Gauge Configuration Groupoids and Moduli

Gauge fields modulo gauge transformations form an action groupoid, not merely a set of orbits. Its objects are connections, its arrows are gauge transformations between connections, and its automorphism groups are the stabilizers of individual fields. The coarse quotient is often enough for gauge-invariant functions, but it discards precisely the isotropy, family, boundary, and gluing data needed near reducible configurations and across topological sectors.

Required background. Principal and associated bundles supplies the bundle action and gauge automorphisms. Connections, curvature, and the Bianchi identity supplies the affine space of connections. Gauge fields, redundancy, and observables distinguishes an equivalence from a physical symmetry. Local potentials and global configurations supplies transition functions and holonomy. Helpful background. Constraints and reduction explains why singular quotients require more than the regular-value theorem.

The action groupoid of connections

Let $P\to M$ be a fixed principal bundle for a compact Lie group $G$ over a compact $d$-manifold. Choose $k>d/2+1$. The Sobolev completion $\mathcal A_k(P)$ is an affine Hilbert space modeled on $H^k\Omega^1(M,\operatorname{ad}P)$, while the $H^{k+1}$ gauge group $\mathcal G_{k+1}(P)$ is a Hilbert Lie group whose multiplication and action are continuous. With a right-action convention,

$$A^u=u^{-1}Au+u^{-1}\mathrm du, \qquad u\in\mathcal G_{k+1}(P).$$

The action groupoid $[\mathcal A_k(P)/\mathcal G_{k+1}(P)]$ has object space $\mathcal A_k(P)$ and arrow space $\mathcal A_k(P)\times\mathcal G_{k+1}(P)$. Source and target are $s(A,u)=A$ and $t(A,u)=A^u$, with composition $(A,u)(A^u,v)=(A,uv)$. At $A$, the automorphism group is

$$\operatorname{Aut}(A)=\mathcal G_A =\{u:A^u=A\}.$$

Its Lie algebra is $\ker d_A\subset H^{k+1}\Omega^0(M,\operatorname{ad}P)$: infinitesimal stabilizers are covariantly constant adjoint sections. This is why irreducibility is a geometric hypothesis, not a stylistic preference. Atiyah and Bott formulate the connection space, gauge group, and its classifying-space topology in Atiyah and Bott 1983, §2, pp. 539–542; Kondracki and Rogulski give the Sobolev action and symmetries of connections in Kondracki and Rogulski 1986, §§1.3–2.2, pp. 11–20.

Passing to the coarse orbit space $\mathcal A_k/\mathcal G_{k+1}$ remembers whether two objects are isomorphic, but forgets how they are isomorphic. It therefore forgets $\mathcal G_A$, and it cannot represent a family whose local representatives glue by gauge transformations satisfying a cocycle law only up to specified arrows. If $M$ has a boundary, the object also depends on whether gauge transformations are unrestricted, fixed at the boundary, or fixed at a base point. Transformations excluded from the quotient may act as genuine boundary symmetries. If bundles of several isomorphism classes are admitted, the full configuration groupoid is a disjoint union over those classes; choosing one $P$ has already selected a topological sector.

Families explain why arrows are indispensable even away from a visibly singular coarse point. Given a parameter cover $\{V_i\}$, local connection families $A_i(v)$ may be related on overlaps by parameter-dependent gauges $u_{ij}(v)$. The identities $u_{ij}u_{jk}=u_{ik}$ are part of the family, and an automorphism $u_{ii}$ can vary even when the orbit-valued map is constant. A coarse map $V\to\mathcal A/\mathcal G$ cannot recover these transition functions. The groupoid is therefore the minimal quotient that supports descent; a higher or derived enhancement is needed only when one also wants higher coherences or obstruction complexes.

Circle connections: holonomy and retained isotropy

The first concrete application is the global formulation developed in Local Potentials and Global Gauge Configurations. Take the trivial $U(1)$ bundle on a circle of circumference $2\pi$ and write the real coefficient of a connection as $A=a(\theta)\,\mathrm d\theta$, suppressing the common factor of $i$ in the anti-Hermitian convention. A gauge transformation $u=e^{i\chi}$ acts by $A\mapsto A+\mathrm d\chi$, where

$$\chi(\theta+2\pi)=\chi(\theta)+2\pi n, \qquad n\in\mathbb Z.$$

The exact part of $A$ can be removed, while a winding-$n$ transformation shifts its average $\alpha=(2\pi)^{-1}\int_{S^1}A$ by $n$. Thus the coarse orbit is labeled by

$$h(A)=\exp\!\left(i\oint_{S^1}A\right)=e^{2\pi i\alpha}\in U(1).$$

Yet every $A$ is fixed by constant $U(1)$ transformations. The coarse quotient is a circle, whereas the groupoid has a $U(1)$ automorphism group over every point. This surviving isotropy matters to equivariant cohomology, state counting, and gluing, even though it does not change $h(A)$.

The calculation also separates two ideas often conflated. Winding transformations are disconnected components of the gauge group on the same trivial bundle; they identify $alpha$ and $alpha+n$. Different principal bundles would be different objects before any such quotient. On $S^1$, all principal $U(1)$ bundles are topologically trivial, so no second label occurs.

Independent check and failure boundary

There are two direct checks. First, $h(A+\mathrm d\chi)=h(A)e^{i2\pi n}=h(A)$, so holonomy descends to the orbit space. Second, $\mathrm d\chi=0$ for constant $chi$, so the stabilizer is visibly $U(1)$. Both conclusions survive every choice of $a(\theta)$.

The adversarial test is to replace the groupoid by the set of holonomy values. That set correctly decides gauge equivalence, but its points have no automorphisms; the constant gauge transformations disappear. The replacement is therefore valid only for a question known in advance to depend solely on the coarse orbit. It is not a converse theorem saying that equal gauge-invariant observables determine all stacky or family data.

Exercises

Show that every smooth $U(1)$ connection on $S^1$ is gauge-equivalent to $\alpha\,\mathrm d\theta$ with $0\leq\alpha<1$, and determine the residual gauge transformations.

Solution

Set $\alpha=(2\pi)^{-1}\int_0^{2\pi}a(\theta)\,\mathrm d\theta$ and $\chi_0(\theta)=\int_0^\theta(\alpha-a(s))\,\mathrm ds$. Its integral over a period vanishes, so $\chi_0$ is periodic and $A+\mathrm d\chi_0=\alpha\,\mathrm d\theta$. A winding-$n$ transformation shifts $\alpha$ by $n$, giving the chosen interval. Transformations preserving the representative have $\mathrm d\chi=0$ after the interval is fixed, hence form the constant $U(1)$ stabilizer.

References

• Atiyah, Michael F., and Raoul Bott. “The Yang–Mills Equations over Riemann Surfaces.” Philosophical Transactions of the Royal Society of London A 308 (1983): 523–615. DOI; Open PDF.
• Kondracki, Witold, and Jan S. Rogulski. On the Stratification of the Orbit Space for the Action of Automorphisms on Connections. Dissertationes Mathematicae 250. Warsaw: Polish Scientific Publishers, 1986. Repository record and PDF.