On the dimension of Bernoulli convolutions (Q2327954)

Let \(\lambda\in (0,1)\) and let \(\{\xi_n : n\in \mathbb{N}_0\}\) be a sequence of random variables with \(\mathbb{P}(\xi_n = -1) = \mathbb{P}(\xi_n = 1) = \frac12\). The Bernoulli convolution \(\mu_\lambda\) with parameter \(\lambda\) is the law of the random variable \(\sum_{n=0}^\infty \xi_n \lambda^n\). Denote by \(\mathcal{P}_d\) the set of real polynomials of degree at most \(d\) all of whose coefficients are either \(-1\), \(0\), or \(+1\). Let \[ E_{d,\alpha} :=\{\eta\in (\tfrac12,1) : \dim\mu_\eta < \alpha \wedge P(\eta) = 0, \text{for some \(P\in \mathcal{P}_d\)}\}. \] The main result of this paper is as follows. Theorem. Let \(\lambda\in (\frac12,1)\) be such that \(\dim\mu_\lambda < 1\). Then for every \(\epsilon > 0\), there exists an \(A > 0\) such that the following holds. For every sufficiently large integer \(d_0\), there exists an integer \[ d \in \left[d_0, \exp^{(5)}(\log^{(5)}(d_0) + A)\right] \] and an \(\eta\in E_{d, \dim\mu_{\lambda+\epsilon}}\) such that \[ |\lambda - \eta| \leq \exp(-d^{\log^{(3)}d}). \] Here, \(\exp^{(m)}\) and \(\log^{(m)}\) denotes the \(m\)-fold iteration of \(\exp\) and \(\log\), respectively. Several corollaries of this result are presented.