How a user code looks like (so far)

[fdrmrc]

Assume we're given a vector of agglomerations. Each agglomeration is described by a vector of cells, so cells_to_be_agglomerated should be a vector of vectors. They can be given all in one shot or one vector after the other.

In the following, I'll try to describe the public interface we have so far by commenting line by line what we have in the tests.

The following three lines mark internally masters and slaves and create BoundingBoxes for every agglomeration.

// cells_to_be_agglomerated and cached_tria are given
FE_DGQ<2> dg_fe(1);
AgglomerationHandler<2> ah(cached_tria,dg_fe);
ah.agglomerate_cells(cells_to_be_agglomerated);

Then, DoFS are distributed in the triangulation on master and standard deal.II cells.

ah.initialize_hp_structure();

After that, for every master cell we store connectivity information about agglomerated faces. That is needed for doing ah.reinit(cell,agglomerated_face) in an assembly procedure.

ah.setup_connectivity_of_agglomeration();

This is all we need to compute the sparsity pattern associated to a Discontinuous Galerkin discretization where some cells are agglomerated together.

SparsityPattern sparsity_pattern;
ah.create_agglomeration_sparsity_pattern(sparsity);

[fdrmrc]

@luca-heltai This is the interface we have so far.

Maybe we should rename initialize_hp_structure with distribute_dofs_on_agglomeration, or similar names.

Do you think that DoFs distribution (initialize_hp_structure()) and setup of connectivity (setup_connectivity_of_agglomeration()) should be merged into one call? Something like setup_agglomeration(), for instance?

[luca-heltai]

I agree on simplifying. We could provide a single user entry point in the public interface, maybe distribute_agglomerated_dofs(), (following the convention in deal.II e.g., distribute_mg_dofs) that would call

ah.initialize_hp_structure();
dh.distribute_dofs(fe);
ah.setup_connectivity_of_agglomeration();

So that a user could do:

// cells_to_be_agglomerated and cached_tria are given
FE_DGQ<2> dg_fe(1);
AgglomerationHandler<2> ah(cached_tria);
ah.agglomerate_cells(cells_to_be_agglomerated);
ah.distribute_agglomerated_dofs(dg_fe);

Notice that I removed the FE from the constructor, and I moved it into the distribute_agglomerated_dofs argument, so that we can maybe change fe, or build it before deciding what fe to use, as in dh(tria); dh.distribute_dofs(fe)

[fdrmrc]

Thanks. I applied the suggestion in 555b9a6