The discrimination theorem for normality to non-integer bases (Q1366958)

If \(\theta\) is a real number \(>1\), the real number \(x\) is normal in base \(\theta\) if \(\{\theta^nx\}\) is uniformly distributed modulo 1. \textit{M. Mendès France} asked [Sémin. Delange-Pisot-Poitou, Théorie des Nombres, Fasc. 1, Exp. 7 (1975; Zbl 0324.10049)] for which bases \(\varphi\) and \(\psi\) normality in one implies normality in the other. In the paper under review, the authors prove that if normality in base \(\varphi\) implies normality in base \(\psi\), then \({\log\varphi\over \log\psi}\) is rational. This had been conjectured by Mendès France only when \(\varphi\) and \(\psi\) are both Pisot numbers. Since the authors and \textit{G. Brown} [Normality with respect to powers of a base, Duke Math. J. 88, 247-265 (1997)] gave a complete description of the values of \(\theta\), \(r\), \(s\) for which normality in base \(\theta^r\) implies normality in base \(\theta^s\), this permits Mendès France's question to be completely solved: Let \(\varphi\) and \(\psi\) be real numbers \(>1\). Then normality in base \(\varphi\) implies normality in base \(\psi\) if and only if there exist positive integers \(r\) and \(s\) with \(\varphi= \theta^s\) and \(\psi=\theta^r\) and either (i) \(\exists k\) such that \(\theta^k\in\mathbb{Z}\) and \(\mathbb{Q}(\theta^r)\subset \mathbb{Q}(\theta^s)\) or (ii) \(\exists k\) such that \(\theta^k\pm \theta^{-k}\in\mathbb{Z}\) and \(s\) divides \(r\). To know more about the history of the problem and to see how Riesz products enter the picture, one should definitely read this nice paper.