Sampling in weighted \(L^p\) spaces of entire functions in \(\mathbb{C}^n\) and estimates of the Bergman kernel (Q5939932): Difference between revisions

The author proves for Fock spaces of \(\mathbb C^n\) necessary conditions for sampling or interpolating sequences which generalize (in a certain case) the results obtained by \textit{J. Ortega-Cerdà} and \textit{K. Seip} [J. Anal. Math. 75, 247-266 (1998; Zbl 0920.30039)]. However, for \(n>1\), those are far from being sufficient conditions, unlike in the case \(n=1\) studied in \textit{B. Berndtsson} and \textit{J. Ortega-Cerdà} [J. Reine Angew. Math. 464, 109-128 (1995; Zbl 0823.30023)]. These questions of sampling are motivated, among other things, by the study of frames of Gabor wavelets. Let \(\varphi\) be a plurisubharmonic function in \(\mathbb C^n\), define for \(1\leq p < \infty\) \[ \|f\|_{p,\varphi}^p := \int |f|^p e^{-p\varphi} dm, \] and let \(F_\varphi^p\) be the space of entire functions with \(\|f\|_{p,\varphi}<\infty\). A sequence \(\Gamma := \{\gamma_j\}\) is said to be separated if the infimum of the distance of two distinct points is positive. It is said to be sampling for \(F_\varphi^p\) if there exist two positive constants \(A\) and \(B\) such that for any \(f \in F_\varphi^p\), \[ A \|f\|_{p,\varphi}^p \leq \|\{f(\gamma_j\} \|_{l(p,\varphi)} := \sum_j |f(\gamma_j)|^p e^{-p\varphi(\gamma_j)} \leq B \|f\|_{p,\varphi}^p . \] It is said to be interpolating for \(F_\varphi^p\) if for any sequence of scalars \(\{c_j\}\) with \(\|\{c_j\}\|_{l(p,\varphi)} < \infty\) (where the norm is defined in the above inequalities), there exists a function \(f \in F_\varphi^p\) such that \(f(\gamma_j) = c_j\) for all \(j\). The necessary conditions are given in terms of the upper and lower densities of the sequence \(\Gamma\), defined as follows : \[ D_\varphi^- (\Gamma) := \liminf_{r\to\infty} \inf_{z\in\mathbb C^n} \frac{\#\Gamma \cap B(z;r)}{\int_{B(z;r)} (i\partial \bar \partial \varphi)^n}, \] and \(D_\varphi^+ (\Gamma)\) defined in an analogous manner, with the ``liminf'' replaced by a ``limsup'' and the ``inf'' by a ``sup''. Assume furthermore that \(\varphi\) is \(2\)-homogeneous and \(\mathcal C^2\) outside the origin. Then the following results are proved, for \(1\leq p\leq\infty\) : Theorem 1. If \(\Gamma\) is sampling for \(F_\varphi^p\), then it contains a separated subsequence \(\Gamma'\) which is also sampling and satisfies \( D_\varphi^- (\Gamma') \geq (\pi^n n!)^{-1}\). Theorem 2. If \(\Gamma\) is interpolating for \(F_\varphi^p\), then it is separated and satisfies \( D_\varphi^+ (\Gamma') \leq (\pi^n n!)^{-1}\). In the case \(n=1\), those results hold with strict inequality, but this is an open question for \(n>1\). Simple examples show that conditions about those densities cannot be sufficient for either interpolation or sampling when \(n>1\). Reduction to the case \(p=2\) is relatively easy, once some technical estimates are obtained. The proof then relies, as in \textit{H. J. Landau} [Acta Math. 117, 37-52 (1967; Zbl 0154.15301)], on an estimate about the number of eigenvalues greater than some \(\gamma \in (0,1)\) of the Toeplitz concentration operator given by \(T_{\chi, \varphi}f= P_\varphi (\chi f)\), where \(P_\varphi\) is the Bergman projection from \(L_\varphi^2 := \{f: \|f\|_{2,\varphi}<\infty\}\) to \(F_\varphi^2\), and \(\chi\) is the characteristic function of a ball \(B(z;r)\). All those eigenvalues are in \([0,1]\) anyway, and it is cleverly shown by comparing the traces of \(T_{\chi, \varphi}\) and \(T_{\chi, \varphi}^2\) that they are close to either \(0\) or \(1\), which permits an estimate of the desired quantity using an estimate of the trace of \(T_{\chi, \varphi}\). This is then estimated using a trace formula involving the Bergman kernel for the space \(F_\varphi^2\), and precise estimates on this kernel which involve delicate estimates on solutions of the \(\bar \partial\) equation from \textit{B. Berndtsson} [J. Geom. Anal. 7, No. 2, 195-215 (1997; Zbl 0923.32014)].