OctoInformBeta

OpenAI preprint, September 2026 · §1. Introduction · §1

§1.1 (setting): the finite model and Brownian excursion tree: Rooted maps with Fortuin-Kasteleyn weights, Gromov-Hausdorff-Prokhorov test functions, and the continuum random tree metric space

Finite Fortuin-Kasteleyn maps and the Brownian excursion tree.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

In the Fortuin-Kasteleyn random cluster model with cluster weight q>4q > 4, the percolation interfaces form macroscopic loops, and the geometry of the underlying map, rescaled by its graph distance, converges in distribution under the Gromov-Hausdorff-Prokhorov topology to the Brownian continuum random tree.

Combinatorial Maps

For an integer n≥0n \ge 0, a planar map with n+1n+1 edges is described by 2(n+1)2(n+1) oriented half-edges called darts. The set Dartn\mathrm{Dart}_n of darts of size nn is Dartn={0,1,…,2n+1},\mathrm{Dart}_n = \{0, 1, \dots, 2n + 1\},(1)✦ indexed as in (1).

The permutation orbit space Orbit(s)\mathrm{Orbit}(s) of a permutation ss of Dartn\mathrm{Dart}_n is the quotient of Dartn\mathrm{Dart}_n by the cycle equivalence relation Orbit(s)=Dartn/∼s,\mathrm{Orbit}(s) = \mathrm{Dart}_n / \sim_s,(2)✦ where two darts are equivalent in (2) when they lie in the same cycle of ss.

Given a permutation ss of Dartn\mathrm{Dart}_n and a dart d∈Dartnd \in \mathrm{Dart}_n, the orbit orbit(s,d)\mathrm{orbit}(s, d) is the equivalence class orbit(s,d)=[d]s∈Orbit(s),\mathrm{orbit}(s, d) = [d]_s \in \mathrm{Orbit}(s),(3)✦ defined in (3).

With permutations aa and ss of Dartn\mathrm{Dart}_n and a subset of orbits A⊆Orbit(a)A \subseteq \mathrm{Orbit}(a), the spanning graph spanningGraph(a,s,A)\mathrm{spanningGraph}(a, s, A) is the simple graph with vertex set Orbit(s)\mathrm{Orbit}(s) whose edges connect distinct orbits linked by darts whose aa-orbit belongs to AA, v∼w⇔∃d∈Dartn,[d]a∈A,[d]s=v,[a(d)]s=w,v≠w,v \sim w \iff \exists d \in \mathrm{Dart}_n, \quad [d]_a \in A, \quad [d]_s = v, \quad [a(d)]_s = w, \quad v \ne w,(4)✦ as formulated in (4).

A pair of permutations (a,s)(a, s) of Dartn\mathrm{Dart}_n satisfies the planar map conditions if aa is a fixed-point-free involution, the spanning graph of all orbits is connected, and the Euler characteristic corresponds to the sphere: a2=id,(∀d∈Dartn,a(d)≠d),spanningGraph(a,s,Orbit(a)) is connected,|Orbit(s)|+|Orbit(s∘a)|=n+3,a^2 = \mathrm{id}, \quad (\forall d \in \mathrm{Dart}_n, \, a(d) \ne d), \quad \mathrm{spanningGraph}(a, s, \mathrm{Orbit}(a)) \text{ is connected}, \quad |\mathrm{Orbit}(s)| + |\mathrm{Orbit}(s \circ a)| = n + 3,(5)✦ prescribed in (5).

A labelled planar map MM of size nn is a pair of permutations (a,s)(a, s) of Dartn\mathrm{Dart}_n satisfying the planar map conditions (5), LabelledMapn={(a,s)∈Perm(Dartn)×Perm(Dartn):(a,s) satisfies the planar map conditions},\mathrm{LabelledMap}_n = \{(a, s) \in \mathrm{Perm}(\mathrm{Dart}_n) \times \mathrm{Perm}(\mathrm{Dart}_n) : (a, s) \text{ satisfies the planar map conditions}\},(6)✦ stated in (6).

Two labelled planar maps M=(aM,sM)M = (a_M, s_M) and N=(aN,sN)N = (a_N, s_N) of size nn are isomorphic, written M∼isoNM \sim_{\mathrm{iso}} N, if there is a permutation rr of Dartn\mathrm{Dart}_n fixing the root dart 00 that conjugates their permutations: r(0)=0,aN∘r=r∘aM,sN∘r=r∘sM,r(0) = 0, \quad a_N \circ r = r \circ a_M, \quad s_N \circ r = r \circ s_M,(7)✦ written as in (7).

A rooted planar map MM of size nn is an equivalence class of labelled planar maps under rooted map isomorphism, RootedMapn=LabelledMapn/∼iso,\mathrm{RootedMap}_n = \mathrm{LabelledMap}_n / \sim_{\mathrm{iso}},(8)✦ given in (8).

The dart reversal involution αM\alpha_M of a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n is the dart involution aMa_M of a representative labelled map (aM,sM)∈M(a_M, s_M) \in M, αM=aM,\alpha_M = a_M,(9)✦ in (9).

The vertex rotation permutation σM\sigma_M of a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n is the rotation permutation sMs_M of a representative labelled map (aM,sM)∈M(a_M, s_M) \in M, σM=sM,\sigma_M = s_M,(10)✦ in (10).

The vertices V(M)V(M) of a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n form the orbit space of its vertex rotation permutation, V(M)=Orbit(σM),V(M) = \mathrm{Orbit}(\sigma_M),(11)✦ in (11).

The edge set E(M)E(M) of a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n is the orbit space of its dart reversal involution αM\alpha_M, E(M)=Orbit(αM),E(M) = \mathrm{Orbit}(\alpha_M),(12)✦ in (12).

For a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n and an edge subset A⊆E(M)A \subseteq E(M), the spanning graph G(M,A)G(M, A) is the simple graph on vertex set V(M)V(M) induced by the edges in AA, G(M,A)=spanningGraph(αM,σM,A),G(M, A) = \mathrm{spanningGraph}(\alpha_M, \sigma_M, A),(13)✦ as defined in (13).

The Fortuin-kasteleyn Measure

The cluster count kM(A)k_M(A) of an edge subset A⊆E(M)A \subseteq E(M) in a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n is the number of connected components of the spanning graph G(M,A)G(M, A), kM(A)=|π0(G(M,A))|,k_M(A) = |\pi_0(G(M, A))|,(14)✦ in (14).

Given a Potts parameter q>0q > 0, a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n, and an edge subset A⊆E(M)A \subseteq E(M), the Fortuin-Kasteleyn weight wq(M,A)w_q(M, A) is the positive real number wq(M,A)=qkM(A)+|A|−|V(M)|2,w_q(M, A) = q^{k_M(A) + \frac{|A| - |V(M)|}{2}},(15)✦ defined in (15).

The Fortuin-Kasteleyn partition function Zn(q)Z_n(q) for Potts parameter q>0q > 0 and map size n≥0n \ge 0 is the sum of weights over all rooted maps of size nn and all edge subsets, Zn(q)=∑M∈RootedMapn∑A⊆E(M)wq(M,A),Z_n(q) = \sum_{M \in \mathrm{RootedMap}_n} \sum_{A \subseteq E(M)} w_q(M, A),(16)✦ given in (16).

Metric Measure Spaces and Test Functions

A metric measure space 𝒳\mathcal{X} is a quadruple (X,dX,ℬX,μX)(X, d_X, \mathcal{B}_X, \mu_X) consisting of a set XX, a distance function dX:X×X→ℝd_X \colon X \times X \to \mathbb{R}, a σ\sigma-algebra ℬX\mathcal{B}_X on XX, and a measure μX\mu_X on (X,ℬX)(X, \mathcal{B}_X), 𝒳=(X,dX,ℬX,μX),\mathcal{X} = (X, d_X, \mathcal{B}_X, \mu_X),(17)✦ as in (17).

A metric measure space 𝒳=(X,dX,ℬX,μX)\mathcal{X} = (X, d_X, \mathcal{B}_X, \mu_X) is valid if (X,dX)(X, d_X) is a compact metric space, ℬX\mathcal{B}_X is its Borel σ\sigma-algebra, and μX\mu_X is a Borel probability measure, (X,dX) is compact metric,ℬX=ℬ(X),μX(X)=1,(X, d_X) \text{ is compact metric}, \quad \mathcal{B}_X = \mathcal{B}(X), \quad \mu_X(X) = 1,(18)✦ in (18).

For a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n, the normalized degree measure μM\mu_M on V(M)V(M) is the probability measure that assigns to each vertex a mass proportional to its degree, μM=12(n+1)∑d∈Dartnδ[d]σM,\mu_M = \frac{1}{2(n + 1)} \sum_{d \in \mathrm{Dart}_n} \delta_{[d]_{\sigma_M}},(19)✦ in (19), so that μM({v})=deg⁡M(v)/(2(n+1))\mu_M(\{v\}) = \deg_M(v) / (2(n + 1)).

Given a scale factor a>0a > 0 and a rooted map M∈RootedMapnM \in \mathrm{RootedMap}_n, the rescaled map metric measure space 𝒳a(M)\mathcal{X}_a(M) is the metric measure space with vertices V(M)V(M), graph distance scaled by aa, discrete σ\sigma-algebra, and normalized degree measure, 𝒳a(M)=(V(M),a⋅dG(M),𝒫(V(M)),μM),\mathcal{X}_a(M) = (V(M), a \cdot d_{G(M)}, \mathcal{P}(V(M)), \mu_M),(20)✦ in (20), where dG(M)d_{G(M)} denotes the shortest-path metric in G(M,E(M))G(M, E(M)).

For two metric measure spaces 𝒳=(X,dX,ℬX,μX)\mathcal{X} = (X, d_X, \mathcal{B}_X, \mu_X) and 𝒴=(Y,dY,ℬY,μY)\mathcal{Y} = (Y, d_Y, \mathcal{B}_Y, \mu_Y), a pseudometric dd on the disjoint union X⊔YX \sqcup Y amalgamates 𝒳\mathcal{X} and 𝒴\mathcal{Y} if it restricts to dXd_X on XX and to dYd_Y on YY, d(x,x′)=dX(x,x′)andd(y,y′)=dY(y,y′)∀x,x′∈X,y,y′∈Y,d(x, x') = d_X(x, x') \quad \text{and} \quad d(y, y') = d_Y(y, y') \quad \forall x, x' \in X, \; y, y' \in Y,(21)✦ as formulated in (21).

The ambient cost cost(𝒳,𝒴;d)\mathrm{cost}(\mathcal{X}, \mathcal{Y}; d) between two metric measure spaces 𝒳\mathcal{X} and 𝒴\mathcal{Y} with respect to an amalgamation pseudometric dd on X⊔YX \sqcup Y is the maximum of the Hausdorff distance between the canonical embeddings ιX:X→X⊔Y\iota_X \colon X \to X \sqcup Y and ιY:Y→X⊔Y\iota_Y \colon Y \to X \sqcup Y and the Levy-Prokhorov distance between the pushforward measures, cost(𝒳,𝒴;d)=max⁡(dH(X,Y),dLP((ιX)*μX,(ιY)*μY)),\mathrm{cost}(\mathcal{X}, \mathcal{Y}; d) = \max \bigl( d_{\mathrm{H}}(X, Y), \; d_{\mathrm{LP}}((\iota_X)_* \mu_X, (\iota_Y)_* \mu_Y) \bigr),(22)✦ in (22).

The Gromov-Hausdorff-Prokhorov distance dGHP(𝒳,𝒴)d_{\mathrm{GHP}}(\mathcal{X}, \mathcal{Y}) between two metric measure spaces is the infimum of the ambient cost over all isometric amalgamations, dGHP(𝒳,𝒴)=infd amalgamates 𝒳,𝒴cost(𝒳,𝒴;d),d_{\mathrm{GHP}}(\mathcal{X}, \mathcal{Y}) = \inf_{d \text{ amalgamates } \mathcal{X}, \mathcal{Y}} \mathrm{cost}(\mathcal{X}, \mathcal{Y}; d),(23)✦ as defined in (23).

A functional FF on metric measure spaces is a Gromov-Hausdorff-Prokhorov test function if it is bounded and continuous with respect to the Gromov-Hausdorff-Prokhorov distance on valid spaces, sup𝒳|F(𝒳)|<∞and∀valid 𝒳,∀ε>0,∃δ>0,∀valid 𝒴,dGHP(𝒳,𝒴)<δ⟹|F(𝒴)−F(𝒳)|<ε,\sup_{\mathcal{X}} |F(\mathcal{X})| < \infty \quad \text{and} \quad \forall \text{valid } \mathcal{X}, \, \forall \varepsilon > 0, \, \exists \delta > 0, \, \forall \text{valid } \mathcal{Y}, \, d_{\mathrm{GHP}}(\mathcal{X}, \mathcal{Y}) < \delta \implies |F(\mathcal{Y}) - F(\mathcal{X})| < \varepsilon,(24)✦ in (24).

The Brownian Continuum Random Tree

For a continuous function e:[0,1]→[0,∞)e \colon [0, 1] \to [0, \infty), the excursion pseudometric ded_e on the unit interval [0,1][0, 1] is de(s,t)=e(s)+e(t)−2infr∈[s∧t,s∨t]e(r),d_e(s, t) = e(s) + e(t) - 2 \inf_{r \in [s \wedge t, s \vee t]} e(r),(25)✦ in (25).

The excursion tree 𝒯e\mathcal{T}_e associated with ee is the quotient metric space of [0,1][0, 1] by the pseudometric ded_e, 𝒯e=[0,1]/{de=0},\mathcal{T}_e = [0, 1] / \{d_e = 0\},(26)✦ in (26).

The canonical projection pe:[0,1]→𝒯ep_e \colon [0, 1] \to \mathcal{T}_e of the excursion tree 𝒯e\mathcal{T}_e sends each time parameter t∈[0,1]t \in [0, 1] to its equivalence class under ded_e, pe(t)=[t]de,p_e(t) = [t]_{d_e},(27)✦ as in (27).

For a continuous function e:[0,1]→[0,∞)e \colon [0, 1] \to [0, \infty), the continuum random tree metric measure space 𝒳(e)\mathcal{X}(e) is the space with underlying set 𝒯e\mathcal{T}_e, metric ded_e, Borel σ\sigma-algebra, and the pushforward measure μe=(pe)*λ\mu_e = (p_e)_* \lambda of the Lebesgue measure λ\lambda on [0,1][0, 1], 𝒳(e)=(𝒯e,de,ℬ(𝒯e),(pe)*λ),\mathcal{X}(e) = (\mathcal{T}_e, d_e, \mathcal{B}(\mathcal{T}_e), (p_e)_* \lambda),(28)✦ in (28).

For a three-dimensional standard Brownian motion B=(B1,B2,B3)B = (B_1, B_2, B_3) on a probability space (Ω,ℱ,ℙ)(\Omega, \mathcal{F}, \mathbb{P}), the standard normalized Brownian excursion eB(ω):[0,1]→ℝe_B(\omega) \colon [0, 1] \to \mathbb{R} is the norm of the three-dimensional Bessel bridge, eB(ω)(t)=∑i=13(Bi(t,ω)−tBi(1,ω))2,e_B(\omega)(t) = \sqrt{\sum_{i=1}^3 \bigl(B_i(t, \omega) - t B_i(1, \omega)\bigr)^2},(29)✦ given in (29).

The Definition

Definition 1. Let q>0q > 0 be the Potts parameter, a>0a > 0 a scale factor, n≥0n \ge 0 an edge parameter, and FF a test functional on metric measure spaces. The finite Fortuin-Kasteleyn expectation 𝔼n,q,a[F]\mathbb{E}_{n, q, a}[F] of FF is 𝔼n,q,a[F]=∑M∈RootedMapn∑A⊆E(M)wq(M,A)Zn(q)F(𝒳a(M)),\mathbb{E}_{n, q, a}[F] = \sum_{M \in \mathrm{RootedMap}_n} \sum_{A \subseteq E(M)} \frac{w_q(M, A)}{Z_n(q)} F(\mathcal{X}_a(M)),(30)✦ as in (30), and for a probability space (Ω,ℱ,ℙ)(\Omega, \mathcal{F}, \mathbb{P}) carrying a three-dimensional standard Brownian motion BB, the Brownian continuum random tree expectation 𝔼ℙ,B[F]\mathbb{E}_{\mathbb{P}, B}[F] of FF is 𝔼ℙ,B[F]=∫ΩF(𝒳(eB(ω)))dℙ(ω),\mathbb{E}_{\mathbb{P}, B}[F] = \int_\Omega F(\mathcal{X}(e_B(\omega))) \, d\mathbb{P}(\omega),(31)✦ as in (31).

In the paper, the result reads: for q>4q > 4 and n≥1n \ge 1, a rooted planar map with edge subset (Mn,An)(M_n, A_n) of size |E(Mn)|=n|E(M_n)| = n is sampled with probability proportional to qkM(A)+(|A|−|V(M)|)/2=qℓ(M,A)/2q^{k_M(A) + (|A| - |V(M)|)/2} = q^{\ell(M, A)/2}, where kM(A)k_M(A) is the number of connected components of (V(M),A)(V(M), A) and ℓ(M,A)\ell(M, A) is the number of Fortuin-Kasteleyn interface loops, graph distance dnd_n uses all edges, normalized degree measure is μn({v})=deg⁡Mn(v)/(2n)\mu_n(\{v\}) = \deg_{M_n}(v)/(2n), and the standard normalized Brownian excursion ee defines the continuum random tree (𝒯e,de,μe)(\mathcal{T}_e, d_e, \mu_e) via de(s,t)=e(s)+e(t)−2min⁡r∈[s∧t,s∨t]e(r)d_e(s, t) = e(s) + e(t) - 2 \min_{r \in [s \wedge t, s \vee t]} e(r).

What Is Checked

The definitions are stated over rooted planar maps with n+1n+1 edges represented by orientation-preserving rotation systems on 2(n+1)2(n+1) darts, Fortuin-Kasteleyn edge weights with parameter q>0q > 0, valid compact metric measure spaces equipped with their normalized degree measures, and the continuum random tree of a standard Brownian excursion formed from a three-dimensional Bessel bridge.