salvus.mesh.tools.transforms

def interpolate_mesh_to_mesh(
    mesh_0: salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh,
    mesh_1: salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh,
    use_layers: bool,
    use_1d_vertical_coordinate: bool,
    verbose: bool = True,
    exclude_filter: Optional[Tuple[str, int]] = None,
    fields_to_interpolate: Optional[str, List[str]] = None,
) -> (
    salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh
): ...

Interpolate element nodal material parameters from mesh_0 to mesh_1.

It is often desirable to interpolate material parameters from one mesh to another. This may be relevant, for example, when changing frequency bands in an ongoing full-waveform inversion, when performing a convergence analysis, when comparing the efficacy of different models, and when using advanced waveform inversion methodologies (i.e. Smoothiesem). In simplest case of a rectilinear -> rectilinear mesh the interpolation is relatively straightforward: interpolate the model stored on mesh_0 to mesh_1 by evaluating the model on mesh_0’s basis at the points defined by mesh_1.

For the case of deformed meshes, or models with discontinuities that must be preserved, the story gets a little more complicated. For example: what if the resolution of the new mesh’s topography is different than that of the old mesh? In this case, you may get aliasing around the discontinuities or along mesh edges. To remedy issues like this, this function implements several heuristic measures, which can be toggled by selecting the corresponding function arguments (see their descriptions).

As a general rule: any required extrapolations will be performed via a “clamping” approach, where the reference coordinates for a candidate model source element with be clamped to the reference interval [-1, +1]. This helps mitigate any “blow-ups” that could occur by evaluating the model basis too far outside of a given element. In the end, this means that the interpolation performed by this function may not be well defined in some cases. However, we feel that the measures chosen mitigate the most substantial issues, and should be sufficient for most use cases.

If any difficulties are encountered when using this function, please feel free to seek advice in Mondaic’s support forum.

def uniformly_refine(
    mesh: salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh,
    tensor_order: int,
    refinement_order: int = 2,
    change_tensor_order_kwargs: Dict = {},
    use_high_order_node_locations: bool = True,
) -> (
    salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh
): ...

Refine an unstructured mesh by subdividing all elements uniformly.

This function works by splitting each element evenly into 4 elements in 2-D, and 8 elements in 3-D, resulting in a new mesh with exactly 4 or 8 times as many elements as were present before. Model re-interpolation (of element-nodal fields) from and to arbitrary tensor_orders are supported. Elemental fields and global variables are copied over as well.

A common and practical use case for this function is to increase the frequency band of an ongoing inversion.

def uniformly_refine_chunkwise(
    mesh: salvus.mesh.data_structures.unstructured_mesh.unstructured_mesh.UnstructuredMesh,
    output_filename: str,
    nchunk: int = 10,
    show_progress: bool = True,
    uniformly_refine_kwargs: Dict = {},
) -> None: ...

Refine an unstructured mesh by subdividing all elements uniformly, working through the mesh in chunks to avoid a memory bottleneck. The mesh is merged and directly written to a file.

See docstring of uniformly_refine for more details