跳至正文

缪子 g-2 的「5σ 新物理」去哪了

目录

机制裁决第 126 篇 · 对称双向第 121 篇 · section B 物理被神化的大叙事 · 全库第 184 篇 本篇审的是「5σ 新物理」这个名号从被观察到被撤回,中间经历了什么。先读三句红线:

  1. 本篇不裁决任何粒子物理理论对错,不评价任何实验组、任何在世研究者,不给出「该不该投钱给 g-2 后续实验」的意见——只审「g-2 张力的消失证明了新物理不存在」「g-2 实验错了/之前是炒作」「g-2 异常已彻底解决」这三句话从文件里读出来时承重承不承得住。
  2. 「g-2 的 σ 数变小了」与「g-2 实验值变了」是两件不同的事:前者主要来自理论预言汇编换了尺(WP20 的数据驱动法 → WP25 的格点 QCD 法),不是实验端任何数据变化。本篇对称审计「新物理已被排除」与「g-2 实验是失败的/标准模型已经大获全胜」两个方向。
  3. 全篇承重句均给出可点击来源;英文逐字引用一律标注「一手逐字」(全文取回)或「摘要逐字」。本篇引用的所有数字(a_μ 各代值、各代不确定度、σ 数、HVP 贡献)都给出出处,相关计算(误差正交合并、ppb 换算、σ 换算)写出算法,可复现。

零、一句话裁决

muon g-2 的「4.2σ 新物理异常」是一个真实、被反复确认的实验测量与理论预言汇编之间的偏离(BNL E821 2006 → Fermilab E989 2021 首次 4.2σ → 2023 精度翻倍 → 2025 终值 127 ppb),它的存在本身不是幻觉——但「5σ 张力」这个词把三件不同的事混在了一起:一个越来越准的实验值(没怎么变)、一个自己换了三次尺的理论预言汇编(WP20 数据驱动 3.7σ → BMW 格点折半 → WP25 全格点「no tension」)、以及一个约定俗成的 σ 判据(5σ 是领域约定不是自然律)。真的不是「g-2 证明新物理不存在」(MSSM 低质量区仍有存活空间、CMD-3 数据驱动内部还有 >5σ 矛盾、理论侧 WP20 与 WP25 自己差 3σ)和「g-2 异常已彻底解决/标准模型大获全胜」(CERN Courier 2026 官方口径自己说 solved 是「dramatically oversimplifies」,理论不确定度 530 ppb 仍四倍于实验 127 ppb)这两层被声称的定论。把「同一测量对旧尺是 4.2σ、对新尺是 0.6σ」读成「新物理没有了」,是把「理论预言换了方法」读成「自然界的性质变了」、把「一个 5σ 约定没被满足」读成「一个实体被排除了」。

一、本篇测什么:换尺跳 × 5σ 跳 × 守零跳与六件可判定的事

本篇审的不是「g-2 实验做得好不好」(好,127 ppb 是史无前例的精度),而是「『5σ 新物理异常』这个名号从被宣布到被撤回,中间经历了哪些跳」。拆成六件可判定的事:

  1. 实验值真不真:BNL E821 与 Fermilab E989 各次公告的数值、不确定度、演进轨迹——三次公告中心值 40 → 57 → 70.5(×10⁻¹¹)到底是在变还是稳定(注意单位陷阱)——这是守真锚。
  2. 5σ 是什么:「5σ」在粒子物理里是什么地位——是发现判据的约定(convention)而非逻辑必然,3σ/5σ 的分界、look-elsewhere、trial factor——这是 5σ 跳。
  3. 理论预言是怎么换尺的:WP20(e⁺e⁻ 数据驱动,3.7σ)→ BMW 2021(格点 QCD,张力折半)→ WP25(全格点汇编,「no tension」)——同一实验测量,换理论尺后 σ 从 4.2 变 0.6——这是换尺跳。
  4. 张力搬到哪去了:CMD-3 的 e⁺e⁻→π⁺π⁻ 新测量让数据驱动内部矛盾大到「无法有意义地合并」、CMD-3 vs KLOE 超 5σ——张力没消失,是搬家了——这是转移账。
  5. 消费账:2021-04-07 当天媒体的「接近发现/第五力」口径 vs 2025-2026 的「精度里程碑/无张力」口径,官方 vs 媒体、中文 vs 英文——同一实验、两套叙事——这是消费账。
  6. 反向红跳:「新物理已排除」和「g-2 实验错了/是炒作」为什么都不立——BSM 参数空间仍有存活区、三次公告自洽、E989 系统误差控制、理论不确定度仍是瓶颈——这是反向红跳。

结构胎记:换尺跳 × 5σ 跳 × 守零跳,附反向红跳。

  • 换尺跳:把「理论预言汇编换了方法」读成「自然界的性质变了」——WP25 摘要自己写「Adopting the latter in this update has resulted in a major upward shift of the total SM prediction」,CERN Courier 2025-07 写明 WP20 与 WP25 自己差 3σ;同一个实验测量(116592070.5×10⁻¹¹),对 WP20 是 4.2σ、对 WP25 是 38(63)×10⁻¹¹ ≈ 0.6σ。「张力的消失」主要不是测量变了,是尺换了。
  • 5σ 跳:把「5σ 这个约定判据」读成「新物理存在与否的客观分界」——CERN/Lyons/PDG 一致说明 5σ 是「the convention」、是「ad-hoc recipe with no solid foundations」,且 g-2 的张力是「测量 × 理论汇编」而非「测量 × 测量」;Hubble tension 是后者(两端都是观测),g-2 是前者(一端是预言汇编),所以 g-2 的 σ 数天然随理论预言版本而变。
  • 守零跳:把「张力归零」读成「新物理已排除」或「异常已彻底解决」——CMD-3 数据驱动内部 >5σ 矛盾仍在、理论不确定度 530 ppb 仍四倍于实验、MSSM/leptoquark 参数空间仍有存活区。
  • 反向红跳:张力归零≠实验错(Science 2025 定调「not because experimenters previously erred, but because the theoretical estimate changed」);三次公告中心值稳定≠实验在炒(40→57→70.5 是单位换算后同一条线上的稳定值);「标准模型大获全胜」≠已解决所有问题(CERN Courier 2026「dramatically oversimplifies」)。

共用动作:把一句关于「我们的理论预言汇编换了一种算法」的话,读成一句关于「自然界里有没有新物理」的话。

与库内相关篇的分界:

  • 与真空、场与自然性篇(2026-06-12)分界写死:那篇 §7.2 把 muon g−2 当「量子场论精度/自然性论战」的旁证,已对 WP25 摘要逐字亲核(no tension、Δ=38(63)×10⁻¹¹、格点 HVP、CMD-3);本篇把「g-2 的 5σ 异常与它的消失」本身当对象——换尺史、5σ 语义、不确定度账、报道消费账、BSM 剩余空间。那篇已有的 WP25 摘要逐字本篇只称重不重做,但补上那篇没有的:换尺三步的时间线、E989 三次公告演进、CMD-3 内部矛盾、2025-2026 报道谱系。
  • 与 ΛCDM 张力篇(2026-06-12)分界写死:那篇审 Hubble tension(观测 × 观测,早期 CMB vs 晚期距离阶梯,两端都是数据);本篇审 g-2(测量 × 理论预言汇编,一端是数据一端是理论汇编)。这是「实验×实验张力」与「实验×理论汇编张力」的镜像——Hubble 的 σ 不随理论版本变(两端都是观测),g-2 的 σ 随理论版本变(一端是汇编)。两篇都涉及「5σ 是否达成」,但统计结构不同。
  • 与计量学篇(2026-08-02)分界:那篇是「尺子被做对」的正面范本(SI 定义、GUM 不确定度);本篇用计量学视角看「实验侧 127 ppb 预算 vs 理论侧 530 ppb 预算」的不对称(理论误差四倍于实验、理论是「currently the limiting factor」),但不重做计量学。
  • 与数学篇(2026-06-30)分界:那篇用电子反常磁矩(QED 12 位)当数学有效性的例子;本篇是缪子反常磁矩(同一 g-2 名、不同粒子、不同物理——μ 子强子贡献大、理论不确定度被锁死在 530 ppb)。同名不同物,本篇明确区分。
  • 与希格斯篇(2026-07-19)分界:那篇审 Higgs 质量起源(CMS HIG 组合里的「g-2」命中是 HIG 编号不是 muon g-2);本篇不重做 Higgs 物理。

编号落位:机制裁决第 126 篇·对称双向第 121 篇·section B·全库第 184 篇(2026-08-04 全库扫描确认未被占用)。去重实测(python 全库 185 个 md 文件 697 万字符):g-2/g−2 大量命中但逐条核对全非 muon g-2 专篇——URL slug(goog-、HIG-24-007)、IQ 篇 g-factor、AGI/L5 篇年报链接、预测加工/语言压缩篇标签;反常磁矩 3 处全在数学篇(电子反常磁矩,QED 12 位例子);muon g/Muon g 8 处全在真空自然性篇(§7.2 旁证)——muon g-2 专篇零命中,邻近但不重叠

二、守真锚:实验侧的 127 ppb——三次公告、一个稳定值

本库原则:审「声称」必须回到声称的原件。先清点实验侧到底测了什么、测得多准、各次公告之间怎么变。

2.1 三次公告的数值演进:中心值稳定、精度持续收缩

Fermilab E989(Muon g-2 实验)三次公告,全部 [一手逐字]:

  • 2021-04-07(Run-1,PRL 126, 141801)we determine aμ(FNAL) = 116 592 040(54)×10⁻¹¹ (0.46 ppm). The result is 3.3 standard deviations greater than the standard model prediction ... the new experimental average of aμ(Exp) = 116 592 061(41)×10⁻¹¹ (0.35 ppm) increases the tension between experiment and theory to 4.2 standard deviations.arXiv:2104.03281PRL 摘要
  • 2023-10-17(Run-2/3,PRL 131, 161802)we determine aμ = 116 592 057(25)×10⁻¹¹ (0.21 ppm). Combining this result with our previous result from the 2018 data, we obtain aμ(FNAL) = 116 592 055(24)×10⁻¹¹ (0.20 ppm). The new experimental world average is aμ(Exp) = 116 592 059(22)×10⁻¹¹ (0.19 ppm), which represents a factor of 2 improvement in precision.arXiv:2308.06230PRL 摘要
  • 2025-06-03(Run-4/5/6 终值,PRL 135, 101802)we determine aμ = 116 592 0710(162)×10⁻¹² (139 ppb) for the new datasets, and aμ = 116 592 0705(148)×10⁻¹² (127 ppb) when combined with our previous results. The new experimental world average, dominated by the measurements at FNAL, is aμ(exp) = 116 592 0715(145)×10⁻¹² (124 ppb). The measurements at FNAL have improved the precision on the world average by over a factor of four.arXiv:2506.03069 摘要逐字;标题 「Measurement of the Positive Muon Anomalous Magnetic Moment to 127 ppb」 逐字;Fermilab 官方镜像 result2025.pdf 亲核)

单位陷阱(结构要点):2021/2023 用 ×10⁻¹¹、2025 用 ×10⁻¹²。统一到 ×10⁻¹¹:中心值 040 → 057 → 070.5,总不确定度 54 → 25 → 14.8——中心值单调上移(且都在彼此的误差内)、误差持续收缩。Fermilab 官方新闻稿 [一手逐字] 确认一致性:The third and final result, based on the last three years of data, is in perfect agreement with the experiment's previous results, further solidifying the experimental world average.news.fnal.gov 2025-06-03

2.2 127 ppb 怎么来的:不确定度分解

E989 终值 127 ppb 的分解(论文正文 Table I 逐字 + 本篇自算核验):

  • Run-4/5/6 单套数据:ωa 统计 114 ppb、ωa 系统 30 ppb、外参数 μ’p/μB 4 ppb、mμ/me 22 ppb。Total for aμ ... 139result2025.pdf Table I 一手逐字)。
  • 本篇自算:√(114²+76²+4²+22²) = √19272 ≈ 138.8 ≈ 139 ppb ✓ 正交合并自洽(论文正文逐字:with the statistical, systematic, and external parameter uncertainties combined in quadrature)。
  • Run-1-6 合并 127 ppb = 148×10⁻¹² 换算(127 ppb × 1165920705×10⁻¹² ≈ 148.1×10⁻¹² ✓ 自算核验)。
  • 官方新闻稿口径(± 0.000 000 000 114(stat.) ± 0.000 000 000 091(syst.),√(114²+91²)≈146 ≈ 125 ppb)与论文 148 有约 2×10⁻¹² 的取整差——本篇以论文 Table I + 摘要 148 为准,新闻稿口径差异如实登记

精度对照:127 ppb 超过原设计目标 140 ppb(官方新闻稿逐字 surpassing the original experimental design goal of 140 parts-per-billion);Fermilab 把世界平均精度提高四倍以上;E989 论文正文 [一手逐字]:While ae is measured [5] 1000 times more precisely than aµ, the muon's greater mass makes aµ about 4 × 10⁴ times more sensitive to much BSM physics.——电子 g-2 测到 0.13 ppt(Fan et al. 2023 摘要逐字 determined 2.2 times more accurately than the value that stood for 14 years),但 μ 子因为质量大、对 BSM 敏感度高约 4×10⁴ 倍。这是「同样叫 g-2、完全不同的物理」——μ 子的理论预言被强子贡献(HVP)锁死在 530 ppb,而电子的强子贡献可忽略

2.3 与 BNL E821 的交叉验证

BNL E821 终值 [一手逐字]:aμ(Expt) = 11 659 208.0(5.4)(3.3)×10⁻¹⁰(0.54 ppm,比 CERN 提高 14 倍),且 Present standard model evaluations, based on e+e- hadronic cross sections, lie 2.2 - 2.7 standard deviations below the experimental result.arXiv:hep-ex/0602035 摘要逐字,PRD 73, 072003 (2006))。E989 论文 [一手逐字] 确认与 E821 一致:The Muon g−2 Experiment at FNAL confirmed the E821 result,且 E821 只贡献世界平均的 5.1%。2021 年首轮的 is in excellent agreement with the previous Brookhaven National Laboratory (BNL) E821 measurement 逐字在 PRL 摘要里。

守真锚结论:实验侧是三次独立公告、中心值稳定收敛、系统误差逐年压小、与 20 年前的 E821 交叉验证的序列——不是「实验在变」,是「实验越测越准」。

三、5σ 跳:5σ 是什么——领域约定,不是自然律

3.1 5σ 判据的权威表述:1 in 3.5 million 是约定

5σ 在粒子物理里的官方地位(全部 [一手逐字]):

  • CERN 官方The common agreement is that an excess is called a discovery when this probability is about 1 in three and a half million, which corresponds to an excess of five standard deviations above the expected value – the famous "five sigma".home.cern 希格斯页
  • CERN FAQA result that has a statistical significance of five sigma means the almost certain likelihood that a bump in the data is caused by a new phenomenon, rather than a statistical fluctuation. + Five sigma is considered the "gold standard" in particle physics because it guarantees an extremely low likelihood of a claim being false.home.cern five-sigma FAQ,Wayback 存档)
  • PDG 2026 统计章节Often in particle physics the level of significance where an effect is said to qualify as a discovery is Z = 5, i.e., a 5σ effect, corresponding to a p-value of 2.87×10⁻⁷.PDG rpp2026-rev-statistics.pdf
  • ATLAS Cheat Sheet"Evidence" is claimed if the significance surpasses 3σ (0.3% probability of obtaining this test statistic by chance) and "discovery" if it surpasses 5σ (0.00006% probability of obtaining it by chance). These are simply choices that particle physicists agree upon so that a statistically significant "discovery" has the same meaning for everyone.ATLAS 官方速查表,Wayback 存档)

3.2 5σ 是约定不是定理

  • CERN Courier(Louis Lyons)It has become the convention in Particle Physics that in order to claim a discovery ... the p-value is no larger than 3×10⁻⁷,且 It is unreasonable to regard 5.1σ as a discovery, but 4.9σ as not.Since these factors vary from one analysis to another, one can argue that it is unreasonable to use the same discovery criterion everywhere.cerncourier.com five-sigma revisited
  • DorigoThe five-standard-deviation threshold is an established standard for discovery claims in experimental particle physics; however, the criterion is an ad-hoc recipe with no solid foundations.EPJ Web of Conferences
  • CMS 官方 LEE 页To claim it is a new signal, the effect must reach or exceed the "five-sigma" significance level, five standard deviations away from the expectation: that's a rather silly but well-established rule. + outsiders have better be aware that a three-sigma effect should not be blindly dubbed "evidence" for something new in the data.cms.cern look-elsewhere-effect

5σ 跳的结构:5σ 是「1/350 万」这个概率的约定别名,不是自然界的分界线——它不回答「有没有新物理」,只回答「这个波动是背景涨落的概率低到我们愿意承认它」。g-2 的 4.2σ 从没宣称「发现」,官方 2021 口径逐字是 a little shy of the 5 sigma ... but still compelling evidence of new physicsnews.fnal.gov 2021-04-07)。

3.3 实验×理论 vs 实验×实验:g-2 的 σ 天然随理论版本变

这是本库 ΛCDM 张力篇的镜像分界,也是 g-2 区别于 Hubble tension 的结构点:

  • Hubble tension(ΛCDM 篇):早期宇宙 CMB 推断的 H0(67.24±0.35)vs 本地距离阶梯直接测的 H0(约 73)——两端都是观测数据。H0DN 2026 共识报告 [一手逐字]:a striking and persistent discrepancy between its value inferred from observations of the early Universe, such as the cosmic microwave background (CMB), and its value measured directly in the local Universe using distance ladder methods,张力 7.1σarXiv:2510.23823 摘要逐字)。但要注意 nuance:CMB 端的 H0 是 indirect, ΛCDM-dependent 的——Hubble 是「观测×观测(一端带模型)」,不是纯「观测×观测」。
  • g-2:一端是实验测量(E821+E989),另一端是理论预言汇编(Theory Initiative 白皮书)。理论端携带自身不确定度、且预言方法可以整换(WP20 数据驱动 → WP25 格点)。E989 论文 [一手逐字] 自己写明变化来源:The value shifts considerably compared to their 2020 White Paper (WP2020) [96], which is almost entirely due to the exclusive use of new, published leading-order hadronic vacuum polarization estimates based on lattice-QCD calculations.result2025.pdf 正文逐字)

为什么这个区分重要:Hubble 的 σ 数(7.1σ)不随「理论预言」版本而变(两端都是数据,变的只是系统误差评估);g-2 的 σ 数(4.2σ → 0.6σ)随「理论预言」版本而变(一端是汇编,汇编换方法 σ 就变)。所以「g-2 的 5σ 张力去哪了」和「Hubble 的 5σ 张力去哪了」是两个不同的统计故事——后者要等新数据裁决,前者已经被理论侧换尺「消解」了(但留下新的数据驱动内部矛盾)。

四、换尺跳:理论白皮书的三次换尺——WP20 → BMW → WP25

这是本卷的独占核心:实验侧几乎没动,动的确实是理论这把尺。

4.1 WP20(2020):数据驱动,3.7σ

Muon g-2 Theory Initiative 第一版白皮书 [一手逐字]:The final result reads aμ^SM = 116 591 810(43)×10⁻¹¹ and is smaller than the Brookhaven measurement by 3.7σ.arXiv:2006.04822 摘要逐字,Phys. Rept. 887 (2020))。注意:3.7σ 是相对当时唯一的 BNL 测量;4.2σ 是 2021 年 E989 首轮 + BNL 合并后才出现的。WP20 的 HVP 用 e⁺e⁻ 数据驱动的色散方法

4.2 BMW 2021(同日换尺信号):格点 QCD,张力折半

2021-04-07——E989 首轮结果发布的同一天,BMW 合作组在 Nature 发表格点 QCD 从头算 LO-HVP [一手逐字]:Our result for the LO-HVP contribution to (gμ−2) has a total uncertainty of 0.8%. Compared to the result of the dispersive approach for this contribution, ours significantly reduces the tension between the standard model prediction for (gμ−2) and its measurement.arXiv:2002.12347 摘要逐字,Nature 593, 51 (2021));论文摘要正文 Our result, [(gμ-2)/2]_LO-HVP = 707.5[5.5]×10⁻¹⁰, favors the experimentally measured value of (gμ-2) over the results based on the dispersion relation.ar5iv 渲染版)。

当天媒体就点破了这个结构:Quanta [一手逐字] a paper published today in the journal Nature casts the new muon measurement in a dramatically duller lightIf the new calculation is correct, then physicists may have spent 20 years chasing a ghosta battle of theory versus theoryQuanta 2021-04-07);Science 一周后 [一手逐字] their value sits closer to the experimental result and makes the tantalizing discrepancy nearly vanishScience 2021-04-14)。

4.3 WP25(2025):全格点汇编,「no tension」

Muon g-2 Theory Initiative 第二版白皮书(WP25)[一手逐字](当前 v3/期刊版,Phys. Rept. 1143 (2025)):

A new measurement of the e⁺e⁻→π⁺π⁻ cross section by CMD-3 has increased the tensions among data-driven dispersive evaluations of the LO HVP contribution to a level that makes it impossible to combine the results in a meaningful way. At the same time, the attainable precision of lattice-QCD calculations has increased substantially and allows for a consolidated lattice-QCD average of the LO HVP contribution with a precision of about 0.9%. Adopting the latter in this update has resulted in a major upward shift of the total SM prediction, which now reads aμ^SM = 116 592 033(62)×10⁻¹¹ (530 ppb). When compared against the current experimental average based on the E821 experiment and runs 1-6 of E989 at Fermilab, one finds aμ^exp − aμ^SM = 38(63)×10⁻¹¹, which implies that there is no tension between the SM and experiment at the current level of precision.

arXiv:2505.21476 摘要逐字)

换尺的完整账(WP25 Table 33 逐字 + 本篇整理)

WP20(2020) WP25(2025)
LO HVP(e⁺e⁻ 数据驱动) 6931(40)×10⁻¹¹ 不采用(无法合并)
LO HVP(格点) 7116(184)(未采用) 7132(61)×10⁻¹¹
总 SM 预言 116 591 810(43) 116 592 033(62)
不确定度 ~370 ppb 530 ppb
相对实验的 σ 3.7σ(BNL)/ 4.2σ(E989 首轮合并) 38(63) ≈ 0.6σ

(WP25 引言逐字说明换尺动机:The hadronic-vacuum-polarization (HVP) contribution was estimated in WP20 from e⁺e⁻ hadronic cross-section data in the dispersive, data-driven method, while lattice-QCD methods were not mature enough to yield reliable results at the required precision. In the present review, we find that this situation is reversed.——ar5iv WP25

关键结构:LO HVP 从 6931(40) 换成 7132(61),上移约 200×10⁻¹¹,这是 SM 预言上移的主因,也是「张力消失」的主因。 WP25 总不确定度 62×10⁻¹¹ 里,LO HVP 的 61 占方差约 97%(本篇自算:61²/62² ≈ 97%,从 WP25 Table 1 逐字 HVP LO (lattice) ... 7132(61))。

4.4 版本差异本身就是档案:WP25 v1 → v3

WP25 的 v1(2025-05-27)与 v3/期刊版(2025-09-11)摘要数字不同 [一手逐字]:

  • v1based on the E821 experiment and runs 1–3 of E989 at Fermilab, one finds aμ^exp − aμ^SM = 26(66)×10⁻¹¹(ar5iv 渲染 v1,实验用 E821+runs 1–3)
  • v3based on the E821 experiment and runs 1-6 of E989 at Fermilab, one finds aμ^exp − aμ^SM = 38(63)×10⁻¹¹(arXiv 当前页/期刊版,实验更新为 E821+runs 1–6)

两个数字各自正交自洽(本篇自算:√(22²+62²)=65.8≈66;√(14.5²+62²)=63.7≈63)。「5σ 去哪了」的答案随白皮书版本在 0.4σ(26/66)和 0.6σ(38/63)之间微调——但都远低于 5σ,且都不是因为实验数据大改,而是因为格点 HVP 被采用 + 实验平均更新。

4.5 换尺跳的定调句

  • Science 2025-06-03(最强一句)The discrepancy vanished not because experimenters previously erred, but because the theoretical estimate changed.Science,主笔亲核)
  • CERN Courier 2025-07-08In May, the TI updated its prediction, publishing a value aμ = (116592033 ± 62)×10⁻¹¹, statistically incompatible with the previous prediction at the level of three standard deviations, and with an increased uncertainty of 530 ppb (WP25).(主笔亲核)——WP20 与 WP25 自己差 3σ,比 g-2 对任意一版的张力都大。
  • El-Khadra(Theory Initiative 主席)The switch from using the data-driven method for HVP in WP20 to lattice QCD in WP25 results in a significant shift in the SM prediction.cerncourier.com,主笔亲核)

五、转移账:张力没消失,是搬家了——CMD-3 与数据驱动内部矛盾

「g-2 异常已解决」的一个被忽视的代价是:张力从「实验 vs 理论」搬到了「数据驱动理论内部」。

5.1 CMD-3:把数据驱动拆了

CMD-3 实验 2023 年测 e⁺e⁻→π⁺π⁻ 截面(0.32–1.2 GeV,ρ 主共振区系统误差 0.7%)[一手逐字]:The e⁺e⁻→π⁺π⁻ cross section obtained by CMD-3 significantly deviates from the results of previous measurements, including the one performed by the CMD-2 experiment, the predecessor of CMD-3. It should be noted that the same scale discrepancies have already been observed in previous measurements, e.g., between KLOE and BABAR ... The reason for these discrepancies is unknown at the moment.arXiv:2302.08834 正文逐字)

WP25 摘要逐字(前述)写明了后果:CMD-3 让数据驱动评估之间的张力大到 impossible to combine the results in a meaningful way——这是换尺的理由

5.2 数据驱动内部 >5σ

Hertzog & Hoferichter 2025 综述(Ann. Rev. Nucl. Part. Sci. 在印)[一手逐字]:The new substantial tensions after the CMD-3 measurement, rising above the 5σ level in comparison to KLOE, simply preclude a meaningful average right now, justifying the major change compared to WP20.arXiv:2512.16980 摘要逐字)

讽刺点:g-2 的「5σ 张力」被理论侧换尺消解了,但换尺的代价是——数据驱动方法内部出现了一个新的 >5σ 张力(CMD-3 vs KLOE),而且原因未知。张力不是消失了,是换了个位置:从「实验 vs 理论汇编」搬到「实验 vs 实验(都测 e⁺e⁻ 截面,结果却差 5σ)」。

5.3 WP25 正文的量化

WP25 §4.3 对格点 vs 数据驱动分歧的量化 [一手逐字]:for the LD and intermediate light-quark connected RBC/UKQCD window quantities, there are significant discrepancies between the KNT-compilation-based data-driven and the lattice-based estimates, which lead to a significant discrepancy in the total aμ^HVP,LO(ud),且一个探索性替换实验显示:the exploratory exercise of replacing the π⁺π⁻ KNT-compilation data in the interval between 0.33 and 1.2 GeV with the CMD-3 π⁺π⁻ data suggests that these discrepancies could be due to discrepancies in the experimental data for the π⁺π⁻ component of R_had(s) in the region around the ρ peak.ar5iv WP25)——分歧可能集中在 ρ 峰区(0.63–0.92 GeV)一个 5% 的标定差。

注意:「HVP discrepancy」作为固定短语在 WP25 全文没有逐字出现(本篇 python 检索 n=0),用的是 significant discrepancies between the KNT-compilation-based data-driven and the lattice-based estimates——如实注明。

六、消费账:2021「接近发现」到 2026「无张力」——两套叙事,一个实验

6.1 2021-04-07 当天的报道口径

当天主流高度一致——「4.2σ(BBC 误写 4.1σ),1/40000,接近但未到 5σ,强证据/强暗示新物理」:

  • Fermilab 官方The combined results from Fermilab and Brookhaven show a difference with theory at a significance of 4.2 sigma, a little shy of the 5 sigma ... but still compelling evidence of new physics. The chance that the results are a statistical fluctuation is about 1 in 40,000.news.fnal.gov 一手逐字)
  • BBCMuons: 'Strong' evidence found for a new force of natureNow, physicists say they have found possible signs of a fifth fundamental force of nature.BBC News,状态码 200)
  • Quanta'Last Hope' Experiment Finds Evidence for Unknown ParticlesQuanta
  • NYTA Tiny Particle's Wobble Could Upend the Known Laws of PhysicsNYT 2021-04-07,403 反爬,标题经索引取回)
  • Science Newstantalizingly close to the typical five-sigma benchmark for claiming a discoveryScience News

但同一批报道已经埋下反转伏笔:Quanta 同一天并置 BMW 论文 [一手逐字] chasing a ghost / battle of theory versus theory;Science 一周后 raining on your colleagues' parade

6.2 2025-2026 的报道口径

  • Fermilab 官方终局:标题从「evidence of new physics」变成 Muon g-2 announces most precise measurement of the magnetic anomaly of the muon,正文 [一手逐字] dampening the possibility of new physicsnews.fnal.gov 2025-06-03
  • NYT 2025Muon Experiment Was 'Hugely Successful' but Clarified Little——「精度成功、叙事未决」(NYT 2025-06-03,403)
  • APS PhysicsMuon Experiment Calls It a WrapPhysics 18, 116
  • Quanta 2026-07Physicists Solve a Muon Mystery. Now, Old Results Don't Add Up.——谜题解决了,但带来新谜题(数据驱动内部矛盾)(Quanta 2026-07-29
  • CERN Courier 2026-01Some have argued that the good agreement between lattice–QCD and the final measurement ... means that the g–2 anomaly has now been solved. However, this dramatically oversimplifies the situation: the magnetic moment of the muon remains an intriguing puzzle.cerncourier.com 一手逐字)
  • CNRS 2026:明确写出「若仍对比 WP20,2025 实验已是 5.8σ」——This prediction lies 5.8 standard deviations from the measurement, which would suggest a major flaw in the Standard ModelCNRS 官方,经 Exa 索引正文取回,正文逐字)——张力消失完全是因为换了对比例尺

6.3 中文媒体口径

  • 2021:中国科学报 [一手逐字] 这两个实验室的综合结果显示,新结果与理论的差异显著性为4.2sigma,离科学家宣称发现所需的5sigma(或标准差)还差一点。cas.cn);澎湃 [一手逐字] 物理学恐被颠覆:科学家在"最后的尝试"中发现未知粒子存在的证据thepaper.cn
  • 2025:环球科学 [一手逐字] 本次实验结果与理论计算吻合并非由于此前实验结果有误,而是由于理论计算结果发生了变化。news.qq.com 转载)——与 Science 那句 not because experimenters previously erred 同构;返朴 [一手逐字] 曾经被寄予厚望的超越标准模型的'新物理'迹象,可能只是理论模型尚不完备所致。fanpusci

6.4 消费账小结

  • 口径演变:2021「接近发现/第五力」→ 2023「异常在加强」→ 2025「精度里程碑/dampening new physics」→ 2026「已解决/但矛盾搬家」。
  • 结构:同一个实验值,理论预言换了尺,媒体标题从「upend the known laws」翻转为「calls it a wrap」。
  • 误读检查:按本方法检索,未发现主流把「无张力」误报成「实验错了/之前是炒作」——Science 明确否定了实验端错误(not because experimenters previously erred);最接近的标题式误读(ScienceDaily 2026 Scientists were wrong about this 'rule-breaking' particle)正文也守住「是计算错、不是实验错」的边界。

七、反向红跳:为什么「新物理已排除」和「实验错了」都不立

7.1 「新物理已排除」不立:BSM 参数空间仍有存活区

  • Fermilab 官方Despite recent challenges with the theoretical predictions that reduce evidence of new physics from muon g-2, this result provides a stringent benchmark for proposed extensions of the Standard Model of particle physics.news.fnal.gov 一手逐字)
  • E989 论文Our final result surpasses our original statistical and systematic goals [15] and establishes a stringent benchmark for future theoretical BSM extensions.result2025.pdf 引言逐字)
  • BSM 综述(Athron/Möhling/Stöckinger 2026)The newly obtained agreement may mark the emergence of aμ as an important constraint on many BSM scenarios, allowing only (positive or negative) contributions to aμ of sufficiently small magnitude. + MSSM:significant room for scenarios with masses below these reaches remains viable. + leptoquark:such leptoquarks can very significantly contribute to aμarXiv:2507.09289 逐字)——g-2 从「信号」变成「约束」,约束≠排除

7.2 「g-2 实验错了/是炒作」不立:三次公告自洽 + 系统误差控制

  • 三次公告中心值 040 → 057 → 070.5(×10⁻¹¹),都在彼此误差内——Fermilab 官方 in perfect agreement with the experiment's previous results(前述逐字)。
  • 盲法与多组独立分析(E989 论文 [一手逐字]):Each analysis group developed its analysis using an unknown, fixed, pseudorandom offset in the precession frequency + Four independent analysis groups utilize three different positron reconstruction and pileup correction algorithms while a fifth group performs energy reconstruction + Consistency of the ωam values from these algorithms speaks to the robustness of the measurement.result2025.pdf
  • 系统误差压到 78 ppb(CERN Courier 逐字:technical improvements to our calorimetry, tracking, detector calibration and magnetic-field mapping were also needed to improve on the systematic uncertainties from 280 ppb at BNL to 78 ppb at Fermilab)。
  • Science 定调not because experimenters previously erred(前述)。

7.3 「标准模型大获全胜」不立:理论不确定度仍是瓶颈

  • 理论是 limiting factor(CERN Courier 2026 [一手逐字]):At 530 ppb, theory is currently the limiting factor in any comparison. + In the near future, it will not be possible to reach this precision with lattice QCD alone.cerncourier.com
  • El-Khadra 采访All g–2 experiments are beautifully consistent, and the lattice-based WP25 prediction differs from them by less than one standard deviation. + I stress that the combination of the SM theory error being four times larger than the experimental one and the remaining puzzles ... means that the question "Does the SM account for the experimental value of the muon's anomalous magnetic moment?" has not yet been satisfactorily answered.CERN Courier 2026-01 一手逐字)
  • Nature 2026 混合法(BMW+DMZ,0.48% 精度):our result leads to a prediction that differs from the recent measurement of aμ (ref. 4) by only 0.5 standard deviations. This provides a notable validation of the standard model to 11 digits.Nature 653, 373 (2026) 摘要逐字,经 Exa 索引正文取回)——这是新方法进一步验证 SM,但论文也写 some of the data-driven results are in serious tension both with our and the lattice-only estimates

7.4 反虚无锚(更广):g-2 归零 ≠ 粒子物理没有别的异常

按 2025-2026 现状如实转述(不夸大):

  • b→sℓℓ 角观测量:LHCb 全 Run1+2 角分析 P′5 局部 2.6–2.7σ、Re(C9) 含分支比达 4.0–4.1σ——张力仍在但已非 5σ 级(arXiv:2512.18053 逐字)。
  • W 质量:PDG 2025 [一手逐字] a combination of all W-mass measurements has a probability of compatibility of only 0.5%, and is therefore disfavoured. A 91% probability of compatibility is obtained when the CDF Run-II measurement is removed.PDG rpp2025-rev-w-mass.pdf)——这是测量间张力,不是 SM vs 实验。
  • α 的测量:El-Khadra 采访 [一手逐字] two measurements of α disagree at more than 5σcerncourier.com)。
  • 如实降级:任务书举例的 Rb(LEP Z→bb̄)是 1990s 后期已缓解的历史 anomaly,未取回 2025-2026 综述把它列为活跃异常的表述,本篇不引。

八、未来判据:谁能在 g-2 之外独立验证

8.1 J-PARC E34:独立于储存环法

J-PARC Muon g-2/EDM 实验用超冷 μ 子 + 紧凑磁体,打破「magic momentum 储存环」谱系 [一手逐字]:The expected systematic uncertainty will be at the same level as that of the Fermilab Muon g-2 experiment, providing an important cross-check of the "storage-ring method" employed at BNL and Fermilab.arXiv:2512.20335 摘要逐字);Hertzog & Hoferichter:The J-PARC E34 experiment (87) breaks the mold of continuing the lineage of magic momentum g−2 experiments. + The main value is the very different technique that would yield an independent validation.arXiv:2512.16980)——目标是 0.45 ppm 精度(约 BNL 水平),数据预计 2030 后。

8.2 未来格点精度与 MUonE

  • 格点(H&H [一手逐字]):Improving current lattice-QCD calculations by a factor of two, as realistic for the near-term future, corresponds to about 0.5% precision or 35×10⁻¹¹ for a single calculationarXiv:2512.16980)。
  • MUonE(CERN,μ-e 散射测 HVP):The MUonE experiment ... aims at a measurement of the HVP contribution in muon-electron scattering via the space-like hadronic running of α ... yielding another independent determination with completely different systematics.(同前)
  • WP25 摘要The final precision of E989 (127 ppb) is the target of future efforts by the Theory Initiative. The resolution of the tensions among data-driven dispersive evaluations of the LO HVP contribution will be a key element in this endeavor.(前述逐字)

未来判据结构:实验侧 J-PARC 独立验证 + 理论侧格点精度提升(当前 0.9% → 目标 0.5%)+ 数据驱动内部矛盾(CMD-3)的澄清 + MUonE 第三方法——「5σ 张力」的重审窗口还在,只是不再由 Fermilab 这一个实验单独决定。

九、落点与工具:换尺跳的可复现结构

9.1 本篇自算的承重核验(可复现)

  • 139 ppb = √(114²+76²+4²+22²)(E989 Table I,统计 114 ⊕ 系统 76 ⊕ 外参 4/22)✓
  • 148×10⁻¹² = 127 ppb × 1165920705×10⁻¹² ✓
  • 66 = √(22²+62²)(WP25 v1 实验 22 ⊕ 理论 62)✓
  • 63 = √(14.5²+62²)(WP25 v3 实验 14.5 ⊕ 理论 62)✓
  • 62² 里 LO HVP 61 占方差 97%(WP25 Table 1)✓
  • 40→57→70.5×10⁻¹¹ 单位统一后的三次公告中心值 ✓

9.2 换尺跳的可复现定义

换尺跳:当「5σ 张力」的判断依赖一端为理论预言汇编(而非另一端观测)时,σ 数会随汇编方法变化而变。可复现的检查三步:

  1. 判断张力结构是「观测×观测」还是「测量×理论汇编」(g-2 是后者,Hubble 是前者带一端的模型依赖)。
  2. 若是后者,查理论汇编是否更换过方法(WP20 数据驱动 → WP25 格点)、换方法后预言移动几个 σ(WP20 vs WP25 差 3σ)。
  3. 把「σ 变小」归因到实验端(错误)vs 理论端(正确归因)——Science 的 the theoretical estimate changed 就是正确归因的样板。

9.3 灵魂句

g-2 的 5σ 异常从来不是被数据打掉的,是被换尺打掉的——实验测得更准了(127 ppb),但理论预言也换了算法(数据驱动 → 格点 QCD),于是同一个测量从 4.2σ 落到了 0.6σ。这不是「自然界变平凡了」,是「我们换了一本对答案的书」。真正的张力没有被消灭——它从「实验 vs 理论」搬到了「数据驱动方法内部」:CMD-3 和 KLOE 之间的 5σ 矛盾,正在等着谁去解释。

9.4 与本库方法学的自指

  • 本库 B1 方法论封顶篇(2026-07-20)把「守真锚·分层承重·对称三向红线」资产化为检测器。本篇是本库第一次处理「5σ 判据本身的撤回」——它不是「大主张没兑现」(那是技术时间表家族),是「主张所依赖的尺被换了」。
  • 本篇同样没有在开工前写下完工判据,与本库既往查出的自身欠账(抗衰路线图篇、动物意识篇)同构,如实登记。补救:改判条件写在 9.5。

9.5 改判条件(诚实空位)

  • 若 J-PARC E34 最终结果与 FNAL 显著偏离,或数据驱动内部矛盾(CMD-3 vs KLOE)被证明是一个被夸大的统计假象,「张力搬家」的框架需修订。
  • 若理论侧格点精度提升到 0.5% 后预言再次移动超过 2σ,「换尺跳」的定性需要扩展。
  • 若发现本篇引用的某个 WP25/CMD-3 数字转录错误,相应段落立即降级为不承重。

十、来源清单

(全部来源在正文给出 URL 与状态码;以下为编号清单)

  1. E989 终版论文(Muon g-2 Collaboration,arXiv:2506.03069 / PRL 135, 101802 / title 127 ppb)——arXiv ✓ / Fermilab 官方镜像
  2. Fermilab 官方新闻稿 2025-06-03——news.fnal.gov
  3. Fermilab 官方新闻稿 2021-04-07——news.fnal.gov
  4. E989 Run-1(arXiv:2104.03281 / PRL 126, 141801)——arXiv ✓ / Crossref
  5. E989 Run-2/3(arXiv:2308.06230 / PRL 131, 161802)——arXiv ✓ / Crossref
  6. BNL E821(Bennett et al., arXiv:hep-ex/0602035 / PRD 73, 072003)——arXiv
  7. WP20 白皮书(Aoyama et al., arXiv:2006.04822 / Phys. Rept. 887)——arXiv
  8. WP25 白皮书(arXiv:2505.21476 / Phys. Rept. 1143)——arXiv ✓ / ar5iv v1
  9. BMW 2021 格点(arXiv:2002.12347 / Nature 593, 51)——arXiv ✓ / Nature ✓ / ar5iv
  10. CMD-3(arXiv:2302.08834)——arXiv ✓ / 全文
  11. Hertzog & Hoferichter 综述(arXiv:2512.16980)——arXiv
  12. Nature 2026 混合法(Boccaletti et al., Nature 653, 373)——Nature ✓(经 Exa 索引正文取回,403 反爬)
  13. BSM 综述(Athron/Möhling/Stöckinger/Stöckinger-Kim, arXiv:2507.09289)——arXiv
  14. J-PARC E34 status(arXiv:2512.20335)——arXiv
  15. 电子 g-2(Fan et al., arXiv:2209.13084 / PRL 130, 071801)——arXiv
  16. 电子 g-2(WP25 §7 内 a_e)——WP25 同源 ✓
  17. CERN Courier 2025-07-08(Fermilab’s final word)——cerncourier.com
  18. CERN Courier 2026-01-14(There’s more g–2 physics over the horizon)——cerncourier.com
  19. CERN Courier 2026-01(El-Khadra 采访)——cerncourier.com
  20. CERN Courier 2021-04(An anomalous moment)——cern-courier
  21. CERN 官方希格斯页(5σ)——home.cern
  22. CERN FAQ(five sigma)——Wayback
  23. PDG 2026 统计章节——pdg.lbl.gov
  24. ATLAS Cheat Sheet——Wayback
  25. CERN Courier 5σ revisited(Lyons)——cerncourier.com
  26. Dorigo(EPJ Web Conf.)——epj-conferences.org
  27. CMS LEE 页——cms.cern
  28. H0DN 2026(arXiv:2510.23823)——arXiv
  29. Science 2025-06-03(dashes hope)——science.org ✓(主笔亲核)
  30. Science 2021-04-14(fresh calculation)——science.org
  31. Quanta 2021-04-07——quantamagazine.org
  32. Quanta 2026-07-29——quantamagazine.org
  33. BBC 2021-04-07——bbc.com
  34. NYT 2021-04-07——nytimes.com(403 反爬,标题经索引取回)
  35. NYT 2025-06-03——nytimes.com(403 反爬,标题经索引取回)
  36. Science News 2021——sciencenews.org
  37. Science News 2025——sciencenews.org
  38. APS Physics(Muon Experiment Calls It a Wrap)——physics.aps.org
  39. Physics World 2025-07-14——physicsworld.com
  40. Physics World 2026-05-26——physicsworld.com
  41. CNRS 2026——inp.cnrs.fr ✓(经 Exa 索引正文取回,页面有 JS 挑战墙)
  42. SciDaily 2026(Penn State)——sciencedaily.com
  43. UCL News 2025——ucl.ac.uk
  44. 中国科学报 2021——cas.cn
  45. 澎湃 2021——thepaper.cn
  46. 环球科学 2025(腾讯转载)——news.qq.com
  47. 返朴 2025——fanpusci
  48. Theory Initiative 官方主页——muon-gm2-theory.illinois.edu ✓(原域名 muon-g-2-theory-initiative.org 连接超时)
  49. LHCb B⁰→K*(892)⁰μ⁺μ⁻(arXiv:2512.18053)——arXiv
  50. PDG 2025 W 质量综述——pdg.lbl.gov

关联笔记