Classical capacities and complementary simulation for qubit Davies channels

Recommendation: Major Revision · Material objections: 1

The manuscript extends the thermal block entropy inequality to the qubit Davies family with additional dephasing, and derives from it a population-resolved bound (Lemma 5, Theorem 4), a Holevo capacity formula with additivity (Theorem 1), a strong convex-roof superadditivity (Theorem 6), a complementary-channel capacity formula (Theorem 9) and parallel simulation results (Theorem 10, Corollary 11, Proposition 12). I checked the block-matrix vertex decomposition and the population/coherence identification in Lemma 5 by hand, and ran random numerical tests of Theorem 4, Lemma 5, and the external Lemma 2 and Lemma 3 (no violation). Theorem 1 capacity values in Section 8 were reproduced numerically. The chain Lemma 5 to Theorem 1 additivity and Theorem 6 is logically consistent. The unproven step is the subadditivity (17) of channel formation, which the text only sketches, and it is needed for the equalities in Theorem 10, eq. (18) and Corollary 11. The Proposition 12 separation additionally needs the lower bound G_N(p) >= J(p) near p = 1/2, which is asserted but not written out. The bibliography also has unverified items (reference [3] not found; reference [2] found by title only), and the paper does not cite arXiv:2609.14621, which overlaps Theorem 1 on the GAD boundary. Scope: this is a review of this exact PDF only, based on my reading and on numerical spot checks, not on a formal verification or on the cited sources in full.

Correctness confidence

★★★☆☆Theorem 4 vertex decomposition (Sec. 2) and Lemma 5 block identification (Sec. 3) were checked by hand and by random numerical tests (no violation, residuals about 1e-15). Lemma 2 (external) and Lemma 3 were spot-checked numerically only. Eq. (17) and the equalities in Theorem 10, eq. (18) and Corollary 11 are not established by the text.

Rigor

★★★☆☆Theorem 1 achievability and converse, Theorem 6 and Theorem 9 follow from Lemma 5 and Lemma 2 with no gap found in the logic. Main gaps: sketched subadditivity (17), a garbled entropy chain in its proof, and an asserted endpoint-chord identity G=J near p=1/2 used in Proposition 12.

Novelty

★★☆☆☆The paper states it does not claim Davies dynamics, the unital case, or the Hellmund-Uhlmann flat-roof criterion. arXiv:2609.14621 (Fang, 2026) reports that the GAD classical capacity equals the one-copy Holevo quantity, which overlaps Theorem 1 on the GAD boundary and is not cited. Novelty is limited to the extension to the full Davies family and the complementary/simulation statements.

Significance

★★☆☆☆Results concern a single-qubit phase-covariant family. The capacity formula reduces to known cases (identity, constant channel). The parallel simulation results depend on the unproven step (17), so their significance is conditional.

Reproducibility

★★★☆☆The capacity table in Section 8 was reproduced to about 1e-9 on a grid (a=0.10, b=0.30 with lambda^2 = 0.60, 0.50, 0.10 and a=0.02, b=0.30 with lambda^2 = 0.514624). The Proposition 12 numbers (J(1/2) = 0.704192 bits, maximizer near 0.51695 with value 0.704713) and the Corollary 11 example (0.600874 bits) were reproduced by hand. The paper states that the scripts do not certify intervals. The OpenAI reference [3] could not be located, and the Tang et al. reference [2] was located only by title.

arr:assessment:83ca3376-8480-4bfd-885e-48dc4d10bf01 · prompt ARR-ASSESS-1.0 · response SHA-256 df75137c4729d0660d4addba10fea2d2de6a6e9de6255926a11ccf3df42d09a5 · canonical PDF SHA-256 a6382c9eb75c5b60b877987e168e4d303f9a12b93ed6159a0aaeb11f4d3c1e3e

Recommendation: Major Revision · Material objections: 1

The manuscript extends a recent triangular block-entropy inequality (Tang et al. 2026; OpenAI 2026) from amplitude damping and generalized amplitude damping on the boundary lambda^2=k to the whole qubit Davies region 0<=lambda^2<=k, via a third vertex obtained from strong subadditivity, and derives: a population-resolved output-entropy bound, strong convex-roof superadditivity, additivity of Holevo information and classical capacity C=chi for the channel and its complement with any partner, an explicit scalar convex-envelope formula, formation entropy in the Stinespring range, and parallel simulation cost of the complementary channel via a reverse-channel closure, plus a nonunital example where simulation cost strictly exceeds Choi-state cost. My hand check of the main derivations (Thm 4, Lemma 5, Thm 1, Prop 12 arithmetic) found them consistent. The central dependence on Lemma 2, a 2026 result I could not verify, and uncertified numerics are the principal reasons for a major-revision rating; no internal error was found. I did not execute code or read supplementary files.

Correctness confidence

★★★☆☆I rechecked by hand the three-vertex decomposition of Theorem 4 (vertex coefficients, weights, determinant u, Lemma 3 entropy g(u)), the Lemma 5 parameter identification (alpha delta - beta eta = k p(1-p) >= lambda^2 p(1-p)), the Theorem 1 converse/achievability logic, and the arithmetic of Proposition 12 (reversed parameters 1/4, 1/8, 5/8, 5/48 sum correctly; 1259712 > 823543). I found no error. However the whole result rests on Lemma 2, an external 2026 inequality that I could not verify; that caps my confidence.

Rigor

★★★★☆Proofs are explicit, boundary cases are treated (zero row mass, u=0, p=0,1, t=0,1), external theorems are stated precisely with their sources, and numerical results are labeled as floating-point illustrations. Some steps are terse (convex-roof concavity/continuity in Section 6, Carathéodory-type extremal-measure arguments in Prop. 8 and eq. (10)) and are asserted rather than fully worked.

Novelty

★★★★☆The extension from the amplitude-damping/pure-output boundary (lambda^2=k) to the full Davies region, strong convex-roof superadditivity, complement capacity additivity, reverse-channel closure inside the Davies family and the nonunital strict Choi-cost separation appear to be new relative to the cited work; the paper is honest that the key inequality is not its own. My web searches (two queries) located arXiv 2609.28592 but gave no evidence of a prior full-Davies additivity proof; I could not read the OpenAI note or the full text of 2609.28592, so priority is unchecked.

Significance

★★★★☆If correct, additivity of Holevo information and a single-letter classical capacity for all qubit Davies channels (including the complementary channel) would settle a long-standing class (generalized amplitude damping) and give a nontrivial simulation-cost result. The value is largely derivative of the Tang et al./OpenAI inequality, and the result is confined to one-qubit memoryless channels, so it is strong but not field-shaping.

Reproducibility

★★★☆☆Formulas are explicit and examples have exact parameters, but the scripts and supplement were not available to me, I ran no code, and the numerical tables (e.g., capacity 0.5039 bits, 4-signal optimizers) are unverified by me. The author states no interval certification was done.

arr:assessment:d03c051d-6563-4683-9e54-004de61da9b6 · prompt ARR-ASSESS-1.0 · response SHA-256 6bf0b4fb7689187a35943ffd0657b4ece76e2156cb08c63bcfc6945bfe38280c · canonical PDF SHA-256 a6382c9eb75c5b60b877987e168e4d303f9a12b93ed6159a0aaeb11f4d3c1e3e