Muon g-2: Final Result
Fermilab measures the muon's magnetic anomaly to 127 parts per billion (June 2025); new lattice calculations bring theory into line.
Open in the interactive tree →The muon acts like a tiny spinning magnet whose strength deviates slightly from the simple value 2 because virtual particles contribute. A mismatch with the prediction would hint at new particles. Fermilab used a 14-m storage ring with a superconducting magnet (moved from Brookhaven) and took data from 2018 to 2023.
As of October 2026
On 3 June 2025 the collaboration released the final result from six years of data, with a precision of 127 ppb, beating the design goal of 140 ppb; it agrees with the 2021 and 2023 results. The new experimental world average of the anomaly is 0.001165920715(145). The Muon g-2 Theory Initiative's 2025 white paper (27 May 2025) replaced electron-positron data with lattice QCD and predicts 0.00116592033(62); the difference is 38(63) in units of 10^-11, well under one sigma, so the earlier gap of 4 to 5 sigma has disappeared. The dispute between lattice and data-driven predictions is not fully settled.
Open steps
- Sharper lattice hadronic term High AI leverageCut the lattice error on hadronic vacuum polarization (isospin breaking, finite volume, noise); theory error of about 530 ppb is four times the experiment's 127 ppb.
- Reconcile electron-positron data Medium AI leverageExplain why the CMD-3 measurement and older e+e- data disagree in the two-pion channel, with new independent measurements and combined fits.
- Muon-electron scattering route (MUonE) Low AI leverageMeasure the hadronic contribution from muon-electron scattering at CERN (MUonE) as a third determination beside lattice QCD and e+e- data.
Where AI could help
Medium AI leverage. The experiment is finished; the open work is the disputed theory input, where AI speeds lattice sampling but data and independent calculations decide.
- Flow-based and equivariant samplers to cut autocorrelation in lattice QCD ensembles
- Machine-learned variance reduction for noisy correlators in hadronic contributions
- Automated cross-checks between lattice and electron-positron data-driven analyses
Shown so far
- In August 2022 a preprint showed how gauge-equivariant normalizing-flow models can be combined to sample QCD field configurations, aimed at critical slowing-down and topological freezing in lattice calculations. source
Prerequisites
- Superconductivity1911
- Quantum Electrodynamics1948The muon g-2 measurement tests QED and the Standard Model to many digits
- Standard Model of Particles1973