On spectral \({N}\)-Bernoulli measures (Q2450111)

Let \(1< N\in \mathbb N\), \(0< \rho < 1\), and let \(\mathcal D :=\{0, d_1, \dots d_{N-1}\}\) be a set of digits in \(\mathbb R\). Let \(\{w_j\mid j = 0, 1, \dots N-1\}\) be a set of probability weights. The authors consider self-similar measures \(\mu\) generated by \((\rho, \mathcal D)\) and \(\{w_j\mid j = 0, 1, \dots N-1\}\), i.e., unique probability measures satisfying the self-referential equation \[ \mu (\cdot) = \frac1N \sum_{j=0}^{N-1} w_j \mu (\rho^{-1}(\cdot - d_j)). \] If the digit set \(\mathcal D\) is such that \(d_j = j\), \(j = 0,1, \dots N-1\), then this measure is denoted by \(\mu_{\rho, N}\). The main result of this paper is the following theorem. \(\mu_{\rho,N}\) is a spectral measure, i.e., \(L^2(\mu)\) has an exponential orthonormal basis, iff \(\rho = 1/q\) for some integer \(q > 1\) with \(N| q\). For \(N = 2\) and \(\rho = 1/(2k)\), \(k\in \mathbb N\), the Cantor measure is obtained as a special case and a recent result about the Cantor measure being the only spectral measure for \(0 < \rho < 1\) is recovered.