OctoInformBeta

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

Lemma 3.2 [Stationary averaging]: Forward shift averages of integrable observables with L1L^1 stability, maximal bounds, boundary gaps under positive density, and a totalized first index

Strong law for stationary product shifts and sublinear boundary gaps.

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 study of ladder cuts on random planar maps, one evaluates real-valued observables along shifts of a two-sided sequence of independent and identically distributed (i.i.d.) random variables indexed by the integers ℤ\mathbb{Z} and tracks marked boundary indices in the nonnegative integers ℕ\mathbb{N}. For ergodic averages, the questions are whether forward shift averages converge almost surely for integrable observables of the product sequence, and whether L1L^1 approximation stability and maximal bounds hold for general measure-preserving transformations. Along a set of marked boundaries of positive limiting density, one further asks for the asymptotic relative gap between an index n∈ℕn \in \mathbb{N} and the last marked boundary up to nn, together with a totalized first index for subsets of ℕ\mathbb{N}.

Definitions

For a probability measure ν\nu on a measurable space SS, the i.i.d. sequence law ℙν\mathbb{P}_\nu on SℤS^{\mathbb{Z}} is the doubly infinite product measure (1) with marginal ν\nu at every coordinate: ℙν=ν⊗ℤ.\mathbb{P}_\nu = \nu^{\otimes \mathbb{Z}}.(1)✦

On the product space SℤS^{\mathbb{Z}} for a set SS, each a∈ℤa \in \mathbb{Z} defines the shift operator θa:Sℤ→Sℤ\theta_a : S^{\mathbb{Z}} \to S^{\mathbb{Z}}, the map acting on x=(xk)k∈ℤ∈Sℤx = (x_k)_{k \in \mathbb{Z}} \in S^{\mathbb{Z}} by (2): (θax)k=xk+a,k∈ℤ.(\theta_a x)_k = x_{k+a}, \qquad k \in \mathbb{Z}.(2)✦

Given a transformation T:Ω→ΩT : \Omega \to \Omega on a set Ω\Omega, a function f:Ω→ℝf : \Omega \to \mathbb{R}, and n∈ℕn \in \mathbb{N}, the stationary average An(T,f):Ω→ℝA_n(T, f) : \Omega \to \mathbb{R} is the function given by (3): An(T,f)(x)={0,n=0,1n∑i=0n−1f(Tix),n≥1,x∈Ω.A_n(T, f)(x) = \begin{cases} 0, & n = 0, \\ \dfrac{1}{n} \sum_{i=0}^{n-1} f(T^i x), & n \ge 1, \end{cases} \qquad x \in \Omega.(3)✦

For a subset C⊆ℕC \subseteq \mathbb{N} of marked boundaries and n∈ℕn \in \mathbb{N}, the marked boundary count NC(n)N_C(n) is the number of marked boundaries in CC before nn.✦ For a subset C⊆ℕC \subseteq \mathbb{N} and n∈ℕn \in \mathbb{N}, the last marked boundary LC(n)L_C(n) is the last index belonging to CC up to nn.✦

The totalized first index ι\iota on subsets of ℕ\mathbb{N} is the map (4) assigning to every subset P⊆ℕP \subseteq \mathbb{N} a natural number ι(P)∈ℕ\iota(P) \in \mathbb{N} that equals the ordinary least index belonging to PP whenever PP is nonempty: ι:𝒫(ℕ)→ℕ.\iota : \mathcal{P}(\mathbb{N}) \to \mathbb{N}.(4)✦

The Claim

Theorem 1.

  1. Let SS be a measurable space with probability measure ν\nu, and let f:Sℤ→ℝf : S^{\mathbb{Z}} \to \mathbb{R} be measurable and ℙν\mathbb{P}_\nu-integrable. Then for ℙν\mathbb{P}_\nu-almost every x∈Sℤx \in S^{\mathbb{Z}}, the forward shift average satisfies (5): limn→∞An(θ1,f)(x)=∫Sℤfdℙν.\lim_{n \to \infty} A_n(\theta_1, f)(x) = \int_{S^{\mathbb{Z}}} f \, d\mathbb{P}_\nu.(5)✦

  2. Let (Ω,μ)(\Omega, \mu) be a probability measure space, let T:Ω→ΩT : \Omega \to \Omega be a measurable, μ\mu-preserving transformation, let f:Ω→ℝf : \Omega \to \mathbb{R} and gk:Ω→ℝg_k : \Omega \to \mathbb{R} (k∈ℕk \in \mathbb{N}) be measurable and μ\mu-integrable, and let ak∈ℝa_k \in \mathbb{R} (k∈ℕk \in \mathbb{N}) and m∈ℝm \in \mathbb{R}. Suppose lim⁡k→∞∫Ω|f−gk|dμ=0\lim_{k \to \infty} \int_\Omega |f - g_k| \, d\mu = 0, lim⁡k→∞ak=m\lim_{k \to \infty} a_k = m, and for every k∈ℕk \in \mathbb{N}, lim⁡n→∞An(T,gk)(x)=ak\lim_{n \to \infty} A_n(T, g_k)(x) = a_k for μ\mu-almost every x∈Ωx \in \Omega. Then for μ\mu-almost every x∈Ωx \in \Omega the limit (6) holds: limn→∞An(T,f)(x)=m.\lim_{n \to \infty} A_n(T, f)(x) = m.(6)✦

  3. Let (Ω,μ)(\Omega, \mu) be a probability measure space, let T:Ω→ΩT : \Omega \to \Omega be a measurable, μ\mu-preserving transformation, let f:Ω→ℝf : \Omega \to \mathbb{R} be measurable and μ\mu-integrable, and let η>0\eta > 0. Then the maximal inequality (7) holds: μ({x∈Ω:∃n∈ℕ,|An(T,f)(x)|>η})≤1η∫Ω|f|dμ.\mu\bigl(\{x \in \Omega : \exists n \in \mathbb{N},\, |A_n(T, f)(x)| > \eta\}\bigr) \le \frac{1}{\eta} \int_\Omega |f| \, d\mu.(7)✦

  4. Let C⊆ℕC \subseteq \mathbb{N} be a subset of marked boundaries and let δ>0\delta > 0 satisfy lim⁡n→∞NC(n)/n=δ\lim_{n \to \infty} N_C(n) / n = \delta. Then the relative boundary gap satisfies (8): limn→∞max⁡(n−LC(n),0)n=0.\lim_{n \to \infty} \frac{\max(n - L_C(n), 0)}{n} = 0.(8)✦

In the paper, the result reads: for an i.i.d. SS-valued sequence Z=(Zi)i∈ℤZ = (Z_i)_{i \in \mathbb{Z}} with marginal ν\nu and translates Yi=f(θiZ)Y_i = f(\theta_i Z) of a measurable, ℙν\mathbb{P}_\nu-integrable function f:Sℤ→ℝf : S^{\mathbb{Z}} \to \mathbb{R}, the averages of YiY_i in each index direction converge almost surely to 𝔼[Y0]\mathbb{E}[Y_0], so that a shift-covariant set of boundaries having positive probability at boundary 00 has positive limiting density in both directions, and when 00 belongs to this set, as k→∞k \to \infty, the last such boundary before kk is k−o(k)k - o(k) and the first one after −k-k is −k+o(k)-k + o(k).

What Is Checked

The claims (5) through (4) are stated over the doubly infinite product law ℙν\mathbb{P}_\nu and unit shift θ1\theta_1 on SℤS^{\mathbb{Z}} for a probability measure ν\nu on a measurable space SS and a measurable, ℙν\mathbb{P}_\nu-integrable function f:Sℤ→ℝf : S^{\mathbb{Z}} \to \mathbb{R}; the stationary averages An(T,⋅)A_n(T, \cdot) under a measurable, μ\mu-preserving transformation TT on a probability measure space (Ω,μ)(\Omega, \mu) for measurable, μ\mu-integrable functions ff and (gk)k∈ℕ(g_k)_{k \in \mathbb{N}} with lim⁡k→∞∫Ω|f−gk|dμ=0\lim_{k \to \infty} \int_\Omega |f - g_k| \, d\mu = 0, real numbers (ak)k∈ℕ(a_k)_{k \in \mathbb{N}} and m∈ℝm \in \mathbb{R} satisfying lim⁡k→∞ak=m\lim_{k \to \infty} a_k = m and lim⁡n→∞An(T,gk)(x)=ak\lim_{n \to \infty} A_n(T, g_k)(x) = a_k for μ\mu-almost every x∈Ωx \in \Omega, and a threshold η>0\eta > 0; and, on ℕ\mathbb{N}, the marked boundary count NCN_C, last marked boundary LCL_C, limiting density δ>0\delta > 0 of NC(n)/nN_C(n) / n as n→∞n \to \infty, and totalized first index ι\iota for subsets C,P⊆ℕC, P \subseteq \mathbb{N}.