AI 中文总结
该研究通过Google量子AI实验结合数值模拟，揭示味自由度的局域无序谱而非仅无序强度，决定无无序定域化，二元与多能级无序的定域化行为存在本质差异。
AI 中文摘要
近期Google Quantum AI实验[Gyawali等人，《科学》393卷，71页（2026）]利用量子并行性模拟无序平均多体动力学，守恒的局域自由度产生有效无序势。我们研究这些静态变量的局域谱如何控制味扩展Z2格点规范理论中的定域化，该理论映射到具有n能级键无序的混合场伊辛链。结合有限尺寸谱、纠缠诊断与无限矩阵乘积态动力学，我们发现二元与多能级无序存在质的区别：n=2时，表观定域化最终让位于热化，长寿命瞬态源于能标分离、简并谱塔与近似希尔伯特空间碎片化；而n=4时呈现一致定域化特征，包括泊松能级统计、面积律本征态纠缠、非热纠缠谱及热力学极限下可及时间内的持久局域记忆。结果表明，尽管二元无序方差更大，但缺乏抑制共振所需的局域振幅多样性，定域化不仅由无序强度决定，还取决于局域无序谱及由此产生的多体希尔伯特空间共振连通性。
英文摘要
A recent \texttt{Google Quantum AI} experiment [\href{https://www.science.org/doi/10.1126/science.adr9680}{Gyawali \textit{et al.}, Science \textbf{393}, 71 (2026)}] has exploited quantum parallelism to emulate disorder-averaged many-body dynamics, with conserved local degrees of freedom generating an effective disorder potential. We investigate how the local spectrum of these static variables controls localization in a flavor-extended ${\mathbb Z}_2$ lattice gauge theory, which maps onto a mixed-field Ising chain with $n$-level bond disorder. Combining finite-size spectral and entanglement diagnostics with infinite matrix-product state dynamics, we find a qualitative distinction between binary and multilevel disorder. For $n=2$, apparent localization ultimately gives way to thermalization; the long-lived transient arises from energy-scale separation, degenerate spectral towers, and approximate Hilbert-space fragmentation. By contrast, $n=4$ displays consistent localization signatures, including Poissonian level statistics, area-law eigenstate entanglement, nonthermal entanglement spectra, and persistent local memory over accessible times in the thermodynamic limit. Our results show that, despite its larger variance, binary disorder lacks the local amplitude diversity needed to suppress resonances. Thus, localization is governed not simply by disorder strength, but by the local disorder spectrum and the resulting resonant connectivity of the many-body Hilbert space.
Comments$22$ pages, $14$ figures
