The long-running muon g-2 mystery has not disappeared so much as changed shape. New calculations of the muon’s magnetic behavior now agree closely with Fermilab’s record-setting measurement, weakening one of particle physics’ most celebrated hints of undiscovered particles.
But that agreement has exposed another problem. The calculations conflict with electron-positron collision measurements collected over roughly four decades, while a newer result from the CMD-3 detector in Russia points in a different direction from many of the experiments that came before it.
The muon’s wobble became a test of particle physics
Muons are short-lived elementary particles with the same electric charge as electrons but about 207 times their mass. Like electrons, they possess a quantum property called spin, which makes them behave somewhat like tiny magnets when placed inside a magnetic field.
The strength of that magnetism is described by a number called the g-factor. Basic relativistic quantum mechanics predicts a value of exactly 2, but interactions with other particles shift it slightly, producing what physicists call the anomalous magnetic moment, written as aμ = (g-2)/2.
That small correction contains contributions from every known particle and force included in the Standard Model. It could also contain contributions from particles that have not yet been discovered, which is why an accurate comparison between measurement and calculation has attracted so much attention.
Brookhaven National Laboratory reported a measurement in the early 2000s that was higher than the accepted Standard Model prediction. The difference was only around one part in a million, but it persisted strongly enough to become one of the field’s leading possible signs of physics beyond the Standard Model.
Fermilab made the experimental number far more precise
To test Brookhaven’s result, scientists transported its 15-meter-wide magnetic storage ring from New York to the Fermi National Accelerator Laboratory in Illinois. The upgraded Muon g-2 experiment began operating in 2017 and collected six runs of data between 2018 and 2023.
Fermilab released its final result in June 2025. The experiment measured the muon’s anomalous magnetic moment with a precision of 127 parts per billion, exceeding its original design target of 140 parts per billion and agreeing with its earlier results from 2021 and 2023.
When combined with Brookhaven’s measurement, the result produced an experimental world average with a precision of approximately 124 parts per billion. As ScienceBlog reported when the final Fermilab result was released, the experimental value had become exceptionally stable, leaving the theoretical prediction as the larger source of uncertainty.
Lattice QCD changed the Standard Model prediction
The hardest part of calculating muon g-2 involves the strong nuclear force, which binds quarks into particles such as protons, neutrons and pions. In particular, physicists must calculate a contribution known as leading-order hadronic vacuum polarization, or LO-HVP.
For years, the most precise estimates used a data-driven method. Researchers measured how often electron-positron collisions produced hadrons and then used those measurements to infer how strongly similar quantum processes should affect a muon.
A major alternative emerged in 2021, when the BMW collaboration published a high-precision calculation using lattice quantum chromodynamics. Lattice QCD places quarks and gluons on a simulated grid of space and time, allowing researchers to calculate strong-force effects from the underlying theory rather than relying primarily on collider measurements.
The BMW calculation produced a larger LO-HVP contribution than the traditional data-driven estimate. That moved the complete Standard Model prediction much closer to the measured muon value and reduced the apparent need for unknown particles to explain the result.
Other lattice groups subsequently obtained compatible results for important portions of the calculation. In 2025, the Muon g-2 Theory Initiative adopted a consolidated lattice average and reported a Standard Model value of 116,592,033(62) × 10-11, corresponding to an uncertainty of 530 parts per billion.
Its difference from the experimental average was just 38(63) × 10-11. Statistically, that is consistent with no disagreement between the Standard Model and the measured muon wobble.
A 2026 calculation strengthens the agreement
The case became stronger in April 2026, when an international team led by BMW researchers published a new hybrid LO-HVP calculation in Nature. The researchers used finer lattice grids and combined their simulation with a small amount of experimental information from a low-energy region where the relevant collider measurements agree.
The result was 715.1(3.4) × 10-10 for the LO-HVP contribution, with an uncertainty of 0.48%. That was 1.6 times more precise than the collaboration’s earlier calculation.
When combined with the other Standard Model contributions, the new calculation gave a total prediction of 11,659,205.2(3.6) × 10-10. It differed from the measured value by only 0.5 standard deviations, which is far too small to constitute evidence of a genuine conflict between theory and experiment.
The calculation was not simply adjusted to match Fermilab. Less than 5% of the result came from experimental collision data, and that contribution was taken from a low-energy tail below the disputed region surrounding the rho meson peak.
Most of the answer still came from lattice QCD. The collaboration also found that its intermediate-distance result agreed with eight other lattice calculations, strengthening the argument that the newer theoretical approach is not an isolated computational outlier.
The new problem lies inside the collider measurements
The agreement between lattice calculations and Fermilab does not explain why the older data-driven calculation produced a smaller number. That method is grounded in real measurements of electron-positron annihilation, especially the reaction e+e– → π+π–, in which an electron and positron annihilate and produce two charged pions.
Pion production near the rho meson resonance supplies a particularly important part of the hadronic contribution. Even relatively small differences in the measured production rate can noticeably shift the final Standard Model prediction for muon g-2.
Experiments including KLOE in Italy, BaBar in California, BESIII in China and earlier detectors in Novosibirsk produced measurements that were not perfectly identical, but they broadly supported the lower data-driven prediction. Those datasets became the foundation of calculations that left a substantial gap between the Standard Model and the muon experiments.
The newer CMD-3 detector at the VEPP-2000 collider in Novosibirsk disrupted that picture. Its measurement, published in 2023 and peer-reviewed in 2024, found a higher pion-production rate that pulls the data-driven prediction toward both the lattice result and the direct muon measurement.
The disagreement among the collider datasets is now too large to ignore. In the 2026 Nature analysis, the lattice result for one carefully defined intermediate-distance quantity differed from the KLOE-based value by 6.2 standard deviations and from the BaBar-based value by 3.5 standard deviations.
By contrast, the difference from the CMD-3-based determination was only 1.3 standard deviations. The spread means physicists cannot simply average every available collider measurement and treat the result as a single consistent dataset.
Why the experiments may disagree
No confirmed explanation has yet emerged. Researchers are examining possible differences in detector calibration, event selection, luminosity measurements and the treatment of photons emitted before or after the electron-positron collision.
Radiative corrections are a particular concern because they are needed to convert what a detector records into the underlying pion-production cross section. A small effect handled differently by separate experiments could produce a systematic shift that becomes significant at the precision required for muon g-2.
The CMD-3 result has been subjected to extensive scrutiny without an obvious error being identified. At the same time, decades of earlier measurements were produced by different teams using different machines and experimental techniques, making it difficult to dismiss all of them as incorrect.
Lattice QCD may help identify where the disagreement occurs by calculating restricted energy or distance ranges that can be compared with individual collider datasets. Related work has already shown how lattice calculations can connect different pion reactions, as described in a separate ScienceBlog report on matter-antimatter collisions.
The anomaly has moved rather than vanished
The original muon g-2 discrepancy is no longer the clean signal of new physics it once appeared to be. Fermilab’s measurement remains extraordinarily precise, but modern lattice calculations can reproduce it using known Standard Model particles and forces.
That does not prove that no undiscovered particles affect the muon. It means the muon measurement, by itself, no longer provides persuasive evidence that they must exist.
The immediate mystery is now more practical and, in some respects, more uncomfortable. Physicists must determine why CMD-3, lattice QCD and the measured muon value tell a broadly consistent story while several long-established collider datasets tell another.
Until that conflict is resolved, the field cannot claim a final Standard Model prediction with precision comparable to Fermilab’s measurement. The muon may be wobbling exactly as the Standard Model predicts, but the experiments used to calculate that prediction are still refusing to line up.
