Adaptive Multigrid #87
Conversation
| ) | ||
|
|
||
| for j, parent in enumerate(f2c): | ||
| n = int(num_children[parent].item()) |
There was a problem hiding this comment.
Is this just to avoid numpy types? What happens if you do?
| n = int(num_children[parent].item()) | |
| n = num_children[parent] |
There was a problem hiding this comment.
The following loop after this will return that n is an unhashable type: 'numpy.ndarray' , but you are right that there is no need for the int()
|
There is a demo.py.rst in your project repository. Could it be added to this PR to include it in ngsPETSc? (In particular I would like to comment on the interface.) |
| if mesh.topological_dimension() <= 2: | ||
| max_children = 4 | ||
| else: | ||
| max_children = 16 |
There was a problem hiding this comment.
Why is this 16 and not 8?
There was a problem hiding this comment.
Setting this < 16 leads to bugs in 3D. The coarse-to-fine mappings generated from ngmesh.parentelements.NumPy() sometimes returns more than 8 children.
There was a problem hiding this comment.
In some iterations I think we do. I somewhat empirically found this upper bound.
| for split_label, _ in source_function_splits.items(): | ||
| transfer_op( | ||
| source_function_splits[split_label], | ||
| target_function_splits[split_label], | ||
| ) | ||
|
|
There was a problem hiding this comment.
We should have a special case that uses assign for the unrefined submesh
…on't perform transfer for unsplit submeshes
|
@AnuragRao1 Why are the ngsolve test importing firedrake now ? :) |
Added
AdaptiveMeshHierarchyandAdaptiveTransferManageralong with corresponding tests.