{"id":"069b4f62-3715-467e-b791-611cd1c86073","arxiv_id":"2607.23450","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Above the coherent information of its own input, any stabilizer code over a product Pauli channel has entanglement fidelity decaying exponentially in block length.","lead":"Stabilizer quantum codes that try to send information faster than their own coherent information suffer exponentially collapsing fidelity over Pauli noise. This settles the strong-converse question inside that code class and shows constant tolerated error buys no rate.","discovery_kind":"extension","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"No significant objection identified beyond the reader's flagged full-sector assumption. The linchpin is Proposition 3.1's exact fidelity identity (resting on the sum rule (26)); close reading of the proof chain found no internal gap, so the concern is scope, not soundness — and the paper itself prov","rationale":"The reader's weakest_assumption — the full-sector requirement making branch fidelities binary and optimal fidelity equal to a single event mass — is indeed the load-bearing point, and my independent tracing of the argument confirms both that it is essential and that the paper handles it honestly (Remark 3.2 locates exactly where it enters; Proposition 7.2 proves the decoder-side route cannot remove it; Problem 7.3 correctly isolates the needed encoder-side statement). I explicitly checked the four technical pillars for the places such proofs usually break: (1) the sum rule derivation uses the twirl (16) plus trace preservation and (18), and the normalization D·(1/D)tr(D_s(Π_0))=tr Π_0/D=1 is right; (2) the McDiarmid mean bound in Lemma 4.1 correctly uses the lower tail at t=Ef; (3) Lemma 5.2's chain rule telescopes exactly and classicality of the flag is what licenses H(G|RB^n)≥0 — the footnote correctly notes a quantum flag would only give −log|G|; (4) Lemma 5.3's twirl argument correctly uses concavity and local-unitary invariance of conditional entropy, and the isotropic spectrum computation gives the linear-in-error continuity that the small-γ regime needs. The assembly in Theorem 6.5 (choice δ=δ_γ, g(δ_γ)=γ/2, the two threshold conditions on n) is arithmetically consistent. The antidegradable corollary's use of purity duality I(R⟩E)=−I(R⟩B) is standard and correct. Disagreement with consensus is not at issue here — the result is new but structurally classical in spirit, and its non-PPT character is argued rather than assumed (Remark 8.2). What remains is process risk: an AI-authored long proof verified by one expert. That justifies the reader's MODERATE confidence and motivates the concrete numerical check of Proposition 3.1, which is cheap, fully decidable at small block length, and would falsify the linchpin if the sign/normalization conventions harbored an error. It does not justify moving the verdict off ACCEPT.","tokens_in":30069,"tokens_out":6719,"duration_ms":46606,"concrete_test":"Numerically verify the linchpin Proposition 3.1 independently: for the [[5,1,3]] code under depolarizing noise (e.g., p=0.3) and a second instance with nontrivial stabilizer phases (the paper's own n=2 example with X⊗X, Z⊗Z generators), enumerate the 4^n error patterns, build the syndrome/logical-class array, and compute μ(A_ML)=Σ_s p_s max_ℓ q(ℓ|s) exactly. Independently, compute the true optimal entanglement fidelity over all decoders via the standard fidelity SDP (maximize ⟨Φ|(id⊗D∘N∘V)(Φ)|Φ⟩ over CPTP D). The claim requires exact agreement; a discrepancy at any p would invalidate the sum rule (26) and collapse the whole converse. Additionally re-derive the twirl identity (16) symbolically for the [[4,1,2]] code to confirm the sign conventions in (14)–(15) leave the sector projectors correct.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the central claim (Theorem 6.5 + Corollaries 6.6/6.7) as resting on four steps: (i) Proposition 3.1 — optimal fidelity equals the mass of the ML event, via the sum rule Σ_ℓ f_{s,ℓ}=1, which follows from the twirl identity (16) and the completeness of the logical Pauli operators as an orthogonal basis of B(C); (ii) the blowing-up lemma 4.1 via McDiarmid, including the mean bound Ef ≤ √((n/2)ln(1/c)) from the lower tail at t=Ef; (iii) the flagged-channel accounting in Lemma 5.2, where the chain rule I(R⟩B^nG)=I(R⟩B^n)+I(R:G|B^n) is exact and the bound I(R:G|B^n)≤H(G)≤log|G| uses only classicality of G for H(G|RB^n)≥0; (iv) the one-shot converse Lemma 5.3 via twirling to an isotropic state, with concavity/invariance of conditional entropy giving I(R⟩L)_σ ≥ I(R⟩L)_σ̃ and data processing for coherent information. Each step checks out line by line, including the edge cases (D=1, the ε>1−D^{−2} trivial branch in Theorem 6.1, the ordered-product phase repair (14) in the stabilizer group, and the Fekete step Q^(n)_stab ≤ n Q_stab in Corollary 6.7). The genie-flag construction is legitimate converse technique since Lemma 5.3 holds for arbitrary channels.\n\nThe genuinely load-bearing assumption is the one the reader identified: the sum rule (26), hence the binary branch-fidelity structure, requires the code space to fill its entire 2^k-dimensional sector (Remark 3.2). For a proper subspace the residual noise is a logical Pauli channel on an arbitrary subspace and the identity F⋆=μ(A_ML) fails — and Proposition 7.2 shows this is not an artifact of the method but a real boundary. However, this is a hypothesis of the theorem, not a hidden assumption: the central claim is about the stabilizer class, the delimitation is explicit, and the escape route is isolated as Problem 7.3. I therefore find no internal inconsistency. The residual risk is not a specific soft spot but the absence of independent verification of a long, AI-authored proof — a correctness-process risk the reade","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The paper proves a strong converse for quantum communication over arbitrary products of (not necessarily identical) qubit Pauli channels, restricted to the class of stabilizer coding schemes whose code space is a full joint eigenspace of the stabilizer group, with arbitrary isometric encoder and arbitrary decoder. The central result (Theorem 6.5) is that any such (n,k) scheme with k ≥ I_c(V) + γn has entanglement fidelity F ≤ exp(−nE(γ)) with E(γ) = (1/2)[g^{−1}(γ/2)]² > 0, beyond an explicit threshold n_0(γ). The proof rests on: (i) an exact identity (Proposition 3.1) between optimal fidelity and the measure of the maximum-likelihood event A_ML in the product error space, via the sum rule Σ_ℓ f_{s,ℓ} = 1 (Eq. (26)), which in turn rests on the completeness of the logical Pauli basis on a full sector (Lemma 2.1, twirl identity (16)); (ii) a finite-blocklength blowing-up lemma (Lemma 4.1) derived from McDiarmid's inequality; (iii) a flagged-channel construction whose classical flag is charged at its entropy against coherent information (Lemmas 5.1–5.2); and (iv) a one-shot converse via twirling to an isotropic state (Lemma 5.3). Corollaries give Q_ε = Q_stab for all ε ∈ (0,1) within the class (6.7), zero capacity for antidegradable factors including depolarizing p ∈ [1/4,3/4] (6.6), and extensions to mixtures, orthogonal superpositions, and perturbations (6.9–6.11). Section 7 proves the full-sector assumption cannot be removed decoder-side (Proposition 7.2) and isolates the enc","tokens_in":30514,"tokens_out":5215,"duration_ms":56284,"significance":"If correct, this is the first strong converse with an explicit exponent for quantum communication over Pauli channels in any nontrivial class, and — as the paper demonstrates in Remark 8.2 — it certifies a strong converse in the antidegradable window p ∈ [1/4,1/2) where all PPT/Rains-type relaxations provably cannot, even after regularization. Particular strengths: the argument is fully explicit and finite-blocklength (thresholds n_0(γ), n_1(γ) given in closed form); it requires no additivity assumptions and no semidefinite relaxations; it covers non-memoryless product noise; the master inequality (Theorem 6.1) and Theorem 6.2 are stated for arbitrary encoders, cleanly separating what is general from what is stabilizer-specific; and Proposition 7.2 constructively proves the scope restriction is intrinsic to the method rather than an artifact, converting a potential objection into a sharp open problem (7.3) with a stated consequence (the full strong converse for memoryless Pauli channels). The qudit extension (Appendix A) is carried through with careful attention to where the F_d-linear structure is used. I verified the load-bearing chain line by line — the sum rule (26) via (16) an","major_comments":[],"minor_comments":[{"comment":"Typo in the proof of Proposition 3.1: \"The conditional fidelity oneis therefore 1\" should read \"one is\".","section":"§3, proof of Prop. 3.1"},{"comment":"Notation for the syndrome map alternates between \"σsyn\" and \"σ syn\" (e.g., Eq. (12) vs. §2.3 text and Remark 3.2). Please uniformize.","section":"§2.3, Eq. (12) and passim"},{"comment":"In Corollary 6.7 the identification Q_ε = Q_stab uses Hamada's achievability [43] for the lower bound; it would help the reader to state explicitly that Hamada's concatenated stabilizer codes satisfy the full-sector convention of Definition 2.3 (they do), so that the classes on the two sides of (61) literally coincide.","section":"Corollary 6.7"},{"comment":"In Eq. (64) the double-bar notation for the ℓ² norm over Kraus indices is nonstandard and the first inequality (triangle inequality applied to the vector of amplitudes ⟨Φ|(I⊗M_a)ψ_e⟩ indexed by a) deserves one clause of explanation.","section":"Corollary 6.10, Eq. (64)"},{"comment":"Several references are missing terminal periods ([15], [16], [17], [34], [35], [39], [44], [46], [51], [55], [57]); a uniform pass over the bibliography is warranted.","section":"References"},{"comment":"The constant C(γ) = e^{n_0(γ)E(γ)} in Theorem 6.5 grows very rapidly as γ ↓ 0; a brief numerical remark (e.g., for the depolarizing channel at a representative γ) on where the bound becomes nontrivial would help readers gauge its finite-blocklength content, complementing the asymptotic statement.","section":"Theorem 6.5"},{"comment":"In Remark 8.2 the step \"R(ρ_p^{⊗n}) ≥ n E_D(ρ_p)\" uses superadditivity of distillable entanglement under tensoring; this is standard but currently implicit, and one line citing it would make the regularization argument self-contained.","section":"Remark 8.2"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript contains an unusually detailed disclosure that the proofs, writing, and literature search were carried out by large language models under the author's direction, with the author asserting full verification and responsibility. I have assessed the mathematics on its merits and found it sound; whether this mode of production is compatible with the journal's authorship policies is an editorial question I leave to you, but it should be settled before acceptance rather than after. Scientifically, I note the result is a strong converse for a restricted code class; the paper is commendably upfront about this, but the editor may wish to ensure the abstract wording (\\\"stabilizer codes\\\" with the full-sector qualification) survives into any publicity, since casual readers could over-read it as settling the Pauli-channel strong converse in general."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing to know: for full-sector stabilizer codes over any product of qubit Pauli channels, rate above the code’s own coherent information forces entanglement fidelity to decay as exp(−n E(γ)), with E explicit. That pins the ε-capacity of the class for every ε < 1, and sets it to zero for antidegradable factors (including depolarizing p ∈ [1/4, 3/4]). On p ∈ [1/4, 1/2) this is strictly stronger than anything Rains/PPT can give, which the paper correctly notes stay positive after regularization.\n\nWhat is actually new is the strong (not weak) converse at finite n, the drop of memorylessness, and the exact identity that optimal fidelity equals the mass of the ML event for an arbitrary isometry onto the code space. Hamada already had Q_stab and the vanishing-error converse; the load-bearing step here is Proposition 3.1 plus classical blowing-up on that event, with the flag charged at its entropy. The chain (sector structure → sum rule → McDiarmid → flagged one-shot converse) reads cleanly; I did not find an internal gap.\n\nSoft spots are mostly scope, and the paper owns them. Full-sector is essential: branch fidelities become binary only then, and Section 7 shows decoder-side sharpening cannot remove it (constant-fidelity schemes whose every near-deterministic core is exponentially light). Two-way assistance is excluded for cause. The general-code extension is isolated as an encoder-side extraction problem rather than papered over. The exponent is not optimized, and whether Q_stab equals Q remains open. AI-assisted authorship under Tomamichel’s check is disclosed; that is a process residual on a long proof, not a mathematical red flag on the page.\n\nThis is for people who care about quantum capacity converses and finite-blocklength structure. The math is pure and checkable; citations are in the right places. I would bring it to reading group, cite the stabilizer strong converse and the zero-capacity corollary, and send it to referees.","headline":"Real strong converse inside the stabilizer class for product Pauli channels, with an explicit exponent and a clean win over PPT/Rains on antidegradable depolarizing noise.","tokens_in":31253,"tokens_out":535,"would_cite":true,"duration_ms":18591,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"For stabilizer codes over Pauli channels, sending above the code's own coherent information forces entanglement fidelity to decay exponentially, so constant tolerated error buys no rate.","keywords":["strong converse","quantum capacity","stabilizer codes","Pauli channels","blowing-up lemma","coherent information","entanglement fidelity","antidegradable channels"],"falsifier":"Exhibit a sequence of full-sector stabilizer codes over a fixed Pauli channel whose rate stays a fixed γ above the code's coherent information while entanglement fidelity remains bounded below by a positive constant independent of block length.","tokens_in":30744,"feed_emoji":"📉","tokens_out":1018,"duration_ms":24768,"temperature":0.7,"pith_summary":"The paper proves a strong converse for quantum communication when both the noise is a product of Pauli channels and the codes are stabilizer codes whose code space fills a full joint eigenspace. If such a code uses more logical qubits than the coherent information of its own input state, its entanglement fidelity falls exponentially in the block length, for any isometric encoder onto that space and any decoder. For memoryless channels this pins the ε-capacity of the stabilizer class to the vanishing-error stabilizer rate for every fixed error below 1; for antidegradable factors, including the depolarizing channel at error probability from 1/4 to 3/4, that capacity is zero. The argument never needs additivity or PPT relaxations: optimal decoding succeeds exactly on one event in a product error space, so the classical blowing-up lemma applies and the cheap side information it produces is charged against the coherent information. The same machinery already rules out heavy near-deterministic cores for arbitrary encoders above the coherent information, and isolates the encoder-side extraction step that would extend the strong converse to all codes.","feed_headline":"Stabilizer fidelity collapses above coherent information","feed_subtitle":"Constant tolerated error buys no rate for stabilizer codes on Pauli noise","key_machinery":"Exact fidelity formula for stabilizer codes: optimal entanglement fidelity equals the measure of the maximum-likelihood event A_ML in the product Pauli-error space. That identity turns decoding success into a classical set to which the blowing-up lemma applies; the resulting low-weight flag is then charged at its entropy against the coherent information via a one-shot converse on the flagged channel.","core_discovery":"Within the class of full-sector stabilizer codes over product Pauli channels, any scheme whose rate exceeds the coherent information of its own input state has entanglement fidelity decaying as exp(−n E(γ)), where γ is the overshoot rate and E(γ) = (1/2)[g^{-1}(γ/2)]² > 0. Consequently the ε-quantum capacity of the class equals the vanishing-error stabilizer rate for every ε ∈ (0,1), and is zero whenever every factor is antidegradable.","pith_inferences":["If the open encoder-side extraction statement (Problem 7.3) holds, the strong converse would extend from stabilizer codes to all isometrically encoded schemes over memoryless Pauli channels.","The erasure channel is the natural next target: its classical pattern is handed to the receiver, so branch fidelities are channel-side and no-cloning may force the sharpness that stabilizer structure supplies here.","Small-blocklength numerical searches for coherently hedging codes that keep constant fidelity above the n-letter coherent information would directly test whether fidelity can survive without heavy cores."],"forward_implications":["Among stabilizer codes, tolerating any fixed constant error does not raise the achievable quantum rate above the vanishing-error stabilizer rate.","For every antidegradable Pauli factor (including depolarizing noise with p ∈ [1/4, 3/4]) the stabilizer ε-capacity is zero for all ε < 1, with an explicit exponential fidelity bound.","Partial-transposition / Rains-type bounds cannot certify this zero-rate strong converse for depolarizing noise on p ∈ [1/4, 1/2), so the result is strictly stronger than PPT methods inside the stabilizer class.","The same exponential bound extends to shared-randomness mixtures, coherent superpositions of subexponentially many orthogonal stabilizer codes, and exponentially small encoder perturbations.","No encoder of any kind can carry a near-deterministic core of non-negligible mass once its rate exceeds its own coherent information."],"fun_headline_variants":["Stabilizer fidelity decays exponentially above coherent information","Excess rate forces exp collapse of stabilizer entanglement fidelity","Strong converse: stabilizer codes gain no rate from constant error","Blowing-up lemma pins stabilizer capacity to coherent information","Pauli stabilizer ε-capacity vanishes for every constant error ε"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The code space must fill an entire joint eigenspace of the stabilizer group; only then are branch fidelities binary and optimal fidelity exactly the mass of one event in the error space.","fun_headline_variants_meta":{"raw":{"variants":["Stabilizer fidelity decays exponentially above coherent information","Excess rate forces exp collapse of stabilizer entanglement fidelity","Strong converse: stabilizer codes gain no rate from constant error","Blowing-up lemma pins stabilizer capacity to coherent information","Pauli stabilizer ε-capacity vanishes for every constant error ε"]},"model":"grok-4.5","effort":"low","cost_usd":0.003847,"raw_usage":{"total_tokens":1251,"prompt_tokens":806,"num_sources_used":0,"completion_tokens":60,"cost_in_usd_ticks":38468000,"prompt_tokens_details":{"text_tokens":806,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":385,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":806,"tokens_out":60,"duration_ms":7791,"temperature":1.0,"reasoning_tokens":385,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T21:48:39.095769+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Exhibit a sequence of full-sector stabilizer codes over a fixed Pauli channel whose rate stays a fixed γ above the code's coherent information while entanglement fidelity remains bounded below by a positive constant independent of block length.","supporting_citations":[],"review_version":1}