| Thread Previous • Date Previous • Date Next • Thread Next |
That seems reasonable. I would propose adding a struct-variable underufc::cell that captures the polynomial map (without quadrature), and some kind of ENUM variable that indicates the order of the map. And a boolean variable to ufc::cell that says whether it is straight or not. Of course, a good data structure for the xml format should be chosen...
Then, I think, there would be a switch inside tabulate_tensor that chooses between straight and curved. The code for straight would be the same as before, and for curved, it could just be quadrature (at first).
This will effect FFC. But will it effect FIAT? Or FErari? - Shawn On Thu, 14 Aug 2008, Martin Sandve Alnæs wrote:
Without having thought this through in any way, I think this should rather be additions to the ufc::cell struct than adding a new struct. -- Martin 2008/8/14 Shawn Walker <walker@xxxxxxxxxxxxxxx>:I'm not against changing the xml format. Actually, it would be nice to have a format that is backwards compatible; that way you could ignore the curved information and just use the mesh as straight if you want. That seems do-able. And then add a local struct to the UFC format that is accessible by the tabulate_tensor routine. This local struct would have the local nonlinear map for each cell and facet in place. I need to look at the UFC manual some more and see how this might work... - Shawn On Thu, 14 Aug 2008, Johan Hoffman wrote:On Wed, 13 Aug 2008, Jed Brown wrote:On Wed 2008-08-13 17:37, Shawn Walker wrote:ok, but how would you adjust the xml mesh format? That should be figured out first. I was thinking one could just add extra data for each cell that specifies the location of the other control points for the triangle sides. Or maybe it would be better to just supply a finite element function (2nd order or higher) for the x,y,z coordinates of the mesh and do it that way?I'm a fan of storing coordinates just like any other function, but you probably want variable order (p-adaptivity) for this since you would like most elements to have an affine mapping (for efficiency and to improve the accuracy of the quadrature).Well, if DOLFIN has p-adaptive functions built into it, then this would not be a problem. If it doesn't have it yet, then it could wait. In the meantime, I would just have a full P2, or P3 (2nd or 3rd order) function for all triangles. And, in addition, provide a boolean function on each cell that says whether it is straight or not.I am also interested in curved boundaries. If we do not want to change the xml format, maybe it would be enough to use a boolean MeshFunction over the boundary to achieve what you describe? /Johan_______________________________________________ FFC-dev mailing list FFC-dev@xxxxxxxxxx http://www.fenics.org/mailman/listinfo/ffc-dev
| Thread Previous • Date Previous • Date Next • Thread Next |
