Executable Implementation and Calibration Validation of the Classical Control Layer: Measurements of 235 Circuits on Tianyan 176
Year: 2026
Type: preprint
Abstract: Objective: to let engineers build quantum applications without needing to understand quantum circuits. This is currently not achievable, because writing a quantum program requires understanding gates, understanding how bits are mapped onto physical qubits, understanding the current noise level, and moreover judging on one's own whether the circuit is reliable—which in effect turns "using a machine" into "understanding a machine". What this paper sets out to verify is the link that fills this gap: given a computation, the system turns it into a gate path on a quantum machine, and before execution it knows how accurate that path is, while upon an execution error it knows which segment went wrong. Engineers write only what they want to compute; the rest is left to calibration and compilation. The approach is to first measure the machine once: prepare a number of known quantum states (called "structure" in this paper), apply candidate gates one by one, measure the probability distribution of each structure (called "symbol" in this paper), record "from which structure, taking which gate, one lands in which structure, with what success probability", and thereby obtain a table; afterwards every computation is compiled into a path on this table. Every segment of the path is measured, so the end-to-end success probability can be multiplied out in advance. XOR was executed on real hardware on 2 qubits: all four inputs were correct, with accuracy 93.2–98.2% (worst single case 0.9322); the maximum deviation between the ex-ante prediction by PTM and the measured values was 0.0153. All data are real-hardware measurements on Tianyan 176 (235 circuits in total), itemized in §8. As for the step of engineers no longer needing to understand quantum circuits, this paper has not performed usability validation (§9.4 row 0). Technically, this paper gives a six-layer implementation on the Tianyan platform, executable versions of three calibration Criteria, and the three conditions of the edge definition; it also divides the objective into three layers (final objective / technical objective / implementation, §1.4), and further splits the technical objective into five testable claims (G1–G5), each paired with an experiment. This paper does not claim: general quantum advantage; algorithms involving coherence; compilation of arbitrary classical programs (the front end is not yet implemented); scale beyond 3 qubits (the implementations are 2-qubit XOR and a 3-qubit half-adder); an analytical bound on multi-hop success probability (only per-edge empirical composition). All experimental designs and criteria are preregistered in §7; data and code are given in the appendix. 目标:让工程师不需要懂量子电路,也能做量子应用。 现在做不到,因为写量子程序须懂门、懂比特如何映射到物理比特、懂当前噪声水平,还须自行判断线路是否可靠——这实际上把「使用一台机器」变成了「理解一台机器」。 本文要验证的是补齐这件事的那一环:给定一个计算,系统把它变成量子机器上的一条门路径,并且在执行之前就知道它有多准、执行出错时知道错在哪一段。工程师只写他要算的东西,其余交给标定与编译。 做法是先把机器量一遍:准备若干已知的量子态(本文叫"结构"),逐个施加候选门,测出每个结构的概率分布(本文叫"符号"),记下"从哪个结构走哪个门会落到哪个结构、成功率多少",得到一张表;之后每个计算都被编译成这张表上的一条路。路的每一段都是实测的,所以端到端成功率可以事先乘出来。 在 2 比特上于真机执行了 XOR:四个输入全对,正确率 93.2–98.2%(最差单条 0.9322);由 PTM 事前预测、与实测相比的最大偏差 0.0153。全部数据为天衍 176 上的真机测量(合计 235 条线路),逐条见 §8。工程师不再需要懂量子电路这一步,本文未做可用性验证(§9.4 第 0 行)。 技术上,本文给出天衍平台上的六层实现、三条标定准则的可执行版本与边定义的三条件;并把目标分三层(最终目标 / 技术目标 / 实现,§1.4),技术目标再拆成五条可检验的声称(G1–G5),逐条配实验。 本文不主张:通用量子优势;含相干的算法;任意经典程序的编译(前端尚未实现);3 比特以上的规模(实做 2 比特 XOR、3 比特半加器);多跳成功率的解析界(只有逐边经验复合)。全部实验设计与判据在 §7 预注册,数据与代码见附录。
Author: chen, kaifa
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225793216
Cites:
Cited by:
Related to: