Black Holes as Approximate Quantum Codes

Calling a black hole a quantum code is meaningful only after specifying what is encoded, which degrees of freedom are lost, and how recovery error is measured. The useful statement is not that gravity protects all bulk information. It is that a selected logical algebra on a finite code subspace may have redundant exterior representations, so specified erasures can be corrected approximately. The approximation can worsen when the erased region, the set of states, or the logical algebra is enlarged. This page develops that criterion and keeps the gravitational interpretation separate from the exact finite-dimensional theorem.

Required background. Quantum Information supplies channels, complementary channels, trace distance, and recovery maps; Gauge Theory and String-Theory Preparation supplies gauge constraints and holographic language needed to interpret the physical encoding.

Helpful background. Semiclassical Gravity supplies the regime in which a bulk geometry is meaningful; Quantum Fields in Curved Spacetime supplies the exterior field algebra; and Black-Hole Information supplies the fine-grained recovery question and its subsystem qualifications.

Logical information and the physical encoding

The logical Hilbert space is mapped into a larger physical Hilbert space; the distinction fixes which transformations count as encoding, noise, and recovery. Let $\mathcal H_L$ be a finite-dimensional logical Hilbert space and $\mathcal H_P=\mathcal H_X\otimes\mathcal H_{\bar X}$ the regulated physical Hilbert space. An isometry

$$V:\mathcal H_L\longrightarrow\mathcal H_P, \qquad V^\dagger V=I_L,$$

defines the code subspace $\mathcal H_{\rm code}=V\mathcal H_L$. The noise channel that erases $X$ is

$$\mathcal N(\rho_L) =\operatorname{Tr}_X(V\rho_LV^\dagger),$$

with output on $\bar X$. A decoder is a completely positive trace-preserving map $\mathcal R:\mathcal B(\mathcal H_{\bar X})\to \mathcal B(\mathcal H_L)$. Full-subspace recovery demands that $\mathcal R\circ\mathcal N$ approximate the identity channel for every code state, including states entangled with a reference. Operator-algebra recovery asks less: it preserves expectation values only for a declared logical algebra $\mathcal A_L$, which may have a nontrivial center.

These distinctions matter in gravity. The logical data might be low-energy fields in one semiclassical wedge, whereas the physical degrees of freedom are boundary or radiation variables. Gauge constraints obstruct an exact local tensor factorization. The finite tensor product above is therefore a regulated model whose algebra and dressing must be stated, not an assertion about the fundamental Hilbert space.

Exact erasure correction from environmental blindness

Choose an orthonormal logical basis $\{|i\rangle_L\}$. The erased subsystem $X$ learns nothing about the logical state exactly when its complementary channel is constant:

$$\operatorname{Tr}_{\bar X} \left(V|i\rangle\langle j|V^\dagger\right) =\delta_{ij}\,\sigma_X$$

for one density operator $\sigma_X$. This is the erasure form of the Knill–Laflamme condition. It says both that all basis states give the same environment state and that logical coherences have no off-diagonal imprint in $X$. Under these hypotheses, the Schmidt data can be chosen so that a channel on $\bar X$ reverses the encoding.

The proof architecture is purification uniqueness. Introduce a reference $R$ maximally entangled with $L$ and encode $L$. If the reduced state on $RX$ factorizes as $\rho_R\otimes\sigma_X$, then $R$ is decoupled from the erased system. Two purifications of that same $RX$ state differ by an isometry on $\bar X$, which supplies the recovery operation. Conversely, exact recovery prevents $X$ from distinguishing logical inputs, because a channel cannot create two copies of arbitrary quantum information. The code-subspace and multiple-reconstruction interpretation is developed in Almheiri, Dong, and Harlow 2015, §§ 2–4.

For a logical algebra rather than the entire matrix algebra, the condition is that products of error operators act in the commutant on the code:

$$V^\dagger E_a^\dagger E_b V\in\mathcal A_L',$$

where the physical error products have been pulled back to the logical Hilbert space before taking the commutant. Equivalently, their code-subspace restrictions commute with the represented logical algebra $V\mathcal A_L V^\dagger$. Central classical information can then remain visible to both complementary regions without violating no cloning. Harlow develops the conventional, subsystem, and operator-algebra erasure conditions in Harlow 2017, §§ 3–5.

Approximate recovery and an explicit error

Exact environmental blindness is too rigid for finite $G$, finite $N$, or a large code subspace. Let $\widehat{\mathcal N}$ be the complementary channel with output on $X$, and let $\mathcal C(\rho)=\sigma_X\operatorname{Tr}\rho$ be a constant channel. A useful state-independent criterion is

$$\left\lVert\widehat{\mathcal N}-\mathcal C\right\rVert_\diamond \le\delta.$$

The channel information–disturbance theorem then implies the existence of a recovery map whose worst-case error is controlled by a function of $\delta$ Kretschmann, Schlingemann, and Werner 2008, Theorem 3. The precise constants depend on whether error is stated as diamond distance, entanglement fidelity, or purified distance. One must not switch norms silently. A weaker average-state statement may look small while hiding a bad state in a growing code subspace.

As an elementary independent check, encode one logical qubit as $|0_L\rangle=|000\rangle$ and $|1_L\rangle=|111\rangle$. Erasure of a qubit in the computational basis reveals whether the logical input was $0$ or $1$; $\sigma_X$ is not constant, so this repetition code does not correct an arbitrary qubit erasure even though it corrects a classical bit flip. By contrast, a genuine quantum erasure code makes the lost subsystem independent of both populations and phases. This counterexample exposes why redundancy of classical labels is insufficient.

What the gravitational analogy commits to

A black-hole application must name at least five items: the logical algebra, the allowed state family, the physical output algebra, the erased subsystem, and the error norm. Entanglement-wedge reconstruction can realize multiple representations of one logical operator on different boundary regions, but only inside the stated code subspace and perturbative regime. Enlarging the subspace until it includes macroscopically distinct geometries can change the center, invalidate a common reconstruction, or make nonperturbative errors order one.

Approximate correction also does not guarantee a feasible decoder. The existence proof may use an isometry of exponential circuit complexity. Nor does it show that an evaporating asymptotically flat hole realizes the same code as an AdS construction. Generic quantum-information theory and explicit string or holographic constructions remain outside this page; stable thermodynamics belongs to Semiclassical Gravity. The bounded result is a criterion that any proposed exterior encoding must satisfy.

Check your understanding

Suppose two orthogonal logical inputs give erased-system states $\rho_X^0$ and $\rho_X^1$ with trace distance $\tfrac12\lVert\rho_X^0-\rho_X^1\rVert_1=0.3$. Can erasure of $X$ be exactly correctable for the full logical qubit?

Solution

No. Exact correction requires the complementary channel to be constant on all logical states, so in particular $\rho_X^0=\rho_X^1$. Their nonzero trace distance means that a measurement on the erased system distinguishes the two inputs with an advantage. That leaked information cannot coexist with exact recovery of arbitrary superpositions from $\bar X$. Approximate recovery may still be possible, but its error must be bounded in a declared norm and cannot be set to zero.

References

• Almheiri, Ahmed, Xi Dong, and Daniel Harlow. 2015. “Bulk Locality and Quantum Error Correction in AdS/CFT.” Journal of High Energy Physics 04, 163. arXiv:1411.7041.
• Harlow, Daniel. 2017. “The Ryu–Takayanagi Formula from Quantum Error Correction.” Communications in Mathematical Physics 354, 865–912. DOI. arXiv:1607.03901.
• Kretschmann, Dennis, Dirk Schlingemann, and Reinhard F. Werner. 2008. “The Information–Disturbance Tradeoff and the Continuity of Stinespring’s Representation.” IEEE Transactions on Information Theory 54, 1708–1717. DOI. arXiv:quant-ph/0605009.