On Tue, 27 Apr 2010, Derek Gaston wrote: > On Apr 23, 2010, at 8:46 AM, Lorenzo Botti wrote: > >> To start smoothly with the exodusII format I've implemented output for >> discontinuous solution, see the attached patch. >> It seems to work with ex21... > > Did anyone receive this patch? I don't see the attachment.
I got the patch on the direct Cc to me; it probably didn't make it through the list filter. It looks like a good start and it shouldn't interfere with other code; I'll commit it shortly. > Note that the limitation isn't that it's a single element type mesh: it's > that one "block" or "subdomain" can only contain one type of element. This > is fairly ingrained in exodus (you literally can't have more than one element > type in a block with exodus) but we could come up with a scheme for dealing > with it (like having "virtual" blocks that are just the regular block number > plus 1000 or something... and then we reconstruct the original subdomain ids > when we read the mesh). > > I'm open to ideas. I think that's a reasonable way to do things given the format limitations. > As for h-refinement for high order approximations... I like the idea, but I > don't know exactly how you would implement that with libMesh. The trouble I > have is that if we change the mesh and project the solution during output... > haven't we just mucked up our solution vector and mesh for when we go back to > solving the problem? But, I seem to remember that some mesh file formats > (GMV) has similar capability now... so I may just not understand how to do > this with libmesh. We store non-projected copies in UniformRefinementErrorEstimator, and could do the same here... the alternative is to refine one element at a time during the output code, which would work great for discontinuous output but which would require tricky consistent node numbering if there's any continuous variables in your system. GMV just copies the whole mesh when outputting an error vector, so it can modify that mesh without messing up the original. I'm lazy and RAM is cheap. --- Roy ------------------------------------------------------------------------------ _______________________________________________ Libmesh-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-devel
