What two-dimensional Yang–Mills teaches

Two-dimensional pure Yang–Mills turns a geometric question about a loop into a group-integration calculation. It teaches how gauge-invariant observables can be defined and computed without claiming that the same calculation constructs four-dimensional Yang–Mills.

Prerequisites: the lattice definition and Haar measure.
Outcomes: calculate one loop expectation; track its area and coupling convention; explain what is special about two dimensions.

The model and its normalization

Use the Euclidean plane, a compact connected gauge group $G$, and no matter. Choose a positive invariant metric on its Lie algebra. This choice fixes the Laplacian $\Delta_G$ and the nonnegative quadratic Casimir eigenvalues $C_2(R)$ of irreducible representations $R$.

Define the group heat kernel by

$$\partial_\tau K_\tau=\tfrac12\Delta_G K_\tau,\qquad K_{\tau\downarrow0}\,dU\Rightarrow\delta_e,\qquad \int_G K_\tau(U)dU=1.$$

Here $dU$ is normalized Haar measure and $e$ is the identity. We put the coupling into the dimensionless heat time $\tau=g_2^2 A$, where $A$ is the area enclosed by a simple loop and $g_2$ is defined in this heat-kernel convention. In two dimensions $[g_2^2]=\text{mass}^2$, so $g_2^2A$ has no units. Changing the invariant metric changes both the Casimir normalization and the coupling convention.

In the planar theory the loop holonomy has this heat-kernel law. Lévy’s arXiv:1912.06246v1, §1.4.2, Driver–Sengupta formula, and §2.1, Eq. (17) supplies the explicit construction and character conventions. This is a particular geometry: on a sphere, the complementary region also affects the loop law.

Work one expectation all the way through

The heat-kernel expansion is

$$K_\tau(U)=\sum_R d_R\chi_R(U)e^{-\tau C_2(R)/2}.$$

A normalized loop in representation $S$ is $W_S(U)=\chi_S(U)/d_S$. Character orthogonality gives

$$\langle W_S\rangle=e^{-\tau C_2(S)/2}.$$

One can see the same answer without expanding the kernel: $\chi_S$ is a Laplacian eigenfunction, and heat evolution starting at $e$ multiplies $\chi_S(e)=d_S$ by $e^{-\tau C_2(S)/2}$. Division by $d_S$ gives the normalized expectation.

For $SU(2)$ in the usual spin convention $C_2(j)=j(j+1)$, the fundamental representation has $j=1/2$. Thus

$$\langle W_{1/2}\rangle=\exp[-3g_2^2A/8].$$

This is area dependence, not a fit to synthetic data. The coefficient has units of mass squared. The assumption of a simple planar loop matters: self-intersections require more detailed loop relations, and compact surfaces impose global constraints.

Exact Casimir scaling and its distinction from center-charge classification are also discussed in the primary paper Nguyen, Tanizaki, and Ünsal, arXiv:2104.01824v2, §2. Do not transplant its representation dependence into four dimensions without a new argument.

Why the construction simplifies

In one spatial dimension, the pure gauge field has no transverse local gluon polarization. Global holonomy and external-charge observables can still carry information. “No propagating local polarization” therefore does not mean “no observables,” and area dependence does not automatically describe a four-dimensional glueball spectrum.

The heat kernel composes by convolution:

$$\int_G K_{\tau_1}(UV^{-1})K_{\tau_2}(V)\,dV =K_{\tau_1+\tau_2}(U).$$

Splitting an area into two pieces is compatible with adding heat times. This identity explains why refining a planar graph can preserve the relevant loop expectations. It is an unusually explicit consistency mechanism. In four dimensions plaquettes participate in a more intricate local field theory; the same one-area reduction is unavailable.

Exercise: geometry and units

A simple planar loop’s area doubles at fixed $g_2$, group metric, and representation. How does its expectation change? Would the same answer follow if only the loop’s perimeter doubled?

Solution. Since $\langle W_S\rangle=e^{-g_2^2AC_2(S)/2}$, doubling $A$ squares the old expectation. Doubling the perimeter does not determine the area, so it does not determine the new expectation. The answer also depends on keeping the heat-kernel coupling convention fixed.

Return to the four-dimensional question

This example provides exact holonomy laws in a continuum lower-dimensional setting. It does not establish the required interacting physical vacuum theory on $\mathbb R^4$ or its positive spectral threshold for every compact simple group. The difference is structural, not merely a larger number of lattice sites.

Continue to strong coupling, confinement and screening, or three-dimensional models.