Gauge Constraints, Centers, and Edge Data

Gauss law prevents the physical Hilbert space of a gauge theory from factorizing naively across a spatial boundary. A gauge-theory subsystem is therefore specified first by a regional algebra of gauge-invariant observables. Its center identifies superselection data visible at the boundary; an extended Hilbert space is one useful tensor-product representation of that choice, not a unique physical factorization.

Required background. Charges, screening, and long-range forces supplies Gauss-law physics; gauge-invariant dressed observables supplies physical operator support; edge modes and factorization supplies boundary extensions; local region algebras supplies the algebraic subsystem.

Helpful background. Lattice gauge Hamiltonians and Gauss law supplies a regulator; support and codimension organizes Wilson and ’t Hooft operators; continuum subsystem choice supplies the continuum limit.

Chapter map. The overview gives the task-to-resource map, the comparison table, and the three gates for an operational claim.

Gauss law and the regional algebra

On a lattice, gauge transformations at a boundary vertex act on links assigned to both $A$ and its complement. Physical states obey the corresponding Gauss constraint, so in general

$$\mathcal H_{\rm phys} \ne \mathcal H_{A,{\rm phys}}\otimes \mathcal H_{\bar A,{\rm phys}}.$$

Tracing an arbitrarily selected set of physical links therefore does not define a unique gauge-invariant reduced state. Instead choose an algebra $\mathcal A_A$ of regional observables. Its center is

$$\mathcal Z_A:=\mathcal A_A\cap\mathcal A_A',$$

where the commutant is taken in the represented physical observable algebra. Simultaneous eigenvalues of $\mathcal Z_A$ label superselection sectors. In a finite regulator the restricted state has the block form

$$\rho_{\mathcal A_A} =\bigoplus_R p_R\rho_R, \qquad S_{\mathcal A_A} =H(p_R)+\sum_Rp_RS(\rho_R)$$

when the trace is normalized separately on each matrix block. Other trace conventions must be stated because they can move representation-dimension weights between terms.

Electric, magnetic, and trivial centers

For an Abelian lattice gauge theory, an electric-center algebra retains the normal electric-flux operators on boundary links. Their eigenvalues form $R$, subject to the boundary Gauss constraints. Wilson lines that would change those eigenvalues are absent from the regional algebra.

A magnetic-center construction instead retains selected boundary Wilson loops while omitting their conjugate boundary electric operators. A trivial-center algebra removes enough boundary generators—conveniently along a maximal tree on the boundary—to leave a full matrix algebra. The tree must be chosen locally on the boundary; a generic global axial gauge can turn a nominally regional variable into a large nonlocal loop.

These are different gauge-invariant algebras, so they can assign different entropies to the same global state. Gauge invariance alone does not select among them. The operational question does: which Wilson lines may terminate at the boundary, which fluxes a regional observer may measure, and whether boundary charges are fixed, dynamical, or excluded.

The Abelian classification, including the localized maximal-tree construction, is Casini, Huerta, and Rosabal 2014, §§2–4. That paper does not by itself establish the general non-Abelian decomposition.

Extended Hilbert space dictionary

The extended-Hilbert-space construction splits each boundary-crossing link and relaxes Gauss law at the new endpoints. In a representation basis, an oriented cut link is embedded schematically as

$$|r,i,j\rangle \longmapsto \frac1{\sqrt{d_r}} \sum_{k=1}^{d_r}|r,i,k\rangle_A|r,k,j\rangle_{\bar A}.$$

Reversing the orientation replaces $r$ by its conjugate representation. Thus the two sides carry matching conjugate boundary data, rather than independent physical charges. The enlarged space factorizes,

$$\mathcal H_{\rm ext} =\mathcal H_A^{\rm ext}\otimes\mathcal H_{\bar A}^{\rm ext},$$

and ordinary partial trace is then available.

Let $R=(r_\ell)_{\ell\in\partial A}$ be the complete tuple of cut-link representations, including all fusion constraints, and define

$$d_R:=\prod_{\ell\in\partial A}d_{r_\ell}.$$

The extended-space entropy is

$$S_{\rm ext} =H(p_R) +\sum_Rp_RS(\rho_R) +\sum_Rp_R\log d_R.$$

Equivalently, the last term is $\sum_{\ell\in\partial A}\langle\log d_{r_\ell}\rangle$. This is the exact cut-link form of Donnelly 2012, Eqs. (28)–(32). Intertwiners and bulk multiplicities belong to the conditional spaces $\rho_R$; they should not be hidden inside an undefined single “boundary component” dimension.

For Abelian groups, $d_R=1$ and the extended-space entropy agrees with the electric-center algebraic entropy in the standard lattice construction. For a non-Abelian group, the representation-index term is supplied by the extension and its canonical tensor-product trace. With the sectorwise-normalized algebra trace above, the algebraic entropy omits that term. The difference is a dictionary between specified prescriptions, not an algebra-independent observable or a supply of independently distillable particles.

Under local gauge-invariant operations, the boundary representation tuple cannot be changed coherently. For a pure lattice state, Van Acoleyen et al. 2016, Eq. (7) and the direct/converse proof on pp. 3–4 find

$$E_D^{\rm gauge} =\sum_Rp_RS(\rho_R),$$

whereas $H(p_R)$ and $\sum_Rp_R\log d_R$ are unavailable as Bell-pair yield for that task.

Exact Z₂ lattice benchmark

Use the topological $Z_2$ state on a square lattice constructed in Casini, Huerta, and Rosabal 2014, §5.2, Eqs. (50)–(55): it is the equal coherent sum generated by all closed Wilson loops. Choose a simply connected region whose boundary contains $L_V=4$ links. There is one boundary Gauss constraint, so only

$$N=L_V-n_\partial=4-1=3$$

electric-center generators are independent. Their eigenvalues $\lambda_i=\pm1$ label $2^3=8$ allowed sectors, each with

$$p_\lambda=\frac18.$$

Every conditional state is pure: all Wilson loops supported in the region have eigenvalue $+1$. The electric-center entropy is therefore

$$S_E=H(p_\lambda)=\log8=3\log2=2.0794415417,$$

in agreement with Casini, Huerta, and Rosabal 2014, Eq. (60).

In the Abelian extended-space embedding, the reduced state consequently has eight nonzero eigenvalues, all $1/8$. It reproduces the same boundary-electric probabilities, the same expectation value $+1$ for every interior Wilson loop, and the same entropy $3\log2$. Since every $Z_2$ irrep has dimension one, there is no representation-dimension term. This executes the electric-center/extended-space comparison on a fully specified finite regulator.

Prescription on the same region	Center data with nonzero weight	Conditional state	Entropy
Electric center	eight flux sectors, $p_\lambda=1/8$	pure in every sector	$3\log2$
Extended Hilbert space	same eight sectors, $d_R=1$	pure in every sector	$3\log2$
Magnetic center	boundary Wilson loop fixed to $+1$	pure	$0$
Trivial center	no center sector	pure	$0$

The magnetic and trivial-center values are Casini, Huerta, and Rosabal 2014, Eqs. (61)–(62).

Adversarial control: change the boundary algebra

Now hold the global state and geometric set of links fixed but replace the electric-center algebra by the magnetic-center algebra. If one incorrectly keeps the electric formula $H(p_\lambda)=3\log2$, one predicts an entropy for flux projectors that are no longer central observables of the new algebra. The correct magnetic center has a single supported Wilson-loop sector and entropy zero. Removing the remaining boundary center gives the same zero for the trivial-center algebra.

The adversary changes the answer by

$$S_E-S_M=3\log2,$$

without changing the global state. The surviving statement is not that the state possesses one unique entropy; it is that the electric algebra and its Abelian extended-space representation agree on their common observables and entropy, while a different boundary algebra defines a different subsystem. The prescription-dependent term is the classical uncertainty of the selected center in this example. The within-sector entropy and gauge-invariant distillable entanglement both remain zero.

Continuum boundary and regulator

In continuum Maxwell or Yang–Mills theory, normal electric flux is distribution-valued and its sharp-boundary fluctuations are ultraviolet divergent. A center probability is then a density relative to a specified functional measure, not a finite discrete Shannon distribution. Casini, Huerta, and Rosabal 2014, §5.1, Eqs. (47)–(49) show explicitly why the entropy of a continuous center changes under a redefinition of its measure.

The continuum comparison must state the boundary conditions, zero-mode and global-flux treatment, center resolution, local counterterms, and limiting procedure. The gauge-field edge-contribution page develops the corresponding Maxwell edge determinant. A universal term may survive a matched change of regulator, but a bare center differential entropy does not.

Common pitfalls

Assuming gauge invariance selects one center. Several regional algebras can be gauge invariant. Name the accessible boundary operators.

Treating the extended space as the physical factorization. It is an embedding with added endpoint representation indices. Match it to a declared algebra.

Writing one $d_R$ without defining $R$. For the cut-link construction, $R$ is the full constrained boundary tuple and $d_R$ is the product of its link dimensions.

Using a discrete flux entropy in the continuum without a measure. A probability density and its differential entropy depend on the regulator and reference measure.

Exercises

1. Count the electric-center sectors

For the $Z_2$ benchmark with four boundary links, show that one Gauss constraint leaves eight sectors and compute their entropy.

Solution

Let the four boundary-electric eigenvalues be $\lambda_i=\pm1$. Multiplying Gauss law over all vertices inside the simply connected region cancels every interior link and imposes one boundary relation,

$$\lambda_1\lambda_2\lambda_3\lambda_4=1.$$

Three signs are independent and the fourth is fixed, so there are $2^3=8$ sectors. The topological state gives equal weight $p_\lambda=1/8$, hence

$$H(p_\lambda)=-8\left(\frac18\log\frac18\right)=\log8=3\log2.$$

2. Derive the representation-dimension term

Fix one cut-link irrep $r$ and show why its extended-space embedding contributes $\log d_r$ to the entropy.

Solution

The cut-link embedding contains the maximally entangled representation-index state

$$|\phi_r\rangle =\frac1{\sqrt{d_r}}\sum_{k=1}^{d_r}|k\rangle_A|k\rangle_{\bar A}.$$

Tracing the outside index gives $I_{d_r}/d_r$, whose entropy is $\log d_r$. Independent cut links tensor together, so their entropies add:

$$\sum_{\ell\in\partial A}\log d_{r_\ell} =\log\prod_{\ell\in\partial A}d_{r_\ell} =\log d_R.$$

Averaging over boundary tuples gives $\sum_Rp_R\log d_R$.

3. Identify the failed invariant

In the $Z_2$ benchmark, suppose someone claims that the value $3\log2$ is invariant under changing from the electric to the magnetic center. Locate the failed assumption and state which quantities do remain unchanged.

Solution

The claim keeps the electric flux-sector probabilities after removing their projectors from the center. That changes the algebra but not the entropy formula, so it compares different subsystems as though they were one.

For the magnetic center, the supported boundary Wilson-loop eigenvalue is $+1$ with probability one and its conditional state is pure; the entropy is therefore zero. The unchanged data are the global state and expectation values of observables common to both algebras. In this example the within-sector entropy and gauge-invariant distillable entanglement are also zero in either prescription. The center Shannon term is not invariant.

References

• Casini, Horacio, Marina Huerta, and José Alejandro Rosabal. “Remarks on Entanglement Entropy for Gauge Fields.” Physical Review D 89 (2014): 085012. DOI. Open PDF.
• Donnelly, William. “Decomposition of Entanglement Entropy in Lattice Gauge Theory.” Physical Review D 85 (2012): 085004. DOI. Open PDF.
• Van Acoleyen, Karel, Nick Bultinck, Jutho Haegeman, Michael Marien, Volkher B. Scholz, and Frank Verstraete. “The Entanglement of Distillation for Gauge Theories.” Physical Review Letters 117 (2016): 131602. DOI. Open PDF.