An output / interpolation problem in parallel

I/O
1

Following code output the gradient of a function.

from pathlib import Path

from mpi4py import MPI
import numpy as np

from dolfinx.io import XDMFFile
from dolfinx.fem import Function, Expression, functionspace
from dolfinx.mesh import create_rectangle

from ufl import grad

msh = create_rectangle(MPI.COMM_WORLD, ((-1, -1), (1, 1)), [100, 100])
space = functionspace(msh, ('Lagrange', 1))

u = Function(space)
u1 = Function(space)

u.interpolate(lambda x: np.tanh(5 * x[0]))
u.x.scatter_forward()
u1.interpolate(Expression(grad(u)[0], space.element.interpolation_points))
u1.x.scatter_forward()

folder = Path("test")

with XDMFFile(MPI.COMM_WORLD, folder / 'u1.xdmf', 'w') as file:
    file.write_mesh(msh)
    file.write_function(u1)

The output is correct in single process,

CorrectOutput
CorrectOutput581×581 3.23 KB

while there’s some things like cracks in the output when I run the code with parallel.

Screenshot from 2025-12-21 18-31-10
Screenshot from 2025-12-21 18-31-10709×704 9.01 KB

It seems that this problem occurs when the degree of the expression and the function(u1) does not match. Can anyone tell me what’s wrong with it?

Besides, I want to know when should I use the function u.x.scatter_forward()?

Dolfinx version: 0.10.0 installed with docker

2

The gradient of `u’ is not in the space you are interpolating into,

Thus this will introduce a dependence on the order of elements traversed. If your u is a P-1 function, the gradient is a DG-0 function.

Scatter forward is needed whenever thou modify the owns part of the DOF vector, but not all ghost entries. For standard interpolation, this is not needed. However, for interpolation onto a subset of cells, it should be called to ensure consistency