Exchangeable random orders and almost uniform distributions (Q1592266)

Let \({\mathcal V}\) \([{\mathcal T}]\) denote the set of all partial [total] orders on \(\mathbb{N}\). For \(U,V\in{\mathcal V}\), let \(U\vee V\) be the smallest partial order containing \(U\cup V\), and put \(\langle V\rangle=\{j \in\mathbb{N}\): there is \(k\neq j\) such that either \((j,k)\in V\) or \((k,j)\in V\}\). Then \(({\mathcal V},\vee)\) is an idempotent Abelian semigroup with neutral element \(D=\{(j ,j):j \in\mathbb{N}\}\) and absorbing element \(\mathbb{N}^2\). Each \(V\in {\mathcal V}\) can be identified with its indicator function \(1_V\in \{0,1\}^{\mathbb{N}^2}\). With this one-to-one correspondence, \({\mathcal V}\) inherits a topological structure as a subspace of the compact metric space \(\{0,1\}^{\mathbb{N}^2}\). Being closed, \({\mathcal V}\) becomes a compact metric space. Also \({\mathcal T}\) is a closed subset of \({\mathcal V}\). Let \({\mathfrak B}({\mathcal V})\) stand for the \(\sigma\)-algebra generated by the open subsets of \({\mathcal V}\). Further, for \(U\subset\mathbb{N}^2\), set \(Q_U=\{W\in {\mathcal V}:W\supset U\}\). Theorem 1. Let \({\mathcal U}=\{U \in{\mathcal V}:\langle U \rangle\) is finite\}. Then a function \(\varphi:({\mathcal U},\vee)\to\mathbb{R}\) is positive definite and normalized (i.e. \(\varphi(D)=1)\) if and only if \(\varphi (U)=\mu (Q_U)\), \(U\in{\mathcal U}\), for a uniquely determined probability measure \(\mu\) on \({\mathfrak B}({\mathcal V})\). Now let \(\Sigma_\infty\) denote the group of all finite permutations of \(\mathbb{N}\). Every \(\sigma\in \Sigma_\infty\) induces a map \(\overline\sigma: {\mathcal V}\to {\mathcal V}\) by \(\overline\sigma (V)=\{(\sigma(j), \sigma(k)): (j,k)\in V\}\), \(V\in{\mathcal V}\). A probability measure \(\mu\) on \({\mathfrak B}({\mathcal V})\) is said to be exchangeable if \(\mu\circ\overline\sigma^{-1}=\mu\) for each \(\sigma\in \Sigma_\infty\). Further, let \({\mathcal W}\) denote the set of all probability measures \(w\) on \([0,1]\) such that \(w(\{t\in [0,1]:w([0,t]) =t \})=1\). For \(w\in{\mathcal W}\), consider \(X_1,X_2,\dots\) i.i.d. random variables on some \((\Omega,{\mathcal A},P)\) with \(P\circ X_1^{-1}=w\). Define \(T:\Omega\to{\mathcal V}\) by \(T(\omega)= \{(j,k)\in \mathbb{N}^2:X_j(\omega)\leq X_k(\omega)\}\), \(\omega\in \Omega\), and put \(\mu_w=P \circ T^{-1}\). Theorem 2. The set of exchangeable probability measures on \({\mathfrak B}({\mathcal T})\) is a Bauer simplex whose extreme points are precisely the probability measures \(\mu_w\), \(w\in {\mathcal W}\).