Approximate Subsystem Erasure Codes

• The paper introduces a framework where only the information-carrying subsystem is recovered, enabling high-fidelity logical restoration even if the gauge subsystem is disturbed.
• The methodology applies a perturbative form of the Knill-Laflamme conditions with the transpose channel to achieve approximate recovery under erasure-like noise models.
• This approach has implications for quantum error correction and holographic applications, linking logical recovery with state-dependent geometric entropy and gravitational backreaction.

Searching arXiv for the cited papers to ground the article in current metadata. Approximate subsystem erasure-correcting codes are quantum error-correcting codes in which logical information is stored only in a designated subsystem and is required to be recoverable with high fidelity even when another subsystem is noisy, disturbed, or effectively lost. In the formulation developed in "Towards a Unified Framework for Approximate Quantum Error Correction" (Mandayam et al., 2012), the code extends approximate subspace quantum error correction to subsystem codes, so that recovery need only restore the reduced state on the information-carrying subsystem rather than the full encoded state. This framework includes approximate erasure correction as a special case when the noise effectively destroys part of the system, and later work has used approximate subsystem erasure-correcting codes to formulate state-dependent geometric entropy in holography-like settings (Cao et al., 13 Mar 2026).

1. Subsystem structure and encoded information

The subsystem formalism begins with a Hilbert-space decomposition

$$\mathcal H=\mathcal H_A\otimes \mathcal H_B\oplus \mathcal K,$$

and defines the code as the set of product states supported on $\mathcal H_A\otimes \mathcal H_B$: $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$ Within this structure, subsystem $A$ is the information-carrying subsystem, whereas subsystem $B$ is the noisy or gauge subsystem whose state may be arbitrary and need not be preserved. Two encoded states that differ only on $B$,

$$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$

represent the same logical information. This is the defining operational distinction between subsystem and subspace coding (Mandayam et al., 2012).

A subspace code is the special case in which $B$ is trivial. In that setting, the whole code space must be recovered. In a subsystem code, by contrast, only the reduced state on $A$ must be recovered: $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$ This relaxation is precisely what makes subsystem coding natural for erasure-like scenarios. If one component of the encoding is allowed to absorb disturbance, then protection of the logical degrees of freedom can remain meaningful even when the full encoded state is not reproducible.

A common misconception is that subsystem codes merely restate subspace codes in different notation. The formalism above shows otherwise: the equivalence class structure on $\mathcal H_A\otimes \mathcal H_B$0 changes both the recovery objective and the correctability conditions. Approximate subsystem erasure correction is therefore not simply approximate recovery on a subspace, but approximate recovery modulo a gauge subsystem.

2. Noise model and approximate erasure criterion

The noise process is a CPTP channel

$\mathcal H_A\otimes \mathcal H_B$1

acting on $\mathcal H_A\otimes \mathcal H_B$2. The framework is not restricted to literal erasure channels, but it is explicitly suited to erasure-like situations because the code is designed to preserve information in $\mathcal H_A\otimes \mathcal H_B$3 even if $\mathcal H_A\otimes \mathcal H_B$4 is severely disturbed or effectively lost (Mandayam et al., 2012).

Approximate correction is quantified by fidelity on the logical subsystem. For pure $\mathcal H_A\otimes \mathcal H_B$5,

$\mathcal H_A\otimes \mathcal H_B$6

The fidelity loss for a state $\mathcal H_A\otimes \mathcal H_B$7 is defined as

$\mathcal H_A\otimes \mathcal H_B$8

and for a code $\mathcal H_A\otimes \mathcal H_B$9,

$$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$0

The optimal recovery error is

$$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$1

A code is perfectly correctable on $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$2 if $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$3. It is $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$4-correctable on $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$5 if

$$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$6

This is the approximate subsystem version of quantum error correction used in (Mandayam et al., 2012). The formulation is operational: correctness is judged only by recoverability of the information-bearing subsystem. In the erasure interpretation, loss of the gauge subsystem is admissible so long as the induced disturbance on $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$7 remains small.

This suggests a useful conceptual distinction between exact erasure correction and approximate subsystem erasure correction. Exact correction requires a strict logical/algebraic decoupling, whereas the approximate notion permits residual coupling to the gauge sector provided its effect on the logical marginal is fidelity-controlled.

3. Exact and approximate correctability conditions

The central recovery map is the transpose channel, defined in representation-independent form by

$$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$8

where $$\mathcal C=\{P=P_A\otimes P_B,\;\forall\,P_A\in\mathcal S(\mathcal H_A),\;P_B\in\mathcal S(\mathcal H_B)\}.$$9 projects onto the code support and

$A$0

uses the inverse on the support of $A$1. Equivalently, its Kraus operators are

$A$2

This map plays a dual role: it yields a compact expression for exact correctability and furnishes a practical recovery for approximate codes (Mandayam et al., 2012).

The exact subsystem condition appears as Theorem 1: $A$3 for all $A$4, with $A$5. This is the subsystem analogue of the Knill-Laflamme conditions, rewritten so that the transpose channel is explicit. The equation states that, within the code support, the noise-recovery overlaps must act trivially on the information subsystem and only nontrivially on the gauge subsystem.

Approximate correctability is obtained by perturbing this exact identity. Theorem 2 assumes

$A$6

for all $A$7, with $A$8. Then $A$9 is $B$0-correctable on $B$1 under $B$2 for $B$3, where

$B$4

The quantity $B$5 is a tight sufficient bound expressed directly in terms of the deviation $B$6 from exact subsystem correctability.

For construction and verification, the paper also derives a simpler corollary: $B$7 This operator-norm condition is easier to check and functions as a practical sufficient criterion for approximate subsystem codes.

A frequent misunderstanding is that approximate criteria merely weaken exact conditions without changing the structure of recovery theory. In this framework, the perturbative form of the subsystem Knill-Laflamme relation is itself the organizing principle: the error terms $B$8 are not auxiliary quantities but the objects that determine the fidelity guarantee.

4. Near-optimality of the transpose channel

A major result of (Mandayam et al., 2012) is that the transpose channel is often nearly as effective as the optimal recovery channel. The paper proves this not in full generality, but in several restricted yet broad scenarios.

For subspace codes, corresponding to trivial $B$9, Theorem 4 gives

$B$0

where for a subspace code of dimension $B$1,

$B$2

as stated in the paper. The significance is that small optimal error implies small transpose-channel error.

When the gauge subsystem is maximally mixed, the relevant subset of code states is

$B$3

with induced channel on $B$4

$B$5

In this case,

$B$6

and since $B$7,

$B$8

hence

$B$9

The paper also notes a state-dependent transpose channel when $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$0 is known to be in a fixed state $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$1, with Kraus operators

$$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$2

If subsystem $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$3 is itself perfectly correctable, Lemma 6 states that for any pure $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$4, the transpose-channel fidelity

$$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$5

is independent of the initial $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$6-state. This yields

$$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$7

The broadest erasure-like statement appears as Condition 1. If there exists $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$8 such that for all $$P_A\otimes T_B \qquad\text{and}\qquad P_A\otimes O_B,$$9,

$B$0

then

$B$1

This condition formalizes the idea that the noise nearly destroys distinguishability on $B$2: once the channel largely erases gauge information, the transpose channel becomes near-optimal for logical recovery on $B$3 (Mandayam et al., 2012).

A plausible implication is that approximate subsystem erasure correction is especially natural when the physical noise itself suppresses access to the gauge degrees of freedom. In that regime, the code architecture and the channel structure are aligned.

5. Erasure interpretation and holographic generalization

In the 2012 framework, approximate erasure correction appears as a special case of approximate subsystem quantum error correction whenever the noise effectively destroys part of the system (Mandayam et al., 2012). A later development, "State-dependent geometries from magic-enriched quantum codes" (Cao et al., 13 Mar 2026), argues that this approximate setting is not merely technically convenient but structurally necessary in holographic applications.

That work begins from exact subsystem complementary recovery, with logical and physical factorizations

$B$4

In the exact case, recovery is described by fixed local unitaries $B$5 and $B$6, and the geometric resource state $B$7 is independent of the logical state. The paper argues that this implies state-independent geometric entropy and therefore cannot encode gravitational backreaction. Approximate subsystem recovery, by contrast, replaces strict factorization by a state-dependent mixture,

$B$8

so that geometry depends on encoded logical content.

The paper then proposes an RT-like decomposition for approximate subsystem erasure-correcting codes. For an encode-recover channel $B$9, the optimal recovery is defined by maximizing coherent information,

$A$0

and the analog bulk matter entropy is the entropy of the optimally recoverable bulk state. The complementary geometric term is the proto-area entropy

$A$1

In exact codes this reduces to the fixed area term of the usual RT/FLM picture, whereas in approximate codes the proto-area becomes state dependent.

For a broad class of skewed codes obtained as small nonlocal perturbations of exact codes,

$A$2

the paper derives expansions showing that, after Haar averaging over local basis changes, the proto-area correction behaves monotonically with bulk entropy. It identifies the origin of this response as tripartite non-local magic in the Choi state of the encoding map, a resource that vanishes in stabilizer codes and controls the leading matter-geometry coupling in approximate subsystem erasure-correcting codes. The authors are careful not to claim an exact identification with quantum extremal surface area, but rather a strong qualitative analogy (Cao et al., 13 Mar 2026).

This later development reframes approximate subsystem erasure correction as a mechanism for controlled state dependence. Exact subsystem recovery corresponds to fixed-background behavior; approximate recovery permits matter-geometry coupling.

6. Related classical analogues and scope limitations

Two classical coding results in the supplied literature are relevant by analogy but are not direct instances of approximate subsystem erasure-correcting codes.

"Correcting an ordered deletion-erasure" (Ganesan, 2018) studies a binary channel with one deletion and at most one erasure, under the promise that the deletion occurs before the erasure. The code is a two-parameter Varshamov-Tenengolts code $A$3 satisfying

$A$4

and for all $A$5 there exists an $A$6-length code $A$7 with redundancy

$A$8

that corrects all ordered deletion-erasure patterns. The lower bound

$A$9

holds for all large $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$0. The result is relevant here because it provides a classical analogue of structured erasure recovery with very low overhead and known erasure location, but the paper itself is not written in quantum-language terms.

"Generalized Concatenated Types of Codes for Erasure Correction" (Blaum et al., 2014) studies generalized concatenated, or integrated interleaved, codes as erasure-correcting array codes with local and global protection. A $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$1-level GC code on an $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$2 array over $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$3 with $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$4 has dimension

$$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$5

and Theorem 2.1 states that it can correct up to $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$6 erasures in any $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$7 rows for every $$\operatorname{tr}_B[(\mathcal R\circ\mathcal E)(\rho)]\approx \operatorname{tr}_B(\rho),\qquad \rho\in\mathcal C.$$8. These codes are explicitly motivated by RAID-like architectures and combine local Reed-Solomon correction with global inter-row coupling. The paper emphasizes, however, that GC/II codes are weaker than PMDS and are not about quantum subsystem codes, gauge subsystems, stabilizers, or approximate recovery maps.

The scope limitation is therefore clear. Approximate subsystem erasure-correcting codes, in the sense established in (Mandayam et al., 2012), are quantum subsystem codes analyzed via fidelity, transpose-channel recovery, and perturbative subsystem Knill-Laflamme conditions. Classical deletion-erasure or layered erasure codes can illuminate structural themes such as locality, side information, and hierarchical recovery, but they do not supply the operator-algebraic notion of subsystem approximate correction.

Taken together, these works place approximate subsystem erasure-correcting codes at the intersection of approximate quantum recovery, subsystem encoding, and erasure-oriented noise models. Their distinctive contribution is to make high-fidelity logical recovery possible even when full-state recovery is impossible, and to do so in a framework where the transpose channel is both analytically tractable and, in several important regimes, nearly optimal.