The filtration of the split-words process (Q438961): Difference between revisions

Let \((r_n)_{n\leq 0}\) be a sequence of integers with \(r_n\geq 2\), and let \(A\) denote an alphabet containing at least 2 letters. Set \(l_0=1\) and \(l_{n-1}=r_n l_n,\;n\leq 0\). Let \((X_n,V_n)_{n\leq 0}\) be a sequence of pairs of random variables satisfying for any \(n\leq 0\) the properties (i) \(X_n\) is uniformly distributed on \(A^{l_n}\); (ii) \(V_n\) is uniformly distributed on \(\{1,\dots ,r_n\}\) and independent of the \(\sigma\)-algebra \(F_{n-1}=\sigma(X_{m},V_{m}:m\leq n-1)\); (iii) if the word \(X_{n-1}\) (with length \(l_{n-1}=r_n l_n)\) is spitted into \(r_n\) subwords of length \(l_n\), then \(X_n\) is the \(V_n\)-th among these subwords. Such a sequence is called a split-word process. The main result of this rather narrative and somewhat negligently written paper asserts that there exists a sequence \((U_n)_{n\leq 0}\) of independent random variables whose filtration \((\sigma(U_m:m\leq n))_{n\leq 0}\) coincides with \((F_n)_{n\leq 0}\) if and only if \(\sum_{n\leq 0}\log(r_n)/l_n=\infty\).