Introduce metric parameters documentation #99
Conversation
| * `hausdorff_number`: Spatial scale factor for the problem. Default: 0.01. | ||
| * `boundary_tag`: Mesh boundary tag to restrict attention to during | ||
| boundary-specific manipulations. Unset by default, which implies all | ||
| boundaries are considered. (Note that this parameter does not currently exist |
There was a problem hiding this comment.
@jwallwork23 It might be good to add how this parameter currently interacts with a similar flag which can be set in a mesh originating from a .gmsh, as Firedrake does seem to preserve that flag when communicating between PETSc and MMG. Is this a different flag altogether?
There was a problem hiding this comment.
This is a different flag. Small clarification in 7b4ec08. This flag is only currently used by the private method _enforce_variable_constraints, which is called by the public method enforce_spd.
animate/metric.py
Outdated
| * `num_iterations`: Number of adaptation iterations in the parallel case. | ||
| Default: 3. |
There was a problem hiding this comment.
The description leaves open the question of what happens in the non-parallel case. Is there only one adaptation iteration to the metric in the non-parallel case?
jwallwork23
changed the title
| if self._plex.metricRestrictAnisotropyFirst(): | ||
| raise NotImplementedError( | ||
| "Restricting metric anisotropy first is not supported in Firedrake." | ||
| ) |
There was a problem hiding this comment.
This should be easy enough to implement (just tweak the arguments to the enforce_spd call in the definition of normalise). However, I'd prefer to address it in a separate issue rather than here, considering the impact with some tests.
|
You happy with the changes @ddundo? |
Closes #98.Fixes the issue with deep-copying metric parameters by always deep-copying these parameters upon setting them. (I can't think of a reason why you wouldn't want this.) Also moves some mechanics behind the scenes and provides a public interface tometric_parameters.Edit: See discussion in the original issue.
Closes #31.
At long last!