Spectral statistics of the Laplacian on random covers of a closed negatively curved surface

Let $(X,g)$ be a closed, connected surface, with variable negative curvature. We consider the distribution of eigenvalues of the Laplacian on random covers $X_n\to X$ of degree $n$. We focus on the ensemble variance of the smoothed number of eigenvalues of the square root of the positive Laplacian $\sqrt{\Delta}$ in windows $[\lambda-\frac 1L,\lambda+\frac 1L]$, over the set of $n$-sheeted covers of $X$. We first take the limit of large degree $n\to +\infty$, then we let the energy $\lambda$ go to $+\infty$ while the window size $\frac 1L$ goes to $0$. In this ad hoc limit, local energy averages of the variance converge to an expression corresponding to the variance of the same statistic when considering instead spectra of large random matrices of the Gaussian Orthogonal Ensemble (GOE). By twisting the Laplacian with unitary representations, we are able to observe different statistics, corresponding to the Gaussian Unitary Ensemble (GUE) when time reversal symmetry is broken. These results were shown by F. Naud for the model of random covers of a hyperbolic surface. For an individual cover $X_n\to X$, we consider spectral fluctuations of the counting function on $X_n$ around the ensemble average. In the large energy regime, for a typical cover $X_n\to X$ of large degree, these fluctuations are shown to approach the GOE result, a phenomenon called ergodicity in Random Matrix Theory. An analogous result for random covers of hyperbolic surfaces was obtained by Y. Maoz.
Comment: 64 pages