On Tue, 11 May 2010, Roy Stogner wrote: > You've still got a CFDLab login, right? Check out > ~roystgnr/pecos/ale/doc/paper.pdf for some discussion of the problem.
Yikes - glancing at that doc again reminded me of the last remaining problem: many of our finite element types depend on geometric information in some way to determine shape function orientations or even values. That all goes to hell when the geometric information is changing underneath you. For some elements (e.g. HERMITE) it's irrelevant because they don't handle distorted meshes anyway, for others (e.g. CLOUGH, XYZ) the changes are smooth and you should actually still get a convergent solution, but for the HIERARCHIC elements there's a sign change when edges rotate and you can get hosed. This is a big enough problem that we may even want to go back to indexing-based sign conventions in the HIERARCHICs. We switched to geometry-based conventions so that we could do more caching, but getting fixed-mesh solutions slightly faster might not be worth getting moving-mesh solutions wrong. --- Roy ------------------------------------------------------------------------------ _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
