Input for mesh refinement with refine_plaza

mesh
,
1

Dear FEniCS community.

I’m trying to use the refine_plaza function in order to consider a criterium for refinement.

I have followed Mesh refinement in dolfinx - #2 by dokken and considered the following MWE.

import dolfinx
from mpi4py import MPI
import numpy as np


def cell_criterion(x):
    """Given mesh coordinates, return if each point 
    satisfies x[0]<0.3 or x[1]>0.4

    :param x: Input coordinates, shape (num_points, 3)
    :returns: Boolean array of shape (num_points, )
    """
    return (x[0] < 0.3) | (x[1] > 0.4)


mesh = dolfinx.mesh.create_unit_square(MPI.COMM_WORLD, 4, 4)

# Compute midpoints for all cells on process
cells_local = np.arange(mesh.topology.index_map(
    mesh.topology.dim).size_local, dtype=np.int32)
midpoints = dolfinx.mesh.compute_midpoints(
    mesh, mesh.topology.dim, cells_local).T

# Check midpoint criterion and find edges connected to cells
should_refine = np.flatnonzero(cell_criterion(midpoints)).astype(np.int32)
mesh.topology.create_entities(1)
local_edges = dolfinx.mesh.compute_incident_entities(
    mesh.topology, should_refine, mesh.topology.dim, 1)
fine_mesh, parent_cell, parent_facet = dolfinx.mesh.refine_plaza(
    mesh, local_edges, False, dolfinx.mesh.RefinementOption.parent_cell_and_facet)

But I get the error on the refine_plaza function

TypeError: refine_plaza(): incompatible function arguments. The following argument types are supported:
    1. (mesh: dolfinx.cpp.mesh.Mesh_float32, redistribute: bool, option: dolfinx.cpp.refinement.RefinementOption) -> Tuple[dolfinx.cpp.mesh.Mesh_float32, numpy.ndarray[numpy.int32], numpy.ndarray[numpy.int8]]
    2. (mesh: dolfinx.cpp.mesh.Mesh_float64, redistribute: bool, option: dolfinx.cpp.refinement.RefinementOption) -> Tuple[dolfinx.cpp.mesh.Mesh_float64, numpy.ndarray[numpy.int32], numpy.ndarray[numpy.int8]]
    3. (mesh: dolfinx.cpp.mesh.Mesh_float32, edges: numpy.ndarray[numpy.int32], redistribute: bool, option: dolfinx.cpp.refinement.RefinementOption) -> Tuple[dolfinx.cpp.mesh.Mesh_float32, numpy.ndarray[numpy.int32], numpy.ndarray[numpy.int8]]
    4. (mesh: dolfinx.cpp.mesh.Mesh_float64, edges: numpy.ndarray[numpy.int32], redistribute: bool, option: dolfinx.cpp.refinement.RefinementOption) -> Tuple[dolfinx.cpp.mesh.Mesh_float64, numpy.ndarray[numpy.int32], numpy.ndarray[numpy.int8]]

Invoked with: <dolfinx.cpp.mesh.Mesh_float64 object at 0x7fa3de0b5df0>, array([ 8, 10, 13, 14, 15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,
       30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46,
       47, 48, 49, 50, 51, 52, 53, 54, 55], dtype=int32), False

If I remove local_edges the code runs but using uniform refinements.

Any help is appreciated.

1 Like
2

I don’t have the scope to run your code, but perhaps the following could help elucidate the arguments to the function: leopart-x/demo/demo_rayleigh-taylor-instability/demo_adaptive-mesh.py at main · LEoPart-project/leopart-x · GitHub

Quickly scanning your code vs mine, your issue could be as simple as needing to pass mesh._cpp_object rather than the python wrapped mesh.

3

Thanks Nate.

I tried your suggestion and got the error:

File "/usr/local/dolfinx-real/lib/python3.10/dist-packages/dolfinx/mesh.py", line 322, in refine_plaza
    mesh1, cells, facets = _cpp.refinement.refine_plaza(mesh._cpp_object, edges, redistribute)
AttributeError: 'dolfinx.cpp.mesh.Mesh_float64' object has no attribute '_cpp_object'

BTW I’m using the dolfinx docker image v.0.7.3

4

Try manually applying the fix that you see in Fix input arguments to dolfinx.mesh.refine_plaza by francesco-ballarin · Pull Request #3018 · FEniCS/dolfinx · GitHub to your docker image by editing python/dolfinx/mesh.py in your installation. It seems to me that there is a mistake in the wrappers of the C++ interface.

1 Like
5

Thanks Francesco.

Manually applying the fix solves the problem.

6

Same error message encountered with the calling of refined_mesh = mesh.refine(old_mesh, edges, redistribute=False). Guess the same issue with it. Not sure how to correct it with the suggested pull request. Any help?

7

Please provide what version of DOLFINx you are running. Can you make a 5 line example (trying to refine a unit square) that reproduces the issue?

8

DOLFINx version 0.7.3

I found that is due to I messed up dolfinx.mesh.Mesh and dolfinx.cpp.mesh.Mesh_float64, where I should use dolfinx.mesh.Mesh mesh1 to do mesh.refine rather than dolfinx.cpp.mesh.Mesh_float64 mesh2. In my implementation, I load once a coarse background mesh mesh0 and then copy it to mesh1 as active mesh for refinement. But mesh0 is changed accordingly as mesh1 changes.

MWE:

from mpi4py import MPI
from dolfinx import mesh, cpp
import numpy as np
mesh0 = mesh.create_unit_square(MPI.COMM_WORLD, 2, 2)                      # 'dolfinx.mesh.Mesh'
mesh1 = mesh.Mesh(mesh0._cpp_object, mesh0.ufl_domain())                   # 'dolfinx.mesh.Mesh'
mesh2 = cpp.mesh.Mesh_float64(mesh0.comm, mesh0.topology, mesh0.geometry)  # 'dolfinx.cpp.mesh.Mesh_float64'
mesh1.geometry.x[:, :] *= 2                      # change mesh1
print("mesh0:\n", mesh0.geometry.x, "\n")        # but mesh0 is also changed
print("mesh1:\n", mesh1.geometry.x, "\n")
print("mesh2:\n", mesh2.geometry.x, "\n")

mesh1.topology.create_entities(1)
new_mesh1 = mesh.refine(mesh1, np.array([1,2,3], dtype=np.int32), redistribute=False)
# mesh2.topology.create_entities(1)
# new_mesh2 = mesh.refine(mesh2, np.array([1,2,3], dtype=np.int32))
# AttributeError: 'dolfinx.cpp.mesh.Mesh_float64' object has no attribute '_cpp_object'
9

So is the issue resolved?
Are you trying to copy a mesh?

10

Yes, resolved. It leads to copy mesh question. What is difference between mesh1 and mesh2? And I thought mesh.Mesh is a copy constructor.

11

The mesh takes in pointers: dolfinx/python/dolfinx/wrappers/mesh.cpp at main · FEniCS/dolfinx · GitHub