Large Gauge Transformations and Topological Sectors

A gauge transformation is large only relative to a specified admissible gauge-transformation group: it lies outside that group’s identity component. This topological condition does not by itself say whether the transformation is a redundancy, a charged boundary symmetry, or a transformation represented by a phase in a quantum sector. Nor does it classify the bundle on which the gauge field lives. This page separates those questions and works them out for compact Yang–Mills theory on a three-ball. Instanton dynamics, classifications of topological actions, and global anomalies remain outside its scope.

Required background. Local Potentials and Global Gauge Configurations supplies the bundle-automorphism language used below. Homotopy, Degree, Winding, and Covering Spaces supplies homotopy classes and winding degree.

Helpful background. Global Form, Matter Representations, and the Faithful Gauge Group explains why the actual global group, rather than only its Lie algebra, is part of the answer.

Large means disconnected in the admissible group

Let $P\to\Sigma$ be a principal $G$-bundle on a spatial manifold. Fix boundary or falloff data $\mathcal B$, including the regularity required of a transformation, and define

$$\mathcal G_{\mathcal B}(P) =\operatorname{Aut}_{\mathcal B}(P).$$

On a nontrivial bundle, these automorphisms are sections of the associated group bundle $\operatorname{Ad}P$; they need not be globally expressible as maps $\Sigma\to G$. Let $\mathcal G_{\mathcal B}^{\circ}(P)$ denote the component containing the identity. The component group is

$$\pi_0\!\left(\mathcal G_{\mathcal B}(P)\right) = \frac{\mathcal G_{\mathcal B}(P)} {\mathcal G_{\mathcal B}^{\circ}(P)}.$$

A transformation $u$ is large precisely when its class in this quotient is not the identity class. The definition therefore changes if one changes $\Sigma$, $P$, the global form of $G$, or $\mathcal B$. A formula for $u(x)$ cannot be called large before those data are declared.

Connectedness and physical role are independent tests. A transformation that is nontrivial at a boundary can be continuously connected to the identity and still carry a nonzero surface charge. Conversely, a transformation that is the identity at the boundary can lie in a disconnected component and still be included among the redundancies. The former is charged but need not be large; the latter can be large but uncharged.

Transformation components are not field sectors

Three different classifications recur in this subject:

• a bundle or configuration sector specifies the global field data, such as an isomorphism class $[P]$ or a characteristic number;
• a transformation component labels a connected component of the admissible automorphism group of a fixed $P$;
• a quantum character specifies how those components act on an equivariant state.

The first two can be separated by elementary counterexamples. For compact $U(1)$ on a linking sphere, with $a=gA$ the faithfully normalized connection,

$$m =\left\langle c_1(P),[S^2]\right\rangle =\frac{1}{2\pi}\int_{S^2} da \in\mathbb Z$$

labels the bundle. An automorphism of that fixed bundle preserves $m$. Moreover, based transformations on compactified three-space have no analogous integer because $\pi_3(U(1))=0$. By contrast, the trivial $\operatorname{SU}(2)$ bundle on a three-ball has only one bundle class but, as shown next, its based transformation group has infinitely many components. Tong’s monopole patches and based Yang–Mills transformations display these two distinct integers in Tong 2018, §§ 1.1.2 and 2.2.2, pp. 6–8 and 43–46, official full-notes PDF.

Based transformations on a three-ball have integer winding

Take $\Sigma=B^3$, the trivial $\operatorname{SU}(2)$ bundle, and the based group

$$\mathcal G_{\partial} = \left\{ u:B^3\to\operatorname{SU}(2) \ \middle|\ u|_{S^2}=\mathbf 1 \right\}.$$

Collapsing the boundary to a point gives $B^3/S^2\simeq S^3$. Therefore

$$\begin{aligned} \pi_0(\mathcal G_{\partial}) &\simeq [S^3,\operatorname{SU}(2)]_* \\ &=\pi_3(\operatorname{SU}(2)) \simeq\mathbb Z. \end{aligned}$$

Choose an orientation of $B^3$ and use the fundamental trace with Hermitian generators $T_a=\sigma_a/2$, so $\operatorname{tr}(T_aT_b)=\delta_{ab}/2$. Fix the remaining sign by calling a standard degree-one map positive. Set $L_i=u^{-1}\partial_i u$. Its winding is

$$\begin{aligned} \nu(u) &=\frac{1}{24\pi^2} \int_{B^3} d^3x\,\epsilon^{ijk} \\ &\quad\times\operatorname{tr}(L_iL_jL_k) \in\mathbb Z. \end{aligned}$$

The expression is unchanged by a homotopy that keeps $u=\mathbf1$ on the boundary. Using $(uv)^{-1}d(uv)=v^{-1}(u^{-1}du)v+v^{-1}dv$, the mixed terms in the cubic form combine into an exact form. Its integral vanishes for based maps, giving

$$\nu(uv)=\nu(u)+\nu(v).$$

Thus $\nu$ identifies the component group with $\mathbb Z$. Reversing the spatial orientation, or choosing the inverse convention for transformations, reverses every displayed winding sign but changes no classification. The integrality, homotopy invariance, and product law are derived in Tong 2018, § 2.2.2, pp. 44–46, official full-notes PDF and Weinberg 1995, § 23.4, pp. 445–450.

For a compact connected simple group, the same based example gives $\pi_3(G)\simeq\mathbb Z$ after the invariant trace is normalized appropriately. A semisimple product can give several integer factors, whereas an Abelian factor contributes none to $\pi_3$. These are examples of the general component definition, not a boundary-independent formula for every gauge group.

The Chern–Simons shift produces a theta character

Continue with $\operatorname{SU}(2)$ and absorb the coupling into the Hermitian spatial connection. In differential-form notation,

$$\begin{aligned} a&=gA, & f&=da-ia\wedge a. \end{aligned}$$

Then $D=d-ia$, and a finite transformation acts by

$$a^u =uau^{-1}-i(du)u^{-1}.$$

In the trace and orientation convention above, define the normalized Chern–Simons functional

$$\begin{aligned} N_{\mathrm{CS}}[a] &=-\frac{1}{8\pi^2} \int_{B^3} \operatorname{tr}\!\Bigl( a\wedge da \\ &\qquad -\frac{2i}{3}a\wedge a\wedge a \Bigr). \end{aligned}$$

Choose the sign of $\nu$ consistently with this definition. Direct substitution then gives, for $u\in\mathcal G_{\partial}$,

$$N_{\mathrm{CS}}[a^u] =N_{\mathrm{CS}}[a]+\nu(u).$$

The general transformation law also contains a surface transgression term. It vanishes here because $u$ is constant on $S^2$. It cannot be dropped when the boundary value of $u$ varies. Tong displays the Chern–Simons functional, the discarded total derivative, and the integer shift in Tong 2018, § 2.2.2, pp. 43–46, official full-notes PDF.

The shift explains the canonical theta character. First quotient the space of connections by the identity component:

$$\widetilde{\mathcal C} =\frac{\mathcal A}{\mathcal G_{\partial}^{\circ}}.$$

The residual component group $\mathbb Z$ acts on this covering space. A theta sector can be represented by a function $\Phi_\theta$ obeying

$$\Phi_\theta[a^{u_n}] =e^{-in\theta}\Phi_\theta[a], \qquad \nu(u_n)=n.$$

Equivalently, $\Phi_\theta$ is a section of a flat line bundle over the full orbit space, with holonomy $e^{-in\theta}$. The product law for winding makes this a consistent one-dimensional representation:

$$e^{-i(n_1+n_2)\theta} =e^{-in_1\theta}e^{-in_2\theta}.$$

One convenient wavefunctional convention restores an invariant scalar by the Chern–Simons dressing

$$\Psi_\theta[a] =e^{i\theta N_{\mathrm{CS}}[a]} \Phi_\theta[a].$$

The two phases cancel under every based $u_n$. Reversing the winding convention reverses both exponents. Because the characters of this integer component group satisfy $\chi_{\theta+2\pi}=\chi_\theta$, this particular based $\operatorname{SU}(2)$ construction has $\theta\sim\theta+2\pi$. The period must be reconsidered if the global group, allowed topological charges, spin assumptions, or redundancy group changes. Weinberg derives the integer topological weighting in Weinberg 1995, §§ 23.5–23.6, pp. 450–457.

There are two related but distinct statements about an action. A three-dimensional or boundary term $S_{\mathrm{CS}}=2\pi kN_{\mathrm{CS}}$ shifts by $2\pi k\nu(u)$ for the based transformations above, so the phase $e^{iS}$ tests the coefficient and the declared redundancy group. By contrast, the four-dimensional density $\operatorname{tr}(F\wedge F)$ is itself gauge invariant. Its Chern–Simons primitive and the patching of states are where the integer shift appears; the bulk theta density does not become gauge noninvariant.

Quotient, charge, and phase require separate decisions

For any proposed transformation, make the following decisions in order.

1. Admissibility. Does it preserve the bundle, field boundary conditions, and the boundary completion of the action? If not, it relates different problems rather than acting within one theory.
2. Component. If it is admissible, does it lie in $\mathcal G_{\mathcal B}^{\circ}(P)$? This answers only whether it is large.
3. Generator. Does its differentiable Hamiltonian generator vanish on the allowed phase space? A vanishing generator identifies a redundancy. A finite, integrable, nonzero boundary generator can define a physical symmetry even in the identity component.
4. Phase. Does the action or state acquire a boundary or topological phase? A consistent character can define a quantum sector. An inconsistent phase is an anomaly obstruction, a different question from largeness.

Boundary conditions and boundary terms are part of the Hamiltonian system, and they determine its allowed transformations and generators. In a Dirichlet Maxwell example, transformations that vanish at the boundary are presymplectic zero modes, whereas transformations approaching a nonzero constant act with an electric-flux generator Harlow and Wu 2020, § 1, pp. 3–4, and § 3.3, pp. 22–23, Open PDF. A theta character is therefore not, by itself, a global anomaly.

A Yang–Mills ball in three descriptions

Consider compact $\operatorname{SU}(2)$ Yang–Mills theory on

$$M=\mathbb R_t\times B^3.$$

Choose boundary data for which transformations in $\mathcal G_{\partial}$ are redundancies: they equal the identity on $S^2$, and their Gauss generators have no surface term. Suppose separately that some transformations with nontrivial boundary value are allowed and have finite, integrable generators. These are two declared classes of transformations, not conclusions obtained from the local field equations.

Orbit description. Quotienting only $\mathcal G_{\partial}^{\circ}$ gives $\widetilde{\mathcal C}$, on which the residual deck group $\mathbb Z$ acts by winding. If every based component is a redundancy, the full classical orbit space is

$$\mathcal C =\frac{\widetilde{\mathcal C}}{\mathbb Z} =\frac{\mathcal A}{\mathcal G_{\partial}}.$$

The theta label does not undo this classical quotient. It specifies a flat line bundle over $\mathcal C$, or equivalently the equivariance of a function on $\widetilde{\mathcal C}$.

Charge description. A based transformation of winding one is large but has no boundary surface charge under the stated conditions. Conversely, a constant boundary rotation can be connected to the identity because $\operatorname{SU}(2)$ is connected. If the boundary conditions allow it, its generator can contain a term of the form

$$Q[\epsilon] \sim \int_{S^2} \operatorname{tr}(\epsilon E^r),$$

with its exact normalization and possible counterterms fixed by the action. It is then a physical boundary symmetry even though it is not large. This comparison makes charge and component visibly independent.

Gauge-fixed description. A local condition such as Coulomb gauge and its Faddeev–Popov operator probe infinitesimal directions near the identity. They do not compute $\pi_0(\mathcal G_{\partial})$, decide whether a residual component is quotiented, or select the character $e^{-in\theta}$. A global gauge-fixing prescription must still impose the component identifications and the theta equivariance. It may also encounter multiple representatives related by disconnected transformations, but a full Gribov analysis lies beyond this page.

The integer Chern–Simons shift is valid in this bounded example because the based condition kills its surface term. Once nonconstant boundary values are allowed, write $\Delta_uN_{\mathrm{CS}}[a]$ for $N_{\mathrm{CS}}[a^u]-N_{\mathrm{CS}}[a]$. Then

$$\Delta_uN_{\mathrm{CS}}[a] =\nu(u)+\text{boundary functional},$$

and the second term must be retained. The boundary conditions, action counterterms, and complete generator then decide whether $u$ is inadmissible, redundant, charged, or represented through a phase.

Scope limits

• The explicit integer calculation uses a trivial $\operatorname{SU}(2)$ bundle on $B^3$. Other manifolds, bundles, global groups, and boundary conditions can give different component groups.
• The Chern–Simons calculation fixes a transformation law; it does not classify all topological actions or their boundary completions.
• Four-dimensional instanton solutions can relate changes in Chern–Simons number between time slices, but their construction, amplitudes, moduli, and vacuum energy are not developed here.
• A consistent theta character is not an anomaly analysis. Global anomaly obstructions require additional fermionic and cobordism data.
• A local Faddeev–Popov operator does not provide a global theorem about the orbit space or resolve the Gribov problem.

Common pitfalls

Calling every nonidentity transformation large. A transformation is large only if it is disconnected from the identity inside the declared admissible group.

Equating large with charged. A based winding transformation can be large and uncharged, while a connected boundary transformation can be physically charged.

Using one integer for two classifications. A first Chern number labels a bundle sector. The winding $\nu(u)$ labels a component of an automorphism group of a fixed bundle.

Dropping the Chern–Simons surface term at a boundary. The pure integer shift follows for a closed spatial manifold or for boundary conditions, such as $u|_{S^2}=\mathbf1$, that make the surface term vanish.

Calling a theta character an anomaly. A character respects the component group law and consistently defines a quantum sector. An anomaly is an obstruction to such a consistent implementation.

Check your understanding

On a three-ball, classify each of the following and state whether it is a bundle label, a transformation component, a redundancy, or a possible boundary charge:

1. an $\operatorname{SU}(2)$ transformation with $u|_{S^2}=\mathbf1$ and $\nu(u)=1$, when every based component is quotiented;
2. an allowed constant nonidentity boundary rotation whose finite, integrable generator is nonzero;
3. a compact-$U(1)$ bundle with $(2\pi)^{-1}\int_{S^2}da=m\neq0$.

For the first case, also determine the phase of $\Phi_\theta$ and explain why the same integer-shift calculation cannot be applied unchanged to the second.

Solution

The first transformation lies in the winding-one component of $\mathcal G_{\partial}$. It is large, but under the stated boundary conditions it is a redundancy with no boundary surface charge. The equivariant representative obeys

$$\Phi_\theta[a^u]=e^{-i\theta}\Phi_\theta[a].$$

The phase is a character of the residual component group, not a new bundle class.

The constant boundary rotation is connected to the identity in the enlarged admissible $\operatorname{SU}(2)$ group, so it need not be large. Its nonzero generator makes it a physical boundary symmetry rather than a redundancy. Because its boundary value is nontrivial, the Chern–Simons transformation law contains the surface functional; no phase follows from $\nu$ alone.

The integer $m$ is the first Chern number of a $U(1)$ bundle. It labels the field sector, not a gauge transformation. Automorphisms of that fixed bundle preserve $m$, and no theta character is determined by the value of $m$ alone.

Where the remaining questions are developed

Principal-Bundle Topological Sectors and Large Gauge Transformations gives a theorem-level treatment of bundle sectors and automorphism groups. Theta Terms, Periodicity, and Vacuum Sectors classifies the topological terms and global-form qualifications suppressed here. Theta Parameters, States, and Sector Sums develops theta states dynamically, while Gauge Instantons: Charge and Moduli constructs the interpolating solutions. Global and Torsion Anomalies handles inconsistent global phases rather than consistent characters.

References

• Harlow, Daniel, and Jie-qiang Wu. “Covariant Phase Space with Boundaries.” Journal of High Energy Physics 10 (2020): 146. DOI. Open PDF
• Tong, David. Gauge Theory. Cambridge Part III Mathematical Tripos lecture notes, 2018. Official course page. Official PDF
• Weinberg, Steven. The Quantum Theory of Fields. Volume II: Modern Applications. Cambridge: Cambridge University Press, 1995. DOI