OctoInform Beta

OpenAI preprint, September 2026 · §3. Ordered arrays and conditional locality · §3.4

§3.4 (setting): the square: At each order nn, the conditional law κn\kappa_n of an array given its parameter array, and the law νn\nu_n of that parameter array

The kernel and the parameter law of an array object.

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

Let TT be a measurable automorphism of a space XX, and let II be a finite index set. The item works with a space YY of states, each state lying over a point of XX and carrying a parameter in a space UU, and with a law mnm_n on the arrays of states of order nn, the families of states indexed by the nn-tuples of elements of II, one for every n≥0n \ge 0. Replacing every entry of such an array by its parameter gives its array of parameters, and the item is about how mnm_n disintegrates over that array. At n=2n = 2 an array is a square of states indexed by I×II \times I, whose rows and columns are arrays indexed by II.

Arrays and Their Planes

An index tuple of length nn over II is a map vv from {0,…,n−1}\{0, \dots, n-1\} to II, written v=(v0,…,vn−1)v = (v_0, \dots, v_{n-1}); these tuples form the product InI^n.✦ An array of order nn over a space YY is a map yy from InI^n to YY, with entry yvy_v at vv, and the arrays of order nn form the product space YInY^{I^n}.✦ The only index tuple of length zero is the empty one, and YI0Y^{I^0} is identified with YY by the measurable isomorphism that sends an array to its entry at the empty tuple, with inverse sending a point to the constant array at that point.✦

The coordinatewise map of a map f:Y→Zf \colon Y \to Z is the map f(n):YIn→ZInf^{(n)} \colon Y^{I^n} \to Z^{I^n} given by (f(n)y)v=f(yv),v∈In.(f^{(n)} y)_v = f(y_v), \qquad v \in I^n.(1)✦ For an axis a∈{0,…,n}a \in \{0, \dots, n\}, a position in the index tuples of length n+1n+1, and a slot i∈Ii \in I, the plane restriction ρa,i:YIn+1→YIn\rho_{a,i} \colon Y^{I^{n+1}} \to Y^{I^n} is the map that fixes axis aa of an array at slot ii: (ρa,iy)v=y(v0,…,va−1,i,va,…,vn−1),v∈In.(\rho_{a,i}\, y)_v = y_{(v_0, \dots, v_{a-1},\, i,\, v_a, \dots, v_{n-1})}, \qquad v \in I^n.(2)✦ For a subset JJ of II, the tuple of planes of an array yy of order n+1n+1 along JJ is the family ρJ(y)=(ρn,jy)j∈J\rho_J(y) = (\rho_{n,j}\, y)_{j \in J} of its planes with the last axis, axis nn, fixed at a slot in JJ.✦

Array Objects

A standard Borel space is a nonempty set with a σ\sigma-algebra under which it is isomorphic, as a measurable space, to a Polish space with its Borel sets.✦

Initial array laws for a measurable automorphism TT of a standard Borel space XX are probability measures mnXm^X_n on XInX^{I^n}, one for each order n≥0n \ge 0, each invariant under T(n)T^{(n)}, that are consistent, (ρa,i)*mn+1X=mnXfor every n, every axis a and every slot i,(\rho_{a,i})_*\, m^X_{n+1} = m^X_n \quad \text{for every } n, \text{ every axis } a \text{ and every slot } i,(3) and proper, (ρJ)*mn+1X=(mnX)⊗Jfor every n and every subset J⊊I,(\rho_J)_*\, m^X_{n+1} = (m^X_n)^{\otimes J} \quad \text{for every } n \text{ and every subset } J \subsetneq I,(4)✦ where g*mg_* m is the image of a measure mm under a map gg.

An array object over initial array laws mXm^X for TT consists of standard Borel spaces YY of states and UU of parameters with measurable automorphisms TYT_Y and TUT_U, measurable maps x:Y→Xx \colon Y \to X and par:Y→U\mathrm{par} \colon Y \to U with x∘TY=T∘xx \circ T_Y = T \circ x and par∘TY=TU∘par\mathrm{par} \circ T_Y = T_U \circ \mathrm{par}, and probability measures mnm_n on YInY^{I^n}, the laws of the object, such that for every nn the law mnm_n is invariant under TY(n)T_Y^{(n)} and lies over the base, x*(n)mn=mnXx^{(n)}_* m_n = m^X_n, the laws are consistent and proper, with mm in place of mXm^X in (3) and (4), and the object is conditionally local and locally independent.✦

The parameter array of an array y∈YIny \in Y^{I^n} is par(n)(y)∈UIn\mathrm{par}^{(n)}(y) \in U^{I^n}, and Lawm(z∣g(z)=p)\mathrm{Law}_m(z \mid g(z) = p) denotes the conditional law of zz given g(z)=pg(z) = p when zz has law mm. An array object is conditionally local when, for every nn and every subset J⊊IJ \subsetneq I, the planes along JJ of an array yy with law mn+1m_{n+1} are independent given its parameter array pp, the plane at slot jj having the conditional law under mnm_n of an array given that its parameter array is ρn,jp\rho_{n,j}\, p: Lawmn+1(ρJ(y)|par(n+1)(y)=p)=⨂j∈JLawmn(z|par(n)(z)=ρn,jp)\mathrm{Law}_{m_{n+1}}\bigl(\rho_J(y) \mathrel{\big|} \mathrm{par}^{(n+1)}(y) = p\bigr) = \bigotimes_{j \in J} \mathrm{Law}_{m_n}\bigl(z \mathrel{\big|} \mathrm{par}^{(n)}(z) = \rho_{n,j}\, p\bigr)(5) for almost every pp under the law of par(n+1)(y)\mathrm{par}^{(n+1)}(y). An array object is locally independent when the maps xx and par\mathrm{par} are independent under m0m_0, read on YY through the identification of YI0Y^{I^0} with YY.

The Definition

Definition 1. Let MM be an array object over initial array laws mXm^X, with states in YY, parameters in UU, parameter map par\mathrm{par} and laws mnm_n, and let n≥0n \ge 0. The kernel of MM given parameters at order nn is the kernel κn\kappa_n from UInU^{I^n} to YInY^{I^n} that gives the conditional law under mnm_n of an array given its parameter array: κn(p,⋅)=Lawmn(y|par(n)(y)=p),p∈UIn.\kappa_n(p, \cdot) = \mathrm{Law}_{m_n}\bigl(y \mathrel{\big|} \mathrm{par}^{(n)}(y) = p\bigr), \qquad p \in U^{I^n}.(6)✦ The parameter law of MM at order nn is the law νn\nu_n on UInU^{I^n} of the parameter array of an array with law mnm_n: νn=(par(n))*mn.\nu_n = (\mathrm{par}^{(n)})_*\, m_n.(7)✦

In the paper, the result reads: the square has level-two states ya,jy_{a,j} for a,j∈Ia, j \in I, the full parameter array EE, the row and column parameter tuples pap_a and qjq_j, and the row and column laws λp\lambda_p, where EE stands for the page’s par(2)(y)\mathrm{par}^{(2)}(y), pap_a and qjq_j for its planes ρ0,a(E)\rho_{0,a}(E) and ρ1,j(E)\rho_{1,j}(E), and λp\lambda_p for κ1(p,⋅)\kappa_1(p, \cdot).

What Is Checked

Both definitions are stated over an array object over initial array laws for a measurable automorphism of a standard Borel space, with a finite index set, at every order n≥0n \ge 0.