Live Webcast 15th Annual Charm++ Workshop

Many applications use unstructured meshes for solving a variety of problems in various fields of science and engineering. Most of these applications solve problems over irregular domains and tend to use unstructured meshing software. Applications could be brain models, climate models or engineering models to study material deformation and weapons strength. Apart from solid meshes, there exists a huge set of applications that solve computational fluid dynamics problems over meshes. Millions of lines of mesh framework source code exists to satisfy these applications. Most of these frameworks do not simultaneously support parallelism and geometric adaptivity - 'coarsening and refinement'. Parallel adaptivity is a challenging problem of significance to application scientists. They would like to be able to adaptively give more importance to one part of the problem being studied, while keeping the computational requirements from blowing up. This requires the ability to refine parts of a mesh while at the same time coarsen other parts. The contribution of this thesis is a novel way to perform parallel adaptivity on large meshes. This thesis introduces an incremental method to perform parallel adaptivity. We define a set of primitive operations on a mesh and use these to design a set of atomic operations to perform adaptivity in parallel. Incremental operations give us a very low level tool for modifying the mesh. This gives a lot of power and flexibility to modify the mesh in any way one wants. The entire adaptive component is implemented as part of ParFUM, a parallel mesh framework. Moreover, scientific and engineering applications are successfully using or trying to use the parallel adaptivity presented in this thesis. These applications include engineering mechanics applications and Spacetime Discontinuous Galerkin applications.