Sampling in de Branges spaces and Naimark dilation (Q521945)

Ortega-Cerdà and Seip characterized sampling sequences in Paley-Wiener spaces by embedding in de Branges spaces in which the sequence becomes a complete interpolating sequence. The present work represents an extension in the sense that the kernel functions associated with a sequence \(\{\lambda_n\}\) form a frame for the Paley-Wiener space if and only if the space embeds into a larger space that allows the kernel functions to become a Riesz basis for the latter. The Hermite-Biehler class \(\mathcal{HB}\) consists of entire functions \(E\) such that \(|E(\bar{z})|<|E(z)|\) for all \(z\) with positive imaginary part. Given \(E\in\mathcal{HB}\) the de Branges space \(\mathcal{H}(E)\) consists of all entire \(f\) such that \(\|f\|^2_E=\int_{\mathbb{R}} \bigl|\frac{f(t)}{E(t)}\bigr|\, dt<\infty\) and \(f/E\) and \(f^\ast/E\) are of bounded type (quotient of two bounded analytic functions in the upper half plane) where \(f^\ast=\overline{f(\bar{z})}\). \textit{J. Ortega-Cerdà} and \textit{K. Seip} proved in [Ann. Math. (2) 155, No. 3, 789--806 (2002; Zbl 1015.42023)] that a separated sequence \(\Lambda\) of real numbers is a set of sampling for \(\text{PW}_\pi\) if and only if there exist \(E,F\in \mathcal{HB}\) such that \(\mathcal{H}(E)\subset \text{PW}_\pi\) and \(\Lambda\) is the set of zeros of \(EF+E^\ast F^\ast\). The authors begin proving the following extension, which is a necessary condition for sampling in \(\mathcal{H}(E_0)\): Let \(E_0\in\mathcal{HB}\). If a separated sequence \(\Lambda\) is a sampling sequence for \(\mathcal{H}(E_0)\) then there exist \(E,F\in \mathcal{HB}\) such that \(\mathcal{H}(E)\simeq \mathcal{H}(E_0)\) and \(\Lambda\) is the set of zeros of \(EF+E^\ast F^\ast\). The space \(\mathcal{H}(E)\) has reproducing kernel \[ K_E(w,z)=\frac{\bar{E}(w) E(z)-E(\bar{w}) E^\ast(z)}{2\pi i (\bar{w}-z)}\, . \] Theorem 2 states that if \(E_0\in\mathcal{HB}\), \(\Lambda\) is a sampling sequence for \(\mathcal{H}(E_0)\) and \(E\) is the Hermite--Biehler function as above, the kernel functions \(\{K_E(\lambda_n,\cdot)\}_n\) form a frame for \(\mathcal{H}(E_0)\) with canonical dual frame \(\{K_{E_0}(\lambda_n,\cdot)\}_n\). Consequently, for each \(f\in \mathcal{H}(E_0)\), \(f(z)=\sum f(\lambda_n) K_E(\lambda_n,z)\) (in \(\mathcal{H}(E_0)\)-norm). The first theorem is re-interpreted as providing an embedding of \(\mathcal{H}(E_0)\) and \(\mathcal{H}(F)\) into a larger space \(\mathcal{H}(EF)\) such that, with an appropriate weighting, corresponding weighted interpolating functions form a Parseval frame for \(\mathcal{H}(E)\). Interpolation formulas are provided that relate to frame expansions and embeddings under which weighted interpolating functions provide Riesz bases for the embedding space. A sufficient condition for sampling in \(\mathcal{H}_0\) is provided when \(E_0\) has no real roots (Theorem 5) and an application to multiplexing is outlined.