OctoInform Beta

OpenAI preprint, September 2026 · §3. Ordered arrays and conditional locality · §3.2

§3.2 (setting): the data fields 𝒫_n, 𝒪_{n,j}, 𝒢_{n,j}: The σ\sigma-algebras generated by the parameter array, by the last-axis planes other than jj, or by both, and the orthogonal projections onto them

Data fields of an array object and their projections.

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 finite index set II, say with three elements. A random 3×33 \times 3 matrix of points of a space YY, each point carrying a parameter in a second space UU, is a random array of order two; its columns are arrays of order one, and an array of any order n+1n+1 is in the same way a stack of |I||I| arrays of order nn. The data fields are the σ\sigma-algebras one conditions on when one column, or one plane in higher order, is set against the rest of the array and its parameters.

Arrays and Array Objects

An index tuple of length nn over II is a word v=(v0,…,vn−1)v = (v_0, \dots, v_{n-1}) of length nn in the letters of II, a point of InI^n.✦ An array of order nn over a space YY is a map y:In→Yy \colon I^n \to Y, a point of the product space YInY^{I^n}.✦ The coordinatewise map fˆ\hat f of a map f:Y→Zf \colon Y \to Z applies ff to every entry: fˆ:YIn→ZIn,fˆ(y)(v)=f(y(v)).\hat f \colon Y^{I^n} \to Z^{I^n}, \qquad \hat f(y)(v) = f\bigl(y(v)\bigr).(1)✦ An array of order zero has a single entry, at the empty word, and is identified with that point of YY.✦

The plane restriction ρa,i\rho_{a,i} fixing axis aa, one of the axes 0,…,n0, \dots, n, at the slot i∈Ii \in I is the map from YIn+1Y^{I^{n+1}} to YInY^{I^n} that keeps the slice of an array on which the coordinate along axis aa equals ii: ρa,i(y)(v)=y(v0,…,va−1,i,va,…,vn−1).\rho_{a,i}(y)(v) = y(v_0, \dots, v_{a-1}, i, v_a, \dots, v_{n-1}).(2)✦ The last-axis planes of yy are the arrays y(j)=ρn,j(y)y^{(j)} = \rho_{n,j}(y) of order nn, and for a subset JJ of II the tuple of planes is ρJ(y)=(y(j))j∈J\rho_J(y) = (y^{(j)})_{j \in J}.

A standard Borel space is a nonempty set with the Borel σ\sigma-algebra of some Polish topology on it.✦ Let XX be one and TT a measurable automorphism of XX. A system of initial laws for TT is a sequence of probability measures μn\mu_n on XInX^{I^n} that is invariant, consistent and proper: for every order nn, axis aa, slot ii and subset J≠IJ \ne I, μn∘Tˆ−1=μn,μn+1∘ρa,i−1=μn,μn+1∘ρJ−1=μn⊗J.\mu_n \circ \hat T^{-1} = \mu_n, \qquad \mu_{n+1} \circ \rho_{a,i}^{-1} = \mu_n, \qquad \mu_{n+1} \circ \rho_J^{-1} = \mu_n^{\otimes J}.(3)✦

An array object over (μn)(\mu_n) is a stationary family of random arrays with parameters. It consists of standard Borel spaces YY and UU with measurable automorphisms SS of YY and RR of UU; measurable maps x:Y→Xx \colon Y \to X and p:Y→Up \colon Y \to U, the base point and the parameter, with x∘S=T∘xx \circ S = T \circ x and p∘S=R∘pp \circ S = R \circ p; and probability measures mnm_n on YInY^{I^n} that satisfy (3) with mnm_n and SS in place of μn\mu_n and TT, lie over the base, mn∘xˆ−1=μnm_n \circ \hat x^{-1} = \mu_n, and satisfy two conditions on the parameters.✦ The first is that for every subset J≠IJ \ne I and almost every value uu of the parameter array pˆ(y)\hat p(y), mn+1(ρJ(y)∈⋅|pˆ(y)=u)=⨂j∈Jκn(u(j)),m_{n+1}\bigl(\rho_J(y) \in \cdot \mathrel{\big|} \hat p(y) = u\bigr) = \bigotimes_{j \in J} \kappa_n\bigl(u^{(j)}\bigr),(4) where κn(w)\kappa_n(w) is the conditional law under mnm_n of an array of order nn given that its parameter array is ww. The second is that under m0m_0, read as a law on YY, the base point and the parameter are independent.

The Definition

Fix an array object over a system of initial laws, with parameter map pp and laws mnm_n, and an order n≥0n \ge 0.

Definition 1. A data kind over II is the parameter kind par\mathrm{par}, the outside kind outj\mathrm{out}_j of some j∈Ij \in I, or the enlarged kind enlj\mathrm{enl}_j of some j∈Ij \in I; the data kinds form the set 𝒦I={par}⊔{outj:j∈I}⊔{enlj:j∈I}.\mathcal{K}_I = \{\mathrm{par}\} \sqcup \{\mathrm{out}_j : j \in I\} \sqcup \{\mathrm{enl}_j : j \in I\}.(5)✦ The data field of kind dd at order n+1n+1 is the σ\sigma-algebra 𝒟n(d)\mathcal{D}_n(d) on YIn+1Y^{I^{n+1}} given by 𝒟n(par)=σ(pˆ(y)),𝒟n(outj)=σ(y(i):i∈I,i≠j),𝒟n(enlj)=𝒟n(par)∨𝒟n(outj).\mathcal{D}_n(\mathrm{par}) = \sigma\bigl(\hat p(y)\bigr), \quad \mathcal{D}_n(\mathrm{out}_j) = \sigma\bigl(y^{(i)} : i \in I,\ i \ne j\bigr), \quad \mathcal{D}_n(\mathrm{enl}_j) = \mathcal{D}_n(\mathrm{par}) \vee \mathcal{D}_n(\mathrm{out}_j).(6)✦ The projection onto the data field of kind dd is the orthogonal projection Pn(d)P_n(d) of the complex Hilbert space L2(mn+1)L^2(m_{n+1}) onto its closed subspace L2(𝒟n(d),mn+1)L^2(\mathcal{D}_n(d), m_{n+1}) of classes with a 𝒟n(d)\mathcal{D}_n(d)-measurable representative: Pn(d)2=Pn(d)=Pn(d)*,Pn(d)L2(mn+1)=L2(𝒟n(d),mn+1).P_n(d)^2 = P_n(d) = P_n(d)^*, \qquad P_n(d)\, L^2(m_{n+1}) = L^2\bigl(\mathcal{D}_n(d), m_{n+1}\bigr).(7)✦

In the paper, the result reads: for n≥1n \ge 1 there are the data fields 𝒫n\mathcal{P}_n, 𝒪n,j\mathcal{O}_{n,j} of the planes other than jj and 𝒢n,j\mathcal{G}_{n,j} of the enlarged jj, which are the page’s data fields of parameter, outside and enlarged kind.

What Is Checked

The definitions (5), (6) and (7) are stated for an array object over a system of initial laws for a measurable automorphism TT of a standard Borel space XX, with its laws mnm_n subject to (3) and (4), its parameter map pp, the coordinatewise maps of (1) and the last-axis planes of (2). The library proves that each data field 𝒟n(d)\mathcal{D}_n(d) is contained in the product σ\sigma-algebra of YIn+1Y^{I^{n+1}}, and that the data kinds over a finite index set form a finite set.