We study the approximate state preparation problem on noisy intermediate-scale quantum (NISQ) computers by applying a genetic algorithm to generate quantum circuits for state preparation. The algorithm can account for the specific characteristics of the physical machine in the evaluation of circuits, such as the native gate set and qubit connectivity. We use our genetic algorithm to optimize the circuits provided by the low-rank state preparation algorithm introduced by Araujo et al., and find substantial improvements to the fidelity in preparing Haar random states with a limited number of CNOT gates. Moreover, we observe that already for a 5-qubit quantum processor with limited qubit connectivity and significant noise levels (IBM Falcon 5T), the maximal fidelity for Haar random states is achieved by a short approximate state preparation circuit instead of the exact preparation circuit. We also present a theoretical analysis of approximate state preparation circuit complexity to motivate our findings. Our genetic algorithm for quantum circuit discovery is freely available at https://github.com/beratyenilen/qc-ga .

翻译：我们研究在含噪中等规模量子（NISQ）计算机上的近似态制备问题，通过应用遗传算法生成用于态制备的量子电路。该算法可在电路评估中考虑物理机器的特定特性，如原生门集和量子比特连接性。我们利用遗传算法优化由Araujo等人提出的低秩态制备算法提供的电路，发现对于有限数量CNOT门制备的Haar随机态，其保真度有显著提升。此外，我们观察到，对于量子比特连接有限且噪声水平显著的5量子比特量子处理器（IBM Falcon 5T），Haar随机态的最大保真度由较短的近似态制备电路而非精确制备电路实现。我们还提出了近似态制备电路复杂性的理论分析以佐证我们的发现。本文的量子电路发现遗传算法可在https://github.com/beratyenilen/qc-ga 免费获取。