{"id":"ccca9b87-c579-4aa2-81f2-71af0b15aeab","arxiv_id":"2607.15151","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Petz–Rényi channel information is strictly superadditive for α in [1/2,1), so joint encodings improve the entanglement-assisted error exponent even though capacity is additive.","lead":"This paper proves that sending over a quantum channel twice with a joint encoder can make errors fall faster, even though the highest reliable rate is unchanged. The effect needs no entangled inputs, only classically correlated ones.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Operational error-exponent claim depends on [20]'s achievability bound; the superadditivity proofs themselves appear sound.","rationale":"I read the appendices carefully. The strict superadditivity proofs are internally consistent. The convexity reduction in B.1 uses Lieb–Ando correctly; the linear response signs in C.42 and D.82 are correct; and the witnesses are genuinely diagonal, classically correlated two-copy states. The one place where the central 'reliability' claim can fail is the external achievability inequality (12) from [20]. The reader already flagged this premise, so my read is largely aligned. I differ from the reader on one point: the unproven tightness of E_r is not needed for the claimed random-coding-exponent enhancement; the achievability upper bound suffices. Since the mathematical theorems stand and [20] is cited as established literature, I would not change the ACCEPT verdict; the proposed check would confirm that Eq. (12) is indeed applicable exactly as used.","tokens_in":23253,"tokens_out":24131,"duration_ms":184667,"concrete_test":"Re-derive [20]'s achievability bound for the two-copy measurement channel M^{⊗2} (d=4, λ=9/25, α=0.8) directly from the operator layer-cake theorem in [20], and check that the exponent is exactly E_r(2R;M^{⊗2}) = max_{1/2≤α<1}(1−α)/α[I_α(M^{⊗2})−2R] with I_α as defined in Eq. (14). If this reproduces Eq. (12) with the same Petz–Rényi information and prefactor 1.5, the operational claim is settled; if it yields a different quantity (e.g., sandwiched information or a different α-range), Eq. (12) must be revised and the reliability claim downgraded.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central mathematical theorems are internally well-supported: the diagonal reductions (C.13–C.16, D.12), the convexity of the reduced one-copy objectives (Lemma C.1, D.2), and the linear-response identities c1(α)=(α−1)(2−α)/α r^2(ξ−θ)^2 (C.42) and dQ/dκ=(α−1)/(α(2−α)) x_α^2 (D.82) have the correct signs, and I found no gap. The load-bearing vulnerability is not inside these proofs but at Eq. (12): ε*(n,R)≤1.5·2^{−E_r(nR;N^{⊗n})}, attributed to [20]. The advertised 'multi-copy enhancement of the random-coding error exponent' is precisely the statement that this achievability bound improves when N is replaced by N^{⊗2}; if [20]'s exponent is not the Petz information I_α of Eq. (14), or is not proved for entanglement-assisted codes, then strict superadditivity of I_α remains true but the operational reliability claim does not follow. Note that the additional 'E_r is tight' expectation cited by the reader is not needed for the named random-coding exponent; only the validity of (12) is.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Petz–Rényi channel information I_α(N) for entanglement-assisted communication. It first proves a convex-optimization reduction for I_α, gives additivity results for special channel classes, and then establishes strict superadditivity I_α(N^{⊗2}) > 2 I_α(N) for every α∈(0,1) for two explicit families: a single-heavy Fourier measurement channel (Theorem 3) and amplitude-damping channels (Theorem 4). The proofs proceed by reducing the one-copy problem to a diagonal single-parameter convex optimization, then perturbing the product of one-copy optimizers along a classically correlated diagonal two-copy path and showing the linear-response coefficient is strictly negative. The paper also provides single-letter upper bounds on the regularized quantity I_α^∞(N). The advertised operational conclusion is that this strict superadditivity yields a genuine multi-copy enhancement of the entanglement-assisted random-coding error exponent for rates below capacity, even though the capacity itself is additive.","tokens_in":23550,"tokens_out":12253,"duration_ms":123106,"significance":"If correct, the mathematical result is significant: it shows that the Petz–Rényi channel information, unlike the sandwiched Rényi information, is not additive in the data-processing range, and that communication reliability can improve under joint use of a channel even when the rate/capacity is strictly additive. A particularly clean feature is that the superadditivity witness is a separable, classically correlated two-copy input, so the effect is not attributed to input entanglement. The analytic proofs are detailed and largely self-contained; the identities (C.42) and (D.82) have the correct signs, and the diagonal reductions and convexity arguments appear sound. The explicit finite-dimensional examples and the numerical certificates strengthen the presentation. The main weakness is not in the superadditivity proof itself but in the operational bridge to the error-exponent claim, which rests on an external achievability bound.","major_comments":[{"comment":"The abstract and Discussion claim a 'genuine multi-copy enhancement of the entanglement-assisted random-coding error exponent.' This conclusion is obtained by combining the new strict superadditivity theorems with the bound ε*(n,R) ≤ 1.5·2^{−E_r(nR;N^{⊗n})} cited to Ref. [20]. The manuscript does not state the hypotheses or exact theorem of [20] that gives this bound. For the operational claim to be established as stated, [20] must apply to entanglement-assisted codes and must define the exponent via exactly the Petz–Rényi information I_α of Eq. (14). Please quote the relevant statement from [20], or at minimum state explicitly which theorem and assumptions are being imported. The mathematical superadditivity theorems stand independently, but the 'error exponent' interpretation is load-bearing for the title and abstract, so this dependency needs to be made verifiable.","section":"Error Exponent and Petz–Rényi Information, Eq. (12)"},{"comment":"Table I and the Discussion state strict non-additivity of I_α for α∈(1,2) as well as for α∈(0,1), and Remark C.4 says the proof of Theorem C.3 'naturally extends' to α∈(1,2). The supplied Appendix C proves Theorem C.3 only for 0<α<1: Lemma C.1, the endpoint sign check (C.26), and the minimizer argument all use 2−α>1 and the perspective convexity valid for α<1; the sign of the linear-response coefficient (C.42) is negative precisely because 0<α<1. Since the claimed extension to α∈(1,2) is not proved and is not needed for the main [1/2,1) result, the authors should either supply the missing proof or restrict the claims in Table I and Discussion. As written, the scope statement overreaches the supplied evidence.","section":"Discussion, Table I, and Remark C.4"},{"comment":"The title 'Convex optimization reduction' is inaccurate for α∈(1,2): Proposition B.1 proves convexity of the relevant map for α∈(0,1) and concavity for α∈(1,2]. Maximizing a concave objective is not a convex optimization in the standard sense. This does not affect the paper's main α<1 results, but the wording should be corrected to avoid a false general statement.","section":"Proposition 1 and Proposition B.1"}],"minor_comments":[{"comment":"There are several typographical issues: 'R WTH Aachen' in the affiliation, 'qusi-norms' in Ref. [19], and the heading 'QUALITY OF COMMUNICATION' in the Introduction.","section":"Throughout"},{"comment":"The maximum is taken over the half-open interval [1/2,1). If the supremum is not attained, the notation should be 'sup' rather than 'max.'","section":"Eq. (13)"},{"comment":"The paper uses E_r(nR;N^{⊗n}) in the achievability bound and E_r(R;N) in the definition. This is correct, but a brief sentence explaining that the exponent is evaluated on the n-fold product channel would help readers.","section":"Eq. (12)–(13)"},{"comment":"Proposition E.2 relies on additivity of the sandwiched Rényi information eI_β for β∈[1/2,1) from Ref. [19]. Since [19] is a preprint, it would strengthen the paper to cite a published version if one becomes available, or at least to state that this particular additivity result is imported from an unpublished source.","section":"Appendix E"},{"comment":"The range of κ for which ρ_{A1A2}(κ) is a valid density operator is not stated. It is clear from the examples, but an explicit condition would improve readability.","section":"Eq. (21)–(23)"},{"comment":"The numerical table reports c_1(α) and κ_quad(α) for a four-dimensional instance. It would help to state explicitly that these numbers are illustrations and that the proof uses only the exact sign of c_1(α), not the numerical values.","section":"Section C.7"}],"recommendation":"major_revision","confidential_remarks":"I found no internal error in the central superadditivity proofs, and the identities (C.42) and (D.82) appear correct. The main reservation is the operational bridge: the advertised error-exponent enhancement depends on an external achievability bound whose exact hypotheses are not stated. If Ref. [20] indeed proves the bound for entanglement-assisted codes with exactly the Petz exponent used here, then the paper would be acceptable after a modest revision. I would not reject on the current evidence, but the scope overclaim for α∈(1,2) and the unstated dependence on [20] should be addressed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuinely new result, and the analytic core is in good shape. They prove strict superadditivity of the Petz–Rényi channel information I_α for α in [1/2,1) (actually (0,1)∪(1,2)) for two explicit families: a Fourier measurement channel and the amplitude-damping channel. That overturns the natural expectation, which the paper carefully documents, that I_α would inherit the additivity of the mutual information and the sandwiched Rényi information. The capacity stays additive; only the reliability exponent improves. The separable, classically correlated witness is the genuinely surprising part, and it is derived in detail.\n\nWhat is good: the proofs are real. The one-copy reduction to a convex one-dimensional problem is done cleanly (Proposition 1, Appendices B–D). The linear-response calculations—c1 formula (C42) for the measurement channel and dQ/dκ formula (D82) for amplitude damping—have the right signs, and the key non-vanishing arguments (t_α ≠ 1/d and x_α ≠ 0) are addressed, not hand-waved. The additivity of I_α at α=1 and α=2 and for the special channel classes in Proposition 2 is a useful calibration. I did not machine-check every algebra step, and the appendices are dense, but I did not find a gap.\n\nThe soft spot is where the math meets the operational claim. The 'multi-copy enhancement of the random-coding error exponent' relies on Eq. (12), ε*(n,R) ≤ 1.5·2^{−E_r(nR;N^{⊗n})}, quoted from [20]. If [20]'s exponent is not exactly the Petz information I_α of Eq. (14), or is not proved for entanglement-assisted codes, then the strict superadditivity theorems remain true as pure channel-information statements, but the communication-reliability conclusion does not automatically follow. The paper also calls E_r 'tight' over the critical rate; that is an expectation grounded in classical results and covariant channels, not something proven here, and it is not needed for the achievability-side claim. This should be checked or explicitly flagged in a revision.\n\nThe self-citations in the background are not excessive, and the cited additivity results are from other groups [17,19], so the attribution looks fair.\n\nWho is this for: anyone working on quantum Shannon theory, error exponents, or entanglement-assisted communication. It corrects a natural additivity assumption and provides a concrete toolset for studying Petz–Rényi information. It deserves a serious referee: the math is substantive, and the operational interpretation just needs to be tightened. I would send it to review, and I would ask the referee to verify the [20] connection and the two linear-response signs.","headline":"First analytic proof that Petz–Rényi channel information is strictly superadditive for entanglement-assisted communication; the math looks solid, but the advertised error-exponent payoff leans on an external achievability bound.","tokens_in":23992,"tokens_out":1950,"would_cite":true,"duration_ms":19699,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","94A17"],"pacs":["03.67.-a","03.67.Hk"],"model":"deepseek-v4-flash","headline":"This paper proves that the Petz–Rényi channel information governing entanglement-assisted random-coding error exponents can be strictly superadditive: joint coding across channel uses improves reliability even though the entanglement-assist","keywords":["entanglement-assisted communication","Petz–Rényi information","superadditivity","random coding error exponent","measurement channels","amplitude damping","quantum channel reliability","additivity"],"falsifier":"Compute I_α(M^⊗2) and 2I_α(M) for the single-heavy Fourier measurement with d=3, λ=1/2, α=1/2 by numerically optimizing Q_α over diagonal two-copy states; the theorem predicts a strictly negative gap I_α(M^⊗2)−2I_α(M). A nonnegative gap would directly refute Theorem 3. Alternatively, exhibit a code for an amplitude-damping channel whose error probability decays faster than 2^{−E_r(nR;N^⊗n)} at a rate below capacity, which would disprove the tightness transfer used in the operational claim.","tokens_in":23160,"feed_emoji":"🔗","tokens_out":5478,"duration_ms":52509,"temperature":0.7,"pith_summary":"Entanglement-assisted communication has an additive single-letter capacity, so correlations across channel uses cannot raise the rate. This paper shows the additive picture fails for reliability: the Petz–Rényi channel information I_α(N) of order α∈[1/2,1), which enters the random-coding error exponent below capacity, is strictly superadditive for certain channels. In particular, two uses of a single-heavy Fourier measurement channel or an amplitude-damping channel give I_α(N^⊗2)>2I_α(N) for every 0<α<1. The improvement is witnessed by a separable, classically correlated two-copy input, so no entanglement between transmitted systems is required. If the cited achievability bound is tight, joint preshared entanglement does not increase how much information can be sent, but it does improve the error probability at rates below capacity.","feed_headline":"Two channel copies beat one for entanglement-assisted reliability","feed_subtitle":"Joint inputs cannot raise capacity, but correlated coding sharpens the error exponent at rates below capacity.","key_machinery":"The load-bearing object is the Petz–Rényi channel information I_α(N), written through the trace functional Q_α(ρ)=Tr[(Tr_A[ρ_A^{1−α}(√ρ_A Γ_N √ρ_A)^α])^{1/α}], with I_α=(α/(α−1))log Q_α. The machinery consists of four steps: (1) Proposition 1, which shows ρ↦Q_α(ρ) is convex for 0<α<1, making I_α a convex optimization and allowing twirling or pinching to diagonal optimizers; (2) a one-parameter diagonal reduction of the one-copy problem; (3) a two-copy ansatz ρ(κ)=ρ_*⊗ρ_*+κΔ, where Δ is a diagonal, traceless, classically correlated perturbation that preserves the one-copy marginals; and (4) an exact linear-response identity showing the first derivative c_1(α)<0 at κ=0, so by Taylor expansion","core_discovery":"The central claim is strict superadditivity of the Petz–Rényi channel information I_α(N): there exist channels for which I_α(N^⊗2)>2I_α(N) for every 0<α<1, and in particular in the data-processing-safe range [1/2,1). The paper proves this analytically for the single-heavy Fourier measurement M with d≥3 and 1/d<λ<1, and for the qubit amplitude-damping channel with 0<γ<1. Because the random-coding error exponent is E_r(R;N)=max_{1/2≤α<1} (1−α)/α [I_α(N)−R], strict superadditivity of I_α gives E_r(2R;N^⊗2)>2E_r(R;N) for every rate R below capacity—a genuine multi-copy enhancement of reliability. The proof reduces I_α to a convex optimization, shows the one-copy optimizer can be chosen diagonal,","pith_inferences":["If the random-coding exponent E_r is indeed tight for general channels, then the true error exponent of entanglement-assisted communication is not single-letter, even though capacity is single-letter; quality of communication would require regularization while quantity does not.","The same linear-response mechanism likely extends to any channel with a diagonal symmetry and a non-uniform one-copy optimizer; a testable conjecture is that superadditivity appears whenever the square (ξ−θ_α)^2 in the Fourier-measurement calculation is nonzero.","Because the witness is classically correlated, the effect may carry over to channel discrimination and Rényi channel entropy settings, where product strategies would be suboptimal—an implication the paper only touches numerically.","The gap vanishes continuously as α→1, matching the additive limit; quantifying the α-dependence of the gap could indicate how large the reliability gain is for practical finite block lengths."],"forward_implications":["For the two example channels, I_α(N^⊗2)>2I_α(N) for every 0<α<1, so the random-coding error exponent for two channel uses strictly exceeds twice the single-use exponent at every rate below capacity.","The strict superadditivity is witnessed by a separable, classically correlated two-copy input, so entanglement between the transmitted systems is not a necessary resource for the reliability enhancement.","The phenomenon occurs already for entanglement-breaking measurement channels, whose unassisted Holevo information is additive; hence it is not tied to channels that already show capacity nonadditivity.","The regularized Petz–Rényi information I_α^∞(N)=lim_{n→∞}(1/n)I_α(N^⊗n) admits computable single-letter upper bounds in terms of sandwiched Rényi information, bounding the ultimate multi-copy advantage.","At α=1 the nonadditivity disappears, matching the additivity of entanglement-assisted capacity, and at α=2 the paper proves strong additivity of I_α."],"fun_headline_variants":["Error exponent goes superadditive: joint coding boosts reliability","No entanglement needed for multi-copy reliability boost","Capacity unchanged, but correlated inputs sharpen error exponent","Two uses beat one: superadditive error exponent, additive capacity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The operational conclusion that strict superadditivity of I_α translates into improved error probability rests on the cited achievability bound ε*(n,R)≤1.5·2^{−E_r(nR;N^⊗n)} and on the expectation that E_r is tight; the mathematical superadditivity itself does not depend on that assumption.","fun_headline_variants_meta":{"raw":{"variants":["Error exponent goes superadditive: joint coding boosts reliability","No entanglement needed for multi-copy reliability boost","Capacity unchanged, but correlated inputs sharpen error exponent","Two uses beat one: superadditive error exponent, additive capacity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000227,"raw_usage":{"total_tokens":1308,"prompt_tokens":744,"completion_tokens":564,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":488,"completion_tokens_details":{"reasoning_tokens":499}},"tokens_in":488,"tokens_out":564,"duration_ms":6549,"temperature":1.0,"reasoning_tokens":499,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T23:59:41.616415+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute I_α(M^⊗2) and 2I_α(M) for the single-heavy Fourier measurement with d=3, λ=1/2, α=1/2 by numerically optimizing Q_α over diagonal two-copy states; the theorem predicts a strictly negative gap I_α(M^⊗2)−2I_α(M). A nonnegative gap would directly refute Theorem 3. Alternatively, exhibit a code for an amplitude-damping channel whose error probability decays faster than 2^{−E_r(nR;N^⊗n)} at a rate below capacity, which would disprove the tightness transfer used in the operational claim.","supporting_citations":[],"review_version":1}