Netgen: curved meshes from Mesh constructor and refine_marked_elements by pbrubeck · Pull Request #4930 · firedrakeproject/firedrake

pbrubeck · 2026-02-27T09:50:15Z

Description

Previously we had 2 methods to construct a curved Mesh out of a netgen mesh:

First creating a firedrake Mesh from a linear netgen mesh, and then calling curve_field(degree) to produce new coordinates, then you could create a firedrake Mesh out of the curve field. The final Firedrake Mesh will not know that it comes from a netgen mesh, and therefore cannot be adaptively refined.

mesh1 = Mesh(netgen_mesh)
meshp = Mesh(mesh1.curve_field(p))

Via the MeshHierarchy constructor, this method produces a MeshHierarchy in which each mesh knows that it comes from a netgen mesh

mesh1 = Mesh(netgen_mesh)
mh = MeshHierarchy(mesh1, refine, netgen_flags={"degree": p})
meshp = mh[0]

This PR enables the more intuitive approach (while not removing the other two)

meshp = Mesh(netgen_mesh, netgen_flags={"degree": p})

Also propagates the mesh order in refine_marked_elements.

connorjward

This is a way better way to build these meshes. Could you amend our demos?

connorjward · 2026-03-18T12:22:05Z

firedrake/mesh.py

+            (ng_dimension, reference_space_points.shape[0], self.geometric_dimension)
+        )
+        self.netgen_mesh.CalcElementMapping(reference_space_points, physical_space_points)
+        # NOTE: This will segfault for CSG!


Concerning...

We should handle this properly somehow.

We could just raise an exception when self.netgen_mesh.getGeometry() is CSG, but what if this get fixed upstream?

I don't think that we should leave this in here. Instead we should:

Raise an error if a user tries to use CSG that points to the upstream issue

Open our own issue, again linking to the upstream one, telling us to revisit this at a later point

I agree we should raise an error if the geometry is CSG. I believe CSG is no longer developed by the NGSolve group in favour of OCC 2D and 3D, hence why it seg faults on new features like the one we use to convert back to a Netgen mesh the DMPlex after refinement.

connorjward · 2026-03-18T16:07:11Z

firedrake/mesh.py

+
+        netgen_mesh = self.netgen_mesh.Copy()
+        refine_faces = netgen_flags.get("refine_faces", False)
+        if self.comm.rank == 0:


Can this be wrapped in a try except block so we don't deadlock if this ends up failing. I did the same thing here: https://github.com/firedrakeproject/firedrake/pull/4913/changes#diff-3b7ad5733ba21ccaa1be5b8aa4027e50a552acea3d0c66425ab7c47bd31ee278

I am calling Barrier() below, and this seems to prevent deadlocks

Actually, the assumption that rank 0 keeps the global netgen mesh is not always true. A netgen MeshHierarchy will distribute the netgen mesh. I think we need a further PR where we make sure that the underlying netgen mesh is also distributed.

I am calling Barrier() below, and this seems to prevent deadlocks

That's not what I mean. If the code that is only running on rank 0 ever fails for some reason then things will hang in a really really hard to debug way (remember this?).

I've added a helper function for doing this in #4972 which is probably good to be merged later today.

I think we need a further PR where we make sure that the underlying netgen mesh is also distributed.

Sure. I agree that that might be out of scope here. It would be helpful if you could open an issue though.

Could you point me to the function?

I think I can remove the branching here... The netgen mesh on the other ranks should also be refined in both the redundant and the distributed cases

Could you point me to the function?

safe_noncollective in #4972

firedrake/mesh.py

firedrake/netgen.py

tests/firedrake/multigrid/test_netgen_gmg.py

firedrake/netgen.py

demos/netgen/netgen_mesh.py.rst

connorjward · 2026-03-19T10:41:16Z

firedrake/mesh.py

+
+        netgen_mesh = self.netgen_mesh.Copy()
+        refine_faces = netgen_flags.get("refine_faces", False)
+        if self.comm.rank == 0:


I am calling Barrier() below, and this seems to prevent deadlocks

That's not what I mean. If the code that is only running on rank 0 ever fails for some reason then things will hang in a really really hard to debug way (remember this?).

I've added a helper function for doing this in #4972 which is probably good to be merged later today.

I think we need a further PR where we make sure that the underlying netgen mesh is also distributed.

Sure. I agree that that might be out of scope here. It would be helpful if you could open an issue though.

connorjward · 2026-03-19T10:43:01Z

firedrake/mesh.py

+            (ng_dimension, reference_space_points.shape[0], self.geometric_dimension)
+        )
+        self.netgen_mesh.CalcElementMapping(reference_space_points, physical_space_points)
+        # NOTE: This will segfault for CSG!


I don't think that we should leave this in here. Instead we should:

Raise an error if a user tries to use CSG that points to the upstream issue

Open our own issue, again linking to the upstream one, telling us to revisit this at a later point

demos/netgen/netgen_mesh.py.rst

firedrake/mesh.py

tests/firedrake/multigrid/test_netgen_gmg.py

pbrubeck · 2026-03-20T12:05:40Z

firedrake/mesh.py

+            (ng_dimension, reference_points.shape[0], self.geometric_dimension)
+        )
+        self.netgen_mesh.CalcElementMapping(reference_points, physical_points)
+        # NOTE: This will segfault for MeshHierarchy on a netgen CSG geometry


Curving a CSG mesh works fine, as long as this mesh has not been uniformly refined. It fails for CSG + MeshHierarchy, where we convert mesh.topology_dm.refine() into a netgen mesh. So this seems to me like a ngsPETSc issue.

The new mechanics,which allow for 3D support, is as follows now. We create a refinement in PETSc, we construct the maps between coarse to fine and vice versa in PETSc which is faster (you really see it in 3D) and eventually I hope also load balanced, then we create the corresponding Netgen mesh. We then snap to the geometry and curve, we thus need to attach geometrical information to the mesh. This last feature is only supported for OCC and not for old school CSG. I believe this is the best way to go and I don't think we should insist in supporting the old CSG, it makes no sense since also Netgen is no longer actively developing it. There is the serious risk we will end up having to keep fixing issues arising from legacy code if we do so. Less is more in this case :)

firedrake/mesh.py

Netgen: curved meshes from Mesh constructor and refine_marked_elements

8c6f918

connorjward requested changes Feb 27, 2026

View reviewed changes

pbrubeck added 9 commits February 27, 2026 10:07

tests

db3396f

Merge branch 'main' into pbrubeck/netgen-bendy-mesh

9fa8a49

update

86a1af2

clean up

3488bc4

sanitise refine_marked_elements

bb668a5

attempt to fix netgen MeshHierarchy

74ea35d

fixes

c592b66

fix

574be15

WIP

f56305c

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch from 9805342 to f56305c Compare March 16, 2026 16:44

Fixes

e877b90

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch from f56305c to e877b90 Compare March 16, 2026 17:47

pbrubeck added 3 commits March 16, 2026 17:48

merge conflict

7be7a75

Fixes

d1b8c4c

tidy

8f404e1

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch 3 times, most recently from bded950 to 2916572 Compare March 18, 2026 11:53

tidy

9250727

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch from 2916572 to 9250727 Compare March 18, 2026 11:55

update demo

f1cd773

pbrubeck requested a review from connorjward March 18, 2026 12:10

Merge branch 'main' into pbrubeck/netgen-bendy-mesh

dcbffae

connorjward requested changes Mar 18, 2026

View reviewed changes

review comments

e58dd9c

pbrubeck commented Mar 18, 2026

View reviewed changes

firedrake/netgen.py Outdated Show resolved Hide resolved

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch 2 times, most recently from 113d7de to 7528fa0 Compare March 18, 2026 22:10

fix

546f11a

pbrubeck force-pushed the pbrubeck/netgen-bendy-mesh branch from 7528fa0 to 546f11a Compare March 18, 2026 22:13

pbrubeck commented Mar 18, 2026

View reviewed changes

demos/netgen/netgen_mesh.py.rst Outdated Show resolved Hide resolved

pbrubeck requested a review from connorjward March 19, 2026 08:49

connorjward requested changes Mar 19, 2026

View reviewed changes

Remove branching

8d917e7

pbrubeck commented Mar 19, 2026

View reviewed changes

demos/netgen/netgen_mesh.py.rst Outdated Show resolved Hide resolved

pbrubeck added 2 commits March 19, 2026 22:37

Update demos/netgen/netgen_mesh.py.rst

942ae51

Update demos/netgen/netgen_mesh.py.rst

d8fc40b

pbrubeck commented Mar 19, 2026

View reviewed changes

firedrake/mesh.py Outdated Show resolved Hide resolved

Update firedrake/mesh.py

2551b9d

pbrubeck commented Mar 19, 2026

View reviewed changes

tests/firedrake/multigrid/test_netgen_gmg.py Outdated Show resolved Hide resolved

pbrubeck added 2 commits March 19, 2026 23:17

Update tests/firedrake/multigrid/test_netgen_gmg.py

a6d3171

Fix

9a4dc3a

pbrubeck commented Mar 20, 2026

View reviewed changes

pbrubeck added 4 commits March 20, 2026 14:37

fix docs

b4dc822

fix

b82acba

fixes

2b116d5

Merge branch 'main' into pbrubeck/netgen-bendy-mesh

c44acf4

pbrubeck commented Mar 24, 2026

View reviewed changes

firedrake/mesh.py Outdated Show resolved Hide resolved

pbrubeck added 2 commits March 24, 2026 17:06

docs

ac9808b

DROP BEFORE MERGE

e5a06e6

Conversation

pbrubeck commented Feb 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Uh oh!

connorjward left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pbrubeck Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pbrubeck Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pbrubeck Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

UZerbinati Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

pbrubeck commented Feb 27, 2026 •

edited

Loading

pbrubeck Mar 18, 2026 •

edited

Loading

pbrubeck Mar 18, 2026 •

edited

Loading

pbrubeck Mar 20, 2026 •

edited

Loading

UZerbinati Mar 20, 2026 •

edited

Loading