The anomalous magnetic moment

The Dirac equation hands us a clean integer: a structureless spin-½ particle in a magnetic field is a magnet with gyromagnetic ratio $g=2$ , no free parameter, derived once and for all in aq-pp-magnetic-moment-g2. Nature disagrees at the fifth decimal. The measured electron has $g/2 = 1.001\,159\,652\,180\,59$ , and the excess — the anomalous magnetic moment $a=(g-2)/2$ — is not a flaw in Dirac's theory but the fingerprint of quantum field theory pressed onto a single number. A real electron is never bare: it endlessly emits and reabsorbs virtual photons, conjures virtual electron–positron pairs, and briefly borrows every heavier particle the vacuum can supply, and each of those fleeting excursions nudges how it couples to a field. Summed, they lift $g$ above 2 by a calculable amount. This article is about that excess as a number you compute and measure. It takes $g=2$ and the vertex form factor $a=F_2(0)$ as recalled results and asks the deeper questions the anomaly makes possible: how the correction organises into an asymptotic series in $\alpha/\pi$ with pure-number coefficients; why five loops — the last drawn from 12,672 Feynman diagrams — match the electron to ten significant figures; how that agreement, read backwards, delivers the best laboratory value of the fine-structure constant; and why the muon, forty-three thousand times more sensitive to heavy virtual states, became the flagship search for physics beyond the Standard Model, with a hadronic uncertainty that is still the hardest number in precision physics.

Key takeaways

The anomalous magnetic moment $a=(g-2)/2=F_2(0)$ is the loop-generated excess over the Dirac value $g=2$ ; every digit of $a$ is quantum field theory.
QED gives $a$ as an asymptotic series in $\alpha/\pi\approx 1/430$ : $a=\tfrac12\left(\tfrac{\alpha}{\pi}\right)-0.3285\left(\tfrac{\alpha}{\pi}\right)^2+1.181\left(\tfrac{\alpha}{\pi}\right)^3-\cdots$ , computed now through five loops.
The electron is almost pure QED — hadronic and weak effects enter only at $\sim10^{-12}$ — so $a_e$ is both the sharpest test of any theory and, run in reverse, the best determination of $\alpha$ .
The muon's sensitivity to a heavy state of mass $M$ scales as $(m_\mu/M)^2$ , enhanced over the electron by $(m_\mu/m_e)^2\approx4.3\times10^4$ , making $a_\mu$ the premier new-physics probe.
For $a_\mu$ the hadronic vacuum-polarisation term dominates the theory uncertainty, and data-driven ( $e^+e^-$ ) versus lattice-QCD evaluations have disagreed — the crux of the muon story.
Key idea: the anomalous magnetic moment turns "how magnetic is a lepton" into a sum over every virtual process the vacuum allows, making one dimensionless number simultaneously the most precise test of QED, the best measurement of α, and a sensitive lever on undiscovered physics.

The core insight

The anomalous magnetic moment $a=(g-2)/2$ is the sum of every quantum loop a lepton can radiate; organised as a series in $\alpha/\pi$ with computable pure-number coefficients, it makes one quantity at once the most stringent test of QED, the best route to the fine-structure constant, and — through its $(m_\ell/M)^2$ reach — a probe of physics beyond the Standard Model.

Where this lives today

The anomalous moment sits at the sharp edge of two frontiers. On the electron side, a single particle stored for months in a cryogenic Penning trap at Northwestern yields $a_e$ to about 0.1 part per billion; combined with Kinoshita-era five-loop QED, the comparison is the tightest confrontation of theory and experiment anywhere, and inverted it gives $\alpha^{-1}$ to ten digits. That determination now referees a dispute about $\alpha$ itself: the Berkeley caesium-recoil ( $\alpha^{-1}=137.035\,999\,046$ ) and Paris rubidium-recoil ( $137.035\,999\,206$ ) atom-interferometry values differ at the several-sigma level. On the muon side, the Fermilab Muon g-2 experiment — reusing the 15-metre superconducting ring barged from Brookhaven — completed its programme in 2025, fixing $a_\mu$ to about 0.13 part per million. The suspense has migrated to theory: for years measured $a_\mu$ sat roughly four standard deviations above the data-driven prediction, but lattice-QCD evaluations of the hadronic vacuum polarisation (led by the BMW collaboration) came out higher, closing much of the gap, and the discrepant CMD-3 $e^+e^-\to\pi^+\pi^-$ measurement threw the data-driven input into question. The MUonE experiment, extracting the hadronic term from muon–electron scattering, and new $e^+e^-$ and lattice programmes now race to decide whether a genuine crack survives.

The physical picture

The bare vertex of aq-pp-magnetic-moment-g2 — a single photon meeting a straight electron line — gives $g=2$ exactly. The anomaly is everything that decorates that vertex. At order $\alpha$ the electron emits a virtual photon, interacts with the external field mid-flight, and reabsorbs it: one triangle diagram, Schwinger's, contributing $a=\alpha/2\pi$ . At the next order seven diagrams appear, then seventy-two, then 891, then 12,672 — each layer weighted by another power of $\alpha/\pi\approx1/430$ , so the tower collapses fast. The first figure makes that collapse quantitative: each loop's contribution to the electron anomaly drops about four orders of magnitude, until at the fifth it sinks beneath the hadronic and weak effects the electron barely feels.

Figure

The electron anomaly as a converging tower of loops: each QED order contributes about α/π ≈ 1/430 times less than the last, from Schwinger's one loop 1.16×10⁻³ down through five loops, so the tiny hadronic (1.7×10⁻¹²) and weak (0.03×10⁻¹²) pieces set the floor of what the electron can currently see

For the electron that tower is essentially the whole story, because the heavier virtual states it can reach are so far off. The muon flips the emphasis. Being 207 times heavier, it couples $(m_\mu/m_e)^2\approx4\times10^4$ times more strongly to any heavy virtual state, so hadronic and even electroweak loops climb out of the noise — and any undiscovered heavy particle would too. The muon anomaly is therefore a sum over all of physics at its mass scale, and comparing measured $a_\mu$ with the Standard-Model total is a search for what is missing. The second figure is the frontier as it stands: the superb Fermilab measurement set against two Standard-Model predictions that differ because the hadronic vacuum-polarisation term is computed two different ways.

Figure

The muon frontier in units of 10⁻¹¹: the Fermilab measurement a_μ ≈ 116592071 is far tighter than the theory, whose spread comes entirely from the hadronic vacuum-polarisation term — the data-driven e⁺e⁻ evaluation sits low near 116591810 while the lattice-QCD evaluation near 116592033 agrees with experiment, so the historic discrepancy now hinges on which hadronic number is right

The core equations

The anomaly is defined against the Dirac baseline. The gyromagnetic ratio is written $g=2(1+a)$ , and the vertex analysis of aq-pp-form-factors and aq-pp-magnetic-moment-g2 identifies $a$ with the Pauli (magnetic) form factor evaluated at zero momentum transfer.

a\equiv\frac{g-2}{2}=F_2(0) \qquad (1)

where: $g$ — gyromagnetic ratio; $a$ — anomalous magnetic moment, dimensionless; $F_2(q^2)$ — magnetic form factor, with $F_2(0)$ its static value. Reading it: tree level gives $F_2(0)=0$ , so $a$ contains only radiative corrections — there is no classical piece to subtract, and every digit is a loop effect. QED delivers $a$ as an asymptotic power series in $\alpha/\pi$ , the natural loop-counting parameter. Each additional factor of $\alpha/\pi\approx2.3\times10^{-3}$ marks one more virtual photon, and the coefficients are calculable pure numbers.

a=C_1\frac{\alpha}{\pi}+C_2\left(\frac{\alpha}{\pi}\right)^2+C_3\left(\frac{\alpha}{\pi}\right)^3+C_4\left(\frac{\alpha}{\pi}\right)^4+\cdots,\qquad C_1=\tfrac12 \qquad (2)

where: $\alpha=e^2/4\pi\hbar c\approx1/137.036$ — fine-structure constant; $C_n$ — the $n$ -loop coefficient. Reading it: $C_1=\tfrac12$ is Schwinger's result, so $a=\alpha/2\pi$ at leading order; the smallness of $\alpha/\pi$ is why a handful of terms suffices. The leading coefficients are known to extraordinary depth — the mass-independent (universal) part shared by every lepton, from one loop to five.

C_1=0.5,\quad C_2=-0.328\,478\,965,\quad C_3=1.181\,241,\quad C_4\approx-1.912,\quad C_5\approx6.7 \qquad (3)

where: $C_2$ (Petermann–Sommerfield, analytic), $C_3$ (Laporta–Remiddi, analytic), $C_4$ and $C_5$ (Kinoshita and collaborators, largely numerical, the five-loop term from 12,672 diagrams). Reading it: the alternating signs make successive loops partly cancel, and each $|C_n(\alpha/\pi)^n|$ falls by roughly $10^{-2.6}$ per order. For a real lepton the full Standard-Model value adds two more sectors to the QED series: electroweak loops (with $W$ , $Z$ , Higgs) and hadronic loops (quarks and gluons, not perturbatively calculable). This decomposition is exact bookkeeping of every virtual contribution.

a^{\text{SM}}=a^{\text{QED}}+a^{\text{EW}}+a^{\text{HVP}}+a^{\text{HLbL}} \qquad (4)

where: $a^{\text{QED}}$ — photon-and-lepton loops (equation 2); $a^{\text{EW}}$ — weak-boson loops; $a^{\text{HVP}}$ — hadronic vacuum polarisation; $a^{\text{HLbL}}$ — hadronic light-by-light. Reading it: for the electron the last three are $\sim10^{-12}$ ; for the muon they are measurable, and $a^{\text{HVP}}$ carries the dominant theoretical uncertainty. The hadronic vacuum-polarisation term cannot be computed from perturbation theory, but the optical theorem ties it to measured data through a dispersion relation: the same virtual quark loop that dresses the photon also produces real hadrons in $e^+e^-$ collisions.

a_\ell^{\text{HVP,LO}}=\frac{1}{4\pi^3}\int_{s_{\text{th}}}^{\infty}\!\! ds\,K(s)\,\sigma^0_{\text{had}}(s),\qquad K(s)\!>\!0,\ K(s)\xrightarrow{s\gg m_\ell^2}\frac{m_\ell^2}{3s} \qquad (5)

where: $\sigma^0_{\text{had}}(s)$ — bare cross section for $e^+e^-\to$ hadrons at energy $\sqrt s$ ; $K(s)$ — a positive QED kernel; $s_{\text{th}}=m_{\pi^0}^2$ — threshold. Reading it: because $K(s)\sim m_\ell^2/s$ falls with energy, the low-energy $\rho$ -meson region ( $e^+e^-\to\pi^+\pi^-$ ) dominates, which is why a single hadronic measurement can swing the whole prediction. Finally, the sensitivity to a hypothetical heavy particle of mass $M$ decouples quadratically, and the electron measurement inverts equation (2) to yield $\alpha$ itself.

\delta a_\ell\sim\left(\frac{m_\ell}{M}\right)^2,\qquad \alpha=\alpha\big(a_e^{\text{meas}}\big)\ \text{by solving (2)} \qquad (6)

where: $\delta a_\ell$ — shift from new physics at scale $M$ ; $m_\ell$ — lepton mass. Reading it: heavy states decouple, so the heavier lepton wins — hence the muon; and since the QED coefficients are known, a measured $a_e$ determines $\alpha$ more precisely than almost any other method.

Derivation AWhy the one-loop anomaly is finite and equals alpha over two pi

The goal is to see why the leading correction to $g$ is a clean, ultraviolet-finite pure number — Schwinger's $\alpha/2\pi$ — even though the same one-loop diagram's charge part is divergent and needs renormalisation. We take the dressed vertex as the sum of a $\gamma^\mu$ structure and an $i\sigma^{\mu\nu}q_\nu/2m$ structure (recalled from aq-pp-form-factors), track the one-loop triangle onto those two structures, and show that gauge invariance quarantines all the infinity in the charge piece, leaving the magnetic piece finite and computable by an elementary Feynman-parameter integral.

Step 1. Write the order- $\alpha$ vertex correction as the triangle in which the electron emits a virtual photon of momentum $k$ , absorbs the external-field photon, and reabsorbs $k$ . Its amplitude is the loop integral $\int\! d^4k\,\gamma^\alpha(\slashed p'-\slashed k+m)\gamma^\mu(\slashed p-\slashed k+m)\gamma_\alpha/[\,(k^2)((p'-k)^2-m^2)((p-k)^2-m^2)\,]$ . Why: this is the unique one-loop dressing of the bare $\gamma^\mu$ , so any deviation from $g=2$ must originate here — nothing at lower order can shift it.

Step 2. Reduce the numerator using the Dirac algebra and combine the three propagator denominators with Feynman parameters $x,y,z$ ( $x+y+z=1$ ), shifting the loop momentum to $\ell=k-(y\,p+z\,p')$ so the denominator becomes a single quadratic $(\ell^2-\Delta)^3$ . Project the result onto the two Lorentz structures $\gamma^\mu$ and $i\sigma^{\mu\nu}q_\nu/2m$ . Why: the Gordon decomposition guarantees the amplitude can only take these two forms, so the whole calculation reduces to reading off two scalar coefficients.

Step 3. Inspect the two coefficients at $q^2\to0$ . The $\gamma^\mu$ coefficient carries a logarithmically divergent piece $\int d^4\ell\,\ell^2/(\ell^2-\Delta)^3$ — the charge renormalisation, absorbed into the physical $e$ , leaving $F_1(0)=1$ . The $i\sigma^{\mu\nu}q_\nu/2m$ coefficient has no $\ell^2$ in its numerator, so its integral $\int d^4\ell/(\ell^2-\Delta)^3$ converges. Why: the moment is protected — a divergent $F_2(0)$ would be an uncomputable infinite moment, and gauge invariance forbids exactly that, which is what makes the anomaly predictable.

Step 4. Evaluate the finite piece. The convergent loop integral gives $1/\Delta$ up to constants, and the remaining Feynman-parameter integral collapses to $F_2(0)=\dfrac{\alpha}{2\pi}\displaystyle\int_0^1\! dz\,(1-z)\cdot2=\dfrac{\alpha}{2\pi}$ , so $a=F_2(0)=\dfrac{\alpha}{2\pi}\approx1.16\times10^{-3}$ . Why: this is Schwinger 1948 — a parameter-free number times $\alpha$ that matched Kusch and Foley and proved the infinities of field theory were bookkeeping, not physics.

Go deeper · Derivation B · 2 minThe anomaly as an asymptotic series and the sum over all physics

Derivation BThe anomaly as an asymptotic series and the sum over all physics

This derivation explains the structure of $a$ beyond the leading term: why it is an asymptotic series in $\alpha/\pi$ with universal pure-number coefficients, why the muon's coefficients differ from the electron's, and how non-QED physics slots in. The point is conceptual bookkeeping — understanding what each term counts — because that structure is what turns $a$ into a probe of everything the vacuum contains, and what tells us how many loops we must compute for a target precision.

Step 1. Count powers. Every extra internal photon adds a factor $e^2\sim\alpha$ and, after the loop integral, a $1/\pi$ , so the expansion is organised in $\alpha/\pi\approx2.3\times10^{-3}$ ; the $n$ -loop coefficient $C_n$ is a rational-plus-transcendental pure number ( $\zeta$ -values, polylogarithms). Why: the tiny loop-counting parameter is why five terms reach $10^{-13}$ — the series is asymptotic, not convergent, and truncation error is set by the first omitted term.

Step 2. Split $C_n$ into a mass-independent (universal) part $A_1^{(2n)}$ , the same for every lepton, and mass-dependent parts $A_2^{(2n)}(m_\ell/m_{\ell'})$ from loops containing a different lepton. For the electron these are tiny ( $m_e\ll m_\mu$ ); for the muon, light virtual electron loops enhance $C_2$ from $-0.328$ to about $+1.09$ . Why: the muon "sees" the electron as a light, easily excited loop, so its low-order coefficients are larger — a first hint that the heavier lepton is more sensitive to other states.

Step 3. Add the sectors QED cannot supply. Weak bosons give $a^{\text{EW}}$ , calculable but suppressed by $(m_\ell/M_W)^2$ ; quark–gluon loops give the non-perturbative hadronic terms $a^{\text{HVP}}$ and $a^{\text{HLbL}}$ . Equation (4) is then the complete Standard-Model prediction. Why: because $a$ receives a contribution from every particle that can appear in a loop, it sums over all of physics at the lepton's mass scale — the reason it doubles as a discovery tool.

Step 4. Read off the strategy. To match an experiment of relative precision $\epsilon$ , compute QED to the order where $|C_n(\alpha/\pi)^n|<\epsilon\,a$ , then add EW and hadronic pieces at the same level. For $a_e$ this means five QED loops and little else; for $a_\mu$ it means five loops plus percent-level control of $a^{\text{HVP}}$ . Why: the series structure shows where the effort must go — for the muon, not into more photons but into the hadronic term of equation (5).

Go deeper · Derivation C · 2 minThe muon lever and the hadronic bottleneck

Derivation CThe muon lever and the hadronic bottleneck

Here we make quantitative the two facts that define the muon programme: that $a_\mu$ is vastly more sensitive to heavy new physics than $a_e$ , and that its Standard-Model prediction is limited not by QED but by the hadronic vacuum-polarisation integral of equation (5). We estimate the new-physics scaling from dimensional analysis of a heavy loop, then trace why the hadronic term resists first-principles calculation and must be pulled from data or the lattice.

Step 1. Model a heavy particle of mass $M$ coupling to the lepton with strength $g_{\text{NP}}$ . Its one-loop contribution must be dimensionless and vanish as $M\to\infty$ , so on dimensional grounds $\delta a_\ell\sim\dfrac{g_{\text{NP}}^2}{16\pi^2}\dfrac{m_\ell^2}{M^2}$ . Why: the magnetic-moment operator flips chirality, costing one power of $m_\ell$ , and a second comes from the external momentum — so the effect is quadratic in the lepton mass, the origin of the whole lever.

Step 2. Take the ratio of muon to electron sensitivity: $\dfrac{\delta a_\mu}{\delta a_e}=\Big(\dfrac{m_\mu}{m_e}\Big)^2=\Big(\dfrac{105.66}{0.5110}\Big)^2=(206.77)^2=4.28\times10^4$ . Why: with everything else common, the mass ratio squared is the entire enhancement — the muon is about forty-three thousand times the better antenna, which is why it, not the more precisely measured electron, leads the search.

Step 3. Now the hadronic term. The vacuum-polarisation blob is a quark loop, but at the low momenta that dominate $K(s)$ the quarks are confined, so perturbative QCD fails. The optical theorem rescues it: $\text{Im}\,\Pi(s)$ is fixed by the total $e^+e^-\to\text{hadrons}$ cross section, giving the dispersion integral (5). Why: an uncomputable loop is traded for a measurable cross section — importing the experimental uncertainty of $\sigma_{\text{had}}$ , especially in the $\rho$ -resonance $\pi^+\pi^-$ channel the kernel weights most.

Step 4. Locate the tension. Two routes evaluate $a_\mu^{\text{HVP}}$ : the data-driven dispersion integral over $e^+e^-$ data, and a direct lattice-QCD computation of the same correlator. They have disagreed by an amount comparable to the entire "muon anomaly", and a discrepant $e^+e^-\to\pi^+\pi^-$ measurement (CMD-3) sharpened the puzzle. Why: the prediction of $a_\mu$ is only as good as $a_\mu^{\text{HVP}}$ , so whether measured $a_\mu$ signals new physics hinges on which hadronic evaluation is correct — a theory problem, not an experimental one.

Limiting cases

Case	Result	Meaning
Tree level (Dirac)	$a=0$ exactly	$g=2$ ; no radiative correction, the baseline the anomaly is measured against
One loop (Schwinger)	$a=\alpha/2\pi\approx1.16\times10^{-3}$	universal leading term, identical for every lepton
Electron	$a_e\approx\dfrac{\alpha}{2\pi}-0.328\big(\tfrac{\alpha}{\pi}\big)^2+\cdots$	essentially pure QED; hadronic + weak only $\sim10^{-12}$ ; used to fix $\alpha$
Muon	$a_\mu\approx1.1659\times10^{-3}$	QED coefficients enhanced by light-lepton loops; HVP + EW measurable; new-physics lever $(m_\mu/m_e)^2$
Tau	$a_\tau\sim(m_\tau/m_\mu)^2$ more sensitive	dominated by hadronic and weak loops, but $\tau$ lifetime too short for a precision measurement
Heavy new physics, mass $M$	$\delta a_\ell\sim(m_\ell/M)^2$	decouples for large $M$ ; the muon is the practical probe, the electron nearly blind

Common trapThe word anomalous in the muon anomalous magnetic moment means a disagreement with the Standard Model

It does not — and conflating the two meanings is the most common error in popular accounts. "Anomalous" refers to the deviation of $g$ from the Dirac value 2, i.e. to the loop-induced quantity $a=(g-2)/2$ itself. This anomaly is entirely expected and precisely predicted by the Standard Model: QED alone gives $a_\mu\approx1.1659\times10^{-3}$ , and the electroweak and hadronic sectors add known amounts. What made headlines is a second, separate thing: a possible gap between the measured $a_\mu$ and the Standard-Model prediction of that same $a_\mu$ — the so-called "muon $g-2$ discrepancy." An electron or muon having $a\neq0$ is not new physics; it is textbook QED. Only a mismatch between the measured $a_\mu$ and the complete theoretical $a_\mu$ would be. And whether that mismatch is real is presently unsettled, because the hadronic vacuum-polarisation input (equation 5) differs between data-driven and lattice evaluations — so the "anomaly" that matters is a comparison of two numbers, not the existence of $a$ .

Worked examples

1 · The electron anomaly, loop by loop (fully worked)

Using $\alpha^{-1}=137.035\,999$ , evaluate the QED series (2)–(3) for $a_e$ term by term and compare with the measured $a_e=1.159\,652\,180\,59\times10^{-3}$ . Show that the series has converged below the measurement's precision by five loops.

Step 1. Form the expansion parameter: $\dfrac{\alpha}{\pi}=\dfrac{1}{137.035999\times\pi}=\dfrac{1}{430.511}=\mathbf{2.322\,819\times10^{-3}}$ . Every loop multiplies the previous scale by roughly this.

Step 2. One loop: $a^{(1)}=\tfrac12\cdot\dfrac{\alpha}{\pi}=\dfrac{\alpha}{2\pi}=\mathbf{1.161\,410\times10^{-3}}$ . This already matches the measurement to three significant figures — Schwinger's triumph — but is too large by about $1.8\times10^{-6}$ .

Step 3. Two loops: $a^{(2)}=-0.328479\,(\alpha/\pi)^2=-0.328479\times5.3955\times10^{-6}=\mathbf{-1.772\times10^{-6}}$ , pulling the total down to $1.159\,637\times10^{-3}$ — now correct to six figures. Three loops: $a^{(3)}=+1.18124\,(\alpha/\pi)^3=+1.480\times10^{-8}$ , giving $1.159\,6522\times10^{-3}$ .

Step 4. Four and five loops: $a^{(4)}=-1.912\,(\alpha/\pi)^4=-5.6\times10^{-11}$ and $a^{(5)}=+6.7\,(\alpha/\pi)^5=+4.6\times10^{-13}$ . Summing all five: $a_e^{\text{QED}}=\mathbf{1.159\,652\,177\times10^{-3}}$ . The measured value is $1.159\,652\,181\times10^{-3}$ ; the residual $\approx4\times10^{-12}$ is filled by the tiny hadronic ( $+1.7\times10^{-12}$ ) and weak ( $+0.03\times10^{-12}$ ) pieces plus mass-dependent QED terms. Five QED loops reproduce $a_e$ to ten significant figures — the most precise agreement between theory and experiment in all of science.

2 · Turning the electron into a ruler for α (fully worked)

Because the QED coefficients are known, the same agreement determines $\alpha$ . Estimate $\alpha^{-1}$ by inverting the series, keeping only the dominant one-loop term, then indicate the precision the full calculation reaches.

Step 1. To leading order $a_e\approx\alpha/2\pi$ , so $\alpha\approx2\pi\,a_e=2\pi\times1.159\,652\times10^{-3}=\mathbf{7.2865\times10^{-3}}$ , i.e. $\alpha^{-1}\approx137.24$ — close, but the higher loops matter at this precision.

Step 2. Include the two-loop term: solve $a_e=\tfrac12 x-0.328479\,x^2+\cdots$ with $x=\alpha/\pi$ . The negative $x^2$ term means the true $x$ must be slightly larger than $2a_e$ to compensate, nudging $\alpha^{-1}$ down toward its true value $\approx137.036$ . Each further loop refines the last digits.

Step 3. With all five loops and the measured $a_e$ , the inversion yields $\alpha^{-1}=137.035\,999\,17(15)$ — a determination at the $2\times10^{-10}$ level, competitive with and independent of atom-recoil methods. The electron anomaly is simultaneously the best test of QED and, read backwards, one of the two best measurements of the fine-structure constant.

3 · Why the muon, not the electron, hunts new physics

Quantify the muon's advantage as a new-physics antenna, and identify which term limits its Standard-Model prediction.

Step 1. New physics of mass $M$ shifts the anomaly by $\delta a_\ell\sim(m_\ell/M)^2$ (Derivation C). The muon-to-electron ratio is $\big(m_\mu/m_e\big)^2=(206.77)^2=\mathbf{4.28\times10^4}$ — the muon is roughly forty-three thousand times more responsive to a heavy state.

Step 2. Yet $a_e$ is measured far more precisely than $a_\mu$ . The trade-off works out in the muon's favour anyway: a part-per-million effect in $a_\mu$ from a TeV-scale particle exceeds the muon measurement error, while the same particle's imprint on $a_e$ sits below even the electron's exquisite precision.

Step 3. The Standard-Model $a_\mu$ decomposes (equation 4) as $a_\mu^{\text{QED}}\approx116\,584\,719\times10^{-11}$ (huge, but known to $10^{-11}$ ), $a_\mu^{\text{EW}}\approx154\times10^{-11}$ , $a_\mu^{\text{HVP}}\approx6850\times10^{-11}$ , $a_\mu^{\text{HLbL}}\approx92\times10^{-11}$ . The QED part dominates the value; the hadronic part $a_\mu^{\text{HVP}}$ , uncertain at the $\pm40\times10^{-11}$ level, dominates the error. The muon's $(m_\mu/m_e)^2\approx4\times10^4$ enhancement makes it the discovery channel, while the hadronic vacuum-polarisation term — not QED — sets the precision of the comparison.

Physics of the portrait

The anomalous moment has two founding numbers: Schwinger's one-loop $\alpha/2\pi$ , and the tower of higher-loop coefficients that let it be tested to ten digits. The second story belongs to the physicist who spent half a century computing those coefficients.

Origins · 1 minToichiro Kinoshita and how this was discovered

Toichiro Kinoshita

1925–2023 · multi-loop QED and the anomalous magnetic moment

Toichiro Kinoshita turned the anomalous magnetic moment from a one-loop curiosity into the most precisely tested prediction in physics. Born in Japan and, from 1960, a professor at Cornell, he devoted decades to computing the higher-order QED coefficients $C_n$ of the electron and muon anomalies — the pure numbers that multiply each power of $\alpha/\pi$ . Where Schwinger's one loop was a single diagram, the fourth-order (two-loop) term needed seven, the sixth-order seventy-two, the eighth-order 891, and the tenth-order a staggering 12,672; no one could evaluate these by hand. Kinoshita's answer was to automate the enterprise, developing with collaborators — notably Makiko Nio, Masashi Hayakawa, and Tatsumi Aoyama — computer-algebra and numerical-integration machinery that generated, renormalised, and integrated the diagrams systematically, culminating around 2012 in the complete five-loop (tenth-order) universal coefficient. That heroic calculation is what makes the ten-significant-figure test of QED possible, and it is what lets the electron anomaly serve, in reverse, as a leading determination of the fine-structure constant. He also mapped the mass-dependent contributions that distinguish the muon from the electron, and his QED evaluation of $a_\mu$ underlies every modern comparison with the Fermilab measurement. Kinoshita worked on the problem essentially until his death in 2023, having built, diagram by diagram, the theoretical edifice on which precision QED still rests.

History

1948

— Schwinger computes the one-loop anomaly a = α/2π

1957

— Sommerfield and Petermann obtain the two-loop coefficient analytically

1996

— Kinoshita completes the four-loop (eighth-order) QED contribution

2012

— the five-loop (tenth-order) QED coefficient is evaluated from 12,672 diagrams

2025

— lattice-QCD hadronic vacuum polarisation reshapes the muon Standard-Model prediction

Going further

Texts. Peskin & Schroeder, An Introduction to Quantum Field Theory, §6.2–6.3 (the electron vertex and the one-loop $g-2$ ) · Schwartz, Quantum Field Theory and the Standard Model, §17 (the anomalous magnetic moment) · Griffiths, Introduction to Elementary Particles, 2nd ed., §7.6 · Halzen & Martin, Quarks and Leptons, §6 · Thomson, Modern Particle Physics, §4.4–4.7 (spin and the Dirac magnetic moment).

Provenance. The Dirac result $g=2$ , the Pauli reduction, and the vertex identity $a=F_2(0)$ are recalled by name from aq-pp-magnetic-moment-g2 and aq-pp-form-factors (not re-derived here). Schwinger's $F_2(0)=\alpha/2\pi$ follows Peskin & Schroeder §6.3 and Schwinger, Phys. Rev. 73, 416 (1948); the two-loop coefficient is Sommerfield/Petermann (1957); the multi-loop programme is Kinoshita, Aoyama, Hayakawa & Nio, and Laporta & Remiddi. The muon Standard-Model decomposition and the HVP dispersion relation follow the g-2 Theory Initiative white papers. A verifier should confirm: (1) $\alpha/2\pi=1.161\,410\times10^{-3}$ ; (2) the loop-by-loop sum $1.161410\times10^{-3}-1.772\times10^{-6}+1.480\times10^{-8}-5.6\times10^{-11}=1.159\,652\,18\times10^{-3}$ ; (3) $(m_\mu/m_e)^2=(206.77)^2=4.28\times10^4$ ; (4) $\alpha/\pi=2.3228\times10^{-3}$ for $\alpha^{-1}=137.036$ . Measured $a_e=1.159\,652\,180\,59\times10^{-3}$ and $a_\mu\approx1.16592071\times10^{-3}$ are standard experimental figures.

Prerequisites. The Dirac Equation · Gamma Matrices and the Clifford Algebra · Plane-Wave Dirac Solutions · Spin from the Dirac Equation. Leads to. Feynman Rules for QED · Renormalisation and the Running Coupling · Electroweak Radiative Corrections.

Ask the tutor: Ask me to walk through Schwinger's Feynman-parameter integral that yields $\alpha/2\pi$ , or to lay out exactly how the hadronic vacuum-polarisation term turns $e^+e^-$ collider data into a prediction for the muon.

What is the difference between g − 2 and the anomalous magnetic moment a?

They are the same physics, differing by a factor of two and a name. The g-factor is the full ratio of moment to spin in units of $q/2m$ ; the Dirac equation fixes its tree value at exactly 2. The quantity $g-2$ is the excess over that prediction, and $a=(g-2)/2$ is that excess halved. The halving is not cosmetic: $a$ equals the magnetic form factor at zero momentum transfer, $F_2(0)$ , precisely the object perturbation theory computes as a series in $\alpha/\pi$ . Experimentalists quote $g/2=1+a$ ; theorists quote $a$ , because tree level contributes exactly zero to it — so every digit of $a$ is a quantum loop effect.

Why do only five loops of QED matter for the electron?

Because the expansion parameter $\alpha/\pi\approx1/430$ is small, each loop shrinks the contribution by about two-and-a-half orders of magnitude. One loop gives $1.16\times10^{-3}$ , two loops correct it by $1.8\times10^{-6}$ , three by $1.5\times10^{-8}$ , four by $6\times10^{-11}$ , five by $5\times10^{-13}$ . The current electron measurement is sensitive at about $10^{-13}$ , so the five-loop term is the last one that matters — the six-loop piece would sit near $10^{-15}$ , below reach. This is why Kinoshita's completion of the five-loop coefficient, from 12,672 diagrams, was exactly the calculation the precision frontier needed.

How can measuring the electron's magnetism tell us the fine-structure constant?

Turn the logic around. The QED series $a_e=\tfrac12(\alpha/\pi)-0.328(\alpha/\pi)^2+\cdots$ expresses the anomaly in terms of $\alpha$ with fully known coefficients. If instead you measure $a_e$ precisely and trust QED, you solve that same equation for the one remaining unknown, $\alpha$ . Because $a_e$ is measured to about $10^{-13}$ and the coefficients are known to five loops, the inversion pins $\alpha^{-1}$ to roughly $2\times10^{-10}$ — among the best determinations anywhere. The quantity built to test QED becomes, granting QED, the sharpest ruler for a fundamental constant, then cross-checked against atom-recoil measurements.

Why is the hadronic contribution so hard to compute?

Because it involves quarks and gluons at low energies, where the strong coupling is large and perturbation theory fails. The hadronic vacuum-polarisation term is a quark loop dressing the photon, but at the momenta that dominate the muon anomaly the quarks are confined, so you cannot expand in the strong coupling as you do with photons. Two escape routes exist: the data-driven method uses the optical theorem to replace the uncomputable loop with the measured $e^+e^-\to$ hadrons cross section (equation 5), and the lattice method computes the correlator directly by simulating QCD on a grid. Both are formidable, and they have not fully agreed — which is why the hadronic term, not QED, limits the muon comparison.

Is the muon g − 2 discrepancy still there in 2026?

It is unresolved rather than confirmed or dead. The Fermilab measurement is superb and stable, fixed to about 0.13 part per million, so the uncertainty is now entirely theoretical and turns on the hadronic vacuum-polarisation term. The long-standing data-driven prediction sat about four standard deviations below the measurement — the famous hint — but lattice-QCD calculations came out higher, largely erasing the gap, and a new $e^+e^-\to\pi^+\pi^-$ measurement (CMD-3) disagreed with the older data the data-driven number relied on. So whether a genuine anomaly exists depends on which hadronic evaluation is correct; reconciling the data-driven and lattice inputs is the central open problem, and experiments like MUonE aim to settle it independently.

Does the anomalous moment mean the electron has internal structure?

No — and this is the beautiful part. The correction $g-2$ does not come from the electron having a size; it comes from the vacuum around a strictly point-like electron. A physical electron is perpetually surrounded by virtual photons and pairs it emits and reabsorbs, and while such a fluctuation is present its coupling to a field is slightly altered. Summing them gives $a=\alpha/2\pi+\cdots$ . The electron stays a mathematical point; the "structure" being probed is that of the quantum vacuum, not the particle. That is why the calculation succeeds to ten digits with a size-zero electron — and why a large anomaly, like the proton's, means something different: genuine constituents, a story owned by aq-pp-form-factors.

Why does the muon's two-loop coefficient differ from the electron's?

Because the muon can put a lighter lepton — the electron — inside its loops. The universal part of each coefficient is common to every lepton, but there are extra mass-dependent pieces from diagrams containing a second lepton species. For the electron no lighter charged lepton exists, and heavier ones are suppressed by $(m_e/m_\mu)^2$ , so its mass-dependent terms are tiny. For the muon a virtual electron loop is light and easily excited, enhancing its two-loop coefficient from the universal $-0.328$ to about $+1.09$ . This is the same physics as the new-physics lever: the muon is more sensitive to other particles in its loops, lighter and heavier alike.