Kernels of conditional determinantal measures and the Lyons-Peres completeness conjecture (Q2031675)

Let $E$ be a locally compact $\sigma$-compact Polish space, equipped with a metric such that any bounded set is relatively compact. Let $\mathrm{Conf}(E)$ be the space of locally finite configurations on $E$. Let $\mu$ be a $\sigma$-finite Radon measure on $E$, let $K$ be the kernel of a locally trace class positive contraction acting on the complex Hilbert space $L^2(E, \mu)$, and let ${\mathbf{P}}_K$ be the corresponding determinantal measure on $\mathrm{Conf}(E)$. Assume moreover that $K$ is a locally trace class orthogonal projection onto a closed subspace $H$ of $L^2(E, \mu)$. For $x\in E$, introduce a function $K_x\in L^2(E, \mu)$ by the formula $K_x(t):= K(t, x)$, $t\in E$. Recall the formulation of the Lyons-Peres completeness conjecture: For ${\mathbf{P}}_K$-almost every $X\in \mathrm{Conf}(E)$, we have $\overline{\mathrm{span}}^{L^2(E, \mu)}\{K_x:\, x\in X\} = H$. A point process on $E$ is by definition a Borel probability measure on $\mathrm{Conf}(E)$. The main result of the article states that Lyons-Peres completeness conjecture holds. If we fix a realization for each $h\in H$ in a such a way that the equation $h(x) = \langle h, K_X\rangle$ holds for every $h\in H$ and $x\in E_0$ ($E_0$ is a Borel subset of $E$ with $\mu(E\setminus E_0)=0$) the main results of the paper can equivalently be reformulated as follows: For ${\mathbf{P}}_K$-almost every $X\in \mathrm{Conf}(E)$, if $h\in H$ satisfies $h|_X=0$, then $h=0$. A corollary of the main result is the following result: Corollary. Almost surely, $Z(f_D)$ is a uniqueness set for $A^2(D)$, where $D$ is the open disc, $f_D(z) =\sum_{n=0}^{\infty} f_n\, z^n$, $Z(f_D) = \{z\in D:\, f_D(z)=0\}$ is the zero set of $f_D(z)$, and $A^2(D)$ is the Bergman space of holomorphic functions on $D$ square-integrable with respect to the Lebesgue measure. This corollary was also proved in a different way in the paper [\textit{R. Lyons} and \textit{A. Zhai}, Ann. Inst. Fourier 68, No. 6, 2311--2328 (2018; Zbl 1410.30004)]. In [\textit{S. Ghosh}, Probab. Theory Relat. Fields 163, No. 3--4, 643--665 (2015; Zbl 1334.60083)], the conjecture under the important additional assumption that the determinantal point process $\mathbf{P}_K$ is number rigid in the sense of Ghosh and Peres is established. While many determinantal point processes are indeed number rigid see [Zbl 1334.60083] for the sine-process, [\textit{S. Ghosh} and \textit{Y. Peres}, Duke Math. J. 166, No. 10, 1789--1858 (2017; Zbl 1405.60067)] for the Ginibre ensemble, [\textit{A. I. Bufetov}, Bull. Math. Sci. 6, No. 1, 163--172 (2016; Zbl 1335.60075)] for processes governed by the Airy, the Bessel and more general integrable kernels, [\textit{A. I. Bufetov} et al., Ergodic Theory Dyn. Syst. 38, No. 7, 2493--2507 (2018; Zbl 1400.37010)] for stationary processes, [\textit{A. I. Bufetov} and \textit{Y. Qiu}, Commun. Math. Phys. 351, No. 1, 1--44 (2017; Zbl 1406.60073)] for generalized Ginibre ensembles), zero set $Z(f_D)$ of the present paper is not: Indeed, \textit{A. E. Holroyd} and \textit{T. Soo} [Electron. J. Probab. 18, Paper No. 74, 24 p. (2013; Zbl 1291.60101)] showed that the point process $Z(f_D)$ is insertion and deletion tolerant, the opposite of being number rigid. For determinantal point processes associated with generalized Bergman spaces on $D$, insertion and deletion tolerance is established in [Zbl 1406.60073] and the Radon-Nikodym derivative of the Palm measure with respect to the initial measure is given explicitly as a generalized multiplicative functional.