Nonperturbative Gauge Measures: Positivity and Configuration-Space Limits

A nonperturbative gauge measure is a countably additive positive measure, or an equivalent positive state, on a specified space of generalized gauge configurations modulo a specified gauge group. A finite lattice Boltzmann weight supplies such a measure at fixed cutoff. It becomes a continuum quantum gauge theory only after uniform tightness or another compactness mechanism, convergence of renormalized observables, Euclidean axioms including reflection positivity, and reconstruction are proved.

Required background. Gauge configuration groupoids fixes the configuration object. Principal-bundle sectors fixes the sector sum and gauge group. Determinant and Pfaffian lines supplies the fermionic sign and trivialization problem. Helpful background. Gauge ensembles and renormalized observables supplies lattice probability measures. Fermion determinants, Pfaffians, and positivity distinguishes positive weights from sign-indefinite effective measures.

What a continuum construction must provide

The formal expression

$$\frac{1}{Z}e^{-S_{\mathrm{YM}}(A)}\,\mathcal DA$$

is not a measure because there is no translation-invariant infinite-dimensional Lebesgue measure $\mathcal DA$. A construction must instead specify a measurable configuration space, a sigma-algebra or observable algebra, sector and boundary data, and a positive normalized functional. A regulator-removal theorem then needs enough of the following, in the exact topology used:

1. gauge invariance or a gauge-covariant quotient construction;
2. tightness of the regulated probability laws, or compactness of all declared correlation functionals;
3. convergence and renormalization of a separating family of observables;
4. Euclidean covariance, reflection positivity, symmetry, regularity, and clustering in the limit;
5. Osterwalder–Schrader reconstruction of a positive Hilbert-space theory;
6. compatibility with determinant or Pfaffian-line data when fermions are present.

No one item implies the rest. In particular, positivity at each lattice spacing is stable under weak limits if a limit exists, but it does not prove tightness or nontriviality.

Two-dimensional heat-kernel Yang–Mills

Two-dimensional Yang–Mills provides a rigorous positive comparison model. For a compact group $G$ and a graph $\Gamma$ embedded in an oriented surface, assign $U_e\in G$ to each edge and normalized Haar measure $\mathrm dU_e$. In the trivial sector on a simply connected region, the heat-kernel lattice law has density

$$\mathrm d\mu_\Gamma(U) =\frac{1}{Z_\Gamma} \prod_{f}K_{g^2A_f}(h_f(U)) \prod_e\mathrm dU_e,$$

where $A_f$ is the face area, $h_f$ its oriented boundary holonomy, and $K_t$ the positive heat kernel on $G$. Gauge invariance follows because $h_f$ changes by conjugation and $K_t$ is central. Positivity and normalization are ordinary finite-dimensional facts.

The decisive consistency check is the semigroup identity

$$\int_G K_s(xy^{-1})K_t(yz^{-1})\,\mathrm dy =K_{s+t}(xz^{-1}).$$

Subdividing a face and integrating the new edge therefore reproduces the original density. Projective consistency constructs continuum holonomy laws rather than merely a sequence of unrelated lattices. Driver constructs the two-dimensional continuum theory, expresses Wilson expectations by heat kernels, and proves lattice convergence in Driver 1989, §§1–6, pp. 575–612. Compact surfaces and nontrivial bundles require the corresponding global constraints; the simple product formula above is not a universal sector formula.

The four-dimensional Wilson sequence

The first application is the continuum question posed by The Wilson Gauge Action and the Continuum Limit. On a finite four-dimensional lattice of spacing $a$, with $G=SU(N)$, the pure-gauge probability law is

$$\mathrm d\mu_{a,L}(U) =\frac{1}{Z_{a,L}} \exp\!\left[-\beta(a)\sum_p \left(1-\frac1N\operatorname{Re}\operatorname{tr}U_p\right)\right] \prod_e\mathrm dU_e.$$

Compactness of $G^{E}$ and positivity of the exponential make this a normalized probability measure for every finite lattice. Gauge invariance is exact. For the Wilson action, the regulated Schwinger functions satisfy physical reflection positivity under the standard reflection setup; Osterwalder and Seiler prove this finite-cutoff property in Osterwalder and Seiler 1978, pp. 440–471.

Those facts do not produce four-dimensional continuum Yang–Mills. One must tune $\beta(a)$, take $a\to0$ and $L\to\infty$, control gauge-invariant local composite fields or a separating observable algebra, prove nontrivial limits with the Euclidean axioms, and reconstruct the Lorentzian theory. Jaffe and Witten state the required four-dimensional existence and axiomatic output in Jaffe and Witten 2000, §§3–4, pp. 5–7. The finite Wilson law settles the regulator, positivity, and gauge-invariance entries; it does not settle the continuum existence entry.

With fermions, integrating Grassmann variables can introduce a determinant or Pfaffian. Even if the pure gauge factor is positive, the effective weight may be complex or sign-indefinite, and a global line trivialization may be obstructed. “Sign-free” is therefore an additional representation- and discretization-dependent theorem, not a property of gauge measures in general.

Independent checks and failure test

The two-dimensional independent check is exact subdivision invariance from the heat-kernel convolution law. The four-dimensional check is more limited: $Z_{a,L}$ is finite and positive because the integration domain is compact and the integrand is continuous and strictly positive. These checks validate different claims.

The adversarial move is to present $\mu_{a,L}$ at one finite lattice as the continuum measure. Demand a topology on continuum configurations, uniform tightness as $a\to0$, convergence of renormalized observables, and a reconstruction theorem. None follows from finite-dimensional normalization. Conversely, failure of one naive gauge-fixed density does not prove that no gauge-invariant algebraic state or alternative constructive limit can exist.

Exercises

Show that inserting one edge to split a heat-kernel face of area $s+t$ leaves the boundary-holonomy distribution unchanged after the new edge variable is integrated out.

Solution

The two new face factors have the form $K_s(xy^{-1})K_t(yz^{-1})$, where $y$ is the new edge variable and $xz^{-1}$ is the original boundary holonomy. Haar integration gives $K_{s+t}(xz^{-1})$ by the heat-kernel semigroup identity. All other face factors and edge measures are unchanged, so the marginal law equals the coarser graph law.

References

• Driver, Bruce K. “YM2: Continuum Expectations, Lattice Convergence, and Lassos.” Communications in Mathematical Physics 123 (1989): 575–616. DOI; Open PDF.
• Jaffe, Arthur, and Edward Witten. “Quantum Yang–Mills Theory.” In The Millennium Prize Problems, 129–152. Cambridge, MA: Clay Mathematics Institute and American Mathematical Society, 2006; problem description originally released 2000. Official PDF.
• Osterwalder, Konrad, and Erhard Seiler. “Gauge Field Theories on a Lattice.” Annals of Physics 110 (1978): 440–471. DOI.