Renormalisation & the running of alpha

The fine-structure constant $\alpha \approx 1/137$ is the most famous pure number in physics, and almost everything a first course says about it is a half-truth. It is not a constant: it is the low-energy value of a quantity that grows as you probe a charge harder — $\alpha^{-1}$ falls to about $128.95$ at the mass of the $Z$ boson. This article is where quantum electrodynamics stops being a tree-level Feynman-rules exercise and becomes a quantum field theory: the vertex coupling $e$ that the rules treat as a fixed input is really a bare parameter, dressed by an infinite cloud of virtual pairs that no measurement can switch off. Two ideas organize everything that follows. The first is renormalisation — the discovery that the ultraviolet divergences generated by loop corrections are not a fatal disease but a bookkeeping artifact: they can be absorbed, once and for all, into a redefinition of the charge and the field normalisations, leaving every physical prediction finite and expressed in terms of one measured coupling. The second is the running coupling — the physical residue of that procedure. Because the dominant loop, the photon's vacuum polarisation, depends on the momentum transfer $q^2$ , the renormalised charge inherits a scale dependence: probe gently and the vacuum's screening cloud hides most of the charge; probe hard and you punch through toward the bare charge, so the effective $\alpha$ climbs. This running is measured at LEP, it dominates the uncertainty budget of electroweak precision fits, and it is the exact template for the opposite-running strong coupling that gives QCD its asymptotic freedom.

Key takeaways

The QED coupling $\alpha=e^2/4\pi$ is a bare input at tree level; loop corrections dress it and generate ultraviolet divergences that renormalisation absorbs into a single measured charge.
The dominant correction is vacuum polarisation: the exchanged photon fluctuates into virtual $f\bar f$ pairs, screening the charge like a dielectric.
Resumming those bubbles (the Dyson series) dresses the propagator into an effective coupling $\alpha_{\text{eff}}(q^2)=\alpha/[1-\hat\Pi(q^2)]$ that depends on the scale probed.
Consequently $\alpha$ runs: $\alpha^{-1}$ drops from $137.036$ at $q^2\to0$ to $\approx 128.95$ at $q^2=M_Z^2$ , a shift $\Delta\alpha\approx0.059$ confirmed at LEP.
The Ward identity ( $Z_1=Z_2$ ) means the running lives entirely in the shared photon line, so every charged particle's coupling runs identically — charge universality.
Key idea: The charge you measure is not the bare charge but one screened by a cloud of virtual pairs; how much screening you see depends on how hard you probe, so the effective coupling depends on energy.

The core insight

The charge you measure is not the bare charge but one screened by a cloud of virtual particle–antiparticle pairs; how much of that screening you penetrate depends on the momentum transfer, so the effective coupling is a function of energy, not a constant.

Where this lives today

The running of $\alpha$ is a live number in the most demanding precision physics being done. At the $Z$ pole, the electroweak fits that turned LEP data into predictions for the top-quark and Higgs masses take $\alpha(M_Z)$ as one of their three defining inputs, and its uncertainty — dominated by the hadronic $\Delta\alpha_{\text{had}}^{(5)}$ — is a leading parametric limit on how sharply the Standard Model can be tested. That same hadronic vacuum polarisation is the crux of the muon $g\!-\!2$ anomaly: the tension between the measured and predicted muon magnetic moment turns on how well $\Delta\alpha_{\text{had}}$ is known, pitting the $e^+e^-\to\text{hadrons}$ dispersion evaluation against newer lattice-QCD results. To break the impasse, CERN's MUonE experiment aims to measure $\Delta\alpha_{\text{had}}$ directly, by reading the running of $\alpha$ off the shape of elastic muon–electron scattering — a beautifully literal use of vacuum polarisation as a laboratory tool. The same one-loop machinery, extended to all three Standard-Model couplings, extrapolates them to $\sim10^{16}$ GeV to ask whether they meet — the first rung toward grand unification.

The physical picture

Place a bare electron in the vacuum and the vacuum does not stay empty. The electric field polarises the sea of virtual electron–positron pairs that continually flicker in and out of existence: each pair orients with its positive end outward, so a cloud of net negative charge migrates toward the core and partially cancels it. From far away — at low momentum transfer — you see the screened charge, and that screened value is what $1/137$ records. Push a probe in close, at high $|q^2|$ , and you penetrate the cloud, seeing more of the unscreened core: the effective charge, and with it $\alpha$ , grows. The vacuum behaves exactly like a dielectric, and renormalisation is the statement that the "true" bare charge at the centre is unobservable — only the screened, scale-dependent charge is.

Figure

The polarised vacuum screens a bare charge: virtual e⁺e⁻ pairs orient with their negative ends inward, so a low-q² probe sees the screened charge α(0) = 1/137.036 while a high-q² probe penetrates the cloud and sees a larger effective charge, α(M_Z) = 1/128.95

The consequence of that screening, made quantitative, is a coupling that slides with energy. If you plot $\alpha^{-1}$ against the logarithm of the momentum transfer, screening turns into a nearly straight descending line: flat and equal to $137.036$ while you are below the electron-pair threshold, then bending downward as each charged fermion's pair-cloud becomes penetrable in turn, until at the $Z$ mass — the scale of every LEP measurement — the coupling has strengthened to $\alpha^{-1}\approx128.95$ . The second figure is the number an electroweak fitter actually uses: the measured running of $\alpha$ from the Thomson limit to the $Z$ pole.

Figure

The measured running of the QED coupling: α⁻¹ falls from its Thomson value 137.036 at q²→0 to 128.95 at the Z mass (91.19 GeV), a total shift Δα ≈ 0.059 built from lepton and hadron loops and confirmed at LEP

The core equations

The coupling itself is the starting point. The fine-structure constant packages the electron charge into a dimensionless measure of how strongly light and matter interact; in the natural (Heaviside–Lorentz) units of field theory it is simply the squared charge over $4\pi$ .

\alpha \equiv \frac{e^2}{4\pi} = \frac{e^2}{4\pi\varepsilon_0\hbar c} \approx \frac{1}{137.035\,999} \qquad (1)

where: $e$ — renormalised electron charge; $\varepsilon_0,\hbar,c$ restore SI units; $\alpha$ — dimensionless coupling. The value $1/137$ is the low-energy one, defined in the Thomson limit $q^2\to0$ ; equation (1) fixes "the charge" only once we say at which scale it is measured. Beyond tree level the photon propagator is dressed by the vacuum polarisation. Resumming the geometric series of self-energy insertions replaces the coupling by an effective, scale-dependent one — the single most important formula in this article.

\alpha_{\text{eff}}(q^2) = \frac{\alpha}{1-\hat\Pi(q^2)} \qquad (2)

where: $\hat\Pi(q^2)$ — renormalised photon self-energy (vacuum polarisation), subtracted so that $\hat\Pi(0)=0$ ; $\alpha$ — the low-energy coupling of (1). Reading: as $\hat\Pi$ grows with $|q^2|$ the denominator shrinks below one, so $\alpha_{\text{eff}}$ rises above $\alpha$ — screening penetrated. The self-energy is computed from a single fermion loop. Regularising and subtracting at $q^2=0$ leaves a finite function; at momentum transfers far above the loop mass it collapses to a logarithm, which is the engine of the running.

\hat\Pi(q^2) = -\frac{2\alpha}{\pi}\int_0^1\! dx\, x(1-x)\,\ln\!\frac{m^2}{m^2-x(1-x)q^2}\ \xrightarrow{\,-q^2\gg m^2\,}\ \frac{\alpha}{3\pi}\!\left[\ln\frac{-q^2}{m^2}-\frac{5}{3}\right] \qquad (3)

where: $m$ — mass of the looping fermion; $x$ — Feynman parameter; $-q^2=Q^2>0$ for a spacelike probe. Reading: below threshold ( $|q^2|\ll m^2$ ) the log vanishes and $\hat\Pi\to0$ — heavy fermions decouple; above threshold each fermion adds a $\ln Q^2$ that never turns off. Feeding (3) into (2) and summing over every charged fermion light enough to be pair-produced gives the renormalisation-group solution for the running coupling — most cleanly written for the inverse coupling, in which the running is exactly linear in the logarithm.

\frac{1}{\alpha(Q^2)} = \frac{1}{\alpha(\mu^2)} - \frac{1}{3\pi}\sum_f N_c^{f}\,Q_f^2\,\ln\frac{Q^2}{\mu^2} \qquad (4)

where: $\mu$ — reference scale where $\alpha(\mu^2)$ is known; $Q_f$ — fermion electric charge in units of $e$ ; $N_c^f$ — colour factor ( $1$ for leptons, $3$ for quarks). Reading: each active fermion steepens the descent of $\alpha^{-1}$ by $N_c Q_f^2$ , so the hadronic sector matters despite fractional charges. Differentiating (4) gives the QED beta function — the rate of running. Its sign is the whole story: positive, so the coupling grows in the ultraviolet.

\beta(\alpha) \equiv \frac{d\alpha}{d\ln\mu^2} = \frac{\alpha^2}{3\pi}\sum_f N_c^{f}\,Q_f^2 \;>\;0 \qquad (5)

where: the sum runs over fermions with $m_f\lesssim\mu$ . Reading: $\beta>0$ is QED screening in one symbol — probe harder, see a larger charge. In QCD the gluon self-coupling flips this sign, giving asymptotic freedom. Finally, the experimentally quoted result packages the total running from zero momentum to the $Z$ mass into a single hadronic-plus-leptonic shift $\Delta\alpha$ , the quantity precision electroweak fits actually input.

\alpha(M_Z^2) = \frac{\alpha(0)}{1-\Delta\alpha(M_Z^2)},\qquad \Delta\alpha \approx 0.0590 \;\Rightarrow\; \alpha^{-1}(M_Z^2)\approx 128.95 \qquad (6)

where: $\Delta\alpha=\Delta\alpha_{\text{lep}}+\Delta\alpha_{\text{had}}^{(5)}+\Delta\alpha_{\text{top}}\approx 0.0315+0.0277-0.0001$ . Reading: leptons contribute a calculable $0.0315$ ; the five-quark hadronic piece $0.0277$ is extracted from data and carries essentially all the uncertainty in $\alpha(M_Z)$ .

Derivation AHow vacuum polarisation dresses the photon into a running charge

The goal is to show how an infinite chain of vacuum-polarisation bubbles inserted into the bare photon line resums into a geometric series, turning the tree-level propagator into a dressed one whose residue defines an effective, momentum-dependent charge — and then to see, through the Ward identity, why this one self-energy is the sole divergence that renormalises the coupling, making the running universal across every charged particle.

Step 1. Compute the one-particle-irreducible photon self-energy $i\Pi^{\mu\nu}(q)$ . Gauge invariance (the Ward identity $q_\mu\Pi^{\mu\nu}=0$ ) forces the transverse structure $\Pi^{\mu\nu}(q)=(q^2 g^{\mu\nu}-q^\mu q^\nu)\,\Pi(q^2)$ , so a single scalar $\Pi(q^2)$ carries all the physics and the $q^\mu q^\nu$ piece vanishes between conserved currents. Why: transversality is what keeps the photon massless — no mass term can be generated at $q^2=0$ — so the correction can only rescale the coupling.

Step 2. Insert the bubble repeatedly into the photon line and sum. The tree propagator $-ig_{\mu\nu}/q^2$ becomes $-ig_{\mu\nu}/q^2\,[1+\Pi+\Pi^2+\cdots]=-ig_{\mu\nu}/[q^2(1-\Pi(q^2))]$ . Why: this Dyson resummation converts an infinite perturbative series into a closed form; the pole stays at $q^2=0$ (the photon is still massless) but the residue is rescaled by $1/(1-\Pi)$ — the charge, not the mass, is what moves.

Step 3. Each end of the propagator carries a bare vertex $e_0$ , so any amplitude scales as $e_0^2/(1-\Pi(q^2))$ . Renormalise by subtracting $\Pi$ at a reference scale (choose $q^2=0$ , so $\hat\Pi(0)=0$ ) and identify the measured low-energy coupling with $\alpha$ ; the observable combination is then $\alpha_{\text{eff}}(q^2)=\alpha/[1-\hat\Pi(q^2)]$ , equation (2). Why: it is the product of charge-squared and propagator residue that any experiment reads out, so the scale dependence is a genuine property of the effective charge, not of an arbitrary field normalisation.

Step 4. Track the renormalisation constants. The vertex and electron self-energy corrections enter through $Z_1$ and $Z_2$ , but the Ward identity guarantees $Z_1=Z_2$ , so in $e=Z_1^{-1}Z_2\,Z_3^{1/2}\,e_0=Z_3^{1/2}\,e_0$ they cancel and only the photon-field renormalisation $Z_3$ survives. Why: this is charge universality — because the running is carried entirely by the shared photon line and not by the species-specific vertex, an electron, a muon and a quark all see their coupling run in lockstep, which is why a single $\alpha(Q^2)$ can describe them all.

Go deeper · Derivation B · 1 minThe one-loop logarithm and the running of alpha

Derivation BThe one-loop logarithm and the running of alpha

The aim is to extract, from the fermion loop, the explicit logarithm that makes the coupling run, promote it to a sum over all charged fermions, and differentiate the result into the QED beta function — establishing both the running law (4) and the positive sign (5) that marks QED as a screening theory whose coupling strengthens at short distance.

Step 1. Evaluate the loop with a Feynman parameter and regularise. Subtracting the value at $q^2=0$ (so the renormalised coupling is defined to be the Thomson-limit charge $\alpha(0)$ ) leaves the finite $\hat\Pi(q^2)=-(2\alpha/\pi)\int_0^1 dx\,x(1-x)\ln[m^2/(m^2-x(1-x)q^2)]$ . Why: subtracting on-shell at $q^2=0$ throws all the cutoff dependence into the unobservable bare charge and pins the renormalised $\alpha$ to the number that Thomson scattering actually measures.

Step 2. Take the deep-Euclidean limit $-q^2=Q^2\gg m^2$ . The parametric integral is then dominated by its logarithm and collapses to $\hat\Pi\to(\alpha/3\pi)[\ln(Q^2/m^2)-5/3]$ . Why: once the probe momentum dwarfs the fermion mass the loop "forgets" the mass except inside the log, and the single large logarithm of the scale ratio is the seed of all the running — it is what never turns off as $Q$ grows.

Step 3. Substitute into $\alpha_{\text{eff}}=\alpha/(1-\hat\Pi)$ and add one independent bubble for every charged fermion above its threshold, each weighted by colour and charge-squared. Rearranging for the inverse coupling gives the linear law $1/\alpha(Q^2)=1/\alpha(\mu^2)-(1/3\pi)\sum_f N_c^f Q_f^2\ln(Q^2/\mu^2)$ , equation (4). Why: distinct fermion species contribute additively at one loop, so the single-species log becomes a sum in which up-quarks ( $Q=+2/3$ , three colours), down-quarks ( $Q=-1/3$ , three colours) and leptons ( $|Q|=1$ ) all pull their weighted share.

Step 4. Differentiate with respect to $\ln\mu^2$ to read off $\beta(\alpha)=d\alpha/d\ln\mu^2=(\alpha^2/3\pi)\sum_f N_c^f Q_f^2$ , equation (5), which is manifestly positive. Why: the positive sign is screening expressed as a differential equation — raising the scale raises the coupling — and it is the exact opposite of the negative QCD beta function, so the two theories run in opposite directions from the same machinery.

Go deeper · Derivation C · 1 minThe Landau pole and why QED screens while QCD anti-screens

Derivation CThe Landau pole and why QED screens while QCD anti-screens

The purpose here is to integrate the one-loop renormalisation-group equation to a closed-form running law, expose the Landau pole where perturbative QED formally breaks down, estimate how absurdly high that scale sits, and contrast the sign of the QED beta function with the non-Abelian QCD case that runs the other way into asymptotic freedom.

Step 1. Write the one-loop RGE with a single effective coefficient $b_0>0$ : $d\alpha^{-1}/d\ln Q^2 = -b_0/2\pi$ with $b_0=(2/3\pi)\sum_f N_c^f Q_f^2$ absorbed, integrating to the straight line $1/\alpha(Q^2)=1/\alpha(\mu^2)-(b_0/2\pi)\ln(Q^2/\mu^2)$ . Why: a first-order ODE with a constant coefficient integrates to a line in the $(\ln Q^2,\,\alpha^{-1})$ plane — the linear descent is the visual signature of one-loop running seen in the second figure.

Step 2. Because $\alpha^{-1}$ falls with $Q$ , it must reach zero at a finite scale — the Landau pole $Q_L$ — where $\alpha$ formally diverges. For pure electron QED, setting $1/\alpha=0$ gives $\ln(Q_L^2/m_e^2)=3\pi/\alpha$ , so $Q_L=m_e\exp(3\pi/2\alpha)\approx 10^{277}\ \text{GeV}$ . Why: the same screening that strengthens $\alpha$ guarantees a formal breakdown of perturbation theory, but at an energy $258$ orders of magnitude above the Planck scale — so for real QED, embedded in the Standard Model well before then, the pole is a curiosity, not a catastrophe.

Step 3. Repeat the exercise for QCD. There the gluon self-interaction contributes a large negative term to the beta coefficient, $\beta_s=-(b_0/2\pi)\alpha_s^2$ with $b_0=11-\tfrac{2}{3}n_f>0$ , flipping the sign relative to QED. Why: the non-Abelian antiscreening of the gluon loops overwhelms the fermion screening, so $\alpha_s$ decreases toward the ultraviolet and vanishes at high energy — asymptotic freedom — instead of diverging at a Landau pole; QED and QCD run in opposite directions from structurally the same calculation.

Limiting cases

Case	Result	Meaning
Thomson limit $q^2\to0$	$\alpha^{-1}=137.036$	fully screened charge; the "textbook" fine-structure constant
Below pair threshold $\|q^2\|\ll(2m_f)^2$	fermion $f$ decouples, $\hat\Pi_f\to0$	heavy loops switch off — only fermions light enough to pair-produce run the coupling
$2m_e\ll Q\ll 2m_\mu$	only the electron loop active	slope set by a single lepton, $N_cQ_f^2=1$
$Q=M_Z=91.19$ GeV	$\alpha^{-1}\approx128.95$	all light fermions active; measured at LEP, the electroweak-fit input
Landau pole $Q_L\sim10^{277}$ GeV	$\alpha\to\infty$	perturbation theory formally breaks; far above $M_{\text{Pl}}$ , so academic for QED
QCD analogue	$\alpha_s$ runs opposite	gluon antiscreening $\Rightarrow$ asymptotic freedom, $\alpha_s\to0$ in the UV

Common trap“α = 1/137 is a fundamental constant of nature.”

It is not — or rather, $1/137$ is only the value at one particular scale. The same physical electron, probed at higher momentum transfer, behaves as though it carries a larger charge because the screening vacuum has been partially penetrated; at the $Z$ mass the effective coupling is $\alpha^{-1}\approx128.95$ , and it keeps climbing above that. Two further traps ride along. First, the running is not an artifact of regularisation or scheme choice — the difference $\alpha(M_Z)-\alpha(0)$ is a scheme-independent, directly measured quantity, confirmed at LEP; the divergences renormalisation removes are unphysical, but the scale dependence it leaves behind is entirely physical. Second, $1/137$ has no known deeper explanation — it is a measured input, not something QED predicts; only its variation with energy is predicted. So "the fine-structure constant" is shorthand for "the coupling in the Thomson limit," and quoting it without a scale is like quoting a refractive index without a wavelength.

Worked examples

1 · The running of α from the Thomson limit to the Z (fully worked)

Using the one-loop vacuum-polarisation formula, compute the leptonic shift $\Delta\alpha_{\text{lep}}(M_Z)$ from the electron, muon and tau, then combine it with the measured hadronic piece to predict $\alpha^{-1}(M_Z)$ . Take $M_Z=91.1876$ GeV, $m_e=0.511$ MeV, $m_\mu=105.66$ MeV, $m_\tau=1776.9$ MeV, and $\Delta\alpha_{\text{had}}^{(5)}=0.02766$ .

Step 1. Each lepton contributes, in the high- $Q$ limit of equation (3), a term $(\alpha/3\pi)[\ln(M_Z^2/m_\ell^2)-5/3]$ . Evaluate the logarithm-minus- $5/3$ factor for each: electron $2\ln(91188/0.511)-5/3=24.18-1.67=\mathbf{22.52}$ ; muon $2\ln(91188/105.66)-5/3=13.52-1.67=\mathbf{11.85}$ ; tau $2\ln(91188/1776.9)-5/3=7.88-1.67=\mathbf{6.21}$ .

Step 2. Sum and multiply by $\alpha/3\pi=(1/137.036)/(3\pi)=7.743\times10^{-4}$ . The three factors sum to $22.52+11.85+6.21=40.58$ , so $\Delta\alpha_{\text{lep}}=7.743\times10^{-4}\times40.58=\mathbf{0.03142}$ (the precise multi-loop value is $0.03150$ ).

Step 3. Add the sectors. The five-quark hadronic contribution is not calculable perturbatively (the loop momenta run through the non-perturbative regime) and is taken from $e^+e^-\to\text{hadrons}$ data as $\Delta\alpha_{\text{had}}^{(5)}=0.02766$ ; the top quark adds $\approx-0.00007$ . Total $\Delta\alpha=0.03142+0.02766-0.00007=\mathbf{0.05901}$ .

Step 4. Insert into equation (6): $\alpha^{-1}(M_Z)=\alpha^{-1}(0)\,(1-\Delta\alpha)=137.036\times(1-0.05901)=137.036\times0.94099=\mathbf{128.95}$ . This matches the LEP/PDG value $\alpha^{-1}(M_Z)=128.95$ , and shows that the running is roughly half leptonic, half hadronic — and that the hadronic half is the one carrying the uncertainty. α⁻¹(M_Z) = 128.95, from Δα = 0.0590.

2 · What the running does to a cross section (fully worked)

A pure-QED photon-exchange process such as $e^+e^-\to\mu^+\mu^-$ has a cross section $\sigma\propto\alpha^2$ . By how much is it enhanced if the relevant coupling is $\alpha(M_Z)$ rather than the naive $\alpha(0)=1/137$ ?

Step 1. The cross section scales as the square of the effective coupling entering the two vertices, so the ratio at the two scales is $\sigma(M_Z)/\sigma(0)=[\alpha(M_Z)/\alpha(0)]^2$ .

Step 2. Form the coupling ratio from the inverse couplings: $\alpha(M_Z)/\alpha(0)=\alpha^{-1}(0)/\alpha^{-1}(M_Z)=137.036/128.95=1.0627$ .

Step 3. Square it: $\sigma(M_Z)/\sigma(0)=(1.0627)^2=\mathbf{1.129}$ . The running strengthens the electromagnetic coupling enough to raise the photon-exchange rate by about $13\%$ at the $Z$ scale. A ~13% enhancement — far larger than experimental precision, so using the low-energy α at collider scales would be a gross error.

3 · The Landau pole of pure QED (order-of-magnitude)

At what scale would the one-loop running of the electron-only QED coupling drive $\alpha\to\infty$ ?

Step 1. Set $\alpha^{-1}(Q_L)=0$ in the single-fermion running law $\alpha^{-1}(Q)=\alpha^{-1}(m_e)-(1/3\pi)\ln(Q^2/m_e^2)$ , giving $\ln(Q_L^2/m_e^2)=3\pi/\alpha=3\pi\times137.036=\mathbf{1291.5}$ .

Step 2. So $\ln(Q_L/m_e)=645.8$ , i.e. $Q_L/m_e=e^{645.8}=10^{280.5}$ . With $m_e=5.11\times10^{-4}$ GeV, $Q_L\approx10^{280.5}\times10^{-3.3}\,\text{GeV}=\mathbf{10^{277}\ GeV}$ .

Step 3. Compare scales: the Planck mass is $\sim10^{19}$ GeV, so the QED Landau pole sits about $258$ orders of magnitude beyond where gravity and unified physics have long since rewritten the theory. Q_L ~ 10²⁷⁷ GeV — the "triviality" of QED is a formality, not a physical limit, because new physics intervenes vastly sooner.

Physics of the portrait

Renormalisation was the crisis that nearly killed quantum electrodynamics and the triumph that saved it: the recognition that the loop infinities could be absorbed into a redefinition of charge and mass — and that the leftover, the running of the coupling, was a real and measurable prediction — turned QED into the most precisely tested theory in science. No one did more to make that program rigorous and communicable than Freeman Dyson.

Origins · 1 minFreeman Dyson and how this was discovered

Freeman Dyson

1923–2020 · the resummation and renormalisation of quantum electrodynamics

Freeman Dyson was the mathematician who taught physicists what they had actually discovered. Trained in pure mathematics at Cambridge, he arrived in the United States in 1947 just as Schwinger, Feynman and Tomonaga were each, in mutually unintelligible languages, wrestling QED's infinities into finite answers. On a now-legendary bus ride across Nebraska in 1948, Dyson saw how to unify them — proving that Feynman's intuitive diagrams and Schwinger's formal operator calculus were the same theory, and that the whole apparatus was renormalisable: the infinities could be absorbed into a finite number of redefinitions, order by order, to all orders. His 1949 papers gave the field its working grammar, including the resummation now called the Dyson series — the very geometric sum of self-energy bubbles that dresses the photon propagator in this article's first derivation. It was Dyson's synthesis that made the running coupling a calculable, predictive object rather than a formal embarrassment. Characteristically, he never shared the 1965 Nobel Prize that went to the three he had unified, a limit of three laureates leaving him out. He remained, to the end, the outsider mathematician who had shown the physicists the structure hidden inside their own equations.

History

1935

— Uehling and Serber compute the vacuum-polarisation correction to the Coulomb potential

1948

— Schwinger, Feynman and Tomonaga independently render QED radiative corrections finite

1949

— Dyson unifies the three formulations and resums the propagator to all orders

1954

— Gell-Mann and Low formulate the running coupling and the renormalisation group

1997

— the TOPAZ experiment at TRISTAN measures a running α well above 1/137

Going further

Texts. Peskin & Schroeder, An Introduction to Quantum Field Theory, Ch. 7 (the vacuum polarisation, Eqs. 7.90–7.96) and §12.2 (the running coupling) · Schwartz, Quantum Field Theory and the Standard Model, Ch. 16 (vacuum polarisation) and Ch. 23 (renormalisation group) · Thomson, Modern Particle Physics, §5.3 and §10.5 (running couplings) · Halzen & Martin, Quarks and Leptons, §7.4 · Griffiths, Introduction to Elementary Particles, 2nd ed., Ch. 7.

Provenance. The tree-level vertex $q\gamma^\mu$ and the photon propagator $-g_{\mu\nu}/q^2$ are recalled by name from aq-pp-feynman-rules-qed and aq-pp-photon-propagator; the spinor structure of the fermion loop rests on aq-pp-dirac-equation. The one-loop vacuum polarisation (3), the effective coupling (2), the running law (4) and the shift $\Delta\alpha(M_Z)\approx0.059$ giving $\alpha^{-1}(M_Z)=128.95$ were re-verified numerically for this article. The opposite-sign running of the strong coupling and its interpretation are owned by aq-pp-running-alpha-s and aq-pp-asymptotic-freedom (forward references, not re-derived here).

Prerequisites. Feynman Rules for QED · The Photon Propagator · The Dirac Equation. Leads to. The Running of α_s · Asymptotic Freedom · Electroweak Radiative Corrections.

Ask the tutor: Ask me to walk through the vacuum-polarisation loop integral, or to show how the hadronic $\Delta\alpha$ feeds the muon $g\!-\!2$ puzzle.

Is the fine-structure constant 1/137 or 1/128?

Both, at different scales — that is the whole point. The number $1/137.036$ is $\alpha$ in the Thomson limit, $q^2\to0$ , which is what you measure in atomic physics and low-energy scattering where the screening cloud is seen in full. At the $Z$ mass, where LEP and LHC physics happens, the relevant coupling is $\alpha^{-1}(M_Z)\approx128.95$ , because a high-momentum probe penetrates the screening. Quoting "the fine-structure constant" without naming a scale is genuinely ambiguous above a few hundred MeV. In everyday usage $1/137$ is meant, but any calculation at collider energies must use the running value or it will be wrong by more than $10\%$ in a cross section.

What is vacuum polarisation, physically?

It is the polarisation of the quantum vacuum by an electric charge, in exact analogy with a dielectric. The vacuum is not empty: virtual electron–positron pairs (and, at higher energies, pairs of every charged fermion) continually fluctuate into existence for times allowed by the uncertainty principle. A bare charge polarises these pairs, so it is surrounded by a screening cloud of opposite charge: from far away you see the reduced, screened charge; up close you see more of the bare charge. In a Feynman diagram this is the photon momentarily becoming a fermion loop; summed to all orders it dresses the propagator and makes the coupling run.

Why does only the photon self-energy Z₃ renormalise the charge, and not the vertex?

Because of the Ward identity, which is a consequence of gauge invariance. The full charge renormalisation is $e=Z_1^{-1}Z_2 Z_3^{1/2}e_0$ , where $Z_1$ renormalises the vertex, $Z_2$ the electron field, and $Z_3$ the photon field. The Ward identity forces $Z_1=Z_2$ exactly, so those two cancel and only $Z_3$ — the photon-field, i.e. vacuum-polarisation, renormalisation — survives. This has a profound consequence: because the running is carried entirely by the shared photon line and not by anything specific to the electron vertex, every charged particle's coupling runs identically. That universality is why we can speak of a single running $\alpha(Q^2)$ rather than a different coupling for each species.

Does the Landau pole mean QED is inconsistent?

Formally, yes — the one-loop running drives $\alpha\to\infty$ at a finite scale, and this "triviality" suggests pure QED has no consistent continuum limit as a standalone theory. But the pole sits at roughly $10^{277}$ GeV, some $258$ orders of magnitude above the Planck scale. In the real world QED is embedded in the electroweak theory above $\sim100$ GeV and unified with the other forces far below the pole, so the divergence is never reached. The Landau pole is a genuine statement about QED-in-isolation as a mathematical object, but it has no bearing on any physical prediction.

How is α(M_Z) actually measured, if you can't calculate the hadronic part?

The leptonic part of the running is calculated from first principles (Worked Example 1 gives $\Delta\alpha_{\text{lep}}=0.0315$ ). The hadronic part cannot be computed perturbatively, because the fermion loops run through the strongly coupled low-energy region of QCD — so it is extracted from data. Using a dispersion relation, one integrates the measured cross section for $e^+e^-\to\text{hadrons}$ over energy to reconstruct the hadronic vacuum polarisation, obtaining $\Delta\alpha_{\text{had}}^{(5)}\approx0.0277$ ; lattice QCD now gives an independent evaluation, and MUonE aims to measure it directly from muon–electron scattering. Combining the pieces gives $\alpha^{-1}(M_Z)=128.95$ , with essentially all the uncertainty in the hadronic term.

Why does the strong coupling run the opposite way?

Because gluons carry colour charge and interact with each other, whereas photons are electrically neutral. In QED only fermion loops dress the photon, and they screen the charge, so $\alpha$ grows toward high energy. In QCD there are additional gluon self-interaction loops that antiscreen — they spread the colour charge outward rather than shielding it — and this effect dominates the quark screening as long as there are fewer than $17$ quark flavours. The net beta-function sign flips, so $\alpha_s$ decreases at high energy (asymptotic freedom, letting quarks behave almost freely inside hadrons) and grows at low energy (confinement). It is the same one-loop calculation as QED, with the extra non-Abelian diagrams turning the answer inside out.