The Best Evidence Yet for a Glueball Is a Decay That Never Happens
On 5 August 2026, at a plenary session of the International Conference on High Energy Physics in Brazil, the BESIII collaboration told the field that the lightest pseudoscalar glueball dominates the X(2370). The measurement carrying that argument found nothing at all. BESIII searched for the X(2370) decaying into a K*(892) and an anti-kaon, saw no signal, and set a limit of 2.7 in ten million on the combined branching fraction. The absence is the evidence. Here is why a missing decay counts, what fifteen years of measurement actually established, and which two questions the announcement did not close.
Cite this article
Quantum Nature Science Desk (2026). The Best Evidence Yet for a Glueball Is a Decay That Never Happens. Quantum Nature. https://quantum-nature.com/x2370-glueball-evidence-2/
What a glueball is
Every particle you have heard of contains quarks. Protons hold three, pions hold a quark and an antiquark. A glueball holds none. It is a particle built entirely out of the force that binds other particles together.
The reason this is possible. Photons carry the electromagnetic force but hold no electric charge, so two photons pass through each other without interacting. Gluons carry the strong force and do hold the corresponding charge, called colour. A gluon therefore pulls on other gluons. That single difference, the non-Abelian structure of quantum chromodynamics, means the force field can bind to itself with no matter present.
QCD has predicted these objects since the 1970s. Nobody has confirmed one. That makes the glueball a fifty-year test of whether the theory’s most distinctive structural feature produces the objects it says it should.
Why it is hard
A glueball carries no quantum number that a conventional meson cannot also carry. No label settles it. Identification has to come from a pattern of properties, none of which is decisive alone.
Only lattice QCD can say where they sit
The strong coupling grows as energies fall, so the usual method of expanding in a small parameter fails exactly where hadrons live. Lattice QCD replaces continuous spacetime with a four-dimensional grid and evaluates the theory numerically. It is the only first-principles route to a glueball mass.
What the lattice predicts. The quenched approximation leaves out the effects of dynamical quarks. Within it, the lightest two-gluon glueballs come out at roughly 1.5 to 1.7 GeV for spin-parity 0++, 2.3 to 2.4 GeV for 2++, and 2.3 to 2.6 GeV for 0-+. Across several independent calculations the pseudoscalar spans about 2.1 to 2.7 GeV.
The most recent dedicated calculation, by Gui and colleagues in 2019, put the pseudoscalar glueball at 2395 ± 14 MeV. A three-gluon version of the same quantum numbers would land near 3.4 to 3.6 GeV, far heavier than anything in this discussion.
What BESIII measured, in sequence
BESIII built the case over fifteen years, one property at a time.
- 2011. A new state appears in J/ψ decaying to a photon plus two pions and an eta prime. BESIII names it X(2370).
- 2020. A second channel confirms it, this time a photon plus two kaons and an eta prime.
- 2024. A partial wave analysis on 10 billion J/ψ events fixes the spin-parity at 0-+, matching the lattice prediction for the pseudoscalar glueball. The mass is 2395 ± 11 (stat) +26/-94 (syst) MeV. The width is 188 +18/-17 (stat) +124/-33 (syst) MeV.
- 2025 and 2026. Further decay modes turn up, and the decay to K*(892) with an anti-kaon does not.
Note the central mass. Gui’s lattice calculation gave 2395 MeV. BESIII measures 2395 MeV. The agreement is genuine and it is also, on its own, worth very little: the systematic uncertainty runs to nearly 100 MeV downward, and several non-glueball explanations predict states in the same region.
Why a missing decay is the strongest evidence
A glueball has no valence quarks, so it has no preference between up, down and strange. It couples to all three identically. Physicists call this a flavour singlet. It is the property that most sharply separates a glueball from a quark-antiquark state, because an ordinary meson draws on specific quark flavours and inherits their asymmetries.
The test. Theory predicts strong suppression when a 0-+ flavour singlet decays to a K* and an anti-kaon. The charmonium state eta-c, whose decays also run through gluons, shows exactly this suppression. So if the X(2370) is a flavour singlet, that channel should sit close to empty.
The result. Using the full 10-billion-event sample, BESIII searched for the neutral K*(892) mode through J/ψ decaying to a photon and two neutral kaons and a neutral pion. No signal appeared. The product branching fraction is below 2.7 × 10-6 at 90 percent confidence. BESIII reports this as the first flavour-singlet light hadron ever observed above 1 GeV.
Why J/ψ is the right hunting ground
Glueball searches keep returning to one process for a structural reason. The J/ψ is a bound charm quark and antiquark. When it decays radiatively, the pair annihilates into a photon plus a gluon-rich hadronic system. The environment is unusually clean and unusually full of glue, precisely the condition a glueball needs to appear in quantity.
The photon also tags the event, giving experimenters a handle on whatever recoiled against it. Beijing’s collider has accumulated 10 billion J/ψ decays for this purpose. That sample size is why a null result at the level of a few parts in ten million is possible at all.
Seven properties, one consistent explanation
BESIII’s argument is cumulative rather than decisive at any single point. The collaboration lists seven measured properties and argues that a dominant 0-+ glueball component is the only interpretation that accommodates all of them together.
The competing explanations each survive some tests and fail others. A fourth radial excitation of the eta or eta prime is a conventional quark-antiquark state, so it is not a flavour singlet and it should show the K* mode. A P-wave tetraquark of four strange quarks carries explicit strangeness, which conflicts with flavour blindness. Baryonium interpretations, which treat the state as a loosely bound baryon and antibaryon, struggle with the narrow partial widths.
That is a real argument. It is also an argument from exhaustion rather than from a positive signature, and its strength rests entirely on whether the list of alternatives is complete.
Figure 1
Seven measured properties, and what each one rules out
BESIII argues cumulatively. No single property below identifies a glueball, and the chips show which rival interpretations each one survives.
Mass
2395 ± 11 (stat) MeV, systematic +26 / −94. Lattice puts the pseudoscalar glueball at 2.1 to 2.7 GeV.
Spin-parity
0−+, from partial wave analysis of 10 billion J/ψ events. Matches the lattice pseudoscalar.
Flavour-singlet behaviour
K*(892) mode absent. Product branching fraction below 2.7 × 10−6 at 90 percent confidence.
Narrow partial widths
Total width 188 +18/−17 MeV, spread across many channels with none dominant.
Decay pattern resembles ηc
Appears in KK̄π, ππη, ππη′ and KK̄η′ with no single mode dominating, as expected when decays proceed through gluons.
Radiative decays to ω and φ suppressed
Another consequence of flavour blindness. An ordinary meson with definite quark content would not show this.
Production rate in J/ψ radiative decay
BESIII counts the high rate as supporting evidence. A 2021 extraction found it roughly ten times larger than lattice predicts for a glueball.
Read the last row against the first six. The property BESIII lists as evidence for a glueball is the one an independent analysis used to argue against it. Six properties exclude conventional quark-antiquark states. None of the seven separates a glueball from a flavour-singlet hybrid.
Sources: BESIII, arXiv:2607.20366 and Phys. Rev. Lett. 132, 181901 (2024); Sun et al., arXiv:2110.08827; Lu, arXiv:2608.22394.
The seven properties BESIII cites for the X(2370), with what each one excludes. All seven are consistent with the lightest pseudoscalar glueball, but six are equally consistent with a flavour-singlet hybrid, and the seventh is actively disputed. Sources: arXiv:2607.20366 · Phys. Rev. Lett. 132, 181901 (2024) · arXiv:2110.08827 · arXiv:2608.22394
Two arguments the announcement did not close
The production rate points the wrong way. BESIII cites the high production rate in J/ψ radiative decay as evidence for a glueball. In 2021, Sun, Dai, Kuang, Qin and Szczepaniak extracted that branching ratio from the invariant mass spectra and found it roughly an order of magnitude larger than the glueball production rate lattice QCD predicts. They concluded that the X(2370) is not likely to be a glueball. A 2026 analysis revisits the problem and asks whether a small charmonium admixture could reconcile the rate while leaving the state overwhelmingly gluonic. That question is open.
Flavour-singlet does not mean glueball. In August 2026, Jianlong Lu tested a flavour-singlet 0-+ radial hybrid against the same data. A hybrid contains quarks and an excited gluon field, so it is a different object from a glueball. It is also a flavour singlet, and Lu finds that it naturally suppresses the K* mode too. The null result therefore rules out conventional quark-antiquark states. It does not distinguish a glueball from a hybrid.
What is contested
Not the measurements. The mass, the spin-parity and the K* limit are experimental results with quoted uncertainties. The contested part is the inference: do the seven properties select a glueball uniquely, or a class of gluon-rich states of which the glueball is one member?
What confirmation would take
Amplitude-level analysis. The published BESIII information consists largely of inclusive three-body product branching fractions. Those do not give fit fractions, relative phases, or the covariance between components. Treating them as independent quantities discards interference information and can manufacture discriminating power that does not exist. Separating a glueball from a hybrid requires the full amplitude decomposition.
Production somewhere else. Model calculations predict that glueball, tetraquark and molecular configurations of the X(2370) would give measurably different yields, transverse momentum spectra and rapidity distributions in proton-proton collisions. Measuring the state at the LHC would test the structure in an environment with no gluon-rich bias built in.
Unquenched lattice results. The predictions above come mostly from quenched calculations that leave dynamical quarks out. In the real world a glueball with the right quantum numbers mixes with quark-antiquark states, so “pure glueball” may not be a well-posed category. Pinning down how much mixing to expect remains an unsolved lattice problem.
The partners. QCD predicts a whole spectrum, not one state. The scalar and tensor glueballs should also be identifiable. Physicists have argued over the scalar candidates f0(1500) and f0(1710) for four decades without resolution. A consistent account of the full multiplet would be far more convincing than one state.
Where it stands
Established. The X(2370) exists, in multiple decay channels, from a sample of 10 billion J/ψ events. Its spin-parity is 0-+. Its mass falls inside the lattice range for the lightest pseudoscalar glueball. Its decay to K*(892) and an anti-kaon falls below 2.7 × 10-6.
Contested. Whether the production rate in J/ψ radiative decay is compatible with a glueball at all. Whether the seven properties, taken together, exclude a flavour-singlet hybrid.
Unproven. That anyone has yet observed a glueball.
BESIII’s own phrasing is careful and worth reading literally. The claim is that a pseudoscalar glueball is the dominant constituent of the X(2370), not that the X(2370) is a glueball. In a theory where mixing is unavoidable, that distinction is not hedging. It may be the most precise statement the physics permits.
Note on sourcing
The mass, width and spin-parity are from peer-reviewed BESIII papers. The 2026 flavour-singlet result is a preprint presented at ICHEP, and the list says so. The two counterarguments are also preprints, and this article names their authors rather than passing them off as consensus. Lattice mass ranges come from quenched calculations, which omit dynamical quarks. The body text carries that caveat rather than burying it here.
References
- BESIII Collaboration, Lightest 0-+ Glueball as Dominant Constituent of X(2370) (July 2026). Preprint, not peer-reviewed; the flavour-singlet result
- BESIII Collaboration, Determination of the spin-parity of X(2370) as 0-+ from J/psi radiative decay, Phys. Rev. Lett. 132, 181901 (2024). Preprint arXiv:2312.05324 doi:10.1103/PhysRevLett.132.181901
- BESIII Collaboration, Observation of the X(2370) in J/psi radiative decay, Phys. Rev. Lett. 106, 072002 (2011). Preprint arXiv:1012.3510 doi:10.1103/PhysRevLett.106.072002
- BESIII Collaboration, Confirmation of the X(2370) in a second channel, Eur. Phys. J. C 80, 746 (2020). Preprint arXiv:1912.11253 doi:10.1140/epjc/s10052-020-8078-4
- Y. Huang, S. Jin & P. Zhang, Discovery of a glueball-like particle X(2370) at BESIII, Int. J. Mod. Phys. A 40, 2530007 (2025). Preprint arXiv:2503.13286 doi:10.1142/S0217751X25300078
- L.-C. Gui et al., Study of the pseudoscalar glueball in J/psi radiative decays, Phys. Rev. D 100, 054511 (2019). The lattice mass prediction; preprint arXiv:1906.03666 doi:10.1103/PhysRevD.100.054511
- C. J. Morningstar & M. J. Peardon, The glueball spectrum from an anisotropic lattice study, Phys. Rev. D 60, 034509 (1999). Preprint arXiv:hep-lat/9901004 doi:10.1103/PhysRevD.60.034509
- X. Sun et al., On the nature of X(2370) (2021). Preprint, not peer-reviewed; argues the production rate disfavours a glueball
- J. Lu, Testing a Flavor-Singlet Radial-Hybrid Interpretation of the X(2370) (August 2026). Preprint, not peer-reviewed; the hybrid alternative
- Z.-L. She et al., X(2370) production in e+e- and pp collisions with the PACIAE model, Phys. Rev. D 110, 054046 (2024). Preprint arXiv:2408.04130 doi:10.1103/PhysRevD.110.054046
Common questions
What is a glueball?
A particle built entirely from gluons, the carriers of the strong force, with no quarks inside. QCD predicts glueballs because gluons carry colour charge and so pull on each other, letting the force field bind to itself with no matter present.
Why is the X(2370) the candidate?
Its mass (about 2395 MeV) and spin-parity (0 minus plus) from 10 billion J/psi events match the lattice prediction for the lightest pseudoscalar glueball, and it behaves as a flavour singlet, coupling equally to all quark flavours.
How can a decay that never happened be evidence?
A flavour-singlet 0-minus-plus state should barely decay to a K*(892) and an anti-kaon. BESIII looked and found nothing, setting the branching fraction below 2.7 in ten million, which rules out ordinary quark-antiquark states that would show that decay.
Does this confirm a glueball?
No. BESIII says a pseudoscalar glueball is the dominant constituent of the X(2370), not that the state is a pure glueball. The absent decay rules out conventional mesons but cannot separate a glueball from a flavour-singlet hybrid.
What are the open problems?
The production rate in J/psi radiative decay is about ten times higher than lattice predicts for a glueball, and a flavour-singlet hybrid (quarks plus an excited gluon field) reproduces the same suppressed decay. Both keep the identification from being unique.
Why search in J/psi decays?
The J/psi is a charm quark-antiquark pair; when it decays radiatively it produces a photon plus a gluon-rich system, a clean, glue-heavy environment ideal for producing glueballs, with the photon tagging the event.
What would settle it?
A full amplitude-level analysis rather than inclusive branching fractions, production of the state at the LHC where there is no gluon-rich bias, unquenched lattice results on mixing, and a consistent account of the scalar and tensor glueball partners.
Responses