{"id":"5958bded-f937-4ae5-996a-b64d40232910","arxiv_id":"2607.24687","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Quantum conditional entropy is continuous in trace distance with optimal modulus h₂(δ)+δ log(d²−1) up to δ=1−d⁻² and 2 log d thereafter, tight when dim B≥d.","lead":"The paper proves the sharpest possible continuity bound for quantum conditional entropy that depends only on the dimension of one subsystem. This closes a known open case left after earlier classical, classical-quantum, and same-marginal quantum results.","discovery_kind":"extension","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"No significant objection identified. The proof of Theorem 1.1 is a short chain of standard finite-dimensional operator inequalities, and every step survives an independent line-by-line re-derivation, including the tightness example.","rationale":"The reader identified the pinching bound (Eq. 1) as the weakest assumption while correctly noting it is tight rather than loose. My independent pass confirms this: the entire upper bound rests on the constant d in X_AB ≤ d·1_A⊗X_B, and the matching entangled saturating example demonstrates that d²−1 cannot be improved, so the assumption is not a soft spot. I verified each nontrivial step of Theorem 1.1 (regularization, positivity and marginal of σ̂, the branch condition δ ≤ 1−d⁻² from the coefficient of σ_AB, the score identity, the data-processing sign control, the operator-monotone-log sandwich, the Jordan-decomposition estimate, and the final algebraic identity), as well as the trace distance, marginal, and entropy computations for the tightness family and the plateau value 2 log d at s = 1−d⁻². The refinements in §2 (κ_σ hierarchy and Schmidt-number interpolation) use the same mechanism and are consistent with the known classical (Alhejji–Smith), CQ (Wilde), and same-marginal ([5],[6]) special cases, which the paper extends rather than reclaims. There is no empirical component, no free parameter, and no circularity. The AI-assistance disclosure does not bear on correctness since the manuscript is fully self-contained and verifiable. The reader's ACCEPT with HIGH confidence and low correctness risk is appropriate; my read does not move the verdict.","tokens_in":6880,"tokens_out":4516,"duration_ms":153752,"concrete_test":"Numerical stress test for small dimensions: for d = dim A = 2 and 3 with dim B = d, maximize |H(A|B)_ρ − H(A|B)_σ| over randomly sampled state pairs (and local optimization from many seeds) subject to T(ρ,σ) ≤ δ on a grid of δ ∈ (0,1]. If any sampled pair exceeds h₂(δ) + δ log(d²−1) (resp. 2 log d on the plateau) by more than numerical tolerance, the bound is false; if the maximized values track g_{d²}(δ) and saturate it at the grid points, the modulus and its tightness are corroborated independently of the proof. As a second, cheaper check, independently re-derive the two-sided sandwich (3) and the marginal identity ρ_B = 1_B/d for the saturating example.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I re-derived the argument rather than relying on the reader's summary. The steps that could plausibly hide an error all check out: (i) the pinching bound (Eq. 1) is the standard X ≤ d·P(X) ≤ d·1_A⊗X_B for a d-block pinching, so σ̂_AB ≥ 0, Tr σ̂ = 1, and σ̂_B = σ_B all hold as stated; (ii) the coefficient of σ_AB in τ_AB is (d²−1−d²δ)/(d²−1), which is nonnegative exactly on the first branch δ ≤ 1−d⁻², so the lower sandwich in (3) is valid precisely where the proof needs it; (iii) the score identity H(A|B)_ρ − H(A|B)_σ = Tr[(ρ−σ)G] + D(σ‖τ) − D(ρ‖τ) + D(ρ_B‖σ_B) expands correctly, and dropping −D(ρ‖τ) + D(ρ_B‖σ_B) is exactly data processing under the partial trace since τ_B = σ_B; (iv) the operator-monotone-log step applied to (3) is legitimate because all operators involved commute with 1_A⊗σ_B; (v) the Jordan-decomposition estimate uses t ≤ δ with a log factor that is nonnegative iff δ ≤ 1−d⁻², matching the branch condition; (vi) the final algebra δ log[(d²−1)(1−δ)/δ] − log(1−δ) = h₂(δ) + δ log(d²−1) is exact. The tightness example also verifies: ρ−σ has eigenvalue −s on Φ and s/(d²−1) on the d²−1 orthogonal directions, giving T = s; ρ_B = 1_B/d, so ΔH(A|B) = H(ρ) = h₂(s) + s log(d²−1); and at s = 1−d⁻² this equals 2 log d, confirming the plateau. The only assumption one could attack is the pinching constant d, but the saturating entangled example proves d²−1 is optimal, so the construction is tight, not loose. The ChatGPT-assistance disclosure is unusual but irrelevant to soundness: the text is self-contained and fully checkable. I find no load-bearing weakness.","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript proves that, for bipartite finite-dimensional states ρ_AB and σ_AB at trace distance at most δ and d=dim A, the sharp dimension-only uniform continuity modulus of H(A|B) is h₂(δ)+δ log(d²−1) up to δ=1−d⁻², followed by the plateau 2 log d. The proof regularizes to full rank, constructs the canonical complement σ̂_AB=(d 1_A⊗σ_B−σ_AB)/(d²−1), and combines a relative-entropy score identity, data processing, a two-sided pinching estimate, and a Jordan-decomposition bound. Matching entangled examples establish optimality when dim B≥d. Section 2 extends the same construction to a conditional-min-entropy refinement and an exact Schmidt-number interpolation, while the appendices give a global proof of the classical Alhejji–Smith bound and a comparison with Winter’s argument.","tokens_in":7382,"tokens_out":6572,"duration_ms":261936,"significance":"If accepted as written, this settles a well-known open case of the fully quantum conditional-entropy continuity problem and realizes Wilde’s conjectured log(d²−1) constant without restricting the B-marginals. The derivation is short, self-contained, and parameter-free, and the lower-bound examples are independent constructions that certify optimality for every δ rather than merely matching the asymptotics. The exact Schmidt-number hierarchy in Proposition 2.1 is a useful additional result connecting the classical, separable, and unrestricted quantum regimes. The elementary nature of the proof should also make the improvement easy to propagate into applications of Alicki–Fannes-type bounds.","major_comments":[],"minor_comments":[{"comment":"With Δ_AB=t(Δ_+,AB−Δ_−,AB) and Tr Δ_±=1, the displayed equality should read Tr[ΔG]=t(Tr[Δ_+G]−Tr[Δ_−G]), with the factor t also multiplying the bracketed logarithms on the right. The next line restores the factor, so this is typographical, but it occurs in the central estimate and should be corrected.","section":"§1, proof of Theorem 1.1, Jordan-decomposition display"},{"comment":"For completeness, add a one-line justification of the second inequality in Eq. (1): if X_AB=∑_{ij}|i⟩⟨j|⊗X_{ij}, positivity gives X_{ii}≤∑_j X_{jj}=X_B for each i, hence the pinched operator is bounded by 1_A⊗X_B.","section":"§1, Eq. (1)"},{"comment":"Equation (7) uses monotonicity of g_K in K, whereas the wording around Eq. (4) most naturally asserts monotonicity in δ. Please state and briefly verify both monotonicities, perhaps also noting continuity at δ=1−K⁻¹. This appears to be a presentation omission rather than a mathematical gap.","section":"§2, Eqs. (4) and (7)"},{"comment":"The funding sentence ends with “(ML4Q-2) and .” and appears to be truncated. Please complete the missing grant or institutional information.","section":"Acknowledgements"},{"comment":"The AI-assistance disclosure is commendably explicit. Depending on journal style, the detailed tool name/version might be better placed in an acknowledgements or disclosure section, accompanied by a statement that the authors checked and take responsibility for the final proof. This would preserve transparency without diverting attention from the theorem in the abstract.","section":"Abstract and Introduction"}],"recommendation":"minor_revision","confidential_remarks":"I found no load-bearing technical problem. The prominent statement that the key idea was developed with assistance from a named generative-AI system does not affect the mathematical assessment, but the editor may wish to verify that the disclosure satisfies the journal’s current AI-use and author-responsibility policy, especially because the manuscript itself frames the submission partly as a prompt for discussion of AI-assisted mathematics."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing to know: this closes the unrestricted fully quantum sharp continuity bound for conditional entropy—the case Wilde conjectured and that the same-marginal papers left open. The modulus is h₂(δ)+δ log(d²−1) up to δ=1−d⁻² and 2 log d after, and it is optimal whenever dim B ≥ d.\n\nWhat is actually new is the one-sided comparison state built only from σ: the canonical complement σ̂ = (d 1_A⊗σ_B − σ)/(d²−1), mixed to τ, then a score identity plus data-processing and an operator sandwich on G. That breaks the symmetry in Winter’s argument and recovers the d²−1 constant instead of d², and h₂(δ) instead of the looser (1+t)h₂(t/(1+t)). The algebra is short and closes exactly; I re-checked the pinching, the branch condition on the coefficient of σ in τ, the Jordan bound, and the final binary-entropy identity—they all hold. Sharpness is by the usual maximally-entangled mixture and saturates for every δ. Proposition 2.1 then gives a clean Schmidt-number hierarchy interpolating classical / separable / full quantum, which is useful packaging.\n\nSoft spots are minor. The whole upper bound rests on the standard pinching constant d; that is tight rather than loose because the entangled example matches. No free parameters, no circularity, citations to Alhejji–Smith, Wilde, Berta–Lami–Tomamichel, and Audenaert et al. are in the right places. The ChatGPT disclosure is unusual but irrelevant—the text is self-contained and line-checkable. Appendices on the classical case and the comparison to Winter are helpful, not padding.\n\nThis is for anyone who uses continuity of conditional entropy in coding, security, or resource theories and wants the optimal dimension-only modulus. It deserves a serious referee; I would accept it for peer review and I would cite the bound. Bring it to reading group if people still quote Winter’s constant out of habit.","headline":"Clean resolution of the remaining open Alicki–Fannes–Winter case; the elementary one-sided complement proof checks out and the modulus is tight.","tokens_in":8484,"tokens_out":535,"would_cite":true,"duration_ms":13834,"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":"The sharp continuity modulus of quantum conditional entropy is h₂(δ) + δ log(d²−1) up to δ = 1−d⁻², then 2 log d.","keywords":["quantum conditional entropy","uniform continuity","trace distance","Alicki–Fannes bound","Schmidt number","conditional min-entropy","sharp modulus"],"falsifier":"Exhibit any pair of states on systems of dimensions d and m ≥ d whose trace distance is at most some δ ≤ 1−d⁻² yet whose conditional-entropy difference strictly exceeds h₂(δ) + δ log(d²−1), or show that the maximally-entangled construction fails to attain equality.","tokens_in":8059,"feed_emoji":"⚛️","tokens_out":1108,"duration_ms":18852,"temperature":0.7,"pith_summary":"Quantum conditional entropy measures how much uncertainty remains about one quantum system once another is known. Earlier bounds on how this quantity can jump when two bipartite states are close in trace distance were not sharp. This paper proves the optimal dimension-only modulus: if the states differ by at most δ and the conditioned system has dimension d, the difference of conditional entropies is at most the binary entropy of δ plus δ times log(d²−1), until δ reaches 1−1/d², after which the bound saturates at the absolute maximum 2 log d. When the conditioning system is at least as large as d, the bound is achieved by an explicit pair of states built from a maximally entangled state and its orthogonal complement. The argument adapts a classical tight proof by constructing a single comparison state from one of the two inputs alone, rather than mixing both toward a common state, and thereby keeps the constant d²−1 instead of the looser d². The same construction yields a hierarchy controlled by conditional min-entropy and Schmidt number, recovering the classical and separable cases as special instances.","feed_headline":"Sharp continuity bound for quantum conditional entropy","feed_subtitle":"The optimal modulus is binary entropy of δ plus δ log(d²−1), tight whenever the conditioning system is large enough","key_machinery":"The canonical complement ˆσ_AB = (d 1_A ⊗ σ_B − σ_AB)/(d²−1), mixed with σ_AB to form the comparison state τ_AB = (1−δ)σ_AB + δ ˆσ_AB. This state shares the B-marginal of σ, sits in a two-sided operator interval with multiples of 1_A ⊗ σ_B, and converts the entropy difference into a controlled linear term plus a relative-entropy remainder bounded by −log(1−δ).","core_discovery":"For bipartite states ρ_AB and σ_AB at trace distance at most δ, with d = dim A ≥ 2, the absolute difference of conditional entropies |H(A|B)_ρ − H(A|B)_σ| is at most h₂(δ) + δ log(d²−1) when 0 ≤ δ ≤ 1−d⁻² and at most 2 log d thereafter. When dim B ≥ d the right-hand side is optimal for every δ in [0,1], attained by mixing a maximally entangled state toward the normalized projector onto its orthogonal complement.","pith_inferences":["Because the plateau begins already at δ = 1−d⁻² rather than at δ = 1, many finite-size security proofs that only need continuity up to moderate distance can now quote a strictly smaller additive error.","Energy-constrained infinite-dimensional extensions suggested in the conclusion would likely replace the global dimension d by an effective dimension set by the energy cutoff, recovering the same functional form of g_K.","A coupling formulation that unifies the classical fibre-wise argument with the present global quantum construction could yield sharp continuity for other channel divergences that lack an obvious pinching bound."],"forward_implications":["Continuity estimates that previously used Winter’s Alicki–Fannes bound can replace the factor 2δ log d + (1+δ)h₂(δ/(1+δ)) by the tighter modulus g_{d²}(δ).","When both states have Schmidt number at most s the sharp modulus shrinks exactly to g_{d s}(δ), interpolating between the classical (s=1) and fully quantum regimes.","Fixed-marginal refinements are controlled by the conditional min-entropy of the reference state alone via the scalar κ_σ = d exp(−H_min(A|B)_{σ|σ}).","The same comparison-state technique is proposed as a route to sharp moduli for quantum mutual information and conditional mutual information."],"fun_headline_variants":["Sharp continuity bound for quantum conditional entropy pinned down","Optimal modulus: h₂(δ)+δ log(d²-1) for quantum conditional entropy","Quantum conditional entropy continuity tight when dim B≥d","Uniform continuity of quantum conditional entropy fully settled","Trace-distance continuity of H(A|B) now sharp for all δ"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The universal pinching inequality that any positive operator on A⊗B is at most d times the identity on A tensored with its B-marginal, which is what forces the comparison constant to be d²−1.","fun_headline_variants_meta":{"raw":{"variants":["Sharp continuity bound for quantum conditional entropy pinned down","Optimal modulus: h₂(δ)+δ log(d²-1) for quantum conditional entropy","Quantum conditional entropy continuity tight when dim B≥d","Uniform continuity of quantum conditional entropy fully settled","Trace-distance continuity of H(A|B) now sharp for all δ"]},"model":"grok-4.5","effort":"low","cost_usd":0.004265,"raw_usage":{"total_tokens":1268,"prompt_tokens":727,"num_sources_used":0,"completion_tokens":72,"cost_in_usd_ticks":42648000,"prompt_tokens_details":{"text_tokens":727,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":469,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":727,"tokens_out":72,"duration_ms":7200,"temperature":1.0,"reasoning_tokens":469,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T07:50:37.911023+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Exhibit any pair of states on systems of dimensions d and m ≥ d whose trace distance is at most some δ ≤ 1−d⁻² yet whose conditional-entropy difference strictly exceeds h₂(δ) + δ log(d²−1), or show that the maximally-entangled construction fails to attain equality.","supporting_citations":[],"review_version":1}