Intermittency of dynamical phases in a quantum spin glass

Vadim N. Smelyanskiy, Kostyantyn Kechedzhi, Sergio Boixo, Hartmut Neven, Boris Altshuler

Published 2019-07-02Version 1

Answering the question of existence of efficient quantum algorithms for NP-hard problems require deep theoretical understanding of the properties of the low-energy eigenstates and long-time coherent dynamics in quantum spin glasses. We discovered and described analytically the property of asymptotic orthogonality resulting in a new type of structure in quantum spin glass. Its eigen-spectrum is split into the alternating sequence of bands formed by quantum states of two distinct types ($x$ and $z$). Those of $z$-type are non-ergodic extended eigenstates (NEE) in the basis of $\{\sigma_z\}$ operators that inherit the structure of the classical spin glass with exponentially long decay times of Edwards Anderson order parameter at any finite value of transverse field $B_{\perp}$. Those of $x$-type form narrow bands of NEEs that conserve the integer-valued $x$-magnetization. Quantum evolution within a given band of each type is described by a Hamiltonian that belongs to either the ensemble of Preferred Basis Levi matrices ($z$-type) or Gaussian Orthogonal ensemble ($x$-type). We characterize the non-equilibrium dynamics using fractal dimension $D$ that depends on energy density (temperature) and plays a role of thermodynamic potential: $D=0$ in MBL phase, $0<D<1$ in NEE phase, $D\rightarrow 1$ in ergodic phase in infinite temperature limit. MBL states coexist with NEEs in the same range of energies even at very large $B_{\perp}$. Bands of NEE states can be used for new quantum search-like algorithms of population transfer in the low-energy part of spin-configuration space. Remarkably, the intermitted structure of the eigenspectrum emerges in quantum version of a statistically featureless Random Energy Model and is expected to exist in a class of paractically important NP-hard problems that unlike REM can be implemented on a computer with polynomial resources.