发表机构
MIGAL–Galilee Research Institute; Tel-Hai University of Kiryat Shmona（MIGAL-加利利研究所; 基里亚特谢莫纳塔尔海大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究提出编译显式的筛查与压力测试方案，基于往返状态回波（RTSE）等指标，在超导、离子阱等量子硬件上开展诊断，结果支持按多维度对量子硬件进行评估。
AI 中文摘要
端到端量子硬件评分无需在工作负载、编译或执行时间之间迁移。我们指定了一个显式编译的筛查与压力测试方案，其首要诊断项为往返状态回波（RTSE）：在路由根处制备四个四面体量子比特态之一，通过交换操作将其移出并返回，在根处应用逆制备操作，最终记录零值。执行快照指的是带日期的提交任务批次及其捕获的能力文档（若可用），而非经过认证的校准周期。在稀疏超导硬件上，字节级相同的通信重运行改变了路由级对比度，尽管聚合RTSE估计值仅相差0.00125。在一项预先冻结的双窗口长度研究中，RTSE及远程逆无操作前驱的根边际均从长度2降至长度10；预先指定的交互作用未支持更优的RTSE保留。在64个删除恢复单元中，IQM执行快照间的平均选中输出返回概率从0.738变为0.624。在一个宣称五个提交虚拟线路间全对全连接的离子阱服务中，两个窗口的恢复值分别为0.911和0.923，超过了预先冻结的三分之二参考值；恢复值与伴随控制的差值分别为0.446和0.443。这些是执行工作负载诊断，而非编码增益、错误抑制、物理损耗、容错或架构排名相关主张。结果支持按工作负载、布局或虚拟线路契约、编译、架构及执行快照进行评估。
英文摘要
Quantum-hardware scores vary across workloads and execution times. We study these variations through repeated communication and deletion-recovery experiments on superconducting and trapped-ion hardware. Our opening test is round-trip state echo (RTSE): prepare one of four tetrahedral qubit states, move it along a route and back by swaps, undo the preparation at the root, and measure the return probability. Repeating identical native communication programs on IQM Emerald changed the differences between tasks on the same routes, while the aggregate RTSE estimates differed by only $0.00125$. A separate two-window experiment found a drop in root return probability from route length 2 to 10 for both RTSE and its remote-inverse do-nothing predecessor; the prespecified directional test did not support better retention by RTSE. Across 64 fixed combinations of placement, input state, and deleted position, the IQM recovery mean fell from $0.738$ to $0.624$. On IonQ's five submitted virtual wires, with advertised all-to-all connectivity, recovery means of $0.911$ and $0.923$ exceeded the prespecified two-thirds reference in both windows. The corresponding differences from a control using the decoder's adjoint were $0.446$ and $0.443$. These measurements characterize complete compiled workloads at dated executions. Recovery is tested after subsystem discard in the circuit model, and each hardware interface is analyzed separately.
Comments41 pages, 11 figures, 11 tables. Revised exposition, organization and methodological detail; numerical results unchanged. Reproducibility package: doi:10.5281/zenodo.21969397
