Running of alpha_s

The strong coupling $\alpha_S$ is not a number you can look up once and reuse. Ask "how strong is the strong force?" and the only honest answer is another question: at what energy? At the scale of a proton's binding, $\alpha_S\sim1$ and quarks are locked into hadrons so tightly that perturbation theory — the workhorse of QED — simply diverges. Yet at the scale of a hard collision at the LHC, the same coupling has fallen to $\alpha_S\approx0.09$ , small enough that quarks and gluons behave almost as free particles and the parton picture works beautifully. One coupling, two worlds, and a single controlled slide between them: this is the running of $\alpha_S$ . The machinery behind the slide is renormalisation. A bare vertex is dressed by an infinite tower of loop corrections to the gluon propagator, and those loops are individually divergent. But the divergences can be absorbed into the definition of the coupling, leaving a finite, calculable dependence on the momentum transfer $q^2$ . In QED the loops are electron–positron pairs that screen charge, so the coupling grows with $q^2$ — confirmed at LEP, where $\alpha$ climbs from $1/137$ at rest to $1/127$ near $200$ GeV. QCD copies the calculation but adds something QED cannot have: gluons carry colour, so gluons loop on gluons. These non-Abelian loops enter with the opposite sign, and for the real world they win. The beta-function coefficient $B=(11N_c-2N_f)/12\pi$ is positive, $\alpha_S$ falls with energy, and the strong force becomes weak at short distance. That single sign flip — antiscreening beating screening — is asymptotic freedom, and it is why QCD is a theory we can calculate with at all.

Key takeaways

The QCD coupling is not constant: $\alpha_S=\alpha_S(q^2)$ depends on the momentum scale of the probe, because loop corrections to the gluon propagator are absorbed into its definition via renormalisation.
Fermion (quark) loops screen colour charge and drive the coupling up with $q^2$ , exactly as in QED; gluon self-interaction loops antiscreen and drive it down.
The one-loop beta coefficient is $B=(11N_c-2N_f)/12\pi=(33-2N_f)/12\pi$ , positive whenever $N_f<\tfrac{11}{2}N_c=16.5$ — so with six quarks $\alpha_S$ decreases at high energy: asymptotic freedom.
The running is fixed by one scale: $\alpha_S(q^2)=1/[B\ln(q^2/\Lambda^2)]$ , with $\Lambda_{\text{QCD}}\sim0.2$ GeV marking where perturbation theory breaks down and confinement sets in.
Independent measurements — $\tau$ decays, quarkonia, deep inelastic scattering, $e^+e^-$ jet rates — trace $\alpha_S$ from $\sim0.3$ at $2$ GeV down to the world average $\alpha_S(m_Z^2)=0.1180\pm0.0009$ .
Key idea: self-energy loop corrections make the coupling depend on the probing energy scale, and in QCD the gluon self-interaction flips the sign relative to QED, so the strong coupling weakens at short distances — asymptotic freedom.

The core insight

Self-energy loop corrections make the coupling depend on the probing energy scale; in QCD, gluon self-interactions flip the sign relative to QED, so the strong coupling weakens at short distances — asymptotic freedom.

Where this lives today

The running coupling is not a historical curiosity; it is a working input to every prediction at a hadron collider. Cross sections at the LHC are expansions in $\alpha_S(q^2)$ evaluated at the scale of the hard process, so an error in $\alpha_S$ propagates directly into Higgs-production rates, top-quark measurements, and searches for new physics — which is why $\alpha_S(m_Z^2)$ is measured to sub-percent precision and remains one of the more contested numbers in the Standard Model (lattice determinations and collider fits differ at the edge of their error bars). The running itself is verified as a curve, not a point: the coupling extracted from hadronic $\tau$ decays at $1.8$ GeV, from bottomonium spectroscopy, from scaling violations in deep inelastic scattering, and from three-jet rates and event shapes in $e^+e^-$ annihilation all fall on the single predicted line across two orders of magnitude in energy — arguably the most stringent test of a quantum field theory's dynamics ever made. At the low-energy end, where $\alpha_S$ diverges and perturbation theory dies, lattice QCD takes over to compute hadron masses and the confining potential from first principles, while heavy-ion experiments at RHIC and the LHC melt hadrons into a quark–gluon plasma — matter in the asymptotically-freer regime that filled the microsecond-old universe. And model-builders reach for the other sign: theories with many fermions can sit in a conformal window with an infrared fixed point, a scale-invariant strong dynamics explored as a route to composite Higgs and dark-sector models.

The physical picture

A gluon does not travel from one quark to another as a bare quantum. In flight it fluctuates: it splits into a virtual quark–antiquark pair and reannihilates, or — and this is the piece with no QED analogue — it splits into two gluons and rejoins. Each fluctuation is a correction to the gluon propagator, and each carries a divergent loop integral. Renormalisation trades those divergences for a coupling that depends on how hard you look. The first figure is the whole mechanism in two diagrams: the quark loop that screens (as in QED) and the gluon loop that antiscreens (unique to a non-Abelian theory), together fixing the sign of the beta function.

Figure

The two one-loop corrections to the gluon propagator. The quark loop screens colour charge and pushes the coupling up (∝ −2N_f); the gluon self-coupling loop antiscreens and pushes it down (∝ +11N_c). Their sum B = (11N_c − 2N_f)/12π is positive for N_c = 3 and N_f ≤ 16, so α_s falls with energy

The two diagrams pull in opposite directions, and which one wins is not a matter of taste but of counting: $11N_c$ against $2N_f$ . For the physical values $N_c=3$ and any $N_f\le16$ , the gluon term dominates, the net coefficient is positive, and the coupling weakens as $q^2$ grows. The second figure shows what that verdict looks like when you actually measure it — the coupling assembled from a dozen kinds of experiment, sliding down a single curve from the few-GeV region to the electroweak scale and pinned at the $Z$ mass to the world-average number.

Figure

The measured running of α_s from the one-loop formula anchored to the world average α_s(m_Z) = 0.1180. The coupling falls from about 0.26 near 2 GeV to about 0.106 at 200 GeV; the amber point marks α_s(m_Z² ) = 0.1180 ± 0.0009 at the Z mass 91.2 GeV

The core equations

The engine of the running is the renormalisation-group equation for the coupling: as you change the scale $\mu$ at which the coupling is defined, it must change in a way that leaves physical predictions untouched. That rate of change is the beta function, and at one loop it is a single quadratic term.

\mu^2\frac{\mathrm{d}\alpha_S}{\mathrm{d}\mu^2}=\beta(\alpha_S)=-B\,\alpha_S^2+O(\alpha_S^3),\qquad B=\frac{11N_c-2N_f}{12\pi} \qquad (1)

where: $\alpha_S=g_S^2/4\pi$ — strong coupling; $\mu$ — renormalisation scale; $B$ — one-loop beta coefficient; $N_c=3$ — number of colours; $N_f$ — number of active quark flavours at the scale $\mu$ . The minus sign in front of $B>0$ is the whole story: a negative beta function means the coupling shrinks as the scale grows. Integrating equation (1) between a reference scale $\mu^2$ and the probe scale $q^2$ gives the coupling at any scale in terms of its measured value at one scale. This is the formula every experiment fits.

\alpha_S(q^2)=\frac{\alpha_S(\mu^2)}{1+B\,\alpha_S(\mu^2)\ln(q^2/\mu^2)} \qquad (2)

where: $\alpha_S(\mu^2)$ — coupling at the reference scale (usually $\mu=m_Z$ ); $q^2$ — momentum transfer squared of the process; the logarithm makes the running slow — you need decades in $q^2$ to move $\alpha_S$ appreciably. Because $B>0$ , the denominator grows with $q^2$ and $\alpha_S(q^2)$ falls. The QED coupling obeys the same form with the opposite sign, which is the sharpest way to see what non-Abelian gauge theory buys you. Here only fermion loops contribute, and they enter with a minus sign in the denominator.

\alpha(q^2)=\frac{\alpha(\mu^2)}{1-\dfrac{\alpha(\mu^2)}{3\pi}\ln(q^2/\mu^2)} \qquad (3)

where: $\alpha\approx1/137$ at $q^2\to0$ ; the coefficient $+1/3\pi$ (per unit-charge fermion) sits with a minus sign in the denominator, so QED's coupling grows with $q^2$ — the vacuum screens charge. QCD's extra gluon loops flip that minus to a plus in (2). The running can be repackaged so that the one free constant is a scale rather than a coupling. Absorbing $\alpha_S(\mu^2)$ into a single mass $\Lambda_{\text{QCD}}$ turns the strength of the strong force into the location where it blows up — this is dimensional transmutation.

\alpha_S(q^2)=\frac{1}{B\,\ln(q^2/\Lambda^2)}=\frac{12\pi}{(33-2N_f)\,\ln(q^2/\Lambda^2)} \qquad (4)

where: $\Lambda\equiv\Lambda_{\text{QCD}}$ — the scale where the one-loop coupling formally diverges; empirically $\Lambda^{(5)}_{\overline{\text{MS}}}\approx0.21$ GeV (a one-loop fit anchored at $m_Z$ gives $\approx0.09$ GeV). For $q\gg\Lambda$ the coupling is small and perturbation theory works; as $q\to\Lambda$ it diverges and confinement takes over. Finally, the coefficient $B$ itself comes from group theory. In the general form it is built from the Casimir invariants of the colour group, and the sign of the whole thing is a competition between the adjoint (gluon) and fundamental (quark) representations.

B=\frac{1}{4\pi}\left(\tfrac{11}{3}C_A-\tfrac{4}{3}T_F N_f\right),\qquad C_A=N_c=3,\ \ T_F=\tfrac12 \qquad (5)

where: $C_A=N_c$ — adjoint Casimir, the strength of the gluon self-coupling; $T_F=\tfrac12$ — the quark colour normalisation; the gluon term $\tfrac{11}{3}C_A$ antiscreens, the quark term $\tfrac43 T_F N_f$ screens. Asymptotic freedom survives as long as $\tfrac{11}{3}C_A>\tfrac43 T_F N_f$ , i.e. $N_f<\tfrac{11}{2}N_c=16.5$ .

Derivation AIntegrating the one-loop beta function into the running coupling

The goal is to turn the differential statement of the renormalisation group — that the coupling changes with scale at a rate $-B\alpha_S^2$ — into the closed-form running law of equation (2), the formula that every measurement of $\alpha_S$ is ultimately compared against. We start from the one-loop beta function, separate variables, integrate between two scales, and invert. The only physics input is the sign and size of $B$ ; everything else is calculus, and the payoff is that a single measured value of the coupling determines it at every accessible energy.

Step 1. Write the one-loop renormalisation-group equation as $\mathrm{d}\alpha_S/\mathrm{d}L=-B\alpha_S^2$ , where $L\equiv\ln(\mu^2)$ is the logarithm of the scale. Why: the coupling only ever depends on the scale through its logarithm, so trading $\mu^2$ for $L$ makes the equation a clean autonomous first-order ODE.

Step 2. Separate variables: $-\alpha_S^{-2}\,\mathrm{d}\alpha_S=B\,\mathrm{d}L$ , which integrates immediately to $\alpha_S^{-1}(q^2)-\alpha_S^{-1}(\mu^2)=B\ln(q^2/\mu^2)$ . Why: the quadratic beta function is exactly the case whose reciprocal is linear in the log of the scale, which is why the inverse coupling — not the coupling — runs as a straight line.

Step 3. Read the result as a statement about $1/\alpha_S$ : the inverse coupling increases linearly with $\ln q^2$ at slope $B$ . Since $B>0$ , $1/\alpha_S$ grows and therefore $\alpha_S$ shrinks as $q^2$ rises. Why: this is the cleanest form of asymptotic freedom — plotted against $\ln q^2$ , the data for $1/\alpha_S$ fall on a rising straight line whose slope measures $B$ and hence counts colours and flavours.

Step 4. Invert to isolate $\alpha_S(q^2)$ : multiply through by $\alpha_S(\mu^2)$ and rearrange to obtain $\alpha_S(q^2)=\alpha_S(\mu^2)/[1+B\alpha_S(\mu^2)\ln(q^2/\mu^2)]$ , equation (2). Why: this is the experimentally usable form — feed in one anchor value $\alpha_S(m_Z^2)=0.118$ and it predicts the coupling at $2$ GeV or $2$ TeV with no further input. *Verified: reproduces $\alpha_S(m_Z^2)=0.118\Rightarrow\alpha_S(10\,\text{GeV})=0.173$ and $\alpha_S(200\,\text{GeV})=0.106$ with $B=23/12\pi$ , $N_f=5$ .*

Go deeper · Derivation B · 1 minDimensional transmutation and the meaning of Λ_QCD

Derivation BDimensional transmutation and the meaning of Λ_QCD

The aim here is to show that the running coupling secretly contains no dimensionless free parameter at all: the apparent freedom to choose $\alpha_S(\mu^2)$ is really just the freedom to choose a single mass scale $\Lambda$ , below which the strong force becomes strong. A classically scale-invariant theory (massless quarks and gluons have no intrinsic scale) thus generates a scale purely from quantum corrections — this is dimensional transmutation, and $\Lambda_{\text{QCD}}$ is the mass it produces. We rewrite the integrated running law in terms of $\Lambda$ and read off where perturbation theory must fail.

Step 1. Return to $1/\alpha_S(q^2)=1/\alpha_S(\mu^2)+B\ln(q^2/\mu^2)$ and define $\Lambda$ by demanding $1/\alpha_S(\mu^2)=B\ln(\mu^2/\Lambda^2)$ , i.e. $\Lambda^2=\mu^2\exp[-1/(B\alpha_S(\mu^2))]$ . Why: this absorbs the entire reference value into one scale — instead of quoting a coupling at a chosen $\mu$ , you quote the single scale $\Lambda$ that says everything.

Step 2. Substitute back: the $\mu$ -dependence cancels and $1/\alpha_S(q^2)=B\ln(\mu^2/\Lambda^2)+B\ln(q^2/\mu^2)=B\ln(q^2/\Lambda^2)$ , giving $\alpha_S(q^2)=1/[B\ln(q^2/\Lambda^2)]$ , equation (4). Why: the cancellation of $\mu$ is the check that $\Lambda$ is physical — the answer cannot depend on the arbitrary scale at which we chose to define the coupling.

Step 3. Read the limits. As $q^2\to\infty$ , $\ln(q^2/\Lambda^2)\to\infty$ and $\alpha_S\to0$ (asymptotic freedom); as $q^2\to\Lambda^2$ , the logarithm vanishes and $\alpha_S\to\infty$ (the infrared Landau pole). Why: $\Lambda$ is literally where the one-loop coupling diverges — the border between the perturbative quark–gluon world and the confining hadronic world sits at a few hundred MeV, the proton's mass scale, generated from nothing but the beta function.

Step 4. Put in numbers: anchoring at $\mu=m_Z=91.19$ GeV with $\alpha_S=0.118$ and $B=23/12\pi$ gives $1/(B\alpha_S)=13.9$ , so $\Lambda^{(5)}=m_Z e^{-13.9/2}\approx0.09$ GeV at one loop. Why: the one-loop value is only indicative — higher orders and flavour thresholds raise it to the standard $\Lambda^{(5)}_{\overline{\text{MS}}}\approx0.21$ GeV — but it correctly lands at the hadronic scale, hundreds of MeV, not GeV or MeV. *Verified: $\mu$ -independence of $\Lambda$ follows from the cancellation in Step 2; one-loop $\Lambda^{(5)}=0.088$ GeV from the anchor values above.*

Go deeper · Derivation C · 2 minWhy the coefficient is 11N_c minus 2N_f

Derivation CWhy the coefficient is 11N_c minus 2N_f

The purpose of this derivation is to make the sign of the beta function intuitive rather than merely quoted: to explain why quark loops screen (as in QED) while gluon loops antiscreen, and why the competition comes out as $11N_c$ versus $2N_f$ . We cannot do the full loop integrals here — they are owned by the renormalisation and QCD-vertex articles — but we can assemble the coefficient from the two physical mechanisms and check the threshold it predicts. The result is the single inequality that decides whether a gauge theory is asymptotically free.

Step 1. Treat the quark loop first. A gluon fluctuates into a virtual $q\bar q$ pair; the pair is a colour dipole that lines up to oppose the source, spreading positive colour charge outward and reducing what a distant probe sees — vacuum polarisation, exactly QED's screening. Each flavour contributes a fixed negative amount, $\propto-\tfrac43 T_F$ per flavour, giving the $-2N_f$ piece for $T_F=\tfrac12$ . Why: quarks live in the fundamental representation and behave just like QED's electrons, so their contribution can only screen and can only push the coupling up with energy.

Step 2. Now the gluon loop, which has no QED counterpart because photons are neutral. Gluons carry colour, so a virtual gluon pair is itself charged and does not simply screen: the dominant effect is that the colour charge is carried outward by the gluon cloud, so probing at short distance finds less charge — antiscreening. This adjoint contribution is large and positive, $\propto+\tfrac{11}{3}C_A$ , giving the $+11N_c$ piece. Why: the non-Abelian self-coupling is precisely what QED lacks, and it enters with the opposite sign and a bigger coefficient — the entire distinction between the two theories lives here.

Step 3. Add the two mechanisms: $B\propto\tfrac{11}{3}C_A-\tfrac43 T_F N_f=\tfrac{1}{3}(11N_c-2N_f)$ , so $B=(11N_c-2N_f)/12\pi=(33-2N_f)/12\pi$ for $N_c=3$ . Why: the sign of this single number is the sign of the beta function; antiscreening wins whenever the gluon term exceeds the quark term.

Step 4. Extract the threshold: $B>0$ requires $N_f<\tfrac{11}{2}N_c=16.5$ . With the six known flavours the world sits comfortably on the asymptotically free side, and even the maximally active $N_f=6$ leaves $B=21/12\pi>0$ . Why: had nature supplied 17 or more light flavours, the quark screening would have overwhelmed the gluons, the coupling would grow with energy like QED, and there would be no perturbative QCD — the margin is real physics, not a coincidence. *Verified: gluon term $+33/12\pi=0.875$ , quark term (N_f=5) $-10/12\pi=-0.265$ , net $B=0.610$ ; threshold $N_f<16.5$ .*

Limiting cases

Case	Result	Meaning
High energy $q^2\to\infty$	$\alpha_S(q^2)\to0$	asymptotic freedom — quarks and gluons act nearly free; perturbation theory converges
Low energy $q^2\to\Lambda^2$	$\alpha_S\to\infty$	infrared slavery — coupling diverges, confinement, only non-perturbative methods (lattice) apply
No gluon loops ( $11N_c\to0$ )	$B<0$ , coupling grows	the QED-like theory — pure fermion screening, a Landau pole in the ultraviolet
Too many flavours $N_f\ge17$	$B<0$	asymptotic freedom lost — screening beats antiscreening, no perturbative short-distance limit
Flavour threshold $q\approx m_q$	$N_f\to N_f\pm1$	$B$ changes as a quark becomes active/inactive; $\alpha_S$ is continuous but its slope kinks
Conformal window (large $N_f<16.5$ )	infrared fixed point	$\alpha_S$ runs to a nonzero constant — a scale-invariant phase, relevant to beyond-Standard-Model model building

Common trapα_s falls with energy "because it is screened like QED but more strongly"

This gets the mechanism exactly backwards. In QED the vacuum screens charge, so the coupling grows with $q^2$ — more screening stripped away at short distance means a larger effective charge. If QCD merely screened harder, $\alpha_S$ would grow too, and there would be no asymptotic freedom. What actually happens is a sign reversal: the extra gluon self-interaction loops, which exist only because the theory is non-Abelian (gluons carry the colour charge they mediate), antiscreen. They spread colour charge outward so that a short-distance probe sees less of it. Antiscreening is not "stronger screening"; it is the opposite effect, and it beats the quark screening by the count $11N_c>2N_f$ . Strip the gluon self-coupling out of the calculation and QCD would run up with energy just like QED. The falling coupling is a signature of the gluons' colour charge, nothing else.

Worked examples

1 · Running α_s from the Z pole down to the b-quark scale (fully worked)

Given the world average $\alpha_S(m_Z^2)=0.118$ at $m_Z=91.19$ GeV, predict $\alpha_S$ at $q=10$ GeV — roughly the $\Upsilon$ /bottomonium scale — using the one-loop formula with $N_f=5$ active flavours.

Step 1. Fix the coefficient. With $N_c=3$ and $N_f=5$ , $B=(33-2\times5)/12\pi=23/12\pi$ . Numerically $23/(12\times3.14159)=23/37.699=\mathbf{0.6101}$ . This is the slope at which $1/\alpha_S$ rises per unit $\ln q^2$ .

Step 2. Evaluate the logarithm of the scale ratio: $\ln(q^2/m_Z^2)=\ln(10^2/91.19^2)=\ln(100/8315.6)=\ln(0.012026)=\mathbf{-4.421}$ . It is negative because we are running down in energy, where the coupling is larger.

Step 3. Assemble the denominator of equation (2): $1+B\,\alpha_S(m_Z^2)\ln(q^2/m_Z^2)=1+(0.6101)(0.118)(-4.421)=1-0.3183=\mathbf{0.6817}$ .

Step 4. Divide: $\alpha_S(10\,\text{GeV})=0.118/0.6817=\mathbf{0.173}$ . The coupling has risen from $0.118$ to $0.17$ over less than one decade downward in energy — running is slow but unmistakable, and the measured value near the $\Upsilon$ ( $\approx0.18$ ) confirms it. α_s(10 GeV) ≈ 0.173, up from 0.118 at m_Z.

2 · The verdict of the sign: does adding a fourth colour or extra flavours matter?

Compare the gluon and quark contributions to $B$ , and find how many flavours it would take to destroy asymptotic freedom.

Step 1. The two pieces of $B$ are the gluon (antiscreening) term $+11N_c/12\pi$ and the quark (screening) term $-2N_f/12\pi$ . For $N_c=3$ : gluon $=33/12\pi=\mathbf{+0.875}$ , quark (per flavour) $=-2/12\pi=-0.0531$ . The gluons contribute a large fixed positive amount; each flavour chips away a small fixed negative amount.

Step 2. For the real world with $N_f=5$ : quark total $=-10/12\pi=\mathbf{-0.265}$ , so $B=0.875-0.265=\mathbf{+0.610}$ , safely positive. Even at $N_f=6$ (all quarks active): $B=(33-12)/12\pi=21/12\pi=+0.557>0$ .

Step 3. Asymptotic freedom is lost when the quark term cancels the gluon term: $2N_f=11N_c=33$ , i.e. $N_f=\mathbf{16.5}$ . So $N_f\le16$ keeps QCD asymptotically free; $N_f\ge17$ would make it screen like QED. The Standard Model's six flavours leave a comfortable margin — antiscreening wins by more than a factor of two.

Physics of the portrait

Asymptotic freedom is the discovery that made QCD calculable — and the one that nearly everyone expected to be impossible, because every quantum field theory anyone had studied did the opposite.

Origins · 1 minasymptotic freedom and how this was discovered

asymptotic freedom

1973 · Gross, Politzer & Wilczek · Nobel Prize 2004 · the negative beta function of non-Abelian gauge theory

By 1970 the strong interaction was a scandal. Deep inelastic scattering at SLAC showed quarks behaving as nearly free particles inside the proton at high momentum transfer (Bjorken scaling), yet no one could confine them or write a theory in which the force weakened at short distance. It was widely believed — and Sidney Coleman had a student check — that no field theory could be asymptotically free; the vacuum always screens. The exception, buried in group theory, was non-Abelian gauge theory. Gerard 't Hooft had computed the crucial negative sign in 1972 but did not publish its physical meaning. In 1973 David Gross and his student Frank Wilczek at Princeton, and independently David Politzer at Harvard, calculated the Yang–Mills beta function, found it negative, and drew the consequence: the coupling of a non-Abelian theory falls to zero at high energy. That same year Fritzsch, Gell-Mann, and Leutwyler assembled these pieces into quantum chromodynamics — an $SU(3)$ colour gauge theory of quarks and gluons — and asymptotic freedom became its defining prediction. It explained Bjorken scaling at a stroke, made perturbative QCD possible, and pointed at confinement as its low-energy mirror image. The three shared the 2004 Nobel Prize. The deeper lesson is representation theory: the antiscreening lives in the adjoint (gluon) representation, absent from any Abelian theory, so asymptotic freedom is the fingerprint of gluons carrying colour.

History

1954

— Yang and Mills formulate non-Abelian gauge theory

1969

— Khriplovich computes the antiscreening sign of the Yang–Mills coupling

1973

— Gross, Wilczek, and Politzer prove QCD is asymptotically free

1973

— Fritzsch, Gell-Mann, and Leutwyler name the coloured gauge theory QCD

2004

— the Nobel Prize honours the discovery of asymptotic freedom

Going further

Texts. Thomson, Modern Particle Physics, §10.5 (running of $\alpha_S$ and asymptotic freedom), Eqs. (10.15)–(10.17), Figs. 10.10–10.14 · Halzen & Martin, Quarks and Leptons, Ch. 7 (asymptotic freedom and the beta function) · Peskin & Schroeder, An Introduction to QFT, Ch. 16–17 (non-Abelian beta function, the $11-\tfrac23 N_f$ coefficient) · Griffiths, Introduction to Elementary Particles, 2nd ed., §2.3 and Ch. 8 · Schwartz, QFT and the Standard Model, Ch. 26 (the QCD beta function and asymptotic freedom).

Provenance. The loop integrals behind the coefficient $B$ , and the renormalisation programme that absorbs the divergences, are owned by aq-pp-renormalisation-running-alpha (the QED case) and aq-pp-qcd-vertices (the three- and four-gluon couplings that supply the antiscreening loop); the diagrams themselves are aq-pp-feynman-diagrams-virtual. This article recalls those results by name and assembles them into the running law, rather than re-deriving the loops. A verifier should confirm: (1) $B=(33-2N_f)/12\pi=0.610$ for $N_f=5$ ; (2) equation (2) with the $m_Z$ anchor gives $\alpha_S(10\,\text{GeV})=0.173$ and $\alpha_S(200\,\text{GeV})=0.106$ ; (3) the threshold $N_f<16.5$ ; (4) one-loop $\Lambda^{(5)}=0.088$ GeV from the anchor values. The world average $\alpha_S(m_Z^2)=0.1180\pm0.0009$ is the PDG value.

Prerequisites. Colour and QCD · Renormalisation and the running of $\alpha$ (QED) · QCD vertices · Virtual particles and loops. Leads to. Asymptotic freedom · Colour confinement · QCD in $e^+e^-$ annihilation · Deep inelastic scattering.

Ask the tutor: Ask me to run $\alpha_S$ between any two energies you choose, or to show exactly how the $11N_c-2N_f$ coefficient falls out of the gluon and quark loops.

Is α_s a constant or not — textbooks quote "α_s ≈ 0.12" as if it were one?

It is not a constant, and the quoted $0.12$ is shorthand for $\alpha_S(m_Z^2)$ , its value at the specific scale of the $Z$ -boson mass, $91.2$ GeV. The coupling changes with the energy of the process: it is about $0.3$ at a couple of GeV, $0.12$ at the $Z$ , and around $0.09$ at a TeV. When someone quotes a single number they are quoting the value at a chosen reference scale, always $m_Z$ by convention, precisely so that different measurements can be compared. Think of $\alpha_S(m_Z^2)=0.118$ the way you would think of a car's fuel economy "at 60 mph" — a standard reference point, not a claim that the number never changes.

How can the coupling "run" — surely the strength of a force is a fixed property of nature?

The bare coupling in the Lagrangian is fixed, but it is not what you measure. Any real measurement probes the charge through a cloud of virtual particles, and how much of that cloud you penetrate depends on how hard you hit it — the momentum transfer $q^2$ . A soft probe sees the charge dressed by the full cloud; a hard probe punches closer to the core and sees a different effective charge. The "running coupling" is just this effective, scale-dependent charge, made precise by renormalisation. Nothing about nature changes with energy; what changes is how much of the vacuum's screening or antiscreening you see. The bare parameter is unobservable; the effective one is what enters predictions.

Why does adding gluon loops flip the sign, when adding more of anything usually just adds up?

Because the gluon loop is not "more of the same" — it is a physically opposite effect. Quark loops are colour dipoles that screen, pulling the effective charge down as you move away (up as you probe closer), exactly like QED. Gluon loops antiscreen: because gluons carry colour, the virtual gluon cloud transports colour charge outward, so a short-distance probe finds less charge, not more. Screening and antiscreening enter the beta function with opposite signs. Adding gluon loops therefore subtracts from the screening rather than adding to it, and because the gluon coefficient ( $11N_c$ ) is large it not only cancels the quark screening but overwhelms it, reversing the direction the coupling runs.

What exactly is Λ_QCD, and why is it around 200 MeV rather than something huge?

$\Lambda_{\text{QCD}}$ is the scale at which the one-loop running coupling formally diverges — the energy where the strong force stops being weak and confinement takes over. It is not put in by hand; it is generated by the running itself, a phenomenon called dimensional transmutation: a theory with no intrinsic mass scale (massless quarks and gluons) produces one purely through quantum corrections. Its value, a few hundred MeV, is fixed by where the coupling measured at high energy extrapolates to strong coupling, and it lands at the hadronic scale — the same few hundred MeV that sets the proton's size and mass. That is no accident: hadron masses are essentially $\Lambda_{\text{QCD}}$ dressed up, which is why almost all the mass of ordinary matter comes from QCD dynamics rather than from the tiny quark rest masses.

Does asymptotic freedom mean quarks inside a proton are actually free?

Not free, but asymptotically freer. At the high momentum transfers of deep inelastic scattering, the relevant $\alpha_S$ is small, so a quark struck hard behaves almost as if unbound for the brief, short-distance instant of the collision — this is exactly what SLAC saw as Bjorken scaling, and it is what "asymptotic" means: freedom is approached only in the limit of infinite energy or zero distance. Pull the quark away, though, and you move to larger distance and smaller $q^2$ , where $\alpha_S$ grows without bound. The force does not fall off with distance the way gravity or electromagnetism does; it stays roughly constant, so the energy needed to separate quarks grows until a new quark–antiquark pair pops out of the vacuum. Confinement and asymptotic freedom are the same running coupling read at its two ends.

Could there be a world where QCD was not asymptotically free?

Yes, and the boundary is sharp. Asymptotic freedom requires $N_f<\tfrac{11}{2}N_c=16.5$ active flavours. With more than sixteen light quark flavours the screening from quark loops would overwhelm the gluons' antiscreening, the beta function would turn positive, and the strong coupling would grow with energy like QED's — no perturbative short-distance description, and a very different early universe. Just below that boundary lies the "conformal window", where the coupling runs not to zero but to a finite infrared fixed point, giving a scale-invariant strong dynamics with no confinement. Our world, with six flavours, sits well inside the asymptotically-free region, but these alternatives are actively studied on the lattice and used to build composite-Higgs and dark-sector theories — the sign of the beta function is a dial that model-builders deliberately turn.