Fluctuations in Various Regimes of Non-Hermiticity and a Holographic Principle

Abstract: We investigate the variance of both smooth linear statistics and rough linear statistics (counting statistics) of random normal matrices. The two classical ensembles of random matrices, the Gaussian unitary ensemble (GUE) and the complex Ginibre ensemble, enjoy an exact map to the ground state of noninteracting Fermions in a harmonic trap in one, respectively two dimensions. Both feature that the variance of the number of Fermions (the number variance) in a set $A$, and the respective entanglement entropy are proportional. For the Ginibre ensemble this statement about the entropy was known for radial symmetric sets $A$, and we prove it in generality for subsets $A$ of the bulk, thus establishing a holographic principle. Moreover, for generic random normal matrix ensembles depending on some potential $V$, not necessarily exhibiting radial symmetry, we show that for large matrix size $n$ the number variance is proportional to the circumference of $A$, $\sim\sqrt{n}|\partial A|$. A special focus is on the variance of linear statistics of the elliptic Ginibre ensemble, which interpolates between the GUE and Ginibre ensemble, in various regimes of non-Hermiticity, where we enlarge the concept of weak non-Hermiticity to mesoscopic scales. Within the elliptic Ginibre ensemble we prove that for smooth test functions $f$ its variance interpolates between that of the GUE and Ginibre ensemble. The interpolation depends on two positive parameters, the rescaling $\alpha$ of the weak non-Hermiticity parameter $\tau=1-\kappa n^{-\alpha}$, and the rescaling $\gamma$ of the test function $f(n^\gamma z)$, for a continuous range of values of $\alpha$ and $\gamma$. In the mesoscopic regime and when $\alpha=\gamma<1$, we prove an interpolating central limit theorem with an adaptation of the method of Ward identities.