QSCI-CMP：基于化学动机预选的量子选择组态相互作用
QSCI-CMP: Quantum-Selected Configuration Interaction with Chemically Motivated Preselection
AI总结：
该研究提出QSCI-CMP量子-经典混合算法，通过预选化学平凡状态优化SQD-AA，使24量子比特系统的量子计算成本最多降低72%，还可结合最优迭代次数进一步减少查询次数。
AI中文摘要：
我们提出了QSCI-CMP，一种用于分子基态计算的量子-经典混合算法，可减少基于采样的量子对角化与振幅放大（SQD-AA）的查询次数和门数。SQD-AA通过放大尚未测量的基态来缓解量子选择组态相互作用（QSCI）的测量瓶颈，但其预言机需逐一列出已测量的状态，导致门数随收集状态的数量增长；此外，量子资源还会被用于那些从化学知识看重要性明确的状态，比如Hartree-Fock参考态的低阶激发态，这类状态本可在初始阶段通过经典方式收集。因此，我们提出预先固定这类化学上平凡的状态，从一开始就将其纳入对角化子空间，并将其排除在放大目标之外，使用低成本预言机通过每个基态的激发水平和资历数来识别这些状态。我们数值验证，对于24量子比特系统，QSCI-CMP相对于SQD-AA在达到化学精度所需的查询次数和门数上分别减少了约68%和72%。化学平凡子空间可在经典计算预算内自由调整，更大的子空间会将更多工作转移到经典求解器，进一步降低量子成本；当子空间足够好地捕获基态时，甚至完全不需要量子采样。我们还指出，从量子搜索分析中得到的查询最优迭代次数可进一步将两种方法的查询次数减少约12%。
英文摘要：
We present QSCI-CMP, a quantum-classical hybrid algorithm for molecular ground-state calculations that reduces both the query count and the gate count of sample-based quantum diagonalization with amplitude amplification (SQD-AA). SQD-AA mitigates the measurement bottleneck of quantum-selected configuration interaction (QSCI) by amplifying the basis states that have not yet been measured. Its oracle, however, specifies the measured states by listing them one by one, so its gate count grows with the number of collected states. Moreover, the quantum resources are spent even on states whose importance is evident from chemical knowledge, such as low-order excitations from the Hartree-Fock reference, which could be collected classically at the outset. We therefore propose to fix such chemically trivial states in advance, include them in the diagonalization subspace from the start, and exclude them from the amplification target, using a low-cost oracle that recognizes them through the excitation level and the seniority number of each basis state. We numerically demonstrate that QSCI-CMP reduces the query count and the gate count required to reach chemical accuracy by up to approximately 68% and 72% relative to SQD-AA for 24-qubit systems. The chemically trivial subspace is freely tunable within the classical computational budget. A larger subspace shifts more work onto the classical solver and increases the reduction in quantum cost, and when it captures the ground state sufficiently well, no quantum sampling is needed at all. We also point out that a query-optimal iteration count known from the analysis of quantum search further reduces the query count of both methods by approximately 12%.
