OctoInform Beta

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

Lemma 3.6 [Parameter geometry]: In the order-two parameter law of a saturated array object with |I|≥2|I| \ge 2, the column qjq_j given the row pap_a has law Ka(ua,j,⋅)K_a(u_{a,j}, \cdot)

Given a row, proper sets of columns are conditionally independent.

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

An array object lifts a measurable automorphism TT of a space XX to stationary random arrays indexed by tuples from a finite set II; each entry is a point of a state space YY and carries a parameter in a space UU. At order two the parameters form a random I×II \times I matrix. Lemma 3.6 of the paper on Rokhlin’s multiple-mixing problem for one transformation (OpenAI, 2026) describes the geometry of that matrix when the object is saturated: what one row determines about the columns that cross it, and how the row and the column through a diagonal entry depend on each other.

Arrays

Fix a finite index set II. An index tuple of order nn is an element v=(v0,…,vn−1)v = (v_0, \dots, v_{n-1}) of InI^n.✦ An array of order nn over a space YY is a function y:In→Yy \colon I^n \to Y, and the arrays of order nn form the product YInY^{I^n}.✦ An array of order zero has a single entry, at the empty tuple, through which YI0Y^{I^0} is identified with YY as a measurable space.✦ The coordinatewise map fn:YIn→ZInf_n \colon Y^{I^n} \to Z^{I^n} induced by a map f:Y→Zf \colon Y \to Z is fn(y)(v)=f(y(v))f_n(y)(v) = f(y(v)).✦

For an axis a∈{0,…,n}a \in \{0, \dots, n\} and a slot i∈Ii \in I, the plane restriction fixing axis aa at slot ii sends an array yy of order n+1n+1 to the array y|a=iy|_{a=i} of order nn given by y|a=i(v)=y(v0,…,va−1,i,va,…,vn−1),v∈In.y|_{a=i}(v) = y(v_0, \dots, v_{a-1}, i, v_a, \dots, v_{n-1}), \qquad v \in I^n.(1)✦ For a subset JJ of II, the last-axis planes of yy along JJ are the family (y|n=j)j∈J(y|_{n=j})_{j \in J}.✦

Array Objects

A standard Borel space is a nonempty measurable space isomorphic to a Polish space with its Borel σ\sigma-algebra.✦ Let TT be a measurable automorphism of a standard Borel space XX. Initial array laws over (X,T)(X, T) are probability measures mnXm^X_n on XInX^{I^n}, one for each n≥0n \ge 0, each preserved by TnT_n, that are consistent and proper: for every order nn, axis aa, slot ii and subset J≠IJ \neq I, (y↦y|a=i)*mn+1X=mnX,(y↦(y|n=j)j∈J)*mn+1X=⨂j∈JmnX.(y \mapsto y|_{a=i})_* \, m^X_{n+1} = m^X_n, \qquad \bigl(y \mapsto (y|_{n=j})_{j \in J}\bigr)_* \, m^X_{n+1} = \bigotimes_{j \in J} m^X_n .(2)✦

An array object MM over mXm^X consists of a standard Borel state space YY with an automorphism TYT_Y, a standard Borel parameter space UU with an automorphism 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}, its laws, such that each mnm_n is preserved by (TY)n(T_Y)_n and lies over the base, (xn)*mn=mnX(x_n)_* \, m_n = m^X_n, the laws are consistent and proper as in (2), conditionally local, and locally independent: under m0m_0, read on YY, the maps xx and par\mathrm{par} are independent.✦ For Markov kernels λj\lambda_j from a space ZZ to spaces WjW_j, with jj in a finite set, the product kernel ⨂jλj\bigotimes_j \lambda_j is the Markov kernel from ZZ to ∏jWj\prod_j W_j that sends z∈Zz \in Z to the product measure ⨂jλj(z)\bigotimes_j \lambda_j(z).✦ Conditional locality of the laws of MM means that for every nn and every J≠IJ \neq I, under mn+1m_{n+1} and for almost every parameter array pp, Law((y|n=j)j∈J|parn+1(y)=p)=⨂j∈Jκn(p|n=j,⋅),\mathrm{Law}\bigl((y|_{n=j})_{j \in J} \mathrel{\big|} \mathrm{par}_{n+1}(y) = p\bigr) = \bigotimes_{j \in J} \kappa_n(p|_{n=j}, \cdot),(3)✦ where κn(w,⋅)=mn(⋅∣parn=w)\kappa_n(w, \cdot) = m_n(\, \cdot \mid \mathrm{par}_n = w) is the conditional law under mnm_n of an array given its parameter array.

An extension ee of MM by an array object NN over the same initial laws is a pair of measurable maps, a state map σ:YN→YM\sigma \colon Y_N \to Y_M and a parameter map ρ:UN→UM\rho \colon U_N \to U_M, that intertwine the automorphisms of the state spaces and of the parameter spaces, satisfy xM∘σ=xNx_M \circ \sigma = x_N and parM∘σ=ρ∘parN\mathrm{par}_M \circ \sigma = \rho \circ \mathrm{par}_N, and carry laws to laws, (σn)*mnN=mnM(\sigma_n)_* \, m^N_n = m^M_n for every nn.✦ The pullback along ee at order nn is the linear isometry e*:L2(mnM)→L2(mnN)e^* \colon L^2(m^M_n) \to L^2(m^N_n), F↦F∘σnF \mapsto F \circ \sigma_n, of complex L2L^2 spaces.✦

For each order nn, an array object carries a data σ\sigma-algebra on YIn+1Y^{I^{n+1}} for each of 2|I|+12|I| + 1 data kinds: the parameter field 𝒫n+1\mathcal{P}_{n+1} and, for each j∈Ij \in I, the outside field 𝒪n+1,j\mathcal{O}_{n+1,j} and the enlarged field 𝒢n+1,j\mathcal{G}_{n+1,j}. The projection EdE_d onto a data field dd is the conditional expectation onto dd, the orthogonal projection of L2(mn+1)L^2(m_{n+1}) onto the closed subspace of dd-measurable functions.✦

An array object MM is saturated if its projections onto data commute with pullback along every extension: for every extension ee of MM by an array object NN, every order nn, every data field dd and every F∈L2(mn+1M)F \in L^2(m^M_{n+1}), EdN(e*F)=e*(EdMF).E^N_d(e^* F) = e^*(E^M_d F).(4)✦

Parameter Laws

The parameter law of order nn of MM is the probability measure πn=(parn)*mn\pi_n = (\mathrm{par}_n)_* \, m_n on UInU^{I^n}, the law of the parameter array of an array drawn from mnm_n.✦ A parameter array of order one is a tuple q=(qr)r∈Iq = (q_r)_{r \in I}, and its plane at a slot a∈Ia \in I is the coordinate qaq_a, a point of UU. For a∈Ia \in I, the parameter coordinate kernel KaK_a is the kernel from UU to UIU^I giving the conditional law under π1\pi_1 of a tuple given its coordinate at aa: Ka(w,⋅)=π1(⋅∣qa=w),w∈U.K_a(w, \cdot) = \pi_1(\, \cdot \mid q_a = w), \qquad w \in U.(5)✦

A parameter array of order two is a matrix u=(ur,s)r,s∈Iu = (u_{r,s})_{r,s \in I}. The row pa=u|0=a=(ua,s)s∈Ip_a = u|_{0=a} = (u_{a,s})_{s \in I} and the column qj=u|1=j=(ur,j)r∈Iq_j = u|_{1=j} = (u_{r,j})_{r \in I} of a parameter array uu of order two are parameter arrays of order one, and they meet in the entry ua,ju_{a,j}, the coordinate of pap_a at jj and of qjq_j at aa.

The Claim

Theorem 1. Let MM be a saturated array object over initial array laws, with an index set II of at least two elements, and let uu be a parameter array of order two drawn from π2\pi_2, with rows pap_a and columns qjq_j. For every a∈Ia \in I and every subset JJ of II other than II itself, Law((qj)j∈J|pa)=⨂j∈JKa(ua,j,⋅),\mathrm{Law}\bigl((q_j)_{j \in J} \mathrel{\big|} p_a\bigr) = \bigotimes_{j \in J} K_a(u_{a,j}, \cdot),(6)✦ and for every i∈Ii \in I, Law((pi,qi)|ui,i)=Ki(ui,i,⋅)⊗Ki(ui,i,⋅).\mathrm{Law}\bigl((p_i, q_i) \mathrel{\big|} u_{i,i}\bigr) = K_i(u_{i,i}, \cdot) \otimes K_i(u_{i,i}, \cdot).(7)✦

In the paper, the result reads: in the level-two parameter law π2\pi_2, for every aa and every proper subset JJ of II, Law((qj)j∈J∣pa)=⨂j∈JKa(ua,j,dqj)\mathrm{Law}((q_j)_{j \in J} \mid p_a) = \bigotimes_{j \in J} K_a(u_{a,j}, dq_j) with Ka(w,dq)=π1(dq∣qa=w)K_a(w, dq) = \pi_1(dq \mid q_a = w), and in particular pip_i and qiq_i are conditionally independent given ui,iu_{i,i}, with the same law Ki(ui,i,⋅)K_i(u_{i,i}, \cdot).

What Is Checked

The claim is stated over a saturated array object MM over initial array laws, with an index set of at least two elements, over its parameter law π2\pi_2 of order two, the plane restrictions that cut a parameter matrix into rows, columns and diagonal entries, the parameter coordinate kernels KaK_a and the product kernel. Beyond the claim, the library proves that every plane restriction is measurable✦.