Superspin Renormalization and Slow Relaxation in Random Spin Systems

TL;DR

This paper introduces a real-space renormalization group method for analyzing the dynamics of disordered spin systems, revealing slow relaxation behaviors and providing a new computational approach for large, complex quantum systems.

Contribution

The authors develop an excited-state RSRG-X formalism applicable to various symmetries and long-range interactions, enabling efficient simulation of dynamics in large disordered spin systems.

Findings

Quantitative agreement with exact diagonalization at low frequencies.

Decay of autocorrelation functions slower than power laws.

Observation of slow late-time decay in two-dimensional long-range systems.

Abstract

We develop an excited-state real-space renormalization group (RSRG-X) formalism to describe the dynamics of conserved densities in randomly interacting spin-$\frac{1}{2}$ systems. Our formalism is suitable for systems with $\textrm{U}(1)$ and $\mathbb{Z}_2$ symmetries, and we apply it to chains of randomly positioned spins with dipolar $XX+YY$ interactions, as arise in Rydberg quantum simulators and other platforms. The formalism generates a sequence of effective Hamiltonians which provide approximate descriptions for dynamics on successively smaller energy scales. These effective Hamiltonians involve ``superspins'': two-level collective degrees of freedom constructed from (anti)aligned microscopic spins. Conserved densities can then be understood as relaxing via coherent collective spin flips. For the well-studied simpler case of randomly interacting nearest-neighbor $XX+YY$ chains, the superspins reduce to single spins. Our formalism also leads to a numerical method capable of simulating the dynamics up to an otherwise inaccessible combination of large system size and late time. Focusing on disorder-averaged infinite-temperature autocorrelation functions, in particular the local spin survival probability $\overline{S_p}(t)$, we demonstrate quantitative agreement in results between our algorithm and exact diagonalization (ED) at low but nonzero frequencies. Such agreement holds for chains with nearest-neighbor, next-nearest-neighbor, and long-range dipolar interactions. Our results indicate decay of $\overline{S_p}(t)$ slower than any power law and feature no significant deviation from the $\sim 1/ \log^2(t)$ asymptote expected from the infinite-randomness fixed-point of the nearest-neighbor model. We also apply the RSRG-X formalism to two-dimensional long-range systems of moderate size and find slow late-time decay of $\overline{S_p}(t)$.