Generalized Tower Graphs in Dimer Theory and Algebra

• Generalized tower graphs are finite, planar bipartite structures defined by shuffling algorithms guided by up-down paths, extending the Aztec diamond and classical tower graph constructions.
• In planar dimer theory, these graphs admit perfect t-embeddings with rigorous rigidity estimates that underpin the convergence of height fluctuations to the Gaussian free field.
• In commutative algebra, generalized tower graphs correspond to generalized tower sets whose edge combinatorics characterize height-2 arithmetically Cohen–Macaulay squarefree monomial ideals.

Generalized tower graphs designate distinct objects in at least two active research settings. In planar dimer theory, they are a broad family of finite, planar, bipartite graphs defined by a shuffling algorithm that extends the Aztec diamond and tower graph constructions; their geometry is encoded by an up-down path and supports perfect $t$-embeddings whose fluctuation theory leads to the Gaussian free field (Keating et al., 23 Sep 2025). In commutative algebra, the phrase refers naturally to simple graphs whose edge sets are generalized tower sets, a codimension-$2$ combinatorial structure characterizing the primary decompositions of arithmetically Cohen–Macaulay squarefree monomial ideals (Favacchio et al., 2014). The shared terminology reflects extension of a more rigid “tower” pattern, but the underlying objects, invariants, and applications are field-specific.

1. Terminological scope and principal meanings

Two meanings dominate the current technical use of the term.

Context	Defining structure	Principal outcome
Planar dimer models	Finite planar bipartite square-hexagon graphs determined by an up-down path	Perfect $t$-embeddings and convergence of height-fluctuation gradients to the Gaussian free field
Commutative algebra	Simple graphs whose edge sets are generalized tower sets	Characterization of height-$2$ aCM squarefree monomial ideals

In the dimer-model setting, generalized tower graphs are introduced as a broad family defined by a shuffling algorithm that generalizes the construction of Aztec diamond and tower graphs; the sequence of columns and column heights is determined by an up-down path, and the uniformly weighted model admits perfect $t$-embeddings satisfying rigidity assumptions sufficient for Gaussian free field fluctuation results (Keating et al., 23 Sep 2025). In the algebraic setting, generalized tower graphs arise from generalized tower sets in $(\mathbb{Z}_+^2)^*$, where the edges of the graph encode minimal primes of a squarefree monomial ideal, and the resulting combinatorics characterize the arithmetically Cohen–Macaulay property in height $2$ (Favacchio et al., 2014).

The two usages are not interchangeable. One belongs to integrable probability, planar bipartite geometry, and dimer asymptotics; the other belongs to combinatorial commutative algebra, Hilbert–Burch theory, and primary decomposition.

2. Generalized tower graphs in planar dimer theory

Generalized tower graphs are defined as a family of finite, planar, bipartite graphs obtained by a shuffling algorithm generalizing the Aztec diamond and tower graph constructions. They are square-hexagon graphs whose sequence of columns and column heights is determined by an up-down path. An up-down path assigns to every integer $c$ a height $H(c)$ such that $H(c+1)-H(c)=\pm 1$, and it is often chosen to approximate a profile $2$0 with $2$1; a rank-$2$2 generalized tower graph can then be arranged so that column heights remain within a fixed error of $2$3 (Keating et al., 23 Sep 2025).

The basic examples are explicit. The Aztec diamond corresponds to the profile $2$4, while the tower graph corresponds to $2$5. More general profiles, including quadratic and sinusoidal ones, produce graphs with locally varying geometry. Construction starts from a minimal rank-$2$6 template and proceeds by shuffling columns through local moves, specifically spider move, gauge transformations, and contraction of degree $2$7 vertices. The labeling of faces by $2$8 and the column indexing encode the relation to the profile function, with boundaries described by $2$9. The construction extends naturally to reduced graphs obtained after contracting all degree-$t$0 vertices, and all reduced generalized tower graphs have degree-$t$1 boundary (Keating et al., 23 Sep 2025).

A precursor study of uniformly weighted Aztec diamonds states that perfect $t$2-embeddings were also constructed and analyzed for a sequence of uniformly weighted finite graphs called tower graphs, and that a simple transformation identifies height fluctuations on the tower graph with those of the Aztec diamond, although not all technical assumptions of the relevant theorem were checked there (Berggren et al., 2023). This suggests that the generalized tower graph framework systematizes and extends a previously isolated tower-graph case.

3. Perfect $t$3-embeddings, rigidity, and fluctuation theory

A $t$4-embedding of a planar bipartite graph $t$5 is an embedding of the dual graph $t$6 in the complex plane such that the sum of the angles at each internal vertex, for black and white faces separately, is $t$7, and the geometric dual-edge lengths are gauge equivalent to the original edge weights. A perfect $t$8-embedding is one in which the outer face of $t$9 maps to a tangential polygon inscribed in a circle and edges adjacent to boundary vertices lie along the bisectors of the boundary angles (Keating et al., 23 Sep 2025).

For generalized tower graphs, the perfect $2$0-embedding is explicitly related to that of the Aztec diamond. For any face $2$1 in the reduced generalized tower graph of rank $2$2, the embedding coordinates satisfy

$2$3

where

$2$4

The construction is recursive through elementary moves such as spider move, contraction, and double-edge splitting, and the resulting embedding is invariant under the choice of shuffling order. In the uniformly weighted setting, all edge weights are $2$5 (Keating et al., 23 Sep 2025).

The associated origami map $2$6 is built from an origami square root function encoding local rotation and folding data. For perfect $2$7-embeddings, this map realizes the model surface in $2$8. The fluctuation theory requires technical assumptions denoted LIP, EXP-FAT, and rigidity. Rigidity is the strongest: for any compact $2$9 in the liquid region, every edge length in $t$0 lying in $t$1 is between $t$2 and $t$3 for some positive constant $t$4 independent of $t$5, and all face angles in $t$6 are uniformly bounded away from $t$7 and $t$8. The proof uses steepest descent analysis on explicit contour integral formulae for local geometry (Keating et al., 23 Sep 2025).

These estimates yield the main asymptotic conclusions. The embedding domains converge in Hausdorff sense to a simply connected planar region; the origami maps converge uniformly to the graph of a smooth function $t$9, a maximal surface in Minkowski space; and the gradients of the $(\mathbb{Z}_+^2)^*$0-point correlation functions of the centered dimer-model height function converge to those of the standard Gaussian free field. In the notation supplied for the model,

$(\mathbb{Z}_+^2)^*$1

Local observables, including edge probabilities and arctic curves, are computable through contour integrals, and the liquid region is described by

$(\mathbb{Z}_+^2)^*$2

The same framework therefore treats geometry, local statistics, and fluctuation asymptotics in one package (Keating et al., 23 Sep 2025).

4. Generalized tower graphs in codimension-$(\mathbb{Z}_+^2)^*$3 combinatorial commutative algebra

The algebraic theory begins with tower sets and tower schemes. A tower set is a finite subset $(\mathbb{Z}_+^2)^*$4 satisfying a nested-fiber condition: for every $(\mathbb{Z}_+^2)^*$5 and all $(\mathbb{Z}_+^2)^*$6 with $(\mathbb{Z}_+^2)^*$7, the nonemptiness of $(\mathbb{Z}_+^2)^*$8 implies $(\mathbb{Z}_+^2)^*$9. Given generic families of forms $2$0, the corresponding tower scheme is cut out by

$2$1

and tower schemes are arithmetically Cohen–Macaulay (Favacchio et al., 2014).

In codimension $2$2, not every arithmetically Cohen–Macaulay squarefree monomial ideal arises from a tower scheme, so the notion is extended. A generalized tower set is a subset $2$3 satisfying four conditions: connectedness; a decomposition $2$4 with $2$5 a tower set; the requirement that for every $2$6, the index $2$7 is new relative to $2$8 while $2$9; and a closure condition stating that if $c$0 and $c$1, then $c$2, where

$c$3

The associated generalized tower scheme is

$c$4

Its primary decomposition is encoded by $c$5 (Favacchio et al., 2014).

The graph-theoretic interpretation is direct. The elements of $c$6 correspond to minimal primes $c$7, so $c$8 can be viewed as the edge set of a simple graph on $c$9 vertices, with no loops because $H(c)$0. In this sense, generalized tower graphs are precisely the combinatorial graphs whose edge sets are generalized tower sets. The principal theorems are exact. If $H(c)$1 is a generalized tower set, then

$H(c)$2

is arithmetically Cohen–Macaulay. Conversely, a squarefree monomial ideal $H(c)$3 of height $H(c)$4 is arithmetically Cohen–Macaulay if and only if its primary support

$H(c)$5

is a generalized tower set (Favacchio et al., 2014).

This produces a precise combinatorial characterization of the primary decompositions of all height-$H(c)$6 arithmetically Cohen–Macaulay squarefree monomial ideals. The canonical example supplied is

$H(c)$7

for which $H(c)$8 is a determinantal ideal of $H(c)$9 minors of a certain matrix; it is not towerizable, but it is a generalized tower set, hence yields an arithmetically Cohen–Macaulay generalized tower scheme (Favacchio et al., 2014).

5. Relation to earlier tower constructions and common disambiguations

The phrase “generalized tower graph” should not be conflated with other graph families carrying the word “tower.” Within the dimer literature, a 2023 abstract identifies “tower graphs” as another sequence of uniformly weighted finite graphs admitting perfect $H(c+1)-H(c)=\pm 1$0-embeddings, with height fluctuations related by a simple transformation to those of the Aztec diamond (Berggren et al., 2023). The later generalized tower graph construction enlarges that picture by allowing up-down paths and profile functions rather than a single fixed geometry (Keating et al., 23 Sep 2025).

Outside the dimer setting, several unrelated constructions coexist. The generalized Tower of Hanoi graph $H(c+1)-H(c)=\pm 1$1 is recursively built from $H(c+1)-H(c)=\pm 1$2 copies of $H(c+1)-H(c)=\pm 1$3, beginning from a simplex, and is used in the study of the dimer-monomer model and entropy per site (Li et al., 2018). The state graphs $H(c+1)-H(c)=\pm 1$4 for the $H(c+1)-H(c)=\pm 1$5-peg Tower of Hanoi encode legal moves by words over an alphabet of peg labels and support graph-theoretic proofs of the Frame–Stewart recurrence for $H(c+1)-H(c)=\pm 1$6 (Žerovnik, 2016). The Path$H(c+1)-H(c)=\pm 1$7 graphs of the restricted Tower of Hanoi problem place pegs along a line so that moves are allowed only between nearest neighbors (Berend et al., 2011). A Fibonacci-variant Tower of Hanoi produces directed graphs $H(c+1)-H(c)=\pm 1$8 that are strongly connected and non-planar for $H(c+1)-H(c)=\pm 1$9 (Rittaud, 2022).

A second source of confusion is the use of “tower” for inverse systems or covering structures rather than for a single graph family. Examples include towers of connected regular Ramanujan graphs arising from modular curves (Sugiyama, 2019), branched $2$00-towers of finite connected graphs studied through the growth of spanning trees (Gambheera et al., 2024), and $2$01-towers of graph coverings arising from constant voltage assignments, with Iwasawa invariants determined by an explicit characteristic power series (Lei et al., 7 Jan 2025). These are towers of graphs indexed by level, not generalized tower graphs in either the dimer or commutative-algebra sense.

6. Mathematical significance

In planar dimer theory, generalized tower graphs provide a solvable family that combines explicit combinatorial construction, exact embedding technology, and asymptotic fluctuation theory. Because the graphs are parametrized by up-down paths approximating a profile function, they interpolate between rigid benchmark models and geometries with locally varying structure. The perfect $2$02-embedding formalism, the origami map, the rigidity estimates, and the contour-integral expressions for local statistics together yield a setting in which arctic geometry and Gaussian free field fluctuation behavior can be analyzed simultaneously (Keating et al., 23 Sep 2025).

In commutative algebra, generalized tower graphs isolate the exact combinatorial data needed to characterize height-$2$03 arithmetically Cohen–Macaulay squarefree monomial ideals. The significance is not merely classificatory. The decomposition $2$04 shows how a tower-like nested core can be enlarged by controlled “flat” edges without losing the Cohen–Macaulay property, and the graph interpretation converts questions about primary decomposition into questions about edge-set structure (Favacchio et al., 2014).

A common misconception is that the word “tower” identifies a unified graph class across disciplines. The literature instead uses it for several unrelated constructions: self-similar state graphs, covering towers, Bratteli-diagram towers, and the two notions of generalized tower graphs described above. The precise meaning is therefore determined by context. In current usage, the technically richest meanings are the planar bipartite square-hexagon family of the dimer model and the edge-set encoding of codimension-$2$05 arithmetically Cohen–Macaulay ideals.