Graph

plot3d.graph.build_weighted_graph_from_face_matches(face_matches: List[dict], n_blocks: int, aggregate: str = 'sum', ignore_self_matches: bool = True) → Tuple[Dict[int, List[int]], Dict[int, Dict[int, int]]]

Convert connectivity_fast face_matches into adjacency + weights.

Note

Aggregate controls how we combine weights when the same two blocks are connected by multiple faces.

Why this matters:

connectivity_fast can return more than one face between the same pair of blocks (e.g. a block is split and has two non-contiguous interfaces with its neighbor). Each face gives a weight = dI * dJ * dK (number of shared nodes). METIS expects one edge per block-pair, with a single weight. So if there are multiple faces, we must decide how to merge them. That’s what aggregate does.

Parameters:

• face_matches (List[dict]) – Output from connectivity_fast

• n_blocks (int) – number of blocks in a mesh

• aggregate (str, optional) – ‘sum’|’max’|’min’. Controls how weights are combined. Defaults to “sum”.

• ignore_self_matches (bool, optional) – ignores self matching (i==j). Defaults to True.

Raises:

ValueError – if aggregate is not one of ‘sum’,’max’,’min’

Returns:

• adj_list (Dict[int, List[int]]): Neighbors for each block.

• edge_w (Dict[int, Dict[int, int]]): Edge weights (u->v).

Return type:

Tuple[Dict[int, List[int]], Dict[int, Dict[int, int]]]

plot3d.graph.csr_from_adj_and_weights(adj_list: Dict[int, List[int]], edge_w: Dict[int, Dict[int, int]]) → Tuple[List[int], List[int], List[int]]

Build CSR arrays (xadj, adjncy, eweights) from adjacency + weights.

Returns:

prefix sum of neighbors adjncy (List[int]): flattened neighbor list eweights (List[int]): edge weights aligned with adjncy

Return type:

xadj (List[int])

plot3d.graph.partition_from_face_matches(face_matches: List[dict], blocks: Sequence[Block], nparts: int, favor_blocksize: bool = True, aggregate: str = 'sum', ignore_self_matches: bool = True) → Tuple[List[int], Dict[int, List[int]], Dict[int, Dict[int, int]]]

Partition a graph derived from face_matches using pymetis.

Note

Aggregate controls how we combine weights when the same two blocks are connected by multiple faces.

Why this matters:

connectivity_fast can return more than one face between the same pair of blocks (e.g. a block is split and has two non-contiguous interfaces with its neighbor). Each face gives a weight = dI * dJ * dK (number of shared nodes). METIS expects one edge per block-pair, with a single weight. So if there are multiple faces, we must decide how to merge them. That’s what aggregate does.

Returns:

• parts (List[int]) – Partition id (0-based) for each block.

• adj_list (Dict[int, List[int]]) – Adjacency list used for the partitioning.

• edge_w (Dict[int, Dict[int, int]]) – Edge weights.

plot3d.graph.write_ddcmp(parts: Sequence[int], blocks: Sequence[Block], adj_list: Dict[int, List[int]], edge_weights: Dict[int, Dict[int, int]] | None = None, filename: str = 'ddcmp.dat') → None

Writes ddcmp.dat and ddcmp_info.txt.

Notes

• parts are 0-based in memory, but written 1-based in the file (to match your C#).

• edge_weights affects the per-partition ‘edge_work’ if provided.