跳至正文

容错量子计算成真了吗

目录

母裁决(四件)

below-threshold 是真(守真锚):2024-12 Google 首次在表面码上演示「逻辑错误率随码距增大而指数下降」(Λ=2.14±0.02、101 物理比特距离-7 码每纠错周期 0.143%±0.003%,Nature 638, 920–926(2025));此后不到两年内,中科大(PRL 135, 260601(2025-12-22),d=7 全微波泄漏抑制、Λ=1.40(6))、哈佛/QuEra(Nature 2025-11-10,448 中性原子、四轮表征电路 2.14(13)×)等多个平台各自跨过阈值——这不是「零进展」,是三十年前阈值定理承诺的指数抑制第一次在硬件上成片复现。 ② 「容错时代已到来」是门槛跳(命门·独占核心):below-threshold 是判据在场、首格达成——但它是整条十格楼梯的第一格。Google 官方路线图自认「currently working towards Milestone 3 (M3), a long-lived logical qubit」(官方 PDF),而 M3–M6(逻辑门、规模化工程、百万比特机)一个都未完成;「below-threshold 达成 → 容错已实现」把第一格读成了终点。 ③ 报数口径在物理/逻辑比特间滑动(命门·jingle):一个距离-7 表面码逻辑比特占 101 个物理比特(Willow 正文逐字),实用容错逻辑比特约需 10³–10⁴ 物理比特(Fowler 2012 逐字)——「105 比特芯片」「12 个逻辑比特」「2,000 个逻辑比特」用的是不同标尺;2025 年流传的「IBM 126 逻辑比特」实际出自第三方公司 Automatski 的 LinkedIn 自述,与 IBM 无关(一手核查,见下)。 ④ 「全是炒作/永不可行」同样未立(镜像虚无跳):阈值定理是数学定理(KLZ 1998 逐字「arbitrarily accurate quantum computation is possible provided that the error per operation is below a threshold value」),2025–2026 年离子阱已演示逻辑错误率低于物理的盈亏平衡双比特非 Clifford 门(Quantinuum 2025-06-26),而「深 100 比特随机电路任何经典方法模拟仍然相当无望」是 Aaronson 2026-07-18 的公开立场——RCS 侧量子优势窗口没有重开经典反超。

灵魂句:below-threshold 是真的、逻辑比特是真的、阈值定理是真的——真的不是「容错时代已经到来」这一层被声称的胜利。判据在场、首格达成、全程未证;把第一格读成终点,是把一条十格的楼梯看成了一级台阶。


〇 母裁决:四件与结构胎记

为什么是这四个问题

本篇是量子霸权篇的纵深化:那篇裁「窄任务采样里程碑 vs 全面颠覆」(supremacy≠advantage≠FTQC,其 4.3 节只在尺度腿提了一句「below-threshold 是真砖块」);本篇把「容错」这整条链拆开——判据(阈值定理)→ 首格(below-threshold)→ 报数(逻辑/物理比特)→ 开销(蒸馏与解码)→ 路线图(2029/2028/2030)→ 替代路线(Majorana)→ 经典反攻——只问一件事:「判据在场且首格达成」之后,离「容错通用机」还有多远,以及各家用什么尺子报进度。

拓扑物质篇分界写死:那篇审 Majorana 零模、拓扑保护的凝聚态物理本质(准粒子存在性、拓扑相变的物理争议);本篇只把微软 Majorana 路线当作一种容错交付主张,审其「声明链」的每一环(发布会→论文→编辑注→质疑→撤稿史→Matters Arising),不裁物理对错。与计量学篇对偶:SI 把「知道自己测不准多少」写成义务;容错量子计算把「知道自己做到哪一格」写成路线图——区别在路线图的格子由造尺人自己画、自己报、自己核。

结构胎记:四跳

  • 门槛跳物理错误率低于阈值 → 逻辑错误率指数下降(判据,真且达成)被读成「容错已经实现」。实则 below-threshold 只是编码保持,完整链还有逻辑门、魔法态、算法级、规模化四格。
  • 报数跳逻辑量子比特物理量子比特两个标尺在宣传口径里互换——「105 比特芯片」是物理比特(只含 1 个 d=7 逻辑比特),「2,000 逻辑比特」是路线图目标(每个需 10³–10⁴ 物理比特)。报进度时换了尺。
  • 首格跳:路线图第一格达成(M1 2019 超经典、M2 2023 纠错、M2.5 2024 below-threshold)→「2029 年容错机」被读成既成时间表。实则日期是声明,不是交付;且 2026 年政府目标(DOE 2028)比 DARPA 自己的评估(2033 才可能出 utility-scale 机)早五年。
  • 镜像虚无跳:「全是炒作、量子纠错永不可行」同样未立——阈值定理数学为真、below-threshold 多平台复现为真、经典侧对深 100 比特电路的模拟「仍然相当无望」(Aaronson 2026-07-18)。

共用动作=把一条路的第一个里程碑读成整条路已经走完。这是本库「判据在场」家族的新形态:L5 篇是没人写判据、边界自定;固态电池篇是判据在场、口径缺席;本篇是判据在场、首格真的达成——所以这篇的刀要特别小心:不能因为「第一格之后还远」就说「第一格是假的」。

一 守真锚:below-threshold 链

1.1 Willow(2024-12-09)——首个 below-threshold 表面码

Google 官方博客逐字:"We tested ever-larger arrays of physical qubits, scaling up from a grid of 3x3 encoded qubits, to a grid of 5x5, to a grid of 7x7 — and each time, using our latest advances in quantum error correction, we were able to cut the error rate in half."(我们测试了越来越大阵列的物理量子比特,从 3×3 编码网格、5×5 网格到 7×7 网格——每一次,借助我们量子纠错的最新进展,我们都能把错误率减半)——blog.google [一手逐字亲核]。同文:"This historic accomplishment is known in the field as 'below threshold' — being able to drive errors down while scaling up the number of qubits."(这一历史性成就在学界被称为「低于阈值」——在扩大量子比特数目的同时把误差压下去);"As the first system below threshold, this is the most convincing prototype for a scalable logical qubit built to date."(作为第一个低于阈值的系统,这是迄今构建的最有说服力的可扩展逻辑量子比特原型)。

论文(arXiv:2408.13687,正式版 Nature 638, 920–926(2025))摘要逐字:"The logical error rate of our larger quantum memory is suppressed by a factor of Λ = 2.14 ± 0.02 when increasing the code distance by 2, culminating in a 101-qubit distance-7 code with 0.143% ± 0.003 per cent error per cycle of error correction. This logical memory is also beyond breakeven, exceeding the lifetime of its best physical qubit by a factor of 2.4 ± 0.3."(把码距增加 2 时,更大量子存储器的逻辑错误率被抑制 Λ=2.14±0.02 倍,最终达到 101 比特、距离-7 的码,每纠错周期错误率 0.143%±0.003%;该逻辑存储器还超越盈亏平衡,寿命为其最佳物理比特的 2.4±0.3 倍)[一手逐字亲核]。2026 年该文有一条勘误(Nature 653, E5(2026),2026-04-28)[多源交叉]。

任务书级纠错①:训练记忆中「ε3/ε5/ε7 ≈ 1.07%/0.49%/0.24%」未在原文取回。实测补充材料 Table S1(神经网络解码列):ε3 = 0.650%±0.006%、ε5 = 0.303%±0.003%、ε7 = 0.143%±0.003%(Libra 解码器列略高:0.71%/0.35%/0.17%)[一手逐字亲核]。1.07%/0.49%/0.24% 这组数字出处不明,本篇弃用。

1.2 多平台复现(2024-12 → 2026-08)

below-threshold 不是单点实验。逐条列账(全部一手):

  • 中科大(超导,2025-12-22)PRL 135, 260601「Experimental Quantum Error Correction below the Surface Code Threshold via All-Microwave Leakage Suppression」摘要逐字:"we demonstrate a quantum memory operating below the threshold by implementing an all-microwave leakage suppression architecture on a distance-7 surface code. We achieve a logical error suppression factor of Λ=1.40(6), definitively reversing the above-threshold scaling (Λ<1) caused by unmitigated leakage."(我们在距离-7 表面码上实现全微波泄漏抑制架构,演示了阈值以下的量子存储器;达到 Λ=1.40(6) 的逻辑误差抑制因子,明确逆转了未缓解泄漏造成的上阈值缩放 Λ<1)——注意这组工作的价值:它证明「未缓解的泄漏」本身会把表面码推回阈值以上(Λ<1),抑制泄漏后才跨过阈值。这是 Willow 论文没有细说的一个真实工程暗礁。[一手逐字亲核]
  • 哈佛/加州理工(中性原子,2025-11-10)Nature「Architectural mechanisms of a universal fault-tolerant quantum computer」摘要逐字:"We first use surface codes to study how repeated QEC suppresses errors, demonstrating 2.14(13)x below-threshold performance in a four-round characterization circuit by leveraging atom loss detection and machine learning decoding."(我们首先用表面码研究重复 QEC 如何抑制误差,借助原子丢失检测与机器学习解码,在四轮表征电路中演示了 2.14(13) 倍的低于阈值性能)——必须保留「four-round characterization circuit」限定:这组是在较短的表征电路中跨过阈值,与 Willow 的长程记忆实验结构不同,但抑制因子 2.14 与 Google 的 Λ=2.14 数字巧合相同。同文:"enabling deep-circuit protocols with dozens of logical qubits and hundreds of logical teleportations with [[7,1,3]] and high-rate [[16,6,4]] codes"(用 [[7,1,3]] 与高码率 [[16,6,4]] 码实现数十个逻辑比特、数百次逻辑隐形传态的深电路协议)。448 原子、图 6h 中最多 96 个 d=4 逻辑比特同时激活(后选择存在,见 Methods)。[一手逐字亲核]
  • Quantinuum(离子阱,2025-06-26):官方博客逐字:"Setting a record magic state infidelity was just the beginning. The paper also presents the first break-even two-qubit non-Clifford gate, demonstrating a logical error rate below the physical one."(创纪录的魔法态非保真度只是开始;论文还呈现了第一个盈亏平衡的双比特非 Clifford 门,证明逻辑错误率低于物理错误率)——quantinuum.com [一手逐字亲核]。论文(arXiv:2506.14688,2025-06-17,H1-1 平台 20 比特,[[6,2,2]] 码 8 物理比特):魔法态非保真度 7⁺³₋₁×10⁻⁵、逻辑 CH 门错误率 ≤2.3×10⁻⁴ 对物理 10⁻³。
  • IBM(超导,2026-02-27)Nature Communications 17, 3281([[4,2,2]] 码 + normalizer dynamical decoupling):"The fidelities we achieve are beyond-breakeven, i.e., they significantly exceed the corresponding fidelities of unprotected entangled qubits in the same setting."(我们达到的保真度超越盈亏平衡,即显著超过同一环境下未保护纠缠量子比特的对应保真度)——注意 IBM 至今未宣称 below-threshold;其 2025-11-12 官方口径是「demonstrates all hardware elements of fault-tolerant quantum computing」(新闻稿)。[一手逐字亲核]
  • 微软+Atom Computing(中性原子,2024-11)arXiv:2411.11822(v3 修订 2025-06-09):"We demonstrate the entanglement of 24 logical qubits encoded into 48 atoms... We also implement the Bernstein-Vazirani algorithm with up to 28 logical qubits encoded into 112 atoms, showing better-than-physical error rates."(我们演示了编码在 48 个原子中的 24 个逻辑比特的纠缠……还实现了最多 28 个逻辑比特编码在 112 个原子中的 Bernstein-Vazirani 算法,错误率优于物理比特)[一手逐字亲核]。

镜像虚无跳的反驳(不虚无化):把 above-threshold 的 2023(Google 2023 年论文 Λ≈1 附近、未能跨阈)到 2024-12 的 Λ=2.14,再到 2025-2026 四大平台(超导×2、离子阱、中性原子)各自跨阈,读成「量子纠错是骗局」需要无视一整套可复现的实验记录。本文明确不写「全是炒作」。

二 判据账:阈值定理与 below-threshold 的确切判据

2.1 阈值定理(1996–1998)——判据的原件

  • Preskill 1998arXiv:quant-ph/9712048,讲义版)摘要逐字:"In principle, an arbitrarily long quantum computation can be performed reliably, provided that the average probability of error per gate is less than a certain critical value, the accuracy threshold."(原则上,只要每个门出错的平均概率低于某个临界值——精度阈值,任意长的量子计算就能可靠地进行)正文第 5 节:Steane 7 位码阈值 "the threshold value of p0 is 1/21";门错误率粗略估计 "εgate,0 ~ 6·10⁻⁴"。[一手逐字亲核]
  • Aharonov & Ben-OrarXiv:quant-ph/9611025)摘要逐字:"We describe fault tolerant quantum computation when the error probability is smaller than some constant threshold. The cost is polylogarithmic in time and space... With our codes, the threshold is about 10⁻⁶."(我们描述错误概率低于某个常数阈值时的容错量子计算;时间与空间开销均为 polylogarithmic……用我们的码,阈值约为 10⁻⁶)[一手逐字亲核]
  • Knill, Laflamme & ZurekScience 279, 342–345(1998),全文镜像)摘要逐字:"It is shown that arbitrarily accurate quantum computation is possible provided that the error per operation is below a threshold value."(本文证明:只要每次操作错误低于一个阈值,任意精度的量子计算都是可能的)末句:"Whether resilient quantum computation can be implemented in practice remains to be seen. However, the results obtained here show that, in principle, noise of a level below the error threshold is not an obstacle for quantum computation."(弹性量子计算能否在实践上实现仍有待观察;但本文结果表明,原则上,低于错误阈值的噪声并非量子计算的障碍)——注意 1998 年作者自己就写了「remains to be seen」。其给出的阈值数字为 3×10⁻⁶(主估计)。[一手逐字亲核]

这三个原件的共同点:阈值定理是数学陈述,不承诺任何硬件在可预见时间内到达阈值——1996–1998 年的阈值估计(10⁻⁶ 量级)比当时任何硬件的错误率低十几个数量级。「定理为真」与「硬件到达阈值」之间隔了二十八年,这正是本篇判据账的核心结构。

2.2 表面码阈值与「below-threshold 的判据定义」

  • Fowler et al. 2012Phys. Rev. A 86, 032324(2012)——任务书级纠错②:该文发表于 PR A 86 而非 Rev. Mod. Phys. 84, 807):阈值与指数下降的判据句逐字:"For p < pth , the logical error rate falls exponentially with d, while for p > pth , PL increases with d. In Fig. 4... the threshold rate is pth = 0.57%."(p<p_th 时逻辑错误率随 d 指数下降,p>p_th 时 P_L 随 d 增大;图 4 模拟下阈值为 0.57%);引言里流行的「约 1%」是转引:"with a per-operation error rate as high as about 1% [22, 23]";0.57% 对应整个测量周期约 4%。任务书级纠错③:「表面码阈值约 1%」是 Fowler 转引 Raussendorf 等人的说法,Fowler 自己的模拟值是 0.57%。 overhead 经典表述逐字:"It takes a minimum of thirteen physical qubits to implement a single logical qubit. A reasonably fault-tolerant logical qubit that can be used effectively in a surface code takes of order 10³ to 10⁴ physical qubits."(实现单个逻辑比特至少需要 13 个物理比特;一个可在表面码中有效使用、合理容错的逻辑比特需要约 10³ 到 10⁴ 个物理比特)同段脚注:"Here we assume an error rate approximately one-tenth the threshold rate, which implies that we need about 14,500 physical qubits per logical qubit"(假设错误率约为阈值十分之一,则每个逻辑比特需要约 14,500 个物理比特)。[一手逐字亲核]
  • Willow 的 below-threshold 判据定义(正文第 I 节逐字):"If the physical operations are below a critical noise threshold, the logical error should be suppressed exponentially as we increase the number of physical qubits per logical qubit. This behavior is expressed in the approximate relation εd ∝ (p/pthr)^((d+1)/2)... with the error suppression factor Λ = εd/εd+2 ≈ pthr/p"(物理操作低于临界噪声阈值时,逻辑错误应随每逻辑比特物理比特数增加而指数抑制……抑制因子 Λ=εd/εd+2≈p_thr/p 表示码距每增 2 时逻辑错误率的下降倍数);"These processors demonstrate Λ > 2 up to distance-5 and distance-7, respectively."(这些处理器分别在距离-5 与距离-7 上展现 Λ>2)。[一手逐字亲核]

判据账小结:below-threshold 的判据是自己写的(Λ>2、逻辑错误率随距离指数下降),2024 年达成,2025–2026 多平台复现——判据在场、且首格真的达成。这与 L5(判据在场、无人达成)、固态电池(判据在场、口径缺席)都不同:本篇的刀不砍「判据假」,砍「首格之后的阶梯被折叠」。

三 报数账:物理比特与逻辑比特——三把尺子的换算

3.1 三套口径并存

  • 旋转表面码(Google 口径)"Here, d is the code distance indicating 2d² − 1 physical qubits used per logical qubit"(d 是码距,表示每个逻辑比特占用 2d²−1 个物理比特)——Willow 引言逐字;d=7 即 97 数据+测量(49 数据+48 测量)+4 泄漏移除 = 101 物理比特:"We operate a distance-7 surface code memory comprising 49 data qubits, 48 measure qubits, and 4 additional leakage removal qubits"。[一手逐字亲核]
  • 未旋转布局(Fowler 口径)nq = (2d − 1)²(d=5 时 81 比特)。
  • 含间隔的补丁(Gidney-Ekerå 口径)"We assume distance-d logical qubits are stored using square patches of 2(d+1)² physical qubits"(距离-d 逻辑比特存于 2(d+1)² 个物理比特的方形补丁中)——d=27 时每逻辑比特 1,568 物理比特。[一手逐字亲核]

同一个「逻辑比特」,三种公式给出不同物理比特数——这是报数账的第一层:宣传口径里的「比特数」几乎总是物理比特数,而「逻辑比特数」需要另行换算。

3.2 换算表(全部一手数字)

声明 物理比特 逻辑比特 出处
Willow「105 比特芯片」 105 1 个 d=7 表面码(101 比特)+ 对比用 d=3/d=5 arXiv:2408.13687
「12 个逻辑比特」(真实出处:微软+Quantinuum 2024-09-10) 56(H2 离子阱,双比特保真度 99.8%) 12(GHZ 纠缠,电路错误率 0.0011,为物理 0.024 的 22 分之一) 微软官方博客——任务书级纠错④:12 逻辑比特的真实出处是 2024-09-10 微软博客,不是记忆中的 2025-06-09(后者恰是 Atom Computing 论文 v3 修订日)
IBM Starling 2029 路线图「200 逻辑比特」 未承诺物理比特数;按 10³/逻辑比特口径隐含约 20 万+ 200(1 亿量子运算) IBM 2025-06-10
「126 逻辑比特」(2025 年流传) 任务书级纠错⑤:出处是第三方公司 Automatski 的 LinkedIn 自述(2026-01-02,声称用 126 逻辑比特分解 N=1,398,488,603),评论区即有「That’s a classical simulation」质疑;IBM 官方渠道无此纪录。本篇不引用该数字作任何承重用途。
微软 Majorana 1「8 个拓扑量子比特」 8(在按百万容量设计的芯片上) 0(未宣称逻辑比特) 微软 2025-02-19——任务书级纠错⑥:是 8 个,不是 12 个
哈佛/QuEra 448 原子 448 数十([[7,1,3]]/[[16,6,4]]),图 6h 最多 96 个 d=4 同时激活(含后选择) Nature 2025-11-10
Quantinuum「Helios 98 比特」 98(离子) 12(2026-07-29 Steane 码 QFT)+ 48 纠错/94 检错(2026-03-04 Iceberg) Quantinuum 博客Skinny 博客

3.3 换算的工程含义:1 逻辑比特的真实成本

  • Google 官方博客逐字:"At current physical error rates, we might need more than a thousand physical qubits per surface code grid to realize relatively modest encoded error rates of 10⁻⁶."(在当前物理错误率下,要实现相对温和的 10⁻⁶ 编码错误率,每个表面码网格可能需要超过一千个物理比特)——research.google/blog/making-quantum-error-correction-work [一手逐字亲核]
  • Willow Outlook 节逐字:"Extrapolating the projections in Fig. 1 d, achieving a 10⁻⁶ error rate would require a distance-27 logical qubit using 1457 physical qubits."(外推图 1d,达到 10⁻⁶ 错误率需要距离-27 逻辑比特,占用 1,457 个物理比特)[一手逐字亲核]
  • Gidney-Ekerå 2021(Quantum 5, 433):2048 位 RSA 分解整板 226 · 63 个逻辑比特(≈14,200),其中 "Approximately 25% of these qubits are being used for distillation"(约 25% 用于蒸馏);d=27 时每逻辑比特 1,568 物理比特,总计约 2,000 万物理比特。[一手逐字亲核]
  • Preskill & Eisert 2025 综述(arXiv:2510.19928):"roughly 10³ physical qubits per logical qubit once ancillary qubits and logical-gate overhead are included, or about 10⁶ physical qubits in total"(计入辅助比特与逻辑门开销后,每个逻辑比特约 10³ 个物理比特,总计约 10⁶ 个物理比特)[一手逐字亲核]

报数账的落点:当「105 比特芯片」「12 个逻辑比特」「200 逻辑比特 2029」「百万比特芯片」出现在同一篇报道里,它们分别是物理比特数、中等规模逻辑比特数、路线图目标、芯片设计容量——四个不同标尺。报进度时换尺,是这篇领域最常见的口径滑动。

四 开销账:从 below-threshold 到跑算法之间有什么

4.1 T 门(魔法态)蒸馏——被省略的大头

  • Bravyi & Kitaev 2005arXiv:quant-ph/0403025——任务书级纠错⑦:正确编号是 quant-ph/0403025,不是 0408176):T 型方案逐字:"The algorithm recursively iterates an elementary distillation subroutine that transforms five copies of an imperfect magic state into one copy having a smaller error probability."(该算法递归迭代一个基本子程序:把 5 份不完美的魔法态转换为 1 份误差更小的态)H 型方案:"It is based on a certain CSS stabilizer code that encodes one qubit into 15"(基于把 1 个量子比特编码进 15 个的 CSS 稳定子码);阈值 ≈ 0.655。[一手逐字亲核]
  • Bravyi & Haah 2012arXiv:1209.2426——任务书级纠错⑧):把 BK2005 的 15 比特 Reed-Muller 方案记为「15-to-1」:"the original distillation protocol of Ref. [13] based on the 15-qubit Reed-Muller code has a distillation cost O(log^γ (1/ε)), where γ = log₃(15) ≈ 2.47.";自己的 (3k+8)→k 协议 "takes as input 3k + 8 magic states with an error rate p and outputs k magic states with an error rate O(p²)";并构造第一个距离-5 横向 T 门码 [[49,1,5]]。[一手逐字亲核]
  • Fowler 2012 的工程数字(为什么 T 门是开销大头):"The ancilla |YL⟩ and |AL⟩ states used in the SL and TL gates... are not easy to perform in the surface code."(S_L 与 T_L 门使用的辅助态在表面码中不容易直接实现);"Each Toffoli gate consumes seven |AL⟩ states. In total, the exponentiation circuit therefore requires approximately 280N³ ≈ 2.2 × 10¹² |AL⟩ states."(每个 Toffoli 门消耗 7 个 |AL⟩ 态;整个幂模电路共需约 2.2×10¹² 个);"generating and purifying one |AL⟩ state takes a surface code area of about 800,000 physical qubits, and takes about 100 µs"(生成并纯化一个 |AL⟩ 态需要约 80 万物理比特的表面码面积,耗时约 100 微秒)[一手逐字亲核]——单个魔法态工厂的占地面积是当前全部实验芯片的数千倍。
  • Gidney-Ekerå 的蒸馏账"we need to distill approximately 3 billion CCZ states"(需蒸馏约 30 亿个 CCZ 态);整板 25% 逻辑比特用于蒸馏。[一手逐字亲核]

4.2 实时解码——「经典逻辑必须显著快于量子比特」

  • Fowler 2012 判据句:"the operation and error detection of the surface code assumes classical logic support, with the classical logic operating significantly faster than the qubits"(表面码的运行与检错假定有经典逻辑支持,且经典逻辑必须显著快于量子比特);"it is difficult to imagine performing the surface code classical processing if the rounds of error detection are applied much faster than 10⁶ to 10⁷ Hz." [一手逐字亲核]
  • Willow 实测:摘要逐字 "achieving an average decoder latency of 63 µs at distance-5 up to a million cycles, with a cycle time of 1.1 µs"(距离-5 上平均解码延迟 63 微秒,最多运行一百万周期,周期时间 1.1 微秒)——任务书级纠错⑨:63μs 是 distance-5 的解码延迟,不是 distance-7;且正文明确「does not yet include feedback into the logical circuit」(尚不包含对逻辑电路的反馈)。实时解码器 ε5=0.35%±0.01%、Λ=2.0±0.1,吞吐要求每周期 <1.1μs,对比神经网络解码器每周期 24μs。Google 博客补充:"We measure a decoder delay time of 50 to 100 microseconds on our device, and anticipate it will increase at larger lattice sizes. This delay could significantly impact the speed of error-corrected operation."(我们测得解码延迟 50–100 微秒,并预计更大晶格会增加;该延迟可能显著影响纠错运行速度)[一手逐字亲核]
  • AlphaQubit(2024-11-20)Nature 635, 834–840——任务书级纠错⑩:正确出处是 Nature 635, 834–840,arXiv 编号 2411.09685 实为无关的 Axion-Maxwell 场论论文):"AlphaQubit... decodes the Sycamore surface code experiments more accurately than any previous decoder"(比任何此前解码器更准确地解码 Sycamore 表面码实验);精度最高(Λ=1.056±0.010)但官方博客自承:"While AlphaQubit is great at accurately identifying errors, it's still too slow to correct errors in a superconducting processor in real time."(AlphaQubit 擅长精确识别错误,但仍太慢,无法在超导处理器上实时纠错)[一手逐字亲核]。后继 AlphaQubit 2arXiv:2512.07737,2025-12-08):"a real-time variant of AQ2 can decode both the surface code and colour code (up to 241 and 181 physical qubits respectively) in under 1 µs per cycle on commercial hardware... reaching logical error rates below 10⁻¹⁰ per cycle"(AQ2 的实时变体可在商用硬件上以每周期 1μs 以下解码表面码与 color code,分别多达 241 与 181 个物理比特;表面码与 color code 每周期逻辑错误率低于 10⁻¹⁰)[一手逐字亲核]。

4.3 需求端:跑一个算法要什么(Gidney 账本)

  • Gidney & Ekerå 2019/2021:物理假设逐字 "a characteristic physical gate error rate of 10⁻³, a surface code cycle time of 1 microsecond, and a reaction time of 10 microseconds"(特征物理门错误率 10⁻³、表面码周期 1 微秒、反应时间 10 微秒);结论 "This estimate implies that factoring a 2048 bit integer will take approximately 7 hours"(分解 2048 位整数约需 7 小时);表 4 精算 5.1 小时/运行、每运行 20 兆比特。任务书级纠错⑪:「8 hours」只出现在标题——正文粗估 7 小时、表 4 为 5.1 小时;2025 版论文摘要回顾时自己写 "I co-published an estimate stating that 2048 bit RSA integers could be factored in eight hours by a quantum computer with 20 million noisy qubits",即「8 小时/2000 万」是标题口径。[一手逐字亲核]
  • Gidney 2025-05-21arXiv:2505.15917):"I estimate that a 2048 bit RSA integer could be factored in less than a week by a quantum computer with less than a million noisy qubits. I make the same assumptions as in 2019: a square grid of qubits with nearest neighbor connections, a uniform gate error rate of 0.1%, a surface code cycle time of 1 microsecond, and a control system reaction time of 10 microseconds."(我估计 2048 位 RSA 整数可用不足一百万个噪声量子比特、在不到一周内分解;假设与 2019 相同)——需求端四年内降了 20 倍(2000 万 → 不足 100 万)。[一手逐字亲核]
  • Oratomic/Caltech 2026-03arXiv:2603.28627,含 J. Preskill、D. Bluvstein):标题即「Shor’s algorithm is possible with as few as 10,000 reconfigurable atomic qubits」(Shor 算法用低半径 LDPC 码、可重构原子比特最少只需 10,000 个)——一年内需求端再降两个数量级。[一手逐字亲核]

需求端的流动性:RSA-2048 的物理比特需求在 2021–2026 五年内从 20 亿(2015 年旧估)→ 2000 万 → 不足 100 万 → 1 万(原子平台口径),变动达四个数量级。这意味着「缺口还有多少倍」本身是一个快速移动的靶——本篇所有缺口数字都标注「截至 2026-08-02 的口径」。

4.4 一个隐藏的工程暗礁:相关错误突发

Willow 正文报告:重复码(repetition code)在逻辑错误率降到约 10⁻¹⁰ 处出现「每小时一次」的相关错误突发,成因未明(本段细节见原文 error-burst 节;A 捆与 C 捆均取回该段落,此处按摘要级转述并标注——该现象若在表面码上同样存在,将直接限制任何深电路算法)。这是 below-threshold 之外、首格账里最值得写进诚实空位的一条。[多源交叉]

五 首格账:第一格之后还有几格

5.1 Google 自己的格子:M1–M6

Google 官方 PDF(Willow 后现行版)逐字:"Our roadmap is shown below, and includes six technical milestones... Google is currently working towards Milestone 3 (M3), a long-lived logical qubit (that is, capable of "living" for 1 million computational steps with less than 1 error)."(路线图含六个技术里程碑……Google 正推进里程碑 3——长寿命逻辑量子比特,即能「存活」100 万计算步、错误少于 1 个)——services.google.com 官方 PDF [一手逐字亲核]。同文:"In 2019, we announced that we were the first to demonstrate 'beyond classical' performance (Milestone 1 on our roadmap)... In 2023, we announced the first-ever experimental demonstration that it is possible to scale quantum error correction (Milestone 2)."(2019 年 M1 超经典;2023 年 M2 首次实验演示量子纠错可扩展)。

首格账第一行:M1(2019 超经典)、M2(2023 可扩展纠错)、M2.5(2024 below-threshold)已达成;M3(长寿命逻辑比特)是官方自认的「currently working towards」;M4(逻辑门)、M5/M6(规模化、百万物理比特机)未完成。Google 自己都没有说第一格之后走完了。

2019 年原版路线图(《Our quantum computing journey》,ai.googleblog.com 原帖)已下线且 Wayback 无快照(CDX 查询为空),六里程碑内容只能从当前官方 PDF、2021 博客与二手综述重建——本文不引用 2019 原文的任何逐字,只在二手基础上标注 [多源交叉]。2021-05-18 博客逐字(blog.google):"we're on a journey to build 1,000,000 physical qubits that work in concert inside a room-sized error-corrected quantum computer"(我们正走在建造百万物理量子比特的路上);"we need to show we can encode one logical qubit — with 1,000 physical qubits"(需要证明 1 个逻辑比特可用 1,000 个物理比特编码)。

5.2 已交付的格子与缺口

截至 2026-08-02,各平台公开演示过的最大逻辑比特数:96 个 d=4 逻辑比特同时激活(哈佛/QuEra,含后选择,二手口径;一手原文为「dozens of logical qubits and hundreds of logical teleportations」)、Quantinuum Helios 48 个纠错逻辑比特(Iceberg 级联码,98 物理比特)、Google Willow 1 个 below-threshold 表面码逻辑比特(d=7)。对照:

需求 逻辑比特 物理比特 出处
RSA-2048(Gidney-Ekerå 2019 口径) ≈14,200(含 25% 蒸馏) 2,000 万 Quantum 5, 433
RSA-2048(Gidney 2025 口径) 约 1,000((0.5+ε)n) 不足 100 万 arXiv:2505.15917
早期容错化学(2025 视角) 25–100 arXiv:2506.19337
当前演示(2026-08 上限) 数十–96 105–448 见上文

码距维度:当前全部公开演示 d≤7;Willow 自估 10⁻⁶ 需 d=27(1,457 物理比特);Gidney 用 d=17/27 双层。缺口是「数十 vs 千余」约一到两个数量级(逻辑比特)与「d≤7 vs d≈27–35」(码距)——但需求端在快速流动(见 4.3)。

5.3 逻辑门:第一格之外的第一个硬格子

below-threshold 只是「编码保持」。逻辑门演示现状(2025–2026):

  • Google color code(2025-05-26)Nature 645, 614–619):"Scaling the code distance from three to five suppresses logical errors by a factor of Λ3/5 = 1.56(4). Simulations indicate this performance is below the threshold of the colour code... We test transversal Clifford gates with logical randomized benchmarking and inject magic states, a key resource for universal computation, achieving fidelities exceeding 99% with post-selection."(码距 3→5 抑制逻辑误差 1.56(4) 倍;模拟表明该性能低于 color code 阈值……我们用逻辑随机基准测试验证了横向量子比特 Clifford 门并注入魔法态——通用计算的关键资源——后选择下保真度超过 99%)——注意「post-selection」限定:保真度 >99% 是后选择下的。[一手逐字亲核] 博客补充:逻辑 Hadamard 门在 color code 约 20ns,表面码上可能慢 1,000 倍(research.google/blog/a-colorful-quantum-future)。
  • Google magic state cultivation(2025-12-15)arXiv:2512.13908,295 位作者):"Cultivation reduces the error by a factor of 40, with a state fidelity of 0.9999(1) (retaining 8% of attempts)."(cultivation 把误差降低 40 倍,态保真度 0.9999(1),保留 8% 的尝试)——保留率 8%:这是魔法态路线的当前真实效率。[一手逐字亲核]
  • Quantinuum 非 Clifford 门(2025-06-26):见 1.2——首个盈亏平衡双比特非 Clifford 门(逻辑 CH ≤2.3×10⁻⁴ 对物理 10⁻³);配套论文(arXiv:2506.14169,Phys. Rev. X 2025-09-18):码切换魔法态非保真度 ≤5.1(2.7)×10⁻⁴,「below the pseudo-threshold」。
  • 中科大 Wukong(2025-11-14)npj Quantum Information 11, 177):距离-2 表面码上实现逻辑 CNOT + 任意单比特旋转 + 逻辑门隐形传态(通用逻辑门集)[一手·摘要级]。
  • 2026 年的新格子:超导平台表面码晶格手术逻辑操作(arXiv:2606.06598,一对距离-3 表面码逻辑比特、每周期错误率 0.0365(2)/0.0282(1)、确定性制备逻辑 Bell 态、逻辑级 Deutsch-Jozsa 算法);离子阱跨码晶格手术真多体纠缠(arXiv:2607.04227,2026-07-05);Quantinuum H2 非阿贝尔任意子编织+融合实现通用拓扑门集(Nature s41586-026-10709-y,2026-07-15,54 比特 S3 量子双群基态、拓扑制备魔法态)。[一手·摘要级/预印本级]

首格账的落点:截至 2026-08-02,「below-threshold 保持」已多平台达成;「容错逻辑门」正在逐个演示(Clifford 先行、非 Clifford 破冰、后选择与保留率仍在);「算法级运行」(千余逻辑比特×数十亿门)没有任何平台接近。Willow 论文 Outlook 末句是最好的自我陈述:"With below-threshold surface codes, we have demonstrated processor performance that can scale in principle, but which we must now scale in practice."(借助低于阈值的表面码,我们展示了原则上可扩展的处理器性能,但现在必须在实践中扩展它)[一手逐字亲核]。

5.4 Preskill-Eisert 的四道缺口

Preskill & Eisert 2025(arXiv:2510.19928,v3 2026-05-21):"We identify four related hurdles along the road ahead: (i) from error mitigation to active error detection and correction, (ii) from rudimentary error correction to scalable fault tolerance, (iii) from early heuristics to mature, verifiable algorithms, and (iv) from exploratory simulators to credible advantage in quantum simulation."(我们识别了四道相关障碍:(i) 从纠错缓解到主动检错与纠错;(ii) 从初级纠错到可扩展容错;(iii) 从早期启发式到成熟、可验证的算法;(iv) 从探索性模拟器到量子模拟中的可信优势);"The path from NISQ to FASQ is likely to be arduous, expensive, and prolonged."(从 NISQ 到 FASQ 的道路可能是艰难、昂贵且漫长的)[一手逐字亲核]。

六 路线图账:各家时间表与兑现史

6.1 IBM:唯一把「承诺—交付」做成可核对年表的公司

  • 2020-09-15 首张路线图逐字:"Our team is developing a suite of scalable, increasingly larger and better processors, with a 1,000-plus qubit device, called IBM Quantum Condor, targeted for the end of 2023."(Condor 1000+ 比特,2023 年底);"Next year, we'll debut our 127-qubit IBM Quantum Eagle processor."(Eagle 127);"In 2023, we will debut the 1,121-qubit IBM Quantum Condor processor"——ibm.com/quantum/blog/ibm-quantum-roadmap [一手逐字亲核]。
  • 兑现史(对照一手新闻稿):Eagle 2021-11-16 发布 ✓、Osprey 433 2022-11-09 发布 ✓、Condor 1121 2023-12-04 引入 ✓、Heron 133 与 System Two 2023-12-04 上云/上线 ✓——前四张图全如期。2023-12-04 更新路线图延伸至 2033:"in 2029, we hit an inflection point: executing 100 million gates over 200 qubits with our Starling processor... This is followed by Blue Jay, a system capable of executing 1 billion gates across 2,000 qubits by 2033."(2029 年 Starling 200 比特 1 亿门;2033 年 Blue Jay 2,000 比特 10 亿门)[一手逐字亲核];2023-05-21 另设「100,000-qubit system by 2033」目标(IBM 博客)。
  • 2025-06-10 更新逐字:"By 2029, we will deliver IBM Quantum Starling — a large-scale, fault-tolerant quantum computer capable of running quantum circuits comprising 100 million quantum gates on 200 logical qubits."(2029 年交付 Starling——200 逻辑比特、1 亿门的容错机);"we expect to achieve quantum advantage sooner—by 2026."(2026 年量子优势);自评 "so far, we have successfully delivered on each of our milestones."(迄今每个里程碑都如期交付——卖方自评)[一手逐字亲核]。2025-11-12:「advantage by the end of 2026, fault-tolerant quantum computing by 2029」+ 实时解码「completed one year ahead of schedule」。
  • 漂移记录:2022 年版路线图承诺的 Kookaburra(2025 年,1,386 比特)滑到 2026;Crossbill/Flamingo 从路线图消失。第三方 tracker 记 IBM 1 次 missed、1 次 softened(见 6.5)。

6.2 Google:无具体年份,只有「本十年末」

Willow 博客(2024-12-09)没有给出 2027/2029/2030 任何年份;Neven 当天对媒体说第六里程碑机器「around the end of the decade」(HPCwire 二手)。2026-03-24 Neven 博客逐字:"We are now increasingly confident that commercially relevant quantum computers based on superconducting technology will become available by the end of this decade."(我们越来越有信心:基于超导技术的商用相关量子计算机将在本十年末可用)——同日宣布新增中性原子路线(blog.google/neutral-atom)[一手逐字亲核]。「Pichai 2029 目标」只有二手出处(quantumzeitgeist 2026-06-27 转述),无一手原文,不承重

6.3 微软:从「years, not decades」到 2029

  • 2023-06-21 六里程碑路线图(量子超算目标:"at least one million rQOPS with an error rate of at most 10⁻¹²")[一手]。
  • 2025-02-19 Majorana 1:"we are on track to build an FTP of a scalable quantum computer—in years, not decades—as part of the final phase of the DARPA US2QC program."(正按计划在「数年而非数十年」内建成可扩展量子计算机的容错原型机——DARPA US2QC 最后阶段)[一手逐字亲核]。
  • 2026-06-02 Majorana 2:"This rapid progress, enabled by AI, has cut our timeline in half for delivering a scalable quantum computer—now anticipated by 2029."(AI 助力的快速进展把我们交付可扩展量子计算机的时间表砍掉一半——现在预期 2029 年)[一手逐字亲核](本条博客正文逐字,见下第七章)。
  • 时间表变迁链:2023「十年内」→ 2025「数年而非数十年」→ 2026「2029」——唯一一家把年份明确写下来、又明确改早的公司;同期外部批评者 Henry Legg 的说法是「centuries, not decades」(见第七章)。

6.4 国家级计划:政府日期 vs 独立评估

  • 美国:NQI 法(2018)2023 年秋到期后进入再授权流程(S.3597 2026-01-08 提出、2026-04-14 参院委员会通过;H.R.8462 2026-04-29 众院委员会通过,拟并入 FY27 NDAA)congress.gov;2026-06-22 行政令 EO 14413(whitehouse.gov):要求 DOE 部署至少一台容错科学级量子计算机(QC-ADDS),白宫官员称「could be completed by 2028」;DOE 回应承诺「world’s first fault-tolerant, scientifically relevant quantum computer by 2028」(IEEE Spectrum 2026-06-30 二手)。对照:DARPA 官员 Micah Stoutimore 2026 年判断「someone will build a utility-scale quantum computer by 2033」(MIT Technology Review 2026-07-14 二手转述,精确链接未保留)——政府目标 2028 比 DARPA 自己评估的 2033 早五年。[二手·多源交叉,不承重为科学结论]
  • 欧盟:Quantum Flagship 2018 启动(预期预算 10 亿欧元)digital-strategy.ec.europa.eu;2025-07-02《Quantum Europe Strategy》(EUR-Lex COM(2025)363):"the EU has invested nearly EUR 2 billion in quantum technologies, complemented by more than EUR 9 billion in additional public funding from Member States";2030 目标「full-stack quantum computing capability」;2026 年拟推 Quantum Act。
  • 中国:十五五规划纲要(2026-03-12 通过,longbridge 转述纲要原文):"develop fault-tolerant universal quantum computers and scalable dedicated quantum computers"(发展容错通用量子计算机和可扩展专用量子计算机);中科院官方口径(国新办 2025-03-04):三步走路线图(量子霸权 → 数百比特模拟器 → 容错通用机),朱晓波:「currently conducting surface code error correction research with a code distance of 7. After making progress, they will extend it to 9 and 11.」(当前做码距 7 的表面码纠错研究,有进展后扩展到 9、11)。

6.5 第三方兑现率审计

  • Quantum Outpost Vendor Roadmap Trackerquantumoutpost.com/roadmaps,2026-05-24 验证):按 (delivered + 0.6×late) ÷ past-due claims 排序——Quantinuum 100%、PsiQuantum 100%、Atom Computing 100%、QuEra 100%、Pasqal 100%、Xanadu 80%、IBM 73%、Google 72%、IonQ 60%、D-Wave 57%、Microsoft(拓扑路线)52%、Rigetti 33%。评语:IBM「mostly hits qubit-count targets; error-correction milestones have slipped」;微软「Twenty years of research, multiple high-profile retractions… The Majorana 1 announcement remains controversial — the underlying physics claims have not been independently replicated.」[二手·但可逐条溯源]
  • PostQuantum 2025-09-11 综述"The gap between roadmap and reality remains large — but it's shrinking... Roadmaps are converging on the late 2020s for fault-tolerance demonstrations."(路线图与现实的差距仍然很大——但在缩小;路线图正收敛到 2020 年代末的容错演示)[二手]。
  • Olivier Ezratty(2024-12)对 Willow 的冷静估价"Moving from 53 to 105 qubits in 5 years is far from being some exponential progress, particularly given millions of physical qubits are required to do something useful."(五年从 53 到 105 比特远非指数级进展,尤其考虑到要做有用的事需要数百万物理比特)[二手·独立专家]。

路线图账的落点:路线图日期(2028/2029/2030)全部是声明不是交付;历史兑现率最高的是「比特数承诺」(IBM 全如期),最弱的是「容错里程碑」;2026 年唯一一家把年份改早的是 Quantinuum(2030→2029)与微软(→2029),而同一时期 DARPA 的独立评估(2033)比所有公司声称都晚。

七 Majorana 账:微软拓扑路线的验证链

7.1 声明链的每一环

  • 2025-02-19 发布会Azure Quantum 博客,Chetan Nayak 署名)逐字:"Today we hit our second milestone, demonstrating the world's first topological qubit. And we've already placed eight topological qubits on a chip designed to house one million."(今天我们达成第二里程碑:演示世界首个拓扑量子比特;已在按百万容量设计的芯片上放置了 8 个拓扑量子比特)[一手逐字亲核]。官方新闻稿(Microsoft Source):"The world's first Topological Core powering the Majorana 1 is reliable by design, incorporating error resistance at the hardware level making it more stable."(Majorana 1 的拓扑核心「可靠源于设计」,硬件层内建抗错误能力)[一手逐字亲核]。
  • 同日 Nature 论文Nature 638, 651–655(2025),arXiv:2401.09549——任务书级纠错⑫:arXiv 编号 2401.09549(2024-01-17 提交),不是 2505.05891(后者是无关的过渡金属二硫属化物论文)):InAs-Al 异质结构单次干涉奇偶性测量,"we extract a dwell time in the two associated states that is longer than 1 ms at in-plane magnetic fields of approximately 2 T"(约 2T 面内磁场下驻留时间超过 1ms);"enable a parity measurement with an assignment error probability of 1%"(赋值错误概率 1%)[一手逐字亲核]。关键自限句(摘要级转述,正文逐字)"By itself, this measurement does not unequivocally distinguish between MZMs in the topological phase and fine-tuned low-energy Andreev bound states in the trivial phase."(单凭这一测量无法明确区分拓扑相中的 MZM 与平凡相中微调的低能 Andreev 束缚态)[一手逐字亲核]。
  • Nature 编辑注(同行评审文件,经 Aaronson 博客 2025-02-20 全文收录):"The editorial team wishes to point out that the results in this manuscript do not represent evidence for the presence of Majorana zero modes in the reported devices. The work is published for introducing a device architecture that might enable fusion experiments using future Majorana zero modes."(编辑团队希望指出:本文结果不构成所报告器件中存在 Majorana 零模的证据;发表此文是为了介绍一种或可支撑未来 MZM 融合实验的器件架构)[一手·编辑注全文]。而微软官方新闻稿同日写的是「peer-reviewed confirmation that Microsoft has not only been able to create Majorana particles」——与编辑注直接矛盾(Aaronson 点名此缝隙:"their claim to have created a topological qubit has not yet been accepted by peer review")。[一手逐字亲核]

7.2 背景链:2018 → 2021 撤稿

  • 2018-03-28 Nature(Zhang et al.「Quantized Majorana conductance」,Delft/微软合作)→ 2020-04 Nature 581, E4 发表关切声明 → 2021-03-08 撤稿Nature 591, E30 撤稿通知逐字):"several inconsistencies were pointed out by Sergey Frolov and Vincent Mourik between the raw measurement data that was made available to them and the figures that were published in the paper... The new conductance calibration shifted the plateau values by 8 per cent, above 2e2/h... We can therefore no longer claim the observation of a quantized Majorana conductance, and wish to retract this Letter."(Frolov 与 Mourik 指出原始数据与论文图件不一致……新标定使平台值上移 8%,超过 2e2/h……我们不能再声称观测到量子化 Majorana 电导,故请求撤稿)[一手逐字亲核]。任务书级纠错⑬:这次撤稿不是「主动」——Frolov/Mourik 声明(Retraction Watch 转述):撤稿通知「omits that the two of us have contacted Nature, not the authors」。Delft 独立调查结论:无造假证据,但「作者自我欺骗风险未加防范」(Quanta Magazine 2021-09-29 二手转述,精确链接未保留)。
  • 2023 链路:微软 TGP 声称(PRB 107, 245423「InAs-Al hybrid devices passing the topological gap protocol」,模拟 349 器件零假阳性)→ Hess, Legg, Loss, Klinovaja 质疑(PRL 130, 207001(2023):平庸 Andreev 带可模拟 TGP 所需的能隙闭合/重开信号)→ 微软反驳(arXiv:2307.15813)→ 再反驳(arXiv:2308.10669)→ PRL 2024 官方 Comment/Reply 收尾(Das Sarma & Pan 站微软侧 vs Hess et al.,PRL 132, 099601/099602)——2023 拉锯实质未分胜负

7.3 2026-06-24:Nature Matters Arising 正式对垒(本篇证据截止前最后一个大事件)

  • Legg 批评Nature,DOI 10.1038/s41586-026-10567-8):核心指控——"Simply shifting measurement windows can alter the protocol's outcome. This causes Microsoft's software to classify the exact same device region as either suitable for quantum computing ('gapped') or not suitable ('gapless') simply because of arbitrary measurement choices."(仅移动测量窗口即可改变协议结果——同一器件区域可因任意测量选择被分类为「有能隙」或「无能隙」;[techxplore 转述])。两处具体编码错误:TGP 绘图软件硬编码过滤器 zbp_cluster_numbers=[1] 只显示最大区域;软件按数组索引(x[::-1])而非物理偏压值翻转数据(The Register 2026-06-24)。Legg 的结论性表述:"I think it's more like centuries, not decades."(我想更像是几个世纪,而不是几十年)[二手·采访引语]。
  • 微软同日回应Nature,DOI 10.1038/s41586-026-10568-7):"Legg centers on a selective examination of transport tune-up procedures... It relies on unsubstantiated claims about our transport spectra while not engaging with the capacitance measurements at the core of our study... Legg offers no alternative physical model capable of reproducing the capacitance signal"(Legg 聚焦于对传输调试程序的片面审视……依赖对我们传输谱的无根据断言,却未面对作为本研究核心的电容测量……未提出能复现电容信号的替代物理模型)——并承认一处:"Legg identified a minor off-by-one-pixel bug in our TGP processing. The correction shifts extracted gap values by less than 5 µeV for more than 96% of pixels... The region used for parity readout remains classified as gapped after the fix."(Legg 在我们的 TGP 处理中发现了一个小的 off-by-one 像素 bug;修正使 96% 以上像素的提取能隙值移动不足 5 μeV;修正后用于奇偶性读出的区域仍被分类为有能隙)[一手逐字亲核]。
  • Frolov 的第三视角Scientific American 2026-06-24):"The 'Matters Arising' makes it painfully apparent that the paper in Nature has no scientific value... And that it likely needs to be retracted, like the other Nature papers associated with Microsoft."(这篇 Matters Arising 痛苦地表明该 Nature 论文没有科学价值……很可能需要像微软其他 Nature 论文一样被撤稿)[二手·引语]。
  • Majorana 2(2026-06-02,提前发布)官方博客逐字:"Majorana 2 contains qubits that are 1,000x more reliable than those in our previous quantum processing unit. The new material stack, which swaps aluminum for lead, creates highly reliable topological qubits with operations on the microsecond scale and lifetimes with a mean of 20 seconds, occasionally exceeding one minute."(Majorana 2 的量子比特比上一代可靠 1,000 倍;新材料堆栈用铅替换铝,平均寿命 20 秒、偶超一分钟);"In the aluminum-based Majorana 1, qubit lifetimes were between one and 12 milliseconds"(Majorana 1 的寿命在 1–12 毫秒之间);时间表 "we have cut our timeline in half and now aim to reach this target by 2029"(时间表砍半,目标 2029)[一手逐字亲核]。技术稿为预印本(arXiv:2606.03884,「20 Second Parity Lifetime in an InAs–Pb Tetron Device」,未同行评审)。Legg 对 Majorana 2 的即时回应"'1,000 times more reliable' refers to the lifetime of a classical bit (the parity of the state). There is no evidence this is a qubit and can hold a superposition."(「可靠 1,000 倍」指的是一个经典比特——状态奇偶性——的寿命;没有证据表明这是能保持叠加态的量子比特)[二手·采访引语]。

Majorana 账的落点(不裁物理对错):截至 2026-08-02,微软「世界首个拓扑量子比特」声明链的每一环都有一处以上的独立缺口——发布声明与 Nature 编辑注直接矛盾、论文自承无法区分拓扑/平庸解释、2018 年同一团队的旗舰结果已撤稿、2026 年 Matters Arising 指控软件编码错误(微软部分承认)、核心操作(X 测量/相干叠加)至今无同行评审发表、无任何独立小组复现。这与超导/离子阱/中性原子三条路线的验证链(arXiv→同行评审→多平台复现)形成结构性对比——本篇把它作为「一种容错交付主张的验证链完整性」审计,不裁决「Majorana 零模是否存在」这一凝聚态物理问题(归拓扑物质篇)。

八 经典反攻与「优势窗口」:与霸权篇的接口

8.1 RCS 侧:窗口没有重开

  • Willow RCS 声称(官方博客逐字):"Willow performed a standard benchmark computation in under five minutes that would take one of today's fastest supercomputers 10 septillion (that is, 10²⁵) years — a number that vastly exceeds the age of the Universe."(Willow 在五分钟内完成基准计算,据估计需要今天最快的超算 10²⁵ 年)[一手逐字亲核]。Aaronson 引 Google 侧数据:最优张量网络算法下该实验经典模拟约 3 亿年(无内存限制)或 10²⁵ 年(有内存限制)(scottaaronson.blog/?p=8525,2024-12-10)。
  • 祖冲之 3.0(PRL 134, 090601(2025-03-03)——任务书级纠错⑭:arXiv 编号是 2412.11924,不是 2503.03088):"This task is estimated to be infeasible on the most powerful classical supercomputers, Frontier, which would require approximately 5.9 × 10⁹ years to replicate the task."(Frontier 需约 59 亿年复现该任务);105 比特芯片、两比特门 99.62%[一手逐字亲核]。
  • 2025-10 起经典侧的理论反攻(噪声可模拟性):Lee et al.(arXiv:2510.06328)基于条件互信息指数衰减的经典模拟算法;Zhang et al.(arXiv:2510.06324):"we present a classical algorithm that runs in n^{polylog(n)} time for simulating quantum circuits under local depolarizing noise, thereby ruling out their quantum advantage in these settings."(局部去极化噪声下 n^{polylog(n)} 时间经典模拟算法,排除了这些设定下的量子优势);Go et al.(arXiv:2607.20804,2026-07-23):给标准局部去极化噪声的可模拟/不可模拟边界。Aaronson 的反驳2026-07-18):噪声可模拟定理的算法「still needs time that’s exponential in the circuit depth」;"Sampling-based quantum supremacy experiments... passed the point about two years ago where, absent a breakthrough in classical algorithms, they quite clearly are beating what can easily be simulated on any existing classical computer."(采样类霸权实验约两年前已越过任何现役经典机可轻易模拟的范围);"simulating deep ~100-qubit random circuits... still seems pretty hopeless with any current classical method."(深 100 比特随机电路对现有经典方法仍相当无望)。
  • 2026-07 的公开论战:Hagar《The NISQ Trap》(arXiv:2607.07530):"every NISQ-era flagship demonstration of 'quantum advantage' has, within eighteen months of its announcement, been classically reproduced, shown to rest on classically tractable structure, or closed by a simulability theorem"(每个 NISQ 时代旗舰「量子优势」演示都在宣布后 18 个月内被经典复现、被证明依赖经典可处理结构、或被可模拟性定理封堵)——注意其自设例外:唯一例外是 Google 2025-10-22 的 Quantum Echoes(可验证量子优势,Nature 646, 825 附近;Google 官方称 13,000 倍加速,research.google/blog/a-verifiable-quantum-advantage)。这是一条活的战线(双方对同一批定理解读相反),本篇不裁决。

8.2 为什么「优势」不是本篇的主角

本篇与霸权篇的接口在此收口:RCS 优势属于霸权篇(窄任务采样 vs 经典);本篇只取一个相关事实——2026 年的经典侧理论(噪声可模拟性)即使成立,也主要影响「采样类」优势叙事,不影响 below-threshold 逻辑比特的物理事实(那是对抗本地噪声的主动纠错,不是采样)。两个问题必须分开审:能不能比经典快纠错是否有效

九 母题收口:三问尺对照与灵魂句

尺子 判据在场? 首格达成? 全程完成? 口径由谁定?
容错量子计算(本篇) 在场且写得极细(阈值定理、Λ>2、d 与错误率的关系、码距需求) 达成(below-threshold 2024–2026 多平台) 未完成(M3–M6 未交付、逻辑门后选择/保留率仍在、算法级相差 1–2 个数量级) 造尺人自己(阈值判据、路线图年份、报数口径都是各家自定)
L5 自动驾驶(07-27 篇 在场(J3016 分级) 未达成(没有人在朝终点报进度) 未完成 被测方自己写边界
固态电池(07-27 篇 在场(USABC 目标表) 部分达成(交付的那支不是叙事里那支) 未完成 买方写判据、卖方报进度
核聚变(07-24 篇 在场(劳森判据) 部分达成(Q_plasma>1) 未完成 物理量(Q)在场、工程量(Q_eng)缺席
计量学(08-02 篇 在场且被做对 达成 达成 SI 制度(定义可复现、不确定度随行、尺子送出去比)

本篇在「判据在场而口径缺席」家族中的位置:L5/固态电池/核聚变是「判据在场、没人达成或口径缺席」;本篇是「判据在场、首格真的达成」——所以本篇的落点不是「判据是假的」,而是:首格达成被读成整条路完成,报数时换尺,日期表被读成交付。

灵魂句(定稿):below-threshold 是真的、逻辑比特是真的、阈值定理是真的——真的不是「容错时代已经到来」这一层被声称的胜利。判据在场、首格达成、全程未证;把第一格读成终点,是把一条十格的楼梯看成了一级台阶。当有人把「2029 年容错机」说成既成时间表,问一句:第一格之后那九格,哪一格有同行评审的交付记录?

十 诚实空位与未决清单

  1. Willow 相关错误突发:重复码在 ~10⁻¹⁰ 处每小时一次的相关错误突发,成因未明——若在表面码同样存在将直接限制深电路(原文段落已取回,机制解释空缺)。
  2. 2019 年 Google 原版路线图逐字:原帖下线、Wayback 无快照,六里程碑仅能二手重建,本文不引用其任何逐字。
  3. 「Pichai 2029 目标」:只有二手转述,无一手原文,不承重。
  4. McKinsey「90% 价值 2035 前不会实现」:常用引文,但一手 PDF 未取回该句,本文仅引用已核实的 $700B/2035 与 $90B+/2040 两数(2022 版)与 2026 版 $2.7T/2035;该表述本身按二手处理。
  5. IBM Starling 的 qLDPC 制造可行性:FT 报道称分析师认为「works in theory but has yet to be proven in manufacturing」,无一手制造数据。
  6. 微软 X 测量(相干叠加证据):至 2026-08-02 无同行评审发表;Majorana 2 技术稿为未审预印本。
  7. DARPA 评估内容:微软反复援引 US2QC 最后阶段入选,但评估细节(含专有数据)不公开,无法作为科学证据链一环审计。
  8. Hagar vs Aaronson 论战(2026-07):RCS 优势窗口是否被噪声可模拟性定理关闭,双方对同一批定理解读相反,本篇不裁决。
  9. 量化缺口数字的时效:RSA-2048 物理比特需求 5 年内变动四个数量级(20 亿→2000 万→<100 万→1 万原子平台口径),本篇所有缺口数字截至 2026-08-02。
  10. IBM「below-threshold」状态:IBM 官方未宣称,第三方口径未采用。
  11. 「12 逻辑比特 2025-06-09」:无任何来源支撑,已改锚 2024-09-10 微软博客。
  12. Automatski「126 逻辑比特」:第三方 LinkedIn 自述、评论区有经典模拟质疑,本篇不承重。

关键来源(编号 1–62)

判据账(阈值定理与判据定义)

  1. Preskill, Fault-tolerant quantum computation(1998),arXiv:quant-ph/9712048 — [一手逐字亲核] https://arxiv.org/abs/quant-ph/9712048
  2. Aharonov & Ben-Or, Fault tolerant quantum computation with constant error rate,arXiv:quant-ph/9611025 — [一手逐字亲核] https://arxiv.org/abs/quant-ph/9611025
  3. Knill, Laflamme & Zurek, Resilient quantum computation, Science 279, 342–345(1998)— [一手逐字亲核](全文镜像 PDF)https://courses.physics.illinois.edu/phys513/sp2016/reading/week12/ResilientQCompScience1998-1.pdf · DOI http://web.archive.org/web/1998id_/https://www.science.org/doi/10.1126/science.279.5349.342
  4. Fowler, Mariantoni, Martinis & Cleland, Surface codes, Phys. Rev. A 86, 032324(2012)— [一手逐字亲核] https://arxiv.org/abs/1208.0928 · http://web.archive.org/web/2014id_/https://journals.aps.org/pra/abstract/10.1103/PhysRevA.86.032324
  5. Acharya et al.(Google Quantum AI), Quantum error correction below the surface code threshold, Nature 638, 920–926(2025)— [一手逐字亲核] https://www.nature.com/articles/s41586-024-08449-y · arXiv https://arxiv.org/abs/2408.13687 · 勘误 https://doi.org/10.1038/s41586-026-10559-8

守真锚(below-threshold 多平台)

  1. Google Quantum AI 博客 Willow(2024-12-09)— [一手逐字亲核] https://blog.google/innovation-and-ai/technology/research/google-willow-quantum-chip/
  2. Google Research 博客 Making quantum error correction work(2024-12-09)— [一手逐字亲核] https://research.google/blog/making-quantum-error-correction-work/
  3. He et al.(USTC), Experimental QEC below the surface code threshold via all-microwave leakage suppression, PRL 135, 260601(2025-12-22)— [一手逐字亲核] http://web.archive.org/web/2026id_/https://journals.aps.org/prl/abstract/10.1103/rqkg-dw31
  4. Bluvstein et al.(Harvard/QuEra/MIT), Architectural mechanisms of a universal fault-tolerant quantum computer, Nature(2025-11-10)— [一手逐字亲核] https://www.nature.com/articles/s41586-025-09848-5
  5. 3000 原子连续运行, Nature(2025-09-15)— [一手逐字亲核] https://www.nature.com/articles/s41586-025-09596-6
  6. Microsoft + Atom Computing, Entanglement of 24 logical qubits, arXiv:2411.11822(v3 2025-06-09)— [一手逐字亲核] https://arxiv.org/abs/2411.11822
  7. Quantinuum 博客 Overcomes last major hurdle… by 2029(2025-06-26)+ 论文 arXiv:2506.14688 / 2506.14169 — [一手逐字亲核] https://www.quantinuum.com/blog/quantinuum-overcomes-last-major-hurdle-to-deliver-scalable-universal-fault-tolerant-quantum-computers-by-2029
  8. Quantinuum 博客 Scaling the signal(2026-07-29)— [一手逐字亲核] https://www.quantinuum.com/blog/scaling-the-signal-what-a-larger-qft-says-about-quantum-progress
  9. Microsoft 博客 12 logical qubits(2024-09-10)— [一手逐字亲核] https://azure.microsoft.com/en-us/blog/quantum/2024/09/10/microsoft-and-quantinuum-create-12-logical-qubits-and-demonstrate-a-hybrid-end-to-end-chemistry-simulation/
  10. IBM, Entangled logical qubits beyond breakeven, Nat. Commun. 17, 3281(2026-02-27)— [一手逐字亲核] https://www.nature.com/articles/s41467-026-70011-3
  11. IBM 新闻稿 Starling(2025-06-10)— [一手逐字亲核] https://newsroom.ibm.com/2025-06-10-IBM-Sets-the-Course-to-Build-Worlds-First-Large-Scale,-Fault-Tolerant-Quantum-Computer-at-New-IBM-Quantum-Data-Center
  12. IBM 新闻稿 Loon/Nighthawk(2025-11-12)— [一手逐字亲核] https://newsroom.ibm.com/2025-11-12-ibm-delivers-new-quantum-processors,-software,-and-algorithm-breakthroughs-on-path-to-advantage-and-fault-tolerance

开销账

  1. Bravyi & Kitaev, Universal quantum computation with ideal Clifford gates and noisy ancillas, PR A 71, 022316(2005)— [一手逐字亲核] https://ar5iv.labs.arxiv.org/html/quant-ph/0403025
  2. Bravyi & Haah, Magic state distillation with low overhead, PR A 86, 052329(2012)— [一手逐字亲核] https://arxiv.org/abs/1209.2426
  3. Gidney & Ekerå, How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits, Quantum 5, 433(2021)— [一手逐字亲核] https://quantum-journal.org/papers/q-2021-04-15-433/ · arXiv https://arxiv.org/abs/1905.09749
  4. Gidney, How to factor 2048 bit RSA integers with less than a million noisy qubits(2025-05-21)— [一手逐字亲核] https://arxiv.org/abs/2505.15917
  5. Google DeepMind, AlphaQubit, Nature 635, 834–840(2024-11-20)— [一手逐字亲核] https://www.nature.com/articles/s41586-024-08148-8
  6. Google DeepMind 博客 AlphaQubit(2024-11-20)— [一手逐字亲核] https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphaqubit-quantum-error-correction/
  7. AlphaQubit 2, arXiv:2512.07737(2025-12-08)— [一手逐字亲核] https://arxiv.org/abs/2512.07737
  8. Preskill & Eisert, Mind the gaps, arXiv:2510.19928(2025-10, v3 2026-05-21)— [一手逐字亲核] https://www.alphaxiv.org/abs/2510.19928
  9. Google magic state cultivation, arXiv:2512.13908(2025-12-15)— [一手逐字亲核] https://arxiv.org/abs/2512.13908
  10. Google color code, Nature 645, 614–619(2025-05-26)— [一手逐字亲核] https://www.nature.com/articles/s41586-025-09061-4 · 博客 https://research.google/blog/a-colorful-quantum-future/
  11. Oratomic/Caltech, Shor’s algorithm with 10,000 reconfigurable atomic qubits, arXiv:2603.28627(2026-03)— [一手逐字亲核] https://arxiv.org/html/2603.28627v1

路线图账

  1. Google Quantum AI 官方 PDF(现行路线图)— [一手逐字亲核] https://services.google.com/fh/files/misc/google_quantum_ai_about.pdf
  2. Google 博客 Unveiling our new quantum AI campus(2021-05-18)— [一手逐字亲核] https://blog.google/innovation-and-ai/products/unveiling-our-new-quantum-ai-campus/
  3. Neven 博客 Neutral atom(2026-03-24)— [一手逐字亲核] https://blog.google/innovation-and-ai/technology/research/neutral-atom-quantum-computers/
  4. IBM 路线图 2020-09-15 — [一手逐字亲核] https://www.ibm.com/quantum/blog/ibm-quantum-roadmap
  5. IBM 路线图 2023-12-04(2033 延伸)— [一手逐字亲核] https://www.ibm.com/quantum/blog/quantum-roadmap-2033
  6. IBM 博客 100k qubit supercomputer(2023-05-21)— [一手逐字亲核] https://www.ibm.com/quantum/blog/100k-qubit-supercomputer
  7. IBM 博客 large-scale-ftqc(2025-06-10)— [一手逐字亲核] https://www.ibm.com/quantum/blog/large-scale-ftqc
  8. Microsoft 博客 Majorana 1(2025-02-19)— [一手逐字亲核] http://web.archive.org/web/2025id_/https://azure.microsoft.com/en-us/blog/quantum/2025/02/19/microsoft-unveils-majorana-1-the-worlds-first-quantum-processor-powered-by-topological-qubits/
  9. Microsoft Source 新闻稿 Majorana 1(2025-02-19)— [一手逐字亲核] https://news.microsoft.com/source/features/innovation/microsofts-majorana-1-chip-carves-new-path-for-quantum-computing/
  10. Microsoft 博客 Majorana 2(2026-06-02)— [一手逐字亲核] https://quantum.microsoft.com/en-us/insights/blogs/majorana-2-scalable-quantum-processor
  11. Aasen et al.(Microsoft), Roadmap to fault tolerant quantum computation using topological qubit arrays, arXiv:2502.12252 — [一手逐字亲核] https://arxiv.org/abs/2502.12252
  12. 美国 EO 14413(2026-06-22)— [一手逐字亲核] https://www.whitehouse.gov/presidential-actions/2026/06/ushering-in-the-next-frontier-of-quantum-innovation/
  13. NQI 再授权 S.3597 — [一手·文本] http://web.archive.org/web/2026id_/https://www.congress.gov/bill/119th-congress/senate-bill/3597/text
  14. 欧盟 Quantum Europe Strategy, COM(2025)363(2025-07-02)— [一手逐字亲核] http://web.archive.org/web/2025id_/https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A52025DC0363
  15. 欧盟 Quantum Flagship 官方页 — [一手逐字亲核] https://digital-strategy.ec.europa.eu/en/policies/quantum-technologies-flagship
  16. 中国 15 五规划纲要(longbridge 转述)— [二手转述] https://longbridge.com/en/news/279522188
  17. 国新办发布会(2025-03-04,朱晓波)— [一手·官方口译] http://web.archive.org/web/20250304100939id_/https://english.scio.gov.cn/chinavoices/2025-03/04/content_117744218.html
  18. Quantum Outpost Vendor Roadmap Tracker — [二手·可溯源] https://quantumoutpost.com/roadmaps/
  19. PostQuantum 路线图综述(2025-09-11)— [二手] https://postquantum.com/quantum-computing-companies/quantum-computing-companies-roadmaps/
  20. Ezratty, Inside Google Willow(2024-12)— [二手·独立专家] https://www.oezratty.net/wordpress/2024/inside-google-willow/

Majorana 账

  1. Aghaee et al.(Microsoft), Interferometric single-shot parity measurement in InAs-Al hybrid devices, Nature 638, 651–655(2025)— [一手逐字亲核] https://www.nature.com/articles/s41586-024-08445-2 · arXiv https://arxiv.org/abs/2401.09549
  2. Aaronson 博客 Majorana(2025-02-20,含 Nature 编辑注全文)— [一手·编辑注收录] https://scottaaronson.blog/?p=8669
  3. 2018 Nature 论文撤稿通知, Nature 591, E30(2021-03-08)— [一手逐字亲核] https://doi.org/10.1038/s41586-021-03373-x
  4. Hess, Legg, Loss & Klinovaja, Trivial Andreev band mimicking topological bulk gap, PRL 130, 207001(2023)— [一手·摘要级] http://web.archive.org/web/2023id_/https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.130.207001
  5. Antipov et al.(Microsoft)反驳, arXiv:2307.15813 + Legg et al. 再反驳 arXiv:2308.10669 — [一手·预印本级] https://arxiv.org/abs/2307.15813 https://arxiv.org/abs/2308.10669
  6. Aghaee et al.(Microsoft), InAs-Al hybrid devices passing the topological gap protocol, PRB 107, 245423(2023)— [一手·摘要级] http://web.archive.org/web/2023id_/https://journals.aps.org/prb/abstract/10.1103/PhysRevB.107.245423
  7. Legg, On the robustness of topological gap detection via transport, Nature(2026-06-24)— [一手逐字亲核] https://doi.org/10.1038/s41586-026-10567-8
  8. Microsoft 回应(2026-06-24, Nature)— [一手逐字亲核] https://doi.org/10.1038/s41586-026-10568-7
  9. The Register 报道 Legg(2026-06-24)— [二手·采访引语] https://www.theregister.com/research/2026/06/24/boffin-claims-microsofts-supposed-quantum-leap-does-not-compute-due-to-basic-python-errors/5260489
  10. Scientific American Microsoft’s topological qubit(2026-06-24)— [二手·引语] https://www.scientificamerican.com/article/top-quantum-computer-expert-claims-microsofts-topological-qubit-doesnt-hold-up/

经典反攻与整体评估

  1. Aaronson, NISQ and quantum supremacy did not fail(2026-07-18)— [二手·领域权威博客] https://scottaaronson.blog/?p=9949
  2. Hagar, The NISQ Trap, arXiv:2607.07530(2026-07-08)— [一手·预印本级] https://arxiv.org/abs/2607.07530
  3. 噪声可模拟性定理三篇:arXiv:2510.06328 / 2510.06324 / 2607.20804 — [一手·预印本级] https://arxiv.org/html/2510.06328 https://arxiv.org/pdf/2510.06324 https://arxiv.org/html/2607.20804v1
  4. Preskill, APS Global Physics Summit 演讲(2025-03-17)— [一手逐字亲核] https://preskill.caltech.edu/talks/MAR-EV40-Preskill-2025.pdf
  5. Preskill, Beyond NISQ: The Megaquop Machine, ACM Trans. Quantum Comput.(2025-04)— [一手逐字亲核] https://www.preskill.caltech.edu/pubs/preskill-2025-megaquop.pdf
  6. Preskill 访谈, New Scientist(2025-02-11)— [一手·访谈原文] http://web.archive.org/web/2025id_/https://www.newscientist.com/article/2466740-how-the-megaquop-machine-could-usher-in-a-new-era-of-quantum-computing/
  7. Reuters, Huang 15–30 years(2025-01-08)— [一手·直接引语] https://www.cnbc.com/2025/01/08/nvidia-ceo-jensen-huang-is-dead-wrong-about-quantum-d-wave-ceo.html
  8. McKinsey Quantum Technology Monitor 2026(2026-04-28)— [一手逐字亲核] http://web.archive.org/web/20260511220752id_/https://www.mckinsey.com/capabilities/mckinsey-technology/our-insights/mckinsey-quantum-technology-monitor-2026-a-commercial-tipping-point
  9. McKinsey 2022 QTM — [一手·数字] http://web.archive.org/web/20250813082702id_/https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/quantum-computing-funding-remains-strong-but-talent-gap-raises-concern
  10. BCG, Long-term forecast(2024-07-18)— [一手逐字亲核] http://web.archive.org/web/20260619190538id_/https://www.bcg.com/publications/2024/long-term-forecast-for-quantum-computing-still-looks-bright
  11. Quantinuum Helios, Nature(2026-06-17)— [一手逐字亲核] https://www.nature.com/articles/s41586-026-10676-4
  12. Quantum Echoes, Nature 646(2025-10-22)+ Google 博客 — [一手逐字亲核] https://research.google/blog/a-verifiable-quantum-advantage/
  13. USTC 祖冲之 3.0, PRL 134, 090601(2025-03-03)— [一手逐字亲核] http://web.archive.org/web/20260517211623id_/https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.134.090601 · arXiv https://arxiv.org/abs/2412.11924
  14. 超导表面码晶格手术, arXiv:2606.06598(2026-06)— [一手·预印本级] https://arxiv.org/abs/2606.06598
  15. 离子阱跨码晶格手术, arXiv:2607.04227(2026-07-05)— [一手·预印本级] https://arxiv.org/pdf/2607.04227
  16. Quantinuum H2 任意子编织, Nature(2026-07-15)— [一手·摘要级] https://www.nature.com/articles/s41586-026-10709-y
  17. 早期容错化学 25–100 逻辑比特视角, arXiv:2506.19337(2025-06-24)— [一手逐字亲核] https://arxiv.org/html/2506.19337
  18. Atom Computing 1,225-site 阵列新闻稿(2023-10-24)— [一手逐字亲核] https://www.prnewswire.com/news-releases/quantum-startup-atom-computing-first-to-exceed-1-000-qubits-301964712.html
  19. Automatski「126 逻辑比特」LinkedIn 自述(2026-01-02,帖子;仓库见链接)— [第三方自述·不承重] https://github.com/adityayadav76/the-first-quantum-cryptography-benchmarks
  20. 中科大 Wukong 逻辑门集, npj Quantum Information 11, 177(2025-11-14)— [一手·摘要级] https://preview-www.nature.com/articles/s41534-025-01118-6

简短结论

母裁决四件再收一次:① below-threshold 真(守真锚:Willow Λ=2.14/0.143%、中科大 Λ=1.40(6)、哈佛/QuEra 2.14(13)×、Quantinuum 非 Clifford 盈亏平衡、IBM 纠缠逻辑比特 beyond-breakeven——2024-12 至 2026-08 五大平台各自跨阈,这不是零进展);②「容错时代已到来」是门槛跳(below-threshold 只是 M2.5,Google 自认「working towards M3」,算法级需求(千余逻辑比特)与当前演示(数十)相差一到两个数量级);③报数口径在物理/逻辑比特间滑动(d=7=101 物理比特=1 逻辑比特、1 逻辑比特≈10³–10⁴ 物理比特、「105 比特」「12 逻辑比特」「200 逻辑比特」「8 拓扑比特」四把尺);④「全是炒作」同样未立(阈值定理数学为真、RCS 深电路经典模拟「仍相当无望」、多平台复现成片)。Majorana 账单独裁决:微软拓扑路线的声明链每一环都有独立缺口(编辑注矛盾、自承无法区分、2018 撤稿先例、2026 Matters Arising 编码错误指控部分获承认、X 测量无同行评审发表、无独立复现)——作为「容错交付主张」,其验证链完整性截至 2026-08-02 是六条路线里最弱的;但这不裁决「Majorana 零模是否存在」的物理问题。灵魂句:below-threshold 是真的、逻辑比特是真的、阈值定理是真的——真的不是「容错时代已经到来」这一层被声称的胜利;判据在场、首格达成、全程未证——把第一格读成终点,是把一条十格的楼梯看成了一级台阶。

与本库的接口:接量子霸权篇(那篇裁窄任务采样 vs 全面颠覆,本篇裁容错链);与固态电池篇(判据在场而口径缺席首用)、L5 篇(判据在场而边界自定)、核聚变篇(Q 值会计)组成「完工判据谱系」四连——本篇是新形态:判据在场、首格达成;与计量学篇对偶(SI 把「知道自己测不准多少」写成义务;量子计算把「知道自己做到哪一格」写成路线图,格子由造尺人自己画、自己报、自己核)。

高风险声明:本篇为计算物理与元科学体检,不构成投资建议、技术选型建议、密码迁移时间表或对任何公司/研究者的品格指控;不裁决 Majorana 物理问题(归拓扑物质篇);Microsoft/DARPA 之争的专有数据不可核查,按诚实空位处理。