Plaquette Random-Cluster Model

• The Plaquette Random-Cluster Model is a higher-dimensional extension of the Fortuin–Kasteleyn model that uses i-dimensional plaquettes with topological weights based on homology.
• It couples exactly to the (i-1)-dimensional q-state Potts lattice gauge theory, establishing rigorous Wilson loop identities and a higher-dimensional Edwards–Sokal correspondence.
• The model exhibits sharp self-dual homological phase transitions and duality relationships, offering deep insights into topological phases and statistical mechanics.

Searching arXiv for the cited PRCM papers and related random-cluster background. arXiv search: "Plaquette Random-Cluster Model" The plaquette random-cluster model (PRCM) is a higher-dimensional analogue of the Fortuin–Kasteleyn random-cluster model in which the random variables are $i$-dimensional plaquettes of a cubical complex, while the topological weight is carried by homology or cohomology rather than by connected components. On a finite cubical complex, the model assigns a probability to each $i$-dimensional percolation subcomplex $P$ that is proportional to a Bernoulli factor in the number of open plaquettes and a topological factor measuring $(i-1)$-dimensional homological complexity. In the prime-field case, the model couples exactly to $(i-1)$-dimensional $q$-state Potts lattice gauge theory, and recent work establishes exact Wilson-loop identities, duality, boundary-condition formalisms, infinite-volume limits, positive association, and a sharp self-dual homological phase transition on $2i$-dimensional tori (Duncan et al., 2022, Duncan et al., 2024).

1. Finite-volume definition and relation to the Fortuin–Kasteleyn model

Let $X$ be a finite $d$-dimensional cubical complex and $0Duncan et al., 2022), the $i$0-dimensional plaquette random-cluster model is the random $i$1-complex $i$2 containing the full $i$3-skeleton of $i$4, with probability measure

$i$5

where $i$6 is the normalizing constant, $i$7 is the number of $i$8-cells in $i$9, $P$0 is the total number of $P$1-cells of $P$2, and $P$3 is the $P$4-st Betti number over the coefficient field $P$5. Equivalently, for a configuration $P$6, one may write

$P$7

This is the direct higher-dimensional analogue of the classical random-cluster weight on a finite graph,

$P$8

where the topological statistic is the number of connected components $P$9 rather than a Betti number (Duminil-Copin et al., 2017). In this sense, PRCM replaces edges by plaquettes and connectivity by homology. The parameter $(i-1)$0 remains the occupation parameter, while $(i-1)$1 weights topological complexity through the factor $(i-1)$2 (Duncan et al., 2022).

A potential source of confusion is to treat the model as merely Bernoulli plaquette percolation. The defining weight shows that the random geometry is not determined solely by the number of open plaquettes: the homological term biases the ensemble toward or against configurations with large $(i-1)$3-dimensional homology, depending on $(i-1)$4.

2. Homology, cohomology, and the role of coefficients

The algebraic-topological content of the PRCM is not an auxiliary reformulation but part of the definition. The 2022 formulation uses Betti numbers over a specified field $(i-1)$5, and explicitly stresses that the coefficient field matters because homology depends on coefficients (Duncan et al., 2022). The 2024 treatment, "Some Properties of the Plaquette Random-Cluster Model" (Duncan et al., 2024), formulates the finite-volume model with coefficients in $(i-1)$6 as

$(i-1)$7

This cohomological form is connected to homology by the universal coefficient theorem. In particular, the paper records

$(i-1)$8

and notes that if $(i-1)$9 vanishes or is free, then

$(i-1)$0

This coefficient dependence is one of the main differences between graph random-cluster theory and its higher-dimensional extension. On graphs, the cluster-counting term is purely combinatorial. In the PRCM, the weight is sensitive to the algebraic topology of the sampled subcomplex, and torsion may enter through the cohomology groups when coefficients are taken in $(i-1)$1 (Duncan et al., 2024). A plausible implication is that higher-dimensional random-cluster phenomena cannot be fully characterized without specifying coefficients, even when the underlying cubical geometry is fixed.

3. Exact coupling to Potts lattice gauge theory

When $(i-1)$2 is a prime integer and $(i-1)$3, the plaquette random-cluster model couples exactly to the $(i-1)$4-dimensional $(i-1)$5-state Potts lattice gauge theory (Duncan et al., 2022). In that gauge theory, spins are assigned to $(i-1)$6-cells; a spin field is a cochain $(i-1)$7; and the Hamiltonian is

$(i-1)$8

where $(i-1)$9 is the coboundary operator and $q$0 is the Kronecker delta.

With the parameter identification $q$1, the joint coupling $q$2 on pairs $q$3 has two marginals. Summing over plaquette configurations yields the Potts lattice gauge theory marginal,

$q$4

while summing over spin fields yields the plaquette random-cluster marginal,

$q$5

This is a higher-dimensional Edwards–Sokal-type correspondence. The prime-field assumption is essential in the proof, because the argument uses linear-algebraic homology and cohomology over a field, and the random-cluster marginal must match the gauge-theory marginal exactly (Duncan et al., 2022). The significance of the correspondence is structural rather than merely formal: it places PRCM within the same representational framework that relates the ordinary random-cluster model to Potts spin systems, but with cocycles, coboundaries, and plaquette degrees of freedom replacing graph spins and bonds.

4. Wilson loop variables and null-homologous cycles

A central theorem of the 2022 paper identifies generalized Wilson loop observables with a null-homology event in the plaquette random-cluster representation (Duncan et al., 2022). Let $q$6 be an $q$7-cycle. Define $q$8 as the event that $q$9 is null-homologous in the plaquette random-cluster model,

$2i$0

and define the generalized Wilson loop variable by

$2i$1

where $2i$2 and $2i$3 is viewed as a $2i$4-th root of unity in $2i$5.

The exact relation proved is

$2i$6

with $2i$7, where $2i$8 is the Potts lattice gauge theory measure and $2i$9 is the plaquette random-cluster measure. More precisely, if $X$0 occurs, then $X$1; if $X$2 occurs, then $X$3 is uniformly distributed over $X$4 viewed in $X$5; hence

$X$6

and therefore

$X$7

The paper presents this as the first rigorous justification of the claim by Aizenman, Chayes, Chayes, Fröhlich, and Russo that Wilson loop observables are exactly tied to the event that a loop is bounded by a surface in an interacting plaquette system (Duncan et al., 2022). In PRCM language, the “loop bounded by a surface” interpretation is realized as the event that the relevant cycle vanishes in the sampled homology group. This removes the distinction between a heuristic gauge-theoretic picture and the topological event in the random-complex representation.

5. Duality and boundary conditions

The 2024 paper develops an exact duality theory for PRCM on cubical complexes in $X$8 (Duncan et al., 2024). Its central statement is that an $X$9-dimensional PRCM is dual to a $d$0-dimensional PRCM, with dual parameter

$d$1

For a box $d$2, if $d$3 is an $d$4-dimensional percolation subcomplex and $d$5 is the complementary dual $d$6-dimensional complex in the shifted dual lattice $d$7, then

$d$8

More generally, if $d$9 is a boundary condition for the $0

$0

The proof uses Alexander duality together with algebraic-topological descriptions of the relevant cohomology groups. A key formula is

$0

and

$0

In particular, there is a constant $0

$0

Boundary conditions are formulated topologically rather than graph-theoretically. For a box $i$00 and boundary condition $i$01, if $i$02 is the inclusion into the complex extended by external plaquettes, then

$i$03

A “wired at infinity” variant is defined using Borel–Moore homology,

$i$04

For free boundary conditions, $i$05 is surjective and one recovers the finite-volume PRCM. For wired boundary conditions on a box, adjoining $i$06 leads to the cohomological term $i$07 (Duncan et al., 2024).

These results establish that PRCM is self-dual in the correct dimension: when $i$08, duality maps the model back to one of the same plaquette dimension. That observation underlies the self-dual transition statement proved on $i$09-dimensional tori in the 2022 paper.

6. Infinite-volume limits, extremality, uniqueness, and positive association

On infinite volume, the free and wired weak limits on growing boxes $i$10 exist: $i$11 and, more generally, if a limit $i$12 exists for some boundary condition $i$13, then its dual limit exists as well and satisfies the corresponding duality relation (Duncan et al., 2024). Among all weak limits, the free and wired measures are extremal: $i$14

The same paper proves positive association, i.e. FKG-type monotonicity, by using the Mayer–Vietoris sequence for cohomology. In finite volume,

$i$15

for increasing events $i$16 and $i$17, and the same holds for infinite-volume limits when they exist (Duncan et al., 2024). This places PRCM in the same monotonicity class as the ordinary random-cluster model, although the proof is topological rather than graph-combinatorial.

For uniqueness, the authors define a pressure-like quantity $i$18 from the partition function, prove that it exists and is independent of exhaustion and boundary condition, and show that it is convex in $i$19. Theorem 23 states that if $i$20 is differentiable at $i$21, then there is a unique infinite-volume PRCM with parameters $i$22; equivalently, the free and wired infinite-volume measures coincide at $i$23 (Duncan et al., 2024). The paper therefore isolates a direct higher-dimensional analogue of the standard random-cluster uniqueness mechanism, with the exceptional set encoded by possible non-differentiability of the pressure-like function.

7. Self-dual homological phase transition and dynamical interpretation

On the $i$24-dimensional torus $i$25, the PRCM exhibits a sharp phase transition at the self-dual point

$i$26

in the sense of homological percolation (Duncan et al., 2022). The theorem is formulated through events $i$27 and $i$28, corresponding respectively to nontriviality and surjectivity of the induced homology map, and asserts

$i$29

as $i$30. Below the self-dual point, the induced homology map is typically trivial in the relevant sense; above it, giant homology classes appear, namely cycles that are nontrivial in the ambient torus (Duncan et al., 2022).

The topological interpretation is that one phase is dominated by local cycles, while the other contains giant cycles spanning the torus. Statistically, this is the higher-dimensional analogue of the sharp phase transition familiar from random-cluster and Potts models, but expressed through homology rather than connectivity (Duncan et al., 2022). In the classical planar random-cluster model on $i$31, the self-dual value is also the critical value for $i$32,

$i$33

though there the transition is formulated in terms of clusters and crossings rather than homology (Mukoseeva et al., 2018). This suggests a structural parallel between planar self-duality in graph models and homological self-duality in $i$34-dimensional PRCM.

Because PRCM couples to Potts lattice gauge theory, the transition also has an algorithmic interpretation. The 2022 paper states that below the transition the generalized Swendsen–Wang updates are effectively local, whereas above the transition giant cocycles or cycles appear and the dynamics becomes non-local (Duncan et al., 2022). The same homological event thus marks both a topological change in the sampled complexes and a qualitative change in Monte Carlo behavior.