Schnyder decompositions for regular plane graphs and application to drawing

Schnyder woods are decompositions of simple triangulations into three edge-disjoint spanning trees crossing each other in a specific way. In this article, we define a generalization of Schnyder woods to $d$-angulations (plane graphs with faces of degree $d$) for all $d\geq 3$. A \emph{Schnyder decomposition} is a set of $d$ spanning forests crossing each other in a specific way, and such that each internal edge is part of exactly $d-2$ of the spanning forests. We show that a Schnyder decomposition exists if and only if the girth of the $d$-angulation is $d$. As in the case of Schnyder woods ($d=3$), there are alternative formulations in terms of orientations (“fractional” orientations when $d\geq 5$) and in terms of corner-labellings. Moreover, the set of Schnyder decompositions on a fixed $d$-angulation of girth $d$ is a distributive lattice. We also show that the structures dual to Schnyder decompositions (on $d$-regular plane graphs of mincut $d$ rooted at a vertex $v^$) are decompositions into $d$ spanning trees rooted at $v^$ such that each edge not incident to $v^*$ is used in opposite directions by two trees. Additionally, for even values of $d$, we show that a subclass of Schnyder decompositions, which are called even, enjoy additional properties that yield a reduced formulation; in the case d=4, these correspond to well-studied structures on simple quadrangulations (2-orientations and partitions into 2 spanning trees). In the case d=4, the dual of even Schnyder decompositions yields (planar) orthogonal and straight-line drawing algorithms. For a 4-regular plane graph $G$ of mincut 4 with $n$ vertices plus a marked vertex $v$, the vertices of $G\backslash v$ are placed on a $(n-1) \times (n-1)$ grid according to a permutation pattern, and in the orthogonal drawing each of the $2n-2$ edges of $G\backslash v$ has exactly one bend. Embedding also the marked vertex $v$ is doable at the cost of two additional rows and columns and 8 additional bends for the 4 edges incident to $v$. We propose a further compaction step for the drawing algorithm and show that the obtained grid-size is strongly concentrated around $25n/32\times 25n/32$ for a uniformly random instance with $n$ vertices.

💡 Research Summary

The paper presents a comprehensive generalization of Schnyder woods—originally defined for simple triangulations—to arbitrary $d$‑angulations (plane graphs whose faces all have degree $d$) for any integer $d\ge3$. The authors introduce the notion of a Schnyder decomposition, which consists of $d$ spanning forests $\mathcal{F}_1,\dots,\mathcal{F}_d$ on a $d$‑angulation $G$. The decomposition must satisfy two key constraints: (i) every internal edge belongs to exactly $d-2$ of the forests, and (ii) the forests intersect in a prescribed cyclic order that mimics the crossing pattern of the three trees in the classical Schnyder wood.

A central theorem proves that such a decomposition exists if and only if the girth of $G$ equals $d$, i.e., the smallest cycle in $G$ is a $d$‑cycle. This condition is both necessary and sufficient, showing that the combinatorial structure of a $d$‑angulation must be “tight” in the sense that no shorter cycles are allowed.

The authors then give three equivalent formulations of the same object:

1. Fractional orientations – For $d\ge5$, each internal edge receives $d-2$ oriented copies, yielding a “fractional” out‑degree/in‑degree balance at each vertex. When $d=3$, this reduces to the classic integer orientation where each interior edge is oriented exactly once.

2. Corner labellings – Each corner of a face is labelled with a number from $1$ to $d$. The labelling must satisfy local consistency rules (the labels increase cyclically around each face) and global constraints that encode the forest structure.

3. Distributive lattice of decompositions – The set of all Schnyder decompositions on a fixed $d$‑angulation forms a distributive lattice under a natural “flip” operation that moves an edge from one forest to another while preserving the defining constraints. This lattice perspective yields a powerful combinatorial framework for enumeration, random sampling, and optimization.

When $d$ is even, the paper defines a subclass called even Schnyder decompositions. In this case the $d$ forests can be paired so that each pair uses every internal edge in opposite directions. For $d=4$, even decompositions coincide with well‑studied structures on quadrangulations: 2‑orientations (each interior vertex has out‑degree 2) and a partition of the edge set into two spanning trees. This specialization dramatically simplifies the representation and leads to more efficient algorithms.

The dual viewpoint is explored in depth. The dual of a Schnyder decomposition on a $d$‑angulation is a decomposition of a $d$‑regular plane graph (minimum cut $d$) rooted at a distinguished vertex $v^{}$ into $d$ spanning trees. Every edge not incident to $v^{}$ appears in exactly two trees, oriented oppositely. This dual structure is the key to the drawing applications.

For the case $d=4$, the authors develop explicit orthogonal and straight‑line drawing algorithms for 4‑regular plane graphs of min‑cut 4. Given such a graph $G$ with $n$ vertices and a marked vertex $v$, they first remove $v$, embed the remaining $(n-1)$‑vertex graph on an $(n-1)\times(n-1)$ integer grid according to a permutation pattern derived from the even Schnyder decomposition, and draw each edge with exactly one bend. The marked vertex $v$ can then be re‑inserted by adding two extra rows and two extra columns; its four incident edges acquire two bends each, for a total of eight additional bends.

A further compaction step eliminates empty rows and columns, yielding a much tighter layout. Empirical analysis on uniformly random instances shows that the final grid size is sharply concentrated around $25n/32 \times 25n/32$, a substantial improvement over the naïve $O(n^2)$ bound.

Overall, the paper makes several substantial contributions:

• It extends the theory of Schnyder woods to all $d$‑angulations, providing necessary and sufficient conditions for existence.
• It offers three interchangeable representations (fractional orientations, corner labellings, lattice structure) that facilitate both theoretical analysis and algorithmic implementation.
• It identifies a rich distributive lattice structure, opening avenues for enumeration, random generation, and optimization of Schnyder decompositions.
• It isolates the even subclass for even $d$, linking it to classical objects on quadrangulations and simplifying the combinatorial description.
• It leverages the dual of even Schnyder decompositions to devise efficient orthogonal and straight‑line drawing algorithms for 4‑regular plane graphs, with provably compact grid embeddings and low bend complexity.

These results not only deepen our understanding of planar graph decompositions but also have practical implications for graph visualization, VLSI layout, and related fields where compact, low‑bend planar drawings are essential.