Eigenvectors of the square grid plus GUE (Q6143295)

The paper under review is to reveal a product structure of eigenvectors of the GUE-perturbed discrete torus with uniform boundary conditions for small perturbations but converge to discrete Gaussian waves for large perturbations, and to answer the question on how much noise need to add to a discrete torus so that the eigenfunctions become Gaussian waves in the limit. A manifestation of quantum ergodicity is the interactions of eigenfunctions resulting in averaging phenomena akin to the ergodic theorem. But such interactions are not expected to happen in regular domains. Theorem 2 considers eigenvalues of the adjacent matrix of the discrete torus with uniform boundary conditions in the interval \((E \pm n^{-\delta})\) for \(E\in (-4, 4)\setminus \{0\}\) under suitable conditions and states the random measure \(\frac{1}{m} \sum_{i=1}^m\delta_{u_j (\cdot + o_n)}\) converges in probability to the law of the complex Gaussian wave \(Z_E\) for \(\gamma<1\) and all \(\delta\) sufficiently close to 0 and for \(\gamma > 1\) and all \(\delta <1\) the law of \(Z_E\) is not a limit point of these measures. The density of the spectrum measure of \(Z^2\) lows up at \(E=0\), and \(E=\pm 4\) expect to a constant vector with a uniform phase or add signs with a checkerboard pattern, and leave in Problem 7. A central ingredient of the proof of Theorem 12 in Section 2 as a main tool for the convergence into complex Gaussian waves is the theorem of \textit{L. Benigni} [Ann. Inst. Henri Poincaré, Probab. Stat. 56, No. 4, 2822--2867 (2020; Zbl 1465.60005)] on eigenvector limits, and the novelty if Theorem 12 is that it works in the locally convergent matrix setting and the regularity of the empirical measure of the eigenvalue is only required at most points of given window instead of at all the points of the window. To apply Theorem 12 for adjacent matrix A of Theorem 2, one needs to have a good control over the spectrum measure of \(A_{n, c, d}\) down to the optimal scale \(n^{-2 + \varepsilon}\) for most choices of the boundary conditions; to prove the non-convergence part of Theorem 2, one needs Theorem 13 in Section 2 on concentration of perturbed eigenvectors relying on the GUE resolvent flow by \textit{P. von Soosten} and \textit{S. Warzel} [Electron. J. Probab. 23, Paper No. 70, 24 p. (2018; Zbl 1409.60025)] where the initial eigenvalues random choice is not essential, The uniform boundary conditions help to prove regularity of the unperturbed eigenvalues, the main obstacle to Theorem 2 on the standard torus without uniform boundary conditions points in Conjecture 8. Section 3 introduces positive semidefinite operators, Weyl's inequality and the Gaussian unitary ensemble, and the spectral measure of \(A_Z\) in Lemma 18, the Gaussian wave \(Z_{\lambda}\) is a complex Gaussian process with covariance matrix \(M=M_{\lambda} =\frac{\rho_{a, b}(\lambda)}{\rho_{0,0}(\lambda)}\). Section 4 on the spectrum of the discrete torus understands the uniform boundary condition otherwise an open problem in Conjecture 21 of number theoretic nature, lists a good control for logarithmic singularity of \(\rho_{o, o}\) at 0, and the main estimate on the regularity of the spectral measure of the discrete torus with uniform boundary condition is given in Proposition 27 by Riemann sum error and Young's convolution inequality and given in Proposition 28 for close eigenvalue pairs. Section 5 proves Theorem 12, where Theorem 33 in Section 5.3 is a simpler deterministic version of Theorem 12, then subdivide the interval \([E\pm l/2]\) into smaller intervals and the continuity properties of the Stieltjies-transform to most of these smaller intervals where Theorem 33 can be applied, finally using the uniform integrability condition and the norm estimate to complete the proof of Theorem 12 with Theorem 33 and those cannot be applied contain a negligible number of eigenvalues. Starting with free convolution with semicircle distribution of \textit{P. Biane} [Indiana Univ. Math. J. 46, No. 3, 705--718 (1997; Zbl 0904.46045)], Theorem 32 of results of \textit{L. Benigni} [Ann. Inst. Henri Poincaré, Probab. Stat. 56, No. 4, 2822--2867 (2020; Zbl 1465.60005)], Theorem 33 (eigenvector limits of GUE-perturbed locally convergent matrices). Section 6 explains how to apply Theorem 12 to obtain the convergence part of Theorem 2, where the main Proposition 47 given in appendix. Section 7 discusses small perturbations, Proposition 48 of \textit{P. von Soosten} and \textit{S. Warzel} [Electron. J. Probab. 23, Paper No. 70, 24 p. (2018; Zbl 1409.60025)] for properties of the GUE resolvent flow, and the proof of Theorem 13 (concentration of eigenvectors) is given in Lemma 51 an previous Lemma 15--17. Section 8 discusses random initial condition in Lemma 52 and applies to the discrete torus, and the non-convergence part of Theorem 2 follows from Proposition 5 with several lemmas. Some technical statements and Lemmas and Propositions are given in Appendix for either special case or detailed estimates.