the filters you want will work 
with face-based information.  Since most algorithms work on a per-cell 
basis, you will get more than 50% there by implementing the new cell 
type.  Other filters which specialize on vtkUnstructuredGrid will have 
to be modified.

>
>> Are these polyhedra the result of the "dual" of a tet mesh?  What solver
>> are you using?
>
>
> The polyhedrals result from dynamic mesh refinement. We are replacing 
> a home-grown solver with an open source solver. OpenFOAM is one we are 
> considering. However, the choice of output file format is a seperate 
> issue - we'll rewrite the output code in the solver as needed.

Are we talking about completely arbitrary polys, or things like "hex 
with corner cut" - i.e. can the polys be characterized as modifications 
to existing cell-types?
-Jeff

>
> Thanks for your thoughts!
> Kent
> Pittsburgh Supercomputing Center
>
>