Island Rule: Semiclassical Entropy in Black Holes
- Island Rule is a semiclassical prescription that computes the fine-grained entanglement entropy of a nongravitating region by incorporating an additional gravitational island.
- It employs a generalized entropy functional that is extremized over candidate island boundaries to yield a transition from monotonically growing Hawking entropy to a Page-curve behavior.
- The method has been applied to black holes, cosmological settings, and algebraic models, providing practical insights into resolving the information paradox.
Searching arXiv for recent and foundational papers on the island rule and related Page-curve literature. The island rule is a semiclassical prescription for computing the fine-grained entanglement entropy of a nongravitating region when it is coupled to a gravitating spacetime. Instead of assigning the entropy solely to quantum fields on , the prescription allows an additional gravitating region , called an island, to contribute through a generalized entropy functional that is first extremized over candidate island boundaries and then minimized over saddles. In the modern formulation, this mechanism converts the ever-growing Hawking entropy into a Page-curve behavior and has been applied to evaporating and eternal black holes, cosmological settings, and operator-algebraic models of bulk reconstruction (Bousso et al., 2021, Piao, 2023, Lin et al., 2024, Deng, 9 Jan 2026).
1. Definition and generalized-entropy prescription
The island rule is stated in the data in several equivalent forms. A standard expression is
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$
with the island boundary and the matter entanglement entropy on (Kim et al., 2021). A closely related form writes
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$
for black-hole radiation coupled to a bath (Lin et al., 2024). In the cosmological application, the same logic appears as
$S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$
implemented in a JT-like model through dilaton and CFT terms (Piao, 2023).
The operational rule is consistent across the cited works: one extremizes the generalized entropy with respect to the island location, i.e. the quantum extremal surface (QES), and then selects the minimal saddle. In the noncommutative-black-hole study this is written explicitly as
$S_R = \min_X \left\{ \ext_X \left[ \frac{\mathcal A(X)}{4G_N} + S_c(\Sigma_R \cup \Sigma_I) \right] \right\},$
or equivalently
0
where 1 denotes the QES and the quantity in brackets is the generalized entropy 2 (Liu et al., 14 May 2025).
This prescription is the semiclassical entropy formula associated with replica-wormhole reasoning. One of the general analyses expresses the relation to the replica definition as
3
and interprets the island rule as the semiclassical realization of that fine-grained entropy computation (Yu et al., 2024). This suggests that the island rule is not a correction to coarse-grained thermodynamics but a proposal for the exact entropy in a semiclassical regime.
2. Page curves and the entropy of Hawking radiation
The central physical role of the island rule in the literature summarized here is to recover the Page curve. Without islands, the matter entropy of Hawking radiation grows monotonically or linearly at late times. With islands, a new saddle takes over after the Page time and prevents indefinite growth.
For a broad class of static black holes, the no-island late-time entropy is
4
whereas the island saddle yields
5
The resulting Page-curve form is summarized as
6
(Yu et al., 2024). The same qualitative structure appears in explicit four-dimensional RN-AdS calculations, where the no-island branch is
7
while the island branch saturates at
8
In asymptotically flat Reissner–Nordström black holes with a heat bath, the non-extremal no-island entropy behaves as
9
while the island saddle produces a constant late-time entropy determined by the outer horizon radius (Kim et al., 2021). In one-sided collapse geometries, the same competition appears between a linearly growing no-island entropy and an island saddle giving 0 up to subleading corrections (Gan et al., 2022). In massive gravity, the no-island entropy again grows linearly, and the island branch yields
1
for neutral and non-extremal charged two-sided geometries (Nam, 2021).
The Page time is identified by the crossover between the Hawking branch and the island branch. Different models supply different estimates. For RN-AdS2,
3
(Lin et al., 2024). For asymptotically flat non-extremal RN,
4
(Kim et al., 2021). In one-sided collapse in 5,
6
with the 7 expression obtained by setting 8 (Gan et al., 2022). The general entropy-bound analysis does not require explicit QES solution and instead characterizes the Page transition by the inequality
9
after the semiclassical Page time (Bousso et al., 2021).
A common misconception is that the island rule merely modifies the late-time asymptotics. The cited calculations instead present a saddle competition that governs the entire entropy history: before the Page time the no-island saddle dominates, and after the Page time the island saddle dominates. In that sense the Page curve is not inserted by hand; it is the result of minimizing generalized entropy (Kim et al., 2021, Lin et al., 2024).
3. Quantum extremal surfaces, Island Finder, and geometric criteria
The island rule is often operationalized through the search for QESs. In explicit black-hole calculations, one introduces candidate island endpoints and extremizes $S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$0 with respect to both temporal and radial data. For example, in the general static-black-hole analysis the late-time extremality conditions imply
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$1
and the radial condition becomes
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$2
(Yu et al., 2024). In RN-AdS$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$3, early-time extremization yields $S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$4 and $S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$5, while at late times the island sits just behind the horizon,
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$6
(Lin et al., 2024). In asymptotically flat RN, late-time extremization gives
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$7
and places the island near the outer horizon (Kim et al., 2021).
A more abstract attempt to avoid explicit QES solving is the Island Finder criterion. The sufficient conditions were formulated as follows: if there exists a region $S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$8 such that
$S(R)=\min\left\{\ext\, S_{\rm gen}\right\},\qquad S_{\rm gen}=\frac{\mathcal{A}(\partial I)}{4G_N}+S_{\rm matter}(R\cup I),$9
and 0 is either quantum normal or quantum anti-normal, then a non-empty island exists and
1
(Bousso et al., 2021). Quantum normality and anti-normality are expressed through the quantum expansion one-form 2. In the formulation quoted in the data,
3
for quantum normality, with reversed inequalities for quantum anti-normality (Bousso et al., 2021).
That sufficiency claim was later refined. “Island mirages” identifies a loophole in the proof and argues that the conditions are not actually sufficient in full generality (Rolph, 2022). The counterexample mechanism is that one may have a detector region 4 satisfying
5
while the true island of 6 remains empty because the relevant maximin slice does not intersect 7 or 8 (Rolph, 2022). This controversy is significant because it shows that entropy and expansion inequalities alone do not universally replace the full extremization problem.
A complementary line of work derives geometric conditions favorable to island existence. For a broad class of static black holes, a sufficient near-horizon criterion is
9
where 0 is the blackening factor in Schwarzschild gauge (Yu et al., 2024). The same paper relates this criterion to the quantum focusing conjecture (QFC), emphasizing that it is sufficient but not strictly necessary. Since in the reduced two-dimensional sector the Ricci scalar is
1
the condition 2 corresponds to positive curvature near the horizon (Yu et al., 2024). This suggests a geometric regime in which island saddles are naturally supported, although the data explicitly warns that sufficiency should not be confused with necessity.
4. Model-dependent realizations
The island rule has been implemented in a wide range of settings, each of which emphasizes a different structural aspect of generalized-entropy minimization.
Black holes in asymptotically flat and AdS spacetimes
In four-dimensional RN-AdS black holes, the generalized entropy for a single island with endpoints 3 is written as
4
This model is further used to study how first-order phase transitions deform the Page curve without spoiling unitarity (Lin et al., 2024).
For asymptotically flat eternal RN black holes in thermal equilibrium with a bath at arbitrary off-shell temperature, the method proceeds through an off-shell generalized entropy 5 and only then imposes the on-shell limit 6 (Kim et al., 2021). This off-shell construction is the key technical step that allows the same framework to treat both non-extremal and extremal geometries.
One-sided asymptotically flat black holes formed by collapse introduce a state dependence absent in eternal Hartle–Hawking setups. In the “in” vacuum, the generalized entropy is still minimized over island candidates, but the matter term changes. A notable result is that the QES location depends on the position of the cutoff surface: when the cutoff is far from the horizon, 7 lies inside and near the horizon; when the cutoff is near the horizon, 8 lies outside and near the horizon (Gan et al., 2022). This differs from the eternal black-hole case, where the data states that 9 is always outside the horizon.
Massive gravity provides another deformation of the standard story. In four-dimensional dRGT massive gravity, the metric function
0
modifies the horizon structure and therefore the Page time, scrambling time, and island location (Nam, 2021). The island rule itself, however, retains the same generalized-entropy form.
Noncommutative black holes
For noncommutative black holes, the paper summarized in the data stresses that noncommutativity alone does not guarantee finite radiation entropy. Without islands, the radiation entropy still grows and can diverge. With islands, the late-time behavior depends on the cutoff placement (Liu et al., 14 May 2025).
When the cutoff surface is far from the event horizon, the matter entropy is approximated by
1
or in conformal coordinates
2
(Liu et al., 14 May 2025). In this regime, the island improves the radiation entropy for both commutative and noncommutative cases, but only the noncommutative black hole achieves a proper Page curve at late times because the temperature tends to zero and evaporation ends in a remnant near 3 (Liu et al., 14 May 2025).
When the cutoff surface is close to the horizon, the matter entropy is approximated instead by
4
and both the commutative and noncommutative black holes reproduce the Page curve once islands are included (Liu et al., 14 May 2025). The Page time for the noncommutative case is greatly delayed, by about four orders of magnitude according to the paper.
Cosmological implementation
A cosmological analogue appears in a JT-like two-dimensional inflation geometry, where a de Sitter patch is glued to a Minkowski “hat” representing the post-inflationary observer’s causal region (Piao, 2023). The matter entropy without islands scales like the number of inflationary e-folds,
5
and this creates an inflationary analogue of the information paradox if one insists on the no-island entropy (Piao, 2023). Including an island inside a neighboring collapsing patch yields a saturated entropy
6
rather than indefinite growth (Piao, 2023). The crossover occurs at
7
This is a Page-curve-like transition in the entropy of primordial perturbations rather than Hawking quanta.
5. Causality, operator algebras, and effective descriptions
The island rule is often described through entanglement-wedge reconstruction, but the data also highlights questions about causal consistency. “Causality Criteria for Island Models” argues that apparent violations of micro-causality in some double-holographic effective theories do not arise from island physics itself, nor from nonlocal reconstruction as such (Deng, 9 Jan 2026). Instead, the problem is a mismatch between effective spacetime separation and bulk causal accessibility.
The paper formulates a structural LDM criterion. For an operator 8 on a codimension-one surface 9 and $S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$0 on the asymptotic boundary $S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$1, the target micro-causality statement is
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$2
This is to follow from bulk micro-causality,
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$3
The three conditions are: no independent propagation channels beyond those of the bulk theory, a local bulk-supported operator dictionary, and consistent matching between effective spacelike separation and dynamically accessible bulk causal curves (Deng, 9 Jan 2026).
In this account, nonlocal reconstruction within a code subspace is explicitly separated from signal propagation. The operator relation
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$4
is not an operator identity on the full Hilbert space but only on the semiclassical code subspace (Deng, 9 Jan 2026). This suggests that entanglement-based interior reconstruction does not by itself imply superluminal influence.
An algebraic variant of the island idea appears in the double-scaled SYK model with an end-of-the-world brane. There, the right boundary algebra $S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$5 remains a type $S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$6 factor with a nontrivial commutant, implying that full bulk reconstruction from the right boundary alone is impossible (Cao et al., 3 Nov 2025). In the semiclassical JT limit, the commutant becomes the canonical purification of the exterior algebra, and the hidden region between the horizon and the brane is termed a “no man’s island” (Cao et al., 3 Nov 2025). A plausible implication is that the island idea can be reformulated in operator-algebraic language without requiring an ordinary second asymptotic boundary.
6. Extensions, observational consequences, and outstanding tensions
The data set includes several directions in which the island rule acquires model-specific consequences beyond the basic Page-curve mechanism.
One extension concerns extremal or near-extremal black holes. In extremal RN, the no-island entropy is divergent because the reference point lies at a singular geometry, while the island prescription gives a finite late-time entropy and places the island near the horizon (Kim et al., 2021). However, the same source emphasizes that in extremal RN the island boundary hits the curvature singularity, so semiclassical control is questionable. The Hayward black hole is introduced as a regular alternative in which the same qualitative behavior—logarithmic early-time growth and constant late-time entropy—appears without the singularity pathology (Kim et al., 2021). Massive gravity exhibits an analogous issue: in the extremal charged case, the no-island entropy is ill-defined, signaling the need for UV completion, whereas the island saddle remains finite (Nam, 2021).
Another extension concerns thermodynamic phase structure. In four-dimensional RN-AdS black holes, the first-order phase transition modifies the shape of the Page curve because the effective temperature becomes branch dependent,
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$7
and the entanglement entropy is modeled as
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$8
(Lin et al., 2024). The data stresses that the Page curve can become monotonic, non-monotonic, or discontinuous, but unitarity is preserved because the entropy is always obtained by minimizing competing saddles.
The cosmological application goes further by proposing an observational imprint. In the inflationary JT-like setup, the usual superhorizon scalar spectrum without islands is
$S(R)=\min_I\,\mathrm{ext}\left[\frac{\mathrm{Area}(\partial I)}{4G_N}+S_{\text{bulk}(R\cup I)\right]$9
while the island picture leads to
$S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$0
which the paper interprets as a Page-like suppression at large scales in the primordial perturbation spectrum (Piao, 2023). This suggests a cosmological observable analogous to the turnover of black-hole radiation entropy.
At the same time, the literature summarized here records substantial conceptual caution. “Island mirages” argues that island-detection criteria may give false positives and also identifies an apparent tension with the quantum Bousso bound (Rolph, 2022). The tension is encapsulated by the chain
$S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$1
which clashes with
$S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$2
if one naively follows a shrinking representative $S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$3 all the way to the tip of $S_{\rm gen}(R\cup I)=\frac{\rm Area}(\partial I)}{4G_N}+S_{\rm mat}(R\cup I),$4 (Rolph, 2022). The resolutions discussed in the data are all limitations of the semiclassical approximation: truncation by singularities, failure of quantum lightsheets, blueshift beyond semiclassical validity, or sub-Planckian regions outside the trusted domain. This suggests that the island rule is powerful but not mechanically separable from the regime of validity of generalized entropy and QFC reasoning.
Taken together, these works define the island rule as a generalized-entropy prescription with broad applicability, a precise role in producing Page curves, and a nontrivial dependence on state, geometry, cutoff placement, and operator-algebraic interpretation. The literature represented here also shows that the rule is not a closed chapter: its sufficiency criteria, causal interpretation, and semiclassical limits remain active subjects of refinement (Bousso et al., 2021, Rolph, 2022, Deng, 9 Jan 2026).
