Centers, Edge Extensions, and Distillable Entanglement

Gauge-theory entropy separates three mathematically distinct contributions: uncertainty in a boundary superselection label, entanglement in the multiplicity spaces inside each sector, and representation-space entropy introduced by an extended-Hilbert-space split. Their sum is a useful entropy, but it is not automatically a supply of Bell pairs. The operational answer depends on the regional algebra, the allowed local operations, and whether boundary carriers are auxiliary or physical.

Required background. Gauge subregions and centers fixes the algebra and flux sectors; gauge-field edge contributions fixes the extension and continuum measure.

Helpful background. Superselection and accessible entanglement supplies the analogous charge-sector decomposition.

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

Boundary sectors and their multiplicity spaces

Consider a pure lattice gauge-theory state and a bipartition whose boundary meets $n$ vertices. Gauge invariance gives a direct-sum decomposition labelled by the vector of boundary irreducible representations

$$\vec r=(r_1,\ldots,r_n), \qquad d_{\vec r}:=\prod_{i=1}^{n}d_{r_i}.$$

After the representation indices have been made explicit, the reduced state has sector weights $p_{\vec r}$ and normalized multiplicity-space states $\rho_A^{\vec r}$. Its extended-Hilbert-space entropy is

$$S_{\rm ext}(A) =H(p_{\vec r}) +\sum_{\vec r}p_{\vec r}\log d_{\vec r} +\sum_{\vec r}p_{\vec r}S(\rho_A^{\vec r}).$$

Here $H(p_{\vec r})=-\sum_{\vec r}p_{\vec r}\log p_{\vec r}$ is classical sector uncertainty. The representation term comes from maximally mixed boundary carrier indices in the chosen extension. The last term is entanglement in the multiplicity spaces on which gauge-invariant regional operations can act. Donnelly 2012, § II, Eqs. (27)–(32) derives this three-term decomposition directly from the boundary-representation form of the reduced state; Soni and Trivedi 2016, § 2.2, Eq. (2.30), pp. 8–9 gives the corresponding non-Abelian formulation.

For the electric-center algebra, the representation term is absent while the sector and conditional terms remain:

$$S_{\rm alg}(A) =H(p_{\vec r}) +\sum_{\vec r}p_{\vec r}S(\rho_A^{\vec r}).$$

Neither expression is wrong. They answer different algebraic questions, so an entropy prescription must be named before two values are compared.

What gauge-invariant LOCC can distill

Suppose Alice and Bob may use local gauge-invariant operations and classical communication. They may measure $\vec r$ and condition a protocol on its value, but their local operations cannot coherently mix different boundary sectors or act on an auxiliary representation carrier as though it were a laboratory system. For the pure-state task above, the asymptotic Bell-pair rate is

$$E_D^{\rm gauge}(|\Psi\rangle) =\sum_{\vec r}p_{\vec r}S(\rho_A^{\vec r}).$$

Van Acoleyen et al. 2016, Eq. (7) and the direct/converse proof on pp. 3–4 derive this rate. Soni and Trivedi 2016, § 4.5, Eqs. (4.69)–(4.70), pp. 27–28 reach the same operational separation for Abelian and non-Abelian cases.

The word asymptotic matters: $E_D$ is the rate obtained from many identical copies. It is not a fractional number of Bell pairs promised by one specimen. Moreover, ordinary global-superselection activation results cannot simply be imported when Gauss constraints act independently on every copy. Any enlarged many-copy operation class or shared reference must be stated explicitly.

Executed two-sector benchmark

Take two boundary sectors, labelled $0$ and $1$, with

$$(p_0,p_1)=\left(\frac13,\frac23\right), \qquad (d_0,d_1)=(1,2).$$

Use a product multiplicity state in sector $0$ and one Bell pair in the multiplicity space of sector $1$:

$$E_0=0, \qquad E_1=S(\rho_A^1)=1\ \text{bit}.$$

A normalized extended-space representative is

$$|\Psi\rangle =\sum_{r=0}^{1}\sqrt{p_r}\, |\Phi_r\rangle_{\rm rep} \otimes|\psi_r\rangle_{\rm mult},$$

where $|\Phi_r\rangle_{\rm rep}$ is maximally entangled with Schmidt rank $d_r$. Since

$$H_2\!\left(\frac13,\frac23\right) =\log_2 3-\frac23 =0.918296\ \text{bits},$$

the three answers are

Quantity	Exact value	Numerical value
Algebraic entropy $S_{\rm alg}$	$\log_2 3$	$1.584963$ bits
Extended entropy $S_{\rm ext}$	$\log_2 3+2/3$	$2.251629$ bits
Gauge-LOCC distillation rate $E_D^{\rm gauge}$	$2/3$	$0.666667$ ebits per copy

The difference is not a small correction: treating $S_{\rm ext}$ as a Bell yield overstates this task by about $1.585$ ebits per copy.

Executed edge-resource adversary

Now promote the rank-two representation carrier in sector $1$ to a genuine, controllable boundary pair and supply the reference needed to manipulate it. Within each fixed sector, allow unrestricted local operations on that carrier and the multiplicity space, while retaining the requirement that every admissible operation commute with the sector projectors. In that enlarged but still block-diagonal physical theory, sector $1$ contains one representation ebit and one multiplicity ebit, so the conditional asymptotic yield becomes

$$E_D^{\rm physical\ edge} =p_1(1+1) =\frac43\ \text{ebits per copy}.$$

The retained sector restriction is essential. If the added reference also enabled coherent operations between sectors, the Shannon term $H_2(p)$ could become an activatable resource, and $4/3$ would no longer be the complete yield for that different task.

Removing the shared reference or again restricting both parties to the original gauge-invariant algebra makes the carrier inaccessible and returns the rate to $2/3$. The Shannon contribution $H_2(p)$ remains classical sector uncertainty in both cases. The strongest statement that survives the adversary is therefore:

A representation-space term is distillable only when the corresponding boundary carrier and its manipulating operations are included as physical resources; changing an auxiliary extension alone does not create entanglement.

This control also explains why a material interface can differ from a bookkeeping extension. A Hamiltonian, charge budget, and localized gauge-invariant couplings must demonstrate that the edge system is physically available.

Common pitfalls

Suppressing the vector sector label. In a non-Abelian cut, $\vec r$ records all boundary irreducible representations and $d_{\vec r}$ is their product. Writing an unexplained power of one dimension can hide the boundary multiplicity.

Calling a formal entropy a protocol yield. Algebraic entropy, extended entropy, and gauge-LOCC distillation are different quantities. State the algebra and operation class before comparing them.

Mixing one-shot and asymptotic claims. The conditional entropy gives an asymptotic pure-state rate under the cited protocol. A one-shot task needs its own error tolerance and one-shot entanglement measure.

Exercises

1. Entropy of a direct sum

Let $\rho=\bigoplus_r p_r\rho_r$, with each $\rho_r$ normalized. Show that

$$S(\rho)=H(p_r)+\sum_r p_rS(\rho_r).$$

Solution

If $\lambda_{rj}$ are the eigenvalues of $\rho_r$, the eigenvalues of $\rho$ are $p_r\lambda_{rj}$. Therefore

$$\begin{aligned} S(\rho) &=-\sum_{r,j}p_r\lambda_{rj}\log(p_r\lambda_{rj})\\ &=-\sum_rp_r\log p_r\sum_j\lambda_{rj} -\sum_rp_r\sum_j\lambda_{rj}\log\lambda_{rj}\\ &=H(p_r)+\sum_rp_rS(\rho_r), \end{aligned}$$

because $\sum_j\lambda_{rj}=1$ in every sector.

2. Reproduce the benchmark

Using the two-sector data above, derive all three entries in the benchmark table without decimal approximations.

Solution

First,

$$H_2\!\left(\frac13,\frac23\right) =-\frac13\log_2\frac13-\frac23\log_2\frac23 =\log_2 3-\frac23.$$

The average representation entropy is $(2/3)\log_2 2=2/3$, and the average multiplicity entanglement is $(2/3)(1)=2/3$. Hence

$$S_{\rm alg}=\log_2 3, \qquad S_{\rm ext}=\log_2 3+\frac23, \qquad E_D^{\rm gauge}=\frac23.$$

3. Auxiliary versus controllable carriers

In sector $1$, suppose the representation factor is a Bell state but every allowed local observable acts as the identity on that factor. Show why it contributes one bit to the extended entropy but zero to the declared distillation rate. What changes if the local algebra is enlarged to the full matrix algebra on the carrier?

Solution

Tracing either half of the Bell carrier gives $I_2/2$, so its contribution to the extended entropy is $\log_2 2=1$ bit. Under the restricted algebra, however, all allowed measurement statistics are independent of the carrier state because every allowed observable is proportional to the identity there. No allowed protocol can address, rotate, or extract that Bell pair, so its contribution to the operational rate is zero.

After the algebra is enlarged to all local matrices on the two carrier halves, the parties can run an ordinary Bell-distillation protocol. The carrier then contributes one ebit whenever sector $1$ occurs, or $p_1=2/3$ ebit per copy on average. This is a change of physical resources, not a reinterpretation of the original algebra.

References

• Donnelly, William. “Decomposition of Entanglement Entropy in Lattice Gauge Theory.” Physical Review D 85 (2012): 085004. DOI. Open PDF.
• Soni, Ronak M., and Sandip P. Trivedi. “Aspects of Entanglement Entropy for Gauge Theories.” Journal of High Energy Physics 2016, no. 1 (2016): 136. DOI. Open PDF.
• Van Acoleyen, Karel, Nick Bultinck, Jutho Haegeman, Michael Mariën, Volkher B. Scholz, and Frank Verstraete. “The Entanglement of Distillation for Gauge Theories.” Physical Review Letters 117 (2016): 131602. DOI. Open PDF.