Fix mapping fe field similarity #9769
Conversation
|
The diff is best watched without whitespace, |
kronbichler
force-pushed
the
fix_mapping_fe_field_similarity
branch
from
87da40a
to
20de795
Compare
| MappingFEField when the underlying mesh had similarities, even though the | ||
| deformed cells did not. This would lead to wrong Jacobians and transformed | ||
| shape function gradients, among others, when running deal.II without threads. | ||
| This is now fixed. |
There was a problem hiding this comment.
Can you update that the same fix applies to MappingManifold (since you also do this in this patch)?
source/fe/mapping_fe_field.cc
Outdated
| } | ||
| for (unsigned int k = 0; k < data.n_shape_functions; ++k) | ||
| { | ||
| unsigned int comp_k = fe.system_to_component_index(k).first; |
There was a problem hiding this comment.
| unsigned int comp_k = fe.system_to_component_index(k).first; | |
| const unsigned int comp_k = fe.system_to_component_index(k).first; |
source/fe/mapping_fe_field.cc
Outdated
| for (unsigned int i = 0; i < spacedim; ++i) | ||
| for (unsigned int k = 0; k < data.n_shape_functions; ++k) | ||
| { | ||
| unsigned int comp_k = fe.system_to_component_index(k).first; |
There was a problem hiding this comment.
While you're there:
| unsigned int comp_k = fe.system_to_component_index(k).first; | |
| const unsigned int comp_k = fe.system_to_component_index(k).first; |
source/fe/mapping_fe_field.cc
Outdated
| for (unsigned int j = 0; j < spacedim; ++j) | ||
| for (unsigned int k = 0; k < data.n_shape_functions; ++k) | ||
| { | ||
| unsigned int comp_k = fe.system_to_component_index(k).first; |
There was a problem hiding this comment.
| unsigned int comp_k = fe.system_to_component_index(k).first; | |
| const unsigned int comp_k = fe.system_to_component_index(k).first; |
To be honest, I don't know a good way of how to construct a proper manifold that would trigger this. Let me make a short test with a spherical manifold in 2d. |
kronbichler
force-pushed
the
fix_mapping_fe_field_similarity
branch
from
20de795
to
6cc3b12
Compare
|
It turned out that it was not too complicated to get that test - just take a refined hyper cube and let the edges be deformed differently with |
We had the cell similarity detection in
MappingFEFieldenabled (in analogy toMappingQGeneric), even though we cannot know for an Euler vector whether there really is a similarity. Therefore, we must completely disable this concept in case of aMappingFEFieldand similarily also forMappingManifold.Fixes #9757