Random number sequences and the first digit phenomenon (Q456243)

Let \(b>1\) an integer. The mantissa in base \(b\) of a positive real number \(x\) is the unique number \(\mathcal{M}_b(x)\in [1,b[\) such that \(x=\mathcal{M}_b(x)b^k\) for some integer \(k\). A sequence \((u_n)_n\) of positive numbers is called natural-Benford in base \(b\) if \[ \lim_{N\to \infty}\frac 1N \sum_{n=1}^N 1_{[1,t[}(\mathcal{M}_b(u_n))=\mu_b([1,t[)=\log_b t \qquad (1\leq t<b), \] where \(1_{B}\) is the indicator function of the set \(B\). Let \((q_n)_n\) be a sequence of numbers in \([0,1]\) summing to infinity, and let \((X_n)_n\) be a sequence of independent Bernoulli random variable such that \(\mathbb{P}(X_n=1)=q_n\). Let \(Y_n\) denote the \(n\)th number in the random set \(\{k: X_k=1\}\). For a sequence of positive numbers \((u_n)_n\), \((U_n)_n=(u_{Y_n})\) denotes the random subsequence of \((u_n)_n\). \textit{E. Janvresse} and \textit{T. de la Rue} [Uniform Distribution Theory 7, No. 2, 35--60 (2012)] showed that if \(u_n=n\) and \(q_n=1/n\), then \((U_n)_n\) is a.s. natural-Benford. The authors extend this property to a larger class of probabilities \(q_n\) and the case \(u_n=p_n\) (\(p_n\) is the \(n\)th prime number). They give an estimate of the rate of convergence for \(u_n=n\), \(u_n=n\log n\) or \(u_n=p_n\) and a large family of sequence of probabilities \((q_n)_n\). Next, they prove that \((\log n)_n\) and \((\log\log n)_n\) are not logarithmic-Benford either and that \((U_n)_n\) is a.s. natural-Benford when \(u_n=\log n\) and \(q_n=1/(n\log n)\) and when \(u_n=\log\log n\) and \(q_n=1/((n\log n)(\log\log n))\). Lastly, they prove that, if a sequence \((Z_n)\) of positive random variables is a.s. natural-Benford and if the law of \(\mathcal{M}_b(Z_n)\) converges weakly to a probability measure \(Q\) as \(n\to\infty\), then \(Q=\mu_b\). Furthermore, the authors show that, if the sequence \((nq_n)\) is non-increasing, then the law of \(\mathcal{M}_b(U_n)\) converges to weakly to \(\mu_b\) in the case \(u_n=n\), and, under additional conditions on \(q_n\), in the cases \(u_n=n\log n\), \(u_n=p_n\) and \(u_n=n\log\log n\).