Regularity of digits and significant digits of random variables (Q2568303)

Consider the expansion of the random variable \(X\) in base \(b\) (\(b\) is a natural number greater than 1), \(X=\sum_{n=-\infty}^{+\infty} (D_n^{(b)}(X)/b^{-n})\). \(X\) is said to be digit-regular in base \(b\) (= d.r. \(b\)) if, for every \(k\geq 1\), the limit of the probability that any block of \(k\) consecutive digits in the above expansion of \(X\), namely, for all \(d_j\in\{0,1,\dots,b-1\}\), \(P(D_{n+j}^{(b)}(X)=d_j, 1\leq j\leq k)\) is equal to \(b^{-k}\) for every \(b>1\). If \(S_n^{(b)}(X)\) denotes the \(n\)th significant digit of \(X\), namely \(S_n^{(b)}(X)=D_{n+m-1}^{(b)}(X)\) for all natural \(n\) on the set \(\{b^{-m}\leq X<b^{-m+1}\}\), then \(X\) is said to be significant-digit-regular in base \(b\). The authors prove that any random variable \(X\) that is d.r. \(b\) is necessarily continuous, but show that \(X\) may be both continuous and almost surely completely normal without being d.r. \(b\) and that it may be d.r. \(b\) without being a.s. normal. Further they show that \(x\) is d.r. \(b\) if, and only if, the sequence of random variables defined by \(b^n X \bmod 1\) converges in distribution to the uniform distribution on \((0,1)\), or, equivalently, if \(X\) has law given by that \(\mu\), \(\varphi_{\mu}(m\,b^n)\) tends to \(0\) for every integer \(m\neq 0\); here \(\varphi_{\mu}\) is the Fourier coefficient of the law \(\mu\), \(\varphi_{\mu}(n):=E(\exp(2\pi\,i\in X))\). If \(X\) has the Cantor distribution on \((0,1)\), \(X=\sum_{k=1}^{\infty} X_k\,3^{-k}\), where the \(X_k\)'s are i.i.d. with law \(P(X_k=0)=P(X_k=2)=1/2\), then \(X\) is both d.r. and normal in base \(2\) but is neither d.r. nor normal in base 3. The properties about significant-digit-regular and strongly digit-regular random variables are also given; here we mention only that every absolutely continuous random variable is both significant-digit-regular and almost surely completely normal.