跳至正文

分歧是真的,危机是假的——动的是汇编,不是常数

目录

机制裁决第 152 篇 · 对称双向第 147 篇 · section B · 全库第 210 篇 本篇审的是「G 是测得最不准的基本常数/G 越测越不准/G 在随时间振荡/引力定律在实验室尺度有危机/分歧已被解决」这组话——以及它背后那十六个用扭秤、单摆、梁秤和冷原子干涉仪做出的测量值、那本五十年出了十届的 CODATA 汇编账、和那台从巴黎郊区运到盖瑟斯堡的三十年老扭秤。先读三句红线:

  1. 本篇不裁决「G 的真实值是多少」——那是实验物理学尚未结案的工作;本篇审的是「分歧」这组事实被读成的三层结论(常数在动/定律危机/分歧已解决)各自的承重方式。本篇不构成任何计量、实验设计或投资判断建议。
  2. 「测量值互相对不齐」与「自然界出了事」是两件不同的事。本篇的裁决落在两者之间的缝里:分歧是真的(官方汇编成文承认),「这是测量账、不是常数账」也是真的(引力定律的检验精度比分歧带窄 9 到 12 个数量级),而「最测不准的常数」这个名号的含义被读错了——它不是无知的深渊,是十六个自称 12–150 ppm 的测量在官方账本上留下的折痕——跳变不发生在扭秤里,发生在「汇编的不确定度」被读成「常数的性质」的那个动作里。
  3. 全篇承重句均给出可点击来源;中英文逐字引用一律取自实际取回并落盘的文件(含 Wayback 原件与 NIST/APS 托管全文 PDF),自算部分写明算式。本篇引用的所有数字(σ 数、ppm、扩张因子、被引次数)都给出出处与口径。

零、一句话裁决

「G 的各次精密测量互相对不齐」是真的——CODATA 2022 平差报告把这个事实写成了最高级:「The value of G has been the least well known of the major fundamental constants for decades」(CODATA 2022,arXiv:2409.03787[一手逐字]);16 个输入测量里最紧的一对互相差 9.9 倍合成标准差(本篇自算),全局带宽 551 ppm,是最佳单实验自称精度(12 ppm)的 46 倍。「这个常数在动」是假的——2015 年那篇宣称 G 随昼长以 5.9 年周期振荡的论文,十五个月内被三篇同行文献系统性否定,而原文作者自己就写下「we do not suggest that G is actually varying by this much, this quickly」;月球激光测距与脉冲星计时把任何真实的 Ḡ/G 压在 10⁻¹³–10⁻¹² /年量级、与零相容。「引力定律因此有危机」是假的——等效原理与平方反比定律的现代检验精度在 10⁻¹³–10⁻¹⁵,比分歧带窄 9 到 12 个数量级;连做复测的人自己都在论文里写「The first hypothesis, unexplained physics, remains remote.」。「分歧已被解决」也是假的——2026 年同一台装置在另一个国家计量院做的首次盲测复测,结果落在原值的另一侧:「Our result does not corroborate either the earlier BIPM value or the current CODATA recommendation.」。真正在动的从来不是这个常数,是汇编它的那本账:官方相对不确定度五十年间从 460 ppm 收到 22 ppm,中途两次不降反升(1998 年放大「about a factor of 12 larger」、2010 年「an increase in uncertainty of 20%」),而现行的 22 ppm 是把 16 个互相打脸的值的不确定度统一乘以 3.9 之后抚平的读数——抚平动作白纸黑字印在同一篇论文里。

本篇的独占格在第五章(张力账:16 值全表逐条回到一手文献核对+配对 σ 表+本篇自算残差三甲与 CODATA 官方逐字逐位对上)、第六章(换尺账:CODATA 十届编年——1986 年「arbitrarily double」的人为加倍、1998 年「chosen」出来的 1500 ppm、2006 年把不确定度当科研激励工具的政策性自陈、2010 年扩张因子 14、2022 年无新输入原样冻结)、第八章(复测账:NIST 用 BIPM 三十年前造的同一台扭秤做盲测复测,偏差从 +200 ppm 翻到 −64 ppm——装置级复测的第一次落地没有收敛分歧,而是把分歧刻进了同一台装置)、第九章(危机叙事账:Anderson「5.9 年振荡」案从宣称到三篇反驳闭合的十五个月全记录——未撤稿、CODATA 零提及、交锋止于 2016)。

灵魂句:常数没动,动的是汇编。22 ppm 不是测出来的,是抚出来的——3.9 倍扩张因子,是十六个互相打脸的精密测量在官方账本上留下的折痕。而这道折痕本身是诚实的:这个体系宁可把分歧印进推荐值的不确定度里,也不肯假装它不存在。

一、本篇测什么

1.1 被审对象

被审的是一组互相咬合的说法及其用法

  • 分歧句:「G 的各次精密测量互相超出对方的误差棒」(事实层——但被读成什么,是后面十章的事)
  • 危机句:「G 在随时间振荡/漂移」「测量分歧说明引力定律在实验室尺度失效」(危机叙事层)
  • 名号句:「G 是测得最不准的基本常数」「G 的测量越测越不准」(消费话术层)
  • 解决句:「分歧已被某个机制解释/已被某次新测量解决」(收敛宣称层)

1.2 结构胎记

换尺跳 × 张力跳 × 收敛跳 + 反向红跳。

  • 换尺跳:把「官方汇编的输入集与扩张因子在变动」读成「这个数本身在动/人类的测量能力在退化」——CODATA 官方不确定度的两次反升(1986→1998 的 128 ppm→1500 ppm、2006→2010 的 100 ppm→120 ppm)都是汇编事件:一次为一个离群值「chosen」出方便的不确定度,一次把每条输入的不确定度统一乘以 14。尺在动,不是数在动(第六章)。
  • 张力跳:把「测量值之间的不一致」读成「常数层面或定律层面的信号」——Anderson 2015 案是把张力读成时间信号的完整样本(宣称→反驳→沉默十五个月走完);而 10⁻¹³–10⁻¹⁵ 量级的定律检验封死了这条读法的物理空间(第九、十一章)。
  • 收敛跳:把「一个机制被确认(丝滞弹性)/一个错误被抓到(JILA 勘误)/一次复测落地(NIST-26)」读成「分歧已解决」——每一件都是真进展,每一件都没能合上这条带;IUPAP 工作组 2025 年报逐字「No method-dependent systematic error can be identified.」(第七、八章)。
  • 反向红跳(第十一章):四句虚无侧读法——「G 完全未知」「G 在振荡」「这是计量学的失败」「分歧说明引力定律错了」——逐句称重,四句全不立,但每一句的诱人之处都给出交代。

1.3 落位:section B 与分界

本篇落 section B(物理),与三篇邻近篇分界写死:

  • 与计量学篇(2026-08-02,全库第 175 篇)互为姊妹:那篇审 CODATA/SI 体系的正面形态——2019 年 SI 修订把七个常数钉成定义值的「定义换尺」;本篇审同一体系里唯一修不好的常数——G 在 2019 年之后仍是「通用常数」里唯一以 ppm 级不确定度实测的,换的是「汇编」这把尺。那篇是正面范本,本篇是负面姊妹:同一本 CODATA 账,一边展示体系能修好一切,一边展示它修不好 G。
  • 与 g-2 篇(2026-08-04,全库第 184 篇)同族:那篇换的是理论预言汇编这把尺(同一实验值对两把理论尺读出 4.2σ 与 0.6σ);本篇换的是输入集+扩张因子这把尺。
  • 与惰性中微子篇(2026-08-12,全库第 202 篇)同族:那篇换的是反应堆通量分母这把尺。三篇构成「换尺家族」三连,详见第十二章。

1.4 去重实测

python 全库扫描 211 个 markdown / 8,468,153 字符(2026-08-16 结题日复扫):引力常数万有引力常数gravitational constantNewtonian constantCavendish卡文迪许torsion扭秤HUSTJILALENS-14SchlammingerKuroda滞弹性anelastic扩张因子expansion factor昼长length of day6.674 全库零命中BIPM 98 处中 89 处在计量学篇(国际计量局机构义)、CODATA 10 处、Quinn 8 处全为别义——G 测量分歧专篇零命中,处女地确认

1.5 边界

  • 不裁决 G 的真实值;不推荐任何测量方案;不评价任何实验室或科学家的业务信誉——纠错编年一章只登记「公开文献里记录的错误与修正」这一层,纠错是科学自我检修的勋章,不是罪状。
  • Anderson 案只审其作为「危机叙事样本」的生命周期,不评价作者个人的学术功过。
  • 中文媒体叙事只登记话术形态,不评价具体报道与记者。
  • 本篇不涉及任何计量实践或实验选购建议。

1.6 方法备案

  • 取证通路:NIST 物理常数站(physics.nist.gov)历届 CODATA 平差报告官方 PDF 直取+现行值页/历史归档页;APS harvest 全文(PRL/PRD/RMP);arXiv abs 页与 PDF;Europe PMC/PMC 全文;Nature 摘要层(meta description 逐字);IOP 开放获取 HTML;Wayback CDX+id_ 原件快照;IUPAP 官方报告 PDF;Crossref 元数据核验卷期与被引数。
  • 承重统计一律 python3 复算;配对张力统一算式 n = |G₁−G₂|/√(u₁²+u₂²)(独立不确定度方和根;已知相关性的配对另注);凡「自算」必标注,官方数字与自算数字分行陈述不混排。
  • 任务书预设勘误 5 处(全部有落盘证据,明细见附录 A):「1998 年不确定度放大 17 倍」证伪(官方逐字为「about a factor of 12 larger」);「BIPM 2022–2024 G 国际比对(CCM 框架首次)」查无实证(CCM.G-K2 实为绝对重力仪即小 g 的比对;G 的制度响应走的是 IUPAP 工作组+装置转移复测路线);「CODATA 输入 11 个(或几个)」修正为 16 个(Table XXX);开题简报把 MSL-03 刊物误作 Meas. Sci. Technol.(实为 PRL 91, 201101);任务书「分歧与定律检验差约八个数量级」按实测口径修正为 9–12 个。

二、核验标记与层速览

核验标记沿用本库四档:[一手逐字]=原文逐字取自实际取回文件(含 Wayback id_ 原件);[文献较稳]=多源一致或权威评估;[需亲核]=正文未全取回或版本存疑;[多源检索]=方法学阴性检索(含「缺席即证据」登记)。

名称 核心问题 裁决方向
守真锚 哪些是真的 六条,一条不动
对象账 G 是什么、为什么难 牛顿没引入它,卡文迪许没测它
张力账 16 个值互相差几个 σ 最紧一对 9.9σ;带宽 551 ppm
换尺账 官方推荐值五十年怎么动 两次反升都是汇编事件;22 ppm 是抚平值
机制账 分歧藏在哪 一个机制确认、一批错误被抓、全局未定位
复测账 同一台装置重做会怎样 从 +200 ppm 翻到 −64 ppm
危机叙事账 「G 在振荡」的一生 十五个月三篇反驳闭合;未撤稿无人跟进
消费账 「最测不准」名号怎么形成 名号成立;意思是折痕不是空洞
十一 反向红跳 四句虚无读法 四句全不立
十二 镜像家族账 换尺三连 尺动了,不是世界动了
十三 尾账 在研与制度现状 五个活口+四个观察点

三、守真锚(这一侧一条都不动)

在拆任何跳变之前,先把真实的那一侧钉死。以下六条本篇不做任何弱化:

守真锚一:分歧为真,且官方成文承认。 CODATA 2022 平差报告 G 节首句逐字:「The value of G has been the least well known of the major fundamental constants for decades. The uncertainties of the 16 values in Table XXX on which the CODATA 2022 recommended value is based required an expansion factor of 3.9 to reduce their inconsistencies to an acceptable level…」(CODATA 2022[一手逐字])。上一届同口径:「Inconsistencies among measurements of the Newtonian constant of gravitation G have long been a problem. This is no different in the 2018 adjustment.」(CODATA 2018,RMP 93, 025010,NIST 托管 PDF[一手逐字])。IUPAP 第 13 工作组 2024 年报量化:「the relative spread of the results is on the order of 100 parts in 10⁶, while the best experiments show relative uncertainties of 12 parts in 10⁶」、「the difference between the smallest and largest measured value of G differ relatively by more than 500 × 10⁻⁶」(IUPAP WG13 2024 年报[一手逐字])。

守真锚二:十六个输入值全是认真做的真测量。 它们全部经同行评议发表于 PRL、Nature、PRD、Phil. Trans. R. Soc. A 等刊;每一个值的每一次修订都有公开文档链(JILA 的公开勘误、HUST 的自我修正、BIPM 的勘误、UWash 经 CODATA 转述的私人通信修正);1995 年那个离群 0.7% 的 PTB 值,是原作者机构自己复查后宣布「the PTB result for G and its uncertainty cannot be considered correct」并撤出的(CODATA 2002,RMP 77, 1[一手逐字])。CODATA 2022 的输入表注逐字公开三组相关系数:「The data are uncorrelated except for three cases with correlation coefficients r(NIST-82, LANL-97) = 0.351, r(HUST-05, HUST-09) = 0.134, and r(HUST-09, HUSTT-18) = 0.068.」[一手逐字]。没有任何一方被指控造假;这个领域的人物特写里最说明问题的一句是 Müller 说的:「One of the wonderful things about this field is that there’s no attempt to sweep the discrepancies under the rug.」(PNAS 2016 特写,PMC5018785[一手逐字])

守真锚三:系统误差未定位完——这也是真的,且官方逐字承认。 综述逐字:「The published values scatter by this factor more than they should based on the reported uncertainties. This gives reason to suspect hidden systematic errors in some of the experiments.」、「There will always be an unknown uncertainty, often referred to as a dark uncertainty.」(Rothleitner & Schlamminger 2017, Rev. Sci. Instrum. 88, 111101,PMC8195032[一手逐字])。WG13 2025 年报逐字:「No method-dependent systematic error can be identified.」[一手逐字]——按方法分组找不到偏差的聚类。

守真锚四:方法学与制度响应为真。 2014 年两次国际会议促成 IUPAP「Big G」工作组成立,宗旨逐字:「to assist in resolving the discrepancy present in G measurements」(RS2017 引)[一手逐字];CIPM 2014 年 11 月决议逐字:「to establish a consortium of national metrology institutes to facilitate new work aimed at resolving the present disagreement among measurements of the Newtonian constant of gravitation」(同上)[一手逐字];统计学家为这套数据专门发展了「dark uncertainty」层级模型(Merkatas et al. 2019, Metrologia 56, 054001[一手逐字]);装置级盲测复测 2026 年首次落地(第八章)。

守真锚五:汇编的输入与规则全部公开。 CODATA 把 16 个输入值连同不确定度、方法、实验室代号、文献引用、三组相关系数整表刊出(Table XXX);计算规则逐字:「Because G is independent of all other constants, it can be determined in a separate least-squares adjustment, which is simply a calculation of their weighted mean.」(CODATA 2022 §XV.A)[一手逐字];收录关门日期逐字:「The cutoff date for accepted data was at the close of 31 December 2018」(CODATA 2018)[一手逐字]。这本账没有任何一页是黑的。

守真锚六:引力定律本身在比分歧小 9–12 个数量级的精度上被反复检验,全部与零相容。 等效原理:MICROSCOPE 卫星 η(Ti, Pt) = [−1.5 ± 2.3(stat) ± 1.5(syst)] × 10⁻¹⁵(Touboul et al. 2022, PRL 129, 121102[一手逐字]);Eöt-Wash 扭秤 η(Earth; Be-Ti) = (0.3 ± 1.8) × 10⁻¹³(Schlamminger et al. 2008, PRL 100, 041101[一手逐字]);平方反比:52 μm–3.0 mm 区间「Newtonian gravity gave an excellent fit to our data」(Lee et al. 2020, PRL 124, 101101[一手逐字]);时间漂移:月球激光测距 Ḡ/G = (4 ± 9) × 10⁻¹³ yr⁻¹(Williams, Turyshev & Boggs 2004, PRL 93, 261101[一手逐字])。这条锚是第十一章全部裁决的地基。

四、对象账:G 是什么、为什么难(被审对象的本体)

4.1 牛顿没有引入这个常数

G 是牛顿万有引力定律里的比例常数——但牛顿本人从未引入它。综述逐字:「G plays the role of a constant of proportionality (i.e., it can be considered as conversion factor). In fact, when Newton wrote the law of gravitation, he did not introduce this proportionality factor because at that time, laws were formulated as ratios rather than as equations. Hence, G was of no significance to Newton.」(Rothleitner & Schlamminger 2017,下称 RS2017[一手逐字])。当时定律以比例式而非等式书写,比例常数没有意义——G 是后世把定律改写成 F = G m₁m₂/r² 时才被造出来的一个数。在 SI 里它的数值「is very small (ten orders below unity)」(同综述[一手逐字])。

4.2 为什么难:弱 38 个数量级、不可屏蔽、且只能做绝对测量

难度的三根支柱全部是 RS2017 逐字:

  • 弱且不可屏蔽:「the gravitational force is about 38 orders of magnitude smaller than the electromagnetic force and since the gravitational interaction, unlike the electromagnetic interaction, cannot be shielded」[一手逐字]——引力比电磁力弱约 38 个数量级,且不能像屏蔽电荷那样把背景质量屏蔽掉。
  • 信号相对本地重力极小:「Although gravimeters can resolve gravity to better than one part in 10⁹, the signal, when compared to g, is only about 1 part in 10⁷.」[一手逐字]——重力仪能分辨 10⁻⁹,而 G 实验的信号只有 g 的 10⁻⁷。
  • 它是绝对实验:「the other three torsion-balance methods require the measurement of an absolute quantity, angle, angular acceleration, or feedback voltage. It is much more difficult to measure these quantities with a relative uncertainty of 10⁻⁵ than it is to measure a time interval with the same relative uncertainty.」、「The time interval is the physical quantity that can be measured with the highest precision, especially in the age of GPS.」[一手逐字]——G 实验要把质量、长度、角度/电压全部独立溯源测准;没有「G 的标准器」可以保存和传递,每一代实验都要从头做一次绝对力学测量。

做实验的人自己的总结(Parks & Faller 2010,JILA 单摆实验导言逐字):「It is a supreme test of an experimental physicist to cleanly pull this signal out of the inevitable sea of perturbing influences.」(PRL 105, 110801[一手逐字])——从无边无际的扰动海洋里干净地捞出这个信号,是对实验物理学家的终极考试。

还要把一个常见混淆钉死:「Currently, G is often also called ‘Big G’ in order to distinguish it from the acceleration due to gravity, which is denoted by the lower-case letter g—hence referred to as ‘Little g’.」(RS2017[一手逐字])。测 G(常数)与测 g(本地重力加速度)是两门生意:g 已被自由落体绝对重力仪测到 10⁻⁹ 量级且有成建制的国际比对(第十一章用这个事实称「计量学失败」句);G 只到 2.2×10⁻⁵——同名引力,差 4 个数量级。

4.3 卡文迪许 1798 原文账:他测的是地球密度,全文没有 G

一手件:Phil. Trans. R. Soc. London 88, 469–526 (1798),DOI 10.1098/rstl.1798.0022,Royal Society 扫描 PDF 经 Wayback 取回(OCR 文本层完整)。

标题与收读日期逐字:「XXI. Experiments to determine the Density of the Earth. By Henry Cavendish, Esq. F.R.S. and A.S. Read June 21, 1798.」[一手逐字]

结论段逐字(OCR 原文,两处记号已核对):「By a mean of the experiments made with the wire first used, the density of the earth comes out 5.48 times greater than that of water; and by a mean of those made with the latter wire, it comes out the same; and the extreme difference of the results of the 23 observations made with this wire, is only ,75; … therefore, it seems very unlikely that the density of the earth should differ from 5,48 by so much as 1/14 of the whole.」[一手逐字]

「全文无 G」的机器验证:用 python 对 OCR 全文检索 constant(不区分大小写),共 2 处命中、均为副词「constantly」(「a room which should remain constantly shut」/「the motion was constantly of the same kind」)——全文没有任何引力常数记号或「constant of gravitation」表述[多源检索]。RS2017 的综述口径逐字:「Thus, originally, Cavendish did not measure G in his famous experiment but the mean density of the Earth, as his article was titled.」[一手逐字] Wikipedia「Cavendish experiment」条同向:「Because of the unit conventions then in use, the gravitational constant does not appear explicitly in Cavendish’s work.」[二手导航级]

装置谱系(卡文迪许自述逐字):「Many years ago, the late Rev. John Michell, of this Society, contrived a method of determining the density of the earth, by rendering sensible the attraction of small quantities of matter; but, as he was engaged in other pursuits he did not complete the apparatus till a short time before his death, and did not live to make any experiments with it. After his death, the apparatus came to the Rev. Francis John Hyde Wollaston … was so good as to give it to me.」[一手逐字]——扭秤是 Michell 设计建造的,他 1793 年去世前没多久才装完、没来得及用;经 Wollaston 转手到卡文迪许。

两个流传叙事的时点核对:「weighing the world」语出卡文迪许本人——「A 1783 letter from Cavendish to Michell contains ‘the earliest mention of weighing the world’.」(Wikipedia 引 Jungnickel & McCormmach 1996[二手导航级]);而卡文迪许发表的 5.48 今算应作 5.448 ± 0.033(Wikipedia 同条:因 1821 年 Francis Baily 发现的一处算术错误)[二手导航级]。

4.4 G 记号史:1885 年首见刊,1894 年定名,1890 年代末通行

记号的两个时点,两条证据并存:

  • 首见刊(1885):RS2017 逐字:「G appeared, in all likelihood, for the first time in a publication by König and Richarz in the year 1885.」(其参考文献 29:König & Richarz, ‘Eine neue Methode zur Bestimmung der Gravitationsconstante’, Ann. Phys. 260(4), 1885)[一手逐字]
  • 定名与通行(1894–1890s 末):Wikipedia「Gravitational constant」条逐字:「’Newtonian constant of gravitation’ is the name introduced for G by Boys (1894)」、「The modern notation involving the constant G was introduced by Boys in 1894 … and becomes standard by the end of the 1890s, with values usually cited in the cgs system.」[二手导航级]

Falconer 2023(Historical Notes: The Gravitational Constant,arXiv:2306.06411)给出 1890 年代观念之争的逐字场景——Boys 在 1894-06-08 皇家研究院演说《The Newtonian Constant of Gravitation》中力主这个常数的普适性:「G… represents that mighty principle under the influence of which every star, planet and satellite in the universe pursues its allotted course… It is in no way dependent on the accidental size or shape of the earth; if the solar system ceased to exist it would remain unchanged…」(一手逐字);六年后 Poynting 转述:「Professor Boys has almost indignantly disclaimed that he was engaged in any such purely local experiment as the determination of the mean density of the earth. He was working for the Universe, seeking the value of G, information which would be as useful on Mars or Jupiter or out in the stellar system as here on the earth…」(同件[一手逐字])。Falconer 的定位句逐字:「G was a very recent innovation; until 1884 finding the mean density of the earth was the main purpose given for gravitational experiments…」[一手逐字]

叙事成形线:1798 卡文迪许测密度(5.48×水);1783 书信已有「weighing the world」;1884 年前「求地球平均密度」一直是这类实验的公开名目;1885 年 G 记号首见刊;1894 年 Boys 定名「Newtonian constant of gravitation」并力主其普适性;1890 年代末记号通行。「测 G」作为一个独立对象,比这个实验本身晚出生了将近一百年。

4.5 方法分类学:八大类仪器,为「不聚类」打底

RS2017 总览逐字:「A large array of different instruments ranging from the simple torsion balance to the sophisticated atom interferometer can be used to determine G.」[一手逐字] 八大类各一句(原理句均 RS2017 逐字,代表实验为第五章 16 值表的成员):

  1. 扭秤·静态偏转法(卡文迪许法):「The static deflection method was used by Cavendish to measure the mean density of the Earth. A static torque on a torsion balance causes a deflection from the equilibrium position…」——代表:卡文迪许 1798;BIPM 装置的自由偏转臂。
  2. 扭秤·周期法(time-of-swing,ToS):「the pendulum’s period is measured with the field masses in two different positions… The gravitational potential adds to the potential of the fiber, thereby increasing the restoring torque.」——代表:NIST-82、TR&D-96、LANL-97、HUST-99/05/09、UCI-14、HUSTT-18。Luther & Towler 1982 自述原理逐字:「The difference in the squares of the frequencies with and without the balls is proportional to G.」(PRL 48, 121[一手逐字])
  3. 扭秤·静电伺服法:「the gravitational torque of a modulated mass arrangement can be compensated by an electrostatic torque. The torsion balance acts as a null detector… Since no torsional excursion occurs, the measurement is not affected by the Kuroda effect…」——代表:BIPM-01/13/14 的伺服臂(Quinn 2013 逐字:「two modes of operation, free deflection (Cavendish) and electrostatic servo control」,PRL 111, 101102[一手逐字]);MSL-03。
  4. 扭秤·角加速度反馈法(AAF):「the torsion balance is mounted on a turntable… A feedback-control loop accelerates the turntable in such a way that the torsion bob does not move with respect to the rotating reference frame.」——代表:UWash-00、HUSTA-18。
  5. 单摆:Parks & Faller 自述逐字:「interferometrically measuring the change in spacing between two free-hanging pendulum masses」——代表:JILA-10/18。
  6. 梁秤:「The gravitational force of two stainless steel tanks filled with 13 521 kg mercury on 1.1 kg test masses was measured using a commercial mass comparator.」(Schlamminger et al. 2002, PRL 89, 161102[一手逐字])——代表:UZur-02/06。
  7. 自由落体:「This experiment does not suspend the test mass from a support system. It is therefore free of many systematic errors associated with supports.」(Schwarz et al. 1998, Science 282, 2230,Wayback 存档[一手逐字])——代表:JILA-98(半吨场源质量)。
  8. 冷原子干涉:「Here we report the precise determination of G using laser-cooled atoms and quantum interferometry.」(Rosi et al. 2014, Nature 510, 518[摘要级一手])——代表:LENS-14。

这一层的作用在第七章显现:分歧不随方法聚类——同一方法内部(HUST 两根丝、BIPM 两种模式、NIST 复测四支)照样对不齐,不同方法之间反而能互洽。

4.6 对象层小结

G 是定律改写时被造出来的比例常数;卡文迪许测的是地球密度;「测 G」作为独立对象 1890 年代才出生;它的测量是绝对力学实验——弱、不可屏蔽、无传递标准。这三个结构性事实此后两百年没变过,它们是第五、六章全部数字的背景重力。

五、张力账:16 个值互相差几个 σ(分歧的定量本体)

5.1 十六个输入值全表(CODATA 2022 Table XXX,本篇逐条回到一手文献核对)

CODATA 2022 平差的 G 输入以 Table XXX 全表刊出(RMP 97, 025002 (2025),APS 全文arXiv:2409.03787[一手逐字])。表头逐字:「The data are uncorrelated except for three cases with correlation coefficients r(NIST-82, LANL-97) = 0.351, r(HUST-05, HUST-09) = 0.134, and r(HUST-09, HUSTT-18) = 0.068.」

# 代号 实验(一手文献) 方法 CODATA-22 列值 (×10⁻¹¹ m³kg⁻¹s⁻²) 相对不确定度
1 NIST-82 Luther & Towler, PRL 48, 121 (1982) 扭秤周期法 6.672 48(43) 6.4×10⁻⁵
2 TR&D-96 Karagioz & Izmailov, Izm. Tekh. No.10, 3–9 (1996)(Springer 英译摘要页,Wayback[摘要级]) 扭秤周期法 6.672 9(5) 7.5×10⁻⁵
3 LANL-97 Bagley & Luther, PRL 78, 3047 (1997) 扭秤周期法(双丝检验 Kuroda 假说) 6.673 98(70) 1.0×10⁻⁴
4 UWash-00 Gundlach & Merkowitz, PRL 85, 2869 (2000) 扭秤角加速度反馈 6.674 255(92) 1.4×10⁻⁵
5 BIPM-01 Quinn et al., PRL 87, 111101 (2001) 扭带秤双模式 6.675 59(27) 4.0×10⁻⁵
6 UWup-02 Kleinevoß 博士论文(Univ. Wuppertal, 2002;原文未上网,经 CODATA-02 转引) 悬浮体位移补偿 6.674 22(98) 1.5×10⁻⁴
7 MSL-03 Armstrong & Fitzgerald, PRL 91, 201101 (2003) 补偿扭秤 6.673 87(27) 4.0×10⁻⁵
8 HUST-05 Hu, Guo & Luo, PRD 71, 127505 (2005) 扭秤周期法 6.672 22(87) 1.3×10⁻⁴
9 UZur-06 Schlamminger et al., PRD 74, 082001 (2006) 梁秤(13.5 t 汞+商用质量比较仪) 6.674 25(12) 1.9×10⁻⁵
10 HUST-09 Luo et al., PRL 102, 240801 (2009)Tu et al., PRD 82, 022001 (2010) 扭秤周期法 6.673 49(18) 2.7×10⁻⁵
11 BIPM-14 Quinn et al., PRL 111, 101102 (2013)勘误 PRL 113, 039901 (2014);详评 Phil. Trans. R. Soc. A 372, 20140032 扭带秤双模式 6.675 54(16) 2.4×10⁻⁵
12 LENS-14 Rosi et al., Nature 510, 518 (2014)[摘要级] 冷原子双干涉梯度计 6.671 91(99) 1.5×10⁻⁴
13 UCI-14 Newman et al., Phil. Trans. R. Soc. A 372, 20140025 (2014)(Wayback 全文 扭秤周期法(低温三丝) 6.674 35(13) 1.9×10⁻⁵
14 HUSTT-18 Li et al., Nature 560, 582 (2018)[摘要级] 扭秤周期法 6.674 184(78) 1.2×10⁻⁵
15 HUSTA-18 同上 扭秤角加速度反馈 6.674 484(77) 1.2×10⁻⁵
16 JILA-18 Parks & Faller, PRL 105, 110801 (2010)勘误 PRL 122, 199901 (2019) 单摆(自由悬挂双摆) 6.672 60(25) 3.7×10⁻⁵

表注三则(本篇核出的「CODATA 列值 ≠ 一手文献原值」处,原委全部在案):①NIST-82 原文 6.6726(5),CODATA-2010 按丝滞弹性修正 15.9(4.6)×10⁻⁶ 后收为 6.672 48(43);②UWash-00 原文 6.674215(92),作者 2002 年私人通信追加磁阻尼器扭矩修正 +6.0×10⁻⁶ 后收为 6.674 255(92)(CODATA-02 逐字转述,PRL 上不存在对应公开勘误——本篇实测 PRL 89(22) 两件勘误号均为他文);③UCI-14 论文摘要与结论自报三丝不带权平均 6.67433(13)(19 ppm),CODATA 表列 6.674 35(13)(1.9×10⁻⁵),数值上等于该文第一根丝的单独结果——两处都在文献中,采纳口径差异如实登记,不推测原因。

5.2 关键配对张力(全部本篇自算;n = |G₁−G₂|/√(u₁²+u₂²))

配对 \ Δ\ (ppm) 合成 u (ppm) 备注
BIPM-14 vs JILA-18 440.5 44.5 9.9σ CODATA 2022 认证的最大两残差点
BIPM-01 vs JILA-18 448.0 55.1 8.1σ
BIPM-14 vs HUSTT-18 203.2 26.7 7.6σ
BIPM-14 vs HUSTA-18 158.2 26.6 5.9σ
HUST-09 vs HUSTA-18 148.9 29.3 5.1σ 与 CODATA 2018 官方口径「about 5.1」一致
NIST-26 vs BIPM-14(同装置复测) 250.2 61.8 4.1σ 第八章
NIST-26 vs BIPM-01 257.7 69.8 3.7σ
HUST-09 vs HUSTT-18 104.0 29.4 3.5σ 与官方「about 3.5」一致
HUSTT-18 vs HUSTA-18(同组两法) 44.9 16.4 2.7σ 与官方「2.7 times」逐字一致
TR&D-96 vs UWash-00 203.0 76.2 2.7σ
NIST-26 vs CODATA-22 推荐值 64.4 61.2 1.1σ 论文自评「不显著异于零」
TR&D-96 vs JILA-18 44.9 83.8 0.5σ 两个「低点」改后几乎重合
BIPM-01 vs BIPM-14(同组同装置两代) 7.5 47.0 0.16σ 自洽
UCI-14 vs HUSTT-18 / HUSTA-18 24.9 / 20.1 22.7 / 22.6 1.1σ / 0.9σ 中段互相自洽
MSL-03 vs NIST-26 0.0 69.8 0.0σ 中心值完全相同(6.673 87),巧合

(自算口径:不确定度取 CODATA Table XXX 列值;NIST-26=6.673 87(38)。已知相关性的配对——NIST-82/LANL-97、HUST-05/HUST-09、HUST-09/HUSTT-18——不在本表上列。)

5.3 全局极差与官方口径互证

本篇自算:16 值中最大 BIPM-01 = 6.675 59,最小 LENS-14 = 6.671 91,极差 0.003 68×10⁻¹¹,相对推荐值 6.674 30 为 551 ppm;各值相对推荐值偏差落在 −358 ppm(LENS-14)到 +193 ppm(BIPM-01)之间,即 ±0.036% 带内;16 值中 9 个落在 ±150 ppm 内。551 ppm ≈ 最佳单实验自称不确定度(12 ppm,HUST-18 两值)的 46 倍

官方与综述口径逐字互证:WG13-2024「the relative spread of the results is on the order of 100 parts in 10⁶, while the best experiments show relative uncertainties of 12 parts in 10⁶」、「the difference between the smallest and largest measured value of G differ relatively by more than 500 × 10⁻⁶」[一手逐字];RS2017「The smallest reported relative standard uncertainty is 14 × 10−6. However, the difference of the largest reported result to the smallest reported result exceeds 500 × 10−6, more than 30 times the smallest uncertainty.」[一手逐字](其图取 14 值集,与 CODATA 16 值集略异)。三个口径互相咬合:分散带 >500 ppm,最佳自称 12–14 ppm,差 30–46 倍。

5.4 本篇自算与官方账的交叉验证(自算可信度锚)

  • 本篇自算(不计相关性的近似)加权平均 6.674 290,χ²=197.8,自由度 15,Birge 比 3.63;CODATA 官方扩张因子 3.9(政策为把全部归一化残差压到 2 以下,且计入三对相关性)——同量级。
  • 本篇自算相对推荐值 6.674 30 的归一化残差前三甲:BIPM-14 +7.75、JILA-18 −6.80、BIPM-01 +4.78——与 CODATA 2022 G 节逐字完全对上:「Before the uncertainties of the 16 values of G in Table XXX were multiplied by 3.9, the largest three normalized residuals were 7.7, 6.8, and 4.8 for BIPM-14, JILA-18, and BIPM-01, respectively.」[一手逐字]
  • HUST-18 两法互差:本篇自算 2.74σ vs CODATA 2018 逐字「the difference between the two new HUST values is 2.7 times the standard uncertainty of their difference」[一手逐字]。
  • BIPM 两值 vs CODATA 推荐值:Speake et al. 2023 摘要逐字「The discrepancy therefore amounts to about seven times the combined uncertainty of the latest BIPM and CODATA values.」(Metrologia 60, 024001[一手逐字]);本篇按 Table XXX 复算约 5.7σ(按 PRL-13 原值约 4.9σ)——「about seven」为约数,量级一致。

5.5 分歧的结构:不是白噪声,是两端拉扯

把 16 值按数值排队看到的不是围绕推荐值的高斯散布,而是有结构的对抗:高簇=BIPM-01(+193 ppm)与 BIPM-14(+186 ppm,同一装置两代、互相 0.16σ 自洽);低簇=LENS-14(−358)、HUST-05(−312)、NIST-82(−272)、JILA-18(−255 ppm);中段自洽簇=UWash-00(−6.7)、UZur-06(−7.5)、UCI-14(+7.5)、HUSTT-18(−17.4)、HUSTA-18(+27.6 ppm)贴着推荐值。同一团队内部照样对不齐:HUST 自己 2018 年用两种方法测出的两个值互差 2.7σ,且都高于自己 2009 年的值(3.5σ/5.1σ)——CODATA 2018 对此的逐字判语是:「Presently, there are no explanations for the inconsistencies.」(CODATA 2018 §XIX.B[一手逐字])

5.6 张力层小结

分歧是真的、定量的、有结构的:最紧的一对互相排斥到 9.9σ——在粒子物理的惯例里,5σ 就够宣布发现(Müller 在 PNAS 特写里逐字:「In particle physics, if you have an anomaly in your signal that’s five times the error bars … then you can claim discovery of a new effect,」「The disagreement in the measurements of Big G passes that threshold.」[一手逐字])。但这条带不是「无知的深渊」:16 个值全部挤在 ±0.036% 以内。这是过度声称的分歧——带是被自称 12–150 ppm 的尺子们撑开的,不是被无校准的粗测量撑开的。「张力该读成什么」是第九、十一章的事;「官方拿这条带怎么办」是下一章的事。

六、换尺账:CODATA 五十年编年——不确定度不是测出来的,是抚出来的

G 在 CODATA 平差里的特殊地位,1986 年报告逐字写死:「There is no established relationship between the gravitational constant G and other physical quantities; it stands completely uncoupled from the remainder of the adjustment.」(CODATA 1986, RMP 59, 1121[一手逐字])——G 与任何其他常数无耦合,每届单独做加权平均。所以 G 的推荐值史,是一部纯粹的「输入集+统计处置」史。

6.1 十届编年总表

平差 推荐值 (×10⁻¹¹ m³ kg⁻¹ s⁻²) 相对不确定度 G 输入数 特别处置 一手出处
1969 6.6732(31) 460 ppm 1(Heyl & Chrzanowski 1942 之值) RMP 41, 375
1973 6.6720(41) 615 ppm 2(Heyl 1930+Heyl & Chrzanowski 1942 加权平均;另 2 值明言排除) JPCRD 2, 663
1986 6.67259(85) 128 ppm 1(Luther & Towler 1982;原报 64 ppm) 人为加倍 64→128 ppm RMP 59, 1121
1998 6.673(10) 1500 ppm 0(不做数值分析,沿用 1986 值) 不确定度放大约 12 倍 RMP 72, 351
2002 6.6742(10) 150 ppm 8 加权均值 Birge 比 2.87,TG 酌定不确定度 RMP 77, 1
2006 6.674 28(67) 100 ppm 8 Birge 比 2.35;政策性定价(6.5) RMP 80, 633
2010 6.673 84(80) 120 ppm 11 扩张因子 14 RMP 84, 1527
2014 6.674 08(31) 47 ppm 14 扩张因子 6.3 ri\
2018 6.674 30(15) 22 ppm 16 扩张因子 3.9 RMP 93, 025010
2022 6.674 30(15)(与 2018 相同) 22 ppm 16(无新输入) 沿用 3.9 arXiv:2409.03787NIST 现行值页

五十年总账:推荐值从 6.6732 走到 6.674 30,漂移 165 ppm;不确定度从 460 ppm 收到 22 ppm,约 21 倍改善——但路径上有两次不降反升,每一次都是「尺」的事件。

6.2 1986:人为加倍——「we shall arbitrarily double the uncertainty」

Luther & Towler 1982(NIST-82)原报 64 ppm。1986 平差采纳它的同时把不确定度翻倍,逐字:「We therefore adopt the Luther and Towler result, Eq. (2.21), as the recommended value, but in view of the difficulties in properly evaluating the systematic uncertainties of this type of measurement and the limited number of runs carried out, we shall arbitrarily double the uncertainty; it therefore appears below in Tables VI and VII as 128 ppm.」(CODATA 1986 §II.3[一手逐字])。这是 CODATA 史上第一次为 G「造」不确定度——理由是这类测量的系统误差难以恰当评估。加倍这个动作本身就是政策,不是统计。

6.3 1998:1500 ppm 是「chosen」出来的——PTB 事件

1995 年 PTB(德国联邦物理技术研究院)报出 6.71540(56)×10⁻¹¹,比 1986 推荐值高 0.7%。CODATA 1998 逐字:「The PTB value for G exceeds the 1986 CODATA recommended value by 42 udiff … and hence the two values are in severe disagreement.」(CODATA 1998 §III.Q[一手逐字])

工作组的处置不是做数值分析,而是直接把不确定度放到 1500 ppm:「The 1998 value, G = 6.673(10) ×10⁻¹¹ m³ kg⁻¹ s⁻² [1.5×10⁻³], is the same as the 1986 value but its uncertainty is about a factor of 12 larger.」[一手逐字]——中心值不动,不确定度放大 11.5 倍(官方措辞「约 12 倍」;任务记忆线索「17 倍」不成立,附录 A 登记)。

放大的方式是把不确定度当成一个可选参数来定,逐字:「(vi) The convenient standard uncertainty u(G98) = 0.010×10⁻¹¹ m³ kg⁻¹ s⁻² meets these requirements. Chosen so that GPTB − G98 ≈ 4 udiff, it has the effect of reducing the discrepancy between the PTB value and the recommended value by a factor of 10 and producing a recommended value that encompasses all other values, except that from PTB, to within about 1.5 times the recommended value’s standard uncertainty.」[一手逐字]——不确定度被「选」成恰好让 PTB 值落在 4 倍合成差的位置。这届平差对 G 甚至没有做数值分析:「Because we do not obtain the 1998 recommended value of G from an in-depth numerical analysis of the available data, we do not give a detailed review of the values of G and their uncertainties …」[一手逐字]

6.4 2002:骤降 10 倍——离群值自我否定后被「安全剔除」

PTB 组自己复查后否定了自己的值。CODATA 2002 逐字:「experimental investigations of several critical aspects of the PTB determination of G recently carried out and reported by PTB researchers (Michaelis et al., 2004) have led these authors to conclude that the PTB result for G and its uncertainty cannot be considered correct.」(CODATA 2002 §III.Q[一手逐字]);总结段逐字:「an earlier, credible result in quite significant disagreement with all other values of G can now be safely omitted from the 2002 adjustment. Although the new values have allowed a factor-of-10 reduction in the uncertainty of the recommended value of G, they are still not completely consistent, thereby implying that some problems in experiments to measure G still remain.」[一手逐字]

8 个新输入(UWash-00 达 14 ppm)加权平均,χ²=57.7、自由度 7、Birge 比 2.87,工作组酌定不确定度 150 ppm——1500→150,一年到位的「精度恢复」同样是汇编事件:是一个输入值被移出样本集造成的,不是谁的实验突然变好了。

6.5 2006:政策性定价——不确定度当科研激励工具

2006 年报告把定价逻辑写成明文,逐字:「In assigning this uncertainty to the 2006 recommended value of G, the Task Group recognized that if the uncertainty was smaller than really justified by the data, taking into account the history of measurements of G, it might discourage the initiation of new research efforts to determine G … Such efforts need to be encouraged … On the other hand, if the uncertainty were too large … then the recommended value would not have reflected the fact that we now have two data that are in excellent agreement, have ur less than 2×10⁻⁵, and are the two most accurate values available.」(CODATA 2006 §X.B[一手逐字])

不确定度定小了会打击新研究的积极性、定大了又辜负那两个最好的实验——官方不确定度在这里公开承认自己是政策工具,不是统计读出。这是换尺跳最硬的一块化石:尺的刻度是开会定出来的。

6.6 2010:不降反升 20%——扩张因子 14

两个新值(JILA-10、HUST-09)互相及与旧值冲突:55 个两两差值中 3 个超过 10 倍合成不确定度,加权均值 χ²=209.6(自由度 10)、Birge 比 4.58。工作组的处置逐字:「the Task Group decided to take as the 2010 recommended value the weighted mean of the 11 values in Table XXIV after each of their uncertainties is multiplied by the factor 14. This yields G = 6.67384(80)×10⁻¹¹ m³ kg⁻¹ s⁻² [1.2×10⁻⁴].」、「In view of the significant scatter of the measured values of G, the factor of 14 was chosen so that the smallest and largest values would differ from the recommended value by about twice its uncertainty」(CODATA 2010 §XIII.B.1[一手逐字])——不是用 Birge 比 4.58,而是主动取 14,让最两端的值都落在 ±2u 内。反升逐字确认:「The 2010 recommended value represents a fractional decrease from the 2006 value of 0.66×10⁻⁴ and an increase in uncertainty of 20%.」[一手逐字]

6.7 2014→2022:回归惯例,然后冻结

2014 年,若沿用 2010 的「±2u 包两端」标准,需要的扩张因子约 16;工作组决定改回惯例:「After due consideration the Task Group decided that it would be more appropriate to follow its usual approach of treating inconsistent data, namely, to choose an expansion factor that reduces each |ri| to less than 2.」(CODATA 2014[一手逐字])——于是因子 6.3、47 ppm。2018 年降到 3.9、22 ppm:「The measurements are inconsistent and an expansion factor of 3.9 is required to bring all residuals to within a factor of two from the 2018 recommended value…」、「The residuals of the data from BIPM-14 and JILA-18 are the largest and determined our expansion factor.」(CODATA 2018 §XIX[一手逐字])——因子的数值由最不听话的两个值(BIPM-14 与 JILA-18,即第五章那对 9.9σ)直接定价

2022 年没有新输入,原样冻结:「Since there is no new value, the same 3.9 expansion factor applied to their uncertainties in 2018 to reduce their inconsistencies to an acceptable level is also used in 2022; the 2022 and 2018 recommended values of G are therefore identical.」(CODATA 2022[一手逐字])——16 个测量值自 2018 年起一个没动过,动的只有账本格式。

6.8 方法论自陈:扩张因子是什么

CODATA 2018 的定义句逐字:「A measure of the consistency of our least-squares adjustment for the ith input datum Xi is its normalized residual ri = (Xi − ⟨Xi⟩)/u(Xi) … An absolute value greater than two is problematic and is reduced to less than two by the application of a multiplicative expansion factor to the initially assigned uncertainties of the input datum in question as well as related input data. … This procedure makes the effective data consistent.」[一手逐字]

最后这句是全章的题眼:扩张因子不消除分歧,它把分歧折算进不确定度,让数据「在效果上」一致。现行 22 ppm 的构成因此可以拆开:扩张前加权均值的原始不确定度约 5.8 ppm(本篇自算:22.5÷3.9),乘上 3.9 得到官方的 22 ppm——22 ppm 不是测出来的,是抚出来的;而抚平的动作白纸黑字印在同一篇论文里

6.9 换尺层小结

五十年里推荐值只漂移了 165 ppm,不确定度从 460 收到 22 ppm;两次反升(1998、2010)都是汇编事件——一次为离群值「chosen」不确定度,一次给所有输入乘 14。消费层那句「G 越测越不准」(第十章)的两次真实对应物,在这里现出原形:那不是测量能力的退化曲线,是汇编政策的定价记录

七、机制账:分歧藏在哪(系统误差排查的四十年)

分歧不是玄学:四十年里每一个被怀疑的系统误差源头都被人提出、检验、部分排除。本章按「已确认/已排除/被抓到的错误/仍开」四栏结账。

7.1 已确认:丝滞弹性——唯一被定量化的机制,一个教科书级科学循环

提出(1995):Kuroda(PRL 75, 2796)指出滞弹性固体模型意味着弹性本身依赖频率,对周期法(ToS)测 G 直接相关:「A consequence of the anelastic solid model is that not only the dissipation but also the elasticity is frequency dependent. … in this paper I point out that it is highly relevant to high-precision quantitative applications such as time-of-swing measurements.」(全文[一手逐字])。后果:丝的频率依赖弹性抬高摆动频率,用牛顿公式反推时 G 被系统性高估,偏估量级 1/(πQ)。Kuroda 自己写下适用范围:「Of course, anelasticity is only one of many possible error sources in measurements of G, and is not relevant to certain other recent measurements, which work on different principles.」[一手逐字]

检验(1997):Bagley & Luther(LANL-97,PRL 78, 3047)论文标题即《…A Test of the Kuroda Hypothesis》。同一摆、同频率、同气压,两根钨丝 Q=950 与 Q=490:「Kuroda has predicted that such determinations have an upward bias inversely proportional to the oscillation Q, and our results support this conjecture.」(全文[一手逐字])

直测修正(2009–2010):HUST-09(Tu et al., PRD 82, 022001)用高 Q 石英丝辅助双盘摆首次直接测量该效应:「the anelastic effect of the torsion fiber is first measured directly by using two disk pendulums with the help of a high-Q quartz fiber.」;修正量逐字:「brings in a correction of −211.80 ppm to our G value and contributes an uncertainty of 18.69 ppm」(全文[一手逐字])——这是该实验误差预算里最大的单项修正(几何项 −24.28 ppm、非线性项 −39.83 ppm 等都小一个量级)。

设计级回应(2014 起):UCI-14 引言传述该模型逐字:「a widely accepted model for linear anelastic behaviour, with reasonable parameter assumptions as investigated by Quinn et al., leads to an upward fractional bias in a G measurement of 1/(π Q)」,然后给出回应——4 K 以下低温摆+三根不同的丝(Q 分别 82,000/120,000/164,000)「minimizing experimental bias from fibre anelasticity」(Newman et al. 2014,Wayback 全文[一手逐字])。BIPM 则从结构上绕开:「97% of the restoring torque of the loaded torsion strip is gravitational, which is lossless, and only 3% of the total stiffness is due to the elasticity of the strip. This essentially eliminates the problem of frequency-dependent stiffness owing to anelasticity.」(Quinn et al. 2014 详评,Phil. Trans. R. Soc. A 372, 20140032,Wayback 全文[一手逐字])

裁决:滞弹性是真实机制——提出→检验→直测→设计规避,二十年走完一个完整循环。但注意它修的是 ToS 法的上偏;它不能解释全局分散,尤其不能解释 BIPM(结构免疫)与 JILA(不用丝的单摆)之间的对立。

7.2 已排除:杂散交变磁场假说(2023)

BIPM 两值比 CODATA 高约 200 ppm,Kibble 曾怀疑是环境交变磁场在装置铁磁部件上产生寄生力矩。Speake et al. 2023(Metrologia 60, 024001)做了专门实验:动机句逐字「The discrepancy therefore amounts to about seven times the combined uncertainty of the latest BIPM and CODATA values.」;结论逐字:「A bias due to the spurious torques due to ambient magnetic fields cannot … resolve the discrepancy between the BIPM results and the CODATA recommended value.」(全文[一手逐字])——环境场下偏估上限 <20 ppm,远小于 Type A 不确定度;且实测力矩符号与假说相反:「observed a reduction in the gravitational torque, countering the intuition of Bryan Kibble」[一手逐字]。一个流传二十年的假说被一个专门实验定量关死。

7.3 被抓到的错误:纠错编年史(每条一手)

事件 量级 一手出处
1995→2002 PTB 值离群 0.7%(42 倍合成差),原作者机构复查后自我否定、撤出 曾驱动 CODATA 1998 放大 12 倍 CODATA 1998CODATA 2002;PNAS 2016 逐字「an experiment in 1995, shockingly, came up with a value for G that differed from the accepted value by 0.7%, in addition to causing a 12-fold increase in its uncertainty … Although the measurement turned out to be flawed…」
2000→2002 UWash-00:丝顶端小质量效应,私人通信修正 +6.0×10⁻⁶ 6.674215→6.674255(92) CODATA-02 转述[一手逐字]
1999→2005 HUST-05 自我修正:源质量质心偏心+空气浮力 6.6699(7)→6.6723(9),+360 ppm HUST-05, PRD 71, 127505 逐字「our value of G should be revised to be 6.6723(9)×10⁻¹¹ m³ kg⁻¹ s⁻², which is 0.036% larger than our previous published value」
2001→2013 BIPM 装置完全重建(Mk II),两法双模式 −21 ppm(6.67559(27)→6.67545(18)) Quinn et al. 2013, PRL 111, 101102
2013→2014 BIPM-13 勘误:源质量密度梯度修正「inadvertently applied twice」等三处错误 +13 ppm(6.67545(18)→6.67554(16),CODATA 表列后者) Erratum, PRL 113, 039901 (2014)[一手逐字]——BIPM-13 与 BIPM-14 两个公布值的差异由此销账
2010→2019 JILA-10 两处计算错误(弹簧常数旋转修正算错+Abbe 误差漏算),装置运到 NIST 备展时被发现 +3.9×10⁻⁵;ur 2.1→3.7×10⁻⁵;6.672 34(14)→6.672 60(25) Erratum, PRL 122, 199901 (2019) 逐字「The result of correcting these two errors is an increase in the value for the gravitational constant of +3.9×10⁻⁵…」;发现者署名:Patrick Egan 与 Jack Stone;CODATA 2018 逐字「During initial preparations at NIST, two calculational errors were discovered…」

这张表的读法:每一个「离谱的值」背后,最后找到的都是具体的、甚至可以说无聊的原因——一处修正用了两次、一个毫米级的轴高差、丝顶端一块小质量、质心偏心。没有任何一例指向新物理。

7.4 仍开:全局分散无机制解释+暗不确定度

WG13 2025 年报逐字:「No method-dependent systematic error can be identified.」[一手逐字]——把 16 值按方法分组,看不到偏差随方法聚类。RS2017 给出结构性原因:「The principal problem is that the set of the known systematic and statistical effects is only a subset of all systematic and statistical effects that can perturb an experiment. Hence, regardless of how much effort is spent, the uncertainty budget can never be complete. There will always be an unknown uncertainty, often referred to as a dark uncertainty.」[一手逐字]

统计学侧的回应已经成文:Merkatas, Toman, Possolo & Schlamminger 2019(Metrologia 56, 054001)用「不同深浅的暗不确定度」潜簇贝叶斯混合模型重建共识:「both reflecting the presence of dark uncertainty … We demonstrate this procedure by deriving a new estimate for G, as a consensus value G = 6.67408 × 10⁻¹¹ …, with u(G) = 0.00024 × 10⁻¹¹ …」[一手逐字]——即 36 ppm(本篇自算换算),比 CODATA 表观的 22 ppm 大 64%:一个不经过扩张因子抚平的独立统计方法,给出的不确定度比官方更大。这套方法随后被 NIST-26 论文原样采用合成其四支测量(第八章)——方法学直接进了最新实验论文。

7.5 机制层小结

已确认的机制(滞弹性)修不掉全局;已排除的假说(磁场)证明排查在认真进行;被抓到的错误全是具体错误;仍开的是「能解释 551 ppm 全局分散的单一机制——无」。收敛跳在这一层现形最清楚:每确认一个机制、每抓到一处错误,新闻稿都可以写「分歧之谜接近解开」,而账本上那条带一毫米都没有变窄。

八、复测账:同一台装置,三十年后,另一个实验室(本篇最硬格)

8.1 一个从不复测的领域

RS2017 的两句逐字是本章的地基:「most measurements are performed by small groups—publications with only two authors are not uncommon in this field. … This makes it difficult to build institutional memory.」、「experiments are not repeated … no two identical G experiments have ever been repeated.」[一手逐字]——G 领域两百年没有过一次「同一台装置、换一个团队」的复测。所有「不一致」都可以被「装置不同」这个万能解释吸收。

8.2 谱系:从巴黎到盖瑟斯堡

2014-11-07 NIST 研讨会上,「Terry Quinn … proposed to send his torsion balance to NIST so that an independent team could attempt a measurement of G using the same instrument」(NIST-26 论文逐字,Metrologia 63, 025012 (2026)[一手逐字])。BIPM-14 所用的 Mk II 装置后来运抵 NIST——CODATA 2022 亦逐字载明:「We note that the balance used in the BIPM-14 experiment was transferred to NIST for use there to measure G (Schlamminger et al., 2022).」[一手逐字]

8.3 盲测程序:一个信封

这是 G 测量史上第一次带盲测程序的实验(NIST 官方新闻,2026-04-16《NIST Weighs In on the Mystery of the Gravitational Constant》[一手逐字]):同事 Abbott 把关键修正数秘密封存进信封;2022 年 Schlamminger 曾因发现气压相关疏漏而临场拒绝揭封——修正后重新分析;「Now, at 3 p.m. on July 11, 2024, Schlamminger was scheduled to report his findings at the annual Conference on Precision Electromagnetic Measurements in Aurora, Colorado.」——在 CPEM 会场当场揭封。新闻稿逐字记下揭封瞬间:「To get the results he expected, the secret number needed to be relatively large and negative. It was.」[一手逐字]

8.4 结果:从 +200 ppm 翻到 −64 ppm

论文摘要逐字(数学式按其 LaTeX 替代文本还原):「We report the first replication of a high-precision measurement of the gravitational constant, G. The experiment employed the torsion balance originally designed and constructed at the International Bureau of Weights and Measures (BIPM) approximately three decades ago. Using the same apparatus and geometry, with several modifications documented in this work, we determined G = (6.67387 ± 0.00038) × 10⁻¹¹ m³ kg⁻¹ s⁻², corresponding to a relative standard uncertainty of 5.7 × 10⁻⁵. The result is lower by 2.5 × 10⁻⁴ relative to the BIPM determination.」(Metrologia 63, 025012 (2026),2026-04-16 上线[一手逐字])

四支独立测定(论文 Table 16,铜/蓝宝石源质量 × 伺服/自由偏转,逐字):铜伺服 6.673 642(23 ppm)、铜自由偏转 6.674 021(30 ppm)、蓝宝石伺服 6.672 637(38 ppm)、蓝宝石自由偏转 6.673 636(94 ppm,×10⁻¹¹ m³kg⁻¹s⁻²)。伺服臂恒低于自由偏转臂:「At present, we do not have a conclusive explanation for this systematic offset.」——且「the Mark II apparatus exhibited a comparable behavior」:BIPM 当年同一台装置就有同方向的内部偏移[一手逐字]。

解读账(本篇自算,注明口径):相对 CODATA 推荐值 6.674 30,BIPM-14 在 +186 ppm,NIST-26 在 −64 ppm——同一台装置在两个实验室的偏差方向相反;NIST-26 与 BIPM-14 差 250 ppm、4.1σ(第五章配对表);与 CODATA 推荐值差 1.1σ,论文自评「the difference between the value we measured for G and the CODATA recommended value does not differ significantly from zero」[一手逐字]。(注意双口径并存:论文摘要写低 2.5×10⁻⁴ 即 250 ppm,NIST 新闻写「0.0235% lower」即 235 ppm——两处均逐字在案,差异源自对比基准是 BIPM-13 原值还是 BIPM-14 勘误后值,本篇不强行合并。)

8.5 裁决句与不确定度自陈

论文「Parting words」逐字:「Our result does not corroborate either the earlier BIPM value or the current CODATA recommendation. Relative to CODATA, the BIPM determination lies at [行内 MathML 数值在文本层丢失] and ours at [同上], placing the two on opposite sides.」(两个丢失数值可由公开值自算补上:+186 ppm 与 −64 ppm)[一手逐字]

正式版 57 ppm 的不确定度比 CPEM 2024 报告的初步值(24.1 ppm,WG13-2025 逐字「turned out to be lower than the current CODATA value … with a preliminary combined uncertainty of 24.1 ppm.」[一手逐字])大了一倍多,原因论文自陈:「This increase does not reflect a degradation in measurement performance, but rather a more complete accounting of systematic limits.」、「By explicitly including the ‘dark uncertainty’ driven by the scatter among our individual datasets, we believe this result provides a more robust estimate of the true accuracy achievable with this apparatus.」[一手逐字]——把四支测量之间的散布显式折进不确定度:CODATA 对 16 个值做的事,NIST 对同一台装置内部的 4 个值做了一遍。同一个动作,两个尺度。

8.6 复测层小结

装置级复测的第一次落地,结果不是收敛,而是把分歧刻进了同一台装置:巴黎的 +200 ppm 与盖瑟斯堡的 −64 ppm 出自同一台扭秤。论文动机句逐字:「This persistent scatter points to two possibilities: new physics or unrecognized instrumental effects. The first hypothesis, unexplained physics, remains remote.」[一手逐字]——做实验的人自己把「新物理」排在远端。收敛跳的最贵反面教材在此:复测不等于收敛,它只是把「哪里不一致」的定位精度提高了一个量级。

九、危机叙事账:「G 随昼长振荡」案的十五个月(张力跳的最贵样本)

9.1 宣称

2015 年 4 月,Anderson、Schubert、Trimble & Feldman(JPL/UCLA/UC Irvine)在 EPL 110, 10002 发表对历次 G 测量值的时间序列分析(arXiv:1504.06604[一手逐字])。摘要逐字:「We find that these values for G are oscillatory in nature, with a period of P = 5.899 ± 0.062 yr, an amplitude of (1.619 ± 0.103) × 10⁻¹⁴ m³ kg⁻¹ s⁻², and mean-value crossings in 1994 and 1997.」;与地球昼长(LOD)的 5.9 年振荡周期与相位相同:「We report the G/LOD correlation, whose statistical significance is 0.99764 assuming no difference in phase, without claiming to have any satisfactory explanation for it.」。正文力度宣称逐字:「Figure 1 appears to provide convincing evidence that there exists a 5.9 year periodicity in the macroscopic determinations of G in the laboratory with variations at the level of ∆G/G ∼ 2.4 × 10⁻⁴ … with a much smaller standard error of 10.3 ppm instead of the CODATA recommended error of 120 ppm.」——即:把 13 个测量值按「实验运行时间」(「our best estimates of when the experiments were run, not the publication dates」)排到时间轴上,正弦拟合的表观不确定度比 CODATA 官方小 12 倍。

9.2 原文自带的退路(危机叙事在诞生处被自己切断)

同一篇摘要里紧接着的限定逐字:「However, we do not suggest that G is actually varying by this much, this quickly, but instead that something in the measurement process varies.」;结论节更彻底:「The least likely explanation is a new-physics effect that could make a difference in the macroscopic and microscopic determinations of G.」[一手逐字]——作者自己把「G 在变」和「新物理」两个读法都预先划掉。后世的危机传播(「常数在振荡」)连原文的第一段都没读完。

9.3 反驳链:十五个月闭合

  1. Schlamminger, Gundlach & Newman,PRD 91, 121101(R),2015-06-11全文arXiv:1505.01774[一手逐字]):汇总 35 年的 G 测量,摘要逐字「A least-squares regression to a sinusoid with period 5.9 years still yields a better fit than a straight line. However, our additions and corrections to the G data reported by Anderson et al. significantly weaken the correlation.」——13 个点扩为 19 个测量区间,指出原文「several points … are not plotted at the right time and one experiment is missing」:时间赋值错了。结论节逐字:「The situation is disturbing—clearly either some strange influence is affecting most G measurements or, probably more likely, the measurements have unrecognized large systematic errors. The need for new measurements is clear.」
  2. Pitkin 评论,EPL 111, 30002,2015-08arXiv:1505.06725[一手逐字]):贝叶斯模型选择,结论逐字「a model containing an additional unknown Gaussian noise component is hugely favoured, by factors of ≳ e³⁰, over two models allowing for a sinusoidal component.」、「I also find that periodic signals can easily be found in random permutations of the data suggesting that the observed periodicity seen in [1] is just a random artifact of the data.」——数据更支持「测量带额外噪声」;在数据的随机排列里也能轻松找到「周期」。
  3. Anderson 等答辩,EPL 111, 30003,2015-08arXiv:1508.00532[一手逐字]):「we were unable to replicate his claims with our own independent analysis」、「Thus, we stand by our conclusions of potential periodic terms in the reported G measurements」——同刊同期评论+答辩结构在案。
  4. Desai,EPL 115, 20006,2016-07arXiv:1607.03845[一手逐字]):频率主义模型比较复验,结论逐字「We find in agreement with Pitkin that the best model is the one with a constant offset in measurements of G along with an unknown systematic offset. Therefore, there is no evidence for any sinusoidal variations in the measurements of G.

从见刊(2015-04)到贝叶斯与频率主义两条独立路径各自闭合(2016-07),十五个月

9.4 媒体当月就泼了冷水

Science News 2015-04-30(原文见刊当月),Andrew Grant 报道标题逐字《Explanation for G’s imprecision stumbles》,副题「Possible link of Earth’s rotation rate to gravitational constant questioned」;deck 逐字:「A surprising new result seems to suggest that subtle changes in Earth’s rotation rate could account for physicists’ difficulty in measuring Newton’s gravitational constant. But some confusion with dates appears to derail the finding.」(Science News[一手逐字])——日期赋值问题在见刊当月就被媒体点破(指当时尚未见刊的 Schlamminger 预印本)。

9.5 现状(检索日 2026-08-16)

  • 未撤稿:Crossref 记录的 relation/update-to 字段皆空,无 retraction/erratum 挂接[多源检索];被引数(Crossref,检索日):Anderson 原文 63,反驳链 33/13/5/12——原文被引仍高于任何一篇反驳。
  • 官方沉默:CODATA 2018 与 2022 全文 python 检索「Anderson」「length of day」「oscillat」(G 语境)均 0 命中[多源检索]——两届平差报告对该宣称不置一词。
  • 交锋止于 2016:Desai 2016 是可查的最后一件交锋文献;arXiv 题名检索仅返回原文/评论/答辩三件[多源检索]。
  • 原作者的 v2 退让(arXiv 附录,回应 Schlamminger 后追加,逐字):「the sample correlation coefficient is reduced to 0.860. … this makes the interpretation of a possible correlation of the G measurements with LOD more problematic, with the similar periods near 5.9 years possibly a coincidence.」[一手逐字]

9.6 真正的 Ḡ/G 约束:两条天体物理通道封死时间变化

变 G 的理论谱系(Williams, Turyshev & Boggs 2004 引言逐字):「The possibility of a time variation of the constant of gravitation, G, was first considered by Dirac in 1938 on the basis of his large number hypothesis and was later developed by Brans and Dicke in their theory of gravitation…」(PRL 93, 261101[一手逐字])——这个想法从 1938 年起就有理论动机,因此有独立的硬约束:

  • 月球激光测距(Williams et al. 2004)摘要逐字:「The search for a time variation in the gravitational constant results in Ḡ/G = (4 ± 9) × 10⁻¹³ yr⁻¹; consequently there is no evidence for local (∼1 AU) scale expansion of the solar system.」;正文自评该不确定度「is 83 times smaller than the inverse age of the Universe」[一手逐字]。
  • 脉冲星计时:Lazaridis et al. 2009(MNRAS 400, 805)逐字:「Ḡ/G = (−0.7 ± 3.3) × 10⁻¹² yr⁻¹ … Our pulsar test therefore restricts Ḡ/G to less than a 20th of the expansion rate of the Universe.」[一手逐字];Zhu et al. 2019(PSR J1713+0747,MNRAS 482, 3249)逐字:「Ḡ/G = (−0.1 ± 0.9) × 10⁻¹² yr⁻¹, which is weaker than Solar system limits, but applies for strongly self-gravitating objects.」[一手逐字]

量级账(本篇自算):Anderson 宣称的表观年际摆动 ∆G/G ∼ 2.4 × 10⁻⁴,对 LLR 给出的真实漂移上限 10⁻¹³ /年——差约 9 个数量级。实验室「分歧」与「G 是否随时间变化」是两本账:前者是测量史现象,后者被观测封死。

9.7 危机叙事层小结

Anderson 案是一个危机叙事完整生命周期的标本:宣称自带退路 → 同行十五个月内两条独立统计路径否决 → 媒体当月点破日期问题 → 不撤稿、不跟进、官方零提及。它的尸体留在文献里(EPL 110, 10002 至今可引),但无人认领。把这条案例与 9.6 合读,「G 在振荡」这个读法的全部承重墙都已拆除——而它的废墟恰好是本篇主题的反面教材:把「测量过程的性质」读成「常数的性质」,就是张力跳本身。

十、消费账:「最测不准的常数」名号化石谱系

10.1 CODATA 官方措辞的四代化石(全文检索,python 计数)

对库内十届 CODATA 平差报告全文做关键词扫描(least precisely / least accurately / least well / most poorly / poorly known,限定 G 语境):

届次 逐字措辞 出处
1986 G 节只有结构句「it stands completely uncoupled from the remainder of the adjustment」,无修辞 RMP 59, 1121
1998 「this measure of the strength of the most pervasive force in the universe is so poorly known … because of gravity’s central role in physics, the large uncertainty of G is disconcerting.」 RMP 72, 351[一手逐字]
2010 「Clearly, there is a continuing problem for the determination of this important, but poorly known, fundamental constant」 RMP 84, 1527[一手逐字]
2014 无最高级;G 节逐字「have not resolved the considerable disagreements that have existed among the measurements of G for the past 20 years」 RMP 88, 035009[一手逐字]
2018 among the most poorly known constants in our 2018 adjustment」 RMP 93, 025010[一手逐字]
2022 「The value of G has been the least well known of the major fundamental constants for decades.」 arXiv:2409.03787[一手逐字]

化石序列:so poorly known(1998)→ poorly known(2010)→ among the most poorly known(2018)→ the least well known(2022)——最高级是 2022 年才进 CODATA 正文的。(1969/1973 两届为 OCR 扫描层,关键词零命中受 OCR 质量限制,只登记「未见」、不证「不存在」,附录 B。)

10.2 论文与综述层:精确最高级的可证起点

  • 1982,Luther & Towler(PRL 48, 121,OCR 扫描层仅引干净句):「Of the several attempts to improve this measurement made during the past 39 years, none seems to have increased significantly the precision of this, the internationally accepted value of G」[一手逐字]——库内最早的「几十年没进步」逐字实证。
  • 1997,Gillies 综述(Rep. Prog. Phys. 60, 151,摘要经 INSPIRE 取回):「Most other constants of nature are known (and some even predictable) to parts per billion, or parts per million at worst. However, G stands mysteriously alone …」(INSPIRE 记录[摘要级])
  • 2000,Gundlach & Merkowitz(PRL 85, 2869 全文):「Unlike most other physical constants, the value of G is not precisely known due to the weakness and nonshieldability of gravity. Since the first laboratory measurement by Cavendish over 200 years ago, the reduction in uncertainty in G has been only a factor of about 10 per century.」[一手逐字]
  • 2018-08-29,Li et al.(Nature 560, 582,摘要层 meta description 逐字):「The Newtonian gravitational constant, G, is one of the most fundamental constants of nature, but we still do not have an accurate value for it. Despite two centuries of experimental effort, the value of G remains the least precisely known of the fundamental constants. A discrepancy of up to 0.05 per cent in recent determinations of G suggests that there may be undiscovered systematic errors in the various existing methods.」[摘要级一手]——精确最高级「the least precisely known of the fundamental constants」的可证起点即此文;一次定向检索(排除 2018 派生件)未见更早用例,命中者全部为 Li 2018 原文或其转引[多源检索]。
  • 2020,Xue et al.(Natl. Sci. Rev. 7, 1803,PMC8290936 全文):「Compared with the uncertainties of previous recommended values, it is improved by a factor of 2. But this value remains the least precisely known among all of the fundamental constants.」[一手逐字]

即:媒体层的「humanity still doesn’t have an accurate value」(2014-10,Gibney)→ 论文摘要层的最高级(2018-08)→ CODATA 正文的最高级(2022)。一条话术的三级跳,每一级都有逐字化石。

10.3 媒体层:分歧叙事在 2018 之前已是常态

  • 2014-02-13,Physics World 特写《The lure of G》(页面):「measurements of big G are in wild disagreement with one another」[一手逐字]。
  • 2014-06-30,Speake & Quinn,Physics Today 67(7), 27(Wayback 快照):「The actual numerical value of G is of little consequence to physics. … What matters, then, is not the value of the constant G but our ability to show that it is, in fact, constant.」[一手逐字]——从业者自己早把重点从「值」挪到「恒」。
  • 2014-10-08,Gibney,Nature 514, 150(Wayback 存档):「humanity still doesn’t have an accurate value for the gravitational constant. And, bafflingly, scientists’ ability to pinpoint G seems to be getting worse.」、「known for sure to only 3 significant figures」[一手逐字]——「越测越不准」话术的媒体化石(其真实对应物=第六章的两次汇编事件)。
  • 2016,PNAS 特写(PMC5018785):「More than 200 years and 350 experiments later, scientists know the gravitational constant to a precision only about 1,000-times better than could be calculated from Cavendish’s data.」;Tino 逐字:「For any other experiment in physics, this would be a real surprise or shame.」[一手逐字]
  • 2018-08-29,Schlamminger 的 Nature News & Views(Wayback 快照):「the gravitational constant, G … is one of the most poorly defined physical constants.」[一手逐字]
  • 2026-04-16,NIST 官方新闻:「Scientists have been trying to measure big G for over 225 years, but the exact value has eluded them.」、「its value remains the least well-known of the four fundamental forces in nature」(按原文照录——该句把「常数」与「力」混用,G 不是四种基本力之一;如实登记为通稿层话术)[一手逐字];配套视频标题《The ‘Black Sheep’ of the Fundamental Constants: Measuring Big G》。

10.4 Wikipedia 层:叙事直接挂一手,倍数写对了

《Gravitational constant》条「History of measurement」节(2026-08-16 检索,wikitext API)逐字:「Published values of G derived from high-precision measurements since the 1950s have remained compatible with Heyl (1930), but within the relative uncertainty of about 0.1% (or 1000 ppm) have varied rather broadly, and it is not entirely clear whether the uncertainty has been reduced at all since the 1942 measurement. Some measurements published in the 1980s to 2000s were, in fact, mutually exclusive.」、「But the continued publication of conflicting measurements led NIST to considerably increase the standard uncertainty in the 1998 recommended value, by a factor of 12, to a standard uncertainty of 0.15%…」(词条[二手导航级])。其承重叙事句的引用直接挂 Gillies 1997 综述与 CODATA 2002 平差报告(引文附注逐字:「Section Q (pp. 42–47) describes the mutually inconsistent measurement experiments from which the CODATA value for G was derived.」)——Wikipedia 的分歧叙事挂在一手文献上,不是转引媒体;1998 年的倍数它写对了(12,不是 17)。另注意其「Constancy」节把公众的「恒定」叙事锚在超新星上:「Analysis of observations of 580 type Ia supernovae shows that the gravitational constant has varied by less than one part in ten billion per year over the last nine billion years.」(引 Mould & Uddin 2014)[二手导航级]。

10.5 中文层:「最差常数」转写+三十年叙事框,但分歧信息没丢

  • 新华网 2018-08-31《我国科学家耗时30年测出目前最精准引力常数G》(链接):「华中科技大学引力中心罗俊院士团队历经30年艰辛工作,测出目前国际上最精准的万有引力常数G值」;杨山清受访逐字:「引力常数G是计算物体间万有引力的关键,但是目前我们并不知道G的精确值是多少,这使得很多与之相关的基础科学难题至今无法解决。」[一手逐字]
  • 新华网 2018-09-01《中国科学家测出最高精度万有引力常数》(链接):「万有引力常数G是人类认识的第一个基本常数,但是,G值的测量精度是目前所有基本常数中最差的。」、「给出了目前国际上最高精度的G值,相对不确定度优于12ppm」[一手逐字]
  • 新华网 2018-10-31 全媒头条《在中国一所大学的山洞里,有这样一群”追引力的科学家”》(链接):「作为人类最早认识的物理学基本常数,G值的测量精度迄今为止仍是最差的一个。」[一手逐字]
  • 科学网 2019-01-03《2018年中国十大科技进展新闻》头条条目(链接):「以往G值测量的相对精度虽然接近10-5,相互之间的吻合程度仅达到10-4水平。」[一手逐字]——分歧以「吻合程度仅 10⁻⁴」的形式保留在通稿里,没有丢失。
  • 人民网(浙江频道)2020-05-27《从引力到引力波,36年专注一个问题》(链接):「在西方科学家主导了300多年的领域里,中国科学家测出了世界最精确的万有引力常数G的值」[一手逐字]

中文层话术特征:英文「least precisely known」被转写为「测量精度最差的/最难测定的」,并叠加「30 年/山洞/赶超/话语权」叙事框——名号被保留、被放大,但它的「折痕」含义没有被转写

10.6 名号称重与消费层小结

「最测不准的基础常数」——真,且是数量级式的真:2019 SI 修订把 h、e、k、N_A、c 等钉成定义值(NIST 全表逐字行「Planck constant 6.626 070 15 e-34 (exact)」等,allascii.txt[一手逐字])之后,仍须实测的通用常数排位为 G(2.2×10⁻⁵)≫ G_F(5.1×10⁻⁷)≫ m_μ(2.2×10⁻⁸)≫ m_e、m_p(3.1×10⁻¹⁰)> α⁻¹(1.6×10⁻¹⁰)≫ R_∞(1.1×10⁻¹²)(全部 NIST 各值页逐字[一手逐字]):G 比次差的 G_F 差约 43 倍、比电子质量差约 7 万倍、比精细结构常数差约 14 万倍(本篇自算倍数)。

但同一张账显示名号的真实含义:不是「无知」,而是「官方把十六个互相打脸的精密测量如实折进 22 ppm」。「越测越不准」的两次真实事件(1998、2010)都是汇编事件;「200 年 350 次实验只比卡文迪许好 1000 倍」(PNAS 2016)对照的是一条 21 倍改善的官方不确定度曲线和一条 165 ppm 的推荐值漂移。名号是诚实的折痕,不是能力的空洞——消费层把折痕读成了空洞。

十一、反向红跳账:四句流行的「虚无读法」逐句称重

对称双向的另一侧:拆完危机话术,还要挡住虚无口号。四句流行的反向读法,逐句裁决。

11.1 「G 完全未知/人类根本不知道引力多强」——不成立(错在数量级)

官方现行值逐字(NIST 值页):「Numerical value 6.674 30 x 10⁻¹¹ m³ kg⁻¹ s⁻² / Standard uncertainty 0.000 15 x 10⁻¹¹ … / Relative standard uncertainty 2.2 x 10⁻⁵ … Source: 2022 CODATA recommended values」[一手逐字]。全部现代测量挤在 ±0.036% 的带内(5.3);最差单项自称 150 ppm、最好者自称 12 ppm。这句读法把「22 ppm 的官方不确定度」偷换成「不知道」——G 的问题从来不是「不知道」,而是「知道到 22 ppm 时,16 个自称 12–150 ppm 的测量互相打脸」。这是过度声称的分歧,不是无知的深渊。

11.2 「G 在振荡/随时间漂移」——不成立(振荡为统计幻影,漂移被封死)

振荡主张 2015 年发表当年即遭同刊贝叶斯否决(e³⁰ 因子),频率主义复验同结论,原文作者自己限定「不是 G 在变」(第九章全部逐字在案);漂移被两条独立天体物理通道约束在 10⁻¹³–10⁻¹² /年量级且与零相容(LLR (4 ± 9) × 10⁻¹³ yr⁻¹;脉冲星 (−0.7 ± 3.3) 与 (−0.1 ± 0.9) × 10⁻¹² yr⁻¹)[一手逐字]。G 测量值的分散是测量史现象,不是常数的时间变化。

11.3 「这是计量学的失败」——不成立(失败的是这个量,不是这个学科)

反例一:同一批国家计量院体系把小 g 比对到 10⁻⁹ 量级国际一致——CCM.G-K2 绝对重力仪国际关键比对(2013,Walferdange 地下实验室)「In total, 25 absolute gravimeters were compared including 7 different types of instruments.」、「Overall, the measurements are all consistent given the declared uncertainties.」(BIPM 官方最终报告 PDF,Wayback[一手逐字];25 台仪器、19 国[一手逐字])。反例二:Kibble 天平把普朗克常数 h 测到 10⁻⁸ 量级、2019 年起 h 成为定义值(跨篇一行,取证见计量学篇)。

G 卡壳的原因写在 G 综述自己的句子里:信号只有 g 的 10⁻⁷、引力不可屏蔽、没有可传递的标准器(第四章)。而面对卡壳,制度的响应是成建制的:IUPAP 工作组、CIPM 决议、盲测复测、暗不确定度统计、扩张因子会计(守真锚四)。失败的是这个量的单次绝对测量,不是这门学科。

11.4 「分歧说明引力定律错了」——不成立(差 9–12 个数量级)

量级账(本篇自算):分歧带全宽 5.5×10⁻⁴,典型两值对立(BIPM-14 vs JILA-18)4.4×10⁻⁴;等效原理检验零结果精度 10⁻¹³(Eöt-Wash 扭秤、LLR)至 10⁻¹⁵(MICROSCOPE);平方反比在 52 μm–3.0 mm 区间与牛顿引力完美相符(Lee et al. 2020[一手逐字]);Ḡ/G ≲ 10⁻¹³ /年。分歧量级 ÷ 定律检验精度 ≈ 10⁹·⁷–10¹¹·⁶——差 9 到 12 个数量级。如果平方反比或等效原理在 10⁻⁴ 量级上出错,那些 10⁻¹³–10⁻¹⁵ 的检验早已爆表;它们全部与零相容。NIST-26 论文把这句话写在自己动机节:「The first hypothesis, unexplained physics, remains remote.」[一手逐字]分歧是测量账,不是定律账。

11.5 反向红跳小结

四句虚无读法共用同一个偷换:把「汇编的不确定度」或「测量间的张力」读成「自然界/学科层面的溃败」。每一条的证伪都不需要新实验——CODATA 自己的表、定律检验自己的摘要、重力仪比对自己的报告,逐字都在案。

十二、镜像家族账:换尺三连——g-2、惰性中微子、G

本篇是「换尺家族」的第三件。三篇共用同一个动作:尺动了,不是世界动了

  • g-2 篇全库第 184 篇):换的是理论预言汇编这把尺——同一个实验值 116592070.5×10⁻¹¹,对 WP20 尺是 4.2σ、对 WP25 尺(格点 QCD)是 0.6σ;Science 定调逐字「The discrepancy vanished not because experimenters previously erred, but because the theoretical estimate changed」。
  • 惰性中微子篇全库第 202 篇):换的是反应堆通量分母这把尺——反应堆数据一次没动,Huber–Müller→Kopeikin/EF→CEA 三次分母换尺让异常从 0.943(0.023)「生」、压到 1.1–1.2σ「死」、再回到 2.2σ「复活」。
  • 本篇(G):换的是汇编输入集+扩张因子这把尺——16 个测量值自 2018 年起一个没动过(CODATA 2022 逐字「there is no new relevant input datum」、「the 2022 and 2018 recommended values of G are therefore identical」[一手逐字]),动的是抚平折痕的系数(14→6.3→3.9)与样本集成员(PTB 进表出表、JILA-10 被 JILA-18 取代、NIST-82 修正后回归);官方 22 ppm 是扩张后的账,扩张前的加权均值原始不确定度约 5.8 ppm(本篇自算)。

家族关系一句:g-2 的 σ 随理论分母变,中微子的异常随通量分母变,G 的官方不确定度随扩张因子变——σ 数与不确定度都是汇编件的读数;把它们读成「自然界的性质变了」是三篇共审的同一个越界动作。与计量学篇(全库第 175 篇)合看则是一副对偶:那篇审「定义换尺」(2019 SI 把七个常数钉成精确值),本篇审「汇编换尺」(G 是同一体系里唯一仍以 ppm 级实测的通用常数)——同一本 CODATA 账的正面与背面。

十三、尾账:截至 2026-08-16 的活口与观察点

13.1 在研测量与制度现状(WG13 2025 年报逐字骨架)

  • IUPUI–Cal Poly Humboldt(Decca & Hoyle,AAF+ToS 双模式,单晶硅源质量约 1 m 直径,钛摆球 18 μm 钨丝):「first results expected in 2025」、「target uncertainty at the 10 ppm level」[一手逐字]——截至本篇取证日未见刊出值。
  • HUST 下一代:丝镀锗/铍抑制静电噪声,「The goal is to push the uncertainty of G to below 20 ppm」[一手逐字]。
  • 佛罗伦萨(Rosi 组):冷原子喷泉+5 t 钢源质量,目标 10⁻⁵ 级并评估 10⁻⁶ 可行性[一手逐字]。
  • ETH Zürich(Dual 组):旋转棒动态引力激发结构共振,「The current uncertainty in G is approximately 0.5%, primarily limited by calibration errors in laser interferometers and lock-in amplifiers.」[一手逐字]
  • 维也纳(Aspelmeyer):毫克级小质量方向延续(Westphal 2021, Nature 591, 225 之后)[一手逐字]。
  • IUPAP WG13:Rothleitner(PTB)接任主席;Living Reviews 综述(Schlamminger & Rothleitner)「should be finished by the first half of 2026」——截至取证日未见刊出;WG13 会议「planned for 2026. The site has to be defined yet.」[一手逐字]

13.2 统计侧线与落选者

  • Muravyov et al. 2025(RusAutoCon 会议论文,DOI 10.1109/RusAutoCon65989.2025.11177218)提出 preferential median,WG13-2025 逐字转引其结果:「Muravyov et al. derive a value of G = 6.674486(48) × 10⁻¹¹ m³ kg⁻¹ s⁻² which shows a much smaller uncertainty than previous CODATA adjustments.」[一手逐字]——会议论文、未经 CODATA 体系采纳,不确定度声明比官方小约 3 倍,登记为侧线。
  • Rinaldi et al. 2022(arXiv:2209.07416,EPJC 83, 891 (2023))重分析:「we find that the uncertainty on this fundamental constant, once systematics are included, is significantly larger than what quoted in CODATA 2018. … we recommend a consensus value for the gravitational constant G = 6.6740 +0.0015 −0.0015 × 10⁻¹¹ m³ kg⁻¹ s⁻².」[摘要级一手]——独立重分析给出的不确定度是 CODATA 的 10 倍。
  • Brack et al. 2022(Nature Physics 18, 952,谐振梁动力学新方法):CODATA 2022 判语逐字「their combined relative standard uncertainty of 1.7×10⁻² is not competitive.」[一手逐字]
  • Lamine et al. 2025(A&A 697, A109,宇宙学通道):WG13-2025 逐字「They find that G is consistent with the standard value within ~2% uncertainty.」[一手逐字]

13.3 什么事件会改变本篇的裁决(观察点)

  1. CODATA 2026 平差是否纳入 NIST-26,以及扩张因子走向:若因子显著小于 3.9,「分歧在收敛」才第一次获得账本级证据。
  2. WG13 的 Living Reviews 综述刊出后的官方口径。
  3. IUPUI–Cal Poly Humboldt 的 10 ppm 级新值落在高簇、低簇还是中段——这将直接检验「装置/团队相关系统误差」假说。
  4. 若 NIST-26 被独立确认且 BIPM 高值被解释为装置相关,则「收敛跳」将获得第一个真实样本——届时本篇第八章的裁决需要改写;在此之前,「分歧已解决」一律按话术处理。

十四、对称金句与三向红线

本篇的结论结构是:「十六个各自诚实、各自校准过的精密测量,占满一条 551 ppm 的带(真)——被读成『常数在动』(张力跳,证伪在第九章)、『人类越测越不准』(换尺跳,原形在第六章)、『分歧已解决』(收敛跳,反例在第八章)」。这个结构本身的承重点在三处:CODATA 十届平差报告全文(逐字引用最多)、十六个输入实验的一手文献(含全部勘误与修正链)、以及 NIST-26 复测论文与 Anderson 案交锋链。凡这三处之外的承重(媒体叙事、Wikipedia、中文通稿、会议论文),本篇一律降格标注。本篇最强的限定词与最强的事实来自同一批文件——这是「对称」二字在本篇的具体含义:拆危机叙事,用的是官方汇编自己的账本;挡虚无口号,用的还是同一本账。

三向红线,存档备查:

  1. 不向危机侧越界:「G 在振荡/漂移」被 e³⁰ 因子的贝叶斯否决、频率主义复验与 10⁻¹³ /年量级的天体物理约束三面封死;「分歧→新物理」的切割在 Anderson 原文自己笔下完成,在 CODATA 历届归因中一致(2018 逐字归因为「the discovery of previously unknown systematic effects in the measurement methods」)。
  2. 不向虚无侧越界:「完全未知」「计量学失败」「定律错了」三句全不立——22 ppm、10⁻⁹ 的 g 比对一致性、10⁻¹³–10⁻¹⁵ 的定律检验,逐字在案。
  3. 不向收敛侧越界:「分歧已解决」不立——WG13-2025「No method-dependent systematic error can be identified.」;NIST-26 把分歧刻进了同一台装置;CODATA 2022 无新输入原样冻结。

灵魂句(重印存档):常数没动,动的是汇编。22 ppm 不是测出来的,是抚出来的——3.9 倍扩张因子,是十六个互相打脸的精密测量在官方账本上留下的折痕。而这道折痕本身是诚实的:这个体系宁可把分歧印进推荐值的不确定度里,也不肯假装它不存在。

附录 A:任务书预设勘误明细(5 处)

  1. 「1998 年不确定度放大 17 倍」——证伪。 官方逐字:「its uncertainty is about a factor of 12 larger」(CODATA 1998 §III.Q);数值核对 1.3×10⁻⁴→1.5×10⁻³ ≈ 11.5 倍。Wikipedia 条目亦写 12 倍——本条以官方 12 倍为准。
  2. 「BIPM 2022–2024 G 国际比对(CCM 框架,史上首次)」——查无实证。 「CCM.G-K2」命名实测为绝对重力仪(测小 g)比对:最终报告题头逐字「Final report — INTERNATIONAL COMPARISON OF ABSOLUTE GRAVIMETERS — CCM.G-K2 Key Comparison」[一手逐字];BIPM KCDB 机读检索与 WebSearch 均无 Big-G 比对条目;Crossref 对 Metrologia 2022–2026 以「gravitational constant」全量检索无比对终报体裁论文[多源检索]。G 的制度化响应实际路线=IUPAP WG13 工作组+装置转移盲测复测(BIPM→NIST),按此改写。
  3. 「CODATA 输入 11 个(或几个)」——修正为 16 个。 2022 平差 G 输入为 Table XXX 全表 16 值(第五章);「11」是 2010 平差的输入数。
  4. MSL-03 刊物名:开题简报误作 Meas. Sci. Technol.,实为 PRL 91, 201101 (2003)
  5. 「分歧与定律检验差约八个数量级」——按实测口径修正为 9–12 个。 本篇自算:5.5×10⁻⁴ ÷ 10⁻¹³–10⁻¹⁵ ≈ 10⁹·⁷–10¹¹·⁶(11.4)。

附录 B:取不回清单(缺席登记,缺席不等于不存在)

  • Nature 2018(HUST-18)与 Nature 2014(LENS-14)正文/Methods:付费墙,仅摘要层可达(meta description 与页内摘要段已逐字录);方法细节由 HUST-14 PMC 综述与 CODATA 2018 G 节转述兜底(均一手)。
  • Kleinevoß 2002 学位论文(UWup-02 一手件):无公开在线全文;已用 CODATA-02 表列值及其参考条目逐字转引代替(「Kleinevoß, U., 2002, Bestimmung der Newtonschen Gravitationskonstanten G, Ph.D. thesis (University of Wuppertal)」)。
  • Karagioz & Izmailov 1996 俄文原刊正文(Izm. Tekh. 1996(10):3–9):无可达文本;Springer 英译页(Meas. Tech. 39(10))经 Wayback 取到摘要级(含值 6.6729±0.0005)。
  • Muravyov et al. 2025 IEEE 全文:IEEE Xplore 直连返 JS 挑战、Unpaywall 无 OA;仅 Semantic Scholar 摘要级+WG13-2025 转述数值。
  • NIST-26 论文正文行内 MathML 数值:IOP HTML 文本层丢失(如「differs relatively from the CODATA value by [丢失]」);摘要数值完整,正文两处丢失量级已由公开值自算补上(+186/−64 ppm);PDF 图像级复核未做。
  • UCI-14 预算表 Table 6「Kuroda anelasticity correction」行各丝数值:pdftotext 双栏串列不可读,图像复核未做;本篇只承重「该行存在」。
  • Boys 1894 原刊扫描件(Phil. Trans. A 186, 1–72 (1895)):未单独抓取;内容经 Wikipedia 脚注与 Falconer 2023 逐字转引双保险记录。
  • CODATA 1969/1973 两届的关键词检索:OCR 扫描层,零命中只能记「未见」、不能证「不存在」;1973 表 24.1 中 Heyl & Chrzanowski 行数值 OCR 乱码未能可靠还原(推荐值本身已由 1986/1998 论文回溯句交叉确认)。
  • Semantic Scholar 引用计数:两次 429;被引数改用 Crossref is-referenced-by-count(检索日 2026-08-16)。
  • Gillies 1997(Rep. Prog. Phys. 60, 151)全文:IOP 403;仅 INSPIRE 官方摘要,其正文是否含「least precisely known」原话未核(摘要层无此短语)。
  • Speake & Quinn 2014(Physics Today 67(7), 27)全文:付费墙;仅 Wayback 2021 快照的 standfirst+首页文字。
  • CODATA 2018 的 16 条逐条归一化残差表:正文只给定性表述与三甲(2022 版);本篇自算版可替代但不冒充官方。
  • Hofmann & Müller 2018(CQG 35, 035015)LLR 更新约束:arXiv 检索两次空返回;Ḡ/G 侧以 Williams 2004 一手为准,不影响裁决。
  • Holme & de Viron 2013(Nature 499, 202)LOD 5.9 年振荡原案的地学界争议:超出本篇范围,未追踪。
  • 南方日报 2018-09-01/文汇网 2019-02-27 两条中文标题:仅检索结果页级命中,原页未取回,不进承重清单。

附录 C:来源清单(全部为本轮实际取回并落盘件;Wayback 链接为 id_ 原件快照)

CODATA 平差报告与 NIST 值页(十届全)

十六个输入实验与其修正链(一手文献)

复测、机制与统计

Anderson 案交锋链

Ḡ/G 与引力定律检验

综述、制度与比对

测量史

消费层与中文通稿


本篇为 Chat Research 机制裁决系列第 152 篇(对称双向第 147 篇,section B,全库第 210 篇)。调研与写作:Kimi。