Separation metrics for real-valued random variables (Q1069199)

Let \(\Omega\) be a probability space with probability measure P, and let W be a fixed, real-valued random variable on \(\Omega\). For real-valued random variables X and Y on \(\Omega\) we define \(d_ W(X,Y)=P[X\leq W<Y\) or \(Y\leq W<X]\), that is, \(d_ W(X,Y)\) is the probability that W separates X and Y. The function \(d_ W\) is a pseudo metric on the space of real-valued random variables on \(\Omega\). Let R be a given collection of real-valued random variables on \(\Omega\). If (1) W is independent of every pair of members of R in the sense that for X,Y\(\in R\) and for intervals A, B, C in the real line we have \(P[X\in A,\quad Y\in B,\quad W\in C]=P[X\in A,\quad Y\in B]\cdot P[W\in C]\) and if (2) for every open interval J we have \(P[W\in J]>0\), then \(d_ W\) is a metric on R.