{"id":"ef3dad15-086b-4e18-a518-b8501a0bf304","arxiv_id":"2507.15661","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Anti-degradable channels have a strong converse for private capacity off the boundary curve in the error-privacy plane, and a pretty strong converse for quantum capacity whenever the error is below 1/sqrt(2).","lead":"This paper proves a strong converse for the private classical capacity of anti-degradable quantum channels: for fixed nonzero error and privacy parameters, the number of reliably and privately transmittable messages stays bounded independently of the number of channel uses. It also gives a simpler proof of a pretty strong converse for quantum capacity and improves the valid error threshold from 1/2 to 1/sqrt(2).","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2's stated domain is invalid: Lemma 1 requires α+β<π/2, i.e., ε²+δ²<1, but the theorem allows sin(α+β)<1, and the counterexample channel falsifies it.","rationale":"The reader identified Lemma 1 and the introductory simplification as the weak spots but did not catch the more serious mismatch: the lemma's domain is α+β<π/2, whereas the theorem states the weaker condition sin(α+β)<1. This is not a minor text issue; the theorem as stated is false, as shown by the constant channel with ε=δ=0.95. The proof itself is sound once the condition is corrected to ε²+δ²<1, which still covers a strong-converse region: for any fixed privacy δ>0 all error thresholds ε<√(1−δ²) are included, and the infimum over ε,δ in the capacity definition still yields zero private capacity. Because the main qualitative contribution is salvageable by a precise restriction of the theorem, the appropriate verdict is CONDITIONAL, but the required revision is substantive: change the condition in Theorem 2, the abstract, the introduction, and Figure 1 from sin(α+β)<1 to α+β<π/2 (equivalently ε²+δ²<1). Without this correction, the central claim is demonstrably false.","tokens_in":50,"tokens_out":27829,"duration_ms":980325,"concrete_test":"Evaluate the claimed theorem on the constant anti-degradable channel N(ρ)=Tr(ρ)|0><0| with U|i>=|0>|e_i>, n=1, ε=δ=0.95. A code with M=3 uses the same input for all messages; the environment is independent of the message, so privacy is 0, and the output decoding to a fixed symbol gives P(ξ,ρ)=√(1−1/9)≈0.943≤0.95. Thus M(1,0.95,0.95)≥3. The stated theorem gives log M≤2log(1/cos(2arcsin0.95))≈0.434, i.e., M≤1.54. This contradiction settles the domain error. Then re-derive the theorem with condition ε²+δ²<1 and check that the proof line 'it holds for α+β<π/2' is consistently used.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 2 invokes Lemma 1 as Hmin^{sin β} ≤ Hmax^{sin α} + log(1/cos²(α+β)), whose stated validity requires α+β<π/2, equivalently ε²+δ²<1. The theorem's condition δ√(1−ε²)+ε√(1−δ²)<1 is sin(α+β)<1, which is strictly weaker because it also holds when α+β>π/2. The proof itself notes 'it holds for α+β<π/2' and then incorrectly concludes the bound on the larger region. Moreover, the bound 2log(1/cos(α+β)) is not even real when α+β>π/2. The error is substantive: for the anti-degradable constant channel N(ρ)=Tr(ρ)|0><0| with isometric extension U|i>=|0>|e_i>, take n=1, ε=δ=0.95. Using the same input for all M=3 messages and a fixed decoder output gives purified error √(1−1/9)=√(8/9)≈0.943≤0.95 and privacy δ=0, so M(1,0.95,0.95)≥3. Theorem 2 would bound log M by log(1/cos²(2arcsin0.95))≈0.434, i.e., M≤1.54. Contradiction. The correct condition for the proof is ε²+δ²<1; the introduction's δ+ε<1 is only a sufficient subregion. The abstract, Theorem 2, Figure 1, and the 'simplified' comment all need correction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies finite-blocklength converse bounds for anti-degradable quantum channels. Its main result, Theorem 2, claims a strong converse for the private classical capacity: for any error epsilon > 0 and privacy delta > 0 with delta sqrt(1 - epsilon^2) + epsilon sqrt(1 - delta^2) < 1, every integer n satisfies log M(n, epsilon, delta) <= 2 log(1/cos(alpha + beta)), where alpha = arcsin(epsilon) and beta = arcsin(delta). The proof combines a smooth min-entropy lower bound from the privacy condition, data processing from the environment to Bob (using anti-degradability), a smooth max-entropy upper bound from the decoding condition, and the smooth min/max entropy inequality of Lemma 1. The paper also proves, in Theorem 3, a pretty strong converse for the quantum capacity of anti-degradable channels: log N(n, epsilon) <= log(1/cos(2 alpha)) for any epsilon < 1/sqrt(2). The quantum result and the proof strategy are transparent, but the domain of validity of Theorem 2 is incorrect as stated.","tokens_in":1636,"tokens_out":2173,"duration_ms":170785,"significance":"If corrected, the proof technique would provide a clean finite-n converse bound for anti-degradable channels using only standard smooth-entropy tools, and the quantum-capacity bound is a nice simplification of earlier work. The paper is explicit and the derivations are easy to follow. However, the central private-capacity theorem as stated is false: the condition in Theorem 2 is strictly weaker than the condition under which Lemma 1 is invoked, and a concrete constant-channel counterexample falsifies the claimed bound. The result can be repaired by replacing the condition with epsilon^2 + delta^2 < 1 (equivalently alpha + beta < pi/2), but the abstract, Theorem 2, Figure 1, and the discussion of the boundary all need substantial revision. With that correction the paper would still be a meaningful contribution, though the claimed sharp boundary would be lost.","major_comments":[{"comment":"The theorem's stated condition is delta sqrt(1 - epsilon^2) + epsilon sqrt(1 - delta^2) < 1, i.e. sin(alpha + beta) < 1. For alpha, beta in [0, pi/2], this also holds when alpha + beta > pi/2, but Lemma 1 is valid only for alpha + beta < pi/2. In the regime alpha + beta > pi/2 the quantity cos(alpha + beta) is negative and the claimed bound 2 log(1/cos(alpha + beta)) is not real. The proof acknowledges this with 'it holds for alpha + beta < pi/2' but then incorrectly concludes the bound on the larger region. The correct condition is alpha + beta < pi/2, equivalently epsilon^2 + delta^2 < 1. The error is load-bearing: for the anti-degradable constant channel N(rho) = Tr(rho)|0><0| with isometry U|i> = |0>|e_i>, taking n = 1, epsilon = delta = 0.95, and M = 3, encoding all messages to the same input and decoding to a fixed output gives purified error sqrt(1 - 1/9) <= 0.95 and privacy delta = 0, so M(1, 0.95, 0.95) >= 3. The theorem as stated would give M <= 1.54, a contradiction. The proof cannot be repaired by a sign convention; the lemma genuinely does not apply when alpha + beta > pi/2.","section":"IV, Theorem 2 and Eqs. (20)-(22)"},{"comment":"The introduction's statement that the bound 'can be simplified to delta + epsilon < 1' is not a simplification of the stated condition sin(alpha + beta) < 1; the latter holds, for example, for epsilon = delta = 0.95, where delta + epsilon > 1. With the necessary correction epsilon^2 + delta^2 < 1, the relation delta + epsilon < 1 is merely a sufficient subregion. Figure 1 currently shades the region sin(alpha + beta) < 1, which includes the invalid region alpha + beta > pi/2; it should instead show the disk epsilon^2 + delta^2 < 1. The abstract's phrase 'sharply defining the boundary' is also overstated: the corrected theorem gives a sufficient condition for a finite-n bound, not a sharp characterization of all parameters for which private communication is impossible.","section":"I and Figure 1"},{"comment":"The parenthetical assertion that the private classical capacity of anti-degradable channels is equal to 0 is made without proof or citation. This fact is used to frame the result as a strong converse for the capacity, so it should be justified, either by a reference or by a short argument from the definition of anti-degradability (the environment can simulate Bob's output).","section":"IV, first paragraph"}],"minor_comments":[{"comment":"In the application of Lemma 1, the roles of alpha and beta are exchanged relative to the lemma statement: the proof bounds H_min^{sin beta} by H_max^{sin alpha} plus the logarithmic term. This is valid after renaming the lemma's parameters, but a sentence noting the exchange would improve readability.","section":"Eq. (21)"},{"comment":"The statement gives log N(n, epsilon) <= log(1/cos(2 alpha)), while the proof derives 2 log N <= log(1/cos^2(2 alpha)); the equivalence is immediate but should be made explicit to avoid confusion.","section":"Theorem 3 and Eq. (25)"},{"comment":"There are minor typos: 'for the later' should be 'for the latter', and the Acknowledgments contain a duplicated 'and'. These do not affect the content.","section":"Discussion and Acknowledgments"},{"comment":"After the condition is corrected to epsilon^2 + delta^2 < 1, the abstract should state the condition explicitly; the current wording 'whenever ... satisfy ...' will otherwise continue to assert a false domain of validity.","section":"Abstract and Theorem 2"}],"recommendation":"major_revision","confidential_remarks":"The false domain in Theorem 2 is serious and could justify rejection if the authors do not weaken the statement. However, the proof strategy is sound on the corrected region epsilon^2 + delta^2 < 1, and the quantum-capacity theorem is unaffected. I therefore recommend major revision rather than rejection, provided the authors revise the theorem, abstract, and figure to the correct condition and remove the claim of a sharp boundary."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper has a real idea and a mostly correct proof structure, but the main theorem overclaims its domain. The stress-test note is right, and the reader's report missed it. \n\nWhat is actually new: a finite-blocklength converse bound for the private capacity of anti-degradable channels, where previously only a pretty strong converse was known, plus a simpler pretty strong converse for quantum capacity with epsilon < 1/sqrt(2). The proof is standard smooth-entropy bookkeeping: privacy gives a min-entropy lower bound, data processing goes from environment to Bob, decoding gives a max-entropy bound, and Tomamichel's Lemma 1 combines them. There is no circularity, no tuned parameters, and the derivation is explicit. That part is sound. \n\nThe soft spot is the domain in Theorem 2. The condition delta*sqrt(1-epsilon^2) + epsilon*sqrt(1-delta^2) < 1 is sin(alpha+beta) < 1. Lemma 1 requires alpha+beta < pi/2, which is equivalent to epsilon^2 + delta^2 < 1. The proof itself says \"it holds for alpha+beta < pi/2\" and then concludes on the larger region. Moreover, the bound 2log(1/cos(alpha+beta)) is not even real when alpha+beta > pi/2. The stress-test counterexample is valid: for the anti-degradable constant channel N(rho)=Tr(rho)|0><0|, encoding all M=3 messages to the same input and using a fixed decoder gives purified error sqrt(1-1/9) = 0.943 <= 0.95, privacy delta=0, so M(1,0.95,0.95) >= 3, while the theorem would force M <= 1.54. So the statement is false as written. \n\nThe fix is straightforward: replace the condition with epsilon^2 + delta^2 < 1, and correct the introduction, where \"this bound can be simplified to delta+epsilon<1\" is not an equivalent simplification; delta+epsilon<1 is merely a sufficient subregion. With that correction the asymptotic conclusion still holds and the paper becomes a clean, useful contribution. It does not resolve a major open problem, but it fills a specific gap. \n\nThis paper deserves peer review: the corrected result is worth publishing and the proof is simple enough to verify. I would not cite the current arXiv version because the main theorem is false, but I would cite the corrected version. For a reading group, it is actually a good case study in checking parameter domains.","headline":"The core idea is right and the proof is mostly clean, but Theorem 2 as stated is false: the parameter domain is too weak and an explicit counterexample kills it.","tokens_in":6279,"tokens_out":7967,"would_cite":false,"duration_ms":86117,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Anti-degradable quantum channels have a strong converse for private classical capacity: for any allowed error and privacy parameters, the largest message a code can carry is bounded by a constant independent of the number of channel uses.","keywords":["anti-degradable channels","private classical capacity","quantum capacity","strong converse","pretty strong converse","smooth min-entropy","smooth max-entropy","quantum channel capacity"],"falsifier":"Exhibit any finite-$n$ private code on an anti-degradable channel (for example the erasure channel with erasure probability at least $1/2$) whose error and privacy satisfy $\\delta\\sqrt{1-\\epsilon^2}+\\epsilon\\sqrt{1-\\delta^2}<1$ but whose message set exceeds $2\\log(1/\\cos(\\alpha+\\beta))$; that single instance would falsify Theorem 2. Alternatively, a direct counterexample to Lemma 1 on a subnormalized bipartite state within its stated domain would break both capacity bounds.","tokens_in":5172,"feed_emoji":"🔐","tokens_out":11710,"duration_ms":116581,"temperature":0.7,"pith_summary":"This paper proves a strong converse for the private classical capacity of anti-degradable quantum channels, meaning channels whose environment can reproduce Bob's output on its own. For any allowed error $\\epsilon>0$ and privacy parameter $\\delta>0$ satisfying $\\delta\\sqrt{1-\\epsilon^2}+\\epsilon\\sqrt{1-\\delta^2}<1$, every $n$-use code has message set bounded by $\\log M(n,\\epsilon,\\delta)\\le 2\\log(1/\\cos(\\alpha+\\beta))$, where $\\alpha=\\sin^{-1}\\epsilon$ and $\\beta=\\sin^{-1}\\delta$. Since the right-hand side does not depend on $n$, positive-rate private communication over such channels is impossible in the strong sense. The same smooth-entropy argument gives a pretty strong converse for the quantum capacity of the same channels for any error $\\epsilon<1/\\sqrt{2}$, with $\\log \\mathcal{N}(n,\\epsilon)\\le \\log(1/\\cos(2\\alpha))$. The result replaces the previous pretty strong converse for private capacity with a full strong converse for an entire class of zero-capacity channels.","feed_headline":"No positive-rate private code survives anti-degradable channels","feed_subtitle":"For any allowed error and privacy, message size is bounded independent of channel uses — private rate is zero.","key_machinery":"The load-bearing object is the pair of smoothed conditional entropies $H^{\\eta}_{\\min}(A|B)_{\\rho}$ and $H^{\\eta}_{\\max}(A|B)_{\\rho}$, defined by optimizing the ordinary min- and max-entropies over states within purified distance $\\eta$. The proof has three steps: (i) in the ideal private code the min-entropy of the classical reference $X'$ given the environment $E^n$ equals $\\log M$, while the max-entropy of $X'$ given Bob's decoded output $X$ is $0$; (ii) data processing moves these entropies between $B^n$ and $E^n$, which is exactly where anti-degradability enters, since the channel's output can be produced from its environment; and (iii) Lemma 1, the inequality $H_{\\min}^{\\sin\\alpha}(A|B)\\le H_{\\max}^{\\sin\\beta}(A|B)+\\log(1/\\cos^2(\\alpha+\\beta))$ for $\\alpha+\\beta<\\pi/2$, converts the two smoothing parameters into a log-cosine constant. The quantum-capacity case is the same chain with $\\beta=\\alpha$ and uses the duality between smooth min- and max-entropy.","core_discovery":"The central claim is that anti-degradability turns the private capacity problem into a sandwich of smooth entropies with an $n$-independent output. Namely, Theorems 2 and 3 show that for a finite-dimensional anti-degradable channel $\\mathcal{N}$, every integer $n$, and any error/privacy pair with $\\delta\\sqrt{1-\\epsilon^2}+\\epsilon\\sqrt{1-\\delta^2}<1$, the private classical code size obeys $\\log M(n,\\epsilon,\\delta)\\le 2\\log(1/\\cos(\\alpha+\\beta))$, with $\\alpha=\\sin^{-1}\\epsilon$ and $\\beta=\\sin^{-1}\\delta$. Because the bound contains no $n$, any family of codes whose rates stay above zero must fail either reliability or privacy; hence the private classical capacity of anti-degradable channels is strongly zero. The same machinery gives the quantum-capacity corollary: for $\\epsilon<1/\\sqrt{2}$, $\\log\\mathcal{N}(n,\\epsilon)\\le \\log(1/\\cos(2\\alpha))$, a pretty strong converse with a simpler proof than the previously known one.","pith_inferences":["An extension left implicit by the paper: because the bound is finite-$n$ and fully explicit, it automatically supplies a one-shot private-capacity upper bound for anti-degradable channels, which could be compared against achievable one-shot rates to test tightness.","The same smooth min/max template may prove strong converses for any channel class in which Bob's output can be simulated from the environment (a hierarchy of 'simulatable' channels), provided the corresponding data-processing chain survives; this goes beyond what the paper itself claims.","Optimizing the smoothing parameters or replacing Lemma 1 by a sharper min/max inequality could move the threshold toward the full boundary $\\delta+\\epsilon<1$ and improve the constant; the authors do not claim their constant is optimal."],"forward_implications":["For every anti-degradable channel, any sequence of private codes whose message size grows faster than a constant must violate either the error or the privacy condition in the limit; in particular, the private classical capacity is strongly zero.","The finite-$n$ bound holds for all codes at once, so it also restricts one-shot and short-block private communication on anti-degradable channels, not just the asymptotic rate.","The quantum-capacity result extends the pretty strong converse for anti-degradable channels from $\\epsilon<1/2$ to $\\epsilon<1/\\sqrt{2}$, with an explicit $n$-independent bound $\\log(1/\\cos(2\\alpha))$.","The sufficient condition $\\delta\\sqrt{1-\\epsilon^2}+\\epsilon\\sqrt{1-\\delta^2}<1$ covers in particular all pairs with $\\delta+\\epsilon<1$, showing that the privacy parameter can be arbitrarily small while the error is close to $1$, and vice versa."],"supporting_citations":[{"why":"Supplies Lemma 1, the smooth min/max entropy inequality from which the constant in Theorems 2 and 3 is obtained.","marker":"[9]"},{"why":"Introduces the pretty strong converse concept and proves it for degradable channels, the baseline the present result strengthens.","marker":"[4]"},{"why":"Gives the earlier pretty strong converse for the quantum capacity of anti-degradable channels for $\\epsilon<1/2$, which Theorem 3 extends to $\\epsilon<1/\\sqrt{2}$ with a simpler proof.","marker":"[5]"},{"why":"Provides the duality between smooth min- and max-entropies used in the quantum-capacity proof.","marker":"[8]"},{"why":"Sets up the smooth min/max entropy formalism and the entropy definitions used throughout the proofs.","marker":"[7]"},{"why":"Defines the private classical capacity and quantum capacity as the asymptotic quantities whose converses are being proven.","marker":"[3]"}],"fun_headline_variants":["Private codes die on anti-degradable channels","Anti-degradable channels collapse private rate to zero","Strong converse: anti-degradable channels forbid private codes","Anti-degradable channels: zero private capacity for any error","Quantum and private rates vanish for anti-degradable channels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole bound rests on the quoted smooth min/max entropy inequality of Lemma 1; if that inequality fails for some subnormalized state or for some pair of smoothing parameters with $\\alpha+\\beta<\\pi/2$, the constant in Theorems 2 and 3 no longer follows.","fun_headline_variants_meta":{"raw":{"variants":["Private codes die on anti-degradable channels","Anti-degradable channels collapse private rate to zero","Strong converse: anti-degradable channels forbid private codes","Anti-degradable channels: zero private capacity for any error","Quantum and private rates vanish for anti-degradable channels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001421,"raw_usage":{"total_tokens":5718,"prompt_tokens":913,"completion_tokens":4805,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":4726}},"tokens_in":529,"tokens_out":4805,"duration_ms":48024,"temperature":1.0,"reasoning_tokens":4726,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:28:13.425489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit any finite-$n$ private code on an anti-degradable channel (for example the erasure channel with erasure probability at least $1/2$) whose error and privacy satisfy $\\delta\\sqrt{1-\\epsilon^2}+\\epsilon\\sqrt{1-\\delta^2}<1$ but whose message set exceeds $2\\log(1/\\cos(\\alpha+\\beta))$; that single instance would falsify Theorem 2. Alternatively, a direct counterexample to Lemma 1 on a subnormalized bipartite state within its stated domain would break both capacity bounds.","supporting_citations":[{"cited_title":"“pretty strong","cited_arxiv_id":null,"evidence_quote":"Supplies Lemma 1, the smooth min/max entropy inequality from which the constant in Theorems 2 and 3 is obtained."},{"cited_title":"(6) A channel N : A → B is called degradable if it can be degraded to its complementary channel, i.e","cited_arxiv_id":null,"evidence_quote":"Introduces the pretty strong converse concept and proves it for degradable channels, the baseline the present result strengthens."},{"cited_title":"This relation is illustrated in Fig","cited_arxiv_id":null,"evidence_quote":"Gives the earlier pretty strong converse for the quantum capacity of anti-degradable channels for $\\epsilon<1/2$, which Theorem 3 extends to $\\epsilon<1/\\sqrt{2}$ with a simpler proof."},{"cited_title":"The private classical capacity and quantum capacity of a quantum channel,","cited_arxiv_id":null,"evidence_quote":"Provides the duality between smooth min- and max-entropies used in the quantum-capacity proof."},{"cited_title":"Capacity of the noisy quantum channel,","cited_arxiv_id":null,"evidence_quote":"Sets up the smooth min/max entropy formalism and the entropy definitions used throughout the proofs."},{"cited_title":"The paper is structured as follows: In Section II we introduce the necessary notation and preliminary results","cited_arxiv_id":null,"evidence_quote":"Defines the private classical capacity and quantum capacity as the asymptotic quantities whose converses are being proven."}],"review_version":1}