On Monday September 2 2013 11:51:58 Jan Blechta wrote: > I just recall what I asked before. Is the approach > > vertex_index*dofs_per_vertex+local_dof > > sustainable? I believe there are plans to abandon 'common cell' > concept, are there? Then `dofs_per_vertex` could (I'm not sure) become > variable across the mesh.
This could be incorporated into the map. > Check also the problem of ghost dofs. These are covered by > `dof_to_vertex_map` by negative and greater-than-local-range values. As > a result this is not invertible and ghosts can't be retained in > an inverse `vertex_to_dof_map`. Again the sole purpose of the vertex_to_dof_map was to be able to be used in parallel. Then ghost dofs need to be excluded so one could use set_local. _You_ needed the opposite mapping together with ghost dofs and added that. I did not see any problems as the original vertex_to_dof_map was retained. This is only a problem if one require the mapping to be bijective, which they by design are not. > Wouldn't it be better to use some more robust map types? This might be handled by some more fine grained mapping for example one for each sub space, but then the initial benefit of mapping any vertex based vector values to a Function is lost. Johan > > Jan > _______________________________________________ > fenics mailing list > fenics@fenicsproject.org > http://fenicsproject.org/mailman/listinfo/fenics _______________________________________________ fenics mailing list fenics@fenicsproject.org http://fenicsproject.org/mailman/listinfo/fenics
