OctoInformBeta

OpenAI preprint, September 2026 · §3. Ladder cuts and a coefficient lower bound · §3

Lemma 3.6 [Output frequencies]: For p∈(1/2,1)p \in (1/2, 1), flexible mass is positive, fixed masses agree, and cut fractions at 00 converge almost surely to 1/21/2 and output masses

Output order masses and uniform inventory count limits on cuts.

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

For Fortuin-Kasteleyn planar maps with Potts parameter q>4q > 4, set t=qt = \sqrt{q} and p=t/(t+2)∈(1/2,1)p = t / (t + 2) \in (1/2, 1). The inventory coding represents map pieces by words over an alphabet of two burger types and three order types (two fixed and one flexible), reduced via burger and order stacks. On the event of a left or right cut at the origin in a stationary bi-infinite sequence of random pieces, the question is what limiting proportions of burger types and order types survive at the bottom of long reduced intervals, and whether the flexible-order weight is positive with equal weights on the two fixed orders.

Inventory Evaluation and Piece Laws

Let ℬ={1,2}\mathcal{B} = \{1, 2\} denote the two burger types, let 𝒪={o1,o2,*}\mathcal{O} = \{o_1, o_2, \ast\} denote the inventory orders consisting of the fixed orders obo_b for b∈ℬb \in \mathcal{B} (with o1,o2o_1, o_2 for types 1,21, 2) and the flexible order *\ast✦, and let 𝒜\mathcal{A} be the alphabet consisting of a burger letter 𝖻b\mathsf{b}_b for each b∈ℬb \in \mathcal{B} and an order letter 𝗈o\mathsf{o}_o for each o∈𝒪o \in \mathcal{O}. An inventory stack evaluation state (O,B)(O, B) is a pair consisting of a list OO of pending orders in 𝒪\mathcal{O} and a list BB of unmatched burger types in ℬ\mathcal{B} stored freshest-first.✦

The evaluation of a finite word ww over 𝒜\mathcal{A} on an evaluation state (O,B)(O, B) is the evaluation state obtained by updating (O,B)(O, B) letter by letter from left to right via the single-step stack transition (returning (O,B)(O, B) on the empty word), in which each order consumes the freshest eligible burger in BB.✦ For a finite word ww over 𝒜\mathcal{A}, the word reduction red⁡(w)\operatorname{red}(w) is the evaluation state obtained by evaluating ww from empty pending orders and an empty burger stack.✦ For a finite list ss of length |s||s| and an integer k≥0k \ge 0, the bottom segment bot⁡k(s)\operatorname{bot}_k(s) is the suffix of ss obtained by dropping its first max⁡(|s|−k,0)\max(|s| - k, 0) entries (leaving the final kk entries, or all of ss when |s|≤k|s| \le k).✦

For p∈(1/2,1)p \in (1/2, 1), the letter mass function πp:𝒜→ℝ\pi_p \colon \mathcal{A} \to \mathbb{R} is the function assigning weight 1/41/4 to each burger letter, p/2p/2 to the flexible order letter 𝗈*\mathsf{o}_\ast, and (1−p)/4(1 - p)/4 to each fixed order letter 𝗈o1\mathsf{o}_{o_1} and 𝗈o2\mathsf{o}_{o_2}✦, and the letter distribution νp\nu_p is the probability distribution on 𝒜\mathcal{A} with weights given by πp\pi_p.✦ For p∈(1/2,1)p \in (1/2, 1) and any index set II, the independent and identically distributed (i.i.d.) inventory stream measure ℙp,I\mathbb{P}_{p,I} on 𝒜I\mathcal{A}^I is the infinite product measure of copies of νp\nu_p, with ℙp=ℙp,ℕ\mathbb{P}_p = \mathbb{P}_{p,\mathbb{N}}.✦ Given any probability measure ν\nu on a measurable space SS, the i.i.d. bi-infinite sequence law ν⊗ℤ\nu^{\otimes \mathbb{Z}} on SℤS^{\mathbb{Z}} is the infinite product measure of copies of ν\nu at every coordinate.✦

The letter increment Δ:𝒜→ℤ\Delta \colon \mathcal{A} \to \mathbb{Z} is the map assigning +1+1 to each burger letter and −1-1 to each order letter.✦ For a one-sided sequence x=(xi)i≥0x = (x_i)_{i \ge 0} in 𝒜ℕ\mathcal{A}^{\mathbb{N}} and n≥0n \ge 0, the letter increment sum Sn(x)S_n(x) is ∑i=0n−1Δ(xi)\sum_{i=0}^{n-1} \Delta(x_i)✦, and for an initial level h≥0h \ge 0 the downward inventory path Dh(n,x)D_h(n, x) is h+Sn+1(x)h + S_{n+1}(x).✦ For an integer-valued sequence Z=(Zn)n≥0Z = (Z_n)_{n \ge 0}, the first hitting time of zero τ0(Z)∈ℕ∪{∞}\tau_0(Z) \in \mathbb{N} \cup \{\infty\} is the least n≥0n \ge 0 with Zn=0Z_n = 0, or ∞\infty if no such nn exists.✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} and h≥0h \ge 0, the first downward hitting time τ↓,h(x)\tau_{\downarrow,h}(x) from level hh is τ0((Dh(n,x))n≥0)\tau_0\bigl((D_h(n, x))_{n \ge 0}\bigr).✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} and n≥0n \ge 0, the revealed word rev⁡n(x)\operatorname{rev}_n(x) of length nn is (xn−1,…,x0)(x_{n-1}, \dots, x_0) (empty for n=0n = 0)✦, and for a map T:𝒜ℕ→ℕ∪{∞}T \colon \mathcal{A}^{\mathbb{N}} \to \mathbb{N} \cup \{\infty\} the stopped word stop⁡T(x)\operatorname{stop}_T(x) is rev⁡m+1(x)\operatorname{rev}_{m+1}(x), where m=T(x)m = T(x) when T(x)<∞T(x) < \infty and m=0m = 0 otherwise.✦ For p∈(1/2,1)p \in (1/2, 1) and x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}}, the reversed right piece word wRrev(x)w_R^{\mathrm{rev}}(x) is stop⁡τ↓,1(x)\operatorname{stop}_{\tau_{\downarrow,1}}(x)✦, the reversed right piece law μR,prev\mu_{R,p}^{\mathrm{rev}} is the pushforward of ℙp\mathbb{P}_p under x↦wRrev(x)x \mapsto w_R^{\mathrm{rev}}(x)✦, and the right piece law μR,p\mu_{R,p} is the pushforward of μR,prev\mu_{R,p}^{\mathrm{rev}} under word reversal.✦

At step n≥0n \ge 0 for x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}}, the survivor indicator χn(x)\chi_n(x) is 11 if the burger stack of red⁡(rev⁡n+1(x))\operatorname{red}(\operatorname{rev}_{n+1}(x)) is empty and 00 otherwise✦, and the first survivor time σ(x)\sigma(x) is τ0((χn(x))n≥0)∈ℕ∪{∞}\tau_0\bigl((\chi_n(x))_{n \ge 0}\bigr) \in \mathbb{N} \cup \{\infty\}.✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}}, set s(x)=σ(x)s(x) = \sigma(x) if σ(x)<∞\sigma(x) < \infty and s(x)=0s(x) = 0 otherwise; the unmatched order count K(x)K(x) is 00 if σ(x)=∞\sigma(x) = \infty and otherwise equals the number of pending orders in red⁡(rev⁡s(x)+1(x))\operatorname{red}(\operatorname{rev}_{s(x)+1}(x))✦, the first survivor type β(x)∈ℬ\beta(x) \in \mathcal{B} is the freshest burger type in the burger stack of red⁡(rev⁡s(x)+1(x))\operatorname{red}(\operatorname{rev}_{s(x)+1}(x)) (defaulting to type 11 if that stack is empty)✦, and the first survivor data (K(x),β(x))(K(x), \beta(x)) is the pair of the unmatched order count and first survivor type.✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} and n≥0n \ge 0, the shifted stream θn(x)\theta_n(x) is defined by θn(x)k=xn+k\theta_n(x)_k = x_{n+k} for k≥0k \ge 0✦, and the first survivor tail θsurv(x)\theta_{\mathrm{surv}}(x) is θs(x)+1(x)\theta_{s(x)+1}(x).✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} and j≥0j \ge 0, the parsing step survivor data at step jj is the pair (K(θsurvj(x)),β(θsurvj(x)))\bigl(K(\theta_{\mathrm{surv}}^j(x)), \beta(\theta_{\mathrm{surv}}^j(x))\bigr), and the parsing step defect count Kj(x)K_j(x) is K(θsurvj(x))K(\theta_{\mathrm{surv}}^j(x)).✦

For a sequence X=(Xj)j≥0X = (X_j)_{j \ge 0} of nonnegative integer defect counts on a space Ω\Omega and a point y∈Ωy \in \Omega, the defect walk at step n≥0n \ge 0 is Wn(X,y)=∑j=0n−1(1−Xj(y)),W_n(X, y) = \sum_{j=0}^{n-1} \bigl(1 - X_j(y)\bigr),(1)✦ and the defect walk survival event at step nn is An(X)={y∈Ω:Wk(X,y)≤0 for all k≤n}A_n(X) = \{y \in \Omega : W_k(X, y) \le 0 \text{ for all } k \le n\}.✦ The defect stopping index τ(X,y)\tau(X, y) of (1) is the least n≥0n \ge 0 with y∉An(X)y \notin A_n(X) (the first step where Wn(X,y)>0W_n(X, y) > 0), or 00 if y∈Ak(X)y \in A_k(X) for all k≥0k \ge 0.✦ For x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} and 𝑲=(Kj)j≥0\mathbf{K} = (K_j)_{j \ge 0}, the left parsing step count NL(x)N_L(x) is τ(𝑲,x)\tau(\mathbf{K}, x).✦ At each parsing step j≥0j \ge 0 for x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}}, a finite word over 𝒜\mathcal{A} (the jj-th raw chunk of xx) reduces to an evaluation state with an assigned normal word over 𝒜\mathcal{A}; for n≥0n \ge 0, the short piece prefix Pshort(n,x)P_{\mathrm{short}}(n, x) is the empty word at n=0n = 0, and at step n+1n + 1 is obtained by prepending the nn-th normal word to Pshort(n,x)P_{\mathrm{short}}(n, x).✦ The short left piece word wL(x)w_L(x) of x∈𝒜ℕx \in \mathcal{A}^{\mathbb{N}} is Pshort(NL(x),x)P_{\mathrm{short}}(N_L(x), x)✦, and for p∈(1/2,1)p \in (1/2, 1) the shortened left piece law μL,p\mu_{L,p} is the pushforward of ℙp\mathbb{P}_p under x↦wL(x)x \mapsto w_L(x).✦

Cut Events, Outputs, and Counts

Given a sequence y=(yj)j≥0y = (y_j)_{j \ge 0} of finite words over 𝒜\mathcal{A} and n≥0n \ge 0, the concatenated piece prefix pref⁡n(y)\operatorname{pref}_n(y) is the empty word at n=0n = 0 and pref⁡n(y)yn\operatorname{pref}_n(y) y_n at n+1n + 1.✦ For a bi-infinite word sequence x=(xj)j∈ℤx = (x_j)_{j \in \mathbb{Z}}, an integer a∈ℤa \in \mathbb{Z}, and n≥0n \ge 0, the concatenated interval word wx(a,n)w_x(a, n) is pref⁡n((xa+j)j≥0)\operatorname{pref}_n\bigl((x_{a+j})_{j \ge 0}\bigr)✦, and its reduction red⁡(wx(a,n))=(Ored(x,a,n),Bred(x,a,n))\operatorname{red}(w_x(a, n)) = \bigl(O_{\mathrm{red}}(x, a, n), B_{\mathrm{red}}(x, a, n)\bigr) gives the reduced interval order stack Ored(x,a,n)O_{\mathrm{red}}(x, a, n) and reduced interval burger stack Bred(x,a,n)B_{\mathrm{red}}(x, a, n). A bi-infinite word sequence xx has a left cut at a∈ℤa \in \mathbb{Z}, written L(x,a)L(x, a), if Ored(x,a,n)O_{\mathrm{red}}(x, a, n) is empty for all n≥0n \ge 0✦, and has a right cut at b∈ℤb \in \mathbb{Z}, written R(x,b)R(x, b), if Bred(x,b−n,n)B_{\mathrm{red}}(x, b - n, n) is empty for all n≥0n \ge 0.✦

For a bi-infinite word sequence xx, integers h,k≥0h, k \ge 0, a burger type b∈ℬb \in \mathcal{B}, and an order o∈𝒪o \in \mathcal{O}, writing count⁡u(s)\operatorname{count}_u(s) for the number of entries equal to uu in a list ss, the bottom burger count Bb(x,h,k)B_b(x, h, k) and the terminal order count Co(x,h,k)C_o(x, h, k) are Bb(x,h,k)=count⁡b(bot⁡k(Bred(x,0,h))),B_b(x, h, k) = \operatorname{count}_b\bigl(\operatorname{bot}_k\bigl(B_{\mathrm{red}}(x, 0, h)\bigr)\bigr),(2)✦ Co(x,h,k)=count⁡o(bot⁡k(Ored(x,−h,h))).C_o(x, h, k) = \operatorname{count}_o\bigl(\operatorname{bot}_k\bigl(O_{\mathrm{red}}(x, -h, h)\bigr)\bigr).(3)✦

For a predicate CC on ℤ\mathbb{Z} and i∈ℤi \in \mathbb{Z}, the left cut distance ℓC(i)∈ℕ\ell_C(i) \in \mathbb{N} and right cut distance rC(i)∈ℕr_C(i) \in \mathbb{N} are the least k≥0k \ge 0 satisfying C(i−k)C(i - k) and C(i+1+k)C(i + 1 + k), respectively, whenever such a kk exists (and nonnegative integers otherwise). Given a predicate CC on ℤ\mathbb{Z}, a list family W(a,n)W(a, n) indexed by a∈ℤa \in \mathbb{Z} and n≥0n \ge 0, a default value dd, and i∈ℤi \in \mathbb{Z}, the cut-interval output out⁡(C,W,d,i)\operatorname{out}(C, W, d, i) is the entry at index ℓC(i)\ell_C(i) of W(i−ℓC(i),ℓC(i)+1+rC(i))W\bigl(i - \ell_C(i), \ell_C(i) + 1 + r_C(i)\bigr) (indexing from 00), or dd when that list has length at most ℓC(i)\ell_C(i).✦ For a bi-infinite word sequence xx and i∈ℤi \in \mathbb{Z}, the right-cut output order Rout(x,i)∈𝒪R_{\mathrm{out}}(x, i) \in \mathcal{O} is Rout(x,i)=out⁡(R(x,⋅),Ored(x),*,i).R_{\mathrm{out}}(x, i) = \operatorname{out}\bigl(R(x, \cdot), O_{\mathrm{red}}(x), \ast, i\bigr).(4)✦ For o∈𝒪o \in \mathcal{O} and a bi-infinite word sequence xx, the output order indicator 𝟏{Rout(x,0)=o}\mathbf{1}_{\{R_{\mathrm{out}}(x, 0) = o\}} is the indicator that equals 11 if Rout(x,0)=oR_{\mathrm{out}}(x, 0) = o and 00 otherwise.✦ For p∈(1/2,1)p \in (1/2, 1) and o∈𝒪o \in \mathcal{O}, the output order mass mp(o)m_p(o) is mp(o)=∫𝟏{Rout(x,0)=o}dμR,p⊗ℤ(x).m_p(o) = \int \mathbf{1}_{\{R_{\mathrm{out}}(x, 0) = o\}} \, d\mu_{R,p}^{\otimes \mathbb{Z}}(x).(5)✦

The Claim

Theorem 1. Let p∈(1/2,1)p \in (1/2, 1). For each burger type b∈ℬb \in \mathcal{B} and μL,p⊗ℤ\mu_{L,p}^{\otimes \mathbb{Z}}-almost every sequence xx satisfying L(x,0)L(x, 0), limk→∞suph≥k|Bb(x,h,k)k−12|=0,\lim_{k \to \infty} \sup_{h \ge k} \left| \frac{B_b(x, h, k)}{k} - \frac{1}{2} \right| = 0,(6)✦ and for each order o∈𝒪o \in \mathcal{O} and μR,p⊗ℤ\mu_{R,p}^{\otimes \mathbb{Z}}-almost every sequence xx satisfying R(x,0)R(x, 0), limk→∞suph≥k|Co(x,h,k)k−mp(o)|=0.\lim_{k \to \infty} \sup_{h \ge k} \left| \frac{C_o(x, h, k)}{k} - m_p(o) \right| = 0.(7)✦ The output order masses satisfy 0<mp(*),0 < m_p(\ast),(8)✦ mp(o1)=mp(o2),m_p(o_1) = m_p(o_2),(9)✦ and for each b∈ℬb \in \mathcal{B}, mp(ob)=1−mp(*)2.m_p(o_b) = \frac{1 - m_p(\ast)}{2}.(10)✦

In the paper, the result reads: almost surely, each burger type b∈ℬb \in \mathcal{B} has frequency 1/21/2 in the stationary L-output, the stationary R-output has almost-sure frequencies (1−mp(*))/2(1 - m_p(\ast))/2, (1−mp(*))/2(1 - m_p(\ast))/2, and mp(*)m_p(\ast) for the two fixed orders o1,o2o_1, o_2 and the flexible order *\ast, respectively, for a deterministic weight mp(*)>0m_p(\ast) > 0, and for a random sequence xx and each b∈ℬb \in \mathcal{B}, lim⁡k→∞sup⁡h≥k|Bb(x,h,k)/k−1/2|=0\lim_{k\to\infty} \sup_{h \ge k} |B_b(x, h, k)/k - 1/2| = 0 on the event L(x,0)L(x, 0) and lim⁡k→∞sup⁡h≥k|Cob(x,h,k)/k−(1−mp(*))/2|=0\lim_{k\to\infty} \sup_{h \ge k} |C_{o_b}(x, h, k)/k - (1 - m_p(\ast))/2| = 0 on the event R(x,0)R(x, 0) (where mp(*)m_p(\ast), Bb(x,h,k)B_b(x, h, k), Cob(x,h,k)C_{o_b}(x, h, k), L(x,0)L(x, 0), and R(x,0)R(x, 0) stand for the paper’s ρ\rho, Bb(h,k)B_b(h, k), Cb(h,k)C_b(h, k), 00 being an L-cut, and 00 being an R-cut).

What Is Checked

The claims are stated for any p∈(1/2,1)p \in (1/2, 1), each burger type b∈ℬb \in \mathcal{B}, and each order o∈𝒪o \in \mathcal{O} over the stationary bi-infinite sequence laws μL,p⊗ℤ\mu_{L,p}^{\otimes \mathbb{Z}} and μR,p⊗ℤ\mu_{R,p}^{\otimes \mathbb{Z}} built from the shortened left piece law μL,p\mu_{L,p} and the right piece law μR,p\mu_{R,p}, on the left-cut event L(x,0)L(x, 0) and right-cut event R(x,0)R(x, 0) for the bottom burger counts (2), the terminal order counts (3), and the output order mass (5) defined via (4).