Local Slices, Gauge Fixing, and Faddeev–Popov Geometry

A Coulomb-type gauge condition defines a local slice only after regularity, boundary conditions, and stabilizer zero modes are controlled. Its linearization along a gauge orbit is the Faddeev–Popov operator. Invertibility on the chosen gauge algebra licenses a local representative near one orbit; it does not license a global representative or exclude distant Gribov copies.

Required background. Gauge configuration groupoids supplies the Sobolev gauge action. Orbit strata and stabilizers identifies reducible points. The Faddeev–Popov construction supplies the formal Jacobian and ghost operator. Helpful background. Elliptic boundary problems and heat kernels supplies Fredholm inverses and boundary domains.

Coulomb slices in Sobolev connection space

Let $X$ be a closed Riemannian $d$-manifold, $P\to X$ a principal compact-$G$ bundle, and $k>d/2+1$. Fix $A_0\in\mathcal A_k(P)$. A nearby connection is $A_0+a$, with $a\in H^k\Omega^1(X,\operatorname{ad}P)$. The Coulomb slice through $A_0$ is

$$\mathcal S_{A_0,\varepsilon} =\{A_0+a:d_{A_0}^*a=0,\ \lVert a\rVert_{H^k}<\varepsilon\}.$$

The tangent orbit is $\operatorname{im}d_{A_0}$, and elliptic Hodge decomposition gives

$$H^k\Omega^1 =\operatorname{im}d_{A_0}\oplus\ker d_{A_0}^*$$

after the gauge algebra is restricted orthogonally to $\ker d_{A_0}$. The missing kernel is not negligible: $\ker d_{A_0}$ is the Lie algebra of the stabilizer. For an irreducible connection of a semisimple group, it vanishes after passing to a based gauge group or dividing out the finite center. At a reducible connection, it contains genuine zero modes.

The local slice theorem states, in one standard form, that for sufficiently small neighborhoods the map

$$\mathcal G_{k+1}\times_{\mathcal G_{A_0}} \mathcal S_{A_0,\varepsilon}\longrightarrow\mathcal A_k, \qquad [u,A]\longmapsto A^u,$$

is a diffeomorphism onto a neighborhood of the orbit. The quotient by $\mathcal G_{A_0}$ is essential. Kondracki and Rogulski prove the Sobolev Hodge decomposition and gauge slice theorem in Kondracki and Rogulski 1986, §§3.1–3.3, pp. 30–42. On a manifold with boundary, the same formula is not a theorem until absolute, relative, Dirichlet, or another elliptic boundary condition is supplied for both fields and gauge parameters.

The Faddeev–Popov differential

Define the gauge-fixing map at $A_0$ by

$$\Phi_{A_0}(A,u)=d_{A_0}^*(A^u-A_0).$$

For $u(t)=e^{t\xi}$ and $A=A_0$, differentiation gives

$$\left.\frac{\mathrm d}{\mathrm dt}\right|_{t=0} \Phi_{A_0}(A_0,e^{t\xi}) =d_{A_0}^*d_{A_0}\xi.$$

Thus the local Faddeev–Popov operator is the nonnegative elliptic operator

$$M_{A_0}=d_{A_0}^*d_{A_0}: H^{k+1}\Omega^0(\operatorname{ad}P)\longrightarrow H^{k-1}\Omega^0(\operatorname{ad}P).$$

Its kernel is exactly $\ker d_{A_0}$. If that kernel is removed by the declared quotient and $M_{A_0}$ has a bounded inverse on its orthogonal complement, the Banach inverse-function theorem supplies the local gauge transformation. Atiyah, Hitchin, and Singer carry out this mechanism for nearby self-dual connections, using the Green operator and implicit-function theorem, in Atiyah, Hitchin, and Singer 1978, §6, pp. 447–449.

This theorem explains the determinant in the formal Faddeev–Popov formula: it is the Jacobian of the orbit-to-gauge-condition map in a local chart. It does not turn the infinite-dimensional path-integral symbol into a measure, and a positive determinant in one region does not prove that the slice intersects each orbit exactly once globally.

If $X$ has a boundary, even the energy identity must be revisited. Integration by parts produces a boundary pairing between $\xi$ and the normal component of $d_A\xi$. Dirichlet gauge parameters, $\xi|_{\partial X}=0$, remove that term but also freeze boundary gauge transformations; a covariant Neumann condition retains a different set of modes. The operator, its kernel, and the residual symmetry therefore depend on one common boundary domain. Combining a field boundary condition from one problem with a ghost boundary condition from another invalidates both the Fredholm statement and its determinant interpretation.

Coulomb gauge around an irreducible connection

The first application is the local geometry behind The Faddeev–Popov Construction. Around an irreducible $A_0$ on a closed spatial manifold, use based gauge transformations so $\ker d_{A_0}=0$. Then

$$\langle\xi,M_{A_0}\xi\rangle_{L^2} =\lVert d_{A_0}\xi\rVert_{L^2}^2,$$

which proves positivity and, by elliptic Fredholm theory plus zero kernel, invertibility. A sufficiently small connection has a unique small based transformation into $d_{A_0}^*(A-A_0)=0$. “Small” refers both to the connection neighborhood and to the transformation neighborhood. A distant transformation can still take a second point of the same orbit into the same condition.

The independent check is dimensional and analytic: $d_{A_0}$ has order one, $d_{A_0}^*$ order one, and $M_{A_0}$ order two with principal symbol $|p|^2\operatorname{id}_{\mathfrak g}$, so it is elliptic. Stabilizer zero modes satisfy both sides of the energy identity with zero.

There is also a nonlinear check on the inverse-function argument. Write $u=e^\xi$ and $A=A_0+a$. Taylor expansion at $(a,\xi)=(0,0)$ gives

$$d_{A_0}^*(A^u-A_0) =M_{A_0}\xi+d_{A_0}^*a+Q(a,\xi),$$

where $Q$ has no term linear in either variable alone beyond those displayed. Applying $M_{A_0}^{-1}$ on the stabilizer-orthogonal subspace turns the gauge condition into a contraction for sufficiently small $a$ and $\xi$. The estimate fails precisely where the inverse is absent or the nonlinear remainder is no longer controlled; it therefore proves a neighborhood theorem and nothing global.

Failure test

At a reducible $A_0$, choose $0\neq\xi\in\ker d_{A_0}$. Then $M_{A_0}\xi=0$, so the derivative required by the inverse-function theorem is not an isomorphism. Applying the theorem anyway is the adversarial error: the proposed chart counts stabilizer directions as distinct gauge motions even though they fix $A_0$. Projecting away the kernel can restore a slice modulo the stabilizer, but it cannot restore a free-action chart.

The converse also fails: invertibility of $M_{A_0}$ proves local transversality, not global uniqueness. Global topology is invisible to this differential at one point.

Exercises

Prove that $M_A=d_A^*d_A$ has zero kernel exactly when the infinitesimal stabilizer is zero, on a closed manifold.

Solution

If $d_A\xi=0$, then $M_A\xi=0$. Conversely, if $M_A\xi=0$, integration by parts gives $0=\langle\xi,M_A\xi\rangle=\lVert d_A\xi\rVert^2$, hence $d_A\xi=0$. The closed-manifold hypothesis removes boundary terms; with a boundary, the same conclusion requires a compatible self-adjoint boundary condition.

References

• Atiyah, Michael F., Nigel J. Hitchin, and Isadore M. Singer. “Self-Duality in Four-Dimensional Riemannian Geometry.” Proceedings of the Royal Society of London A 362 (1978): 425–461. 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.