Lattice determination of the higher-order hadronic vacuum polarization contribution to the muon $g-2$

This paper presents the first sub-percent precision lattice QCD calculation of the next-to-leading order hadronic vacuum polarization contribution to the muon anomalous magnetic moment, yielding a result that is twice as precise as the 2025 White Paper estimate and exhibits a 4.6$\sigma$ tension with data-driven evaluations based on pre-CMD-3 hadronic cross-section measurements.

The Gist

The Big Picture: The Muon's "Wobble"

Imagine the muon as a tiny, spinning top. In the world of quantum physics, this top doesn't just spin in a vacuum; it's constantly bumping into a chaotic sea of virtual particles popping in and out of existence. These bumps cause the top to wobble slightly. This wobble is called the anomalous magnetic moment (or $g-2$).

Scientists have measured this wobble with incredible precision. However, when they try to calculate what the wobble should be using the Standard Model (our best rulebook for physics), there is a mismatch. It's like trying to predict the path of a boat in a storm, but your map of the ocean currents is slightly off.

The biggest source of this "map error" comes from the Hadronic Vacuum Polarization (HVP). Think of this as the "fog" in the ocean. The muon interacts with this fog, and because the fog is made of complex particles (quarks and gluons), it's very hard to calculate exactly how it affects the wobble.

The Problem: Two Maps, Two Results

For years, scientists have tried to map this fog in two ways:

1. The Data-Driven Map: They look at real-world experiments (like smashing particles together) to see how the fog behaves.
2. The Lattice Map: They use supercomputers to simulate the fog from the ground up, using pure math (Quantum Chromodynamics or QCD).

Recently, the "Data-Driven Map" has become confusing. Different experiments are giving different results, creating a fog of their own. The "Lattice Map" has been getting sharper, but until now, it could only map the main part of the fog (the Leading Order). It couldn't map the subtle, secondary ripples (the Next-to-Leading Order or NLO) with enough precision to be useful.

The Breakthrough: Mapping the Ripples

This paper presents the first time scientists have successfully mapped those secondary ripples (NLO) with sub-percent precision using the Lattice method.

Here is how they did it, using some creative analogies:

1. The "Noise-Canceling" Trick

The biggest challenge in these calculations is noise. Imagine trying to hear a whisper in a stadium full of screaming fans.

• The Old Way: You try to measure the whisper (NLOa) and the background noise (NLOb) separately. Both are loud and messy.
• The New Way: The authors discovered that the whisper and the noise are actually opposites. When you add them together, they cancel each other out perfectly, like noise-canceling headphones.
• The Result: Instead of hearing a loud, messy stadium, they hear a near-silent room. This "cancellation" allowed them to ignore the messy long-distance parts of the calculation that usually ruin the precision. This is the secret sauce that made their result so accurate.

2. The "Time-Momentum" Camera

To take a picture of these particles, they used a technique called Time-Momentum Representation.

• Imagine you are trying to understand a movie by looking at a single, long strip of film.
• Instead of looking at the whole movie at once (which is too blurry), they broke the film into three sections:
  • Short Distance (The Close-Up): Where the action is fast and the math is tricky. They used "subtraction" techniques to remove the blurry parts.
  • Long Distance (The Wide Shot): Where the signal is weak. Because of the "noise-canceling" trick mentioned above, this part became much easier to see.
  • The Middle: The bridge between the two.

3. The "Fog" Correction

Even with a perfect simulation, the computer grid they used is finite (like a pixelated image). Real life is infinite.

• They had to correct for the fact that their "ocean" was too small. They used a clever two-step method to mathematically "stretch" their small ocean to the size of the real universe, ensuring the results weren't just an artifact of the simulation size.

The Verdict: What Did They Find?

After all this hard work, they produced a number: $-101.57 \times 10^{-11}$.

• Precision: They are 0.6% sure of this number. This is twice as precise as the previous best estimate from the "Data-Driven" method.
• The Conflict: Their number is slightly lower than the current "Data-Driven" average (by about 1.4 standard deviations).
• The Tension: If you compare their result to the older data-driven numbers (before a recent controversial experiment called CMD-3), the difference is huge (4.6 standard deviations).

Why Does This Matter?

1. Consistency: For the first time, we have a calculation for both the main part of the fog and the ripples using the same method (Lattice QCD). Before, the main part was from Lattice, but the ripples were from Data. Mixing methods was messy; now we have a consistent picture.
2. The Truth is Out There: The fact that their result disagrees with the older data-driven results suggests that the "Data-Driven" method might be missing something or that the new experimental data (CMD-3) is pointing toward a new reality.
3. New Physics: If the Lattice calculation is right and the Data-Driven calculation is wrong, it strengthens the case that the Standard Model is incomplete. The "wobble" of the muon might be caused by a brand new particle we haven't discovered yet.

In a Nutshell

This paper is like upgrading from a blurry, hand-drawn map of a stormy sea to a high-definition, 3D satellite image. By finding a mathematical trick to cancel out the noise, the authors managed to measure a tiny, subtle effect with incredible clarity. Their result challenges the old maps and suggests that the universe might be even stranger than we thought.

Technical Summary

1. Problem Statement

The muon anomalous magnetic moment ($a_\mu$) is a premier observable for testing the Standard Model (SM) and searching for new physics. While the experimental precision has reached 124 ppb (Fermilab E989), the theoretical prediction is limited by the Hadronic Vacuum Polarization (HVP) contribution.

• The Discrepancy: There is a significant tension between the SM prediction and experimental data. The dominant uncertainty in the SM prediction comes from the Leading Order (LO) HVP ($a_\mu^{\text{hvp, lo}}$).
• The NLO Gap: The Next-to-Leading Order (NLO) HVP contribution ($a_\mu^{\text{hvp, nlo}}$) is currently estimated using data-driven dispersive approaches (based on $e^+e^- \to \text{hadrons}$ cross-sections). These data-driven estimates suffer from the same experimental tensions (e.g., discrepancies between CMD-3 and earlier measurements) that plague the LO determination.
• The Need: To provide a consistent, first-principles check of the SM, both LO and NLO HVP contributions must be calculated using the same framework (Lattice QCD). Prior to this work, a sub-percent precision lattice calculation of the NLO contribution did not exist.

2. Methodology

The authors performed the first lattice QCD calculation of $a_\mu^{\text{hvp, nlo}}$ with sub-percent precision using the Time-Momentum Representation (TMR).

A. Formalism and Diagrams

The calculation covers three classes of NLO diagrams (illustrated in Fig. 1):

1. NLOa: Photon lines and muon loops (dominant negative contribution).
2. NLOb: Electron and tau lepton loops (positive contribution, partially cancels NLOa).
3. NLOc: Two QCD insertions (subleading).

The contribution is expressed via the zero-momentum projected vector correlator $G(t)$:
$$a_\mu^{\text{hvp, (i)}} = \left(\frac{\alpha}{\pi}\right)^3 \int_0^\infty dt_1 \dots dt_m \, \tilde{f}^{(i)}(\hat{t}_1, \dots) G(t_1) \dots G(t_m)$$
where $\tilde{f}^{(i)}$ are space-like NLO time-kernel functions.

B. Lattice Setup

• Ensembles: 35 CLS (Coordinated Lattice Simulations) gauge ensembles with $N_f = 2+1$ flavors of non-perturbatively $\mathcal{O}(a)$-improved Wilson fermions.
• Parameters: Six lattice spacings ($a \approx 0.039 - 0.097$ fm) and pion masses ranging from $\sim 430$ MeV down to the physical point.
• Scale Setting: Determined via the gradient-flow observable $\sqrt{t_0}$ with $\sqrt{t_0^{\text{ph}}} = 0.1440(7)$ fm.
• Currents: Two implementations of the electromagnetic current (local and point-split conserved) were used to control discretization errors.

C. Key Technical Strategies

1. NLOa & NLOb Cancellation: A critical feature is the strong cancellation between the NLOa and NLOb time-kernels at large Euclidean times ($t \gtrsim 1.5$ fm). By computing the sum NLOa&b directly, the authors suppress the long-distance (LD) contribution, which is typically the source of the largest statistical noise and finite-volume (FV) systematics in LO calculations.
2. Short-Distance (SD) Treatment: The NLOa kernel contains logarithmically enhanced $\mathcal{O}(a^2 \ln^2 a)$ artifacts. These were removed by subtracting the leading kernel behavior and restoring the piece perturbatively using QCD vacuum polarization up to $\mathcal{O}(\alpha_s^4)$.
3. Long-Distance (LD) Noise Control: Addressed via:
   • Low-mode averaging (LMA) for light quarks.
   • Spectral reconstruction of the correlator tail using GEVP analysis.
   • Standard bounding methods.
4. Finite-Volume (FV) Corrections: Applied using a two-step procedure: the Hansen–Patella formalism for small $t$ and the Meyer–Lellouch–Lüscher method for large $t$, followed by a continuum extrapolation using NNLO chiral perturbation theory.
5. Isospin Breaking (IB): Since ensembles are isospin-symmetric, IB corrections (electromagnetic and strong) were estimated phenomenologically using the spacelike representation. Crucially, the NLOa/NLOb cancellation also suppresses the IB shift, resulting in a small total correction ($\Delta_{\text{IB}} \approx 0.06 \times 10^{-11}$).

3. Key Results

The authors obtained the following results in the continuum limit (units of $10^{-11}$):

• NLOa: $-216.83(81)(87)(99)(81)(34)(31)[1.81]$
• NLOb: $+111.51(55)(64)(64)(58)(16)(25)[1.24]$
• NLOc: $+3.75(6)(6)(3)(3)[10]$
• Total NLO HVP:
  $$a_\mu^{\text{hvp, nlo}} = -101.57(26)_{\text{stat}}(29)_{\text{syst}} \times 10^{-11}$$
  • Total Relative Error: 0.6% (combining statistical and systematic uncertainties).

Breakdown of Uncertainties:
The total uncertainty is dominated roughly equally by:

1. Long-distance statistical noise.
2. Model-average spread (chiral-continuum extrapolation).
3. Scale-setting uncertainty ($t_0$).
4. Isospin-breaking corrections.

4. Significance and Comparison

• Precision: This result is two times more precise than the estimate in the 2025 White Paper (WP25) update, which relied entirely on data-driven methods.
• Consistency Check: The result provides a fully consistent lattice determination for both LO and NLO HVP, removing the methodological inconsistency where only LO was lattice-based in previous SM predictions.
• Tension with Data-Driven Estimates:
  • The result is 1.4$\sigma$ below the WP25 average ($-99.6 \pm 1.3$).
  • It shows a 4.6$\sigma$ tension with the KNT19 data-driven evaluation (which excludes CMD-3 data).
  • This pattern mirrors the tension observed in the LO HVP between lattice calculations and pre-CMD-3 dispersive results, suggesting that the CMD-3 data may be an outlier or that the lattice results are robust against the experimental spread.

5. Conclusion

This work represents a milestone in muon $g-2$ physics by delivering the first sub-percent precision lattice determination of the NLO HVP contribution. The success relies heavily on the cancellation between NLOa and NLOb kernels, which renders the calculation intrinsically less sensitive to long-distance systematics than the LO calculation. The result challenges data-driven evaluations that do not incorporate the CMD-3 measurement and strengthens the case for a potential discrepancy between the Standard Model and experimental data.