Zeros of Gaussian analytic functions (Q1588351): Difference between revisions

Let \(G\subset\mathbb{C}\) be a plane domain and \(\Psi(z)= \{\psi_j(z)\}^N_{j=1}\) \((N\leq+\infty)\) be a vector analytic function in \(G\) with \(\|\Psi(z)\|= \sum^N_{j=1} |\psi_j(z)|^2\). If \(N=+\infty\), the above series converges locally uniformly in \(G\). Let \(\omega= \{\omega_j\}^N_{j=1}\) be a system of independent complex Gaussian random variables such that \({\mathfrak E}\{\omega_j\}= 0\) and \({\mathfrak E} \{|\omega_j|^2\}= 1\). The related probability space is \(\mathbb{C}^N\) equipped with the Gaussian product measure \(d\nu\). Consider the Gaussian analytic function \(\psi(z,\omega)= (\Psi,\omega)= \sum_j\omega_j \psi_j(z)\) and let \(n_\omega\) be a counting measure of zeros (according to their multiplicities) of \(\psi(z,\omega)\). The following results are proved: 1) \({\mathfrak E}\{n_\omega\}= \Delta\log\|\Psi\|dm(z)/2\pi\), where \(\Delta\) is a distributional Laplacian and \(dm(z)\) is the plane Lebesgue measure. 2) Let \(\Psi^{[k]}(z)= \{\psi^{[k]}_j(z)\}^{N^{[k]}}_{j=1}\) \((k= 1,2)\) be two vector analytic functions in \(G\) with linearly independent components and let \({\mathfrak E}\{n^{[k]}_\omega\}\) be the average measures of zeros of \((\Psi^{[k]}(z),\omega)\). If these two measures are equal, then the dimensions of \(\Psi^{[1]}\) and \(\Psi^{[2]}\) are equal to \(N\) and there exist an analytic function \(g(z)\neq 0\) in \(G\) and a unitary transformation \(U\) in \(\mathbb{C}^N\) such that \(\Psi^{[2]}(z)= g(z) U\Psi^{[1]}(z)\). 3) Let \(\mu\) be the Riesz measure of \(\log\|\Psi(z)\|\) and let \(\varphi\in C^\infty_0(G)\) be an arbitrary test function with compact support in \(G\). Then, for every \(\lambda> 0\), \[ \text{Pr}\Biggl(\Biggl\{\omega: \Biggl|\int_G \varphi(dn_\omega- {\mathfrak E}\{dn_\omega\})\Biggr|\geq \lambda\Biggr\}\Biggr)\leq 3\exp(- 2\pi\lambda/\|\Delta\varphi\|_{L^1}). \] Some possible extensions are indicated.