Mixing operators and small subsets of the circle (Q294262)

The authors investigate the problem of determining continuous linear operators on a complex separable Fréchet space which are weakly mixing or strongly mixing with respect to some nondegenerate Gaussian measure. It turns out that this problem is intimately related to two collections of small subsets of the unit circle \(\mathbb{T}\), namely: the collection \(\mathcal{C}\) of the countable sets and the collection \(\mathcal{U}_0\) of the so-called sets of uniqueness for Fourier-Stieltjes series. \textit{E. Flytzanis} [Geom. Funct. Anal. 5, No. 1, 1-13 (1995; Zbl 0827.46043)] discovered the intimate relationship between ergodicity of the operator with respect to a nondegenerate Gaussian measure and its eigenvectors associated to unimodular eigenvalues (the \(\mathbb{T}\)-eigenvectors). Building on Flytzanis' ideas, \textit{F. Bayart} and \textit{S. Grivaux} [Trans. Am. Math. Soc. 358, No. 11, 5083-5117 (2006; Zbl 1115.47005); Proc. Lond. Math. Soc. (3) 94, No. 1, 181-210 (2007; Zbl 1115.47006 ] obtained sufficient conditions for weak and strong mixing properties of operators on Banach spaces in terms of the \(\mathbb{T}\)-eigenvectors. However, their sufficient conditions depended on some geometrical property of the underlying Banach space or on some regularity property of the \(\mathbb{T}\)-eigenvectors. In the present work, the authors succeed in completely removing these additional hypotheses. They obtain sufficient conditions that are valid on arbitrary complex separable Fréchet spaces and that are ``optimal'' in the sense that they are necessary in any Banach space with cotype 2. More precisely, the main result in the paper is the following. Theorem. Let \(T\) be a continuous linear operator on a complex separable Fréchet space \(X\). {\parindent=0.6cm \begin{itemize} \item[(1)] If the eigenvectors of \(T\) associated to unimodular eigenvalues are perfectly spanning (i.e., for any \(D \in \mathcal{C}\), \(\mathrm{span}(\bigcup_{\lambda \in \mathbb{T} \backslash D} \mathrm{ker}(T - \lambda))\) is dense in \(X\)), then \(T\) is weakly mixing with respect to some Gaussian probability measure on \(X\) with full support. \item [(2)] If the eigenvectors of \(T\) associated to unimodular eigenvalues are \(\mathcal{U}_0\)-perfectly spanning (i.e., for any \(D \in \mathcal{U}_0\), \(\mathrm{span}(\bigcup_{\lambda \in \mathbb{T} \backslash D} \mathrm{ker}(T - \lambda))\) is dense in \(X\)), then \(T\) is strongly mixing with respect to some Gaussian probability measure on \(X\) with full support. \item [(3)] In (1) and (2), the converse implications are true if \(X\) is a Banach space of cotype 2. \end{itemize}}