On 6 August 2015 at 07:49, Lizao Li <lixx1...@umn.edu> wrote: > Dear all, > > I plan to implement edge integral assembly in 3D in FEniCS. >
Nice. We have an open feature issue on this, https://bitbucket.org/fenics-project/dolfin/issues/106. > It is good for a number of things, for example assembling the canoniacl > projection for 3D Nedelec edge elements. One issue is that there are more > than 3 cells intersecting at an edge in 3D. > My concern has been whether we can do edge integration efficiently without clever analysis of the form. For example, if an edge integral doesn't need all data from all connect cells (and there might be a lot of connected cells), will an assembler that gets all data be performant? > At the level of UFL, my design is to add, > ds('m') - arbitrary choice of one of the cell values (min cell > index, for example) > ds('avg') - average over all the adjacent cell values > ds('jump') - sum over the jumps at all facets around the edge in the > right-handed direction (which happens to be the one I care about the most) > > Suggestions, hints, and pointers to a good starting point in particular > are welcome~ > > I think we need something other than ds. Perhaps we need to be able to pass the topological dimension to dx. Take a look in measure.py from UFL for background. Garth > Best regards, > Larry > -- > Lizao (Larry) Li > Univeristy of Minnesota > > _______________________________________________ > fenics mailing list > fenics@fenicsproject.org > http://fenicsproject.org/mailman/listinfo/fenics > >
_______________________________________________ fenics mailing list fenics@fenicsproject.org http://fenicsproject.org/mailman/listinfo/fenics
