Universal Terms and Entangling-Surface Geometry

Entanglement entropy contains ultraviolet terms tied to the entangling surface and, in special settings, a logarithmic coefficient or finite remainder that survives a change of regulator. The important distinction is between the regulated entropy and the surviving coefficient. A sharp continuum region is described by a local algebra, usually of type III, rather than by a tensor factor with a trace-class reduced density matrix. In plain language, continuum fields have degrees of freedom at arbitrarily short distances, so the usual finite-system reduced-density-matrix entropy is not defined without a cutoff or controlled split. Thus $S_\Sigma^{\mathcal R}(\epsilon)$ presupposes a UV regulator or a split/type-I replacement. The universal datum can nevertheless be intrinsic to the continuum theory when it is invariant under the declared class of such replacements. Witten 2018, §§6.1–6.5, pp. 57–64 explains this algebraic distinction.

Required background. Use anomaly coefficients and central charges to identify the CFT data in logarithmic terms, conformal geometry and maps to follow the sphere map, UV divergences and the area law to enumerate local subtractions, and mutual information to recognize a comparison that does not require assigning a sharp-region entropy. Helpful background. Defect entropy and monotonicity supplies examples of defect data, while entropy counterterms makes the allowed scheme freedom explicit.

Local terms and the invariant remainder

Let $\Sigma$ be a smooth entangling surface of characteristic size $L$ in $d$ spacetime dimensions. A local regulator $\mathcal R$ can produce the schematic expansion

$$S_\Sigma^{\mathcal R}(\epsilon) =\sum_{p=1}^{d-2}\frac{c_p^{\mathcal R}[\Sigma]}{\epsilon^p} +s_{\log}[\Sigma]\log\frac{L}{\epsilon} +c_0^{\mathcal R}[\Sigma]+o(1).$$

The sum is empty in $d=2$, and symmetries can force some of the displayed coefficients to vanish. The induced metric $\gamma_{ab}$ is the metric obtained by restricting spacetime distances to $\Sigma$. The extrinsic curvatures $K^A_{ab}$ measure how $\Sigma$ bends in each normal direction $A$, while ambient curvature is the spacetime curvature evaluated on $\Sigma$. Local divergent coefficients are integrals over $\Sigma$ of scalars built from these objects. When an Euler density appears, it is a particular intrinsic-curvature combination whose integral on a closed even-dimensional surface depends only on its topology.

For the vacuum of a CFT across a smooth closed surface, the standard symmetry-preserving classification depends on the parity of the spacetime dimension. In even $d$, the logarithmic coefficient $s_{\log}[\Sigma]$ is the universal datum, while an additive constant can change when the cutoff is rescaled or a permitted finite local term is added. In odd $d$, a smooth surface has no conformal-anomaly logarithm, and the constant $c_0^{\rm univ}[\Sigma]$ left after subtracting local powers is the universal datum within the declared symmetry-preserving scheme class. A finite constant can be shifted only when the theory and that scheme class actually admit a dimensionless local surface invariant; one must exhibit such an allowed counterterm rather than assume that every $c_0$ is arbitrary. Singular surfaces, boundaries, masses, and parity-odd contact terms can change this classification and must be treated separately.

A candidate $U[\Sigma]$ is universal only if it is unchanged when one:

1. keeps the theory, state, algebra, and physical geometry fixed;
2. changes between regulators that approach that same continuum problem;
3. adds every symmetry-allowed local surface counterterm in the declared scheme class; and
4. refits all local divergent terms before comparing the remainder.

The chapter structure map shows where this geometric question sits among the chapter’s other information measures. The following compact table gives the most common smooth examples; the broader canonical comparison table records their state, algebra, and regulator assumptions.

Setting	Regulated form	Invariant datum	Scheme-dependent part
One interval in a $d=2$ CFT vacuum	$\frac{c}{3}\log(\ell/\epsilon)+b_{\mathcal R}$	Coefficient $c/3$	$b_{\mathcal R}$
Round sphere in an even-$d$ CFT vacuum	local powers $+(-1)^{d/2-1}4A\log(R/\epsilon)+\cdots$	Type-A anomaly coefficient $A$ in the stated normalization	Power terms and an additive constant
Round disk in a $d=3$ CFT vacuum	$\alpha_{\mathcal R}R/\epsilon-F+o(1)$	$F=-\log\lvert Z_{S^3}\rvert=-\operatorname{Re}\log Z_{S^3}$ after symmetry-preserving subtraction	$\alpha_{\mathcal R}$
General smooth shape in an even-$d$ CFT vacuum	local powers $+s_{\log}[\Sigma]\log(L/\epsilon)+c_0^{\mathcal R}[\Sigma]+\cdots$	$s_{\log}[\Sigma]$, including intrinsic and extrinsic geometry	Power terms and $c_0^{\mathcal R}$
General smooth shape in an odd-$d$ CFT vacuum	local powers $+c_0^{\rm univ}[\Sigma]+o(1)$	$c_0^{\rm univ}[\Sigma]$ in the stated symmetry and scheme class	Local power terms

The even-dimensional sphere coefficient follows from the ball-to-hyperbolic-space map and the integrated trace anomaly Casini, Huerta, and Myers 2011, §4.2, eq. (4.16), p. 29, PDF. For the odd-dimensional sphere, the same map identifies the universal entropy constant with the real sphere free energy; the absolute value above makes a possible partition-function phase irrelevant to this real quantity. For a general smooth surface in four dimensions, both intrinsic topology and extrinsic geometry enter the logarithm Solodukhin 2008, §2, eq. (2.11), pp. 3–4, PDF. None of these statements transfers unchanged to a singular surface or to a theory with an additional physical scale.

Two matched regulators for a free Dirac interval

Consider the vacuum of a massless Dirac CFT on the line, for which $c=1$, and intervals of physical lengths $\ell$ and $2\ell$. Endpoint point splitting gives

$$S_{\rm ps}(\ell,\epsilon) =\frac13\log\frac{\ell}{\epsilon}+b_{\rm ps}, \qquad q_{\rm ps} =\frac{S_{\rm ps}(2\ell,\epsilon)-S_{\rm ps}(\ell,\epsilon)}{\log 2} =\frac13.$$

The interval formula and its regulator-dependent additive constant follow from the twist-field two-point function Calabrese and Cardy 2004, §III.A, eqs. (16)–(19), pp. 9–10, PDF.

For an independent type-I regulator, take the half-filled infinite nearest-neighbor fermion chain with lattice spacing $a$ and an interval of $N=\ell/a$ sites. Its restricted correlation matrix is

$$(C_N)_{ij} =\begin{cases} \dfrac12, & i=j,\\[4pt] \dfrac{\sin[\pi(i-j)/2]}{\pi(i-j)}, & i\ne j, \end{cases} \qquad S_N=-\operatorname{tr}\!\left[C_N\log C_N+(1-C_N)\log(1-C_N)\right].$$

This is a finite matrix computation of a genuine reduced density matrix; the correlation-matrix construction is derived in Peschel 2003, eqs. (5)–(13), pp. L206–L208. Match the two cutoff lengths by $\epsilon=\kappa a$, where the finite calibration $\kappa$ only changes the additive constant. Direct diagonalization gives

$N$	$q_N=(S_{2N}-S_N)/\log 2$	$u_N=\lvert q_N-q_{N/2}\rvert$
32	0.333350999	$5.37\times10^{-5}$
64	0.333337739	$1.33\times10^{-5}$
128	0.333334434	$3.31\times10^{-6}$

Here $u_N$ is a transparent refinement estimate, not a statistical error bar. At $N=128$, the lattice result differs from $1/3$ by $1.10\times10^{-6}$, smaller than $u_N$. The point-split and lattice regulators therefore agree on the logarithmic coefficient while leaving their additive constants scheme dependent and unconstrained. The matrix, estimator, and refinement rule make the comparison reproducible rather than dependent on a fit chosen after seeing the answer.

Adversarial scheme changes

Now deliberately try to destroy the conclusion. A constant rescaling of the point-splitting length gives

$$\epsilon\longmapsto e^\eta\epsilon \quad\Longrightarrow\quad b_{\rm ps}\longmapsto b_{\rm ps}-\frac{\eta}{3},$$

and a finite endpoint counterterm $\Delta S=\lambda\int_{\partial A}1=2\lambda$ shifts the constant by $2\lambda$. Neither operation changes $q_{\rm ps}$. A smooth regulator deformation such as $\epsilon\mapsto\epsilon[1+\alpha(\epsilon/\ell)^r]$ with $r>0$ changes only terms that vanish as $\epsilon/\ell\to0$. By contrast, inserting an explicitly $L$-dependent “counterterm” would change the physical observable rather than merely change a local UV scheme.

This failure injection is the practical content of universality in this even-dimensional interval example: the additive constant fails, while the logarithmic slope survives. The validity map lists the other choices—state, geometry, algebra, and singularities—that must remain fixed before the same conclusion can be drawn.

Common pitfalls

Treating every finite term as universal—or as arbitrary. In even-dimensional smooth CFT examples, an additive constant can move while the logarithmic coefficient survives. In odd-dimensional smooth CFT examples, the constant is instead the universal term within the stated symmetry-preserving scheme. State the dimension, surface regularity, and allowed counterterm class, then demonstrate whether an admissible local term can shift the candidate quantity.

Comparing bare constants across regulators. Matching $\epsilon$ to $a$ fixes a physical short-distance convention only up to a finite factor. Slopes, anomaly coefficients, properly subtracted odd-dimensional universal constants, or other explicitly subtracted quantities are the appropriate cross-regulator targets.

Using a density matrix intrinsically in the continuum. The lattice or split construction has a type-I algebra and a reduced density matrix. The sharp local continuum algebra need not; formulate the intrinsic statement in terms of the surviving coefficient or algebraic quantities.

Exercises

1. For a $d=2$ interval, show explicitly how a constant cutoff rescaling changes the additive constant but not the logarithmic coefficient.

Solution

Implement the new prescription by replacing $\epsilon$ with $e^\eta\epsilon$ while retaining the same reference cutoff label. Then

$$S'=\frac{c}{3}\log\frac{\ell}{e^\eta\epsilon}+b =\frac{c}{3}\log\frac{\ell}{\epsilon} +\left(b-\frac{c\eta}{3}\right).$$

The fitted constant has shifted by $-c\eta/3$. Differentiating with respect to $\log\ell$ gives $c/3$ in either prescription, so the slope is invariant. If instead one chooses the rescaled length itself as the new variable, the same relation is written with the opposite relabeling of the constant; no observable conclusion changes.

2. Use the table above to estimate the convergence order of $q_N-1/3$ when $N$ doubles.

Solution

The errors are approximately $1.77\times10^{-5}$, $4.41\times10^{-6}$, and $1.10\times10^{-6}$ for $N=32,64,128$. Each doubling reduces the error by about a factor of four. Thus the leading correction is consistent with $q_N-1/3=O(N^{-2})$. The refinement estimate $u_N$ is larger than the actual error in these rows, so it is conservative for this sequence.

3. In the anomaly normalization of the comparison table, a two-dimensional CFT has $A=c/12$. Check that the even-dimensional sphere formula reduces to the interval coefficient.

Solution

For $d=2$, the sign factor is $(-1)^{d/2-1}=1$. Therefore

$$4A=4\frac{c}{12}=\frac{c}{3},$$

which is precisely the coefficient of $\log(\ell/\epsilon)$ for an interval. In one spatial dimension the entangling “surface” consists of the two endpoints, so the round-sphere language reduces to the same two-point twist geometry.

References

• Calabrese, Pasquale, and John Cardy. “Entanglement Entropy and Quantum Field Theory.” Journal of Statistical Mechanics: Theory and Experiment 2004, no. 06 (2004): P06002. DOI. Open PDF.
• Casini, Horacio, Marina Huerta, and Robert C. Myers. “Towards a Derivation of Holographic Entanglement Entropy.” Journal of High Energy Physics 2011, no. 05 (2011): 036. DOI. Open PDF.
• Peschel, Ingo. “Calculation of Reduced Density Matrices from Correlation Functions.” Journal of Physics A: Mathematical and General 36, no. 14 (2003): L205–L208. DOI. Open PDF.
• Solodukhin, Sergey N. “Entanglement Entropy, Conformal Invariance and Extrinsic Geometry.” Physics Letters B 665, no. 5 (2008): 305–309. DOI. Open PDF.
• Witten, Edward. “Notes on Some Entanglement Properties of Quantum Field Theory.” Reviews of Modern Physics 90, no. 4 (2018): 045003. DOI. Open PDF.