Bisection Graph Analysis

• Bisection graph is a directed multigraph representing triangle similarity classes and their evolution under longest-edge bisection.
• The spectral analysis reveals terminal quadruple cycles with a spectral radius of 2, ensuring exponential decay of nonterminal classes.
• This framework underpins advanced mesh refinement techniques by providing precise estimates on area distribution and high-quality mesh generation.

A bisection graph encodes the discrete dynamical system underlying the similarity-class evolution of triangles or higher-dimensional simplices under longest-edge bisection (LEB). In the context of triangle bisection, the bisection graph is the directed multigraph whose vertices correspond to the similarity classes of triangles generated from a fixed initial seed, with directed arcs encoding the refinement transitions induced by the two canonical LEB maps. This structure provides a combinatorial–spectral framework for analyzing the evolution, stability, and statistical properties of triangle shapes produced by repeated refinement. The bisection graph concept is central to recent advances in shape-regularity analysis, terminal quadruple localization, and asymptotic area distribution laws for LEB-based mesh generation (Kalmanovich et al., 20 Jan 2026).

1. Definition and Construction of the Bisection Graph

Given an initial triangle $\Delta_0$, the iterative application of LEB produces a finite set of similarity classes %%%%1%%%%: each new triangle at every refinement step is mapped (by normalization) to a unique class. The bisection graph $G_z$ (for normalized initial data $z$) is constructed as follows:

• Vertices: Each vertex represents a similarity class in the refinement orbit of $\Delta_0$.
• Edges (Arcs): Each vertex corresponding to class $i$ has two outgoing arcs, labelled $L$ and $R$, pointing to the classes $j=L(i)$ and $j=R(i)$ obtained by applying the left and right LEB refinement maps, respectively.
• Adjacency Structure: After splitting self-loops (arising when $L(i) = R(i) = i$) to yield a simple digraph, the adjacency matrix $A$ of $G_z$ records the connectivity.

This directed graph structure is well-defined because, by Adler's Theorem, the refinement process generates only finitely many similarity classes from any initial triangle (Kalmanovich et al., 20 Jan 2026). The bisection graph reflects the full transition dynamics of LEB, independent of mesh geometry or spatial embedding.

2. Spectral Properties and Terminal Quadruples

The bisection graph’s transition matrix $A$ captures the area flow among similarity classes under repeated LEB. The key spectral facts are:

• Spectral Radius: The spectral radius of $A$ is exactly 2; the number of linearly independent eigenvectors for eigenvalues $\lambda = \pm2$ equals the number $q(z)$ of terminal quadruples in the refinement orbit.
• Asymptotic Dominance: The vertices corresponding to terminal quadruples form strongly connected subgraphs (cycles of length four contained in the "fat" region $A$ of shape space), and these subgraphs are precisely those with spectral radius 2. All other components have spectral radius less than 2 and thus vanish exponentially in area share as refinement proceeds.

Terminal quadruples—sets of four distinct, non-degenerate similarity classes forming a cycle under the LEB maps—dominate the asymptotic distribution of mesh triangles. For generic acute triangles, $q(z) = 1$, while for certain pathological initial conditions, $q(z) \rightarrow \infty$ is possible.

3. Dynamical Systems Interpretation

The bisection graph encodes the symbolic dynamics of the LEB process as a noncommutative shift system on shape space. Each walk on the graph corresponds to a possible sequence of refinements (left/right choices) mapping the initial triangle to a descendant similarity class. Iterating the refinement thus corresponds to following (random or deterministic) paths in $G_z$.

The discrete-time evolution of the area-mass vector $w^{(n)} \in \mathbb{R}^\ell$, representing the proportion of total area in each similarity class after $n$ steps, is governed by powers of $A$: $w^{(n)} = \frac{A^n e_1}{\|A^n e_1\|_1}$ where $e_1$ localizes the initial class (Kalmanovich et al., 20 Jan 2026). The leading eigenspaces determine the stabilization onto the terminal quadruple cycles.

4. Exact and Asymptotic Distribution of Similarity Classes

The bisection graph provides the structure necessary for rigorous analysis of the distribution of triangle types under repeated LEB:

• Terminal Dominance Theorem: For every initial triangle, the area fraction occupied by terminal quadruple similarity classes converges to one at an exponential rate:

$$1 - \frac{A_n}{A_0} = O(\xi^n), \qquad \xi = \sqrt{2} / 2$$

where $A_n$ is the total area in terminal quadruples after $n$ steps, and $\xi$ is determined by the spectral gap of $A$ (Kalmanovich et al., 20 Jan 2026).

• Eigenspace Decomposition: Decomposing $\mathbb{R}^\ell$ into $\lambda = \pm2$ eigenspaces and the contracting subspace produces explicit exponential-decay estimates for the nonterminal class distribution.
• Multiplicity of Terminal Quadruples: For $z$ in regions $I \cup II \cup III \cup IV$ of normalized shape space, exactly one terminal quadruple occurs; in more degenerate settings, arbitrarily many may arise.

5. Geometric and Topological Properties

Within the bisection graph, terminal quadruples correspond to “fat” cycles whose similarity classes are uniformly separated from degeneration:

• Cycle Geometry: Each terminal quadruple lies on a hyperbolic circle centered at $\zeta = 1/2 + 1/2i$, and thus on a Euclidean circle with center at $\mathrm{Re}(z) = 1/2$. All members of a quadruple share the same hyperbolic distance to $\zeta$.
• Quality Metrics: Every member of a terminal quadruple attains a uniformly positive lower bound on all triangle angles and areas, ruling out sliver formation.

6. Extensions and Practical Implications

The bisection graph framework extends beyond two dimensions to the analysis of simplex bisection processes in tetrahedra and higher simplices, provided the finiteness of similarity classes is preserved. The approach is also adaptable to weighted and locally refined (adaptive) bisection strategies, if the strict-contraction bounds and terminal class identification can be verified.

In mesh refinement practice, the core consequences are:

• Mesh Quality: Almost all of the mesh area comes to reside in a finite, explicit set of well-shaped (terminal) triangle types, regardless of initial conditions.
• Complexity Control: The area-dominant cycles are limited in number ($\leq 4\cdot q(z)$), ensuring asymptotic regularity even if the total number of similarity classes grows logarithmically with the sharpness of initial geometries.
• Algorithmic Predictability: The spectral analysis of the bisection graph enables a priori estimates on refinement depth required to achieve a desired area-fraction coverage by high-quality triangles.

7. Relationship to the Perdomo–Plaza Hyperbolic Framework

The bisection graph formalism sharpens and extends the Perdomo–Plaza model, in which the normalized triangle shape-space $D$ in the hyperbolic half-plane admits piecewise Möbius bisection maps $L$, $R$. The spectral/graph-theoretic approach clarifies the contraction dynamics on $D$’s subregions and delivers both the geometry of terminal shapes and the exponential convergence outside the terminal cycles. New commutation relations among the Möbius maps, as well as refined subdivision of shape space, further elucidate the detailed classification and invariance properties observed in practice.

The bisection graph is, accordingly, the analytical backbone for understanding LEB mesh refinement regularity via explicit combinatorial and spectral techniques, unifying shape dynamics and mesh-quality guarantees in finite-element mesh generation (Kalmanovich et al., 20 Jan 2026).