AI 中文总结
研究量子随机采样中相干空间无序影响，通过精确张量网络模拟方形晶格IQP架构，发现无序增加会致两次转变趋向经典可模拟性，退相干降低复杂性，还通过标度定律表征计算困难区域，为近期设备提供误差预算界限。
AI 中文摘要
虽然已知退相干会削弱量子随机采样中的经典硬度，但相干空间无序的影响仍是一个悬而未决的问题。我们使用精确的张量网络模拟研究了一个受两比特门角无序和单比特退相干影响的方形晶格瞬时量子多项式时间（IQP）架构，模拟规模高达576个量子比特。对于无退相干的有限系统，无序增加会驱动两次连续的转变趋向经典可模拟性：输出分布首先失去反集中性，然后随着纠缠被抑制，张量网络模拟成本从指数下降到多项式。有限尺寸标度塌缩与大系统极限下的连续转变一致。退相干进一步降低了复杂性。我们通过标度定律表征了计算困难区域，为实际的近期设备提供了定量误差预算界限。
英文摘要
While decoherence is known to erode classical hardness in quantum random sampling, the impact of coherent spatial disorder remains an open question. We study a square-lattice instantaneous quantum polynomial-time (IQP) architecture subject to two-qubit gate-angle disorder and single-qubit dephasing using exact tensor-network simulations up to 576 qubits. For finite systems without dephasing, increasing disorder drives two consecutive crossovers toward classical simulability: the output distribution first loses anticoncentration, and then the tensor-network simulation cost drops from exponential to polynomial as entanglement is suppressed. The finite-size scaling collapses are consistent with continuous transitions in the large-system limit. Dephasing further reduces the complexity. We characterize the computationally hard regime through scaling laws that provide quantitative error-budget bounds for realistic near-term devices.
