Perfect t-Embeddings in Dimer Models

• Perfect t-embeddings are geometric realizations of augmented dual graphs that encode dimer model edge weights using gauge-equivalent metrics and discrete angle-balancing.
• They pair with origami maps to form discrete surfaces in Lorentzian/Minkowski space, enabling convergence analysis toward Lorentz-minimal surfaces and Gaussian free field limits.
• Explicit constructions in hexagons, Aztec diamonds, and tower graphs verify rigidity conditions, linking discrete conformal structures to continuum minimal surfaces.

Perfect $t$-embeddings are a class of geometric realizations of augmented dual graphs of weighted planar bipartite graphs, developed in the study of dimer models. They encode edge weights through a gauge-equivalent metric structure, impose a discrete angle-balancing condition at interior vertices, and impose a distinguished tangential boundary geometry. In the modern formulation, a perfect $t$-embedding is paired with an associated origami map, and the pair is interpreted as a discrete surface in Lorentzian or Minkowski space. This framework has become a central tool for relating finite dimer models to continuum objects such as Lorentz-minimal or maximal surfaces and to Gaussian free field limits for height fluctuations (Chelkak et al., 2021).

1. Definition and finite-graph structure

For a finite weighted bipartite planar graph $G=(B\cup W,E)$, one works on the augmented dual graph $G^*$, obtained by augmenting the outer dual face. A $t$-embedding is a proper embedding of $G^*$ with straight edges and convex inner faces such that, at each inner vertex, the sum of angles of black faces adjacent to that vertex equals $\pi$, and likewise the sum of angles of white faces equals $\pi$. In parallel, the dual edge lengths reproduce the original edge weights up to gauge equivalence (Chelkak et al., 2021).

The adjective “perfect” refers to an additional boundary normalization. In the formulation used across the dimer literature, the outer face is a tangential polygon to a circle or, in a normalized version, to the unit circle, and all non-boundary edges adjacent to boundary vertices lie on the bisectors of the corresponding boundary angles. One equivalent formulation states that each boundary-adjacent inner vertex lies on the bisector of the corresponding boundary angle (Berggren et al., 6 Aug 2025).

This boundary geometry is not ornamental. It rigidifies the discrete conformal data and is the finite-graph condition that makes the associated origami map compatible with a Lorentzian surface interpretation. In the gauge-theoretic language, a perfect $t$-embedding is equivalent to a perfect Coulomb gauge after a suitable global complex scaling and additive normalization of the primitives (Chelkak et al., 2021).

2. Coulomb gauges, origami maps, and boundary parametrization

The standard construction uses two complex gauge functions, often written $\mathcal F^\bullet:B\to\mathbb C$ and $t$0, together with a real Kasteleyn matrix $t$1. The embedding and the origami map are defined by the edge increments

$t$2

These formulas occur in several constructions and make explicit that the same discrete data simultaneously determine the planar embedding and its Lorentzian companion (Chelkak et al., 2021).

A complementary description uses an origami square root function $t$3, from which the origami differential is defined piecewise on black and white faces. In this form, the origami map is $t$4-Lipschitz, and for a perfect $t$5-embedding the boundary is mapped to a line, usually normalized to the real axis (Chelkak et al., 2021).

The boundary of a perfect $t$6-embedding can be parametrized by angles $t$7 and $t$8 through

$t$9

The identities

$G=(B\cup W,E)$0

express the tangency of the boundary edges to the unit circle. The corresponding planar domain is

$G=(B\cup W,E)$1

where $G=(B\cup W,E)$2 is the piecewise linear interpolation of the boundary data (Chelkak et al., 2021).

The pair $G=(B\cup W,E)$3 is naturally regarded as a discrete surface in $G=(B\cup W,E)$4. On the boundary, the image lies on the one-sheet hyperboloid

$G=(B\cup W,E)$5

with Minkowski product

$G=(B\cup W,E)$6

This hyperboloid formulation is one of the key links between discrete dimer geometry and Lorentzian surface theory (Chelkak et al., 2021).

3. Scaling limits, Lorentz-minimal surfaces, and height fluctuations

The central analytical program is to study sequences of perfect $G=(B\cup W,E)$7-embeddings whose mesh tends to zero. In the foundational framework, one assumes that the corresponding origami maps converge uniformly on compact subsets and that the limiting graph $G=(B\cup W,E)$8 is the Lorentz-minimal surface $G=(B\cup W,E)$9 solving the Plateau problem in $G^*$0 with the prescribed boundary contour (Chelkak et al., 2021).

Under very mild technical assumptions, this setup yields convergence of the gradients of the height correlation functions to those of the Gaussian Free Field defined in the intrinsic metric of the surface $G^*$1. The statement is explicitly about gradients of height correlations rather than necessarily the correlation functions themselves. This distinction is important because the limiting object may be defined only up to an additive global random variable (Chelkak et al., 2021).

Two technical conditions recur in the theory. The first is a local distortion condition, sometimes denoted LIP, requiring a strict contraction estimate for the origami map on compact subsets at mesoscopic scales. The second is an exponential fatness exclusion, EXP-FAT, which rules out pathological microscopic geometry after suitable face splittings and removal of very tiny fat faces. These hypotheses are used to obtain regularity of $G^*$2-holomorphic functions and uniform control on inverse Kasteleyn entries (Chelkak et al., 2021).

Later work shows that in explicit families these conditions can often be verified through a stronger property called rigidity. In such cases, the discrete complex-analytic machinery applies without taking the convergence of the geometry as a black box. This is how perfect $G^*$3-embeddings become a practical route from exact finite formulas to Gaussian free field scaling limits (Berggren et al., 2024).

4. Explicit constructions and model families

Several graph families now admit explicit perfect $G^*$4-embeddings or explicit formulas for their associated $G^*$5-surfaces.

Family	Main result	arXiv
Regular hexagons of the hexagonal lattice	First explicit perfect $G^*$6-embedding for outer face degree $G^*$7; exact contour integral formulas; origami maps converge to a maximal surface in $G^*$8	(Berggren et al., 2024)
Doubly periodic Aztec diamonds	$G^*$9-surfaces converge to space-like maximal surfaces; gas regions create interior light-like cusps	(Berggren et al., 6 Aug 2025)
Two-periodic Aztec diamond	$t$0-embedding and origami coordinates are expressed as sums of density functions from the octahedron equation	(Berggren et al., 8 Aug 2025)
Uniformly weighted generalized tower graphs	Rigidity is verified, implying Gaussian free field convergence for gradients of height fluctuations	(Keating et al., 23 Sep 2025)

The regular-hexagon construction is historically notable because it provides the first example, and hence proves existence, for graphs with an outer face of degree greater than four. The construction is based on the inverse Kasteleyn matrix and relies only on symmetries of the graph. After normalization, the boundary polygon becomes a regular hexagon, and the origami boundary collapses to alternating values on the real axis (Berggren et al., 2024).

For the two-periodic Aztec diamond, the coordinates of the perfect $t$1-embedding and the origami map are written in the form

$t$2

$t$3

where the functions $t$4 are expressed as sums of density functions arising from the octahedron equation. The density functions admit the dimer-probabilistic interpretation

$t$5

so the embedding coordinates are assembled directly from local dimer statistics (Berggren et al., 8 Aug 2025).

Generalized tower graphs extend the Aztec-diamond-type shuffling framework. Their perfect $t$6-embeddings are identified with those of suitable Aztec diamonds at varying effective rank, which allows one to import contour-integral asymptotics and prove rigidity. This family includes the Aztec diamond, the classical tower graph, and further examples defined by functions such as $t$7 and $t$8 (Keating et al., 23 Sep 2025).

5. Periodic Aztec diamonds, gas regions, and extrinsic geometry

The large-scale geometry of perfect $t$9-embeddings is especially transparent in doubly periodic Aztec diamonds. In the $G^*$0-periodic case there are no gas regions. The associated $G^*$1-surface converges to a unique space-like maximal surface in $G^*$2 with boundary a rhombus, and all $G^*$3 frozen regions collapse to the same four boundary points of that rhombus, regardless of how many frozen components there are (Berggren et al., 6 Aug 2025).

The $G^*$4-periodic case is qualitatively different. Here there are $G^*$5 gas regions, and the limit is a space-like maximal surface in $G^*$6 with $G^*$7 light-like cusps in the interior. Each gas region collapses to a distinct interior point $G^*$8, while the four frozen regions still collapse to the four boundary points of the limiting quadrilateral. In this regime the image of the origami map is generally not one-dimensional, and the limiting surface generally cannot be embedded in $G^*$9 (Berggren et al., 6 Aug 2025).

This dichotomy shows that the extrinsic geometry of the $\pi$0-surface is sensitive to the detailed weight distribution. The boundary rhombus depends only on a single edge-weight parameter $\pi$1, but the cusp positions depend on the full spectral data, including the ordering of the weights and the standard divisor parameter $\pi$2. By contrast, the global conformal structure is robust and coincides with the Kenyon–Okounkov conformal structure. The papers therefore separate two levels of geometry: a universal conformal structure associated with the liquid region, and finer extrinsic data carried by the maximal surface and, in particular, by the cusp locations (Berggren et al., 6 Aug 2025).

A conjectural aspect of this picture is that the cusp apices encode the vector $\pi$3 that shifts the discrete Gaussian component in global fluctuations. In the notation of the periodic Aztec-diamond analysis, this vector is the spectral parameter $\pi$4. The two-periodic case provides explicit evidence for this dependence through the motion of the cusp apex under shifts of the weights (Berggren et al., 6 Aug 2025).

6. Rigidity, universality, and open problems

Rigidity is the main regularity property that upgrades perfect $\pi$5-embeddings from a geometric formalism to a usable analytical tool. For a compact $\pi$6, rigidity requires edge lengths in $\pi$7 to remain comparable to the mesh scale and angles to stay uniformly bounded away from $\pi$8 and $\pi$9. In the generalized tower-graph setting, rigidity implies both LIP and EXP-FAT with

$\pi$0

which is precisely the scale needed to invoke the general Gaussian free field convergence theorem (Keating et al., 23 Sep 2025).

The regular-hexagon case provides a parallel bulk-rigidity statement: for every compact $\pi$1, dual edge lengths are of order $\pi$2 and all angles are uniformly nondegenerate. These bounds are obtained from steepest-descent asymptotics of the gauge functions and are what permit the application of the general convergence theorem to lozenge tilings (Berggren et al., 2024).

A major conceptual consequence is that perfect $\pi$3-embeddings furnish a unifying discrete conformal structure across several classes of dimer models. In each explicit family, the embedding, the origami map, and the inverse Kasteleyn matrix interact tightly enough to identify a continuum maximal surface and then to transfer Gaussian free field fluctuations from the discrete model to the intrinsic conformal geometry of that surface. This suggests a broad universality mechanism, although that broader statement is interpretive rather than formally proved in a single theorem.

The main unresolved issue remains general existence and uniqueness. The foundational theory explicitly notes that existence and uniqueness of perfect $\pi$4-embeddings remain open in general (Chelkak et al., 2021). Subsequent work has expanded the list of explicit examples—most notably to regular hexagons, periodic Aztec diamonds, and generalized tower graphs—but these examples also show that the extrinsic Lorentzian geometry can vary sharply with periodicity, gas regions, and spectral data. Perfect $\pi$5-embeddings therefore occupy a dual role: they are both a discrete analytic device for proving Gaussian limits and a geometric invariant that detects finer structure beyond conformal type.