Scalable parallel generation of partitioned, unstructured meshes

Publisher Summary This chapter focuses on the Parallel Generation of Partitioned, Unstructured Meshes for use in the finite element solution of computational fluid dynamics problems on parallel distributed memory computers. The use of unstructured meshes allows the straightforward representation of geometrically complicated domains and is ideally suited for adaptive solution techniques provided the meshes are sensibly distributed across the processors. The chapter describes an algorithm, which generates well-partitioned unstructured grids in parallel and then discusses the quality of this mesh and its partition, and how this quality can be maintained as the mesh is modified adaptively. The algorithm is implemented using message passing interface (MPI) so as to ensure portability. It runs successfully on a variety of platforms, including a distributed memory computer, a shared memory computer, and also on a cluster of 16 Silicon Graphics (SGI) Indy workstations. The chapter emphasizes a number of extensions of partitioned mesh, including its application to time-dependent problems using adaptive solution method through local h-refinement and de-refinement.