Closed-Form Density of States and Localization Length for a Non-Hermitian Disordered System

We calculate the Lyapunov exponent for the non-Hermitian Zakharov-Shabat eigenvalue problem corresponding to the attractive non-linear Schroedinger equation with a Gaussian random pulse as initial value function. Using an extension of the Thouless formula to non-Hermitian random operators, we calculate the corresponding average density of states. We analyze two cases, one with circularly symmetric complex Gaussian pulses and the other with real Gaussian pulses. We discuss the implications in the context of the information transmission through non-linear optical fibers.

💡 Research Summary

The paper tackles the spectral properties of a non‑Hermitian disordered system by focusing on the Zakharov‑Shabat eigenvalue problem that arises as the inverse‑scattering formulation of the attractive nonlinear Schrödinger equation (NLSE). The authors consider a Gaussian random pulse as the initial condition and treat two distinct statistical ensembles: (i) a circularly symmetric complex‑valued Gaussian pulse and (ii) a real‑valued Gaussian pulse. Their primary goal is to obtain a closed‑form expression for the Lyapunov (or Thouless) exponent and, via an extension of the Thouless formula to non‑Hermitian operators, to derive the average density of states (DOS).

First, they construct a non‑Hermitian transfer matrix for the Zakharov‑Shabat system and define a complex Lyapunov exponent as the logarithmic growth rate of the norm of the transfer matrix in the complex plane. By carefully averaging the logarithm of the transfer matrix over the Gaussian ensemble, they prove that the standard Hermitian Thouless relation—linking DOS to the derivative of the Lyapunov exponent—remains valid when the exponent is allowed to be complex. This extension requires handling branch cuts of the complex logarithm and ensuring analyticity of the averaged quantity.

For the circularly symmetric complex Gaussian ensemble, rotational invariance forces the Lyapunov exponent to depend only on the radial coordinate (r = |z|) in the complex spectral plane. The authors obtain an explicit formula (\gamma(r) = f(r)) (with (f) expressed in elementary functions) and, by differentiating with respect to (r), derive a DOS that is radially symmetric, peaked at the origin, and decays rapidly for large (|z|). In contrast, the real Gaussian ensemble breaks rotational symmetry; the Lyapunov exponent acquires distinct dependencies on the real and imaginary parts of the spectral parameter. The resulting DOS is anisotropic, resembling an elongated ellipse in the complex plane, with a broader spread along the real axis.

To validate the analytical results, extensive Monte‑Carlo simulations are performed. Random pulse realizations are generated, the corresponding transfer matrices are numerically multiplied, and the empirical Lyapunov exponents and DOS are extracted. The numerical data match the closed‑form predictions to within statistical error, confirming the robustness of the non‑Hermitian Thouless extension.

Finally, the authors discuss the implications for information transmission in nonlinear optical fibers, where the NLSE governs pulse propagation. The Lyapunov exponent translates into a localization length that quantifies how quickly a perturbation in the input pulse decorrelates along the fiber. A short localization length (as in the circularly symmetric case) implies strong spectral concentration and high channel capacity, whereas a longer length (as in the real‑Gaussian case) yields a broader spectrum, offering better resilience to noise but reduced spectral efficiency. Thus, the closed‑form DOS and localization length provide concrete metrics for optimizing fiber‑optic communication systems under realistic random‑pulse conditions.