Problem with adaptive refinement in the circle domain

1

Hi all, I want to use adaptive refinement in a circle domain. As the boundary cells be refined I expect the new nodes to be added on the boundary but by using refine(mesh, cell_markers), they be added on the polygonal domain.
Is there a way to resolve this problem in refine?
Thank you in advance.

2

There is functionality to “snap” new nodes to a SubDomain. It should be demonstrated by this demo

https://bitbucket.org/fenics-project/dolfin/src/master/python/demo/undocumented/smoothing/demo_smoothing.py

although the functionality is not properly accessible via Python in 2019.1. I fixed this in the development version (pull request 521), but you can also implement it yourself through the Python interface, by defining a function like

def py_snap_boundary(mesh, sub_domain):
    boundary = BoundaryMesh(mesh,"exterior")
    dim = mesh.geometry().dim()
    x = boundary.coordinates()
    for i in range(0,boundary.num_vertices()):
        sub_domain.snap(x[i,:])
    ALE.move(mesh,boundary)
1 Like
3

Many thanks for your answer. I would be grateful if you could give a more explanation about considering sub_domain in py_snap_boundary(mesh, sub_domain).
In fact, I have already used refine(mesh, cell_markers), so I don’t know how I should consider sub_domain here.

4

Take a look at the demo I linked. To get the correct behavior in that demo, you would add the definition of py_snap_boundary from my previous post and replace both instances of

mesh.snap_boundary(hole)

with

py_snap_boundary(mesh,hole)
6

I imagine it should be fine to just use a SubDomain corresponding to the entire circular boundary for purposes of mesh snapping (i.e., to be used analogously to hole from the demo). On parts of the boundary where the mesh was not refined, there just won’t be any new nodes to move, so the snapping won’t do anything.

1 Like
7

Thank you very much, it worked well.