hello I have noticed that, in order to define the 1D mesh size there are two possibilities depending on the setting of "lcFromCurvature" flag. In case it is set (lcFromCurvature=1) the parameter "minCircPoints" defines the mesh size on a curved surface. This is an isotropic refinement so that with a cylindrical surface it is done in both circumferential and axial direction. In my opinion this is not a good strategy because on a long cylinder characterized by a small radius (like a thin wire) it produces an excessive (and not useful) refinement in the axial direction. The situation becomes even worse in the meshing of a solid with rounded edges having small radius. In this case the two curves which limit the rounding region are divided in a big number of small edges (dictated by the small radius radius) and than, in the following 2D meshing process, the high refinement is propagated to the triangles of the adjacent faces. So in the end a small detail (the rounding of a corner) which should have a negligible impact on the numerical simulations (I am considering electromagnetic problems but this general considerations should remain valid also in other applications) produces a very bug change in the global mesh. The number of triangles can increase of a big factor and the numerical simulation can easily become unfeasible. I think that all these problems would be avoided by a simple change of the 1D meshing with a mesh size that depends only on the linear curvature of the underlying geometrical curve and does not depend on the surface curvature in the orthogonal direction.
Walter Steffè _______________________________________________ gmsh mailing list [email protected] http://www.geuz.org/mailman/listinfo/gmsh
