Colour, confinement and asymptotic freedom

The Δ⁺⁺ is uuu with spin 3/2 and no orbital angular momentum. Its space part is symmetric because the state is a ground state, its spin part is symmetric because all three spins are aligned, and its flavour part is symmetric because all three quarks are up. Three symmetric factors give a symmetric total, which the exclusion principle forbids for identical spin-½ particles. The only escape is a further label. It must carry at least three values, so that the totally antisymmetric combination εᵢⱼₖ exists and can absorb the required sign; and exactly three, because every observed hadron is a colour singlet of SU(3) — three-quark baryons, quark–antiquark mesons, nothing carrying net colour. An independent check comes from π⁰ → γγ, whose amplitude runs through a quark loop and therefore scales as N_c, making the rate scale as N_c². The measured width of 7.8 eV is nine times the 0.87 eV that a single-coloured quark would give.

In e⁺e⁻ annihilation through a virtual photon, making μ⁺μ⁻ and making a quark pair use the same QED vertex; only the fractional charge and the number of available copies differ. Once a quark pair exists it hadronises with probability one, so the hadronic cross-section is settled at the quark level and the ratio is clean: R = σ(hadrons)/σ(μ⁺μ⁻) = N_c Σ_q Q_q², summed over the flavours light enough to pair-produce at that √s. Below open charm the sum is 4/9 + 1/9 + 1/9 = 2/3, giving R = 2. Above the charm threshold add 4/9 for R = 10/3; above bottom add 1/9 for R = 11/3. The data step up at exactly those energies and sit near 2.2, 3.6 and 3.9 once the small (1 + αₛ/π) correction is included — three times what a colourless quark model predicts. No fit and no free parameter: a ratio of two cross-sections counts colours.

In QED a bare charge polarises the vacuum: virtual e⁺e⁻ pairs align and partly cancel it, so a closer probe sees more charge, and α runs from 1/137 at zero momentum to about 1/128 at the Z. QCD has the same fermion loops — n_f of them — screening in the same direction. It also has something QED does not. Gluons carry colour, so gluon loops spread the colour charge outward instead of shielding it, and a closer probe finds less of it. The one-loop beta function weighs the two effects against each other, Q dαₛ/dQ = −[(11N_c − 2n_f)/6π] αₛ², where 11N_c comes from gluon loops and 2n_f from quark loops. With N_c = 3 and n_f = 6 the bracket is 33 − 12 = 21, comfortably positive, so antiscreening wins and αₛ falls as Q rises. Seventeen flavours would be needed to flip the sign — the physical world is nowhere near that.

Integrate the beta function and anchor it on a measurement rather than on Λ: 1/αₛ(Q) = 1/αₛ(M_Z) + [(11N_c − 2n_f)/6π] ln(Q/M_Z). Starting from αₛ(M_Z) = 0.118 with n_f = 5, this gives 0.203 at 5 GeV and 0.088 at 1 TeV (0.089 if the top quark is counted above 173 GeV) — a factor of 2.3 across 2.3 decades of energy, which is how slowly a logarithm moves. What it buys is a perturbation series that works: at 1 TeV each further order costs roughly a factor of 0.088, so a few terms give a per-cent answer, and jet rates, Drell-Yan spectra and Higgs cross-sections are all obtained this way. Run the same formula backwards and it prices its own expiry. At 1 GeV it returns 0.34, at 0.5 GeV 0.47, it passes 1 near 0.2 GeV, and it diverges at Λ. Long before the divergence the series has stopped converging, because each new term is as large as the last.

Knock a quark out of a hadron and the colour field between it and the remnant does not fan out like a Coulomb field. The gluons' mutual attraction collapses the flux into a tube of roughly fixed cross-section, so the stored energy grows linearly: V(r) → k r with k ≈ 0.90 GeV per fm, a constant 1.4 × 10⁵ N — the weight of about fifteen tonnes — that never falls off with distance. After roughly 0.7 fm the tube holds enough energy to create a light quark–antiquark pair, so it breaks instead of stretching, and the process repeats down the chain. What emerges is a collimated spray of colour-singlet hadrons following the original parton's direction, with only a few hundred MeV/c of transverse momentum about it; the spread that hadronisation adds therefore narrows as 1/E. A hard radiated gluon makes a third jet at a rate proportional to αₛ, and the three-jet events seen at PETRA in 1979 were the first direct sight of the gluon.

Nothing above derives confinement. The running formula says only that the perturbative expansion fails below about 1 GeV, and the failure of a method is not a theorem about the theory. What stands in its place is evidence of two kinds. Experimentally, no free quark has ever been isolated: Millikan-style searches for fractional charge bound their abundance below roughly one per 10²⁰ nucleons. Computationally, lattice QCD places the fields on a Euclidean grid of spacing a ≈ 0.05–0.1 fm and samples the path integral numerically; Wilson loops return a static quark–antiquark potential with a linear tail of tension k ≈ 0.9 GeV/fm, and the hadron spectrum comes out right to a few per cent. Those results carry their own extrapolations — a → 0, infinite volume, physical quark masses — and the analytic statement behind them, that pure SU(3) Yang–Mills has a mass gap, remains one of the Clay Millennium problems.