Funfional delaunay refinement.pdf

Functional Delaunay RenementXiang-Yang LiDepartment of Computer ScienceUniversity of Illinois at Urbana-ChampaignUrbana, IL 61801xli2@cs.uiuc.eduAbstract:Given a complex of vertices, constraining segments (and planar straight-line constraining facets in 3D) and an -Lipschitz control spacing functionf() over the domain, an algorithm presented herein can generate a conform-ing mesh of Delaunay triangles (tetrahedra in 3D) whose circumradius-to-shortest-edge ratios are no greater than p2 (2 in 3D). The triangle (tetra-hedron) size is within a constant factor of f(). An implementation in 2Ddemonstrates that the algorithm generates excellent mesh.Keywords: unstructured mesh generation, delaunay renement, controlspacing, mesh conformity.1 IntroductionSimplex mesh generation has many applications, including numerical meth-ods such as the nite element method, computer graphics and geographicinformation system. To ensure accurate result, the simplices of the meshmust be well-shaped, in the sense that they have small aspect ratios, i.e.,the smallest angles are bounded from below [1, 15]. To get accurate nu-merical simulation, we also expect to have small element size. On the otherhand, we prefer to have larger element size due to the time complexity ofthe simulation. Hence elements should have properly chosen size and shapethat adapt to the complex geometry and solution accuracy.A control spacing function species the desired element size at eachpoint of the domain. Several heuristics and algorithms had been developedto generate a mesh whose element sizes conform well to a given controlspacing function. Splitting the longest edge, or subdividing the simplicesare the most used heuristics. Shimada [14] used the particle simulation tond a good mesh vertices set, then constructed the nal mesh by Delaunaytriangulation. There are no any theoretical quality guarantees given byabove algorithms. The algorithm by Miller et al [11] uses the maximalindependent set of a large random