Pinwheel Graph: Structures and Applications

• Pinwheel graph is a recurring geometric motif defined by cyclic, chiral, and hub-based arrangements, appearing in varied formulations across graph theory, tiling, and condensed matter physics.
• In condensed matter, it models dimer networks on the deformed kagome lattice that mediate triplon propagation and enable tunable thermal Hall conductivity.
• Extensions include higher-order wheel graphs, substitution and neighbor graphs in pinwheel tilings, and interaction graphs in pinwheel artificial spin ice, each governed by local structural rules.

“Pinwheel graph” does not denote a single standardized object across the arXiv literature. The available literature instead uses, or naturally motivates, several graph-theoretic and graph-like constructions organized by a pinwheel pattern: a strong-bond dimer subgraph on the deformed kagome lattice, wheel-derived critical graphs with split spokes, substitution and neighbor graphs for pinwheel tilings, cyclic scheduling graphs for pinwheel-type periodic schedules, strip-transition graphs in polygonal outer billiards, and interaction graphs in pinwheel artificial spin ice (Calonge-Martínez et al., 8 Jun 2026, Zeps, 2011, Frank et al., 2010, Schwartz, 2010, Paterson et al., 2019). In all of these settings, the common feature is not a universal definition but a recurrent geometric motif: a chiral or cyclic arrangement whose local rules are encoded by a graph.

1. Terminological scope and non-standardization

Several sources explicitly indicate that the phrase itself is not fixed. In the graph-theoretic paper on higher-order wheels, the term “pinwheel graph” does not appear, but wheel graphs, higher-order wheels, and wheels with split edges are identified as the structurally closest objects (Zeps, 2011). In the literature on pinwheel tilings, the phrase likewise does not appear explicitly; instead, the relevant structures are substitution graphs, adjacency graphs, orientation graphs, neighbor graphs, and transformation groupoids associated with the pinwheel substitution and its fractal variants (Frank et al., 2010, Bandt et al., 2016, Moustafa, 2010).

By contrast, the study of triplons on the deformed kagome lattice uses “pinwheel” in an explicitly graph-like way: spins are vertices, exchange couplings are edges, and the strongest bonds $J_1$ form a pinwheel-shaped dimer network that acts as the backbone for triplon propagation (Calonge-Martínez et al., 8 Jun 2026). This suggests that “pinwheel graph” functions primarily as a structural descriptor whose precise meaning depends on the underlying field.

2. Pinwheel graph on the deformed kagome lattice

In the most literal condensed-matter usage, the pinwheel graph is the subgraph formed by the strongest $J_1$ bonds in the pinwheel valence-bond-solid phase of the deformed kagome lattice compound Rb$_2$Cu$_3$SnF$_{12}$. The lattice has four inequivalent nearest-neighbor couplings $J_1,J_2,J_3,J_4$ and a 12-site magnetic unit cell. The $J_1$ bonds are the strongest, with $J_1 = 18.6$ meV, and the quoted ratios are $J_2 = 0.95 J_1$, $J_3 = 0.85 J_1$, and $J_1$0. The ground state is a 12-site pinwheel VBS with six dimers per unit cell; the dimer orientations wind clockwise or counter-clockwise in a chiral pattern that preserves $J_1$1 rotation while breaking translations, reflections, inversion, and mirror symmetry. In graph language, the vertices are Cu spin-$J_1$2 sites, the $J_1$3 edges are dimer edges, and the weaker $J_1$4 edges decorate the pinwheel and mediate hopping between dimers (Calonge-Martínez et al., 8 Jun 2026).

The corresponding spin Hamiltonian is a nearest-neighbor Heisenberg model with out-of-plane Dzyaloshinskii–Moriya interactions and magnetic field,

$J_1$5

with $J_1$6 and $J_1$7 for the compound. Using bond-operator mean-field theory, the triplon problem becomes an $J_1$8 bosonic BdG problem because there are $J_1$9 dimers and $_2$0 triplet flavors per unit cell. The paper shows that Dzyaloshinskii–Moriya interactions and external field are both needed to endow the triplon bands with nontrivial Chern numbers; the two lowest isolated bands typically carry $_2$1, and an applied field can isolate the lowest triplon Chern band, producing a tunable thermal Hall conductivity in accessible temperature and field regimes. The calculated low-energy dynamical structure factor agrees qualitatively with neutron-scattering data (Calonge-Martínez et al., 8 Jun 2026).

A related study of the Heisenberg model on the same pinwheel-distorted kagome lattice treats the bond-anisotropic graph as an interpolation between a strongly dimerized pinwheel phase and the uniform kagome-lattice Heisenberg model. Writing $_2$2 and $_2$3, numerical linked-cluster expansions and Lanczos-based zero-temperature NLCE provide strong evidence for a phase transition before $_2$4, implying that the uniform kagome model is likely not pinwheel dimerized and is stable to finite pinwheel-dimerizing perturbations (Khatami et al., 2011).

3. Wheel graphs, higher-order wheels, and pinwheel-like critical graphs

In graph theory, the closest formal relatives of a “pinwheel graph” are ordinary wheel graphs and their higher-order variants. An ordinary wheel is the standard graph

$_2$5

with a cycle $_2$6 as rim and a universal hub vertex $_2$7 joined to all rim vertices. The paper on “4-critical wheel graphs of higher order” treats odd wheels $_2$8, $_2$9, as “ordinary [zero order] wheels,” and introduces a sequence of 4-critical graphs $_3$0 that may be considered first-order higher wheels. These satisfy

$_3$1

and

$_3$2

The relations are realized by edge contractions, specifically by contracting successive side edges (Zeps, 2011).

The same paper organizes these graphs by minor brackets. For first-order higher wheels $_3$3, the brackets are $_3$4 and $_3$5, while ordinary wheels $_3$6, $_3$7, have brackets $_3$8 and $_3$9. The free-Hadwiger class is described by

$_{12}$0

and for $_{12}$1 the first-order higher wheels lie outside $_{12}$2 because they all contain $_{12}$3 as a minor (Zeps, 2011).

Although the paper does not define “pinwheel graph,” it explicitly identifies wheels with split edges as structurally very close to what other sources would call pinwheel-like graphs. Splitting a spoke of $_{12}$4 produces an explicit 4-critical example, and the operation is stated to preserve 4-criticality in general. This supports a graph-theoretic usage in which a pinwheel graph is a wheel-derived planar graph with subdivided spokes or analogous wheel-preserving modifications, especially when the hub-and-rim geometry remains visually pinwheel-like (Zeps, 2011).

4. Pinwheel tilings, substitution graphs, and diffraction

The classical pinwheel tiling is built from a right triangle with side lengths $_{12}$5. Inflation by factor $_{12}$6, together with a rotation by an angle $_{12}$7, subdivides each triangle into five congruent copies of the original triangle. Because $_{12}$8 is irrational, repeated substitution produces tiles in infinitely many orientations, dense on the circle. This substitution admits several natural graph encodings: substitution graphs on tile or chirality classes, adjacency graphs of tiles or control points, and orientation graphs that record how orientation classes propagate under substitution (Grimm et al., 2011, Frank et al., 2010).

The oriented-chirality substitution can be written, modulo $_{12}$9, by the matrix

$J_1,J_2,J_3,J_4$0

with the recursion

$J_1,J_2,J_3,J_4$1

and symmetry $J_1,J_2,J_3,J_4$2. In this sense, the pinwheel substitution is already a directed multigraph whose vertices are orientation/chirality states and whose edges are substitution incidences (Grimm et al., 2011).

The same tiling supports a radial diffraction theory because the autocorrelation is circularly symmetric. For weighted pinwheel point sets, the autocorrelation is written in the form

$J_1,J_2,J_3,J_4$3

where $J_1,J_2,J_3,J_4$4 is the distance set, $J_1,J_2,J_3,J_4$5 are autocorrelation coefficients, and $J_1,J_2,J_3,J_4$6 is the uniform probability measure on the circle of radius $J_1,J_2,J_3,J_4$7. Its Fourier transform is correspondingly radial, producing sharp diffraction rings and possibly a continuous radial component. The balanced weighting $J_1,J_2,J_3,J_4$8 removes the central peak and makes the continuous background easier to detect numerically (Grimm et al., 2011).

The computational-geometry literature extends this tiling viewpoint to meshing. The algorithm PINW is a two-dimensional meshing algorithm based on an extension of pinwheel tilings, and the abstract states that it can generate meshes that accurately approximate the distance between any two domain points by paths composed only of cell edges, while proving that the algorithm produces triangles of bounded aspect ratio [0407018].

Graph structure	Vertices or states	Edges or transitions
Substitution graph	tile types, chiralities, or orientation classes	substitution incidences
Adjacency graph	tiles or control points	shared edges or shared vertices
Orientation graph	discrete orientation classes	one-step orientation changes under substitution

5. Fractal pinwheel tiles, neighbor graphs, and operator-algebraic encodings

A fractal version of the pinwheel tiling introduces 13 basic fractiles up to reflection, or 18 when reflections are distinguished. These fractiles arise from the “aorta,” a substitution-invariant fractal curve embedded in each triangle, and the corresponding substitution matrix $J_1,J_2,J_3,J_4$9 has Perron–Frobenius eigenvalue $J_1$0. The paper gives a left eigenvector of relative areas,

$J_1$1

and an approximate right eigenvector of tile frequencies

$J_1$2

This gives a concrete 13-vertex weighted directed graph whose adjacency matrix is the substitution matrix $J_1$3 (Frank et al., 2010).

A later construction replaces the 13-tile fractal system by a single fractal pinwheel tile. At the tile-type level, the substitution matrix is simply $J_1$4, but the geometry is encoded by a much richer neighbor graph. The paper constructs a finite neighbor graph whose vertices are isometries $J_1$5 with $J_1$6, and whose directed edges are generated by conjugation $J_1$7. After pruning to vertices on cycles, the graph has 81 vertices including the identity; among them are 11 edge neighbors and 69 point neighbors. Two edge-neighbor states are irrational rotations, and these irrational rotations are the graph-theoretic mechanism behind infinitely many orientations and statistical circular symmetry (Bandt et al., 2016).

At a more abstract level, the pinwheel tiling also supports a groupoid and $J_1$8-algebraic encoding. The continuous hull $J_1$9 of the pinwheel tiling carries an action of $J_1 = 18.6$0, and the associated crossed product $J_1 = 18.6$1 acts as the operator-algebraic analogue of a graph or Bratteli-type substitution object. Its $J_1 = 18.6$2-group is described as

$J_1 = 18.6$3

with the image of the $J_1 = 18.6$4 summand under the canonical trace equal to zero and the image of the $J_1 = 18.6$5 summand contained in the module of patch frequencies. The remaining cohomological summand is related to a measured index theorem (Moustafa, 2010). In this operator-algebraic setting, the “pinwheel graph” is best understood as a graph-like dynamical system: a transformation groupoid supplemented by substitution and frequency data.

6. Cyclic scheduling graphs and strip-transition graphs

In discrete Bamboo Garden Trimming, the enhanced pinwheel algorithm organizes tasks by partitions $J_1 = 18.6$6 arranged in a periodic cycle. The paper does not define a graph explicitly, but it presents a natural cycle-with-spokes structure: a directed cycle of partition nodes $J_1 = 18.6$7, together with bamboo or task nodes attached to the partition in which they are serviced. This is the most direct graph interpretation of pinwheel scheduling. The construction supports an approximation ratio converging to $J_1 = 18.6$8 when $J_1 = 18.6$9, and a global worst-case bound of $J_2 = 0.95 J_1$0 (Croce, 2020).

In polygonal outer billiards, the pinwheel map produces another graph-like object. A convex polygon with no parallel sides yields a cyclic family of pinwheel strips $J_2 = 0.95 J_1$1 and associated vectors $J_2 = 0.95 J_1$2. On the layered phase space $J_2 = 0.95 J_1$3, the pinwheel map $J_2 = 0.95 J_1$4 advances through strip states and translates points by $J_2 = 0.95 J_1$5 until they enter the current strip. The paper also introduces spokes $J_2 = 0.95 J_1$6 and admissible paths $J_2 = 0.95 J_1$7, with a bijection between forward-partition tiles and admissible paths. The main Pinwheel Theorem states that for each point $J_2 = 0.95 J_1$8 there exists a section $J_2 = 0.95 J_1$9 and an integer $J_3 = 0.85 J_1$0 such that

$J_3 = 0.85 J_1$1

This gives a bounded-time orbit correspondence between outer billiards and the pinwheel map, so the relevant “pinwheel graph” is a strip-transition or spoke-adjacency graph governing symbolic dynamics (Schwartz, 2010).

7. Interaction graphs in pinwheel artificial spin ice

Pinwheel artificial spin ice begins with square artificial spin ice and rotates every elongated island about its center. At rotation angles around $J_3 = 0.85 J_1$2, the geometry changes from the closed square-ice vertex to an open chiral pinwheel vertex. The resulting system is naturally described as an interaction graph: islands are nodes, and dipolar couplings are edges. The strongest graph motifs are the T-shaped dimer and the four-island pinwheel vertex, while larger arrays are built from two interleaved square subarrays (Paterson et al., 2019).

Micromagnetic modeling shows that the effective nearest-neighbor graph is not the one predicted by a point-dipole Ising description. End-states at the island tips generate a Heisenberg pseudo-exchange interaction that strongly renormalizes the nearest-neighbor edges and makes them field-angle dependent. This produces collective anisotropy axes that are misaligned with the geometrical axes: the T-shaped pair exhibits a merged switching feature near $J_3 = 0.85 J_1$3, the pinwheel vertex near $J_3 = 0.85 J_1$4, large asymmetric arrays saturate near $J_3 = 0.85 J_1$5, large symmetric arrays near $J_3 = 0.85 J_1$6, and periodic boundary conditions restore the anisotropy axis to exactly $J_3 = 0.85 J_1$7. The reversal process is corner-mediated: corner islands, as boundary nodes with reduced stabilizing environment, nucleate avalanches that propagate through nearest-neighbor cascades, with an average domain-wall speed of about $J_3 = 0.85 J_1$8 m/s (Paterson et al., 2019).

Taken together, these uses indicate a recurring structural pattern. A pinwheel graph is typically a graph whose local data are arranged around a hub, a cycle, a chiral vertex, or a substitution center, and whose dynamics are controlled by repeated local rules. This suggests that “pinwheel graph” is best understood not as a single invariant combinatorial class but as a family of graph constructions organized by pinwheel geometry, chirality, and multiscale recursion.