Semiclassical Limits of Islands and Page Curves

A generalized-entropy calculation becomes an information-recovery claim only after it specifies a radiation region and algebra, includes the relevant gravitational replica saddles, extremizes and selects among island candidates, and interprets the selected value as the fine-grained entropy of Hawking radiation. Semiclassical extremality is necessary input to that construction, but it does not by itself establish a Page curve, unitarity, or reconstruction.

Required background. Quantum Information supplies fine-grained entropy, purification, subsystem algebras, and the operational meaning of a Page curve.

Helpful background. Quantum Fields in Curved Spacetime supplies Hawking-state entanglement on a fixed background; Einstein Equations supplies the mean geometry; Black-Hole Geometry and Mechanics supplies horizon cuts; and Semiclassical Gravity supplies generalized entropy and backreaction within their controlled limit.

Units. This page restores $\hbar$ in generalized-entropy formulas while keeping $c=k_{\rm B}=1$; setting $\hbar=1$ recovers the site-wide convention.

Three increasingly strong claims

The following statements use similar formulas but require different evidence.

Claim	Required data	What follows
Generalized entropy of a cut	Mean geometry, state, outside algebra, regulator, renormalized gravitational couplings	A finite semiclassical functional $S_{\rm gen}[X]$
Quantum extremal surface	The same data plus normal deformations and anchoring or boundary conditions	A stationary semiclassical saddle, $\delta S_{\rm gen}/\delta X^a=0$
Island formula for radiation	A specified radiation region $R$, replica path integral, allowed saddles, homology, extremization, and dominance rule	A candidate fine-grained radiation entropy and, across time, a candidate Page curve

The first two statements belong to semiclassical horizon physics. The third adds information-theoretic and gravitational path-integral structure. In schematic notation it is often written

$$S(R) =\min_I\operatorname*{ext}_I \left[ \frac{A(\partial I)}{4G\hbar} +S_{\rm bulk}(R\cup I) \right].$$

Here $I=\varnothing$ is also a candidate, and the displayed area term is the leading Einstein contribution; the complete renormalized gravitational entropy is needed when higher-curvature terms are retained. The “ext” operation supplies the quantum extremality equation; “min” compares admissible extrema for the same radiation region, state, and boundary problem. It is not an unconstrained minimization over every spacetime surface. Neither operation is encoded in the local semiclassical Einstein equation alone.

The perturbative QES construction is developed in Engelhardt and Wall 2015, §§ 2–6, PDF. The later replica-wormhole derivation of the island rule is a separate input: Almheiri et al. 2020, §§ 1.1 and 2.1–2.2, PDF analyze JT gravity coupled to a CFT and a nongravitating region. These specified models supply more than a geometric stationarity equation; they do not make an arbitrary four-dimensional QES an island.

Where the semiclassical limit stops

This volume can ask whether $S_{\rm gen}$ is renormalized, whether its shape derivative is finite, whether a candidate surface lies inside the curvature and fluctuation hierarchy, and whether competing saddles solve the same boundary problem. It can also diagnose failure: a Planckian cut, uncontrolled replica continuation, unknown contour, large metric fluctuations, or a saddle outside the state construction invalidates the semiclassical inference.

It cannot infer microscopic information recovery merely because a QES exists. A Page curve is a statement about the fine-grained entropy of a time-dependent radiation algebra. Unitarity is a statement about the full quantum evolution. Reconstruction is a statement about operator algebras or codes. These claims require factorization, ensemble, replica-continuation, dominance, and nonperturbative assumptions that are not fixed by the generalized-entropy stationarity equation.

Accordingly, microscopic information recovery belongs to Black-Hole Information, together with replica saddles, island formulas, fine-grained Page curves, and reconstruction. That neighboring volume compares the information-theory argument and its controlled models. The scientific boundary here is not a dismissal: semiclassical generalized entropy supplies essential geometric input, while the interpretation of that input as recovered microscopic information is a stronger claim.

Checks that prevent overclaiming

No-island branch. Setting $I=\varnothing$ reduces the candidate to the bulk entropy of $R$. This verifies that the formula contains ordinary Hawking-state entropy as one branch, but it does not prove that an island branch exists or dominates.

Extremum-versus-dominance check. A surface satisfying $\delta S_{\rm gen}=0$ may have a larger generalized entropy than another saddle or may carry an unstable mode. Stationarity alone cannot choose the Page-curve branch.

Regime check. The area term scales like $1/(G\hbar)$ while bulk entropy is order unity per field species in a usual expansion. A candidate displacement can be perturbative even when the entropy correction is physically important. Near a cutoff-scale surface or with parametrically large stress fluctuations, the mean-metric saddle may cease to be meaningful.

Independent interpretation check. Ask what the left-hand side denotes. If it is merely $S_{\rm gen}$ of a cut, the result is semiclassical thermodynamics. If it is asserted to be $S(R)$ for collected radiation, the calculation has crossed into fine-grained information and must provide the extra replica and algebraic evidence.

For a simple illustration of branch selection, suppose an already justified entropy prescription gives $S_{\varnothing}(t)=\gamma t$ and a second admissible extremal branch $S_I(t)=S_*$, with $\gamma\gt0$ and dimensionless $S_*\gt0$. The selected leading entropy is $\min(\gamma t,S_*)$, with a transition at $t_*=S_*/\gamma$. Computing $\delta S_I=0$ says nothing about $\gamma$, $S_*$, or whether the empty branch wins before $t_*$. This arithmetic illustrates the extra selection step; a plateau is not the full evaporation Page curve, and these toy functions are not a gravitational solution.

The fixed-background Hawking derivation remains in QFT in Curved Spacetime, and a microscopic endpoint remains a Quantum Gravity question. This page only marks the transition from a geometric entropy saddle to a claimed radiation-entropy computation.

Exercise

A calculation finds two surfaces with $S_{\rm gen}=100$ and $S_{\rm gen}=80$, but it specifies neither a radiation region nor a replica contour. What can be concluded, and what cannot?

Solution

One may compare two semiclassical generalized-entropy saddles if both solve the same stated geometric variational problem; the second has the lower value. One cannot call either an island, identify 80 as the fine-grained radiation entropy, infer a Page curve, or claim unitarity. Those conclusions require a radiation algebra, replica construction, allowed-saddle rule, and controlled dominance argument.

References

• Almheiri, A., Hartman, T., Maldacena, J., Shaghoulian, E., and Tajdini, A. (2020). “Replica Wormholes and the Entropy of Hawking Radiation.” JHEP 05, 013. DOI. Open PDF.
• Engelhardt, N., and Wall, A. C. (2015). “Quantum Extremal Surfaces: Holographic Entanglement Entropy beyond the Classical Regime.” JHEP 01, 073. Open PDF.