The Evolution of α_s · Elementary Particle Physics

The theory becomes easy exactly where experiments are hard, and impossible at the energies where all ordinary matter sits. Asymptotic freedom is a convenience and a bill.

🎯 Why this matters

It is why every calculation in the remaining chapters is done at high energy, and why the mass of a proton needs a computer rather than a formula. The regime that matters most is the one perturbation theory cannot reach.

§5.8 ran the vacuum-polarization argument for QED and found α growing as you look closer. This section runs the identical argument for QCD.

Everything is the same — the bare coupling, the loop expansion, the geometric series, the straight line in $\ln Q^2$ — except that the diagrams include gluon loops as well as quark loops, and bosonic and fermionic loops carry opposite signs. That one fact reverses every conclusion.

The sign flip

Fig. 6.22 — the QCD vertex expansion has a diagram QED does not

Click a vertex or an internal line.

Compare Fig. 5.28. The structure is identical and one bubble is new. Quark loops screen the colour charge, exactly as in QED; gluon loops smear it — antiscreening — and the two have opposite signs.

\htmlClass{t-inv}{\alpha_s^{-1}(|Q|^2)} = \alpha_s^{-1}(\mu^2) + \frac{\htmlClass{t-b}{33 - 2n_f}}{12\pi}\,\htmlClass{t-log}{\ln\frac{|Q|^2}{\mu^2}}

(6.53)

Bettini p. 250. Set beside Eq. (5.52), α⁻¹(Q²) = α⁻¹(μ²) − (z_f/3π)ln(Q²/μ²), the only difference is the sign in front and what multiplies the log.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

the running of α_s · 1/α_s is linear in ln Q², with a kink at every threshold

α_s held constant at its Z value
1/α_s(Q²), one loop
measurements

probe scale Q = 91.2 GeV

1/α_s there

8.482

n_f active

slope d(1/α_s)/d lnQ²

0.4509

contributing

e u d s μ c τ b

The slope is positive: 1/α_s rises, so α_s itself falls at short distance. The 33 in (33 − 2n_f) comes from gluons coupling to gluons, and it beats the 2n_f from quark loops — antiscreening. Same machinery as α, opposite sign, and that one sign is asymptotic freedom.

The same widget that drew α in §5.8, with theory="qcd" and nothing else changed — which is the point. 1/α_s is still a straight line in ln Q² with a kink at every flavour threshold; it just slopes the other way. Flip back to §5.8 and put the two plots side by side: QED's slope is −0.71 and QCD's is +0.61, near-equal magnitudes and opposite signs.

the two slopes, side by side

import numpy as np

print("QCD, Eq. (6.53):  d(1/alpha_s)/dlnQ^2 = +(33 - 2 n_f)/12 pi")
for nf in (3, 4, 5, 6):
    print(f"   n_f = {nf}: {(33-2*nf)/(12*np.pi):+.4f}    "
          f"(gluons {33/(12*np.pi):+.3f}, quarks {-2*nf/(12*np.pi):+.3f})")
print()
print("QED, Eq. (5.52):  d(1/alpha)/dlnQ^2 = -z_f/3 pi")
print(f"   z_f = 6.67: {-6.667/(3*np.pi):+.4f}")
print()
print("Near-equal magnitudes, opposite signs.  The entire difference is that")
print("photons carry no charge and gluons do, so QCD has a +33 that QED lacks.")
print("Take the 33 away and the bracket would be -2n_f: negative, screening,")
print("and QCD would behave like QED with no asymptotic freedom and no jets.")

prints

QCD, Eq. (6.53):  d(1/alpha_s)/dlnQ^2 = +(33 - 2 n_f)/12 pi
 n_f = 3: +0.7162    (gluons +0.875, quarks -0.159)
 n_f = 4: +0.6631    (gluons +0.875, quarks -0.212)
 n_f = 5: +0.6101    (gluons +0.875, quarks -0.265)
 n_f = 6: +0.5570    (gluons +0.875, quarks -0.318)

QED, Eq. (5.52):  d(1/alpha)/dlnQ^2 = -z_f/3 pi
 z_f = 6.67: -0.7074

Near-equal magnitudes, opposite signs.  The entire difference is that
photons carry no charge and gluons do, so QCD has a +33 that QED lacks.
Take the 33 away and the bracket would be -2n_f: negative, screening,
and QCD would behave like QED with no asymptotic freedom and no jets.

Λ_QCD, and a number that does not check out

Because the coupling grows at low energy, the natural way to write the running is to name the scale where it diverges. Eq. (6.55):

\alpha_s(Q^2) = \frac{12\pi}{\htmlClass{t-b}{(33-2n_f)}\, \htmlClass{t-log}{\ln\!\big(|Q|^2/\htmlClass{t-l}{\Lambda_{\text{QCD}}^2}\big)}}

(6.55)

The running of the strong coupling, written around the scale at which it blows up rather than around a reference measurement. One dimensionful number replaces one dimensionless one — the trade the whole section is about.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

$\Lambda_{\text{QCD}}$ is QCD’s one dimensionful parameter, and it is generated out of a dimensionless coupling — dimensional transmutation, one of the strangest things in the Standard Model. The book quotes Eq. (6.56): $\Lambda(5) \approx 215$ MeV, $\Lambda(4) \approx 300$ , $\Lambda(3) \approx 340$ .

⚠️ Λ_QCD is not a measurement, and the book’s own numbers do not close at one loop

Feed $\alpha_s(M_Z^2) = 0.1179$ into the book’s own Eq. (6.54) and you get $\Lambda(5) = 87$ MeV, not the 215 MeV of Eq. (6.56). Go the other way — put $\Lambda(5) = 215$ MeV into Eq. (6.55) — and you get $\alpha_s(M_Z^2) = 0.136$ , not 0.118.

Neither is an error. $\Lambda_{\text{QCD}}$ is scheme- and order-dependent, and (6.54)–(6.55) are the one-loop formulae while 215 MeV is the two-loop MS-bar value. The two simply are not the same quantity, and mixing them is a trap the book does not signpost.

The practical rule: quote $\alpha_s$ at a stated scale, not $\Lambda$ . The book says why in one sentence — Λ depends exponentially on α_s — and the snippet below puts a number on it: a 1 % shift in α_s moves Λ by 7 %. α_s is what is measured; Λ is a derived quantity whose value depends on how you derived it.

why Λ is a bad thing to quote

import numpy as np
MZ, aZ = 91.1876, 0.1179

def Lam(nf, mu, a):                      # Eq. (6.54), one loop
    return mu * np.exp(-6*np.pi / ((33 - 2*nf) * a))

def als(nf, Q, L):                       # Eq. (6.55), one loop
    return 12*np.pi / ((33 - 2*nf) * np.log(Q**2 / L**2))

print(f"one loop from alpha_s(M_Z) = {aZ}:  Lambda(5) = {Lam(5,MZ,aZ)*1000:.0f} MeV")
print(f"the book's Eq. (6.56) prints                     215 MeV")
print(f"and running the book's 215 MeV back up: alpha_s(M_Z) = {als(5,MZ,0.215):.4f}")
print(f"against the measured                                  {aZ}")
print()
print("Both gaps are the same thing: (6.54)-(6.55) are ONE-loop and 215 MeV is")
print("the two-loop MS-bar value.  Lambda is scheme- and order-dependent.")
print()
print("how violently Lambda responds to alpha_s:")
for d in (0.01, 0.02, 0.05):
    print(f"   a {d*100:.0f}% shift in alpha_s -> "
          f"{(Lam(5,MZ,aZ*(1+d))/Lam(5,MZ,aZ)-1)*100:+5.1f}% in Lambda")
print()
print("alpha_s across the range, from the book's own Lambda values:")
for Q, nf, L in [(1.0,3,0.340), (2.0,4,0.300), (10.0,4,0.300),
                 (MZ,5,0.215), (2000.0,5,0.215)]:
    print(f"   Q = {Q:7.1f} GeV (n_f={nf}): alpha_s = {als(nf,Q,L):.3f}")
print()
print("The book says alpha_s at 2 TeV 'is about 0.03'. Its own Eq. (6.55) gives")
print(f"{als(5,2000.0,0.215):.2f}, the accepted value is about 0.082, and the FIGURE on")
print("the same page has its axis floor at 0.05.  See the erratum on the page.")

prints

one loop from alpha_s(M_Z) = 0.1179:  Lambda(5) = 87 MeV
the book's Eq. (6.56) prints                     215 MeV
and running the book's 215 MeV back up: alpha_s(M_Z) = 0.1355
against the measured                                  0.1179

Both gaps are the same thing: (6.54)-(6.55) are ONE-loop and 215 MeV is
the two-loop MS-bar value.  Lambda is scheme- and order-dependent.

how violently Lambda responds to alpha_s:
 a 1% shift in alpha_s ->  +7.1% in Lambda
 a 2% shift in alpha_s -> +14.6% in Lambda
 a 5% shift in alpha_s -> +39.2% in Lambda

alpha_s across the range, from the book's own Lambda values:
 Q =     1.0 GeV (n_f=3): alpha_s = 0.647
 Q =     2.0 GeV (n_f=4): alpha_s = 0.397
 Q =    10.0 GeV (n_f=4): alpha_s = 0.215
 Q =    91.2 GeV (n_f=5): alpha_s = 0.135
 Q =  2000.0 GeV (n_f=5): alpha_s = 0.090

The book says alpha_s at 2 TeV 'is about 0.03'. Its own Eq. (6.55) gives
0.09, the accepted value is about 0.082, and the FIGURE on
the same page has its axis floor at 0.05.  See the erratum on the page.

Erratum — α_s at 2 TeV

The text on p. 251 states: “The value of $\alpha_s$ at a few hundred MeV (not shown in the figure) is around 10; at 2 TeV it is about 0.03.”

The first half is right — Eq. (6.55) with $\Lambda(3) = 340$ MeV gives $\alpha_s = 10$ at about 365 MeV. The second is not. Three independent checks:

The book’s own equation. Eq. (6.55) with $\Lambda(5) = 215$ MeV gives $\alpha_s(2\ \text{TeV}) = 0.090$ .
The book’s own figure, on the same page. Fig. 6.23(b) plots α_s out to 2 TeV and its vertical axis starts at 0.05. A value of 0.03 would be below the bottom of the frame. The plotted curve ends near 0.08.
The accepted value is $\alpha_s(2\ \text{TeV}) \approx 0.082$ .

It should read about 0.08. The error matters a little more than a typo would, because a reader taking 0.03 at face value would conclude α_s falls about twice as fast as it does, and the whole point of the section is the rate.

Asymptotic freedom, and its price

💡 What this really says — Wilczek’s picture, and why it forces confinement

Put a single quark in the vacuum. The cloud of virtual pairs and gluons around it antiscreens : the colour charge you measure grows the further away you look. Extrapolate and the charge — and the energy — would grow without bound.

That is impossible, so something must stop it. What stops it is another colour charge nearby whose cloud cancels the first where the two overlap. A quark and its antiquark can coexist in a finite-energy system; a lone quark cannot exist at all.

So asymptotic freedom and confinement are not two facts. They are one fact read at the two ends of the same curve:

$Q \to \infty$ , distance $\to 0$ : the clouds have not built up, α_s → 0, quarks behave as free — which is exactly the impulse approximation Feynman had to assume in §6.2 and could not justify. This section justifies it.
$Q \to \Lambda_{\text{QCD}}$ , distance ~ 1 fm: α_s → ∞, perturbation theory fails, and the field organises itself into a tube.

Note the contrast with QED’s Landau pole (§5.8). There the coupling diverges at $10^{35}$ GeV — absurdly far away, harmless, and a signal that the theory is incomplete. Here it diverges at 300 MeV — in the middle of everyday matter, and rather than breaking the theory it is the theory’s most important prediction.

pulling a pair apart · electric field vs colour field

separation r = 0.60 fm

string tension k

0.9 GeV/fm

energy in the tube, k·r

0.54 GeV

cost of a new pair

0.7 GeV

cheaper to break at

0.78 fm

Slide further. The electric field above spreads out, so its energy density falls as 1/r⁴ and the total converges — pull those charges to infinity and it costs a finite amount, which is why free electrons exist. The colour tube does not spread: the same energy density occupies a longer volume, so the cost climbs linearly and without bound. That is confinement, and it is a statement about the shape of the field, not about the strength of a force.

Figs. 6.24 and 6.25 driven by one slider. The string tension 0.9 GeV/fm is the standard lattice value; the pair cost is the mass of the lightest quark–antiquark pair the vacuum can conveniently make.

⚙️ Engineer’s bridge — the shape of the field, not the strength of the force

It is tempting to say quarks are confined because the strong force is strong. That is not it, and the widget shows why: the electric and colour panels are driven by the same slider and differ in geometry, not magnitude.

An electric field between two charges spreads. Energy density falls as $1/r^4$ , the volume grows as $r^3$ , and the integral converges — so separating two charges to infinity costs a finite amount. Make the coupling a hundred times stronger and it still converges; you would just pay a hundred times more. Strength does not confine.

A colour field does not spread. It collapses into a flux tube of fixed cross-section, so the energy density stays constant while the length grows: $V = kr$ , linear and unbounded. That is a statement about field configuration.

An engineer meets the same distinction in shielding and in transmission lines: what determines whether a field escapes is the boundary condition and the geometry, not how hard you drive it. A coaxial cable confines its field not because the voltage is high but because the return path is arranged so the field has nowhere else to go — and the energy per unit length is then constant, exactly as in the tube.

Where it breaks: a coaxial cable’s return path is imposed by construction — someone put the shield there. Nothing imposes the flux tube; it forms because the gluon field couples to itself, which is dynamics you can neither arrange nor remove. And the tube does something no cable does: stretch it far enough and it breaks by creating charge, because making a new quark–antiquark pair becomes cheaper than extending the field. A cable’s energy per unit length rises without limit; a flux tube’s is capped by pair production, and that cap is exactly why you get two hadrons instead of a free quark.

	QED (§5.8)	QCD (§6.5)
what runs in the loop	charged fermions only	quarks and gluons
sign of the slope of 1/coupling	−z_f/3π = −0.71	+(33 − 2n_f)/12π = +0.61
behaviour at short distance	coupling grows — screening	coupling falls — asymptotic freedom
where the formula diverges	Λ_EM ≈ 10³⁵ GeV, above the Planck scale	Λ_QCD ≈ 0.2–0.34 GeV, inside a proton
what the divergence means	the theory is incomplete at absurd energies — harmless	confinement — the theory's central prediction
can the charge be isolated?	yes — free electrons exist	no — the energy cost is unbounded
value at the Z	α = 1/128.9	α_s = 0.1179 ± 0.0009

Read the second row and then the fourth. The sign flip is a small change in a formula and it relocates the divergence by thirty-five orders of magnitude — from a place no experiment will ever reach to the middle of ordinary matter. That is the whole difference between a world with free charges and a world with hadrons.

🔑 If you remember only three things

One integer decides it: 33 − 2n_f. With six flavours it stays positive and the coupling falls; enough extra flavours would tip it and QCD would screen like QED.
A scale appears in a theory that had none. Nothing in the Lagrangian puts Λ there; the running generates it, and every hadron mass is measured against it.
A free quark would cost without limit. The field between two of them does not spread out, so separating them is an expense that never stops growing.

Where this goes next

§6.6 runs the same machinery on the quark masses, which are scale-dependent for the same reason.
§6.7 uses the linear potential of the widget above to get the proton’s mass and radius from almost nothing.
§6.10 is what you do when α_s is too large to expand in — and it is the source of the LO-HVP number that dominates §5.9b’s muon anomaly.
§5.8 is the QED half of this argument, and the widget above runs in both modes for the comparison.

✅ Check yourself — asymptotic freedom and confinement

0/5 answered · 0 correct

1.The QCD beta-function coefficient is 33 − 2n_f. Where does each piece come from, and what would happen without the 33?
2.Why is quoting Λ_QCD a worse way to report the strong coupling than quoting α_s at a stated scale?
3.In the widget, the electric field and the colour field are driven by the same slider. What actually differs between them?
4.Pull the pair past about 0.8 fm and the tube snaps. What has actually happened?
5.QED's coupling diverges at 10³⁵ GeV and QCD's at about 0.3 GeV. Why is one an embarrassment and the other a triumph?