Renormalized Entanglement and Scheme Dependence

Renormalized entanglement entropy is a prescription, not a single automatic number. Local counterterm subtraction, differential operators in the region size, and mutual-information regulators can remove the same leading divergences while retaining different finite local terms or geometric conventions. A comparison is meaningful only after the scheme, region family, and endpoint normalization are fixed.

Required background. Regulator removal and renormalized predictions supplies the distinction between a regulator and a renormalization scheme; separating scales fixes the limiting procedure.

Helpful background. Entropy counterterms and renormalization classifies local entangling-surface terms.

Local divergences and finite ambiguities

For a smooth entangling surface $\Sigma$ in $d$ spacetime dimensions, a regulated entropy has the schematic expansion

$$S_\epsilon(A) =\sum_k \frac{c_k[\Sigma]}{\epsilon^{d-2-k}} +c_{\log}[\Sigma]\log(\mu\epsilon) +S_{\rm finite}(A;\mu).$$

The coefficients $c_k[\Sigma]$ are integrals of local intrinsic and extrinsic geometry, masses, and couplings. Removing divergences by local surface counterterms can leave finite local counterterms. Consequently, $S_{\rm finite}$ is not universally scheme independent. Universal logarithmic coefficients at fixed points, or carefully chosen differences such as mutual information of separated regions, are more robust.

The structure diagram shows where the scheme choice enters: before any monotonicity or crossover claim.

Scheme choices belong to the fixed-observable stage. Differential, counterterm, and mutual-information prescriptions may agree on fixed-point anchors while differing in finite crossover terms. Schematic and not to scale.

Differential prescriptions

For a one-parameter family of spheres or disks, a polynomial in $R\partial_R$ can annihilate local power terms. In 2+1 dimensions,

$$\mathcal F(R)=(R\partial_R-1)S(R)$$

removes the perimeter divergence. More generally one writes

$$\mathcal S_d(R)=P_d(R\partial_R)S(R),$$

where the roots of $P_d$ are chosen to kill the powers allowed in the selected geometry. This is convenient and local in scale, but derivatives amplify numerical noise and the result is tied to the one-parameter family.

A finite counterterm whose $R$ dependence survives $P_d$ changes $\mathcal S_d$. Thus “cutoff independent” does not always mean “scheme independent.” One must check the allowed local terms in the dimension and theory under study.

The differential operators and their dimension-dependent roots are given in Liu and Mezei 2013, § 2.

Counterterm subtraction

Replica effective actions organize divergences as local terms on the conical defect. Renormalizing the bulk gravitational couplings and allowed defect terms yields a finite entropy in a chosen scheme. This method keeps contact terms and geometric dependence explicit, and it is natural in curved backgrounds.

The local counterterm structure and its geometric coefficients are reviewed in Solodukhin 2011, §§ 3 and 6.

Its cost is that a finite renormalization can shift the answer. Matching conditions—such as a fixed CFT normalization, vanishing in a trivial gapped phase, or agreement with a separated-region regulator—are needed to compare calculations. Counterterms must be fixed once for the whole family; fitting a new constant independently at every $R$ erases the flow information.

Mutual-information regulator

Replace a sharp boundary by two nearby boundaries separated by $\delta$ and form a mutual information. The leading local divergences cancel because the regions are disjoint. One can then take $\delta\to0$ and identify divergent and finite pieces.

This construction has an operationally clear finite quantity at nonzero separation, but the map from its small-$\delta$ expansion to a “renormalized entropy” depends on how the split surfaces approach one another. Corners, curvature, and state dependence matter. It is a regulator family, not a proof that all finite counterterm schemes coincide.

The cancellation and residual geometry dependence for separated regions are worked out in Casini and Huerta 2009, §§ 2–3.

Comparing three schemes on one dataset

For free-field sphere data, a fair comparison uses identical values of $mR$, the same continuum extrapolation, and one endpoint normalization. Then compute:

1. the differential quantity from a correlated fit to $S(R)$;
2. a counterterm-subtracted finite part with coefficients fixed globally; and
3. a mutual-information construction with several split distances $\delta/R$.

Agreement of universal endpoint constants is expected. Disagreement in the crossover may be allowed and should be reported as scheme dependence rather than averaged away. If the difference equals an allowed local function of $mR$, it is not a numerical failure.

Validity map

Validity map for renormalized entropy. Changing a subtraction or split-surface family changes the observable family. Such curves can be compared only after the finite local ambiguity and endpoint matching are made explicit. Schematic and not to scale.

Common pitfalls

Equating finiteness with universality. A finite local counterterm can shift a finite answer. Identify which coefficients are protected.

Differentiating unconverged data. Differential operators amplify lattice and interpolation errors. Extrapolate or fit with the full covariance and repeat under cutoff refinement.

Changing the scheme along the flow. A scale-dependent fitting convention can manufacture monotonicity. Fix counterterms and geometry once.

References

• Casini, Horacio, and Marina Huerta. “Remarks on the Entanglement Entropy for Disconnected Regions.” Journal of High Energy Physics 2009, no. 3 (2009): 048. DOI.
• Liu, Hong, and Mark Mezei. “A Refinement of Entanglement Entropy and the Number of Degrees of Freedom.” Journal of High Energy Physics 2013, no. 4 (2013): 162. DOI.
• Solodukhin, Sergey N. “Entanglement Entropy of Black Holes.” Living Reviews in Relativity 14 (2011): 8. DOI.