Add MatrixFreeTools::ElementBirthAndDeathMatrixFree #14121
Conversation
|
@sebproell and I have been working on resolving issue #12102. We propose not to integrate the notion of FYI @Rombur you have open the original issue. Would this approach work for you? |
include/deal.II/matrix_free/tools.h
Outdated
| * FENothing assigned to them. In the following we call such cells | ||
| * invalid. All other cells are valid. In contrast to MatrixFree, | ||
| * this class skips invalid cells and faces between valid and | ||
| * invalid cells are treated as boundary faces. |
There was a problem hiding this comment.
Maybe mention how to figure out if a face is an actual boundary face or a valid-invalid boundary face inside the user-supplied boundary operation?
There was a problem hiding this comment.
I would use this function: #14122 :)
| const unsigned int type = | ||
| static_cast<unsigned int>(category.first == fe_index_valid) + | ||
| static_cast<unsigned int>(category.second == fe_index_valid); | ||
|
|
There was a problem hiding this comment.
Would Assert(type != 0, ExcInternalError()) be useful here?
There was a problem hiding this comment.
That is invalid face and nothing is done and the face is skipped. I have added a comment.
include/deal.II/matrix_free/tools.h
Outdated
|
|
||
| /** | ||
| * Loop over all valid cells and faces. Faces between valid and | ||
| * invalid cells are treated as boundary faces. |
There was a problem hiding this comment.
I agree that it's what most people want but it is very different than what we do everywhere else. That's fine with me but that's something we need to think about. Since the face is treated as a boundary, which boundary ID is used?
There was a problem hiding this comment.
I agree that it's what most people want but it is very different than what we do everywhere else. That's fine with me but that's something we need to think about.
That is why it is not part of MatrixFree actual. This is helper class tailored for such purposes.
Since the face is treated as a boundary, which boundary ID is used?
I would say: numbers::internal_face_boundary_id. See #14122
peterrum
force-pushed
the
birth_and_death_mf
branch
2 times, most recently
from
0001d73
to
06ceda0
Compare
|
/rebuild |
| is a wrapper around MatrixFree, which helps users to deal with | ||
| DoFHandlers, where there are cells without degrees of freedom, | ||
| i.e., with FENothing assigned to them. The class helps to implement | ||
| "element birth and death technique". |
There was a problem hiding this comment.
Is this a term used anywhere else on the literature? If not, then I would suggest to with something like activation/ deactivation instead.
There was a problem hiding this comment.
You mean ElementActivationAndDeactivationMatrixFree?
There was a problem hiding this comment.
Yes, something like that sounds better to me. Would that be acceptable?
peterrum
force-pushed
the
birth_and_death_mf
branch
2 times, most recently
from
1ca3b8e
to
9c3e1e3
Compare
| if (fe_collection[i].n_dofs_per_cell() > 0) | ||
| valid_fe_indices.insert(i); | ||
|
|
||
| AssertDimension(valid_fe_indices.size(), 1); |
There was a problem hiding this comment.
Can we allow more than one valid_fe_indices? Or create an invalid_fe_index and check for it later?
There was a problem hiding this comment.
Can we allow more than one valid_fe_indices?
Yes, we can. Also, I guess what you need in your case is to not loop over faces neighboring FE_Nothing.
Before making the modifications I would like to get to a status where we agree that the class is useful, we found a name, and maybe merged the first version so that we can improve it from there.
There was a problem hiding this comment.
Also, I guess what you need in your case is to not loop over faces neighboring FE_Nothing.
Yes, this should only be optional from my perspective. For me this class can definitely be useful, currently I skip FENothing cells (or cell batches) manually.
peterrum
force-pushed
the
birth_and_death_mf
branch
from
9c3e1e3
to
f51943b
Compare
peterrum
force-pushed
the
birth_and_death_mf
branch
from
f51943b
to
56856ae
Compare
|
With #14122 being merged, I have rebased this PR! Ready from my side. |
| * deactivated. All other cells are activated. In contrast to MatrixFree, | ||
| * this class skips deactivated cells and faces between activated and | ||
| * deactivated cells are treated as boundary faces. |
There was a problem hiding this comment.
I think it would be important to say that this class considers only the first FE in an hp::FECollection that has non-zero number of unknowns (if I read the code correctly).
There was a problem hiding this comment.
I have added a comment. To lift this restriction is the next step; see #14121 (comment).
peterrum
force-pushed
the
birth_and_death_mf
branch
from
56856ae
to
1500599
Compare
|
@kronbichler I have incorporated your suggestion! Thanks. |
Add MatrixFreeTools::ElementBirthAndDeathMatrixFree
references #12102
depends on #14122