DELAUNDO creates triangular grids based on the Frontal Delaunay Method (Frod). The triangulation algorithm employed is Bowyer's/Watson's using the circumcircle criterion. The ambiguity due to round-off error is handled by using double precision coordinates and radii and by defining a thickness of the circumcircle with all nodes found to be on the circle not breaking the triangle. This leads to a certain minimum node distance that must be exceeded in order to guarantee that the thickness of the circle remains small in comparison to the radius. This minimum distance is calculated as 10E-10 times the maximum distance in x or y. For CFD purposes ten orders of magnitude ought to be sufficient. Delaundo generates an initial Delaunay triangulation from a set of input boundary points. This initial grid is used as the basis of a background grid to interpolate a local length scale between the nodes of the triangles in the grid. As the connectivity in this background grid is uniquely determined by the distribution of the boundary nodes, unwanted interpolation between finely discretized concave geometries may occur. To prevent FroD from using a too fine scale, the user can require no connectivity between specific geometry segments using the ANTICO statements in .ctr and .pts files. Extra nodes are then inserted that are only present in the background grid and remove the illicit connections. The spacing at these inserted nodes is extrapolated from the surfaces with an average gradient. It is recommende to view the stretching and spacing distribution of the background mesh before generating the mesh by using the OUTTYP b option in the .ctr file. In the case of triangulations with stretched regions, extra nodes are inserted at a user defined distance DELTAS around each frontal surface. This defines a rim of triangles around the frontal surfaces where the stretching magnitude decays from the user specified maximum aspect ratio MAXASP at the boundary to the isotropic value of 1. at the outer edge of the rim. In this stretched region rectangular boxes along the surface are built with a height of the maximum facelength of the boundary node distribution divided by the locally interpolated stretching magnitude. The boxes are being built as long as the corners of the box remain in the rim of the frontal surface they are associated with and as long as the height of the box is smaller than its width. In order to achieve a smooth transition from the stretched node generation process to the isotropic one, the boxes ought to exhibit isotropy where the process ceases. This occurs automatically at the outer edge of the stretched layer where the local aspect ratio tends to unity. At trailing edges or corners the boundary node distribution has to be chosen fine enough to revert to isotropy here as well. Isotropy cannot be achieved where stretching rims of different surfaces inter- fere with each other and the stretching magnitude does not decay to isotropy. To handle these cases, up to two of the outermost layers of stretched wedges can be opened for retri- angulation of the isotropic process using ISMOOT in the .ctr file. Further grid quality improvement can be achieved by swapping grid diagonals to minimize the maximum angle speci- fying FLATSW in the .ctr file. The isotropic node generation process generates nodes from frontal faces that are detected as faces shared between acceptable nearly equilateral triangles and non-acceptable skewed triangles. The ensuing node distribution is smoothed by requiring a spacing disk around each node that may not be violated by any other node before new nodes are inserted. The process stops when no more grid improvement by node insertion can be achieved. The resulting triangulation is extremely regular except where fronts merge or collapse. Hence, it is recommended to keep outer boundaries non frontal using the ITYPBN statement in the .pts file in order to have the 'sew-up' as far away from the physical boundaries as possible. Boundary conformality is enforced by angular swapping at various stages of the generation process. However, the currently employed algorithm is rather crude and not fool- proof. It can fail on very crude boundary discretizations where non-conformality across several consecutive nodes occurs. In that case, refine your boundaries in highly curved regions, or wait for the next update. In case of problems, mail to muller@engin.umich.edu. If you want to know more about the method, read the references: [1] J.-D. M\"uller, H. Deconinck, P.L. Roe, ``A Frontal Approach for Node Generation in Delaunay Triangulations'', AGARD R 787. [2] J.-D. M\"uller, H. Deconinck, P.L. Roe, ``A Frontal Approach for internal node generation in Delaunay Triangulations.'', Submitted to Int. J. Num. Meth. Fluids. [3] J.-D. M\"uller, ``Proven Angular Bounds and Stretched Triangulations with the Frontal Delaunay Method'', to be presented at the 11th AIAA CFD Conference, Orlando, 1993. Enjoy, Jens. ------------------------------------------------------------------------ | Yabbadynamics: n., progressive force which enables Fred Flintstone | | to power a stone automobile with just his feet. | | | | Jens-Dominik Mueller, University of Michigan, Ann Arbor | | muller@engin.umich.edu | ------------------------------------------------------------------------