Errors when interpolating symmetric tensor element in mixed space

dolfinx
1

I am trying to extend a following code https://github.com/FEniCS/dolfinx/issues/3165 to include mixed space elements.

import basix.ufl
import numpy as np
from dolfinx.mesh import create_unit_square
from dolfinx.fem import functionspace, Function
from mpi4py import MPI

mesh = create_unit_square(MPI.COMM_WORLD, 10, 10)

def tensor(x):
    mat = np.array([[1], [0], [0], [2], [3], [0], [4], [5], [6]])
    return np.broadcast_to(mat, (9, x.shape[1]))
def tensor2(x):
    mat = np.array([[1], [2], [3],  [4], [5], [6]])
    return np.broadcast_to(mat, (6, x.shape[1]))

element = basix.ufl.element("DG", mesh.basix_cell(), 0, shape=(3, 3), symmetry=True)
space = functionspace(mesh, element)
space_mix = functionspace(mesh, basix.ufl.mixed_element([element,element]))
subspace,_ = space_mix.sub(0).collapse()
f = Function(subspace)
# f = Function(space)
# f.interpolate(lambda x: tensor2(x))
f.interpolate(lambda x: tensor(x))
print(f.eval([[0, 0, 0]], [0]))

  1. When using a standard space, function with full tensor (tensor(x)) works properly (prints [1. 2. 4. 2. 3. 5. 4. 5. 6.]), while the other approach with only symmetrical part (using tensor2) gives Interpolation data has the wrong shape/size. error.

  2. When using the mixed space, the first approach (tensor(x)) gives the Interpolation ... error, while the second (using tensor2(x)) sometimes ends up with Ubuntu (invalid pointer) or MPI (SEGV) memory errors. Regardless of errors print always executes printing incorrect result: [1.0000e+000 4.0000e+000 9.5750e-321 4.0000e+000 5.0000e+000 1.0563e-320]

  3. Both approaches (with standard and mixed space) work well when symmetry=False

Am I doing something wrong?