Proof Strategies for the Generalized Second Law

The sharpest semiclassical route to the generalized second law is monotonicity of relative entropy under restriction to a smaller algebra. It converts a quantum-information theorem into increase of horizon area plus exterior entropy. The conversion is conditional: it requires a causal horizon, a suitable null-surface algebra and reference state, minimally coupled matter, controlled semiclassical backreaction, and compatible renormalization. This page states that result exactly and distinguishes it from less general quasistationary and hydrodynamic arguments.

Required background. Quantum Information supplies relative entropy, modular Hamiltonians, restriction to subalgebras, and data processing.

Helpful background. Quantum Fields in Curved Spacetime supplies KMS states and horizon quantization; Einstein Equations supplies the semiclassical constraint; Black-Hole Geometry and Mechanics supplies causal horizons and their generators; and Semiclassical Gravity supplies the backreaction expansion and renormalized generalized entropy.

Conditional generalized-second-law statement

Result. Let $H$ be a future causal horizon whose unperturbed geometry is stationary, and let $\Lambda_1(y)$ and $\Lambda_2(y)$ be complete cuts with $\Lambda_2(y)\geq\Lambda_1(y)$ on every generator. Assume:

1. gravity is Einstein gravity treated semiclassically, with minimally coupled quantum matter and metric response determined to first nontrivial order by $G_{\mu\nu}=8\pi G\langle T_{\mu\nu}\rangle$;
2. the matter theory has a nontrivial horizon algebra satisfying determinism, ultralocality, local Lorentz invariance, and stability;
3. the null-translation vacuum restricted above every cut is a faithful KMS reference state for the dilation generator about that cut;
4. information outside the horizon evolves into the later horizon algebra together with the complete escaping system, with no additional exposed sink;
5. the stress tensor, gravitational entropy, and exterior entropy are renormalized in one scheme for which the relative-entropy/free-energy identity continues to hold.

Assume also that the perturbation settles sufficiently rapidly in the future for the weighted flux integrals below to converge and for the late expansion to vanish. The derivation here treats quantum matter with a classical mean metric; quantized gravitational radiation requires including its shear-squared flux at the same order.

Choose an affine parameter $\lambda$ increasing toward the future on each generator. Use the same normalization in the null-energy flux and its affine weight; their rescalings compensate in the dilation energy below. For noncompact cuts, use a common transverse regulator and require finite differences when it is removed. With these hypotheses and terms retained through the order at which $\langle T_{\mu\nu}\rangle$ changes the area, the theorem is

$$S_{\mathrm{gen}}[\Lambda_2]\geq S_{\mathrm{gen}}[\Lambda_1], \qquad S_{\mathrm{gen}}[\Lambda] =\frac{A[\Lambda]}{4G_{\mathrm{ren}}} +S_{\mathrm{out}}^{\mathrm{ren}}[\Lambda].$$

Wall states the assumptions, establishes horizon thermality and the relative-entropy reduction, and extends the result to the full outside system in Wall 2012, §§2.1–2.8, printed pp. 6–21, PDF. The free-field, spin, and interaction analyses that support—and delimit—the horizon-algebra assumption are in Wall 2012, §§3–5, printed pp. 22–42, PDF.

Relative entropy supplies the monotonicity

For a finite-dimensional system—or a type-I regulated approximation—with density matrices $\rho$ and $\sigma$, relative entropy is

$$D_{\mathcal A}(\rho\Vert\sigma) =\operatorname{tr}(\rho\log\rho) -\operatorname{tr}(\rho\log\sigma).$$

A local horizon algebra in continuum QFT is generally type III, so it has no density matrix or trace with which to use this formula literally. There the proof uses Araki relative entropy, defined by the relative modular operator. It agrees with the displayed trace expression in the type-I case, remains meaningful for type-III algebras, and has the same monotonicity under restriction. Wall makes the finite-dimensional/type-III distinction in Wall 2012, §§2.4–2.5, printed pp. 16–18, PDF.

If $\mathcal A_2\subseteq\mathcal A_1$, restriction loses distinguishing power, so data processing gives

$$D_{\mathcal A_1}(\rho\Vert\sigma) \geq D_{\mathcal A_2}(\rho\Vert\sigma).$$

Moving a horizon cut to the future removes an earlier segment from the accessible algebra. Thus the later algebra is the smaller one. Assumption 3 states that the null-translation vacuum $\sigma$ restricts to a KMS state for the cut-dependent dilation generator $K_\Lambda$. In Wall’s derivation this thermality follows from the horizon-algebra structure: local Lorentz symmetry supplies dilations, stability supplies the positive null-translation generator needed for the spectral condition, and ultralocality permits independent cuts on different generators. Local Lorentz symmetry alone would not imply KMS. The argument is given in Wall 2012, §2.4, printed pp. 14–16, PDF. The relative entropy is therefore a modular free-energy difference,

$$D_\Lambda(\rho\Vert\sigma) =2\pi\bigl(\langle K_\Lambda\rangle_\rho -\langle K_\Lambda\rangle_\sigma\bigr) -\bigl(S_\rho(\Lambda)-S_\sigma(\Lambda)\bigr).$$

This identity packages otherwise divergent energy and entropy terms into a finite algebraic quantity. It is precisely the monotonicity theorem—not a pointwise energy condition—that supplies the irreversible sign.

From modular energy to horizon area

The normalization of the modular energy is fixed by the affine parameter used above. With the transverse horizon area element denoted by $d\mu_\perp(y)$, define

$$K_\Lambda =\int d\mu_\perp(y) \int_{\Lambda(y)}^\infty d\lambda\, \bigl(\lambda-\Lambda(y)\bigr)T_{kk}(\lambda,y).$$

Here $k^\mu=dx^\mu/d\lambda$, and $T_{kk}=T_{\mu\nu}k^\mu k^\nu$. This is the formal boost-energy functional about the cut, with no factor of $2\pi$ absorbed into its definition. In the continuum, the sharply restricted integral need not define a standalone operator on the horizon Hilbert space. The proof uses its controlled expectation differences, obtained from the renormalized stress tensor; see Wall 2012, §3.3, Eq. (88), PDF. Under $\lambda'=a\lambda$, $T_{k'k'}=a^{-2}T_{kk}$, while $d\lambda'$ and $\lambda'-\Lambda'$ each supply a factor $a$. Thus $K_\Lambda$ is invariant, as a dimensionless dilation generator should be.

The remaining proof step is gravitational. At the retained order the quadratic expansion and classical shear terms drop out of Raychaudhuri:

$$\frac{d\theta}{d\lambda}=-8\pi G\langle T_{kk}\rangle, \qquad \theta(\infty)=0.$$

Integrating twice, using the unperturbed transverse area element, gives

$$\begin{aligned} \theta(\lambda,y) &=8\pi G\int_\lambda^\infty \langle T_{kk}(\lambda',y)\rangle\,d\lambda',\\ A[\infty]-A[\Lambda] &=8\pi G\langle K_\Lambda\rangle. \end{aligned}$$

The late area is an integration constant, not generally zero. For a state-minus-reference difference the precise identity is

$$2\pi\,\Delta\langle K_\Lambda\rangle =\frac{\Delta A[\infty]-\Delta A[\Lambda]}{4G}.$$

This reproduces Wall 2012, §2.2, Eqs. (19)–(24), PDF. The term $\Delta A[\infty]$ is independent of the moving cut and hence cancels in the GSL comparison. Together with the cut-independent reference entropy, it is absorbed into $C$ when the modular identity becomes

$$D_\Lambda(\rho\Vert\sigma) =C-S_{\mathrm{gen}}[\Lambda]$$

for a cut-independent constant $C$ for the chosen pair of states. Relative entropy decreases as the cut advances, so generalized entropy increases. The sign can be checked without assuming the conclusion: positive future-directed null energy makes the event-horizon generators begin expanding in anticipation, leaving a larger late cut, while the same positive energy raises the modular Hamiltonian. The minus sign above is what makes both descriptions consistent.

To include radiation that escapes rather than crosses $H$, enlarge the later algebra by the complete future exterior system and choose a product reference state: the horizon vacuum times any faithful state on the escaping factor. That factor’s modular energy is independent of the horizon cut. Unitary propagation identifies its joint entropy with the entropy outside the corresponding spacelike slice. This is the outside-system step in Wall 2012, §2.7, printed pp. 19–20, PDF, and it is essential for an evaporating geometry.

What the alternative strategies establish

Strategy	Key input	Strength	Boundary of the argument
Quasistationary free energy	A thermal reference preserved by slow evolution and the horizon first law	Transparent near-equilibrium route	The first-law replacement of area by energy fails for rapid, finite departures
Horizon algebra plus relative entropy	Nested algebras, KMS modular energy, data processing, and the semiclassical constraint	Arbitrary ordered cuts and rapidly varying fields within the stated theory class	Requires the four null-algebra axioms and a compatible renormalization scheme
Hydrodynamic entropy current	Local equilibrium, constitutive relations, and nonnegative entropy production	Useful coarse-grained long-wavelength interpretation	Does not control fine-grained exterior entropy or generic nonhydrodynamic states
Classical focusing	A suitable classical energy condition and Raychaudhuri evolution	Establishes the area theorem in its classical domain	Quantum fields can violate pointwise null energy conditions; area alone need not increase

Wald reviews the quasistationary GSL, entropy bounds, and the limits of traditional thermodynamic reasoning Wald 2001, §§4.1–4.2, PDF. The relative-entropy proof is stronger in time dependence, but narrower in its explicit algebraic and coupling hypotheses. These are complementary architectures, not interchangeable slogans.

Counterexample to overextension

Take a future outer trapping horizon in a strongly dynamical spacetime. Its location is quasilocal and foliation dependent; it need not be the boundary of the past of a complete future worldline. Advancing between its preferred cuts therefore need not produce nested causal-horizon algebras, and there need be no null-translation vacuum that is KMS for every cut. Even if a separately defined generalized entropy happens to increase, Wall’s proof cannot be invoked.

Likewise, nonminimal curvature coupling or a higher-curvature gravitational action changes both the null stress-energy relation and the geometric entropy. Wall discusses these extensions and unresolved points in Wall 2012, §5.3, printed pp. 39–42, PDF. Replacing $A/(4G)$ by a guessed entropy functional is not sufficient. One must rederive the constraint-to-entropy step and establish the relevant horizon algebra. These limitations are counterexamples to an unconditional proof claim, not counterexamples to every possible generalized second law.

Exercise

Let $\Lambda_2$ be later than $\Lambda_1$, and suppose for a controlled Rindler perturbation that $D_{\Lambda}=C-S_{\mathrm{gen}}[\Lambda]$. If data processing gives $D_{\Lambda_1}-D_{\Lambda_2}=0.18$, compute the generalized-entropy change. Then explain why the same arithmetic does not prove a GSL for an arbitrary trapping horizon.

Solution

Subtracting the two identities gives

$$D_{\Lambda_1}-D_{\Lambda_2} =S_{\mathrm{gen}}[\Lambda_2]-S_{\mathrm{gen}}[\Lambda_1].$$

Hence $\Delta S_{\mathrm{gen}}=0.18\gt0$. The numerical sign follows from data processing only because advancing the causal-horizon cut restricts to a nested subalgebra and because the modular-energy/area identity has already been established. An arbitrary trapping horizon supplies neither fact: its cuts may not define nested causal complements, and its generator need not admit the required KMS reference state. The calculation therefore diagnoses the proof architecture rather than extending its domain.

References

• Wald, R. M. (2001), “The Thermodynamics of Black Holes,” Living Reviews in Relativity 4, 6. Open PDF.
• Wall, A. C. (2012), “A Proof of the Generalized Second Law for Rapidly Changing Fields and Arbitrary Horizon Slices,” Physical Review D 85, 104049; revised 2018. Open PDF.

Dated Research assessment. For open questions and recent results, follow the link below. Each assessment states the date through which it considers results and the limits of its conclusions.

• Scope of the Generalized Second Law