One gluon at a time: the X(2370) letter, page by page

Abstract & setup — the claim, the theory, the résumé

• The abstract's result: searched X(2370) → K*(892)⁰K̄⁰ + c.c. via J/ψ → γK_S⁰K_S⁰π⁰ with 10 billion J/ψ events. No evidence; B[J/ψ→γX]·B[X→K*K̄→K_S⁰K_S⁰π⁰] < 2.7×10⁻⁶ (90% C.L.). Suppression ⇒ flavor-singlet ⇒ first flavor-singlet light hadron observed above 1 GeV/c². Then the thesis: mass, 0⁻⁺ spin-parity, high J/ψ radiative rate, η_c-like decay pattern, flavor-singlet nature, narrow width and suppressed γω/γφ decays all match the lightest 0⁻⁺ glueball, while other interpretations are disfavored.
• Why glueballs matter: QCD is non-Abelian, so gluons self-interact and can bind into glueballs — finding one is a direct test of the non-Abelian strong force. Lattice QCD puts the 0⁻⁺ ground state at 2.3–3.0 GeV/c².
• The production trick: J/ψ → γ + glueball is the gluon-rich hunting ground (LQCD rate (2.31±0.90)×10⁻⁴), and a tiny 2°–5° mixing with c c̄ — also predicted by LQCD — lifts the observable rate to (2–20)×10⁻³ while the mixture keeps the glueball's properties.
• X(2370)'s rap sheet: first seen 2011 in J/ψ → γπ⁺π⁻η′ (>6.4σ, 225M J/ψ): mass 2376.3±8.7(+3.2/−4.3) MeV/c², width 83±17(stat) +44/−6(syst) MeV; confirmed at 8.3σ; 0⁻⁺ established >9.8σ with the 10-billion-event sample.

The test — why K*K̄ is the cleanest knife, and the 0.1σ nothing

• Why this mode: comparing branching fractions across channels is muddy (huge systematics). But generalized G-parity conservation forbids a 0⁻⁺ flavor-singlet from decaying to K*(892)⁰K̄⁰ — so suppression of this single mode is an unambiguous flavor-singlet signature. Crucial test, run.
• The sample & selection: (10087±44)×10⁶ J/ψ events; event criteria reused from the companion analysis (ref. [28]); backgrounds estimated negligible.
• Fig. 1: K_S⁰K_S⁰π⁰ spectrum shows X(2370) near 2.3 GeV plus the η_c peak; K_S⁰π⁰ shows the K*(892)⁰; after requiring a K* combination, clear peaks at 1.4 and 1.9 GeV/c² — and no X(2370) bump.
• The fit: unbinned maximum-likelihood, efficiency-corrected Breit–Wigner × Gaussian resolution, third-order polynomial background; X(2370) fixed at 2359 MeV/c², Γ = 170 MeV (combined values from ref. [28]).
• The nothing, quantified: statistical significance 0.1σ; B_K*K̄ = (0.1±1.2(stat)±1.1(syst))×10⁻⁶; ratio R = 0.003±0.040±0.026; upper limits B_K*K̄ < 2.7×10⁻⁶ and R < 0.081 at 90% C.L.

A 0.1σ fit is a jar that fails the sniff test — you lean in expecting off-gassing and get vacuum. In physics, a clean null this sharp is itself a measurement.

Punchline of the page: “The X(2370) is the first flavor-singlet light hadron observed above 1 GeV/c², and it is consistent with the glueball interpretation.”

The silence, weighed — what a <2 MeV partial width kills

• Fig. 2 shows the fit in the K* region; Table I collects the measured product branching fractions (×10⁻⁴): K K̄π 3.25±0.25(+0.73/−0.75), ππη 3.2±0.1(+0.9/−1.0), ππη′ 1.94±0.04(+0.33/−0.88), K K̄η′ 0.39±0.05±0.10 — i.e. X(2370) is produced plentifully, with B[J/ψ→γX] > 1×10⁻³.
• Division: big production, absent K*K̄ ⇒ B[X(2370)→K*(892)K̄] < 1.6% ⇒ partial width < 2 MeV.
• η-η′ molecule / excitation — dead: that interpretation expects 15–200 MeV in K*K̄; a naive extrapolation from the η(1405)/η(1475) (for which K*K̄ is a major mode) alone gives ≥10 MeV, and 30× larger phase space near 2.37 GeV pushes it above 20 MeV. Measured: <2 MeV.
• Ordinary q q̄ — wobbles: LQCD predicts no flavor-singlet uū+dd̄/ss̄ meson anywhere near 2 GeV, and no other flavor-singlet light meson above 1 GeV has ever been observed.
• Setup for the next page: the strong γω/γφ suppression says both the ss̄ and the uū+dd̄ contents inside X(2370) must be very small.

The verdict, the near-miss hypotheses, and the to-do list

• Table II (J/ψ→γR→γγω / γγφ, ×10⁻⁶; caption verbatim: "BFs or their upper limits at 90% C.L." — a mixed table): η(1405) 3.54 ± 0.20 (+0.65/−0.24) / 3.57 ± 0.18 (+0.59/−0.61) (measured); X(1835) <0.11 / 3.37; X(2370) <0.04 / <0.11 (upper limits) — roughly 30–90× below η(1405) on central values, widening to ~25–110× if the η(1405) uncertainties are folded in (my arithmetic, not the paper’s), consistent with theoretical expectations for a pseudoscalar glueball. One more angle, sharpened in review (my gloss, not the paper’s words): X(1835) is measured at 3.37 in γφ — right where X(2370) sits below 0.11 — so the suppression tracks X(2370)’s couplings specifically (structural: flavor-singlet), not a generic radiative or mass-scaling effect. Two provenance notes from the same source-check (my read of the table’s citation structure): the 3.37 benchmark carries ± 0.19 (+0.78/−1.10) — its spread is almost entirely systematic; and the two columns are cited to earlier analyses (ω column → ref. [24], a MESON2026 conference talk on light meson spectroscopy at BESIII; φ column → ref. [3], Ablikim et al., Phys. Rev. D 111, 052011, a partial-wave analysis of J/ψ→γγφ), so the η(1405)/X(1835) benchmarks are quoted from prior work while the X(2370) limits are this sample’s own. And X(1835)’s own γω entry is a limit too — which is why the channel-matched γφ contrast is the clean one.
• Narrow widths as a fingerprint: flavor symmetry spreads the rate over S-wave quasi-two-body modes (f₀(980)η′, a₀(980)π⁰, K₀*(1430)K̄, f₀(1500)η) at only ~1–10% each ⇒ partial widths of a few MeV, vs ~100 MeV for a typical OZI-allowed decay. Normal q q̄, multiquark or hybrid states all start with light quarks and get OZI-allowed exits; a glueball doesn't — OZI bites at only one vertex. A naive OZI estimate lands exactly at “a few MeV”.
• ΣΣ̄ baryonium — skewered: J/ψ radiative decays to baryon pairs run ~10⁻⁴ or less, no threshold distortion shows on the line shape, and a baryonium wouldn't be a flavor-singlet anyway. (Also noted: some phenomenology predicted a K K̄π rate 50× above K K̄η′ — inconsistent with BESIII's measured BFs — though its high-production conclusion still favors the glueball read.)
• Still to do: partial-wave analyses including interference with nearby X(2120), X(2260), X(2600) and threshold effects; new searches in ωω, φφ (flavor-symmetric for a singlet), ωφ (OZI-suppressed) and K*(1410)K̄ (G-parity-forbidden) to nail the fractions down.

The letter's closing claim: “A dominant 0⁻⁺ glueball component is essential for a natural and complete explanation of the properties of the X(2370) … This supports that the lightest 0⁻⁺ glueball is the dominant constituent of the X(2370).”

Acknowledgments open — with thanks to theorists K.T. Chao, Y. Chen, F.-K. Guo, T. Huang, X. Liu, J.-P. Ma, C.-F. Qiao, J.-J. Wu, Y.-B. Yang and Q. Zhao.

Receipts — funding and the 39-reference shelf

• The grant ledger closes out: National Key R&D Program and NSFC (China), CAS programs, ANID (Chile), INFN (Italy), Knut and Alice Wallenberg Foundation (Sweden), DPT (Turkey), NRF (Korea), the Mongolia S&T fund, Polish NCN, STFC (UK) and U.S. DOE.
• References worth spotting: [1–6] the lattice glueball masses; [7–9] why J/ψ radiative decays are the place to hunt; [10–11] the c c̄-mixing production boost; [16] Robson's OZI treatment of glueball decays (1977); [20] the 2011 discovery (PRL 106, 072002); [26–28] BESIII's earlier X(2370) results, including the 0⁻⁺ determination (PRL 132, 181901) and the companion analysis arXiv:2605.26495; [31–32] Lipkin's generalized G-parity; [36] the PDG. The author list starts at the bottom.

Every ferment has a recipe card. This one's card is 39 papers deep, from 1977 OZI folklore to 2026 lattice arithmetic.

The roll call — authors and 91 institutions

• Pages 6–7: the author list runs home alphabetically (Cossio … Zu) — several hundred names in all.
• Pages 7–9: 91 numbered affiliations, from the Institute of High Energy Physics in Beijing outward: universities and institutes across China, Germany, Russia, the USA, Italy, Korea, Turkey, the UK, Poland, Sweden, Chile, Pakistan, Mongolia, Thailand and India.
• Page 10: the last “also at” footnotes (EPFL, Helmholtz Institute Mainz, Hangzhou UCAS, Chengdu University of Technology).
• Small human details: a dagger (†) marks deceased colleagues, and each superscript maps a name to institutes — a reminder that a “BESIII result” is a whole ecosystem, pickled together.