Topological Entanglement Entropy Preview

Topological entanglement entropy is a universal, subleading contribution to the spatial entanglement of a suitable gapped two-dimensional ground state. In the standard vacuum-sector setting, it isolates the total quantum dimension of the phase:

$$\gamma = \ln\mathcal D, \qquad \mathcal D = \sqrt{\sum_a d_a^2}.$$

Here $a$ labels anyon types, $d_a$ is the quantum dimension of type $a$, and natural logarithms are used. The formula is compact; its hypotheses are not. A trustworthy extraction must control the region geometry, correlation length, system size, topological sector, entropy convention, and numerical convergence. It must also distinguish the universal anyonic contribution from corners, physical boundaries, symmetry-breaking constraints, gauge-theory edge conventions, and other nonlocal structures.

A useful calculation cancels local boundary terms, scales the regions beyond the correlation length, and compares the remaining constant with independent properties of the phase. Raw entropy combinations can contain additional, non-topological constants; the anyon formula below must be distinguished from such an uncorrected estimate.

Helpful background. Entanglement entropy explains reduced states, area laws explains boundary scaling, and topological order introduces anyon sectors and quantum dimensions.

Gapped ground states and spatial regions

Unless stated otherwise, assume:

• a local or sufficiently short-range Hamiltonian in two spatial dimensions;
• a pure zero-temperature ground state;
• a nonzero bulk gap and finite bulk correlation length $\xi$;
• regions whose linear dimensions and separations are much larger than $\xi$;
• an entangling boundary far from physical edges, defects, and quasiparticles;
• ordinary intrinsic topological order with a finite set of anyon sectors;
• a bosonic spin-system realization for the basic examples; fermionic theories need their transparent-fermion and sector conventions specified;
• natural logarithms, so entropy is measured in nats;
• a specified tensor-factor or operator-algebra convention for the spatial cut.

For a region $A$ and complement $\bar A$,

$$\begin{aligned} \rho_A &= \operatorname{Tr}_{\bar A} \lvert\Psi\rangle\langle\Psi\rvert, \\ S(A) &= -\operatorname{Tr} \left( \rho_A\ln\rho_A \right). \end{aligned}$$

The Rényi entropy of index $n>0$, $n\ne1$, is

$$S_n(A) = \frac{1}{1-n} \ln\operatorname{Tr}\rho_A^n,$$

with $S(A)=\lim_{n\to1}S_n(A)$.

The symbol $a$ will denote a microscopic cutoff when it appears in a ratio such as $L/a$. The same letter is also conventional for an anyon label. Context distinguishes the two; when both occur in one equation, the cutoff is written $a_0$.

The Universal Constant Behind the Area Law

Local entanglement gives the leading term

For a smooth simply connected region with boundary length $L_{\partial A}$, a gapped ground state typically has an expansion

$$S(A) = \alpha \frac{L_{\partial A}}{a_0} - \gamma + \delta S(A).$$

The coefficient $\alpha$ is nonuniversal. It depends on the regulator, microscopic couplings, and precise placement of the cut. The correction $\delta S(A)$ contains finite-size, curvature, corner, and other geometry-dependent terms. In the intended scaling limit,

$$\frac{L_{\partial A}}{\xi} \longrightarrow \infty,$$

while the shape is scaled without introducing new sharp features, $\delta S(A)$ approaches zero or a controlled known form.

The leading term is local because short-range entangled degrees of freedom within $O(\xi)$ of the cut contribute independently along most of the boundary. A schematic local expansion is

$$\begin{aligned} S_{\mathrm{local}}(A) &= \int_{\partial A}ds\, \mathcal L_{\partial A}(s), \\ \mathcal L_{\partial A}(s) &= c_0(a_0)+c_1\kappa(s) +c_2\kappa(s)^2+\cdots, \end{aligned}$$

where $\kappa(s)$ is boundary curvature. For a pure state with the same cut convention on both sides, $S(A)=S(\bar A)$ constrains this expansion. Reversing the boundary normal changes the sign of signed curvature, so an isolated odd-curvature term is excluded when it is the only orientation-odd quantity. It cannot be used to introduce an arbitrary disk constant. More general permitted terms depend on symmetries and regulator details. The important point is that they are built from local data on the entangling surface.

The constant is not an ordinary intercept

If one computes $S(A)$ for a handful of small disks and fits

$$S(A) \stackrel{?}{=} mL_{\partial A}+b,$$

then identifying $b=-\gamma$ is generally unjustified. The same fitted intercept can absorb:

• lattice-scale shape changes;
• corners and triple junctions;
• finite correlation length;
• proximity to a physical boundary;
• quasiparticles or topological flux;
• symmetry-breaking cat-state entropy;
• numerical truncation and fit-window bias.

Topological-entanglement constructions use several regions with matched local boundaries so these terms cancel before the large-region limit is taken.

Local circuits and subtraction limits

A constant-depth local circuit cannot change the underlying anyon theory, but it can change a raw entropy subtraction. Gates far inside either region do not affect its bipartite entropy; gates near the cut can. It does not follow that their combined entropy change is always a sum of independent local boundary terms. Conditional-information constraints can retain an extra constant even when ordinary correlations have zero range.

For the usual fixed-point states and suitable generic representatives, the subtraction isolates the anyonic contribution. Subsystem-symmetry and replica examples show why a nonzero raw subtraction alone does not establish long-range entanglement or identify a phase.

Total Quantum Dimension

Individual quantum dimensions

An anyon type $a$ has quantum dimension $d_a\ge1$. Operationally, $d_a$ controls the asymptotic growth of fusion spaces containing many anyons of type $a$. Abelian anyons have

$$d_a=1,$$

while non-Abelian anyons have $d_a>1$.

The total quantum dimension is

$$\mathcal D = \sqrt{ \sum_a d_a^2 }.$$

For the vacuum sector, $d_1=1$. The sum runs over all superselection sectors of the anyon theory, not over microscopic particles or lattice sites.

Standard disk formula

For an ordinary two-dimensional intrinsic topological phase in the vacuum sector, the standard asymptotic result is

$$S(A) = \alpha \frac{L_{\partial A}}{a_0} - \ln\mathcal D + \cdots.$$

Thus

$$\gamma_{\mathrm{anyon}} = \ln\mathcal D.$$

The minus sign in the entropy is physical: global topological-charge consistency removes boundary configurations that a purely local count would treat as independent.

The anyonic contribution and a raw estimate

It is useful to separate three statements that are often compressed into one:

Statement	Status under stated assumptions
Fixed-point string-net, quantum-double, and topological-field-theory calculations give $\gamma=\ln\mathcal D$	standard result
A controlled Kitaev–Preskill or Levin–Wen extraction in a generic representative of the same ordinary phase approaches the same anyonic value	standard physical expectation with finite-size qualifications
Every constant returned by every entropy subtraction equals $\ln\mathcal D$	false

For the von Neumann annular definition $\gamma_{\mathrm{raw}}=\tfrac12 I(A:C\mid B)$ in the large-region limit, a lower-bound theorem makes this distinction precise:

$$\gamma_{\mathrm{raw}} \ge \ln\mathcal D.$$

Kim, Levin, Lin, Ranard, and Shi proved this for states related by constant-depth circuits to appropriate quantum-double, string-net, or entanglement-bootstrap reference states. Levin gave a complementary argument from physical assumptions about distinguishable anyon sectors, local indistinguishability, and fusion operations. Thus the result is not a theorem that follows from a spectral gap alone. Equality additionally requires control of the extra contribution. Without it, $e^{\gamma_{\mathrm{raw}}}$ bounds $\mathcal D$ from above rather than determining it. This claim concerns the annular von Neumann definition, not every Rényi or disk intercept.

Benchmark Values

The total quantum dimension gives a useful compact benchmark.

Theory or phase	Quantum dimensions	$\mathcal D$	Standard $\gamma$
trivial short-range-entangled phase	$d_1=1$	$1$	$0$
$\mathbb Z_2$ toric-code order	$d_1=d_e=d_m=d_\epsilon=1$	$2$	$\ln2$
Ising anyon theory	$d_1=d_\psi=1$, $d_\sigma=\sqrt2$	$2$	$\ln2$
bosonic Laughlin order, positive even $m$	$m$ Abelian sectors	$\sqrt m$	$\tfrac12\ln m$
Fibonacci anyon theory	$d_1=1$, $d_\tau=\varphi$	$\sqrt{1+\varphi^2}$	$\tfrac12\ln(1+\varphi^2)$

Here

$$\varphi = \frac{1+\sqrt5}{2}.$$

The toric-code and Ising rows expose an essential limitation:

$$\gamma_{\mathrm{toric}} = \gamma_{\mathrm{Ising}} = \ln2,$$

even though one theory is Abelian and the other contains a non-Abelian sector. Topological entanglement entropy determines one aggregate number, not the fusion rules, braiding matrices, topological spins, or chiral central charge.

Two subtraction geometries

Two region arrangements give different signs and factors. In the disk construction every nonempty union is simply connected, so its topological contribution is $-\gamma$; the seven-term sum leaves $-\gamma$. In the annulus, $B_1$ and $B_2$ separate $A$ from $C$ along both paths. The four-term von Neumann combination leaves $2\gamma$ in the standard vacuum setting. Boundaries are schematic: matched corners and junctions need controlled smoothing or lattice cancellation. Extra spurious contributions must be checked before setting $\gamma=\ln\mathcal D$.

Kitaev–Preskill Disk Subtraction

The entropy combination

Choose three adjacent regions $A$, $B$, and $C$ whose union is a topological disk. Every region and relevant union should have linear dimensions much larger than $\xi$. Define the tripartite information

$$I_3(A:B:C) ={} S_A+S_B+S_C - S_{AB}-S_{AC} - S_{BC}+S_{ABC},$$

where

$$S_{AB} := S(A\cup B),$$

and similarly for the other unions.

In the standard asymptotic geometry,

$$I_3(A:B:C) \longrightarrow -\gamma.$$

Some authors define the positive quantity

$$\gamma_{\mathrm{KP}} := -I_3(A:B:C).$$

Always state the sign convention. “Topological entropy” is used for both $I_3$ and its negative in the literature.

Why the boundary law cancels

Suppose the local contribution can be decomposed into pieces associated with interfaces and junction neighborhoods. Inclusion–exclusion assigns each local piece a net coefficient zero. For example, a segment separating $A$ from the exterior appears with the same total weight in positive and negative terms. The same is true for the corresponding smooth curvature corrections when the geometries are matched.

The topological constant behaves differently. Every nonempty simply connected region in the idealized configuration contributes $-\gamma$. There are three positive single-region terms, three negative double-region terms, and one positive triple-region term, so

$$(-\gamma) \left( 3-3+1 \right) = -\gamma.$$

The result survives because topology does not decompose into independent local contributions along each boundary segment.

Geometric requirements

A reliable Kitaev–Preskill calculation checks:

1. Every arm and interface is thick compared with $\xi$.
2. The union $A\cup B\cup C$ is contractible and lies in the bulk.
3. Corresponding boundary segments use the same lattice convention.
4. Triple-junction and corner neighborhoods are scaled or smoothed consistently.
5. No quasiparticle, defect, or physical boundary enters one entropy but not its cancellation partner.
6. Region sizes are varied while preserving shape.

The cancellation is geometric, not magical. If two nominally matched boundaries cut different microscopic bonds or encounter different edge physics, their nonuniversal terms need not cancel numerically.

Levin–Wen Annular Subtraction

Conditional mutual information

Let $A$, $B_1$, $C$, and $B_2$ form consecutive sectors of a thick annulus, and define

$$B = B_1\cup B_2.$$

Then $B$ separates $A$ from $C$ along both paths around the annulus. The conditional mutual information is

$$I(A:C\mid B) ={} S_{AB}+S_{BC} - S_B-S_{ABC}.$$

Strong subadditivity guarantees

$$I(A:C\mid B) \ge 0.$$

For the standard topological annulus in the asymptotic vacuum-sector setting,

$$I(A:C\mid B) \longrightarrow 2\gamma.$$

The factor of two reflects the two connected entangling boundaries of the annular geometry. An equivalent signed entropy combination tends to $-2\gamma$.

Why conditioning matters

In a generic short-range state, sufficiently thick $B_1$ and $B_2$ screen correlations between $A$ and $C$. One therefore expects

$$I(A:C\mid B) \longrightarrow 0$$

as all buffer widths exceed the correlation length. A topologically ordered state retains a global charge-consistency relation around the annulus. Local data in $B$ do not erase that nonlocal constraint, leaving the topological remnant.

This is a useful interpretation, but not a universal theorem for every gapped state and every partition. The annulus must be scaled, the buffers must be thick, and spurious long-range structures must be excluded.

Relation to the original annular prescription

The original Levin–Wen construction is often written using four region entropies $S_1,S_2,S_3,S_4$ chosen so that all local boundary pieces cancel. Region labels vary among sources. The conditional-information notation makes the sign and positivity transparent:

$$\gamma_{\mathrm{LW}} = \frac12 I(A:C\mid B).$$

The equation is meaningful only with the annular topology shown above. An arbitrary conditional mutual information elsewhere in the sample is not automatically a topological entropy.

Exact Toric-Code Boundary Counting

Fixed-point structure

The toric-code ground state is a clean benchmark because its correlation length vanishes at the exactly solvable point and the entanglement spectrum across a simple cut is flat. The Hamiltonian, loop operators, ground sectors, and anyons are developed on Topological Order Preview. Here only the reduced-state count is needed.

Consider a simply connected bulk region $A$. In a common lattice convention, let $N_{\partial}$ count independent binary boundary variables before the global topological-charge constraint is imposed. Local boundary entanglement would suggest

$$2^{N_{\partial}}$$

allowed boundary configurations. Closed-loop consistency removes one independent binary choice, leaving

$$\operatorname{rank}\rho_A = 2^{N_{\partial}-1}.$$

At the fixed point the nonzero eigenvalues are equal:

$$\lambda_j = 2^{-(N_{\partial}-1)}.$$

Therefore

$$\begin{aligned} S(A) &= -\sum_j\lambda_j\ln\lambda_j \\ &= (N_{\partial}-1)\ln2 \\ &= N_{\partial}\ln2-\ln2. \end{aligned}$$

The first term is the boundary law. The missing one bit gives

$$\gamma = \ln2.$$

Why one constraint is global

No bounded patch of the boundary determines the missing bit. It arises because the total topological charge crossing a contractible entangling loop must fuse to the vacuum when the disk contains no quasiparticle. The constraint couples boundary labels around the entire loop.

Changing the microscopic cut changes $N_{\partial}$ and the leading coefficient. It does not change the missing $\ln2$ in the controlled large-region limit.

Rényi check at the fixed point

Because the reduced state is flat on its support,

$$\operatorname{Tr}\rho_A^n = 2^{N_{\partial}-1} \left( 2^{-(N_{\partial}-1)} \right)^n.$$

Hence

$$S_n(A) = (N_{\partial}-1)\ln2$$

for every $n>0$. The topological term is Rényi-index independent at this fixed point. Generic finite-size corrections need not share that independence.

Anyon Content Inside the Region

Charge-sector shift

If the disk contains a definite anyon of type $a$, the universal constant changes. In the standard anyonic convention,

$$S_{\mathrm{top}}(a) = -\ln\mathcal D + \ln d_a.$$

Equivalently, define the positive sector-dependent deficit

$$\gamma_a = \ln\frac{\mathcal D}{d_a}.$$

Then

$$S(A) = \alpha \frac{L_{\partial A}}{a_0} - \gamma_a + \cdots.$$

For an Abelian anyon, $d_a=1$, so the disk constant is unchanged. For a non-Abelian anyon, $d_a>1$, and the entropy is larger than in the vacuum sector by $\ln d_a$.

The total enclosed charge is what matters

If several quasiparticles lie inside $A$, the relevant label is their total fusion channel, not merely the list of particle types. If the state is a coherent superposition or classical mixture of total-charge sectors, additional Shannon and fusion-space terms can appear. One should then retain the full sector probability distribution rather than assigning a single $d_a$ by inspection.

This is one reason phase identification belongs to a package of observables. A measured shift $\ln d_a$ can constrain quantum dimensions, but extracting fusion multiplicities and braiding requires richer data.

Connected Boundary Components

The number of connected components of the entangling surface matters. A disk in the plane has one circular component. An annulus has two. A cylindrical region cut from a torus also has two.

At a simple fixed point with vacuum constraints independently associated with $b$ boundary components, one may encounter a contribution resembling

$$-b\ln\mathcal D.$$

That mnemonic is useful but not a universal formula for arbitrary topology. Global charge constraints can correlate different components, and punctures, physical boundaries, defects, or nontrivial ground-state superpositions modify the answer. A reliable calculation derives the sector structure for the actual geometry.

The factor of two in

$$I(A:C\mid B) \longrightarrow 2\gamma$$

is a controlled example: the annular construction isolates two topological boundary contributions.

Direct Cylinder Extrapolation

One entangling circle

On a long or infinite cylinder of circumference $L_y$, cut the system into left and right half-cylinders. There is one circular entangling boundary. For a minimally entangled state with definite anyon flux $a$ through the cylinder, the expected asymptotic form is

$$S_a(L_y) = \alpha L_y - \ln\frac{\mathcal D}{d_a} + \delta_a(L_y).$$

For a gapped phase,

$$\delta_a(L_y) \longrightarrow 0$$

as $L_y/\xi\to\infty$, although the approach need not be a single exponential over accessible sizes.

For Abelian $\mathbb Z_2$ order, every $d_a=1$, so a minimally entangled sector gives

$$S_a(L_y) = \alpha L_y-\ln2+\cdots.$$

Two entangling circles

If a torus is divided into two cylinders, the boundary has two connected circles. For a definite minimally entangled flux sector and under the simplest assumptions,

$$S_a(L_y) = \alpha L_y - 2\ln\frac{\mathcal D}{d_a} + \cdots,$$

where $\alpha L_y$ includes both local cuts. Confusing the one-cut and two-cut geometries creates an immediate factor-of-two error.

Minimally entangled states

On a cylinder or torus, the ground space can contain several topological sectors. A minimally entangled state for a chosen cut is a ground-state basis vector with definite topological flux through the cycle dual to that cut. Such states make the universal constant directly interpretable in terms of $d_a$.

A generic superposition

$$\lvert\Psi\rangle = \sum_a c_a \lvert\Xi_a\rangle$$

need not have the same constant. Its reduced state contains sector probabilities

$$p_a = |c_a|^2,$$

and the entropy can acquire a classical contribution

$$H(\{p_a\}) = -\sum_a p_a\ln p_a,$$

together with sector-dependent quantum-dimension terms. The precise expression depends on the geometry and entropy index.

Why DMRG often helps

On long cylinders, density-matrix renormalization group calculations tend to favor low-entanglement representatives of a quasi-degenerate ground space. This can select a minimally entangled sector and make the intercept method practical. It is a tendency, not an identity. Near-degenerate optimization, initialization, bond dimension, cylinder length, and boundary pinning can select or mix sectors.

One should therefore diagnose the sector using loop observables, flux insertion, boundary conditions, or repeated initializations whenever possible.

Fit model and scaling window

A useful candidate fit is

$$S(L_y) = \alpha L_y - \gamma + \frac{b}{L_y} + c e^{-L_y/\xi_*} + \cdots.$$

The terms displayed are not mandatory or exhaustive. Their purpose is to show why a straight-line intercept from two or three small circumferences can be biased. A robust analysis:

• varies the minimum circumference included;
• compares linear and controlled correction models;
• reports covariance between slope and intercept;
• checks $L_y/\xi$ rather than only $L_y$;
• repeats the calculation at larger bond dimension;
• and tests more than one cylinder length or boundary termination.

For known $\mathbb Z_2$ benchmarks, the von Neumann entropy can converge accurately once the circumference is many correlation lengths, while higher Rényi entropies may retain much larger finite-size corrections. That empirical scale is a benchmark, not a universal guarantee.

Rényi Topological Terms

Definition

For a disk or subtraction geometry, define the corresponding Rényi combination by replacing every von Neumann entropy with $S_n$. The resulting four-entropy combination is not generally a nonnegative conditional mutual information: ordinary Rényi entropies do not satisfy strong subadditivity. Neither the von Neumann positivity check nor the annular lower-bound theorem transfers automatically to $n\ne1$. At many nonchiral topological fixed points,

$$\gamma_n = \ln\mathcal D$$

is independent of $n$.

This does not mean the full Rényi entropy is index independent. The boundary coefficient generally satisfies

$$\alpha_n \ne \alpha_m$$

for $n\ne m$.

Fixed-point universality versus finite-size convergence

Separate the asymptotic statement

$$\lim_{L/\xi\to\infty} \gamma_n(L) = \ln\mathcal D$$

from the finite-size estimate $\gamma_n(L)$. Different Rényi indices weight the eigenvalues of $\rho_A$ differently. For $n>1$, the largest eigenvalues dominate, and short-distance distortions of the entanglement spectrum can produce large intercept errors even when ordinary correlation functions appear converged.

Agreement among several $n$ values is a useful consistency check. Disagreement is a diagnostic of finite size, sector mixing, numerical bias, or physics beyond the assumed fixed-point class. It should not be hidden by averaging the indices.

Replica correlation length

Rényi constructions probe replicated density matrices. The length scale governing their finite-size corrections can be much larger than the ordinary two-point correlation length. A state can therefore look short ranged to local correlators while an $n\ge2$ topological-Rényi extrapolation remains contaminated.

This motivates a conservative hierarchy:

$$\begin{gathered} L\gg\xi_{\mathrm{two\ point}} \\ \text{is necessary but may not be sufficient} \\ \text{for Rényi convergence.} \end{gathered}$$

Numerical convergence and cancellation errors

Choose the state class, topological sector, spatial cut convention, and region geometry before interpreting an intercept. Vary every narrow arm or buffer as well as the outer size. A large disk with a neck of fixed microscopic width is not the required large-region limit.

Keep the individual entropies as well as their combination. For $Q=\sum_i c_iS_i$, statistical error propagation gives

$$\sigma_Q^2=\sum_{i,j}c_ic_j\,\operatorname{Cov}(S_i,S_j).$$

If the estimates are independent, this reduces to $\sum_i c_i^2\sigma_i^2$. Dropping unknown covariances is not automatically conservative. A covariance-independent upper bound on the standard deviation is $\sigma_Q\le\sum_i|c_i|\sigma_i$. Systematic state-truncation and finite-size errors require separate checks.

Increase bond dimension, sample count, and solver accuracy while holding the entropy index fixed. Increasing the Rényi index changes the observable; it is not a convergence control. Check positivity of entropy, pure-state complement equality, and nonnegative von Neumann conditional mutual information. Scale the geometry and vary fit windows until the small residual is stable despite cancellation of much larger terms.

Finally compare with a bulk gap, ground-state sectors, local indistinguishability, loop observables, or braiding information, as appropriate. Topological order explains what these independent tests establish.

Finite-Size and Geometry Contaminants

Correlation-length crossover

If a region width is comparable to $\xi$, opposite boundary segments communicate through ordinary short-range correlations. Then the entropy is not a sum of independent local boundary terms plus a topological constant. A schematic contamination is

$$\delta S \sim c_1 e^{-w/\xi} + c_2 e^{-r/\xi} + \cdots.$$

Near a phase transition, $\xi$ grows and a geometry that was adequate deep in the phase may cease to be adequate.

Corners and junctions

Corners can contribute constants or logarithms. A subtraction cancels them only when corresponding angles and microscopic placements match. Triple junctions deserve the same care. On a lattice, rotating or translating a region can change the number and type of cut bonds even when the continuum shapes look identical.

Physical edges

A physical edge can carry gapless modes, symmetry-protected modes, or a gapped boundary condition with its own topological data. If an entangling region touches that edge, the constant is a boundary topological entropy, not automatically the bulk disk value $\ln\mathcal D$.

Aspect ratio and thin handles

Topological formulas assume every neck and annular thickness remains macroscopic. Sending the outer radius to infinity while keeping a fixed thin neck does not realize the intended scaling limit. Report all independent aspect ratios.

False Positives and Incomplete Invariants

Spurious cylinder intercepts

Short-range-entangled two-dimensional states can yield

$$S_n(L_y) = \alpha_n L_y-\gamma_n$$

with an apparently nonzero $\gamma_n$ for $n\ge2$ over cylinder extrapolations. The effect can be controlled by a replica correlation length much larger than the ordinary correlation length. A cylinder intercept alone is therefore not a theorem of topological order.

Subsystem-symmetry contributions

Special subsystem symmetries or long string-order patterns can survive Kitaev–Preskill or Levin–Wen cancellation even in short-range-entangled states. Such contributions may persist at zero ordinary correlation length. Varying the partition orientation, breaking accidental subsystem symmetries, and comparing with a circuit-invariant diagnostic can expose the problem.

Symmetry-breaking cat states

In a finite system, a symmetry-preserving superposition of macroscopically distinct branches can contribute an $O(1)$ entropy to many regions. That constant reflects a shared branch label, not intrinsic topological order. Add a weak symmetry-breaking field, compare branch-selected states, and inspect a local order parameter before assigning a topological meaning.

Invertible topological phases

A vanishing anyonic topological entropy means

$$\mathcal D=1$$

within the assumed anyon framework. It does not imply that the phase is completely trivial. Invertible phases can possess chiral response or protected boundary structure while having no nontrivial anyon sectors and hence $\gamma=0$.

Thus

$$\begin{gathered} \gamma=0 \\ \nRightarrow \\ \text{no topological physics}. \end{gathered}$$

Equal constants, inequivalent phases

Because $\gamma$ only determines $\mathcal D$, distinct anyon theories can share it. The toric-code and Ising examples both have $\mathcal D=2$. More generally, $\mathcal D$ does not determine:

• the number of anyon types;
• which anyons are Abelian;
• fusion multiplicities;
• braiding phases;
• topological spins;
• modular matrices;
• chiral central charge.

Gauge-Theory Factorization

Why the ordinary tensor product can fail

In a lattice gauge theory, Gauss-law constraints couple degrees of freedom across a spatial boundary. The physical Hilbert space need not factorize as

$$\mathcal H_{\mathrm{phys}} \stackrel{?}{=} \mathcal H_A \otimes \mathcal H_{\bar A}.$$

One can instead use an extended Hilbert space with edge modes, or assign an operator algebra to the region and specify its center. Different choices redistribute entropy into boundary Shannon, representation-dimension, and distillable-correlation pieces.

What remains universal

Many factorization ambiguities are local to the entangling boundary. Carefully matched mutual-information or subtraction constructions can cancel them in an appropriate continuum or lattice limit. That cancellation must be shown for the chosen algebra; it should not be assumed from notation alone.

A gauge-theory result should state:

• which links or operators belong to $A$;
• whether an extended Hilbert space is used;
• the boundary center or edge-mode convention;
• whether superselection-sector Shannon terms are included;
• which part of the result is invariant under changing that convention.

Entanglement in field theory discusses these spatial-factorization issues: a spatial partial trace is part of the physical definition, not a harmless implementation detail.

Temperature and Mixed States

The ground-state formula does not transfer unchanged to

$$\rho_\beta = \frac{e^{-\beta H}}{Z}.$$

A thermal state contains both quantum and classical correlations, mobile quasiparticles, and extensive entropy. Mutual-information combinations can cancel leading volume terms, but the remainder is a mixed-state diagnostic with its own order of limits.

For two-dimensional toric-code-like systems at any fixed nonzero temperature, thermally excited anyons proliferate in the thermodynamic limit and destroy the zero-temperature topological memory. Finite samples can nevertheless show crossover plateaus on length scales shorter than the thermal defect spacing. One must state whether the limit is

$$L\to\infty \quad\text{before}\quad T\to0,$$

or the reverse. The two need not agree.

Mixed-state topological order, topological negativity, recoverability, and finite-temperature memory are separate research topics. A ground-state TEE value should not be advertised as a finite-temperature lifetime.

Critical and Gapless Systems

At a quantum critical point, a Fermi surface, or a phase with Goldstone modes, the entropy can contain universal logarithms and geometry-dependent constants unrelated to intrinsic gapped topological order. The assumed expansion

$$S(A) = \alpha L_{\partial A}-\gamma+o(1)$$

may be incomplete.

Examples of competing terms include

$$\beta\ln L_{\partial A},$$

corner functions, and shape-dependent universal constants. Before applying a topological subtraction, verify the bulk gap and test whether the candidate plateau survives moving away from the critical regime.

Entanglement and criticality explains critical scaling, while Area Laws compares the principal violations and subleading structures.

Worked Example: Inclusion–Exclusion on a Disk

Assume every nonempty simply connected union $X$ in a Kitaev–Preskill geometry has

$$S_X = L_X-\gamma,$$

where $L_X$ denotes the complete local boundary functional, not merely Euclidean perimeter. Then

$$I_3 ={} L_A+L_B+L_C - L_{AB}-L_{AC}-L_{BC} +L_{ABC} - \gamma(3-3+1).$$

Geometric matching gives

$$0 ={} L_A+L_B+L_C-L_{AB} - L_{AC}-L_{BC}+L_{ABC}.$$

Therefore

$$I_3 = -\gamma.$$

The derivation explains both the power and the limitation of the protocol. If the local functionals fail to match, the first line does not vanish. If one of the unions has different topology or charge content, its constant is not the same $-\gamma$.

Worked Example: Toric Code Versus Ising Anyons

For toric-code order,

$$\mathcal D_{\mathrm{TC}} = \sqrt{ 1^2+1^2+1^2+1^2 } = 2.$$

For Ising anyons,

$$\mathcal D_{\mathrm{Ising}} = \sqrt{ 1^2+1^2+(\sqrt2)^2 } = 2.$$

Thus both give

$$\gamma = \ln2.$$

Yet the spectra of quantum dimensions differ. The Ising theory contains a non-Abelian anyon $\sigma$ with $d_\sigma=\sqrt2$, while all toric-code anyons are Abelian. The equal TEE is a demonstration of information loss, not an accidental numerical coincidence.

Worked Example: Anyon-Sector Shift

Suppose a disk in an Ising topological phase encloses a definite $\sigma$ charge. The vacuum-sector deficit is

$$\gamma_1 = \ln2.$$

For the $\sigma$ sector,

$$\begin{aligned} \gamma_\sigma &= \ln\frac{\mathcal D}{d_\sigma} \\ &= \ln\frac{2}{\sqrt2} \\ &= \frac12\ln2. \end{aligned}$$

The universal constant in $S(A)$ is therefore less negative by

$$\ln d_\sigma = \frac12\ln2.$$

This shift diagnoses a quantum dimension only if the enclosed topological charge is known and the physical boundary convention is unchanged.

Worked Example: Fit-Window Drift

Consider synthetic one-cut cylinder data generated by

$$S(L) = 0.4L - \ln2 + 0.8e^{-L/2}.$$

A two-point straight-line fit using $L=4$ and $L=6$ gives approximately

$$S(L) \approx 0.3658L-0.4480.$$

The inferred topological entropy would be only $0.4480$, far below

$$\ln2 \approx 0.693.$$

Using $L=10$ and $L=12$ instead gives approximately

$$S(L) \approx 0.3983L-0.6707.$$

The later window is closer but still not exact. The example shows why a plausible straight line does not establish the asymptotic regime; the intercept can drift strongly while the slope looks nearly converged.

Exercises

1. Kitaev–Preskill sign

Assume all seven nonempty regions in the Kitaev–Preskill construction are simply connected and each has a topological constant $-\gamma$. Compute the topological contribution to

$$I_3 ={} S_A+S_B+S_C-S_{AB} - S_{AC}-S_{BC}+S_{ABC}.$$

Solution

The three single-region terms contribute

$$-3\gamma.$$

The three double-region entropies enter with a minus sign, so their constants contribute

$$-3(-\gamma) = +3\gamma.$$

The triple union contributes $-\gamma$. Therefore

$$I_3^{\mathrm{top}} = -3\gamma+3\gamma-\gamma = -\gamma.$$

The positive convention is $\gamma_{\mathrm{KP}}=-I_3$.

2. Quantum-dimension benchmarks

Compute $\mathcal D$ and $\gamma$ for:

1. four Abelian sectors;
2. $m$ Abelian sectors;
3. sectors with dimensions $1,1,\sqrt2$.

Solution

For four Abelian sectors,

$$\mathcal D = \sqrt{4} = 2, \qquad \gamma = \ln2.$$

For $m$ Abelian sectors,

$$\mathcal D = \sqrt m, \qquad \gamma = \frac12\ln m.$$

For dimensions $1,1,\sqrt2$,

$$\mathcal D = \sqrt{1+1+2} = 2,$$

so again $\gamma=\ln2$. The first and third answers have equal TEE but different anyon content.

3. Fibonacci charge in a disk

For Fibonacci anyons, $d_1=1$ and $d_\tau=\varphi$, where $\varphi^2=\varphi+1$. Find the vacuum deficit $\gamma_1$, the $\tau$-sector deficit $\gamma_\tau$, and their difference.

Solution

The total quantum dimension is

$$\mathcal D = \sqrt{1+\varphi^2} = \sqrt{\varphi+2}.$$

Therefore

$$\gamma_1 = \ln\mathcal D = \frac12\ln(\varphi+2).$$

For a definite $\tau$ charge,

$$\gamma_\tau = \ln\frac{\mathcal D}{\varphi}.$$

Their difference is

$$\gamma_1-\gamma_\tau = \ln\varphi.$$

Thus the entropy of the $\tau$ sector is larger than the vacuum-sector entropy by $\ln\varphi$ for the same local boundary geometry.

4. Annular positivity and normalization

Suppose an annular calculation gives

$$I(A:C\mid B) = 1.3861$$

in nats. Under the standard Levin–Wen normalization, estimate $\gamma$ and $\mathcal D$. Is the sign consistent with strong subadditivity?

Solution

The normalization is

$$I(A:C\mid B) = 2\gamma.$$

Hence

$$\gamma \approx 0.69305 \approx \ln2,$$

and

$$\mathcal D = e^\gamma \approx 2.$$

The conditional mutual information is positive, as required by strong subadditivity. This value supports $\mathcal D\approx2$ under the protocol assumptions; it does not identify the complete anyon theory.

5. One cut or two?

A minimally entangled Abelian sector has $\mathcal D=2$. Write the expected universal constant for:

1. a half-infinite-cylinder bipartition with one circular cut;
2. a torus partitioned into two cylinders with two circular cuts.

Solution

For one cut,

$$S(L_y) = \alpha L_y-\ln2+\cdots.$$

For two cuts,

$$S(L_y) = \alpha L_y-2\ln2+\cdots,$$

where the second $\alpha$ convention includes both local boundaries. The intercept magnitudes differ by a factor of two.

6. Fit-window bias

For

$$S(L) = 0.4L-\ln2+0.8e^{-L/2},$$

perform two-point linear fits on $L=4,6$ and on $L=10,12$. Compare the inferred $\gamma$ values with $\ln2$.

Solution

The first pair gives approximately

$$\begin{aligned} S(4)&\approx1.0151, \\ S(6)&\approx1.7467. \end{aligned}$$

The fitted slope is about $0.3658$ and the intercept about $-0.4480$, so

$$\gamma_{4,6} \approx 0.4480.$$

The later pair gives

$$\begin{aligned} S(10)&\approx3.3122, \\ S(12)&\approx4.1088. \end{aligned}$$

The fitted slope is about $0.3983$ and the intercept about $-0.6707$, so

$$\gamma_{10,12} \approx 0.6707.$$

Both estimates are below

$$\ln2 \approx 0.6931,$$

but the later window is much closer. A sequence of fit windows is needed to see the drift.

7. Diagnose the claim

A short-range-entangled state has zero ordinary two-point correlation length. A Rényi-$2$ cylinder fit nevertheless returns a stable-looking negative intercept corresponding to positive $\gamma_2$ over the available circumferences. Does this establish intrinsic topological order? Name three further checks.

Solution

No. A Rényi cylinder intercept can be contaminated by a replica correlation length or subsystem-symmetry structure even when ordinary two-point correlations vanish.

Useful checks include:

• repeat the calculation with the von Neumann entropy;
• enlarge the circumference and test moving fit windows;
• use Kitaev–Preskill and annular geometries;
• rotate or deform the partition;
• break accidental subsystem symmetries;
• test loop sectors, local indistinguishability, and topology-dependent ground states;
• compare with a circuit-invariant entropic diagnostic.

Agreement among independent protocols is substantially stronger than the original intercept.

8. Combine independent tests

Design a minimal set of numerical tests for a proposed $\mathbb Z_2$ spin liquid on a cylinder. Include at least one entanglement test, one bulk test, one sector test, and one convergence test. State which properties of the phase these tests leave undetermined.

Solution

One useful set is:

1. Entanglement: demonstrate a stable one-cut intercept near $-\ln2$ and, if feasible, an independent disk or annular subtraction near the corresponding value.
2. Bulk: extrapolate a nonzero singlet and spin gap and show that local spin and valence-bond correlations are short ranged without a hidden ordering plateau.
3. Sector: prepare or diagnose distinct flux sectors using boundary conditions or loop operators, and verify that local bulk observables agree between them up to finite-size corrections.
4. Convergence: increase cylinder length, circumference, bond dimension, and fit-window minimum; report entropy and correlation-length drift rather than energy convergence alone.

The microscopic phase diagram, competing orders, response properties, excitation content, and anyon statistics still require separate study. These entropy and sector tests constrain intrinsic order but do not determine all those properties.

Further Reading and Cross-Links

• Topological Order Preview for the phase-level package of local indistinguishability, ground sectors, loops, and anyons.
• Topological Order for the classification package that combines $\mathcal D$ with modular, fusion, ground-space, charge, and thermal data.
• Anyons and Braiding for the fusion and braid data that one entropy constant cannot reconstruct.
• Area Laws for leading boundary scaling and its exceptions.
• Entanglement Spectrum for level-resolved information beyond the single entropy constant.
• Mutual Information in Many-Body Systems for general correlation and cancellation properties.
• Matrix Product States Preview and Tensor Networks Preview for numerical state representations and truncation diagnostics.
• Quantum Phase Transitions for gap closing and finite-size drift near phase boundaries.
• Topological Invariants for the mathematical distinction between invariant data and complete classification.
• Surface Code for a compact code-family reference related to $\mathbb Z_2$ topological order.
• Condensed Matter Roadmap for the staged route into band topology and interacting topological phases.

References

• P. Bonderson, C. Knapp, and K. Patel, “Anyonic Entanglement and Topological Entanglement Entropy”, Annals of Physics 385, 399–468 (2017).
• H. Casini, M. Huerta, and J. A. Rosabal, “Remarks on Entanglement Entropy for Gauge Fields”, Physical Review D 89, 085012 (2014).
• C. Castelnovo and C. Chamon, “Entanglement and Topological Entropy of the Toric Code at Finite Temperature”, Physical Review B 76, 184442 (2007).
• X. Chen, Z.-C. Gu, and X.-G. Wen, “Local Unitary Transformation, Long-Range Quantum Entanglement, Wave Function Renormalization, and Topological Order”, Physical Review B 82, 155138 (2010).
• S. Dong, E. Fradkin, R. G. Leigh, and S. Nowling, “Topological Entanglement Entropy in Chern–Simons Theories and Quantum Hall Fluids”, Journal of High Energy Physics 2008, 016 (2008).
• W. Donnelly, “Decomposition of Entanglement Entropy in Lattice Gauge Theory”, Physical Review D 85, 085004 (2012).
• S. T. Flammia, A. Hamma, T. L. Hughes, and X.-G. Wen, “Topological Entanglement Rényi Entropy and Reduced Density Matrix Structure”, Physical Review Letters 103, 261601 (2009).
• A. Hamma, R. Ionicioiu, and P. Zanardi, “Bipartite Entanglement and Entropic Boundary Law in Lattice Spin Systems”, Physical Review A 71, 022315 (2005).
• S. V. Isakov, M. B. Hastings, and R. G. Melko, “Topological Entanglement Entropy of a Bose–Hubbard Spin Liquid”, Nature Physics 7, 772–775 (2011).
• H.-C. Jiang, Z. Wang, and L. Balents, “Identifying Topological Order by Entanglement Entropy”, Nature Physics 8, 902–905 (2012).
• H.-C. Jiang, R. R. P. Singh, and L. Balents, “Accuracy of Topological Entanglement Entropy on Finite Cylinders”, Physical Review Letters 111, 107205 (2013).
• K. Kato and P. Naaijkens, “An Entropic Invariant for 2D Gapped Quantum Phases”, Journal of Physics A: Mathematical and Theoretical 53, 085302 (2020).
• I. H. Kim, M. Levin, T.-C. Lin, D. Ranard, and B. Shi, “Universal Lower Bound on Topological Entanglement Entropy”, Physical Review Letters 131, 166601 (2023).
• A. Kitaev and J. Preskill, “Topological Entanglement Entropy”, Physical Review Letters 96, 110404 (2006).
• A. Y. Kitaev, “Fault-Tolerant Quantum Computation by Anyons”, Annals of Physics 303, 2–30 (2003).
• M. Levin and X.-G. Wen, “Detecting Topological Order in a Ground State Wave Function”, Physical Review Letters 96, 110405 (2006).
• M. A. Levin and X.-G. Wen, “String-Net Condensation: A Physical Mechanism for Topological Phases”, Physical Review B 71, 045110 (2005).
• M. Levin, “Physical Proof of the Topological Entanglement Entropy Inequality”, Physical Review B 110, 165154 (2024).
• C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, “Non-Abelian Anyons and Topological Quantum Computation”, Reviews of Modern Physics 80, 1083–1159 (2008).
• D. J. Williamson, A. Dua, and M. Cheng, “Spurious Topological Entanglement Entropy from Subsystem Symmetries”, Physical Review Letters 122, 140506 (2019).
• Y. Zhang, T. Grover, and A. Vishwanath, “Topological Entanglement Entropy of $\mathbb Z_2$ Spin Liquids and Lattice Laughlin States”, Physical Review B 84, 075128 (2011).
• Y. Zhang, T. Grover, A. Turner, M. Oshikawa, and A. Vishwanath, “Quasiparticle Statistics and Braiding from Ground-State Entanglement”, Physical Review B 85, 235151 (2012).
• L. Zou and J. Haah, “Spurious Long-Range Entanglement and Replica Correlation Length”, Physical Review B 94, 075151 (2016).