Parallel Domain Decomposition Techniques Based On Tetrahedral Mesh Generation Study

With the development of the parallel computer and applied software, application in reality has become more and more complex. For the application of the unstructured mesh, meshes in excess of107elements have become common for production runs in large scale and complex numerical simulation in thousands of cores. The expectation is that in the near future meshes in excess of108or even more elements will be required. As mesh cell numbers become as large as this, the process of mesh generation on a serial computer becomes problematic in terms of computational time as well as memory requirements. So research on parallel unstructured mesh is regarded as significant in real applications.Research on parallel mesh generation can go back to90s last century, with the devel-opment of most twenty years, many research productions have been done by researches, who come from all over the world. There are many research productions in the2D un-structured mesh generation. But the research for3D unstructured mesh generation is still in the stage of researching. And there are many problems to be solved, for instance, con-sistency of the mesh in the interface and the quality of meshes. In this thesis, we will do some research on the unstructured tetrahedral mesh generation and its parallelization in depth.A parallel tetrahedral mesh generation method based on domain decomposition is presented. It is applied for the3-dimension complex geometries. The algorithm con-sists of the following major steps:firstly, constructing the global surface mesh for the3D computational domain; secondly, based on the divided and conquer principle, the global surface mesh is divided into many non-overlapping sub-domains; finally, with the iterative technology, sub-meshes are generated in parallel by constrained Delaunay triangulation.Finally, some numerical examples are given to proof the effectiveness of the al-gorithm from the convergence, efficiency of the parallel computing time and quality of tetrahedral meshes. Experimental results show that millions of or even more tetrahe-dral meshes can be constructed in parallel on parallel computing environment; compared with the sequential algorithm, the time of mesh generation has been decreased a lot and it can also acquire a stable speed-up of parallel computing time; ant the consistency of the meshes in the interface and the quality of tetrahedral meshes has been guaranteed by this scalable parallel tetrahedral mesh generation method.