{"id":"0eae8b33-901a-455d-bfa0-f99ac1463653","arxiv_id":"2608.09347","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A single theorem involving the pretty good measurement rederives all four binary code bounds and yields new channels that strictly improve the MRRW bounds for every relative distance in (0,1/2).","lead":"The paper proves that many classic rate-distance bounds for binary codes are consequences of a single quantum-information criterion, and it uses new quantum channels to strictly improve the two McEliece-Rodemich-Rumsey-Welch bounds. It introduces the pretty good criterion: if the error of a quantum measurement on a channel is below the code's distance fraction, then the code's rate is capped by the channel capacity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the proof hinges on Lemma 10's PGM data-processing inequality, but the Petz-based argument is sound.","rationale":"The reader's weakest-assumption identification is correct: Lemma 10 is the most delicate structural step, since it converts bit-level PGM error into block-level expected Hamming distance. I read the appendix proof in detail and found the Petz-recovery argument coherent. The PGM success is a sum of weighted Gram norms, and the Petz–Gram contraction gives exactly the needed monotonicity; the support restrictions are handled by defining inverses on the support of the average state. I also checked the surrounding proof of Theorem 1: Lemma 9 is a direct corollary of PGM coarse-graining, Lemma 11 follows from the auxiliary classical register and data processing, and Theorem 7 is a standard cq strong converse with the output-symmetry assumption used correctly. The derivations of the four classical bounds from BEC, BSC, PSC, and mPSC are consistent, and the strict-improvement arguments for MQC and 2MQC rely on valid local perturbations rather than the heuristic low-rate expansion. I therefore found no live technical objection. The absence of shipped code and the paper's own heuristic remarks in Section 6.2 reduce certainty but are not grounds for changing the verdict.","tokens_in":70779,"tokens_out":22509,"duration_ms":238423,"concrete_test":"Randomized numerical audit of Lemma 10: generate 100 random ensembles of 3–5 states in dimensions d=2,3,4 and 100 random CPTP maps via Stinespring isometries; compute PGM success probabilities exactly (matrix square-root and inverse on support) and verify p_s(original) ≥ p_s(post-processed). Also test the binary composition identity (18) at 100 random (α,η,σ0,σ1). Any violation would be a counterexample to Theorem 1; a clean pass over a broad sweep would corroborate the appendix proof without replacing it.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The pretty good criterion (Theorem 1) collapses if Lemma 10 fails: Theorem 17's expectation bound E[d(c,ĉ)] ≤ n·p_e(σ0,σ1) is exactly the bit-to-block step, and without it Corollary 18's constant success probability and the strong-converse application both disappear. I focused on Appendix A.2. The proof writes PGM success as Σ_x ⟨A_x,A_x⟩_{σ̄} with A_x = π_x σ_x, and compares with the post-processed ensemble using the Petz–Gram contraction Fact 32(iii), which gives 0 ≤ ⟨A,K(A)⟩ ≤ ⟨A,A⟩ for K = Φ^♯∘Φ. The contraction is justified by self-adjointness of K in the weighted inner product plus eigenvalues in [0,1], shown via Stinespring and unitality of K*; I see no missing support, normalization, or positivity condition. The other components of Theorem 1 (Lemma 9 coarse-graining, Lemma 11 uniform-prior maximization, Theorem 7 strong converse) are standard and internally consistent. The strict improvements in Props 25 and 26 use legitimate local perturbations around endpoints and do not depend on the heuristic Section 6.2 expansion. The only caveat is that no machine-checked or independent code is provided, but this is a verification gap, not an identified error.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a general sufficient condition, called the \"pretty good criterion\" (Theorem 1), for upper-bounding the asymptotic rate of binary codes with relative distance δ. The criterion states that if a binary-input output-symmetric classical-quantum channel has PGM bit error rate below δ, then every length-n code of relative distance δ has rate at most the channel's Holevo information plus O(n^{-1/2}). The authors prove this by deriving an expected-Hamming-distance bound from bitwise PGM properties, converting it into a constant block-decoding success probability, and then invoking a cq-channel strong converse. They show that the BEC, BSC, pure-state channel, and masked pure-state channel rederive the Plotkin, Elias-Bassalygo, first MRRW, and second MRRW bounds, respectively. They then introduce two new channel families, the mixed-qubit channel (MQC) and the masked mixed-qubit channel (2MQC), and prove strict asymptotic improvements over the first and second MRRW bounds for every δ in (0,1/2). The paper also sketches q-ary extensions and gives an LDPC-code refinement.","tokens_in":70967,"tokens_out":38680,"duration_ms":456348,"significance":"If correct, this is a substantial contribution: it provides the first asymptotic improvements over the long-standing MRRW bounds, and it unifies four classical rate-distance bounds as consequences of one channel-coding criterion. The main proof is structurally sound: Theorem 1 rests on the established cq strong converse and on self-contained appendix proofs for PGM coarse-graining, the PGM data-processing inequality via the Petz map, and the uniform-prior maximization lemma. The strict-improvement arguments in Propositions 25 and 26 are analytic and cover every δ in (0,1/2), not just numerically verified points. The paper is also transparent about what is rigorous and what is heuristic: the low-rate expansion in Section 6.2 is explicitly labeled as numerical/heuristic, and Table 2 is described as floating-point estimates rather than certified optima. The numerical tables are therefore not load-bearing for the main claims. The q-ary and LDPC sections broaden the framework and provide a promising research direction, although they are less developed than the binary results.","major_comments":[],"minor_comments":[{"comment":"In the proof of Theorem 7, the equality -Tr(σ0 log \\bar σ)=S(\\bar σ) is asserted without explanation; it follows from the output-symmetry relations Uσ0U†=σ1 and U\\bar σU†=\\bar σ, and this step should be spelled out for the reader.","section":"Appendix A.4"},{"comment":"The q-MC row of Table 3 and the associated spectral formulas are stated without derivations; since Section 7 is explicitly a blueprint, please mark these entries clearly as unproven sketches so that they are not mistaken for theorems.","section":"Section 7"},{"comment":"The low-rate expansion is presented in the main text but is heuristic; consider moving it to an appendix or adding a stronger disclaimer that it is not part of the proof of Proposition 25.","section":"Section 6.2"},{"comment":"The references to specific GPT models and to OpenAI's concurrent work appear in the mathematical narrative; these statements might be better placed in the acknowledgments or a separate remarks section to keep the technical exposition neutral.","section":"Sections 1.7 and 1.9"},{"comment":"The factorization in Eq. (34) is correct, but writing it as p_e = 1/2[1-(1-2η)^2(1-2r)] would match the form of Eq. (19) and remove a small notational hurdle for the reader.","section":"Eq. (34)"},{"comment":"The footnote correctly observes that the output-symmetry lift preserves p_e and χ, but it should explicitly note that the O(n^{-1/2}) constant in Theorem 1 may change under the lift while the asymptotic rate bound is unchanged.","section":"Section 1.1, footnote 2"}],"recommendation":"accept","confidential_remarks":"I see no circularity or parameter-fitting concern: Theorem 1 is proved from independent strong-converse and PGM lemmas, and the channel parameters in the MQC/2MQC constructions are optimization variables rather than fitted quantities. The numerical tables are not certified, but the strictness claims are proven analytically, so the tables are only illustrative. The manuscript is well within the scope of the journal. The AI-use and concurrent-work sections are transparent and do not affect the mathematical assessment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is the real thing. The pretty good criterion (Theorem 1) is a single, clean statement: if the PGM bit error rate of a binary-input output-symmetric cq channel is below δ, then every binary code of relative distance δ has rate at most the channel capacity, up to O(n^{-1/2}). From that one theorem they recover Plotkin, Elias–Bassalygo, and both MRRW bounds, and then go further: the MQC and 2MQC channels give strict asymptotic improvements over the first and second MRRW bounds for every δ in (0,1/2). That is the first improvement over MRRW in roughly five decades, if it holds.\n\nWhat the paper does well: the proof is coherent and unusually self-contained. The expected Hamming distance bound (Theorem 17) is the load-bearing step, and it rests on the PGM data-processing inequality (Lemma 10). The appendix proof via the Petz map is sound; I checked the weighted-inner-product contraction argument and found no gap. The strict improvement arguments are genuinely analytic, not just numerical: local perturbations around endpoints give inequalities valid for every δ, not sampled points. The rederivations of the four classical bounds are elegant and useful in themselves.\n\nSoft spots, in proportion: there is no shipped code or machine-checked proof, and the numerical tables are floating-point estimates, not certified optima. That is a verification gap, not an identified error. The formal expansion in Section 6.2 is explicitly heuristic and does not carry the theorems. The concurrent-work note about OpenAI and the AI-use statement are unusual but transparent; the mathematics is proved in the paper, so I do not treat those as flaws. The citation pattern is fair; self-citations appear mainly in narrative sections and are not load-bearing.\n\nThe main risk is that long, intricate proofs can hide a sign error or a missing case, and the payoff is large enough that independent verification is warranted. But on reading, the central claim is well-supported and the weakest structural step is proved, not assumed.\n\nWho this is for: coding theorists, quantum information theorists, and anyone tracking the rate-distance problem. It deserves a serious referee. My recommendation is to accept subject to careful verification of Lemma 10 and the perturbation analyses, and to send it to a referee comfortable with both quantum information and classical coding bounds.","headline":"A genuine breakthrough: one quantum criterion recovers all four classic binary rate bounds and strictly improves both MRRW bounds, and the core proof holds up on inspection.","tokens_in":71544,"tokens_out":1373,"would_cite":true,"duration_ms":18785,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94B65","81P45","94A17"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a single \"pretty good criterion\" that recovers all four classical binary code rate bounds and yields strict improvements over both MRRW bounds.","keywords":["binary codes","rate-distance bounds","pretty good measurement","classical-quantum channels","MRRW bounds","Holevo information","posterior sampling","channel masking"],"falsifier":"Evaluate the variational formulas for $R_{\\mathrm{MQC}}(\\delta)$ and $R_{\\mathrm{2MQC}}(\\delta)$ at a fixed distance such as $\\delta=1/4$; if either exceeds the corresponding MRRW value, Theorems 2 and 3 are false, and more directly, any binary code of relative distance $\\delta$ whose rate exceeds $\\chi(\\sigma_0,\\sigma_1)$ for a channel with $p_e(\\sigma_0,\\sigma_1)<\\delta$ would refute the pretty good criterion itself.","tokens_in":70528,"feed_emoji":"🧮","tokens_out":9087,"duration_ms":94659,"temperature":0.7,"pith_summary":"The paper proposes a single information-theoretic criterion for upper-bounding the rate of binary error-correcting codes. It says that if a binary-input output-symmetric classical-quantum channel has pretty-good-measurement bit error rate below $\\delta$, then every length-$n$ binary code of relative distance $\\delta$ has rate at most the channel's Holevo capacity, up to an $O(n^{-1/2})$ correction. Asymptotically this gives $R_2(\\delta)\\le \\chi(\\sigma_0,\\sigma_1)$. Choosing four familiar channels rediscovers the Plotkin, Elias-Bassalygo, and both MRRW bounds, while choosing two new mixed-state channels gives an upper bound strictly smaller than each MRRW bound for every $\\delta\\in(0,1/2)$. If correct, these are the first asymptotic improvements over the fifty-year-old MRRW benchmark, and the proof turns a hard combinatorial problem into a channel-design optimization.","feed_headline":"Quantum channels beat the 50-year-old MRRW code bounds","feed_subtitle":"One bit-error criterion rederives all four classic bounds, and two new channels improve the best two.","key_machinery":"The pretty good measurement, the square-root quantum analog of posterior sampling, is the object that carries the argument. Its key structural lemma is PGM data processing: applying any quantum channel to the outputs cannot decrease the PGM bit error rate, a fact proved in the paper with a quantum recovery map. Together with closure under coarse-graining and the fact that the uniform prior maximizes PGM bit error, this yields the expected-Hamming-distance bound $\\mathbb{E}[d(\\hat{c},c)]\\le np_e(\\sigma_0,\\sigma_1)$; the cq channel-coding strong converse then converts the distance-driven constant success probability into the rate bound. The new channels are the mixed-qubit channel MQC, obtained from pure-state outputs by an X-Pauli bit-flip, and its masked version 2MQC, obtained by channel masking.","core_discovery":"The central claim is Theorem 1, the pretty good criterion: fix a binary-input output-symmetric classical-quantum channel with output states $\\sigma_0,\\sigma_1$ and uniform-prior PGM bit error rate $p_e(\\sigma_0,\\sigma_1)<\\delta$; then every binary code $\\mathcal{C}\\subseteq\\{0,1\\}^n$ with minimum distance at least $\\delta n$ satisfies $R(\\mathcal{C})\\le \\chi(\\sigma_0,\\sigma_1)+O(n^{-1/2})$, and asymptotically $R_2(\\delta)\\le\\chi(\\sigma_0,\\sigma_1)$. The derivation goes through the expected Hamming distance between the transmitted codeword and the block PGM output, which is at most $n p_e(\\sigma_0,\\sigma_1)$; the distance assumption turns that bound into constant block-decoding success, and the cq strong converse converts constant success into the capacity upper bound. Instantiating the criterion with the BEC, BSC, pure-state channel, and masked pure-state channel recovers Plotkin, Elias-Bassalygo, and the two MRRW bounds, while the mixed-qubit channel MQC and masked mixed-qubit channel 2MQC strictly improve the first and second MRRW bounds respectively throughout $(0,1/2)$.","pith_inferences":["The strictness proofs use a logarithmic small-argument expansion at one endpoint of the optimization; this suggests a reusable perturbation mechanism for any channel family with a pure-state endpoint, which is my reading rather than a claim the paper makes.","Since the paper notes that MQC covers all output-symmetric qubit channels up to symmetry, the only remaining room for improvement inside this framework must come from four-dimensional or higher outputs; that programmatic conclusion is my inference from their Remark 5.2.","The same criterion is applied to structured code families and to $q$-ary alphabets, so one could try further structured channels for other code families; the paper leaves that search open."],"forward_implications":["For every $\\delta\\in(0,1/2)$, $R_2(\\delta)\\le R_{\\mathrm{MQC}}(\\delta)