In-Principle versus Feasible Verification

A prediction can name a perfectly valid observable and still be unverifiable by the physical system under discussion. The obstruction may occur because accessible data do not identify the prediction, identical states cannot be prepared, the estimator needs too many copies, the decoder takes exponential time, or the apparatus destroys the regime it was meant to test. These are different failures. A useful verdict therefore reports the first failed layer instead of hiding every limitation inside the phrase “in principle.”

Required background. Quantum Information for Black-Hole Physics supplies states, channels, measurements, and error criteria; What the Equivalence Principles Do and Do Not Imply supplies operational clocks, detectors, energy, and disturbance.

Helpful background. Semiclassical Gravity supplies evaporation and backreaction scales; Quantum Fields in Curved Spacetime supplies radiation and detector observables; and Black-Hole Information supplies the recovery claims being compared.

Six layers of verification

Let a proposed experiment distinguish hypotheses $H_0$ and $H_1$ about a black-hole process. Its verification claim should be tested along six explicit axes.

Layer or comparison axis	Question that must be answered	Typical failure	Warranted conclusion
1. Definability	Is there a gauge-invariant observable or channel on a specified algebra and state domain?	Only a coordinate-local symbol or an operator outside the code subspace is given.	The prediction has not yet been made operationally precise.
2. Identifiability	Do the allowed outcome distributions differ under $H_0$ and $H_1$?	All accessible marginals agree even though inaccessible global states differ.	No amount of repetition of those measurements can decide the hypotheses.
3. State access	Can the observer collect the subsystem or prepare the independent copies assumed by the estimator?	One evaporation supplies correlated time bins, not repeatable copies of one state.	The statistical theorem’s access model is unavailable.
4. Computational tractability	Is there an algorithm with stated sample, gate, memory, and precision costs?	A recovery channel exists, but every known implementation is exponential.	Recoverability is established only information-theoretically.
5. Physical implementability	Can the protocol finish within causal, time, energy, storage, and backreaction budgets?	The computer outlives the hole or its stress-energy changes the background.	The algorithm is not a feasible experiment in that regime.
6. Evidential discrimination	Do signal, uncertainty, and alternatives support the claimed inference?	Noise or a coarse near-horizon model reproduces the statistic.	Data constrain a narrower dynamical claim, not fine-grained recovery.

For compact notation, write the six predicates as $D,I,A,C,P,E$. A feasible verification requires

$$V_{\mathrm{feasible}}=D\wedge I\wedge A\wedge C\wedge P\wedge E.$$

Definability alone does not imply the conjunction. Nor does failure of $C$ or $P$ refute the predicted observable: it limits the experiment, not the mathematical theory. Conversely, experimental difficulty cannot rescue an internally inconsistent prediction. The six-axis diagnosis preserves both directions of that logic.

Identifiability depends on the accessible algebra

For an allowed measurement family $\mathfrak M$, define its distinguishability of two states by

$$\Delta_{\mathfrak M}(\rho,\sigma) =\sup_{\{M_x\}\in\mathfrak M} \frac12\sum_x \left\lvert\operatorname{Tr}\!\left[M_x(\rho-\sigma)\right]\right\rvert.$$

If $\Delta_{\mathfrak M}=0$, the two hypotheses produce exactly the same data for every permitted measurement; increasing the number of shots cannot help. If it is nonzero but very small, bounded-outcome sample-mean estimation gives a sufficient shot count of order the inverse square of the separation, up to a confidence logarithm. That is not a universal lower bound: rare-event discrimination and collective measurements can have different scaling.

An exact finite-dimensional check uses

$$\lvert\mathrm{GHZ}_{\pm}\rangle =\frac{\lvert0\rangle^{\otimes n} \pm\lvert1\rangle^{\otimes n}}{\sqrt2}.$$

Every proper-subsystem reduced state is identical for the plus and minus states. If $\mathfrak M$ contains only measurements supported on fewer than $n$ qubits, then $\Delta_{\mathfrak M}=0$. Enlarging the measurement family to include $X^{\otimes n}$ distinguishes the states perfectly because its expectation is $+1$ or $-1$. The states, dynamics, and number of samples have not changed; only the access algebra has. This counterexample shows why “arbitrarily precise local observation” is not equivalent to access to a global fine-grained invariant.

The access restriction must be literal. Measuring $X$ locally on all $n$ qubits of the same preparation and multiplying the recorded outcomes also measures $X^{\otimes n}$; no entangling detector is needed. Thus the example restricts total support or access to joint records, not merely the locality of each elementary detector. Separate small-subsystem marginals contain no such parity information.

Gravity sharpens the first layer as well. Around flat spacetime, perturbative boundary-anchored dressings make operators gauge invariant but carry long-range fields and dressing-dependent nonlocal commutators Donnelly and Giddings 2016, §§ 2–5. That result obstructs naive compact tensor factors in its symmetry-rich background; it is not a theorem excluding every perturbatively compact relational algebra. Antonini and collaborators argue that backgrounds without exact isometries admit compactly supported gauge-invariant operators to all orders in perturbation theory, while also leaving exact nonperturbative compact support unresolved Antonini et al. 2025, §§ 4–5. Consequently, an identifiability claim must use the actual dressed algebra and state domain, not tensor factors inferred from a coordinate diagram.

Copies, algorithms, and physical resources

Suppose a radiation density operator on dimension $d$ is well defined. Entropy estimation is then mathematically meaningful, but entropy is nonlinear in the state. Finite-dimensional upper bounds and task-specific lower bounds quantify the number of independent copies needed for von Neumann or Rényi estimation Acharya et al. 2020, § 1.3 and Theorems 1–4, 7–8. The lower bounds beyond integral Rényi orders in that paper apply to the empirical estimator, not to every possible measurement strategy. These results assume an independent-copy access model. They do not manufacture identical evaporations, collect all modes, store them coherently, or implement the collective measurement.

Antonini and collaborators propose that Page- and Rényi-type observables remain definable for radiation in specified gravitating and massless-gravity settings through sufficiently many preparations Antonini et al. 2025, §§ 2–4. On this page’s comparison, that is a frontier advance at the definability and state-access boundary. It relies on a specified radiation algebra, background symmetries, perturbative control, and very large copy and precision resources. It is not evidence that astrophysical black holes can be reset to the same microscopic state.

Even when samples are supplied, computation is separate. Harlow and Hayden argue that the radiation-distillation task relevant to a firewall experiment is computationally obstructed under stated complexity assumptions Harlow and Hayden 2013, §§ 2–4. The conclusion is not a proved universal exponential lower bound for every physical black-hole decoder. It demonstrates why existence of a recovery map does not specify a feasible circuit.

Finally, a circuit description does not guarantee a gravitational implementation. The Margolus–Levitin bound constrains the rate of orthogonal state evolution by the energy above the ground state Margolus and Levitin 1998, Eqs. (4)–(6), pp. 190–191. Saturating that bound does not construct a computer, route gates, correct faults, cool memory, or keep its stress-energy perturbative. Time, energy, precision, collection, and backreaction must therefore remain independent columns in a resource claim.

A layer-by-layer decision procedure

Apply the comparison in order.

1. Specify the complete hypotheses, state family, physical algebra, and error norm. Reject undefined coordinate observables before counting resources.
2. Exhibit an allowed measurement whose distributions differ under the hypotheses. Report the separation and a coarse alternative that might mimic it.
3. State how subsystems and independent preparations are obtained. Do not relabel correlated radiation bins as copies.
4. Give an algorithm and count samples, gates, memory, communication, and precision separately. Label conjectural complexity assumptions.
5. Compare runtime with causal and evaporation deadlines, and compare apparatus energy and localization with the tolerated backreaction.
6. Propagate detector noise, loss, model systematics, and multiple-testing choices to the final discrimination. State what narrower claim survives.

The first failed step supplies the diagnosis. For example, a many-copy swap observable can pass definability and identifiability while failing state access. A formal decoder can pass the first three layers and fail computation. An executable simulator can pass through physical implementation for its own circuit yet fail evidential transfer to gravity because a different microscopic model produces the same reported statistic.

Results and open questions through 2026-08-19

This discussion reflects results available through 2026-08-19. A permissive position calls an observable verifiable in principle whenever each finite system admits a finite protocol, even if its time or copy count grows exponentially. A conservative operational position requires a realizable protocol within the same semiclassical regime and a nonzero discrimination from credible alternatives. The strongest intermediate position reports separate labels—defined, identifiable, sample-accessible, computable, implementable, and evidenced— instead of forcing one binary verdict. These competing positions are partly terminological, but the resource and algebra questions they expose are physical and remain unresolved.

Evidence or calculation and assumptions	Result	Scope or limitation and affected layer
Perturbative gravitational dressing in Donnelly and Giddings 2016, §§ 2–5	Boundary-anchored observables can be gauge invariant, with dressing-dependent nonlocal commutators.	Supports definability in perturbation theory; denies an automatic exact local subsystem for identifiability.
Finite-dimensional entropy estimation in Acharya et al. 2020, § 1.3 and Theorems 1–4, 7–8	Copy-complexity upper bounds exist for the stated entropy tasks; Theorem 1 gives matching integral-Rényi lower bounds, while Theorems 7–8 bound the empirical estimator.	Assumes independent copies; collection, gates, memory, precision, and backreaction are outside the result.
Gravitating-radiation proposal in Antonini et al. 2025, §§ 2–4	Page- and Rényi-type observables are argued to be definable using many preparations in specified gravity settings.	Frontier proposal with algebra, symmetry, perturbative, copy, and precision assumptions; physical state access is not established.
Radiation distillation in Harlow and Hayden 2013, §§ 2–4	An information-theoretic task can face a conjecturally prohibitive decoding cost.	Complexity model and assumptions matter; not a universal quantum lower bound or a claim of nonunitarity.
Energy-limited evolution in Margolus and Levitin 1998, Eqs. (4)–(6), pp. 190–191	Energy bounds the fastest orthogonal evolution of a specified system.	Necessary speed limit only; it supplies neither an architecture nor a weak-backreaction implementation.

The controlled results occupy different regimes: finite-dimensional quantum statistics, perturbative gravitational observables, a frontier many-copy gravity proposal, conjectural circuit complexity, and a general quantum speed limit. No one row completes all six layers for an evaporating astrophysical black hole.

Interpretation and open questions

“Not feasible” should always name a system size, access model, error tolerance, resource ceiling, and deadline. An exponential scaling can be finite for every finite input yet physically hopeless in the regime of interest. “Not identifiable” is stronger relative to a fixed measurement family: the allowed data distributions coincide. “Not defined” is earlier still. Keeping these verdicts separate prevents an operational limitation from being mistaken for information loss.

Generic quantum-information theory, QFT detector construction, stable semiclassical thermodynamics, and explicit string or holographic models are treated in their specialist resources. This page compares only the layers that connect such results to a bounded verification claim.

The conclusions could change after a gravitational many-copy observable is implemented with quantified copy and precision scaling, or a decoder lower bound is proved for a physically specified channel and error norm, or a protocol completes all six layers within causal, energy, collection, and backreaction budgets while excluding a matched coarse alternative.

Exercise

A proposal defines a radiation purity observable using two copies, proves that the two hypotheses give expectation values separated by $0.2$, and provides a polynomial-size circuit. The source is a single astronomical evaporation that cannot be repeated. Which layers pass, which first fails, and what conclusion is justified?

Solution

Definability passes because the two-copy observable is specified. Identifiability also passes in the hypothetical two-copy access model because the predicted distributions have a nonzero separation. The polynomial circuit would support computational tractability if its gate, precision, and error claims were supplied. State access fails first: one evaporation is not two independent preparations of the same radiation density operator. Therefore the proposal is an identifiable and potentially efficient many-copy test, but not a feasible test of that astronomical event. No conclusion about its purity follows from the unavailable protocol.

References

• Acharya, Jayadev, Ibrahim Issa, Nirmal V. Shende, and Aaron B. Wagner. 2020. “Estimating Quantum Entropy.” IEEE Journal on Selected Areas in Information Theory 1, 454–468. DOI. arXiv:1711.00814.
• Antonini, Stefano, Chang-Han Chen, Henry Maxfield, and Geoff Penington. 2025. “An Apologia for Islands.” arXiv:2506.04311. arXiv.
• Donnelly, William, and Steven B. Giddings. 2016. “Observables, Gravitational Dressing, and Obstructions to Locality and Subsystems.” Physical Review D 94, 104038. DOI. arXiv:1607.01025.
• Harlow, Daniel, and Patrick Hayden. 2013. “Quantum Computation versus Firewalls.” Journal of High Energy Physics 06, 085. DOI. arXiv:1301.4504.
• Margolus, Norman, and Lev B. Levitin. 1998. “The Maximum Speed of Dynamical Evolution.” Physica D 120, 188–195. DOI.