From: John P. <pet...@cf...> - 2004-09-16 13:18:28
|
Andrea Pollini writes: > I have also taken a look to src/geom/elem_refinement.C and I have > already implemented a Tet4 that do bisecting refinement. The problem is > that I think pheraps it would be better to design another way to do > refinement. I try to scketch my ideas. I think that the refinement > algorithm have to know how to divide elements, and elements haven't do > know it. In this way one can implement the refinement strategy he like > without having to derive a new element for every strategy, but simply > passing the mesh in the constructor of the refinement class. So I have > for example > > class RefinementPartitioning { ... } > class RefinementDelaunay { ... } > class RefinementAnisotropic { ... } > > > Or pheraps using a Strategy Pattern in order to implement algorithms > better. > > What do you think about this? If you agree Ii can try to design it > out.... I want to do this because I'm studying a two phase transition > problem (namely a Steffan problem with Allen-Cahn approximation) and I > wanted to use libmesh to make some comparison about ifferent adaptive > meshing strategies It sounds good, but personally I am not familiar with the Strategy Design pattern, so I can't comment on it. The classes which you describe above sound similar to the MeshRefinement class. Do you mean them to be replacements (or derived from) that class? Since libMesh is open source, we welcome contributions from other developers, but we may not have the man-power to review complex patches in a timely manner. We are interested in alternative refinement strategies, and I will try to help you with any questions you have about the library along the way. Good luck! John |