Entropy Bounds, Species, and Regulator Dependence

An entropy–energy bound is not a single formula until it declares the entropy, region, operator algebra, ultraviolet regulator, number of field species, reference state, and role of gravity. A raw subregion entropy in continuum QFT is divergent and normally grows with the number of fields. A relative entropy can remain finite because it compares two states on one algebra, but that is a different object with a reference-dependent modular energy.

Required background. Relative-entropy Bekenstein bounds gives the regulator-safe regional inequality whose species behavior is tested below.

Helpful background. Entropy counterterms and renormalization explains which local surface terms a common subtraction can cancel.

For the chapter-wide orientation, use the entry map, the comparison of claim domains, and the controls that change conclusions.

Why absolute subregion entropy counts species

Regulate a smooth spatial region $A$ with a short-distance scale $\epsilon$. For $d>2$, a typical leading term for $N$ decoupled fields is

$$S_A^{(\epsilon)} =N\,\kappa_{d-2} \frac{\operatorname{Area}(\partial A)}{\epsilon^{d-2}} +\text{subleading powers and logarithms}+S_{\mathrm{finite}}.$$

The coefficient $\kappa_{d-2}$ depends on the field content and regulator, so displaying a common factor $N$ assumes identical decoupled species regulated in the same way. In two dimensions the leading interval term is logarithmic rather than an area power. Srednicki’s coupled-oscillator calculation provides an early explicit area-law example for a free scalar Srednicki 1993, pp. 667–669.

The divergence has a simple origin: modes on opposite sides of the cut are entangled at every wavelength down to $\epsilon$, and each independent field contributes its own modes. Therefore neither increasing $N$ at fixed $\epsilon$ nor decreasing $\epsilon$ at fixed $N$ can be treated as an innocuous change when the left-hand side is a raw entropy.

A continuum local algebra supplies no preferred tensor-factor density matrix whose von Neumann entropy is finite. Gauge theories sharpen this warning: choosing an electric center, magnetic center, or extended Hilbert space changes edge contributions. Those choices can all be useful, but a comparison is meaningful only after fixing one algebra and using it for every state.

The finite regional comparison

Let $\rho_A$ and $\sigma_A$ be states on the same algebra, with $\sigma_A$ the reference. Algebraic relative entropy is

$$D(\rho_A\Vert\sigma_A) =\Delta\langle K_\sigma\rangle-\Delta S_A\geq0.$$

When a regulator description is used, $\rho$ and $\sigma$ must have matching local ultraviolet structure, the same cutoff surface, and the same counterterm prescription. Then the state-independent area terms cancel in $\Delta S_A$. More generally, $D$ is the primary algebraic object even when the two terms on the right are not separately defined.

For independent species in product states,

$$D\!\left(\bigotimes_{i=1}^{N}\rho_i \middle\Vert \bigotimes_{i=1}^{N}\sigma_i\right) =\sum_{i=1}^{N}D(\rho_i\Vert\sigma_i).$$

The modular Hamiltonian and entropy differences are likewise additive when the reference and algebra factorize. This does not make the result independent of $N$; it makes the dependence controlled and prevents an uncancelled vacuum entropy from being mistaken for localized information. Casini 2008, §4 and Appendix gives the relative-entropy resolution of the naive species problem.

Benchmark: many CFT species at fixed total energy

Consider $N$ decoupled two-dimensional CFTs, each with central charge $c=1$, on the infinite line. Use the interval $A=(-R,R)$ and the product vacuum as reference. Put every species in a thermal state of inverse temperature $\beta_N$. The total thermal energy density is

$$\Delta\langle T_{00}\rangle_N =\frac{N\pi}{6\beta_N^2}.$$

To compare with one species at inverse temperature $\beta_1$ while holding the total energy density fixed, choose

$$\beta_N=\sqrt N\,\beta_1.$$

Define $z_1=2\pi R/\beta_1$. Each species then has $z_N=z_1/\sqrt N$. Because the vacuum modular Hamiltonian is additive, the total modular-energy difference is

$$\Delta\langle K_A\rangle_N =N\frac{z_N^2}{18} =\frac{z_1^2}{18},$$

independent of $N$. The thermal interval entropy of Calabrese and Cardy 2004, §3 gives

$$\Delta S_{A,N} =\frac N3\log\!\left[ \frac{\sinh(z_1/\sqrt N)}{z_1/\sqrt N} \right],$$

and hence

$$D_N =\frac{z_1^2}{18} -\frac N3\log\!\left[ \frac{\sinh(z_1/\sqrt N)}{z_1/\sqrt N} \right].$$

At large $N$,

$$\Delta S_{A,N} =\frac{z_1^2}{18}-\frac{z_1^4}{540N}+O(N^{-2}), \qquad D_N=\frac{z_1^4}{540N}+O(N^{-2}).$$

Thus the finite entropy difference approaches the fixed modular-energy bound from below. It does not grow like the raw vacuum entropy.

For the reproducible choice $z_1=1$, the values are

Species $N$	$\Delta K_A$	$\Delta S_A$	$D_N$
1	0.0555556	0.0538131	0.00174244
4	0.0555556	0.0550998	0.000455749
16	0.0555556	0.0554403	0.000115284

By contrast, at cutoff $\epsilon$ the product-vacuum entropy is

$$S_{A,\mathrm{vac}}^{(\epsilon)} =\frac N3\log\!\left(\frac{2R}{\epsilon}\right)+N s_0,$$

where $s_0$ is scheme dependent. This grows linearly with $N$ even though the benchmark’s excitation energy and modular energy are fixed. The two behaviors answer different questions: the raw expression counts vacuum entanglement, while $\Delta S_A$ and $D_N$ compare thermal and vacuum states on the same interval.

Regulator, algebra, and order-of-limits controls

Match the regulator before subtracting. Compute both states on the same lattice or with the same covariant cutoff and surface counterterms. Subtracting two independently fitted area laws does not define relative entropy.

State the order of limits. The sequences $N\to\infty$ at fixed $\epsilon$ and $\epsilon\to0$ at fixed $N$ probe different families of theories. A simultaneous limit needs an explicit scaling relation $N(\epsilon)$ and cannot inherit either fixed-parameter conclusion automatically.

Fix the subsystem algebra. Changing a gauge-theory center or adding edge modes moves entropy between bulk and boundary terms. Relative entropy is robust only after the algebra itself is fixed; changing the algebra invokes data processing rather than equality.

Separate finite quantities. Mutual information between separated regions, vacuum-subtracted entropy for a controlled state family, and relative entropy can each be finite. They are not interchangeable: mutual information depends on two regions, while relative entropy depends on a state pair and one algebra.

Hold the right energy fixed. Ordinary energy, energy density, and modular energy need not track each other. The benchmark fixes total energy density and obtains fixed modular energy only because the state is translation invariant and the vacuum ball weight is known.

Where gravity enters

In semiclassical gravity one often combines a renormalized exterior entropy with an area term,

$$S_{\mathrm{gen}} =\frac{\operatorname{Area}}{4G_{\mathrm{ren}}} +S_{\mathrm{out}}^{\mathrm{ren}} +\text{higher-curvature terms}.$$

Matter-loop divergences can be related to renormalization of gravitational couplings; Susskind and Uglum 1994, §§ 2–4 develops this relation for black-hole entropy. But this is additional gravitational input: it assumes dynamical geometry, a gravitational effective action, a compatible entropy prescription, and the relevant counterterms. Nonminimal couplings, gauge fields, contact terms, and higher-curvature operators require further care. One may not invoke “renormalization of $G$” to turn an unspecified fixed-background entropy into a universal theorem.

The nongravitational conclusion is narrower and cleaner: for an admissible state pair with finite relative entropy, positivity gives a regulator-independent regional state-comparison bound once the algebra and reference are fixed. It neither defines an absolute matter entropy nor proves a black-hole entropy formula.

Decisive counterchecks

Vary $N$ while keeping the cutoff and ordinary energy fixed; raw entropy and finite state differences respond differently. Halve $\epsilon$ at fixed $N$; a raw entropy changes while a properly converged relative entropy does not. Change the reference temperature or the region; the modular Hamiltonian changes. Change the gauge center; the entropy convention changes. Add an area term; the claim has crossed into a gravitational theory. Any proposed bound that survives only by leaving one of these operations implicit has not specified its domain.

Exercises

1. Show how the raw two-dimensional vacuum entropy changes when the cutoff is halved.

Solution

Replacing $\epsilon$ by $\epsilon/2$ gives

$$S^{(\epsilon/2)}_{A,\mathrm{vac}}-S^{(\epsilon)}_{A,\mathrm{vac}} =\frac N3\log2.$$

The shift grows linearly with species number and contains no excitation energy.

2. Derive the fixed-energy temperature scaling in the benchmark.

Solution

Equate $N\pi/(6\beta_N^2)$ to the one-species value $\pi/(6\beta_1^2)$. This gives $\beta_N^2=N\beta_1^2$, hence $\beta_N=\sqrt N\beta_1$ for positive inverse temperature.

3. Obtain the leading large-$N$ relative entropy.

Solution

Use

$$\log\!\left(\frac{\sinh x}{x}\right) =\frac{x^2}{6}-\frac{x^4}{180}+O(x^6)$$

with $x=z_1/\sqrt N$. Multiplication by $N/3$ gives $\Delta S=z_1^2/18-z_1^4/(540N)+O(N^{-2})$. Subtracting from $\Delta K=z_1^2/18$ yields $D_N=z_1^4/(540N)+O(N^{-2})$.

4. Prove additivity of relative entropy for product species.

Solution

Use $\log(\rho_1\otimes\rho_2)=\log\rho_1\otimes\mathbf1+\mathbf1\otimes\log\rho_2$ on the product support. Taking the trace against $\rho_1\otimes\rho_2$ separates the two terms, giving $D(\rho_1\otimes\rho_2\Vert\sigma_1\otimes\sigma_2)=D(\rho_1\Vert\sigma_1)+D(\rho_2\Vert\sigma_2)$. Iterate over $N$ species.

References

• Calabrese, Pasquale, and John Cardy. “Entanglement Entropy and Quantum Field Theory.” Journal of Statistical Mechanics: Theory and Experiment 2004, no. 6 (2004): P06002. DOI. Open paper.
• Casini, Horacio. “Relative Entropy and the Bekenstein Bound.” Classical and Quantum Gravity 25, no. 20 (2008): 205021. DOI. Open paper.
• Srednicki, Mark. “Entropy and Area.” Physical Review Letters 71, no. 5 (1993): 666–669. DOI. Open paper.
• Susskind, Leonard, and John Uglum. “Black Hole Entropy in Canonical Quantum Gravity and Superstring Theory.” Physical Review D 50, no. 4 (1994): 2700–2711. DOI. Open paper.