clifford qc：用于量子模拟的Python工具包
clifford qc: A Python Toolkit for Quantum Simulation
- Institut Teknologi Bandung（万隆理工学院）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
该工具包通过稀疏泡利代数和共享缓存实现量子模型构建与求解器比较，支持可重现方法开发，但未证明硬件性能或通用加速。
AI中文摘要：
clifford_qc是一个Python研究工具包，用于构建量子模型、比较变分和子空间本征求解器，以及估算测量资源。一个通用的稀疏泡利代数将哈密顿量、密度算子、门和候选选择可观测量联系起来，而专用接口则处理电路程序、执行和统计估计。我们通过一个带梯度的示例来介绍该表示，并追踪一个分子计算过程，涵盖可复用的模型准备、独立求解和可选的参考验证。共享测量缓存可在同一参考态上的不同可观测量之间重用结果；打包的系数存储和流式处理揭示了内存与重计算之间的权衡。已提交的小型系统基准测试展示了这些接口如何在明确的精度和成本假设下支持比较。完全对易的测量减少了认证固定一阶能量泛函所需的射击次数，但在示例设备模型下，纠缠门时间和连接性可能抵消节省或使协议不可接受。其他研究确定了子空间偏差下限，该下限无法通过更多射击消除；费米子编码在不同实例间的比较结果不确定；以及一个近似限制筛选，其所需比较在采样前即失败。生成的表格、来源记录和手稿检查将所报告的声明与其输入联系起来，尽管证据标签尚未统一验证。该包支持可重现的方法开发；这些结果既未确立硬件性能，也未表明普遍的模拟加速。
英文摘要：
clifford_qc is a Python research toolkit for constructing quantum models, comparing variational and subspace eigensolvers, and estimating measurement resources. A common sparse Pauli algebra connects Hamiltonians, density operators, gates, and candidate-selection observables, while dedicated interfaces handle circuit programs, execution, and statistical estimates. We introduce the representation with a worked gradient example and trace a molecular calculation through reusable model preparation, an independent solve, and optional reference validation. Shared measurement caches reuse outcomes across observables on the same reference state; packed coefficient storage and streaming expose a memory-recomputation tradeoff. Committed small-system benchmarks show how these interfaces support comparisons with explicit accuracy and cost assumptions. Fully commuting measurement reduces the shots needed to certify a fixed first-order energy functional, but entangling-gate time and connectivity can offset the saving or make the protocol inadmissible under illustrative device models. Other studies identify a subspace bias floor that more shots cannot remove, an indeterminate comparison of fermionic encodings across instances, and an approximate-restriction screen whose required comparison fails before sampling. Generated tables, provenance records, and manuscript checks connect the reported claims to their inputs, although evidence labels are not yet validated uniformly. The package supports reproducible method development; these results establish neither hardware performance nor a general simulation speedup.
