--- title: Non-Hermitian Random Band Matrices url: https://www.emergentmind.com/topics/non-hermitian-random-band-matrices type: topic --- # Non-Hermitian Random Band Matrices Searching arXiv for recent papers on non-Hermitian random band matrices and closely related topics. Non-Hermitian random band matrices are random matrices whose nonzero, or dominant-variance, entries are concentrated near the diagonal or on a sparse banded support, while the matrix lacks Hermitian symmetry and therefore has a genuinely complex spectrum. In current work, the term covers several model classes: inhomogeneous matrices with doubly stochastic variance profiles, Gaussian ensembles with exponentially decaying covariance kernel \(J=(-W^2\Delta+1)^{-1}\), regular-graph-supported sparse bands, power-law banded models, diluted bands, and sparse two-banded matrices with directional bias. Across these settings, the central questions concern the bandwidth threshold for global spectral laws, the singular-value control needed for Hermitization, crossover in local characteristic-polynomial correlations near \(W\sim \sqrt N\), and the competition between disorder-induced localization and non-Hermitian delocalization mechanisms [2508.18143] [2510.04255] [2604.15355]. ## 1. Ensemble classes and structural parameters The contemporary literature does not use a single canonical definition of a non-Hermitian random band matrix. Instead, it studies several families whose common feature is spatially localized coupling together with non-selfadjoint randomness. In the variance-profile setting, a typical model has the form \(X=(b_{ij}x_{ij})\), where the \(x_{ij}\) are independent, mean-zero, variance-one random variables and the deterministic profile \(S=(b_{ij}^2)\) is band-like and doubly stochastic. In Gaussian kernel models, the covariance profile decays exponentially away from the diagonal through \(J=(-W^2\Delta+1)^{-1}\). In regular-graph formulations, the band is encoded by a \(d_N\)-regular directed graph. Physics-oriented variants replace compactly supported bands by algebraically decaying hoppings, explicit dilution, or narrow two-band connectivity patterns [2508.18143] [2510.04255] [1904.11098] [2408.00567]. | Ensemble class | Band parameter | Representative statement | | --- | --- | --- | | Inhomogeneous independent-entry band matrices | \(W\) with \(\sup_{i,j} b_{ij}^2 \le C_W W^{-1}\) | Circular law for \(W\ge N^{1/2+c}\) [2508.18143] | | Gaussian band kernel \(J=(-W^2\Delta+1)^{-1}\) | Effective bandwidth \(W\) | Characteristic-polynomial crossover at \(W\sim \sqrt N\) [2510.04255] | | Critical Gaussian band kernel | \(W=\kappa_*\sqrt N(1+o(1))\) | Critical operator limit \((e^{\mathbb A_0}\mathbf 1,\mathbf 1)\) [2604.15355] | | Variance-profile band matrices | \(b_n\), \(c_n=2b_n+1\), \(\nu=\lim c_n/n\) | CLT for linear eigenvalue statistics [1904.11098] | | Regular-graph-supported sparse bands | Degree \(d_N\) | Elliptic confinement and finite-rank outliers [2408.00567] | Additional model classes broaden the subject’s scope. A non-Hermitian power-law random banded matrix has hoppings \(j_{mn}\sim 1/|m-n|^\alpha\) and complex diagonal disorder [2302.00015]. The non-Hermitian diluted banded random matrix ensemble uses real non-symmetric Gaussian entries inside a band of width \(b\), with random removal of off-diagonal entries controlled by sparsity \(\alpha\) [2406.15426]. Two sparse, two-banded, non-Hermitian ensembles inspired by neural networks use SSH-chain and ladder geometries with random sign disorder and directional bias \(g\) [2605.19072]. ## 2. Global spectral laws and the circular-law regime The sharpest global spectral result currently available in the supplied literature concerns the empirical spectral measure \[ \mu_X=\frac1N\sum_{i=1}^N \delta_{\lambda_i}. \] For inhomogeneous square random matrices \(X=(b_{ij}x_{ij})\) with independent entries, mean \(0\), finite variance, doubly stochastic variance profile \(S=(b_{ij}^2)\), and band-like structure of bandwidth \(W\), the empirical spectral measure converges in probability to the circular law whenever \(W\ge N^{1/2+c}\) for any fixed \(c>0\), provided the entries have a bounded density and a subgaussian tail [2508.18143]. The limiting measure is the circular law \[ \mu_c=\frac{1}{\pi}\mathbf 1_{\{|z|\le 1\}}\, d^2z, \] and the convergence is stated in the weak form \[ \int f\,d\mu_X \xrightarrow{\mathbb P} \int f\,d\mu_c \] for every continuous compactly supported \(f:\mathbb C\to\mathbb R\). This result is significant because \(W\sim N^{1/2}\) is identified there as the natural delocalization threshold in \(1\)-d band matrix theory, analogous to the Hermitian case. The paper explicitly states that it extends the previously known circular-law thresholds with exponents \(\frac56\), \(\frac89\), and \(\frac{33}{34}\) all the way down to \(N^{1/2+c}\). In the special case of block band matrices, the bounded-density assumption is removed and the moment condition is relaxed to finite moments of all orders, still with the conclusion \(\mu_X\xrightarrow{\mathbb P}\mu_c\) for \(W\ge N^{1/2+c}\). A distinctive structural input is a stability estimate on the inverse of a \(2N\times 2N\) operator built from the variance profile \(S\). In the paper’s formulation, this inverse-norm control quantitatively encodes the band structure and is used to make the linearized self-consistent equations for the Hermitized resolvent stable at the optimal delocalization scale. The same work also proves stronger invertibility estimates for profiles with nontrivial mass near the diagonal, so the circular-law theorem is accompanied by information on \(\sigma_{\min}(X-zI)\), not only on eigenvalue counting. ## 3. Characteristic polynomials and the \(W\sim \sqrt N\) crossover A separate line of work studies local spectral structure through characteristic-polynomial correlations rather than the empirical spectral measure. For Gaussian non-Hermitian random band matrices with covariance kernel \(J=(-W^2\Delta+1)^{-1}\), the central object is \[ \Theta_k(z_1,\dots,z_k)=\mathbf E\left\{\prod_{s=1}^k \det(H_N-z_s)\det(H_N-z_s)^*\right\}, \] and the principal regime is the microscopic bulk scaling \[ z_1=z+\frac{\zeta}{\sqrt N},\qquad z_2=z-\frac{\zeta}{\sqrt N},\qquad |z|<1 \] with \(\zeta\) in a compact set [2510.04255]. The main result is a sharp crossover at \(W\sim \sqrt N\). If \(W^2\gg N\log^2 N\) and \(W\le N^{1-\varepsilon_0}\), the normalized second correlation function converges to the Ginibre expression \[ \frac{1-e^{-4|\zeta|^2}}{4|\zeta|^2}. \] If \(W>N^{\varepsilon_0}\) and \(W^2\ll N/\log N\), then \[ \frac{\Theta(z_1,z_2)}{\Theta(z,z)}\to 1, \] so the second correlation function factorizes asymptotically. The mechanism is spectral: after representing \(\tilde\Theta(z_1,z_2)\) as \((\mathcal K_\zeta^{\,N-1}g,g)\), the eigenvalues of the transfer operator satisfy \[ \left|\frac{\lambda_1(\mathcal K_\zeta)}{\lambda_0(\mathcal K_\zeta)}\right|\sim 1-\frac{c}{W^2}. \] The competition is therefore between the matrix size \(N\) and the transfer-operator scale \(W^2\). When \(N/W^2\to\infty\), only the top mode survives and the correlation factorizes; when \(N/W^2\to 0\), many modes contribute and the Ginibre limit is recovered. The critical regime \(W=\kappa_*\sqrt N(1+o(1))\) has now also been analyzed [2604.15355]. In that case, the normalized second correlation function converges to \[ (e^{\mathbb A_0}\mathbf 1,\mathbf 1), \] where \((\cdot,\cdot)\) is the \(L^2([-1,1])\) inner product and \[ \mathbb A_0\varphi(z)=\frac{1}{8(\kappa_*u_*)^2}\Big((1-z^2)\frac{d^2}{dz^2}-2z\frac{d}{dz}\Big)\varphi(z)+2z\,\varphi(z),\qquad u_*=\sqrt{1-|z|^2}. \] The limiting operator is a Legendre-type differential operator plus a linear potential. The same paper frames this as a genuine interpolation between the Ginibre regime \(W\gg \sqrt N\) and the factorized regime \(1\ll W\ll \sqrt N\). It also makes explicit that this is a result about characteristic-polynomial correlations, not yet a full proof of an eigenvalue Anderson-type transition. ## 4. Hermitization, singular values, and invertibility For non-Hermitian band matrices, spectral questions are tightly linked to singular-value analysis through Girko’s Hermitization. The quantity to control is \[ \frac1N\log|\det(X-zI)|=\frac1N\sum_{i=1}^N \log \sigma_i(X-zI), \] so lower bounds on small singular values are indispensable. In the circular-law paper, the main innovation is a lower bound showing that \(X-zI\) cannot have too many singular values in \([0,W^{-1}]\): for \(|z|\) bounded away from \(0\) and \(1\), with overwhelming probability there are at most \(\frac{N}{W}N^\epsilon\) such singular values [2508.18143]. This is derived from a local law for the Hermitized resolvent \[ G(w)=(Y_z^*Y_z-w)^{-1},\qquad Y_z=X-zI, \] whose diagonal entries are shown to be close to a deterministic scalar \(m_c(w,z)\). The same paper gives a representative least-singular-value estimate of the form \[ \mathbb{P}\!\left(\sigma_{\min}(X-zI)\le \epsilon \exp\!\left(-N^{3\kappa}\frac{N}{W}\right),\ \|X\|\le K\right) \le C_{K,\kappa}\epsilon+\exp(-N^{1+\kappa/4}), \] uniformly for \(|z|\le K\), when \(W\ge N^{1/2+c}\). A complementary discrete-direction result proves nonsingularity for non-Hermitian band matrices with sublinear bandwidth and \(\{\pm1\}\) entries [2507.05417]. In the main theorem, if \(d_n\ge n^\alpha\), the entries inside the band \(|i-j|\le d_n\) are uniformly distributed on \(\{\pm1\}\), and the remaining entries are either \(0\) or arbitrary integer-valued random variables, then for \(\alpha>\frac23\), \[ \mathbb{P}(A_n\text{ is singular})\le \exp(-Cn^{\alpha/2}). \] Equivalently, the band matrix is invertible with probability \(1-\exp(-\Omega(n^{\alpha/2}))\). The paper emphasizes that before this result, even invertibility with probability \(1-o(1)\) was not known for these band matrix models except in the very special case of block band matrices. The proof method is distinct from continuous-density least-singular-value arguments. It reduces singularity over \(\mathbb R\) to singularity over \(\mathbb Z_p\), with a prime \(p\) of size roughly \(\exp(\rho n^{\alpha/2})\), and then applies finite-field inverse Littlewood–Offord theory to kernel vectors of block restrictions. The threshold \(\alpha>\frac23\) is stated as a limitation of the current method rather than the natural threshold; the author explicitly conjectures extension down to \(\alpha>\frac12\). The same paper also states bounded-density results of the form \[ \mathbb{P}(\sigma_{\min}(A_n)\le \epsilon e^{-n/d_n}) \le C\epsilon + e^{-Cn} \] for the continuous part of the theory. ## 5. Fluctuations, elliptic confinement, and finite-rank perturbations Beyond laws of large numbers, non-Hermitian random band matrices also admit fluctuation results for linear statistics. For band matrices with variance profile \(w\), bandwidth \(b_n\), and \(c_n=2b_n+1\), the regime is parameterized by \[ \nu=\lim_{n\to\infty}\frac{c_n}{n}\in[0,1]. \] If \(f\) is analytic on a neighborhood of the disk containing the spectrum, then the normalized linear statistic \[ \sqrt{\frac{c_n}{n}}\,L_f^\Delta(M),\qquad L_f^\Delta(M)=\sum_{i=1}^n f(\lambda_i(M))-nf(0), \] converges to a centered Gaussian law \(N(0,\sigma_f^2(\nu))\), and the full vector of such statistics converges to a centered complex Gaussian vector with pseudo-covariance \(\Upsilon=0\) [1904.11098]. The limiting variance is explicit in terms of the Fourier transform of the variance profile. For \(\nu\in(0,1]\), the covariance is expressed as a sum over \(\hat w_\nu(k)\); for \(\nu=0\), it becomes an integral over \(\hat w_0(t)\). The full-matrix case \(\nu=1\) recovers the Rider–Silverstein formula, and \(\sigma_f^2(\nu)\to \sigma_f^2(0)\) as \(\nu\downarrow 0\). A different extension concerns sparse and inhomogeneous band matrices perturbed by bounded-rank deterministic matrices [2408.00567]. Under the moment conditions \[ \mathbb E[X]=0,\qquad \mathbb E[XX^*]=\mathbb E[X^*X]=\mathbf 1,\qquad \mathbb E[X^2]=\rho \mathbf 1, \] together with mild sparsity and moment assumptions, the spectrum of \(X\) is confined near the support of the elliptic law with parameter \(\rho\). In the regular-graph band specialization, \(X_N\) is supported on the edges of a \(d_N\)-regular directed graph, and after normalization by \(d_N^{-1/2}\) one obtains \(\mathbb E[XX^*]=\mathbb E[X^*X]=\mathbf 1\). The paper covers, in particular, all Gaussian band matrices on regular graphs with degree at least \((\log N)^3\). The same framework yields a bounded-rank outlier theory. In the independent-entry case \(\rho=0\), outliers of \((d_N)^{-1/2}X_N+C_N\) converge to the outlying eigenvalues of the perturbation \(C_N\). In the elliptic case, outliers are governed by the map \[ \lambda\mapsto \lambda+\frac{\rho}{\lambda}. \] This extends Tao’s i.i.d. outlier result and the O’Rourke–Renfrew elliptic outlier result to highly sparse and inhomogeneous random matrices, including band matrices on regular graphs. The paper also stresses an important point of scope: the existence of a limiting density is largely unknown for many such inhomogeneous models, yet outliers can still be determined under very general conditions. ## 6. Localization, non-Hermitian delocalization, and model-dependent transitions Localization theory for non-Hermitian banded systems is more varied than the global circular-law picture. In the non-Hermitian power-law random banded matrix, the off-diagonal hoppings decay as \(1/|m-n|^\alpha\) and the diagonal disorder is complex, mimicking random gain and loss [2302.00015]. The paper’s central conclusion is that non-Hermiticity does not simply destroy localization. Instead, it produces two competing effects: the complex on-site disorder spreads levels over a two-dimensional region in the complex plane and suppresses ordinary resonances, but it also permits “bad resonances,” meaning level attraction under non-Hermitian hybridization. The resulting Anderson transition occurs at a disorder-dependent \(\alpha_c(W)\) satisfying \[ \frac d2\le \alpha_c(W)\le d. \] Within the localized phase, the eigenfunctions remain algebraically localized, with decay exponent essentially set by \(\alpha\), even for \(\alpha