{"id":"d26bab33-25d1-43ce-b26f-04d4c870cdb5","arxiv_id":"2607.22995","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Approximate quantum error correction acquires an error-set model: bounded-mixing linear families of noise channels are uniformly correctable from a single geometric code parameter.","lead":"This paper gives approximate quantum error correction the same 'error-set' structure that makes exact quantum codes powerful: one condition on a code protects it against an entire family of noise channels. It then builds good code families for deletion errors, fermionic systems, and Rydberg atom chains, areas that previously lacked such codes.","discovery_kind":"paradigm_shift","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Normalization-dependence of ℰ-controlled channels makes the error-set model non-intrinsic; without a canonical scaling rule, uniform guarantees are properties of (ℰ, scaling), not of the error set alone.","rationale":"The reader's weakest assumption identifies precisely the normalization-dependence of Definition 3, and my analysis agrees that this is the most load-bearing soft spot in the central claim. The uniform guarantee in Theorem A is only meaningful once the scaling of ℰ is fixed; different scalings produce different controlled channel families and different ζ parameters, so the same physical error set can yield different coverage. This is not merely a technical nuisance: it breaks the analogy with exact QEC, where linear span is invariant under rescaling, and it means that the phrase 'ε-AQEC for the error set ℰ' is not fully determined by ℰ alone. I checked that this is not merely an artifact of dependent error sets: the two-projector example is linearly independent, so Observation 1 does not rescue it. The paper's own Remark 2 acknowledges this and adopts model-dependent normalizations, which is an honest limitation but a real one. I do not think this invalidates the theorems: for any fixed normalization, the proofs appear consistent, and the paper's examples choose normalizations that include their intended physical channels. The concern therefore strengthens the case for the CONDITIONAL verdict already reached, rather than changing it. The necessary fix is to either prove scale-invariance of the relevant conclusions under a natural normalization (e.g., requiring the error set to be a Kraus set of a CPTP map, or imposing an ℓ2 normalization tied to the channel family) or to state explicitly that the model is parameterized by (ℰ, scaling) and not by ℰ alone. The concrete test above would settle whether the concern lands by exhibiting the non-invariance in the simplest nontrivial case.","tokens_in":65820,"tokens_out":18833,"duration_ms":191485,"concrete_test":"Compute the following qubit example: fix Q=ℂ², ℰ={Π0,Π1}, ℰ'={Π0/√2,Π1/√2}, and N_dep with Kraus operators {(Π0+Π1)/√2, (Π0−Π1)/√2}. Verify (i) ‖C‖∞=1 for N_dep with respect to ℰ and √2 with respect to ℰ', so N_dep∈N(ℰ) but N_dep∉N(ℰ'); and (ii) ζ(ℰ,Q) and ζ(ℰ',Q) differ exactly by a factor of 2. If both hold, the controlled family and the Theorem 3 sufficient threshold are normalization-dependent, confirming that the claimed 'error-set model' requires a canonical scaling rule to be an intrinsic model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Definition 3 makes N(ℰ) depend on the operator normalization; Observation 1 only removes dependence on the Kraus representation when ℰ is linearly independent, not on rescaling of the operators in ℰ. Remark 2 concedes this. This is load-bearing because every uniform guarantee — Theorems 3, 11, 15 and the Section 5 asymptotics — is a statement about N(ℰ). Concretely, on ℂ² take ℰ={Π0,Π1} with Π_i=|i⟩⟨i|, and the depolarizing channel N_dep with Kraus operators (Π0+Π1)/√2 and (Π0−Π1)/√2. Its coefficient matrix with respect to ℰ has rows (1/√2,1/√2) and (1/√2,−1/√2), so ‖C‖∞=1 and N_dep∈N(ℰ). But with ℰ'={Π0/√2,Π1/√2}, the coefficient matrix doubles entrywise and has norm √2>1, so N_dep∉N(ℰ'). Thus rescaling changes the family of channels to which Theorem 3 applies. Moreover, ζ(ℰ',Q)=ζ(ℰ,Q)/2 for any code Q, because B'_{kl}=(1/2)B_{kl} after the corresponding rescaling of λ; hence the sufficient threshold ε²/2 changes under rescaling. The paper fixes normalizations per platform (unitaries for Pauli/Majorana, binomial-normalized for photon loss), but this makes the central claim conditional on an externally supplied normalization rule rather than a property of the error set alone. In exact QEC, Span(ℰ) is scaling-invariant, so the advertised analogy is weakened exactly where the model needs to be canonical.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a theory of approximate quantum error correction built around an error-set model. A channel is called ℰ-controlled if it has a Kraus representation whose operators are linear combinations of a finite error set ℰ with coefficient matrix of spectral norm at most 1. The main sufficient condition (Theorem 3) states that if the environment-leakage distance ζ(ℰ,Q) ≤ ε²/2, then Q is an ε-AQEC code for every ℰ-controlled channel, via the Bény–Oreshkov framework. An average-case analog (Theorem 11) is proved using a new Knill–Laflamme Hellinger distance and contraction properties of the Petz map. The paper also proves an approximate erasure/general-error equivalence (Theorem 15), links the framework to subsystem variance and circuit complexity, and constructs partition-based codes in qudit, Rydberg-blockaded, Majorana fermionic, constant-excitation Fock, and permutation-invariant systems, claiming the first asymptotically good families for deletions, Majorana fermions, and Rydberg chains.","tokens_in":66170,"tokens_out":13333,"duration_ms":131852,"significance":"If correct, the results would provide a structural, adversarial alternative to channel-by-channel AQEC and extend several exact-QEC organizing principles—distance, erasure/error equivalence, and asymptotic code families—to the approximate case. The paper has clear strengths: the proofs are detailed; Theorem 3 is derived, not assumed, from Bény–Oreshkov; the Hellinger-distance interpretation of the Petz-map criterion is elegant; and the partition-code framework is broadly applicable. However, the central model is not canonical: the definition of ℰ-controlled channels depends on the normalization of ℰ, as Remark 2 concedes, and the claimed uniform guarantee is per-channel recoverability rather than a single universal decoder. These caveats temper the advertised analogy to exact QEC. The paper is a serious candidate for publication if the normalization issue is addressed or the claims are appropriately reframed.","major_comments":[{"comment":"N(ℰ) is not invariant under rescaling of ℰ, and Remark 2 concedes this. Every uniform guarantee in the paper (Theorems 3, 11, 15; Section 5) is a statement about N(ℰ), so the model is a pair (ℰ, scaling), not an intrinsic property of Span(ℰ). Example: on C², ℰ={Π0,Π1}; the depolarizing channel with Kraus operators (Π0±Π1)/√2 has coefficient matrix of norm 1 and belongs to N(ℰ). With ℰ'={Π0/√2,Π1/√2}, its coefficient matrix is [[1,1],[1,-1]], norm √2, so it is not in N(ℰ'), and ζ(ℰ',Q)=ζ(ℰ,Q)/2. The per-platform normalizations are justified only by the intended channel family; the paper should prove a canonical scaling rule or restate the main theorems for normalized pairs.","section":"Definition 3, Remark 2"},{"comment":"The advertised 'adversarial error-set model' guarantees per-channel recoverability: for each N∈N(ℰ) there exists a recovery map, possibly depending on N. Exact QEC's error-set model provides a single decoder that corrects all channels in Span(ℰ). Universal decoders are listed as an open problem in §1.3, so this is acknowledged, but the abstract and Section 1 should not claim the full adversarial model without this caveat. The technical results are unaffected, but the framing overstates the analogy.","section":"Definitions 2–3, §1.3"}],"minor_comments":[{"comment":"The supplied full text truncates in Appendix A.2 at Eq. (96). If this is the complete manuscript, the omitted appendices are essential: the Section 5 rate constraints use Lemmas 27–31 and 33–36. Please include them.","section":"Appendix A"},{"comment":"The definition of the uniform erasure channel contains an unused parameter `α∈C`; remove it.","section":"Proposition 13"},{"comment":"The `maximum t` formulation assumes the family E is nested. State this assumption explicitly, or define the distance via thresholding.","section":"Definition 5"},{"comment":"The repeated claim of 'first known asymptotically good code families' needs a precise comparison with the existing literature on quantum deletion codes and Majorana codes to substantiate 'first known'.","section":"Section 5"},{"comment":"After Eq. (24), `ℬ` is used without subscript for `ℬ^ℰ_{λ,Q}`; define it for readability.","section":"Theorem 3 proof"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about normalization-dependent N(ℰ) lands: the central theorem is a property of (ℰ,scaling), not of Span(ℰ). I recommend the authors either prove a canonical scaling principle or explicitly frame the theory as applying to normalized error-set pairs. The rest of the technical core appears sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis is a serious paper and, conditional on the long proofs holding up, the AQEC error-set model it builds is the structural backbone the field has been missing. The core move — replacing full Knill-Laflamme linearity with ℰ-controlled channel families defined by a spectral constraint on mixing coefficients — works, and the sufficient conditions are derived from Bény–Oreshkov rather than assumed. The environment-leakage distance, the Knill–Laflamme Hellinger distance, the approximate erasure/general-error equivalence, and the partition-code machinery with the metric–error hierarchy are all genuinely new. I checked the steps the reader flagged: the norm chain in Theorem 3, the Hellinger projection identity, the completeness identities; they are consistent. This is not a circular construction.\n\nThe biggest soft spot is the normalization dependence. Remark 2 concedes it, and the stress-test example lands: on C^2, depolarizing is ℰ-controlled for ℰ={Π0,Π1} but not for ℰ'={Π0/√2,Π1/√2}, and ζ scales with the error operators. The uniform guarantee is therefore a property of (ℰ, scaling), not of the error set alone. For Pauli and Majorana errors the unitary normalization is canonical, and for photon loss the binomial normalization is sensible, but the analogy with exact QEC is weaker than the abstract suggests. The paper should either prove invariance beyond linear independence or state the normalization rule as an explicit modeling choice up front. This is a conceptual caveat, not a fatal flaw.\n\nThe other soft spots are proportionate. I could not verify the long appendix proofs — Theorem 14 and Proposition 5 in particular — nor the imported classical bounds (Chvátal–Sankoff, [48], Tverberg). The \"first known\" claims for deletion, Majorana, and Rydberg families need an explicit prior-art audit; the deletion literature is cited but not systematically compared. The constants in Theorem 15 grow exponentially in the linear-distance regime; the paper shows the resulting loss is polynomial in code dimension, which keeps the reduction meaningful, but it is coarse.\n\nThis is for AQEC researchers and code theorists. It deserves a serious referee. I would send it out and ask for the normalization issue to be addressed and the \"first\" claims audited. The central argument is intact.","headline":"A serious AQEC error-set theory with a real but surmountable normalization caveat; referee it, but demand a scaling rule and a prior-art audit.","tokens_in":66818,"tokens_out":5042,"would_cite":true,"duration_ms":47804,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P70","81P45","94B65"],"pacs":["03.67.Pp"],"model":"deepseek-v4-flash","headline":"Approximate quantum error correction has a working error-set model, with a single distance-like parameter controlling every channel generated by an error set.","keywords":["approximate quantum error correction","error-set model","environment-leakage distance","approximate code distance","erasures versus general errors","partition codes","asymptotically good quantum codes","deletion errors"],"falsifier":"For a fixed platform expressing its noise channel in the paper's chosen normalized error operators, compute the spectral norm of the coefficient matrix of the channel's Kraus representation; if the norm exceeds one for the natural normalization, that channel lies in the linear span of the error set but outside the controlled family, and the uniform guarantee does not cover it.","tokens_in":65576,"feed_emoji":"⚛️","tokens_out":7750,"duration_ms":81194,"temperature":0.7,"pith_summary":"The paper sets out to disprove the long-held view that approximate quantum error correction has no structural theory: no linearity, no distance, no erasure-versus-error equivalence. It argues that a restricted version of the exact theory's linearity does survive. Fix an error set; a channel is 'controlled' by it if its Kraus operators are linear combinations of the set with a coefficient matrix of spectral norm at most one. The paper proves that a single code parameter—the environment-leakage distance—controls approximate correction uniformly for every controlled channel, and a companion Hellinger distance does the same for the average-case channel-fidelity criterion. On this basis it defines approximate code distance, proves an approximate erasure/general-error equivalence, and constructs partition-based code families that it claims are the first asymptotically good quantum codes for deletions, Majorana fermions, and one-dimensional Rydberg-blockaded chains.","feed_headline":"Approximate QEC gets a distance and a worst-case error-set model","feed_subtitle":"The paper restores linearity, erasure-vs-error equivalence, and the first asymptotically good deletion and Majorana code families.","key_machinery":"The environment-leakage distance is the infimum over constant matrices lambda of the diamond norm of the superoperator that sends rho to the sum over k,l of tr((P E_k-dagger E_l P minus lambda_{k,l} P) rho) on |k><l|, where P is the code projector; it measures how far the error set's action on the code is from exact-correctability form. The companion average-case object is the Hellinger distance from the error-correlation matrix to the subspace of matrices I-tensor-lambda, which the paper proves contracts under conjugation by the channel's coefficient matrix—this contraction is what converts a single-error-set condition into a uniform guarantee over the whole controlled family. The partition","core_discovery":"The central claim is that if the environment-leakage distance of a code is at most epsilon-squared over two, then the code approximately corrects every channel whose Kraus operators are linear combinations of the error set with a coefficient matrix of spectral norm at most one. The mechanism is restricted linearity: mixing error operators by a contraction cannot increase the deviation from the scalar form that exact correction requires, just as in exact QEC any channel in the linear span of a correctable error set is automatically correctable. The same uniform-control idea extends to the average-case criterion, where the relevant quantity is the Hellinger distance from the error-correlation","pith_inferences":["The normalization of the error set is a variational modeling choice, so a natural extension is to optimize that scaling to maximize the physically relevant channel family covered by a given code.","The Hellinger criterion is optimization-free and directly computable from the error-correlation matrix, so it could serve as a cheap objective for numerical code search, with the worst-case distance checked afterward.","The erasure-to-general reduction suggests a practical benchmarking protocol: measure erasure fidelity at twice the target error weight and translate it into bounded-weight guarantees, provided the erasure error decays faster than the threshold the paper computes.","If a platform's natural noise channel fails the spectral-norm condition under the chosen normalization, the uniform guarantee silently misses that platform's principal noise, so the model's reach is bounded by where controlled families are physically faithful."],"forward_implications":["Any code with environment-leakage distance at most epsilon-squared over two simultaneously protects against every controlled channel built from the same error set, so code design no longer needs to enumerate channels one by one.","Approximate code distance is now defined for arbitrary indexed error families, and the erasure-to-general-error equivalence lets erasure tests certify bounded-weight-error correction up to stated constants.","Subsystem variance, a quantity already linked to circuit complexity, now quantitatively controls approximate correction against erasures and, through the equivalence, against general errors.","The partition construction yields asymptotically good rate-distance families for deletion errors, Majorana fermion systems, and one-dimensional Rydberg-blockaded chains, and for amplitude damping it beats the nondegenerate Hamming bound in some parameter regimes.","Random partition codes inherit typicality properties such as balancedness and bounded per-mode occupancy from the sampling distribution, making the construction compatible with additional physical constraints."],"fun_headline_variants":["Approximate QEC gains a distance and error-set model","New distances unify approximate and exact QEC","Error-set model brings exact QEC structure to approximate","First asymptotically good codes for fermions and deletions","Restricted linearity extends QEC structure to approximate regime"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire uniform guarantee rests on the claim that the physically relevant noise channels admit a Kraus representation whose coefficients in the chosen error set form a matrix of spectral norm at most one—a property that, as the paper concedes, changes when the error operators are rescaled.","fun_headline_variants_meta":{"raw":{"variants":["Approximate QEC gains a distance and error-set model","New distances unify approximate and exact QEC","Error-set model brings exact QEC structure to approximate","First asymptotically good codes for fermions and deletions","Restricted linearity extends QEC structure to approximate regime"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000705,"raw_usage":{"total_tokens":3067,"prompt_tokens":850,"completion_tokens":2217,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":594,"completion_tokens_details":{"reasoning_tokens":2140}},"tokens_in":594,"tokens_out":2217,"duration_ms":13917,"temperature":1.0,"reasoning_tokens":2140,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:56:47.307443+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed platform expressing its noise channel in the paper's chosen normalized error operators, compute the spectral norm of the coefficient matrix of the channel's Kraus representation; if the norm exceeds one for the natural normalization, that channel lies in the linear span of the error set but outside the controlled family, and the uniform guarantee does not cover it.","supporting_citations":[],"review_version":1}