Aztec castles are the finite subgraphs of the dP3 brane tiling — the doubly periodic bipartite dimer lattice dual to the dP3 quiver, a string-theory quiver gauge theory on D3-branes probing the cone over the degree-6 del Pezzo surface — whose weighted perfect matchings realize the cluster variables generated by τ-mutation sequences; every Aztec castle has exactly 2k2^k perfect matchings. Leoni–Musiker–Neel–Turner posed as their Problem 5.2 the graph-theoretic and statistical analysis of these matchings: the distribution of a uniformly random perfect matching of a large castle — in particular whether an Aztec-diamond-type frozen/liquid phase separation (an 'arctic circle') occurs and what the explicit boundary curve is — and the action on castle matchings of the Cottrell–Young domino shuffle for the dP3 lattice, including whether shuffling gives a simple explanation of the power-of-two enumeration and produces new subgraph families with power-of-two perfect-matching counts. The problem remains open.

Background

Cluster algebras, introduced by Fomin and Zelevinsky, are commutative rings generated by cluster variables obtained by iterated mutation of an initial seed; a seed can be encoded by a quiver, a directed graph with no loops or two-cycles, and mutating at a vertex reverses its arrows while adding and deleting arrows along two-paths through it. In string theory, quiver gauge theories on D3-branes probing a toric Calabi–Yau threefold are encoded combinatorially by brane tilings: doubly periodic bipartite planar graphs, i.e. dimer models, whose faces carry the quiver vertices. A central example is the dP3 quiver, associated with the cone over the del Pezzo surface of degree 6 (CP2\mathbb{CP}^2 blown up at three points); its brane tiling, the dP3 lattice, is the hexagonal lattice superimposed on its dual graph.

For such quivers, cluster variables often admit combinatorial interpretations: weighted sums of perfect matchings (also called dimer configurations — sets of edges covering every vertex exactly once) of explicit finite subgraphs of the brane tiling equal the Laurent expansions of cluster variables. Sicong Zhang treated one period-6 mutation sequence on the dP3 quiver through subgraphs called Aztec dragons; Leoni–Musiker–Neel–Turner extended this to all τ-mutation sequences, meaning compositions of mutations with vertex relabelings that satisfy the Coxeter relations of the affine symmetric group A~2\tilde{A}_2, so that a mutation sequence becomes a walk on the triangular Coxeter lattice and the clusters it produces correspond to alcoves (the triangular cells of that lattice). Each cluster variable equals the weighted sum of perfect matchings of an associated subgraph of the dP3 lattice, called an Aztec castle, and the paper's factorization formula implies that the number of perfect matchings of every Aztec castle is exactly a power of 2, 2k2^k, with kk growing along the lattice. As for any planar bipartite graph, a matching can be encoded by a height function, an integer-valued function on the faces whose edge-to-edge increments are fixed by the matching.

The classical analogue is the Aztec diamond: by the arctic circle theorem of Jockusch–Propp–Shor, whose liquid-region local statistics and independent proof were supplied by Cohn–Elkies–Propp, a uniformly random domino tiling of a large Aztec diamond is frozen — locally deterministic, with a brick-wall pattern — outside an inscribed circle and disordered inside it, with an explicit phase boundary. Domino shuffling, the recursive local-move algorithm going back to Elkies–Kuperberg–Larsen–Propp, both generates uniform tilings and explains bijectively why the number of Aztec diamond tilings is 2n(n+1)/22^{n(n+1)/2}. Cottrell and Young adapted domino shuffling to the dP3 lattice: their shuffle counts perfect matchings of finite subgraphs whose boundary conditions are compatible with the algorithm, reproving an enumerative theorem of Ciucu and producing a related family of subgraphs with exactly 2(n+1)22^{(n+1)^2} perfect matchings. The general theory of limit shapes for the dimer model (Kenyon–Okounkov) shows that on any periodic bipartite graph the large-scale statistics of random perfect matchings are governed by a variational principle, with frozen phases controlled by the amoeba of the dimer spectral curve (the amoeba being the image of that curve under the coordinatewise map z↦(log⁡∣z1∣,log⁡∣z2∣)z \mapsto (\log|z_1|, \log|z_2|)); applying it to a concrete family, however, requires a family-specific computation, and none exists for the castles.

Aztec castles are not rescaled dilates of one region: they are indexed by alcoves of a Coxeter lattice, grow along several lattice directions at once, and typically have non-convex boundaries, so neither the classical arctic-circle analysis nor the Cottrell–Young shuffle applies to them directly. Leoni–Musiker–Neel–Turner explicitly left this analysis open as their Problem 5.2.

Problem Statement

Problem 5.2 of Leoni–Musiker–Neel–Turner: determine the graph-theoretic and statistical properties of the Aztec castles, the finite subgraphs of the dP3 brane tiling whose weighted perfect-matching sums realize the cluster variables generated by τ-mutation sequences on the dP3 quiver. The problem comprises two coupled parts, both in direct analogy with the classical theory of the Aztec diamond. Throughout, "Aztec castle" refers to the full alcove-indexed family of subgraphs constructed by Leoni–Musiker–Neel–Turner, of which the Aztec dragons are only one one-parameter subfamily. A resolution must address both parts: in part (1) the affirmative branch requires both the occurrence verdict and the explicit frozen-boundary curve in alcove coordinates, while the negative branch requires both the limiting height-function profile and the identified preventing feature; in part (2), determining the shuffle's action on castle matchings and its enumerative consequences is required.

(1) Random-matching statistics and arctic behavior. For a perfect matching of an Aztec castle chosen uniformly at random, describe the distribution of the matching as the castle grows along its natural growth regimes in the A~2\tilde{A}_2 Coxeter lattice (for example, along sequences of alcoves with unbounded coordinates). Does an Aztec-diamond-type 'arctic circle' phenomenon occur — a separation between frozen regions, in which the matching is forced or asymptotically deterministic, and a liquid region with non-degenerate local statistics? If it occurs, give the explicit frozen-boundary (arctic) curve in terms of the alcove coordinates of the castle; if it does not, characterize the limiting height-function profile and identify which feature of the castles (for example their non-convex boundaries or multi-directional growth) prevents the standard frozen/liquid phenomenology.

(2) Action of the dP3 domino shuffle. Cottrell and Young defined a domino shuffle on the dP3 lattice, valid for finite subgraphs with shuffle-compatible boundary conditions. Determine what this shuffle does to the perfect matchings of Aztec castles: does it map the matchings of a castle bijectively to the perfect matchings of another Aztec castle, or to the perfect matchings of some new class of subgraphs of the dP3 brane tiling — or is there no shuffle-compatible way to apply or extend it to castle boundaries at all? Does the shuffle furnish a simple explanation of the theorem that every Aztec castle has exactly 2k2^k perfect matchings and, if the shuffling does not give perfect matchings of other Aztec castles, might it still produce a new class of subgraphs of the brane tiling whose numbers of perfect matchings are powers of 2?

Both parts serve one scientific objective: a probabilistic and dynamical understanding of the same castle ensemble — the first through limit-shape analysis of the uniform measure on its perfect matchings, the second through the recursive shuffling structure that would explain the power-of-two enumeration — mirroring how the arctic circle theorem and domino shuffling complete each other for the Aztec diamond.

References

  1. M. Leoni, G. Musiker, S. Neel, P. Turner, 'Aztec Castles and the dP3 Quiver', Journal of Physics A: Mathematical and Theoretical 47 (2014) 474011; DOI 10.1088/1751-8113/47/47/474011; arXiv:1308.3926; https://arxiv.org/abs/1308.3926
  2. S. Fomin, A. Zelevinsky, 'Cluster algebras I: Foundations', Journal of the American Mathematical Society 15 (2002) 497–529; DOI 10.1090/S0894-0347-01-00385-X; https://doi.org/10.1090/S0894-0347-01-00385-X
  3. S. Zhang, 'Cluster Variables and Perfect Matchings of Subgraphs of the dP3 Lattice', arXiv:1511.06055 (2015); https://arxiv.org/abs/1511.06055
  4. H. Cohn, N. Elkies, J. Propp, 'Local statistics for random domino tilings of the Aztec diamond', Duke Mathematical Journal 85 (1996) 117–166; DOI 10.1215/S0012-7094-96-08506-3; arXiv:math/0008243; https://arxiv.org/abs/math/0008243
  5. W. Jockusch, J. Propp, P. Shor, 'Random Domino Tilings and the Arctic Circle Theorem', Annales de l'Institut Henri Poincaré D, online first (2026); DOI 10.4171/aihpd/233; https://doi.org/10.4171/aihpd/233; preprint arXiv:math/9801068 (1998); https://arxiv.org/abs/math/9801068
  6. N. Elkies, G. Kuperberg, M. Larsen, J. Propp, 'Alternating-Sign Matrices and Domino Tilings (Part I)', Journal of Algebraic Combinatorics 1 (1992) 111–132; DOI 10.1023/A:1022420103267; https://doi.org/10.1023/A:1022420103267
  7. M. Ciucu, 'Perfect Matchings and Perfect Powers', Journal of Algebraic Combinatorics 17 (2003) 335–375; DOI 10.1023/A:1025005023573; https://doi.org/10.1023/A:1025005023573
  8. C. Cottrell, B. Young, 'Domino shuffling for the Del Pezzo 3 lattice', arXiv:1011.0045 (2010); https://arxiv.org/abs/1011.0045
  9. R. Kenyon, 'An introduction to the dimer model', arXiv:math/0310326 (2003); https://arxiv.org/abs/math/0310326
  10. R. Kenyon, A. Okounkov, 'Limit shapes and the complex Burgers equation', Acta Mathematica 199 (2007) 263–302; DOI 10.1007/s11511-007-0021-0; https://doi.org/10.1007/s11511-007-0021-0
  11. T. Lai, G. Musiker, 'Beyond Aztec Castles: Toric Cascades in the dP3 Quiver', Communications in Mathematical Physics 356 (2017) 823–881; DOI 10.1007/s00220-017-2993-8; arXiv:1512.00507; https://arxiv.org/abs/1512.00507
  12. P. Di Francesco, R. Soto-Garrido, 'Arctic curves of the octahedron equation', Journal of Physics A: Mathematical and Theoretical 47 (2014) 285204; DOI 10.1088/1751-8113/47/28/285204; arXiv:1402.4493; https://arxiv.org/abs/1402.4493
  13. J. H. Chiang, G. Musiker, S. Nguyen, 'Minimal Matchings for dP3 Cluster Variables', The Electronic Journal of Combinatorics 33(2) (2026) P2.44; DOI 10.37236/14192; arXiv:2409.05803; https://arxiv.org/abs/2409.05803
  14. Q. Chen, G. Musiker, 'Gale-Robinson Quivers and Principal Coefficients', arXiv:2602.19402 (2026); https://arxiv.org/abs/2602.19402