Implement quadrilateral sequencing on the extended deep embedding

Adds extended_deep_embedding (subdividing the outer face with an outer-cap
vertex), quadrilateral_decomposition (pairing faces across level edges),
and quadrilateral_sequencing which runs the anchor drop / level add /
join / ring completion precedence with bottommost-on-the-canonical-
boundary-walk tiebreaks and a lex-smallest move-code-string choice for
the initial quad.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
2026-05-15 02:44:48 -04:00
parent 83914a6a20
commit dbb1cbcfe5
+419
View File
@@ -12,6 +12,16 @@ class DeeplyEmbeddedGraph(TypedDict):
deep_embedding: Graph
class QuadrilateralSequence(TypedDict):
deep_embedding: Graph
triangular_faces: list[frozenset[Any]]
depth_labelling: dict[Any, int]
outer_cap_vertex: Any
quadrilaterals: list[frozenset[frozenset[Any]]]
sequence: list[frozenset[frozenset[Any]]]
move_codes: list[int]
def get_plane_depth_labelling(g: Graph, outer_cycle: list[Any]) -> dict[Any, int]:
"""Return the plane depth of each vertex relative to the given outer cycle."""
# equivalent to the commented out naive implementation:
@@ -55,6 +65,411 @@ def deep_embedding(g: Graph, outer_cycle: list[Any], plane_depth_labelling: dict
return g_prime
def _triangle_type(face: frozenset[Any], depth_labelling: dict[Any, int]) -> str:
"""Return 'up' or 'down' for a triangular face (up: {d, d+1, d+1}; down: {d, d, d+1})."""
depths = sorted(depth_labelling[v] for v in face)
if depths[0] == depths[1]:
return 'down'
return 'up'
def _level_edge_of_face(face: frozenset[Any], depth_labelling: dict[Any, int]) -> frozenset[Any]:
"""Return the unique level edge of an up/down triangular face."""
vs = list(face)
for i in range(3):
for j in range(i + 1, 3):
if depth_labelling[vs[i]] == depth_labelling[vs[j]]:
return frozenset([vs[i], vs[j]])
raise ValueError(f"Face {set(face)} has no level edge (not up or down)")
def _quad_vertices(quad: frozenset[frozenset[Any]]) -> frozenset[Any]:
"""Return the 4 vertices of a quadrilateral."""
f1, f2 = list(quad)
return f1 | f2
def _quad_perimeter_edges(
quad: frozenset[frozenset[Any]],
depth_labelling: dict[Any, int],
) -> list[frozenset[Any]]:
"""Return the 4 perimeter edges (non-level) of a quadrilateral."""
f1 = next(iter(quad))
level_edge = _level_edge_of_face(f1, depth_labelling)
edges: list[frozenset[Any]] = []
for f in quad:
vs = list(f)
for i in range(3):
edge = frozenset([vs[i], vs[(i + 1) % 3]])
if edge != level_edge:
edges.append(edge)
return edges
def _quad_type(quad: frozenset[frozenset[Any]], depth_labelling: dict[Any, int]) -> str:
"""Return 'shallow_diamond', 'deep_diamond', or 's_quad' for a quadrilateral."""
types = tuple(sorted(_triangle_type(f, depth_labelling) for f in quad))
if types == ('up', 'up'):
return 'shallow_diamond'
if types == ('down', 'down'):
return 'deep_diamond'
return 's_quad'
def extended_deep_embedding(
g: Graph,
outer_cycle: list[Any],
plane_depth_labelling: dict[Any, int] | None = None,
) -> tuple[Graph, list[frozenset[Any]], dict[Any, int], Any, dict[Any, list[Any]]]:
"""
Return the extended deep embedding of g (including the outer face).
Subdivides every neutral triangular face --- including the outer face,
which is always neutral for a maximal planar graph --- by adding a new
interior vertex. The vertex added inside the outer face is the
outer-cap vertex x*.
Returns (g_prime, faces, depth_labelling, outer_cap_vertex, embedding) where:
- g_prime: the deep embedding graph G'
- faces: list of triangular faces of G' (each a frozenset of 3 vertices),
derived from Sage's planar embedding of g_prime
- depth_labelling: depth of every vertex in G' (extends the original)
- outer_cap_vertex: the new vertex placed inside the outer face
- embedding: the CCW rotation system of g_prime (vertex -> ordered
neighbors) as computed by Sage's planarity routine
"""
if plane_depth_labelling is None:
plane_depth_labelling = get_plane_depth_labelling(g, outer_cycle)
outer_vertices = frozenset(outer_cycle)
embedding_g = g.get_embedding()
if embedding_g is None:
g.is_planar(set_embedding=True)
embedding_g = g.get_embedding()
g_prime = g.copy()
next_vertex = max(g.vertices()) + 1
depth_labelling = dict(plane_depth_labelling)
outer_cap_vertex: Any = None
for face in g.faces(embedding_g):
face_vertices = [u for u, v in face]
a, b, c = face_vertices
if depth_labelling[a] == depth_labelling[b] == depth_labelling[c]:
x = next_vertex
next_vertex += 1
g_prime.add_vertex(x)
g_prime.add_edges([(x, a), (x, b), (x, c)])
depth_labelling[x] = depth_labelling[a] + 1
if frozenset(face_vertices) == outer_vertices:
outer_cap_vertex = x
g_prime.is_planar(set_embedding=True)
embedding = g_prime.get_embedding()
assert embedding is not None, "g_prime must be planar after construction"
faces = [frozenset([u for u, v in face]) for face in g_prime.faces(embedding)]
return g_prime, faces, depth_labelling, outer_cap_vertex, embedding
def quadrilateral_decomposition(
faces: list[frozenset[Any]],
depth_labelling: dict[Any, int],
) -> tuple[list[frozenset[frozenset[Any]]], dict[frozenset[Any], frozenset[frozenset[Any]]]]:
"""
Pair each face with the face on the other side of its level edge.
Returns (quads, quad_of_face) where:
- quads: list of {F1, F2} pairs (each a quadrilateral of the decomposition)
- quad_of_face: dict mapping each face to its containing quad
"""
faces_by_edge: dict[frozenset[Any], list[frozenset[Any]]] = {}
for face in faces:
vs = list(face)
for i in range(3):
edge = frozenset([vs[i], vs[(i + 1) % 3]])
faces_by_edge.setdefault(edge, []).append(face)
quads: list[frozenset[frozenset[Any]]] = []
quad_of_face: dict[frozenset[Any], frozenset[frozenset[Any]]] = {}
seen: set[frozenset[Any]] = set()
for face in faces:
if face in seen:
continue
level_edge = _level_edge_of_face(face, depth_labelling)
adjacent = faces_by_edge[level_edge]
assert len(adjacent) == 2, f"Level edge {set(level_edge)} has {len(adjacent)} adjacent faces"
other = adjacent[0] if adjacent[1] == face else adjacent[1]
quad = frozenset([face, other])
quads.append(quad)
quad_of_face[face] = quad
quad_of_face[other] = quad
seen.add(face)
seen.add(other)
return quads, quad_of_face
def _boundary_edges(
slice_faces: set[frozenset[Any]],
faces_by_edge: dict[frozenset[Any], list[frozenset[Any]]],
) -> set[frozenset[Any]]:
"""Return the set of edges that lie on the boundary of the slice."""
boundary: set[frozenset[Any]] = set()
for edge, fs in faces_by_edge.items():
inside = sum(1 for f in fs if f in slice_faces)
if inside == 1:
boundary.add(edge)
return boundary
def _face_of_wedge(
embedding: dict[Any, list[Any]],
) -> dict[tuple[Any, Any, Any], frozenset[Any]]:
"""For each (v, n_a, n_b) where n_a, n_b are CCW-consecutive in v's rotation,
return the triangular face at that wedge."""
fow: dict[tuple[Any, Any, Any], frozenset[Any]] = {}
for v, rotation in embedding.items():
k = len(rotation)
for i in range(k):
n_a = rotation[i]
n_b = rotation[(i + 1) % k]
fow[(v, n_a, n_b)] = frozenset({v, n_a, n_b})
return fow
def _find_starting_boundary_edge(
slice_faces: set[frozenset[Any]],
embedding: dict[Any, list[Any]],
face_of_wedge: dict[tuple[Any, Any, Any], frozenset[Any]],
) -> tuple[Any, Any]:
"""Find a directed boundary edge (v, n_b) with the slice on its LEFT."""
for v, rotation in embedding.items():
k = len(rotation)
for i in range(k):
n_a = rotation[i]
n_b = rotation[(i + 1) % k]
n_c = rotation[(i + 2) % k]
left_wedge = face_of_wedge[(v, n_a, n_b)]
right_wedge = face_of_wedge[(v, n_b, n_c)]
if left_wedge in slice_faces and right_wedge not in slice_faces:
return (v, n_b)
raise RuntimeError("No boundary edge found for slice")
def _boundary_walk(
slice_faces: set[frozenset[Any]],
embedding: dict[Any, list[Any]],
face_of_wedge: dict[tuple[Any, Any, Any], frozenset[Any]],
) -> list[tuple[Any, Any]]:
"""Trace the slice's boundary CCW with the slice on the LEFT.
Returns an ordered list of directed edges forming the closed boundary walk.
"""
start = _find_starting_boundary_edge(slice_faces, embedding, face_of_wedge)
walk: list[tuple[Any, Any]] = [start]
curr_u, curr_v = start
while True:
rotation = embedding[curr_v]
k = len(rotation)
i = rotation.index(curr_u)
next_v = None
for j in range(1, k + 1):
n_jm1 = rotation[(i + j - 1) % k]
n_j = rotation[(i + j) % k]
if face_of_wedge[(curr_v, n_jm1, n_j)] not in slice_faces:
next_v = n_jm1
break
if next_v is None:
raise RuntimeError(f"All wedges at {curr_v} are in slice; no boundary edge")
next_edge = (curr_v, next_v)
if next_edge == start:
break
walk.append(next_edge)
curr_u, curr_v = next_edge
return walk
def _canonicalize_walk(
walk: list[tuple[Any, Any]],
depth_labelling: dict[Any, int],
) -> list[tuple[Any, Any]]:
"""Rotate walk to start at the (smallest depth, smallest vertex id) position.
This fixes the canonical 'top' of the slice for the top-to-bottom scan.
"""
starts = [(depth_labelling[u], u) for u, _ in walk]
canon_idx = min(range(len(walk)), key=lambda i: starts[i])
return walk[canon_idx:] + walk[:canon_idx]
def _attachment_position(
quad: frozenset[frozenset[Any]],
walk: list[tuple[Any, Any]],
depth_labelling: dict[Any, int],
) -> int:
"""Return the latest index in walk at which a perimeter edge of quad appears.
Realizes 'bottommost attachment on the right boundary scanned top-to-bottom'.
Returns -1 if the quad has no perimeter edge on the boundary walk.
"""
perimeter = {frozenset(e) for e in _quad_perimeter_edges(quad, depth_labelling)}
latest = -1
for i, (u, v) in enumerate(walk):
if frozenset([u, v]) in perimeter:
latest = i
return latest
def _boundary_deep_diamonds(
quads: list[frozenset[frozenset[Any]]],
outer_cycle: list[Any],
outer_cap_vertex: Any,
) -> list[frozenset[frozenset[Any]]]:
"""Return the three boundary deep diamonds (each spans an edge of C)."""
outer_set = set(outer_cycle)
diamonds: list[frozenset[frozenset[Any]]] = []
for q in quads:
vs = _quad_vertices(q)
if outer_cap_vertex in vs and len(vs & outer_set) == 2:
diamonds.append(q)
assert len(diamonds) == 3, f"Expected 3 boundary deep diamonds, got {len(diamonds)}"
return diamonds
def _run_sequence(
initial_quad: frozenset[frozenset[Any]],
quads: list[frozenset[frozenset[Any]]],
depth_labelling: dict[Any, int],
faces_by_edge: dict[frozenset[Any], list[frozenset[Any]]],
embedding: dict[Any, list[Any]],
face_of_wedge: dict[tuple[Any, Any, Any], frozenset[Any]],
) -> tuple[list[frozenset[frozenset[Any]]], list[int]]:
"""Run the sequencing loop from initial_quad and return (sequence, move_codes)."""
sequence: list[frozenset[frozenset[Any]]] = [initial_quad]
move_codes: list[int] = []
slice_quads: set[frozenset[frozenset[Any]]] = {initial_quad}
slice_faces: set[frozenset[Any]] = set(initial_quad)
while len(slice_quads) < len(quads):
slice_v: set[Any] = set().union(*slice_faces)
boundary = _boundary_edges(slice_faces, faces_by_edge)
walk = _canonicalize_walk(
_boundary_walk(slice_faces, embedding, face_of_wedge),
depth_labelling,
)
anchor_drop: list[frozenset[frozenset[Any]]] = []
level_add: list[frozenset[frozenset[Any]]] = []
join: list[frozenset[frozenset[Any]]] = []
ring_completion: list[frozenset[frozenset[Any]]] = []
for q in quads:
if q in slice_quads:
continue
vs = _quad_vertices(q)
k = len(vs & slice_v)
edges = _quad_perimeter_edges(q, depth_labelling)
j = sum(1 for e in edges if e in boundary)
qt = _quad_type(q, depth_labelling)
if qt == 's_quad' and k == 2 and j == 1:
anchor_drop.append(q)
if k == 3 and j == 2:
level_add.append(q)
if qt == 'deep_diamond' and k == 2 and j == 1:
join.append(q)
if k == 4:
ring_completion.append(q)
def pick(cands: list[frozenset[frozenset[Any]]]) -> frozenset[frozenset[Any]]:
return max(cands, key=lambda q: _attachment_position(q, walk, depth_labelling))
if anchor_drop:
next_quad = pick(anchor_drop)
next_code = 0
elif level_add:
next_quad = pick(level_add)
next_code = 1
elif join:
next_quad = pick(join)
next_code = 2
elif ring_completion:
next_quad = pick(ring_completion)
next_code = 3
else:
raise RuntimeError(
f"No applicable move at step {len(sequence)}; slice has {len(slice_quads)}/{len(quads)} quads"
)
sequence.append(next_quad)
move_codes.append(next_code)
slice_quads.add(next_quad)
slice_faces.update(next_quad)
return sequence, move_codes
def quadrilateral_sequencing(
g: Graph,
outer_cycle: list[Any],
plane_depth_labelling: dict[Any, int] | None = None,
) -> QuadrilateralSequence:
"""
Build the quadrilateral sequence of g relative to outer_cycle.
Constructs the extended deep embedding G' (including the outer-cap vertex
x* in the outer face), decomposes G' into quadrilaterals via level-edge
pairing, and produces the deterministic sequence Q_1, ..., Q_N by
repeatedly applying the move-selection rule:
anchor drop (0) > level add (1) > join (2) > ring completion (3).
Within each move's candidate list, the bottommost attachment on the
canonical boundary walk is selected (the largest index in the walk where
a perimeter edge of the candidate appears). Among the three boundary deep
diamonds, Q_1 is the start that produces the lexicographically smallest
move-code string.
Simplification: anchor drop / join detection uses (k, j) = (vertices in
slice, perimeter edges on slice boundary) only; the orientation-specific
'left edge = right edge' clause is not separately enforced.
"""
g_prime, faces, depth_labelling, outer_cap_vertex, embedding = extended_deep_embedding(
g, outer_cycle, plane_depth_labelling
)
quads, _ = quadrilateral_decomposition(faces, depth_labelling)
faces_by_edge: dict[frozenset[Any], list[frozenset[Any]]] = {}
for face in faces:
vs = list(face)
for i in range(3):
edge = frozenset([vs[i], vs[(i + 1) % 3]])
faces_by_edge.setdefault(edge, []).append(face)
face_of_wedge = _face_of_wedge(embedding)
candidates = _boundary_deep_diamonds(quads, outer_cycle, outer_cap_vertex)
best_sequence: list[frozenset[frozenset[Any]]] | None = None
best_codes: list[int] | None = None
for q1 in candidates:
seq, codes = _run_sequence(q1, quads, depth_labelling, faces_by_edge, embedding, face_of_wedge)
if best_codes is None or codes < best_codes:
best_sequence = seq
best_codes = codes
assert best_sequence is not None and best_codes is not None
return QuadrilateralSequence(
deep_embedding=g_prime,
triangular_faces=faces,
depth_labelling=depth_labelling,
outer_cap_vertex=outer_cap_vertex,
quadrilaterals=quads,
sequence=best_sequence,
move_codes=best_codes,
)
def generate_example(n: int) -> DeeplyEmbeddedGraph:
"""Generate a random maximal planar graph of size n and return the triangulation, outer cycle, and deep embedding."""
g = graphs.RandomTriangulation(n)
@@ -68,6 +483,10 @@ def generate_example(n: int) -> DeeplyEmbeddedGraph:
if __name__ == "__main__":
example = generate_example(10)
result = quadrilateral_sequencing(example['graph'], example['outer_cycle'])
canonical, graph_dir = canonize_and_save_graph(example['graph'])
(graph_dir / "plane_depth_sequence").mkdir(parents=True, exist_ok=True)
print(canonical)
print(f"Number of quadrilaterals: {len(result['quadrilaterals'])}")
print(f"Sequence length: {len(result['sequence'])}")
print(f"Move codes: {result['move_codes']}")