LHCb Reports First Single-Experiment Evidence of a B⁰ Decay That Occurs Once in 10 Billion

The LHCb experiment at CERN's Large Hadron Collider (LHC) has captured statistical evidence, for the first time in a single experiment, of an extremely rare particle decay that physicists have been searching for over more than 40 years. The process is the decay of a neutral $B^0$ meson into a muon and an antimuon, written $B^0\to {\mu }^{+}{\mu }^{-}$. The Standard Model predicts it occurs only about once in 10 billion decays, and the new measurement reached a statistical significance of 3.2σ (sigma).

The research team posted a preprint of the paper on arXiv on October 30, 2026, and CERN and LHCb formally announced the result on November 3. The paper has been submitted to the journal Physical Review X but has not yet completed peer review.

The result does not mean a new particle has been discovered. The measured decay probability agrees with the Standard Model prediction within uncertainties, so the existing theory has in effect been confirmed at a new level of precision. Still, this decay can be influenced through quantum effects by unknown particles too heavy to be produced directly, so improving measurement precision matters for searches for "new physics" beyond the Standard Model.

$B^0$ vs. $B_s^0$: Why a Once-in-10-Billion Decay Is Special

To understand the result, it helps to know the difference between the $B^0$ meson and the similarly named $B_s^0$ meson. Both are neutral particles containing a bottom (beauty) quark, but they are made of different quarks and have very different probabilities of decaying into a muon pair.

The ordinary $B^0$ meson consists of a down quark ($d$) and an anti-bottom quark ($\bar{b}$). The $B_s^0$ meson, by contrast, is a combination of a strange quark ($s$) and an anti-bottom quark.

\[ B^0=|d\bar{b}\rangle,\qquad B_s^0=|s\bar{b}\rangle \]

Both can decay into a muon–antimuon pair, but their branching fractions, which express how often the decay occurs, differ by a factor of about 30. The Standard Model predicts $(3.66\pm 0.14)\times 10^{-9}$ for $B_s^0$ and $(1.03\pm 0.05)\times 10^{-10}$ for $B^0$. These correspond to roughly one in 300 million and one in 10 billion, respectively.

The difference is tied to the Cabibbo–Kobayashi–Maskawa (CKM) matrix, which describes mixing between quark generations. The $B^0$ decay involves the matrix element $|V_{td}|$ and the $B_s^0$ decay involves $|V_{ts}|$, and the former is smaller than the latter. Because the decay probability depends on the square of these elements, the $B^0$ decay is more strongly suppressed, an effect known as "CKM suppression."

For $B_s^0\to {\mu }^{+}{\mu }^{-}$, a joint LHCb–CMS analysis in 2015 had already reached a significance above 5σ, establishing its observation. For $B^0\to {\mu }^{+}{\mu }^{-}$, however, no single experiment had previously reached the level of evidence. What makes the new result significant is not another confirmation of the known $B_s^0$ decay, but that LHCb alone has statistically captured a signal for the $B^0$ decay, which is about 30 times rarer.

Results from 13.6 TeV Collision Data

The core of the analysis is proton–proton collision data collected in 2024–2025 during Run 3 of the LHC. The collision center-of-mass energy is 13.6 TeV and the integrated luminosity reaches 18 fb${}^{-1}$. Integrated luminosity is a measure of the amount of data obtained from particle collisions.

The team analyzed the Run 3 data on its own and also performed a combined analysis with results from Run 1 (2011–2012, 7–8 TeV, 3 fb${}^{-1}$) and Run 2 (2015–2018, 13 TeV, 6 fb${}^{-1}$). It is important to distinguish the two: the values measured from Run 3 alone differ from the final Run 1–3 combined values.

The branching fraction of $B^0\to {\mu }^{+}{\mu }^{-}$ from the Run 1–3 combined analysis is as follows.

\[ \mathcal{B}(B^0\to\mu^+\mu^-)= \left(1.14^{+0.42+0.06}_{-0.39-0.05}\right)\times10^{-10} \]

The first superscript/subscript value is the statistical uncertainty and the second is the systematic uncertainty. The central value is $1.14\times 10^{-10}$, consistent with the Standard Model prediction of $(1.03\pm 0.05)\times 10^{-10}$.

The combined branching fraction of $B_s^0\to {\mu }^{+}{\mu }^{-}$, measured at the same time, is:

\[ \mathcal{B}(B_s^0\to\mu^+\mu^-)= \left(3.25^{+0.26+0.13}_{-0.25-0.12}\right)\times10^{-9} \]

This too agrees with the Standard Model prediction of $(3.66\pm 0.14)\times 10^{-9}$ within uncertainties. The team also measured the ratio of the two branching fractions, ${\mathcal{R}}_{\mu \mu }$. This quantity reflects, among other things, the ratio of CKM matrix elements $|V_{td}/V_{ts}{|}^2$, and offers a clue to quark interaction structures that differ from the Standard Model.

\[ \mathcal{R}_{\mu\mu}\equiv \frac{\mathcal{B}(B^0\to\mu^+\mu^-)} {\mathcal{B}(B_s^0\to\mu^+\mu^-)} =0.035^{+0.014+0.002}_{-0.012-0.001} \]

The Standard Model predicts $0.0280\pm 0.0016$, and no significant discrepancy was found for this ratio either.

The effective lifetime of $B_s^0\to {\mu }^{+}{\mu }^{-}$, derived from the decay-time distribution, was also measured. The Run 1–3 combined result is $1.60\pm 0.16\pm 0.01$ picoseconds (ps), consistent with the Standard Model prediction. The effective lifetime probes a different property from the branching fraction and could detect the influence of new physics that does not change the decay probability.

The main results, split into Run 3 alone and the Run 1–3 combined analysis, are summarized below.

Source: LHCb Collaboration paper (arXiv:2611.00230v1). Uncertainties are listed as statistical, then systematic.

{
  "type": "bar",
  "title": "B中間子の希少崩壊：標準模型と測定値の比較",
  "unit": "×10^-10",
  "categories": ["B0 → μ+μ−", "Bs0 → μ+μ−（÷20）"],
  "series": [
    {
      "name": "標準模型の予測値",
      "values": [1.03, 1.83]
    },
    {
      "name": "LHCb結合測定値",
      "values": [1.14, 1.625]
    }
  ],
  "caption": "Run 1〜3結合解析の中心値を比較。Bs0の分岐比は表示上20分の1に縮小している。棒グラフには誤差範囲を表示していないが、両崩壊の測定値は標準模型の予測と不確かさの範囲内で整合する。",
  "source": "LHCb collaboration, arXiv:2611.00230v1"
}

In the new result, the $B^0$ measurement still carries a relatively large statistical uncertainty. Meanwhile, the $B_s^0$ branching fraction and effective lifetime were measured with the highest precision of any single experiment. Neither shows a significant deviation from the Standard Model, which strengthens the constraints on theoretical models that assume unknown interactions.

A Detector 5 Millimeters from the Beam, and Machine Learning

Detecting a decay that occurs only about once in 10 billion requires picking out the target signal accurately from a huge volume of proton collision data. Proton collisions at the LHC produce many particles, so the detection of two muons alone is not enough to conclude that a $B^0$ meson decayed. Unrelated particle combinations and muon pairs arising from other decay processes must be removed as background.

One factor that made the measurement possible is "Upgrade I," a major overhaul of the LHCb detector. According to the University of Manchester, the upgrade lets the experiment collect collision data at up to five times the rate of the previous apparatus. The detectors that measure particle tracks and the particle identification systems were also renewed, enabling precise measurements even at the higher collision rate. However, the fivefold increase in data-taking capacity does not mean that every measurement became five times more precise.

A particularly important role was played by the silicon pixel detector VELO (Vertex Locator). VELO sits just 5 millimeters from the proton beam and records particle tracks with high spatial resolution. $B$ mesons live only about 1.5 picoseconds from production to decay, but because they travel at high speed at the LHC, they can decay several millimeters from where they were produced.

VELO is used to distinguish the primary vertex, where the protons collided, from the secondary vertex, where the $B$ meson decayed. The team combined muon tracks, decay vertices, and flight directions to select candidate events originating from $B$ meson decays. In addition, the upgraded Ring Imaging Cherenkov detectors (RICH) and muon detection system improved the ability to remove events in which other particles are misidentified as muons.

In the software analysis, boosted decision trees (BDTs), a type of machine learning, played an important role. The team suppressed background events using, among other things, the geometric relationships between particle tracks and decay vertices, and classified candidates according to the BDT output. It then fit a statistical model to the invariant-mass distribution of the two muons to estimate the numbers of $B^0$ and $B_s^0$ signal events.

Background events cannot be eliminated completely, however. According to the paper, the largest systematic uncertainty in the Run 3 $B^0$ branching fraction measurement came from estimating backgrounds from hadronic and semileptonic decays, with an effect of about 4% of the branching fraction. For example, events in which $\pi$ or $K$ mesons are misidentified as muons, and backgrounds from decays involving neutrinos, must be modeled properly.

For the $B_s^0$ branching fraction, the largest systematic uncertainty came from the ratio $f_s/f_d$ of the probabilities that a bottom quark hadronizes into a $B_s^0$ versus a $B^0$, with an effect of about 3%. The result was achieved by combining higher-precision detectors, better particle identification, machine-learning background suppression, and an improved statistical model.

3.2σ Is "Evidence," Not "Discovery"

In particle physics, statistical significance is expressed in σ (standard deviations) to assess the likelihood that an observed signal arose from a chance statistical fluctuation. By convention, exceeding 3σ is called "evidence," and reaching 5σ is called "discovery." At 5σ, the probability of seeing a deviation at least as large as observed, assuming background only, is about one in 3.5 million in a one-sided test.

In the analysis using only the 2024–2025 Run 3 data, the $B^0\to {\mu }^{+}{\mu }^{-}$ signal showed a significance of 3.0σ. Adding the earlier Run 1 and Run 2 data raised it to 3.2σ. The value was calculated by comparing the likelihoods of a signal-plus-background model and a background-only model based on Wilks' theorem, and was also checked with pseudo-experiments.

Even so, the result cannot be called a "discovery" of the decay. The distinction matters because 3σ evidence for the same decay was reported before, and its significance later declined with additional data.

In 2015, LHCb and CMS published a joint analysis of Run 1 data in Nature, reporting 3.0σ evidence for $B^0\to {\mu }^{+}{\mu }^{-}$. But analyses using data through Run 2 did not maintain that significance. There remained a possibility that a statistical fluctuation had made the early signal look stronger than it really was.

The new result is described as the "first evidence in a single experiment" because it exceeded 3σ using LHCb's own data alone, without combining results from separate experiments such as LHCb and CMS. In light of past experience, what matters now is whether the significance holds up as more data accumulate. The 3.2σ result is major progress in a search spanning more than 40 years, but final confirmation requires further measurements.

Agreement with the Standard Model Narrows the Room for New Physics

The importance of the result lies not merely in capturing a rare decay. $B^0\to {\mu }^{+}{\mu }^{-}$ has long drawn attention as a process in which the effects of new physics beyond the Standard Model could show up.

The decay is classified as a "flavor-changing neutral current" (FCNC) interaction. In the Standard Model, neutral-current interactions that change quark type do not occur at the simplest, tree level. The decay therefore requires higher-order quantum processes involving $W$ bosons, top quarks, and other particles, which are represented by Feynman diagrams known as "box diagrams" and "penguin diagrams."

Unknown particles too heavy to be produced directly at the LHC could influence these higher-order quantum processes indirectly. For example, if additional Higgs bosons, $Z'$ bosons, or leptoquarks beyond the Standard Model exist, there are theoretical models in which they would alter the decay probability or lifetime. The paper explains that, depending on conditions, such decays can be used to probe the effects of new physics at energy scales up to around 100 TeV.

But the measured branching fractions agreed with the Standard Model predictions. This does not prove that unknown particles do not exist, but it does limit the parameter regions of theoretical models that would have effects large enough to conflict with the measurement. As measurement precision improves, the range allowed for unknown interactions can be narrowed further.

The announcement also used the term "Zeptoverse." It is a conceptual name for the unexplored physics associated with the extremely small spatial scale of one zeptometer ($10^{-21}$ meters). However, the experiment did not directly observe space at the zeptometer scale, nor were any unknown particles actually found there. The term is used in the sense of precisely measuring the decay of known particles to indirectly probe unknown physics at extremely high energy scales.

LHCb plans to accumulate considerably more data through the early 2030s. According to the University of Manchester, there is also a concept for a future additional upgrade that would raise the data-taking rate by roughly another factor of 10. This is a future plan, however, and does not guarantee when the significance of the current result might reach 5σ.

Capturing a decay that occurs about once in 10 billion in a single experiment is an important milestone for particle physics. No sign of new physics has been found so far, but a new means of testing Standard Model predictions more rigorously has been gained. The next questions are whether future data will turn this evidence into a firm discovery, and whether improved measurement precision will reveal a new discrepancy with the Standard Model.