dokken
November 21, 2024, 4:41pm
2
rsz7ncf24p:
Am I overlooking something or is this intended behavior? If so, why, and can I, in an inexpensive post-processing step, reorder the DOFs such that they are aligned with the mesh coordinates again?
Thank you.
It is intended behavior as:
the function is designed to work with arbitrary order function spaces and meshes, you cannot rely on this 1-1 correspondence.
It is supposed to work in parallel (on data partitioned with mpi), which introduces a whole other set of considerations
Why do you rely on this ordering being the same?
Here is a vectorized version of the code above (credit to @dokken )
def vertex_to_dof_map_original(mesh, V):
num_vertices = (
mesh.topology.index_map(0).size_local + mesh.topology.index_map(0).num_ghosts
)
vertex_to_dof_map = np.empty(num_vertices, dtype=np.int32)
num_cells = (
mesh.topology.index_map(mesh.topology.dim).size_local
+ mesh.topology.index_map(mesh.topology.dim).num_ghosts
)
c_to_v = mesh.topology.connectivity(mesh.topology.dim, 0)
…
I don’t have time to read that user manual.
Assuming that what is in NSET describes the original nodes in the mesh you would like to constrain, you need to:
Create a map from the input node index to a vertex on your process, this is simply a matter of inverting mesh.geometry.input_global_indices.
Once you have a map from the global indices (in your case a map from 97, 96, 95, ... to what is local on the process), you remove all the results that are -1.
You then map from geometry to topology …
(And to some extent in: Renumbered .msh imported data giving wrong subdomains - #4 by dokken )
I would start with mapping dofs in the submesh to the nodes of the geometry in the submesh, and then use the last output argument of ‘create_submesh` to map these geometry indices from the submesh to the parent mesh.https://docs.fenicsproject.org/dolfinx/v0.9.0/python/generated/dolfinx.mesh.html#dolfinx.mesh.create_submesh
1 Like