Strong coupling and controlled expansions

Prerequisites: the lattice definition and Haar measure.

You will learn to calculate a leading group-integral contribution, explain how link integration selects surfaces, and distinguish a finite-regulator expansion from a continuum theorem.

Strong coupling means a small expansion parameter

For the Wilson action with $\beta=2N/g_0^2$, strong bare coupling means small $\beta$. At $\beta=0$, link variables have independent normalized Haar distributions. Expanding the Boltzmann weight around that exactly defined measure supplies a natural starting point.

This is an expansion at fixed regulator. The weak-bare-coupling region associated with the usual asymptotically free continuum trajectory lies at large Wilson $\beta$, so a small-$\beta$ result cannot be continued there merely by drawing a smooth curve.

Worked example: one isolated SU(3) plaquette

After eliminating redundant link integrations on a single open plaquette, its holonomy $U$ is Haar distributed. Let

$$w(U)=\frac13\operatorname{tr}U,\qquad Z(\beta)=\int_{SU(3)} e^{(\beta/3)\operatorname{Re}\operatorname{tr}U}\,dU.$$

The constant part of the Wilson action cancels in the normalized expectation. Character orthogonality gives

$$\int\operatorname{tr}U\,dU=0,\quad \int\operatorname{tr}U\,\operatorname{tr}U^\dagger\,dU=1,\quad \int(\operatorname{tr}U)^2\,dU=0.$$

The last equality also follows because two fundamentals of $SU(3)$ contain no singlet. Expand the numerator to first order:

$$\begin{aligned} \int w(U)e^{(\beta/3)\operatorname{Re}\operatorname{tr}U}\,dU &=\frac{\beta}{18}\int \left[(\operatorname{tr}U)^2+ \operatorname{tr}U\,\operatorname{tr}U^\dagger\right]dU +O(\beta^2)\\ &=\frac{\beta}{18}+O(\beta^2). \end{aligned}$$

Since $Z=1+O(\beta^2)$, $\langle w\rangle=\beta/18+O(\beta^2)$. This is a local group-integral calculation with a stated action, normalized trace and group. It is not a value of the continuum string tension.

For $SU(2)$ the fundamental trace is real and its square contains a singlet. Repeating the calculation gives $\beta/4+O(\beta^2)$ with $w=\operatorname{tr}U/2$. Copying the $SU(3)$ coefficient to every group would lose this representation dependence.

Why surfaces appear

Expand each plaquette factor in the product over a finite lattice. Haar integration kills terms whose representation factors on a link cannot combine to a singlet. A nontrivial loop insertion therefore needs plaquette terms arranged so that its boundary representation factors can be paired. A sheet spanning the loop is a natural lowest-order contribution.

This explains the geometric origin of an area dependence in a formal strong-coupling calculation. To prove an area-law bound, one must control all allowed contributions, their multiplicities, cancellations and volume dependence. Counting one minimal sheet alone does not bound the remainder.

On a fixed finite lattice, compactness makes the integrals well behaved for real $\beta$, and expansions near zero can be justified locally. That statement does not give a convergence region uniform in the number of plaquettes. Polymer or cluster methods organize connected contributions and can address precisely such volume dependence when their hypotheses hold.

A genuine theorem is stronger than the picture

Cao–Nissim–Sheffield, arXiv:2509.04688v2, Theorem 1.6 and Definition 1.4 gives an area-law estimate for rectangular loops in its stated gauge groups and small-coupling regime, using a dynamical method. The torus side length is $L$ in lattice sites, and loop sides are restricted to at most $L/2$.

Their action uses $N\beta_{\mathrm{CNS}}\operatorname{Re}\operatorname{Tr}U_p$ in the exponent. For $SU(N)$, matching our exponent $\beta_{\mathrm W}\operatorname{Re}\operatorname{Tr}U_p/N$ gives

$$\beta_{\mathrm W}=N^2\beta_{\mathrm{CNS}}.$$

An inequality on one beta is not an inequality on the other without this translation. See the full dossier for group restrictions and exact thresholds. The theorem does not remove the lattice cutoff.

Exercise: diagnose an invalid continuation

An area estimate is proved for $0\leq\beta<\beta_*$. A numerical plot of the same observable looks smooth up to $10\beta_*$. Does this prove the estimate in that larger interval or the existence of a continuum limit?

Solution. No. Smooth sampled values do not control a series remainder or prove a uniform inequality between samples, let alone under changing volume and cutoff. One needs an argument extending the estimate with its dependencies intact. The plot can motivate that question without settling it.

Continue toward the continuum

The foundational formulation is Wilson, Physical Review D 10, 2445–2459 (1974). Next read continuum scaling to see why the physical units and limiting trajectory must change along with the bare coupling.