Random-matrix model for thermalization

Hans A Weidenmüller
DOI: https://doi.org/10.1088/1751-8121/ad389a
2024-03-28
Journal of Physics A Mathematical and Theoretical
Abstract:We show that for a system governed by a random-matrix Hamiltonian (a member of the time-reversal invariant Gaussian Orthogonal Ensemble (GOE) of random matrices of dimension N), all functions Tr(Aρ(t)) in the ensemble thermalize: For N → ∞ every such function tends to the value Tr(Aρeq(∞)) + Tr(Aρ(0))g2(t). Here ρ(t) is the time-dependent density matrix of the system, A is a Hermitean operator standing for an observable, and ρeq(∞) is the equilibrium density matrix at infinite temperature. The oscillatory function g(t) is the Fourier transform of the average GOE level density and falls off as 1⁄|t|3⁄2 for large t. With g(t) = g(-t), thermalization is symmetric in time. Analogous results, including the symmetry in time of thermalization, are derived for the time-reversal non-invariant Gaussian Unitary Ensemble (GUE) of random matrices. Comparison with the "eigenstate thermalization hypothesis" of Ref. [1] shows overall agreement but raises significant questions.
physics, multidisciplinary, mathematical
What problem does this paper attempt to address?