The Setting
Take a finite index set and a standard Borel space of states, each state carrying a parameter. A random array of order two is a square matrix of states; an array of order is the same square with cells that are themselves arrays of order . An array object prescribes the laws of such arrays at every order.
The question is how the parts of the square depend on one another once the parameters are known: which families of rows and which families of columns are conditionally independent, and on what a single cell, or a block of cells in a group of rows, depends when the parameters, some rows and some columns are given.
Arrays and Their Laws
An array of order over a space , indexed by , is a function from the index tuples to ; the arrays form the product ✦, and an array of order zero is identified with its single entry, a point of .✦ A map acts on arrays coordinatewise, .✦ For an array of order , an axis and an index , the plane restriction is the array of order whose entry at is the entry of at the tuple obtained by inserting into at position .✦ We write and for the planes through the first and the last axis, and for subsets and of we write and , the tuple of planes along .
For random variables and under a probability measure and a Markov kernel , we write , or , when is a version of the conditional law of given ; a variable named is an array distributed by the measure in the subscript. For Markov kernels from to , with in a finite set, the product kernel sends to the product measure .✦
Fix a measurable automorphism of a standard Borel space . Initial array laws over are probability measures on , one for each , that are invariant under acting coordinatewise, consistent, every plane restriction carrying to , and proper: for every proper subset of , the planes are distributed under as the product of copies of .✦
An array object over initial array laws consists of nonempty standard Borel spaces of states and of parameters with automorphisms and , measurable maps and with and , and probability measures on , the laws of , subject to the following conditions.✦ Each is invariant under and is carried by to ; the laws are consistent and proper as above; under , read on , the maps and are independent; and is conditionally local: for every and every proper subset of , with the parameter array of , obtained by applying coordinatewise, (1) where is the conditional law of an array of order given that its parameter array is . The Markov kernel from to is the own kernel of .✦
Extensions and Saturation
An extension of an array object by an array object over the same initial array laws is a pair of measurable maps, a state map from the states of to those of and a parameter map from the parameters of to those of , that intertwine the automorphisms and satisfy and , with , acting coordinatewise, carrying to for every .✦ The pullback along an extension of by at order is the linear isometry from complex to complex given by , with acting coordinatewise.✦
On its arrays of order , an array object has three kinds of data field, each a sub--field of the measurable sets of : the parameter field , and for each the outside field and the enlarged field . The projection onto a data field is the orthogonal projection of onto the -measurable functions, the conditional expectation .✦
An array object is saturated if its projections onto data fields commute with pullback along every extension: for every extension of , every , every kind of data field and every , (2)✦ where and are the data fields of that kind of and of .
The Square
Let be an array object and an array of order over its states, with parameter array . Row of is the plane and column is the plane , both of order ; the cell is an array of order . For sets and of indices, is the block of cells, with and for the blocks of a single row and of a single column✦, and is the set of indices outside ✦; the rows and columns of are written in the same way. Conditional laws without a subscript are taken under . A family of columns or of rows is proper when its index set is a proper subset of .
For a set of rows, are the parent rows✦ and is the parent data. The parent entries of a column are the cells . The raw data of a column is , the parameters with the columns outside . The reveal data of for a set of rows is , the raw data of together with the cells . For sets and of rows and of columns, is the group data and the whole group data. The group selection sends the whole group data to the group data, the column restriction taking from the columns their cells in the rows of , which form the block .✦
The row family kernel of a set of rows is the Markov kernel sending a parameter array of order to .✦ The group marginal of on a set of columns is the image of under the row restriction .✦
For a Markov kernel from to and a measurable map on , the observed kernel of and sends to the joint law of for drawn from . For a parameter array of order and a set of indices, the observed kernel of is the observed kernel of and ; it sends to the joint law of for drawn from .✦ The posterior of given the rows in , for arrays of order , is the conditional law of drawn from given that its rows in are , a disintegration of the observed kernel of at .✦ The column kernel of a column is the Markov kernel sending the parent data to the posterior .✦
The old data of an array of order at an index is , and the row data map of a row sends the raw data to the old data of row .✦ The row kernel of at sends the raw data to the conditional law under of given the old data of , evaluated at the row data of .✦
The Claim
Theorem 1. Let be a saturated array object over initial array laws, with finite index set , let , let be an array of law and let . For every proper subset of , (3)✦ For every set of rows with and every proper subset of , (4)✦ and, for , for every row and every column , (5)✦ If , then for every row and every column , (6)✦ and (7)✦ Let and be disjoint sets of rows with and . For disjoint sets and of columns with , (8)✦ If moreover , then for every column and every split of into disjoint sets and , (9)✦ and for every split of into disjoint sets and with nonempty, (10)✦
In the paper, the result reads: in , simultaneously for in one conull set, (i) given , proper families of columns are independent with laws and proper families of rows are independent with laws , (ii) for , given and the parent rows , proper families of columns are independent, each with conditioned on its parent entries, and (iii) the law of given , and the other columns is conditioned on the rest of row , while in grouped form, for partitions of the rows into , , and of the columns into , , , the grouped coordinates are pairwise independent and in every ordering, the paper writing for , for and for . The library states these laws at every order , with cells that are arrays of order , except (5), stated at order two, and the column half of (i) is the conditional locality (1) of the object.
What Is Checked
The claim is stated over a saturated array object over initial array laws of a measurable automorphism of a standard Borel space , with finite index set : over its laws and own kernel , the planes of its arrays, and the row family kernels, group marginals, column kernels and row kernels built from them. The plane restrictions, the row data maps and the group selection are measurable✦.