OctoInformBeta

OpenAI preprint, September 2026 · §6. From the block tree to the map · §6

§6.4 (proof of Theorem 1.1): assembly: Weak Gromov-Hausdorff-Prokhorov convergence for Potts parameter q>4q > 4 via deterministic comparison bounds and marked tree coding

Brownian continuum random tree limit of Fortuin-Kasteleyn maps above four.

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

Fix a Potts parameter q>4q > 4 and sample a rooted planar map MnM_n with n+1n + 1 edges together with a subset AA of its edges under the Fortuin-Kasteleyn (FK) random-cluster measure. Rescaling graph distances on MnM_n by c/n+1c / \sqrt{n + 1} for an appropriate constant c>0c > 0 and weighting each vertex by its normalized degree produces a sequence of random compact metric probability spaces. The question is whether these rescaled FK-decorated maps converge weakly in the Gromov-Hausdorff-Prokhorov (GHP) topology as n→∞n \to \infty to the Brownian continuum random tree (CRT).

Planar Maps and the Fortuin-kasteleyn Measure

For n∈ℕn \in \mathbb{N}, the dart set DnD_n of a map with n+1n + 1 edges is the set {0,…,2(n+1)−1}\{0, \dots, 2(n+1)-1\} indexing its 2(n+1)2(n+1) directed edges.✦ For a permutation π\pi on DnD_n, the orbit [d]π[d]_\pi of a dart d∈Dnd \in D_n is the equivalence class of darts belonging to the same cycle of π\pi✦, and the orbit set 𝒪π\mathcal{O}_\pi is the set of all orbits of π\pi.✦ Given permutations aa and ss on DnD_n and a subset A⊆𝒪aA \subseteq \mathcal{O}_a, the spanning subgraph G(a,s,A)G(a, s, A) is the simple graph on 𝒪s\mathcal{O}_s in which distinct orbits v,w∈𝒪sv, w \in \mathcal{O}_s are adjacent whenever some dart d∈Dnd \in D_n with [d]a∈A[d]_a \in A satisfies ([d]s,[a(d)]s)=(v,w)([d]_s, [a(d)]_s) = (v, w) or (w,v)(w, v).✦

A pair (a,s)(a, s) of permutations on DnD_n satisfies the planar map conditions if aa is a fixed-point-free involution, the all-edge spanning subgraph G(a,s,𝒪a)G(a, s, \mathcal{O}_a) is connected, and the numbers of orbits under ss and orbits under s∘as \circ a sum to n+3n + 3.✦ A labelled planar map (a,s)(a, s) of size nn is a pair of permutations on DnD_n satisfying the planar map conditions.✦ An isomorphism rr of rooted maps between two labelled planar maps (a,s)(a, s) and (a′,s′)(a', s') of size nn is a permutation of DnD_n fixing dart 00 (the root dart) such that a′∘r=r∘aa' \circ r = r \circ a and s′∘r=r∘ss' \circ r = r \circ s.✦ A rooted planar map MM of size nn is an equivalence class of labelled planar maps under the equivalence relation generated by root-preserving map isomorphisms.✦

Given a representative (a,s)(a, s) of a rooted planar map MM of size nn, the dart reversal involution aMa_M of MM is aa✦, the vertex rotation permutation sMs_M of MM is ss✦, the edge set E(M)E(M) of MM is the orbit set 𝒪aM\mathcal{O}_{a_M}✦, and the vertex set V(M)V(M) of MM is the orbit set 𝒪sM\mathcal{O}_{s_M}.✦ For an edge subset A⊆E(M)A \subseteq E(M), the spanning subgraph of MM induced by AA is GM(A)=G(aM,sM,A)G_M(A) = G(a_M, s_M, A).✦ A spanning tree TT of a rooted planar map MM is a subset of E(M)E(M) whose spanning subgraph GM(T)G_M(T) is connected and acyclic. A tree-decorated rooted map (M,T)(M, T) of size nn is a pair consisting of a rooted planar map MM of size nn and a spanning tree T⊆E(M)T \subseteq E(M).✦

For an edge subset A⊆E(M)A \subseteq E(M) on a rooted planar map MM of size nn, the cluster count k(M,A)k(M, A) is the number of connected components of GM(A)G_M(A).✦ At real Potts parameter qq, the FK-decorated map weight wq(M,A)w_q(M, A) of (M,A)(M, A) is given by (1): wq(M,A)=qk(M,A)+(|A|−|V(M)|)/2,w_q(M, A) = q^{k(M, A) + (|A| - |V(M)|)/2},(1)✦ and the finite FK partition function Zn(q)Z_n(q) at size nn sums (1) over all rooted planar maps MM of size nn and all subsets A⊆E(M)A \subseteq E(M).✦ The normalized vertex degree measure μM\mu_M on V(M)V(M) is the probability measure assigning mass 12(n+1)\frac{1}{2(n+1)} per dart d∈Dnd \in D_n to its vertex [d]sM[d]_{s_M}, giving each vertex v∈V(M)v \in V(M) with degree deg⁡M(v)=|v|\deg_M(v) = |v| mass deg⁡M(v)2(n+1)\frac{\deg_M(v)}{2(n+1)}.✦

Gromov-hausdorff-prokhorov Convergence and the Brownian Tree

A metric measure space presentation X=(𝒳,ρX,ℱX,νX)X = (\mathcal{X}, \rho_X, \mathcal{F}_X, \nu_X) consists of a carrier set 𝒳\mathcal{X}, a distance function ρX:𝒳×𝒳→ℝ\rho_X : \mathcal{X} \times \mathcal{X} \to \mathbb{R}, a σ\sigma-algebra ℱX\mathcal{F}_X on 𝒳\mathcal{X}, and a measure νX\nu_X on (𝒳,ℱX)(\mathcal{X}, \mathcal{F}_X).✦ A metric measure space presentation X=(𝒳,ρX,ℱX,νX)X = (\mathcal{X}, \rho_X, \mathcal{F}_X, \nu_X) is a compact metric probability space if ρX\rho_X is the distance of a compact metric space structure on 𝒳\mathcal{X}, ℱX\mathcal{F}_X is the Borel σ\sigma-algebra of that metric space, and νX(𝒳)=1\nu_X(\mathcal{X}) = 1.✦ Given two metric measure space presentations XX and YY, an isometric amalgamation pseudometric mm between XX and YY is a pseudometric on the disjoint union 𝒳⊔𝒴\mathcal{X} \sqcup \mathcal{Y} that agrees with ρX\rho_X on 𝒳\mathcal{X} and with ρY\rho_Y on 𝒴\mathcal{Y}.✦ For an isometric amalgamation pseudometric mm on the disjoint union 𝒳⊔𝒴\mathcal{X} \sqcup \mathcal{Y} of two presentations XX and YY, the GHP ambient embedding cost is the maximum of the Hausdorff distance between the embedded images of 𝒳\mathcal{X} and 𝒴\mathcal{Y} and the Lévy-Prokhorov distance between the pushforwards of νX\nu_X and νY\nu_Y under the canonical inclusions into (𝒳⊔𝒴,m)(\mathcal{X} \sqcup \mathcal{Y}, m) with its Borel σ\sigma-algebra.✦ The GHP distance dGHP(X,Y)\mathrm{d}_{\mathrm{GHP}}(X, Y) between two metric measure space presentations XX and YY is the infimum in [0,∞][0, \infty] of the GHP ambient embedding cost over all isometric amalgamation pseudometrics on 𝒳⊔𝒴\mathcal{X} \sqcup \mathcal{Y}.✦

For a rooted planar map MM of size nn and a scale factor λ∈ℝ\lambda \in \mathbb{R}, the rescaled map metric measure space ℳλ(M)\mathcal{M}_\lambda(M) is the presentation on V(M)V(M) with distance λdM(v,w)\lambda \, d_M(v, w), where dMd_M is the graph distance in GM(E(M))G_M(E(M)), discrete σ\sigma-algebra on V(M)V(M), and normalized vertex degree measure μM\mu_M.✦ A real-valued function FF on metric measure space presentations is a bounded continuous GHP test if FF is uniformly bounded on all presentations and, for every compact metric probability space XX and every ε>0\varepsilon > 0, there exists δ>0\delta > 0 such that |F(Y)−F(X)|<ε|F(Y) - F(X)| < \varepsilon whenever YY is a compact metric probability space with dGHP(X,Y)<δ\mathrm{d}_{\mathrm{GHP}}(X, Y) < \delta.✦ The finite FK map expectation 𝔼q,λ,nFK[F]\mathbb{E}_{q, \lambda, n}^{\mathrm{FK}}[F] of a bounded continuous GHP test FF at parameter q∈ℝq \in \mathbb{R}, scale λ∈ℝ\lambda \in \mathbb{R}, and size nn is defined by (2): 𝔼q,λ,nFK[F]=∑M∑A⊆E(M)wq(M,A)Zn(q)F(ℳλ(M)).\mathbb{E}_{q, \lambda, n}^{\mathrm{FK}}[F] = \sum_{M} \sum_{A \subseteq E(M)} \frac{w_q(M, A)}{Z_n(q)} \, F\bigl(\mathcal{M}_\lambda(M)\bigr).(2)✦

For a function e:[0,1]→ℝe : [0, 1] \to \mathbb{R}, the excursion tree pseudometric ded_e on [0,1][0, 1] is defined by (3): 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).(3)✦ For a function e:[0,1]→ℝe : [0, 1] \to \mathbb{R}, the excursion real tree 𝒯e\mathcal{T}_e coded by ee is the quotient of [0,1][0, 1] by the equivalence relation generated by de(s,t)=0d_e(s, t) = 0, with canonical excursion tree projection πe:[0,1]→𝒯e\pi_e : [0, 1] \to \mathcal{T}_e.✦ For a function e:[0,1]→ℝe : [0, 1] \to \mathbb{R}, the excursion tree metric measure space (𝒯e,de,μe)(\mathcal{T}_e, d_e, \mu_e) of the excursion real tree 𝒯e\mathcal{T}_e is the presentation on 𝒯e\mathcal{T}_e with the quotient distance induced by ded_e, the Borel σ\sigma-algebra of the quotient topology, and the pushforward μe\mu_e of Lebesgue measure on [0,1][0, 1] under πe\pi_e.✦ Given three independent Brownian motions B=(B0,B1,B2)B = (B_0, B_1, B_2) on a probability space (Ω,μ)(\Omega, \mu), the normalized Brownian excursion eω:[0,1]→ℝe_\omega : [0, 1] \to \mathbb{R} at ω∈Ω\omega \in \Omega is the three-dimensional Bessel bridge (4): eω(t)=∑i=02(Bi(t)(ω)−tBi(1)(ω))2,t∈[0,1].e_\omega(t) = \sqrt{\sum_{i=0}^{2} \bigl(B_i(t)(\omega) - t B_i(1)(\omega)\bigr)^2}, \qquad t \in [0, 1].(4)✦ For a bounded continuous GHP test FF and three independent Brownian motions BB on a probability space (Ω,μ)(\Omega, \mu), the Brownian CRT test expectation 𝔼μ[F(𝒯e,de,μe)]\mathbb{E}_\mu[F(\mathcal{T}_e, d_e, \mu_e)] is the integral ∫ΩF(𝒯eω,deω,μeω)dμ(ω)\int_\Omega F(\mathcal{T}_{e_\omega}, d_{e_\omega}, \mu_{e_\omega})\, d\mu(\omega).✦

The Inventory Bijection and Connected Cores

Given an involution aa on DnD_n and an edge-orbit subset A⊆𝒪aA \subseteq \mathcal{O}_a, the partial dart reversal a‾A\bar{a}_A is the permutation on DnD_n that applies aa to darts with [d]a∈A[d]_a \in A and fixes all other darts.✦ A canonical rooted tree map (a,s,A)(a, s, A) of size nn is a labelled planar map (a,s)(a, s) of size nn equipped with an edge-orbit subset A⊆𝒪aA \subseteq \mathcal{O}_a such that the tour permutation s∘a‾As \circ \bar{a}_A equals the standard cyclic permutation (01…2(n+1)−1)(0\; 1\; \dots\; 2(n+1)-1) on DnD_n.✦ Relabelling darts of a representative of a tree-decorated rooted map (M,T)(M, T) along its tour yields a unique canonical representative called its tree normal form✦, identifying tree-decorated rooted maps of size nn bijectively with canonical rooted tree maps of size nn.✦

An inventory FK model parameter tuple PP is a parameter specification fixing a probability p∈(1/2,1)p \in (1/2, 1), which determines a flexible order parameter tP=1+2p/(1−p)t_P = 1 + 2p/(1-p) and an inventory radius parameter uP=(1−p)/16u_P = (1-p)/16.✦ The inventory alphabet consists of two burger types, two fixed orders (one for each burger type), and one flexible order.✦ An inventory stack evaluation state is a pair consisting of a list of pending orders and a freshest-first stack of burgers✦; evaluating a word of inventory letters starts from empty pending orders and an empty burger stack and updates the state letter by letter, and a word of inventory letters is empty-reducing if this reduction ends with both lists empty.✦ Under parameter pp, the letter distribution assigns mass 1/41/4 to each burger type, p/2p/2 to the flexible order, and (1−p)/4(1-p)/4 to each fixed order.✦ Under the independent and identically distributed (i.i.d.) stream measure of this distribution on infinite letter sequences, p2mp_{2m} is the probability that the first 2m2m letters form an empty-reducing word✦, and the total empty-word probability τP\tau_P under PP is the sum ∑m≥0p2m\sum_{m \ge 0} p_{2m}.✦

A typed burger-order letter is a burger or an order carrying one of the two burger types.✦ Typed stack execution runs a word of typed letters against a freshest-first stack of burger types, where an order consumes the freshest eligible burger and fails on an illegal closure✦; a stack-emptying typed word is a typed word whose execution from an empty stack terminates successfully with an empty stack.✦ For a typed word ww, the flexible closing count f(w)f(w) is the number of flexible order closures encountered during typed stack execution from an empty stack✦, and the word monomial weight of ww at parameters (t,z)(t, z) is tf(w)z|w|t^{f(w)} z^{|w|}.✦ Given a canonical rooted tree map (a,s,A)(a, s, A) of size nn, its canonical tour matching word is the typed word assigning to each dart i∈Dni \in D_n a burger if i<a(i)i < a(i) and an order if a(i)<ia(i) < i, of the type indicating whether [i]a∈A[i]_a \in A.✦ A successful dart word of size nn is a stack-emptying typed word of length 2(n+1)2(n+1)✦, and the canonical tour matching word yields a bijection between tree-decorated rooted maps of size nn and successful dart words.✦

A non-crossing arch matching on a list of positions is a collection of disjoint pairs of positions that form nested non-crossing arches while possibly leaving unmatched elements skipped.✦ A two-stack matching certificate for a typed word decomposes its positions into two non-crossing arch matchings covering all positions, one pairing burgers with orders of the first type and the other pairing those of the second type✦, whose mate permutation is the product of all arch transpositions.✦ For a permutation π\pi on {0,…,K−1}\{0, \dots, K-1\}, the chord openers of π\pi are the indices ii such that i<π(i)i < \pi(i)✦, two openers ii and jj cross if i<j<π(i)<π(j)i < j < \pi(i) < \pi(j) or j<i<π(j)<π(i)j < i < \pi(j) < \pi(i)✦, and the chord intersection graph of π\pi is the simple graph on the chord openers in which two openers are adjacent whenever they cross.✦ A connected typed core cc is a typed word of length KK admitting a two-stack matching certificate whose mate permutation has an opener at index 00 from which every opener is reachable in the chord intersection graph. For a connected typed core cc of length KK, the core vertex rotation permutation is the cyclic shift (01…K−1)(0\; 1\; \dots\; K-1) composed with the transpositions of the first-type arches of the core stack matching✦, the isolated core map graph of cc is the simple graph on the cycle vertices of this permutation with an edge between distinct vertex cycles whenever the mate permutation maps a position in one cycle to the other✦, and the core corner metric increment δc(j)\delta_c(j) at corner j∈{0,…,K−1}j \in \{0, \dots, K-1\} is the graph distance in this isolated core graph from the vertex of the entry corner 00 to the vertex of the corner obtained by applying the cyclic shift (01…K−1)(0\; 1\; \dots\; K-1) to jj.✦

An optional core label is either empty, with arity 00, or a connected typed core cc, with arity |c||c|.✦ An ordered tree of cores TT is a rooted plane tree whose leaves carry the empty label and whose internal nodes carry a connected typed core cc together with |c||c| ordered child subtrees✦, whose label sequence lists these node labels in preorder✦ and whose core tree degree sequence lists their arities.✦ For an ordered tree of cores TT, the maximum core degree maxdeg⁡(T)\operatorname{maxdeg}(T) is the largest core length appearing in TT (00 if TT is a leaf)✦, and inserting child subwords into the corners of each core maps ordered trees of cores bijectively to stack-emptying typed words.✦

For m≥0m \ge 0 and t∈ℝt \in \mathbb{R}, the core generating polynomial coefficient cm(t)=∑wtf(w)c_m(t) = \sum_w t^{f(w)} sums over connected typed cores ww of length 2m2m✦, the block generating polynomial coefficient is bm(t)=𝟏m=0+cm(t)b_m(t) = \mathbf{1}_{m=0} + c_m(t)✦, and the actual FK block mass under PP is b̃m=bm(tP)(uPτP2)m/τP\tilde{b}_m = b_m(t_P)(u_P \tau_P^2)^m / \tau_P.✦ A critical even offspring distribution ν\nu is a probability mass function on ℕ\mathbb{N} supported on even integers with ν(0)>0\nu(0) > 0, ν(2)>0\nu(2) > 0, mean ∑k≥0kν(k)=1\sum_{k \ge 0} k \nu(k) = 1, and finite variance σ2=∑k≥0(k−1)2ν(k)\sigma^2 = \sum_{k \ge 0} (k-1)^2 \nu(k) (with standard deviation σ=σ2\sigma = \sqrt{\sigma^2}).✦ Doubling block offspring probability weights (wm)m≥0(w_m)_{m \ge 0} via k↦wk/2k \mapsto w_{k/2} on even kk and 00 on odd kk produces a critical even offspring distribution whenever w0>0w_0 > 0, w1>0w_1 > 0, ∑m≥0mwm=1/2\sum_{m \ge 0} m w_m = 1/2, and ∑m≥0m2wm<∞\sum_{m \ge 0} m^2 w_m < \infty✦; applying this construction to (b̃m)m≥0(\tilde{b}_m)_{m \ge 0} yields the critical FK corner offspring law νP\nu_P (with standard deviation σP\sigma_P).✦ The doubled block mass sequence of a sequence bb evaluates at kk to b(k/2)b(k/2) when kk is even, and to zero otherwise.✦

For an inventory FK model parameter tuple PP, the root core mark distribution κP\kappa_P on optional core labels is the probability distribution assigning weight 1/τP1/\tau_P to the empty label and tPf(c)(uPτP)|c|/τPt_P^{f(c)}(\sqrt{u_P}\,\tau_P)^{|c|}/\tau_P to each connected typed core cc.✦ For an inventory FK model parameter tuple PP and k∈ℕk \in \mathbb{N}, the conditional core mark distribution κP,k\kappa_{P, k} is the probability distribution on optional core labels of arity kk obtained by conditioning the root core mark distribution κP\kappa_P on labels of arity kk (taking the point mass at the empty label when no label has arity kk).✦

Marked Tree Codes, Height Paths, and Spine Sums

For a sequence d:{0,…,N−1}→ℕd : \{0, \dots, N-1\} \to \mathbb{N}, its total step is ∑i=0N−1(di−1)\sum_{i=0}^{N-1}(d_i - 1)✦ and its degree sequence partial walk at step jj is Wd(j)=∑i<j(di−1)W_d(j) = \sum_{i < j}(d_i - 1).✦ A discrete bridge sequence dd of length NN is a map from {0,…,N−1}\{0, \dots, N-1\} to {0,…,N−1}\{0, \dots, N-1\} with Wd(N)=−1W_d(N) = -1✦, and a Lukasiewicz tree code dd of length NN is a discrete bridge sequence satisfying Wd(j)≥0W_d(j) \ge 0 for all j<Nj < N (equivalently, the degree increment word (di−1)0≤i<N(d_i - 1)_{0 \le i < N} satisfies the discrete excursion condition of summing to −1-1 with all proper prefix sums nonnegative). A compatible marked tree code x=(d,ℓ)x = (d, \ell) of length NN is a pair consisting of a Lukasiewicz tree code dd of length NN and an assignment ℓ\ell of an optional core label ℓi\ell_i to each i∈{0,…,N−1}i \in \{0, \dots, N-1\} satisfying arity⁡(ℓi)=di\operatorname{arity}(\ell_i) = d_i.✦ For a compatible marked tree code x=(d,ℓ)x = (d, \ell) of length 2(n+1)+12(n+1)+1, the reconstructed core tree TxT_x is the ordered tree of cores reconstructed from ℓ\ell✦, and the associated rooted planar map MxM_x is the planar map underlying the tree-decorated rooted map of size nn obtained from TxT_x via the core-tree and rooted-map word bijections. A marked tree code sample x=(d,ℓ)x = (d, \ell) of size nn is a pair consisting of a Lukasiewicz tree code dd of length 2(n+1)+12(n+1)+1 and any assignment ℓ\ell of optional core labels on {0,…,2(n+1)}\{0, \dots, 2(n+1)\}.✦

Given a critical even offspring distribution ν\nu and m∈ℕm \in \mathbb{N}, conditioning the i.i.d. product measure ν2m+1\nu^{2m+1} on total step −1-1 produces the conditioned offspring bridge law✦, and pushing this law forward under cyclic rotation by the unique index achieving the minimum prefix sum to form a valid tree code defines the conditioned tree code law on Lukasiewicz tree codes of length 2m+12m+1.✦ Given a critical even offspring distribution ν\nu, m∈ℕm \in \mathbb{N}, and a mark kernel κ\kappa, the marked tree code law is the probability distribution on pairs (d,ℓ)(d, \ell) obtained by sampling a Lukasiewicz tree code dd of length 2m+12m+1 from the conditioned tree code law and assigning independent node marks ℓi\ell_i distributed according to κdi\kappa_{d_i} at each index ii.✦ For an inventory FK model parameter tuple PP and n∈ℕn \in \mathbb{N}, the law of marked code draw ℙP,n\mathbb{P}_{P, n} on marked tree code samples of size nn is the marked tree code law specialized to ν=νP\nu = \nu_P, κk=κP,k\kappa_k = \kappa_{P, k}, and m=n+1m = n+1.✦

For a degree sequence dd of length NN and a step j∈ℕj \in \mathbb{N}, the tree vertex height Hd(j)H_d(j) is the number of indices k<jk < j (with k<Nk < N) such that Wd(k)≤Wd(r)W_d(k) \le W_d(r) for every r∈{0,…,N}r \in \{0, \dots, N\} with k≤r≤jk \le r \le j.✦ Given a sequence (ai)0≤i≤N(a_i)_{0 \le i \le N} of real numbers, its piecewise-linear interpolation on [0,1][0, 1] is ∑i=0NaiϕN,i\sum_{i=0}^{N} a_i \, \phi_{N, i}✦, where ϕN,i(t)=max⁡(0,1−|Nt−i|)\phi_{N, i}(t) = \max(0, 1 - |Nt - i|) is the interpolation hat function.✦ For a critical even offspring distribution ν\nu with standard deviation σ\sigma and a Lukasiewicz tree code dd of length NN, the normalized preorder height path hν,d:[0,1]→ℝh_{\nu, d} : [0, 1] \to \mathbb{R} is the piecewise-linear interpolation of ai=σHd(i)2Na_i = \frac{\sigma H_d(i)}{2\sqrt{N}} for 0≤i≤N0 \le i \le N.✦ For an inventory FK model parameter tuple PP and a marked tree code sample x=(d,ℓ)x = (d, \ell) of size nn, the actual tree height path H̃P,n(x)\tilde{H}_{P, n}(x) is the rescaled path 1+12(n+1)hνP,d\sqrt{1 + \frac{1}{2(n+1)}}\, h_{\nu_P, d} on [0,1][0, 1].✦ For a nonnegative integer nn, the height scaling factor is 1+12(n+1)\sqrt{1 + \frac{1}{2(n+1)}}.✦

Given a Lukasiewicz tree code dd of length NN and a step v≤Nv \le N, reversing the prefix of length vv gives the stream (ym)m≥0(y_m)_{m \ge 0} with ym=dv−1−my_m = d_{v-1-m} for m<vm < v and ym=0y_m = 0 for m≥vm \ge v, whose partial walk is Sy(m)=∑k=0m−1(yk−1)S_y(m) = \sum_{k=0}^{m-1} (y_k - 1). The spine records of dd at step vv are the weak ascending record times s∈{1,…,v}s \in \{1, \dots, v\} where Sy(j)≤Sy(s)S_y(j) \le S_y(s) for all j≤sj \le s✦; each record ss corresponds to the spine vertex v−sv - s and carries the spine record type tp(d,v,s)=(ys−1,max⁡(0,Sy(s)−max⁡0≤j≤s−1Sy(j)))\mathrm{tp}(d, v, s) = (y_{s-1}, \max(0, S_y(s) - \max_{0 \le j \le s-1} S_y(j))). For a pair (k,j)∈ℕ×ℕ(k, j) \in \mathbb{N} \times \mathbb{N} and an optional core label ℓ\ell, the core label metric increment gcore((k,j),ℓ)g_{\mathrm{core}}((k, j), \ell) is the real number equal to δc(k−1−j)\delta_c(k - 1 - j) when ℓ\ell is a connected typed core cc with |c|=k|c| = k and j<kj < k, and equal to 00 otherwise.✦ For a Lukasiewicz tree code dd of length NN, an assignment ℓ\ell of optional core labels, and a step v≤Nv \le N, the marked spine sum Sv(d,ℓ)S_v(d, \ell) is defined by (5): Sv(d,ℓ)=∑sgcore(tp(d,v,s),ℓ(v−s)).S_v(d, \ell) = \sum_{s} g_{\mathrm{core}}\bigl(\mathrm{tp}(d, v, s),\, \ell(v - s)\bigr).(5)✦

For an infinite sequence X=(Xm)m≥0X = (X_m)_{m \ge 0} in ℕ\mathbb{N}, the offspring walk partial sum is UX(m)=∑k=0m−1(Xk−1)U_X(m) = \sum_{k=0}^{m-1} (X_k - 1)✦, and the ladder epoch condition holds at index mm if UX(m+1)≥0U_X(m+1) \ge 0 or if UX(r+1)<0U_X(r+1) < 0 for all rr and m=0m = 0.✦ For an infinite sequence X=(Xm)m≥0X = (X_m)_{m \ge 0} in ℕ\mathbb{N}, the first weak ascending ladder epoch L(X)L(X) is the smallest index mm satisfying the ladder epoch condition✦, the weak ladder overshoot of XX is max⁡(0,UX(L(X)+1))\max(0, U_X(L(X)+1))✦, and the spine ladder variable pair of XX is (XL(X),max⁡(0,UX(L(X)+1)))(X_{L(X)}, \max(0, U_X(L(X)+1))).✦ For a critical even offspring distribution ν\nu, the offspring spine ladder distribution ℙνsp\mathbb{P}_\nu^{\mathrm{sp}} on ℕ×ℕ\mathbb{N} \times \mathbb{N} is the pushforward of the infinite i.i.d. product measure ν∞\nu^\infty under the spine ladder variable pair.✦ For an inventory FK model parameter tuple PP, the mean core metric increment βP\beta_P is the integral of gcoreg_{\mathrm{core}} first against κP,z1\kappa_{P, z_1} and then against ℙνPsp(dz)\mathbb{P}_{\nu_P}^{\mathrm{sp}}(dz).✦

For an inventory FK model parameter tuple PP and n∈ℕn \in \mathbb{N}, the CRT scaling constant cPc_P is σP22βP\frac{\sigma_P}{2\sqrt{2}\,\beta_P}✦ and the edge metric rescaling factor sP,ns_{P, n} is cPn+1\frac{c_P}{\sqrt{n+1}}.✦ For an inventory FK model parameter tuple PP, n∈ℕn \in \mathbb{N}, and a∈ℝa \in \mathbb{R}, the bad metric event BadP,n,a\mathrm{Bad}_{P, n, a} is the set of all marked tree code samples x=(d,ℓ)x = (d, \ell) of size nn satisfying at least one of three conditions: (i) arity⁡(ℓi)≠di\operatorname{arity}(\ell_i) \ne d_i for some i∈{0,…,2(n+1)}i \in \{0, \dots, 2(n+1)\}; (ii) arity⁡(ℓi)=di\operatorname{arity}(\ell_i) = d_i for all ii and maxdeg⁡(Tx)>a2(n+1)+1\operatorname{maxdeg}(T_x) > a\sqrt{2(n+1)+1}; or (iii) |Sj(d,ℓ)−βPHd(j)|>a2(n+1)+1|S_j(d, \ell) - \beta_P H_d(j)| > a\sqrt{2(n+1)+1} for some j∈{0,…,2(n+1)}j \in \{0, \dots, 2(n+1)\}.✦ For a scale factor λ∈ℝ\lambda \in \mathbb{R}, size n∈ℕn \in \mathbb{N}, and bounded continuous GHP test FF, the marked tree map test observable Φλ,n,F\Phi_{\lambda, n, F} is the real-valued function on marked tree code samples x=(d,ℓ)x = (d, \ell) of size nn defined by Φλ,n,F(x)=F(ℳλ(Mx))\Phi_{\lambda, n, F}(x) = F(\mathcal{M}_\lambda(M_x)) if arity⁡(ℓi)=di\operatorname{arity}(\ell_i) = d_i for all ii, and Φλ,n,F(x)=0\Phi_{\lambda, n, F}(x) = 0 otherwise.✦

The Claim

Theorem 1.

  1. Let ν\nu be a critical even offspring distribution with standard deviation σ\sigma, let n∈ℕn \in \mathbb{N}, and let x=(d,ℓ)x = (d, \ell) be a compatible marked tree code of length 2(n+1)+12(n+1)+1 with reconstructed core tree TxT_x and rooted planar map MxM_x. Let β,λ,a,η,δ∈ℝ\beta, \lambda, a, \eta, \delta \in \mathbb{R} with λ>0\lambda > 0 and δ>0\delta > 0, and set h(s)=2λβ2(n+1)+1σhν,d(s)h(s) = \frac{2\lambda \beta \sqrt{2(n+1)+1}}{\sigma}\, h_{\nu, d}(s) for s∈[0,1]s \in [0, 1]. Suppose that |Sj(d,ℓ)−βHd(j)|≤a2(n+1)+1|S_j(d, \ell) - \beta H_d(j)| \le a\sqrt{2(n+1)+1} for every j∈{0,…,2(n+1)}j \in \{0, \dots, 2(n+1)\}, that |h(s)−h(u)|≤η|h(s) - h(u)| \le \eta whenever s,u∈[0,1]s, u \in [0, 1] satisfy |s−u|≤12(n+1)|s - u| \le \frac{1}{2(n+1)}, and that 4λmaxdeg⁡(Tx)+4λa2(n+1)+1+4η≤2δ4\lambda \operatorname{maxdeg}(T_x) + 4\lambda a \sqrt{2(n+1)+1} + 4\eta \le 2\delta. Then the GHP bound (6) holds: dGHP((𝒯h,dh,μh),ℳλ(Mx))≤δ.\mathrm{d}_{\mathrm{GHP}}\bigl((\mathcal{T}_h, d_h, \mu_h),\, \mathcal{M}_\lambda(M_x)\bigr) \le \delta.(6)✦

  2. Given FK model parameters PP, n∈ℕn \in \mathbb{N}, a compatible marked tree code x=(d,ℓ)x = (d, \ell) of length 2(n+1)+12(n+1)+1 with core tree TxT_x and rooted planar map MxM_x, a continuous function f:[0,1]→ℝf : [0, 1] \to \mathbb{R}, and real numbers a,r,η,δa, r, \eta, \delta with δ>0\delta > 0, suppose that |Sj(d,ℓ)−βPHd(j)|≤a2(n+1)+1|S_j(d, \ell) - \beta_P H_d(j)| \le a\sqrt{2(n+1)+1} for every j∈{0,…,2(n+1)}j \in \{0, \dots, 2(n+1)\}, that sup⁡t∈[0,1]|H̃P,n(x)(t)−f(t)|≤r\sup_{t \in [0,1]} |\tilde{H}_{P,n}(x)(t) - f(t)| \le r, that |f(s)−f(u)|≤η|f(s) - f(u)| \le \eta whenever s,u∈[0,1]s, u \in [0, 1] satisfy |s−u|≤12(n+1)|s - u| \le \frac{1}{2(n+1)}, and that 4sP,nmaxdeg⁡(Tx)+4(sP,na2(n+1)+1+r)+4η≤2δ4 s_{P,n} \operatorname{maxdeg}(T_x) + 4\bigl(s_{P,n} a \sqrt{2(n+1)+1} + r\bigr) + 4\eta \le 2\delta. Then the nearby path comparison (7) gives: dGHP((𝒯f,df,μf),ℳsP,n(Mx))≤δ.\mathrm{d}_{\mathrm{GHP}}\bigl((\mathcal{T}_f, d_f, \mu_f),\, \mathcal{M}_{s_{P,n}}(M_x)\bigr) \le \delta.(7)✦

  3. Given FK model parameters PP, a bounded continuous GHP test FF, a real number ε>0\varepsilon > 0, and a continuous function f:[0,1]→ℝf : [0, 1] \to \mathbb{R}, there exist r>0r > 0, a>0a > 0, and N∈ℕN \in \mathbb{N} such that for every n≥Nn \ge N and every marked tree code sample x∉BadP,n,ax \notin \mathrm{Bad}_{P, n, a} of size nn whose height path ex=H̃P,n(x)e_x = \tilde{H}_{P, n}(x) satisfies sup⁡t∈[0,1]|ex(t)−f(t)|<r\sup_{t \in [0,1]} |e_x(t) - f(t)| < r, the observable error satisfies (8): |ΦsP,n,n,F(x)−F(𝒯ex,dex,μex)|≤ε.\bigl|\Phi_{s_{P,n}, n, F}(x) - F(\mathcal{T}_{e_x}, d_{e_x}, \mu_{e_x})\bigr| \le \varepsilon.(8)✦

  4. Given FK model parameters PP, a probability space (Ω,μ)(\Omega, \mu) carrying three independent Brownian motions B=(B0,B1,B2)B = (B_0, B_1, B_2), and a bounded continuous GHP test FF, the marked map test expectations follow (9): limn→∞∫ΦsP,n,n,F(x)dℙP,n(x)=𝔼μ[F(𝒯e,de,μe)].\lim_{n \to \infty} \int \Phi_{s_{P,n}, n, F}(x)\, d\mathbb{P}_{P, n}(x) = \mathbb{E}_\mu\bigl[F(\mathcal{T}_e, d_e, \mu_e)\bigr].(9)✦

  5. For every Potts parameter q>4q > 4, there exists a scaling constant c>0c > 0 such that on every probability space (Ω,μ)(\Omega, \mu) carrying three independent Brownian motions B=(B0,B1,B2)B = (B_0, B_1, B_2) and for every bounded continuous GHP test FF, the finite FK map expectations converge according to (10): limn→∞𝔼q,c/n+1,nFK[F]=𝔼μ[F(𝒯e,de,μe)].\lim_{n \to \infty} \mathbb{E}_{q,\, c/\sqrt{n+1},\, n}^{\mathrm{FK}}[F] = \mathbb{E}_\mu\bigl[F(\mathcal{T}_e, d_e, \mu_e)\bigr].(10)✦

In the paper, the result reads: rescaled map metric comparison with the exploration walk excursion metric dfNd_{f_N}, correspondence RNR_N between map vertices and excursion tree points, coupling of measures via uniform time, and convergence to (𝒯e,de,μe)(\mathcal{T}_e, d_e, \mu_e), where H̃P,n\tilde{H}_{P, n} stands for fNf_N and dH̃P,nd_{\tilde{H}_{P, n}} for dfNd_{f_N}.

What Is Checked

The five claims are stated over rooted planar maps with n+1n+1 edges (as equivalence classes of fixed-point-free dart involutions and vertex rotations of genus zero), the FK cluster measure at Potts parameter q>4q > 4, GHP distance via isometric amalgamations, bounded continuous GHP test functionals on compact metric probability spaces, and the Brownian CRT constructed from the three-dimensional Bessel bridge of three independent Brownian motions. On the coding side, they are stated over critical even offspring distributions arising from FK block offspring weights, compatible marked tree codes and their ordered trees of connected typed cores, the law of marked code draw, normalized and actual tree height paths, marked spine record sums, the mean core metric increment, the edge metric rescaling factor, the marked tree map test observable, and the bad metric event.

For the underlying constructions, conjugating a labelled planar map by any dart permutation preserves the planar map conditions✦, and partially reversing darts on an edge-orbit subset yields an involution.✦ Numbering darts along a single-cycle tour from the root dart is bijective.✦

Every stack-emptying typed word admits disjoint non-crossing arch matchings for the two burger types covering all its positions.✦ For an ordered tree of cores, its core tree degree sequence satisfies the discrete excursion condition✦ and core tree reconstruction from a compatible marked code of length NN preserves the label length NN.✦ Every sequence in ℕ\mathbb{N} admits a time satisfying the ladder epoch condition.✦ Under any FK model parameters PP, the actual FK block masses are nonnegative at all sizes✦ and strictly positive at sizes 00 and 11.