Limits of Page-Curve Diagnostics

A Page-shaped entropy curve is a stringent consistency check, but it is not by itself a proof of a unitary S-matrix, a microscopic decoding mechanism, or operational information recovery. Its force depends on the state, radiation algebra, regulator, ensemble, and dynamics. Several inequivalent theories can produce the same scalar function $S_R(t)$ while storing—or destroying—information in different ways.

Required background. Quantum Information supplies channels, entropy, distinguishability, ensembles, and recovery criteria used in the diagnostics.

Helpful background. Semiclassical Gravity supplies evaporation and gravitational-entropy regimes; Quantum Fields in Curved Spacetime supplies the mode and regulator choices; Black-Hole Information supplies the complete-system formulation against which a curve is interpreted.

What the curve establishes

For a declared state $\rho_R(t)$, a computed curve establishes the value of one nonlinear spectral functional,

$$S_R(t)=-\operatorname{Tr}\!\left[\rho_R(t)\ln\rho_R(t)\right],$$

to the accuracy of the calculation. If the complete state on $RB$ is pure, $S_R=S_B$ follows from purification; the equality alone does not establish global purity. If earlier radiation remains isolated and the only change is adding a new mode, a downward step requires correlations absent from independent thermal emission. Entropy can also fall under cooling, resetting, or a change of retained algebra, so that conclusion needs the stated passive-collection assumption. If $S_R$ reaches zero and $R$ is the complete final system, the particular final state is pure.

None of these statements alone tests preservation of inner products for every input. A unitary S-matrix must map an orthonormal basis to an orthonormal basis, including phases and reference entanglement. Entropy of one state, even at all times, does not reconstruct that map.

Page’s random-state result is exact for normalized Haar measure Page 1993, Eqs. (1)–(11), with the exact formula proved by Sen Sen 1996, pp. 1–3 of the published article. Applying its shape to evaporation adds energy-shell, locality, charge, and mixing assumptions. Those additions need independent evidence.

Four ambiguity axes

Subsystem and algebra. Radiation may mean outgoing wave packets, a bath interval, an asymptotic algebra, or detector records. Moving a boundary or including a gravitationally encoded interior changes the state being traced. A turnover for one algebra need not be a turnover for another.

Regulator and subtraction. Continuum entropy is divergent. A fixed cutoff, renormalized difference, or generalized entropy must be declared. Comparing values computed with different counterterms can manufacture a transition. Rényi indices also need compatible prescriptions.

Ensemble and realization. An ensemble average can be smooth even when individual realizations fluctuate, and a gravitational path integral may compute an ensemble-like observable rather than a fixed microscopic theory. Conversely, typicality can justify concentration only after the relevant measure and constrained Hilbert space are known.

Formal and operational recovery. Small $S_R$ says the radiation state is nearly pure, not that an observer can identify which pure state was produced or decode a reference before evaporation. Recovery fidelity, sample complexity, circuit complexity, energy, causal time, and backreaction are separate quantities.

These axes also separate coarse-grained thermodynamic entropy from fine-grained entropy. The former can keep increasing while the latter decreases; agreement before the Page time does not make them identical definitions.

Counterexamples to overinterpretation

Consider the replacement channel

$$\Phi(\rho)=\lvert0\rangle\langle0\rvert$$

for every input. Its output entropy is zero, so its “curve” could end at the unitary value. Yet it maps every pair of distinct inputs to the same state and destroys all reference correlations. Final purity is therefore not sufficient for unitarity.

Conversely, an isometry $\lvert i\rangle\mapsto\lvert r_i\rangle_R\lvert h_i\rangle_H$ can be exactly unitary on the enlarged system while $S(R)$ remains nonzero because a remnant $H$ persists. Final radiation purity is not necessary for enlarged unitarity unless radiation is declared complete.

A third counterexample uses spectra. Two three-level density matrices can share the same von Neumann entropy but have different purities and different responses to measurements. Matching $S_1(t)$ therefore does not imply matching Rényi curves, eigenvalue distributions, or correlation functions. A robust diagnosis combines entropies with reference-sensitive channel tests.

A diagnostic checklist

Before interpreting a Page curve, ask:

1. What is the complete global state and which system purifies $R$?
2. Is $R$ a tensor factor, an observable algebra, a bath region, or a detector record?
3. Which regulator, subtraction, and logarithm convention define the entropy?
4. Is the result for one theory and state, a typical-state claim, or an ensemble average?
5. Which saddle, replica continuation, and approximation regime produce the curve?
6. What independent check tests trace, positivity, endpoint purity, and reference entanglement?
7. Does a decoder exist, and can it operate within causal, complexity, energy, and backreaction bounds?

Almheiri and collaborators distinguish the controlled entropy statements from their broader microscopic interpretations across the relevant gravitational models Almheiri et al. 2021, §§ 4–10 and Appendix A. The checklist is designed to keep that boundary visible.

Results and open questions as of 2026-08-19

It remains unresolved which collection of Page-curve diagnostics is sufficient for realistic four-dimensional black-hole information recovery. Competing positions assign different evidentiary weight to entropy turnover, single-theory factorization, and operational decoding.

Evidence or diagnostic	Result	Scope and limitation
Haar average Page 1993, Eqs. (1)–(11) and Sen 1996, published pp. 1–3	Gives an exact mean bipartite entropy benchmark	Imposes no evaporation dynamics, energy constraint, or spatial locality
Leading Hawking curve Almheiri et al. 2021, §§ 2–4	Independent pairs yield monotonic radiation entropy	Valid in the stated semiclassical factorization regime, not at an uncontrolled endpoint
Controlled gravitational Page transitions Almheiri et al. 2021, §§ 5–8	Fine-grained entropy of a declared bath region can turn over	Most explicit in low-dimensional semiclassical models; microscopic single-theory meaning requires additional work
Rényi and spectral agreement	Tests more of the density-matrix spectrum than $S_1$ alone	Finite moments can be insufficient, noisy reconstruction is ill-conditioned, and continuation is assumption-dependent
Reference-entangled recovery test	Distinguishes an information-preserving channel from a fixed pure-state replacement	Formal recovery may still be computationally or gravitationally inaccessible

The permissive interpretation regards a controlled Page transition as strong evidence that semiclassical gravity knows about unitary fine-grained entropy. The conservative interpretation accepts that result while withholding a claim about a unique microscopic S-matrix or feasible decoder. A stronger operational position requires an explicit channel protocol. Their disagreement depends on formulation and regime, not on the definition of entropy alone.

The conclusions could change after a single-theory microscopic model matches von Neumann and Rényi data with controlled errors, or after a theorem links a gravitational entropy curve to reference-preserving channel fidelity, or after an executable decoder passes causal, complexity, energy, and backreaction tests.

Exercise

A proposed evaporation map sends both $\lvert0\rangle$ and $\lvert1\rangle$ to the pure radiation state $\lvert r\rangle$. Its final radiation entropy is zero. Which diagnostic rejects the claim that the map is unitary?

Solution

Inner-product preservation rejects it: orthogonal inputs have overlap zero, but their outputs have overlap one. Equivalently, entangle the input with a reference. The replacement map destroys the reference–input entanglement and produces a product state, so its entanglement fidelity is poor despite zero output entropy. The example shows why endpoint purity for one marginal must be supplemented by a channel or reference-sensitive test.

References

• Almheiri, Ahmed, Thomas Hartman, Juan Maldacena, Edgar Shaghoulian, and Amirhossein Tajdini. 2021. “The Entropy of Hawking Radiation.” Reviews of Modern Physics 93: 035002. DOI.
• Page, Don N. 1993. “Average Entropy of a Subsystem.” Physical Review Letters 71: 1291–1294. Open text.
• Sen, Siddhartha. 1996. “Average Entropy of a Quantum Subsystem.” Physical Review Letters 77: 1–3. DOI. Open PDF, preprint titled “Average Entropy of a Subsystem”.