Generalized Laws for Cosmological Horizons

A generalized second law for a cosmological horizon is meaningful only after naming the observer, the causal horizon, its future generator, the side whose field algebra is retained, and the state and subtraction used for entropy. Pure de Sitter supplies a stationary equilibrium test. Extending that result to an FLRW apparent horizon or a rapidly evolving observer patch requires additional hypotheses and is not a consequence of the area formula alone.

Helpful background. Quantum Fields in Curved Spacetime supplies horizon algebras and states; Einstein Equations supplies flux–geometry response; Black-Hole Geometry and Mechanics supplies causal and trapping horizons; Semiclassical Gravity supplies renormalized generalized entropy; and Causality and Global Structure supplies observer-dependent domains.

Observer, horizon, and algebra

Choose a future-inextendible timelike observer curve $\gamma$. Its future causal horizon is $\partial I^-(\gamma)$ when that boundary is nonempty; one can also form a two-sided patch using the observer’s past endpoint in an appropriate extension. The horizon uses the observer’s whole future and is therefore global. A marginal surface satisfying a local expansion condition on an FLRW slice is a different object, even when symmetry gives a preferred cosmological foliation.

For a spatially flat expanding FLRW model, the apparent-horizon areal radius is $R_A=H^{-1}$. A future event horizon instead has

$$R_E(t)=a(t)\int_t^\infty\frac{dt'}{a(t')},$$

provided the integral converges and that expanding future is actually part of the spacetime. For $a(t)\propto t^p$ with $p\gt1$, $R_E=t/(p-1)$ whereas $R_A=t/p$. For $0\lt p\le1$ the same power-law spacetime has no future event horizon at all, despite finite $H^{-1}$ at finite time. Only special geometries, such as exact de Sitter with $a\propto e^{Ht}$, make these two radii coincide.

Let $\mathcal A_{\rm patch}$ denote the algebra of observables accessible inside the patch on a cut $\sigma$. With this declared side convention, write

$$S_{\rm gen}[\sigma] =S_{\rm grav}^{\rm ren}[\sigma] +S_{\rm patch}^{\rm ren}[\sigma].$$

For Einstein gravity $S_{\rm grav}=A/(4G_{\rm ren})$; higher-curvature terms contribute the corresponding renormalized entropy functional. Entanglement divergences in $S_{\rm patch}$ are canceled by the same coupling renormalization. Reversing the chosen side changes the matter entropy and flux orientation, so it must not be done midway through a monotonicity argument.

Flux balance near equilibrium

In a nearly static de Sitter patch, a small positive energy $\delta E$ placed inside shrinks the cosmological horizon. With the central generator normalized to unit norm at $r=0$,

$$\delta E=-T_c\,\delta S_{\rm grav}, \qquad T_c=\frac{\kappa_c}{2\pi}.$$

If energy leaves the accessible patch, $\delta E_{\rm patch}\lt0$ and the horizon entropy increases. The matter entropy inside may decrease at the same time. For a sufficiently isolated packet of Killing energy $E\gt0$ crossing outward, let $S_{\rm packet}$ denote the decrease of the patch’s renormalized matter entropy in that process. The quasistatic balance is

$$\Delta S_{\rm gen} \simeq\frac{E}{T_c}-S_{\rm packet}.$$

For a thermal reference state $\sigma=e^{-\beta H_{\rm patch}}/Z$, relative entropy gives

$$S(\rho\Vert\sigma)=\beta\Delta E-\Delta S\ge0, \qquad \beta=T_c^{-1}.$$

The energy and entropy differences must refer to the same algebra and reference state. This explains the sign in the small-perturbation balance; it is not a pointwise bound on arbitrary matter. If the packet is correlated with what remains, its standalone entropy is not the entropy lost by the patch: use the actual entropy change, including correlations. The area response and matter comparison must also be evaluated at the same backreaction order.

The Noether-charge first law determines the stationary gravitational term under specified boundary conditions Iyer and Wald 1994, §§ 3–6, PDF. It does not by itself prove monotonic generalized entropy. Conversely, a field-theory entropy inequality on a fixed background does not by itself include the area response. A genuine semiclassical law matches both sides at the same order.

Results and open questions through 2026-08-19

Evidence through 2026-08-19 supports stationary de Sitter thermality and generalized laws for particular causal horizons under explicit algebraic assumptions. A universal law for every cosmological apparent, trapping, or event horizon remains unresolved and conjectural; the available results do not give a universal verdict.

Evidence and exact locator	Result and assumptions	Limitation or missing step
Gibbons and Hawking 1977, pp. 2739–2743	Euclidean de Sitter has $T=H/(2\pi)$ and leading entropy $A/(4G)$ for the regular stationary construction	Establishes equilibrium thermodynamics, not a nonequilibrium monotonicity theorem
Iyer and Wald 1994, §§ 3–6, PDF	Diffeomorphism-invariant theories yield a Noether-charge first law for stationary horizons and controlled perturbations	Does not supply fine-grained matter entropy or an arbitrary dynamical-cosmology GSL
Wall 2012, §§ 2.1–2.8 and 3–5, PDF	Relative-entropy and horizon-algebra hypotheses prove a GSL for causal horizons in the stated semiclassical setting	Must match the cosmological horizon, algebra, thermality, and renormalization assumptions; not a trapping-horizon theorem
Bousso et al. 2016, §§ 3–5, PDF	The QFC organizes null variations of generalized entropy and conditionally implies entropy bounds	The general QFC is a conjecture, and applying it to cosmology requires a specified cut, state, and deformation

The conservative position applies a theorem only to a true causal horizon with the theorem’s algebra and state hypotheses. The competing geometric position assigns entropy to a preferred FLRW apparent horizon and seeks a local balance law from its expansions. The second formulation can be useful in symmetric models, but it is not equivalent: changing foliation moves the apparent horizon, and a local temperature may be ambiguous away from adiabatic evolution.

Adversarial tests

First vary the slicing while holding the observer’s causal patch fixed. A causal-horizon result should not change, whereas an apparent-horizon construction may. Second take the exact de Sitter KMS limit: generalized entropy should be stationary in equilibrium after consistent renormalization. Third send $G\to0$ while keeping the field state smooth. The area response and matter entropy must be ordered consistently; retaining a finite geometric production term from a higher semiclassical order signals mismatched bookkeeping.

An adverse state is equally important. A non-Hadamard excitation can make the stress or entropy subtraction ill-defined at the horizon, so it is not a counterexample to a theorem that assumes Hadamard regularity. A valid counterexample must satisfy the same field class, state, horizon, and algebra assumptions. Multiple temperatures in Schwarzschild–de Sitter provide another boundary: unless the surface gravities match or a nonequilibrium state is specified, there is no single KMS ensemble for both horizons.

Nonequilibrium limits and open questions

For a slowly evolving patch, record the event or apparent horizon definition, generator normalization, adiabatic parameter $|\dot\kappa|/\kappa^2$, renormalized entropy functional, accessible algebra, flux direction, and stress fluctuations. A claimed monotonic law should be tested against constraint evolution and a finite counterterm shift. When the horizon moves on the same timescale as equilibration, “temperature times entropy change” is only an organizing approximation.

The conclusions could change after a theorem extends or limits causal-horizon entropy monotonicity, a state-selected cosmological backreaction calculation changes the area–entropy balance, an explicit slicing counterexample defeats an apparent-horizon proposal, or a corrected QFC result changes the allowed deformation class; a reproducible nonequilibrium multi-horizon calculation or new entropy-renormalization result could also change the conclusion.

Exercise

A packet with Killing energy satisfying $E/T_c=0.03$ and entropy $S_{\rm packet}=0.02$ exits a quasistatic de Sitter patch. Compute the leading change in generalized entropy. Which assumption prevents this arithmetic from becoming a universal theorem?

Solution

The horizon entropy rises by $E/T_c=0.03$, while accessible matter entropy falls by $0.02$. Therefore

$$\Delta S_{\rm gen}=0.03-0.02=0.01\gt 0.$$

The calculation assumes a quasistatic process, the central Killing normalization, a regular state, a definite accessible algebra, and the leading first-law response. Arbitrary dynamical apparent horizons or non-KMS states need not satisfy the same balance without a separate theorem.

References

• Bousso, R., Fisher, Z., Leichenauer, S., and Wall, A. C. (2016). “A Quantum Focussing Conjecture.” Physical Review D 93, 064044. DOI. Open PDF.
• Gibbons, G. W., and Hawking, S. W. (1977). “Cosmological Event Horizons, Thermodynamics, and Particle Creation.” Physical Review D 15, 2738–2751. DOI.
• Iyer, V., and Wald, R. M. (1994). “Some Properties of Noether Charge and a Proposal for Dynamical Black Hole Entropy.” Physical Review D 50, 846–864. DOI. 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. DOI. Open PDF.