Random-cluster interfaces: conformal shapes, natural length, and perturbations

← All digests

Family 223 · Probability and statistical mechanics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

Draw a large random network on a fine square grid, and trace the boundary between an open cluster and the surrounding closed connections. The microscopic path makes innumerable right-angle turns. Does its large-scale law remember the grid, or does it become invariant under smooth angle-preserving changes of coordinates?

The Fortuin–Kasteleyn random-cluster model, usually abbreviated FK, asks this question in one family containing independent percolation, Ising connectivity, and Potts connectivity. The predicted continuum objects are Schramm–Loewner evolution, or SLE, for one distinguished interface, and conformal loop ensembles, or CLE, for collections of loops.

Family 223 claims those limits across broad ranges of the cluster parameter. It also asks two more demanding questions. Does counting lattice steps converge to the continuum curve’s natural measure? And what survives when Ising couplings acquire weak fixed randomness or the temperature moves away from criticality on the scale of the mesh?

The six manuscripts address these questions with different hypotheses. A curve’s shape, its nesting among other loops, and its occupation measure are separate conclusions.

2. Background and essential definitions

A probability law for connected clusters

On a finite graph with edge set $E$, a configuration is a subset $A\subseteq E$ of open edges. Its FK probability is proportional to

$$p^{|A|}(1-p)^{|E|-|A|}q^{k_\xi(A)}.$$

Here $q>0$ rewards or penalizes the number of connected components, and $\xi$ specifies which boundary vertices are identified when counting them. Isolated vertices count as components. At $q=1$, the cluster factor disappears and edges are independent. At $q=2$, the model represents Ising spin connectivity.

A wired boundary arc is counted as a single connected block; a free arc has no such identifications. These bookkeeping connections must be distinguished from an actual open path using graph edges. Dobrushin boundary conditions put one wired and one free arc between two marked boundary points. A medial-lattice interface separates the primal open cluster from the dual cluster made by crossing closed primal edges.

The self-dual square-lattice parameter and predicted SLE parameter are

$$p_q=\frac{\sqrt q}{1+\sqrt q}, \qquad \kappa(q)=\frac{4\pi}{\arccos(-\sqrt q/2)}.$$

For $q\geq1$ up to four, the first quantity is the known critical edge probability. Below one, the corresponding manuscript states its theorem directly at the self-dual parameter without assuming an infinite-volume critical-point identification.

Curves and their continuum laws

Chordal SLE describes a random curve between two boundary points of a simply connected domain. In the upper half-plane, the conformal maps removing the growing curve satisfy

$$\partial_t g_t(z)=\frac{2}{g_t(z)-W_t}, \qquad W_t=\sqrt\kappa B_t,$$

where $B_t$ is standard Brownian motion. Conformal maps transport this law to other domains. The parameter $\kappa$ controls the curve’s roughness; for $\kappa\leq8$, its dimension is $d=1+\kappa/8$.

CLE describes a compatible random collection of loops, including loops nested inside others. Retaining traversal order matters: two curves can trace the same set while visiting its pieces in different orders. Likewise, a collection can contain two distinct loops with the same limiting image. The source statements retain this information.

For a fractal curve, ordinary arc length is generally infinite. Its $d$-dimensional Minkowski content instead rescales the area within distance $r$ of the trace by $r^{d-2}$, then sends $r$ to zero. This produces the natural occupation measure relevant to counting lattice steps.

3. Development history and previous partial results

Fortuin and Kasteleyn’s representation united several statistical-mechanical models through random connectivity. Exact solutions and Coulomb-gas calculations then predicted exponents and relations between cluster weights and continuum parameters. Schramm’s introduction of SLE turned a conformal-invariance prediction into a precise candidate law for interfaces.

Smirnov proved conformal invariance and Cardy’s crossing formula for critical triangular-lattice site percolation. FK–Ising on the square lattice became another major example through fermionic observables and the interface theorem of Chelkak, Duminil-Copin, Hongler, Kemppainen, and Smirnov. These achievements did not settle general square-lattice FK weights.

A substantial intermediate advance was rotational invariance for critical planar FK models by Duminil-Copin, Kozlowski, Krachun, Manolescu, and Oulamara. Rotation invariance is weaker than identifying the complete SLE or CLE law. M1 also acknowledges overlapping square-bond percolation claims by Tsai–Yam–Zhou and Zhou, with the latter’s stated boundary regularity and separation conditions. The present family should be read in that context.

For occupation measures, the dimension prediction was only a first step. Lawler and collaborators developed SLE natural parametrization and its identification with Minkowski content. Garban–Pete–Schramm and Holden–Li–Sun connected triangular-site percolation interface measures to that content; Lawler–Viklund did so for loop-erased random walk. These were important precedents, with different lattice models and normalizations.

Away from criticality, Makarov and Smirnov formulated the massive-SLE program. Park’s massive FK–Ising observable theorem supplied near-critical observables and crossing estimates. M4 claims the further steps needed to identify a unique interface law. Weak random bonds raise another difficulty: the Ising disorder is marginal, so microscopic randomness can decay only slowly under changes of scale.

4. What the new manuscripts claim

Critical curves and all nested loops

M1 claims that, for every fixed $1\leq q<4$, square-lattice Dobrushin interfaces converge to chordal $\mathrm{SLE}_{\kappa(q)}$. M6 includes the endpoint $q=4$. The domain may be any bounded Jordan domain, meaning its boundary is a simple closed curve, with lattice boundaries and marked points converging through uniform boundary parametrizations.

The convergence is of the entire oriented curve, modulo increasing changes of traversal speed. For the free infinite-volume lattice law, the complete nested loop collection converges to whole-plane $\mathrm{CLE}_{\kappa(q)}$, retaining multiplicities and traversals. A spherical metric controls the plane together with its point at infinity. These statements hold along every mesh sequence for each fixed parameter.

Thus $q=1$ gives $\mathrm{SLE}_6$, $q=2$ gives $\mathrm{SLE}_{16/3}$, and $q=4$ gives $\mathrm{SLE}_4$. The endpoint uses a separate field argument; it is not obtained by silently sending $q\uparrow4$ in estimates valid strictly below four.

M2 claims the Dobrushin curve limit for every fixed $0<q<1$, where $6<\kappa(q)<8$, in bounded smooth Jordan domains with its specified polygon approximations. It does not state the same general-Jordan or complete-loop theorem as M6.

Square-bond crossings and exponents

At $q=1$, M1 and M6 claim Cardy’s formula for actual open crossings with all boundary edges free and independently sampled. Map four successive boundary marks $a,b,c,d$ to $0,x,1,\infty$ in the upper half-plane. The limiting probability of connecting arc $[a,b]$ to arc $[c,d]$ is

$$\frac{\displaystyle\int_0^x[t(1-t)]^{-2/3}\,dt} {\displaystyle\int_0^1[t(1-t)]^{-2/3}\,dt}.$$

M1 additionally states square-bond exponent consequences. At criticality, the probability of an open connection from the origin to distance $R$ is $R^{-5/48+o(1)}$. For $2k$ disjoint, cyclically alternating primal-open and dual-open arms across an annulus with fixed sufficiently large inner radius, the exponent is $(4k^2-1)/12$.

Its near-critical consequences include an infinite-cluster density exponent $5/36$, a finite-cluster susceptibility exponent $43/18$, and a second-moment correlation-length exponent $4/3$. Susceptibility here counts only finite clusters. These stated exponent corollaries concern $q=1$, not every FK parameter or every choice of arm colors.

Natural occupation measure

M5 works in the unit square, with opposite marked corners and its precise wired/free drawing convention, for fixed $1\leq q<4$. Let $N_n$ be the number of full medial-edge traversals at mesh $1/n$, and $z_{n,j}$ their midpoints, projected into the closed square when necessary. It claims a constant $c(q)>0$ such that

$$\left(\eta_n,\;c(q)n^{-d}\sum_{j=1}^{N_n}\delta_{z_{n,j}}\right) \Longrightarrow (\eta,\mu_\eta), \qquad d=1+\frac{\kappa(q)}8.$$

Here $\delta_z$ is unit point mass, $\eta$ is SLE, and $\mu_\eta$ is the Minkowski-content measure of that same sample. In particular, $c(q)n^{-d}N_n$ converges to its total mass. Boundary traversals count; the two terminal half-edges do not. The constant depends on these units. A single constant times a power is stronger than knowing only the exponent, because it excludes an undetermined slowly varying correction.

Disorder and thermal perturbations at $q=2$

M3 takes independent fair signs $\xi_e$ and ferromagnetic couplings $J_e=1+\varepsilon\xi_e$. For sufficiently small fixed positive $\varepsilon$, it claims the critical FK interface still converges to $\mathrm{SLE}_{16/3}$. Criticality is the physical plus-state magnetization threshold; the FK edge probabilities are $1-e^{-2\beta_cJ_e}$.

The environment is fixed before sampling the interface. Its conditional curve law converges in probability over environments to the deterministic SLE law, in every deterministically approximated bounded Jordan domain of the stated class. This is stronger than averaging the curve over disorder, but does not claim almost-sure convergence along the full sequence. The companion weak-disorder critical-point and scale-transformation results discussed in family 218 are substantive inputs.

M4 instead changes the thermal parameter on the mesh scale, retaining a fixed nonzero continuum mass $m$. It treats angle-bounded isoradial lattices: faces lie on circles of the common mesh radius, and the associated rhombi stay uniformly nondegenerate. Its elliptic weights use a nome $r_\delta$ satisfying $r_\delta/\delta\to|m|/2$, with the negative branch defined by duality.

For every bounded simply connected marked limit domain, M4 claims a lattice-independent massive $\mathrm{SLE}_{16/3}$ limit. Domain convergence is measured conformally from an interior point; approximating domains may have unbounded diameters. Curves are compared after mapping them to the unit disk.

For positive mass, the driving Brownian motion keeps variance rate $16/3$ and acquires a drift determined by a boundary problem in the evolving physical domain:

$$\Delta h=-4m|\nabla h|,$$

with value one on the primal-wired bank and zero on the dual-wired bank. The manuscript claims uniqueness in a class where the drift has locally finite squared time integral. Negative mass follows by duality and reversal. The mass records a physical length scale, so this law is a deformation of critical SLE rather than ordinary conformal invariance at fixed mass.

5. Why the advance matters

Importance: 5/5. Identifying square-lattice FK interfaces throughout the claimed ranges would settle central conformal-limit predictions beyond the previously established special models. The complete nested ensemble would determine how interfaces fit together, not merely what one boundary looks like.

Natural occupation adds a quantitative geometric observable: the limiting amount of interface inside each region. The disorder and thermal results ask how robust the critical picture is and what replaces it when a macroscopic correlation scale remains.

Together these claims connect microscopic probabilities, continuum random curves, and measurable geometric quantities. Their breadth supports the highest relative rating, while the separate parameter ranges remain essential to interpreting that breadth.

6. The main idea

A central device is a boundary observable with four changes of boundary condition. Its limiting value depends on the conformal positions of four marked points. After part of an interface is revealed, conditional crossing probabilities become martingales. Two suitable tests identify the variance and drift of the limiting Loewner driver.

The difficult passage is from a formula on a fixed shape to one valid in a random slit domain. The sources develop physical crossing estimates, local comparisons, and control of narrow passages, preserving the difference between actual open paths and artificial boundary identifications. Analytic winding observables then supply the conformal information.

Below $q=1$, the usual positive-association inequality is unavailable. M2 builds restricted connection and gluing comparisons instead. At $q=4$, M6 uses a whole-plane height-field input, including the six-vertex field result discussed in family 225, and develops the additional finite-domain and conditional field arguments needed to identify interfaces.

For occupation, M5 must calibrate individual-edge visits against continuum neighborhood area. Quantitative comparisons between successive scales produce an actual amplitude, while boundary estimates prevent mass from disappearing near the square’s edges. For disorder, fixed-domain comparisons are transferred to adaptive explorations without assuming a uniform rate over every possible slit. For the massive law, energy estimates identify the drift and then support uniqueness.

7. Scope, limitations, and open questions

The family does not supply one theorem covering all combinations of lattice, cluster weight, disorder, and temperature. The broad critical curve theorem is square-lattice; the occupation theorem is restricted to a square and $q<4$; the below-one theorem assumes a smooth limiting boundary. Neither endpoint-uniform estimates nor a $q=4$ occupation theorem follow from these statements.

The random-bond result concerns weak symmetric two-valued disorder at $q=2$, with deterministic domain approximations. The thermal result concerns its prescribed isoradial near-critical weights and nonzero mass. It does not address arbitrary off-critical perturbations of every FK model.

Natural questions include extending occupation convergence to broader domains and parameters, treating other disorder laws, and obtaining useful rates of convergence. The manuscripts’ claims are reported as stated; the distinctions between curve, loop, and measure convergence are part of their mathematical scope.

8. Sources and further reading

The primary historical links above describe the Ising interface theorem, rotational invariance, earlier square-bond claims, and the massive observable input. For the continuum occupation concept, see Lawler and Rezaei, Minkowski content and natural parametrization for SLE.

M1 and M6 cover critical interfaces and nested loops, with M6 including four. M2 treats parameters below one. M3 treats fixed weak disorder, M4 thermal mass, and M5 natural occupation. These labels correspond to the six source manuscripts below.

Source manuscripts

• M1. OpenAI, Square-lattice FK interfaces and nested loops for 1 ≤ q < 4 (September 23, 2026).
• M2. OpenAI, Self-dual random-cluster interfaces below one (September 23, 2026).
• M3. OpenAI, Quenched SLE Universality for Weakly Disordered FK–Ising Interfaces (October 5, 2026).
• M4. OpenAI, Thermal FK–Ising interfaces and massive SLE (October 5, 2026).
• M5. OpenAI, Natural Occupation Measures for Critical Square-Lattice FK Interfaces (October 5, 2026).
• M6. OpenAI, Conformal Limits of Critical Square-Lattice Random-Cluster Interfaces (October 5, 2026).

Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.