Gauge-Invariant and Dressed Observables

A gauge-covariant field becomes part of a physical observable only after its unphysical gauge variation is removed: covariant fields can be contracted to a local singlet, an open transporter can be completed at its endpoints, or a charged insertion can be attached to a reference system, a boundary, or infinity. The result must be invariant under the zero-generator subgroup $\mathcal G_0$, although it may still carry a genuine boundary or global symmetry charge. In a $3+1$-dimensional Abelian theory, Gauss law makes that charge measurable by arbitrarily distant electric flux, so a charge-changing operator cannot be compactly localized like a neutral local observable. This page develops these statements for formal Wilson composites and a bounded Maxwell example; renormalized line operators, infrared scattering dressings, and rigorous charged-sector algebras remain outside its scope.

Required background. Gauge Orbits, Gauss Constraints, and Stabilizers supplies the Gauss constraint, the zero-generator subgroup $\mathcal G_0$, and the distinction between redundancy and charged boundary symmetry.

Helpful background. Parallel Transport and Holonomy supplies the geometric transporter, while Fields, Observables, and Interpolating Operators separates a useful field variable from an observable.

Observables descend to the redundancy quotient

Fix a gauge group, a matter representation, a bundle sector, boundary conditions, and the corresponding admissible field space. A classical functional descends to the physical quotient only if

$$\mathcal O[A^h,h\psi] =\mathcal O[A,\psi], \qquad h\in\mathcal G_0.$$

This requirement concerns transformations whose differentiable generators have vanishing variation throughout the allowed constraint surface; a state-independent constant can then be normalized to zero. An admissible transformation with a nonconstant, variable surface generator is instead a physical symmetry. An operator may therefore be invariant under $\mathcal G_0$ and still transform in a nontrivial representation of that symmetry.

Local singlet contractions pass the first test. For matter in a unitary representation, $\psi^\dagger\psi$ is invariant. In non-Abelian theory the curvature matrix is only covariant, $F_{\mu\nu}^h=hF_{\mu\nu}h^{-1}$, but $\operatorname{tr}_R(F_{\mu\nu}F^{\mu\nu})$ is invariant. By contrast, $A_\mu$ and $\psi$ are useful local fields on the auxiliary configuration space, not functions on the quotient by themselves.

This is a transformation-law test, not a complete quantum construction. Coincident composites require renormalization, and a formally invariant line may have endpoint, global, or ultraviolet qualifications. Here “observable” means an object well defined after quotienting redundancies. In algebraic QFT, one often reserves that word for neutral operators acting within one superselection sector and calls a charge-changing object a charged field or intertwiner; the distinction does not alter the Gauss-law argument below.

Endpoint covariance completes open transport

Let $\gamma$ be an oriented path from $x$ to $y$, and let the connection act in the representation of $\psi$. With $D_\mu=\partial_\mu-igA_\mu$, define

$$\begin{aligned} U_\gamma(y,x) &=\mathcal P\exp\!\left(ig\int_\gamma A\right), \\ \gamma(0)&=x, \qquad \gamma(1)=y. \end{aligned}$$

The parallel-transport equation and the finite gauge law give

$$U_\gamma^h(y,x) =h(y)U_\gamma(y,x)h(x)^{-1}.$$

The two endpoint factors are precisely what an open line needs to become a neutral bilocal composite:

$$\begin{aligned} \mathcal B_\gamma(y,x) &=\psi^\dagger(y)U_\gamma(y,x)\psi(x), \\ \mathcal B_\gamma^h(y,x) &=\mathcal B_\gamma(y,x). \end{aligned}$$

The endpoints and path are part of this operator; gauge invariance has not made it local. Closing the path removes the endpoints, but the holonomy matrix still transforms by conjugation, so an invariant function such as $\operatorname{tr}_R U_C$ is required. Tong derives the curvature and endpoint transformation laws and the traced closed holonomy in Tong 2018, §§ 2.1.2–2.1.3, pp. 31–34, official full-notes PDF.

If a path crosses trivializing patches, transition functions must be inserted at the crossings; the global parallel transporter remains the same geometric map. Local Potentials and Global Gauge Configurations develops that patchwise description. The full quantum treatment of line operators and their renormalization belongs to Wilson Lines and Loops.

A boundary endpoint can carry physical charge

Now specialize to compact $U(1)$ with a unit-weight matter field of charge $g>0$:

$$A\longmapsto A+d\lambda, \qquad \psi(x)\longmapsto e^{ig\lambda(x)}\psi(x).$$

Let $x$ lie in a spatial region $\Sigma$ and let a path $\gamma$ run from $x$ to $b\in\partial\Sigma$. Define the boundary-dressed field

$$\Psi_\gamma(b,x) =U_\gamma(b,x)\psi(x).$$

Its transformation is concentrated at the other endpoint:

$$\begin{aligned} U_\gamma(b,x) &\longmapsto e^{ig\lambda(b)}U_\gamma(b,x)e^{-ig\lambda(x)}, \\ \Psi_\gamma(b,x) &\longmapsto e^{ig\lambda(b)}\Psi_\gamma(b,x). \end{aligned}$$

For the bounded field space used below, choose the boundary data so that $\mathcal G_0$ is precisely the based subgroup $\lambda|_{\partial\Sigma}=0$, while an allowed constant boundary value has a finite, integrable surface generator that varies on phase space. Then $\Psi_\gamma$ is constant along every $\mathcal G_0$-orbit. If a different boundary phase space has additional zero-generator transformations, invariance under them must be checked as well. For an allowed value $\lambda|_{\partial\Sigma}=c$ outside $\mathcal G_0$, $\Psi_\gamma$ transforms as a physically charged operator. There is no contradiction: redundancy has been removed, while a true boundary symmetry still acts. A boundary degree of freedom $\chi(b)\mapsto e^{ig\lambda(b)}\chi(b)$ can instead complete the neutral relational operator

$$\chi^\dagger(b)U_\gamma(b,x)\psi(x).$$

The dressing is not unique. For two paths with the same endpoints, the Abelian line factors obey

$$U_{\gamma_1}(b,x)U_{\gamma_2}(b,x)^{-1} =\exp\!\left(ig\oint_{\gamma_1-\gamma_2}A\right).$$

Their difference is a closed holonomy, so changing the path need not leave the operator unchanged. In non-Abelian theory the corresponding based holonomy is conjugated at $b$, and its boundary index must be contracted or otherwise retained as physical data. Thus a dressing is a specified nonlocal completion, not a unique local redefinition of $\psi$. A path-supported Wilson dressing gives a stringlike completion, while another admissible dressing may distribute the compensating electric flux. Formal gauge invariance alone neither identifies those field profiles nor decides whether either construction has finite energy.

Gauss law forbids compactly localized charge

The sharp localization statement is Abelian. Consider QED on $\mathbb R^{3,1}$, and assume that Gauss law holds on the physical domain and that the regulated large-sphere flux converges to the total charge:

$$\begin{aligned} Q_R &=\int_{S_R^2}dS_i\,E^i =\int_{B_R}d^3x\,\rho, \\ Q&=\lim_{R\to\infty}Q_R. \end{aligned}$$

Suppose a physical operator $\mathcal X$ were compactly localized in a bounded spatial region at fixed time. If gauge-invariant observables satisfy ordinary spacelike commutation, then $E^i$ on a sufficiently large sphere is spacelike separated from $\mathcal X$. Consequently,

$$\begin{aligned} [Q,\mathcal X] &=\lim_{R\to\infty} \int_{S_R^2}dS_i\,[E^i,\mathcal X] \\ &=0. \end{aligned}$$

This contradicts $[Q,\mathcal X]=q\mathcal X$ when $q\neq0$. Therefore an operator that changes the Abelian Gauss-law charge cannot be compactly localized in the same sense as a neutral local observable. Its completion must retain support or relational data extending to an opposite charge, a boundary, or infinity. Neutral local singlets are not obstructed.

The argument assumes the flux limit and commutators exist on a suitable domain; it is not an informal manipulation of unrestricted operator-valued distributions. Mund, Rehren, and Schroer review the Gauss-law flux relation and this localization obstruction in Mund, Rehren, and Schroer 2020, §§ 1.1–1.2, arXiv:1906.09596v2, preprint pp. 2–4, Open PDF.

One bounded Maxwell–matter insertion in three descriptions

Extend the finite-region Maxwell system on the prerequisite pages by the unit-weight charged matter field $\psi$. Take $M=\mathbb R\times\Sigma$ with smooth connected $\partial\Sigma$, fix the pullback of $A$ on the timelike boundary, and work in one bundle sector. Choose matter boundary conditions that admit constant $U(1)$ phases. Define the charge density $\rho$ so that, with $E^i=F^{i0}$ and $E^\perp=n_iE^i$ for the outward normal, the matter Gauss constraint and its integrated form are

$$\begin{aligned} \mathcal C &=\partial_iE^i-\rho \approx0, \\ Q_{\partial} &=\int_{\partial\Sigma}d^2S\,E^\perp, \\ Q_{\partial} &=\int_\Sigma d^3x\,\rho \quad\text{on }\mathcal C=0. \end{aligned}$$

Allow this flux to vary. The constant boundary phase then has a finite, integrable, nonconstant generator $Q_{\partial}$ and lies outside $\mathcal G_0$; for this example $\mathcal G_0$ is the based subgroup. The same $\Psi_\gamma(b,x)$ has three compatible descriptions.

Orbit description. Based transformations have generator variations that vanish and act trivially on $\Psi_\gamma$. After the state-independent constant is normalized away, they are exactly $\mathcal G_0$. The dressed field is therefore well defined on the quotient, even though neither $\psi(x)$ nor the open transporter is separately invariant.

Charge description. Let $\mathcal U(c)=e^{icQ_{\partial}}$ implement an allowed constant boundary symmetry, with the convention $\mathcal U(c)\Psi_\gamma\mathcal U(c)^{-1} =e^{igc}\Psi_\gamma$. Then

$$[Q_{\partial},\Psi_\gamma] =g\Psi_\gamma.$$

Acting with $\Psi_\gamma$ therefore maps a physical state to the boundary-charge sector whose $Q_{\partial}$ eigenvalue is larger by $g$ in this convention. Attaching $\chi^\dagger(b)$ produces a neutral operator instead. The exact set of admitted boundary transformations and fluxes is part of the boundary theory, not fixed by the local Maxwell equations alone.

Gauge-fixed description. When an admissible gauge choice imposes $\dot\gamma^iA_i=0$ along this particular path, its line factor is one and the representative of $\Psi_\gamma$ looks like the bare field $\psi(x)$. The gauge condition has hidden the path and boundary completion in the choice of representative; it has not made the physical charged operator compactly local. Residual transformations and global obstructions still have to be checked.

Tong derives the Gauss constraint, the physical-state condition, and the charge interpretation of transformations that remain nontrivial at infinity in Tong 2018, § 2.2.1, pp. 40–42, official full-notes PDF. The finite boundary zero-generator test used here is the one established on Gauge Orbits, Gauss Constraints, and Stabilizers; an asymptotic argument does not by itself choose a finite-boundary phase space.

What the construction does not establish

The endpoint law holds for ordinary non-Abelian transporters, but the sharp flux-localization proof above is a $3+1$-dimensional Abelian statement. It does not establish the existence of isolated gauge-invariant color charges in a confining theory, nor does it decide which non-Abelian dressing is admissible in a given phase.

Formal gauge invariance also does not prove that a Wilson composite is a finite renormalized operator, that a chosen dressing has finite energy, or that it defines an infrared-safe scattering state. Those questions require additional dynamics and regulators. Finally, a fixed boundary reference is extra physical structure: if it transforms, it must be included in the system, and if it is held fixed, that choice restricts the boundary symmetry.

Common pitfalls

Saying that gauge theory has no local observables. Neutral singlets such as properly renormalized versions of $\psi^\dagger\psi$ can be local. Gauss law obstructs compact localization of charge-changing operators, not all observable content.

Calling an open Wilson line gauge invariant. It transforms at both endpoints. Matter, boundary data, or another endpoint must cancel or retain those transformations deliberately.

Quotienting a charged boundary symmetry. Invariance is required under $\mathcal G_0$. A transformation with a finite, integrable surface generator that varies on phase space acts physically and may assign charge to a dressed operator.

Assuming every dressing is equivalent. Two paths with the same endpoints can differ by a closed holonomy. The path, reference, and boundary conditions are part of the operator definition.

Mistaking a gauge-fixed field for a local physical operator. A gauge can make a chosen dressing factor equal to one. The nonlocal constraint and boundary information remain encoded in that gauge choice.

Using gauge invariance as a renormalization theorem. The transformation law is only the first test. Finiteness, operator domains, and regulator dependence require a separate analysis.

Check your understanding

1. Starting from the endpoint law for $U_\gamma(y,x)$, show that $\psi^\dagger(y)U_\gamma(y,x)\psi(x)$ is invariant. Then take $y=b$ on a boundary and show that $U_\gamma(b,x)\psi(x)$ is invariant under based transformations but charged under an allowed constant boundary symmetry.
2. Suppose $Q=\lim_{R\to\infty}\int_{S_R^2}dS_iE^i$ and a physical operator $\mathcal X$ is compactly localized. Under the stated microcausality and domain assumptions, show that $[Q,\mathcal X]=0$. Which operators evade the conclusion, and why does a gauge-fixed bare charged field not provide a counterexample?

Solution

For a unitary representation, $\psi^\dagger(y)\mapsto\psi^\dagger(y)h(y)^{-1}$ and $\psi(x)\mapsto h(x)\psi(x)$. Write $h_x=h(x)$ and $h_y=h(y)$. Then

$$\begin{aligned} \mathcal B_\gamma^h &=(\psi^\dagger h_y^{-1}) (h_yU_\gamma h_x^{-1})(h_x\psi) \\ &=\psi^\dagger U_\gamma\psi =\mathcal B_\gamma. \end{aligned}$$

For the boundary dressing, the factor at $x$ cancels but the one at $b$ remains:

$$\Psi_\gamma(b,x) \longmapsto e^{ig\lambda(b)}\Psi_\gamma(b,x).$$

It is invariant when $\lambda(b)=0$. If an allowed constant boundary value $c$ has a finite, integrable generator that varies on phase space, the remaining phase is the action of a physical symmetry and identifies the charge of the operator.

For a compact localization region, choose $R$ so large that the equal-time sphere $S_R^2$ is spacelike separated from it. Microcausality gives $[E^i(\mathbf z),\mathcal X]=0$ on that sphere. Integrating first and then taking the regulated limit gives $[Q,\mathcal X]=0$. Neutral local operators are consistent with this result. A charge-changing operator must instead have noncompact support or relational data reaching an opposite charge, the boundary, or infinity. A bare charged field in a gauge-fixed presentation is not an operator on the redundancy quotient by itself; if it represents a dressed physical field in that gauge, the nonlocal completion has merely become implicit.

Where dressings lead

Global Form, Matter Representations, and the Faithful Gauge Group determines which matter and endpoint representations are actually allowed. Proper and Improper Gauge Transformations develops boundary-preserving transformations and the differentiable-generator criterion for a nonzero surface charge.

Dressed States and Infrared-Finite Scattering owns scattering dressings and their infrared dynamics. Gauss-Law Charges and Infrared Sectors gives the theorem-level charged-sector and localization treatment.

References

• Mund, Jens, Karl-Henning Rehren, and Bert Schroer. “Gauss’ Law and String-Localized Quantum Field Theory.” Journal of High Energy Physics 01 (2020): 001. DOI. Open PDF, arXiv:1906.09596v2
• Tong, David. Gauge Theory. Cambridge Part III Mathematical Tripos lecture notes, University of Cambridge, 2018. Official course page. Official PDF