Interplay of Confinement and Localization in a Programmable Rydberg Atom Chain

TL;DR

The study investigates how confinement and localization interact to shape non-equilibrium dynamics in a programmable Rydberg-Ising chain. By combining real-device Aquila experiments with noisy JUQAS emulations, it demonstrates that confinement can dominate transport when a longitudinal field is strong, while emergent disorder from hardware noise induces localization that can saturate correlations at finite distances. The work identifies spatial inhomogeneity in the longitudinal field as the primary error source, and shows that confinement and localization act cooperatively to produce non-ergodic dynamics on near-term quantum hardware. This framework enables controlled exploration of the crossover between coherent confinement and disorder-induced localization with potential implications for quantum simulation and information storage on noisy devices.

Abstract

Analog quantum simulators promise access to complex many-body dynamics, yet their performance is ultimately set by how device imperfections compete with intrinsic physical mechanisms. Here we present an end-to-end study of correlation spreading in a programmable Rydberg-atom chain realizing a longitudinal-field transverse-field Ising model, focusing on the joint impact of confinement and effective disorder. Experiments performed on QuEra's Aquila quantum processor are benchmarked against large-scale coherent emulations using the Juelich Quantum Annealing Simulator (JUQAS), enabling the controlled inclusion of realistic hardware imperfections. In the ideal coherent limit, a tunable longitudinal field induces confinement of domain-wall excitations into mesonic bound states, leading to a progressive truncation of the correlation light cone. When experimentally relevant inhomogeneities and fluctuations are included, correlations instead saturate at finite distance even in the nominally deconfined regime, revealing localization driven by emergent disorder. The close quantitative agreement between noisy emulations and experimental data allows us to attribute the observed saturation to specific hardware error channels and to identify the dominant contribution. Our results establish a practical framework for diagnosing and modeling error-induced localization in Rydberg quantum processors, while demonstrating that confinement remains a robust and programmable mechanism for engineering non-ergodic dynamics on near-term quantum hardware.

Figures (4)

• Figure 1: Experimental geometry, control sequence, and correlation spreading. (a) Schematic of the 35-atom Rydberg chain used in the experiment. The atoms are arranged in a quasi-one-dimensional geometry with two $60^\circ$ turns to fit the optical-tweezer trapping region. The lattice spacing $a$ and blockade radius $R_b$ determine the nearest-neighbour interaction scale. During initialization, sites with $h_i = 0$ (orange) and $h_i = 1$ (blue) experience opposite local detunings $\pm \Delta_{\mathrm{loc}}$, preparing a Néel-ordered configuration. (b) Temporal pulse sequence applied in the experiment. The global Rabi frequency $\Omega(t)$ (top) and detunings $\Delta_{\mathrm{glo}}(t)$ (middle) and $\Delta_{\mathrm{loc}}(t)$ (bottom) define two stages: state preparation ($0 < t < t_1$) and quantum simulation ($t_1 < t < t_2$). Dashed vertical lines mark the transition between stages, while shaded regions indicate finite quench ramps arising from hardware constraints. (c) Illustration of the measured correlation spreading dynamics. The surface plot shows the time- and distance-resolved staggered connected correlations $(-1)^d C_{d(t)}$, with colour indicating correlation strength. Superimposed spin configurations highlight the propagation and subsequent saturation of correlations during the quantum evolution.
• Figure 2: Confinement and noise–induced suppression of correlation spreading. Time evolution of the staggered connected correlations $(-1)^d C_d(t)$ for a 35-atom chain at three longitudinal-field strengths $h^z = 0$ (left), $0.04$ (middle), and $0.1$ (right). Black curves denote the semiclassical single-meson prediction for the maximal correlation front of the ideal confined system. Rows correspond to: (a) ideal coherent JUQAS simulations without noise, showing ballistic spreading at $h^z = 0$ and progressively stronger confinement-induced suppression as $h^z$ increases. (b) Noisy JUQAS simulations with the experimentally implemented, spatially varying $h_i$ pattern ($...0101010...$) with added noise, that translates to a homogeneous longitudinal field with added noise in the ferromagnetic quantum Ising model interpretation of the Rydberg Hamiltonian. Both confinement ($h_z=0.1$) and noise-induced localization ($h_z=0$) restrict the propagation of correlations. (c) Noisy JUQAS simulations with a uniform longitudinal-field pattern ($h_i = 1$) with added noise. The pattern translates to a staggered longitudinal field in the ferromagnetic quantum Ising interpretation of the Rydberg Hamiltonian, thereby removing confinement so that only localization limits the spread; as $h^z$ increases, the magnitude of the noise also increases in the same way as in (b), and localization becomes stronger, but propagation is still suppressed less than in (b). (d) Experimental data from the Aquila neutral-atom quantum simulator, showing qualitatively similar saturation of the correlation front due to noise-induced localization at $h_z=0$ and confinement dominated suppression at $h_z=0.1$.
• Figure 3: Qualitative comparison between JUQAS emulations and Aquila experiment. Time evolution of the distance-resolved connected correlations $(-1)^d C_d(t)$ for (a) noisy JUQAS emulations and (b) experimental data from the Aquila Rydberg simulator. (c) Comparison at $d=12$ between ideal emulations, emulations including only atom-position and field errors, emulations including only $h$-pattern errors, emulations including all error sources listed in Table \ref{['table:error specs']}, and the experimental results. The noisy emulations employ error amplitudes scaled by 1.5 above the nominal device specifications, yielding the best quantitative agreement with experiment. Shaded regions indicate standard deviations from multiple noise realizations and experimental repetitions. The close agreement across all distances demonstrates that the enhanced noise model captures the effective inhomogeneity and initialization errors of the device, with the $h$-pattern emerging as the dominant error source.
• Figure 4: Correlation dynamics in a larger Rydberg chain. Time evolution of the staggered connected correlations $(-1)^d C_d(t)$ measured on the Aquila device for a chain of $L = 89$ atoms. (up) Deconfined regime with $h^z = 0$ and (down) confined regime with $h^z = 0.04$. In both cases, correlations extend only over short distances relative to the system size and quickly saturate, indicating strong decoherence and signal loss consistent with the effects of device noise. The absence of sustained propagation or oscillatory behavior, together with the rapid damping of correlations, suggests that increasing system size leads to a reduced signal-to-noise ratio and enhanced effective localization of excitations.