{"id":"db071770-2d28-4023-9a0a-f455a91bbdb8","arxiv_id":"2507.05696","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Regularization of the Umegaki relative entropy over polar-closed convex families is unnecessary if and only if a single-copy optimizer satisfies a phase-averaged supremum condition equal to 1.","lead":"This paper proves that, for a large class of quantum optimization problems, the hard many-copy limit equals the easy single-copy answer exactly when a single-copy optimizer passes a simple algebraic test. That test decides when key error exponents in quantum hypothesis testing and certain entanglement or magic measures become computable.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 10's sufficiency rests on the generalized Stein lemma imported from the unrefereed preprint [20]; if that lemma is unavailable or has hidden extra hypotheses, the single-copy criterion may cease to be sufficient.","rationale":"The reader's verdict CONDITIONAL is appropriate. I checked the internal logic of Theorem 7 and Theorem 10 and found no contradiction: the two directions use the optimizer characterization of Lemma 6, the submultiplicativity of support functions from Lemma 1, and the Fekete-style lower bound regD ≤ D/n exactly as written. The use of Assumption (1.C) is explicit, and the paper honestly states that SEP/PPT fail it, so this is a stated scope limitation rather than a hidden assumption. The delayed-additivity qubit example is indeed supported by numerics rather than a proof of the threshold, but it is not needed for the central iff. The most load-bearing unresolved point is the external generalized Stein lemma (Theorem 5), on which Proposition 9 and therefore the sufficiency direction of the main theorem depend. This is a verification gap: the manuscript gives no proof and cites only an unrefereed preprint. Since the argument is otherwise coherent and the gap is addressable by checking published literature or producing a direct proof, the reader's CONDITIONAL verdict should remain unchanged.","tokens_in":47690,"tokens_out":23750,"duration_ms":282689,"concrete_test":"Independently verify Theorem 5 under Assumption 1. Concretely: (a) check whether the published generalized quantum Stein lemma of Lami [35] directly implies Theorem 5 for families satisfying Assumptions 1, 2, and 3; or (b) attempt a self-contained proof of Proposition 9's superadditivity using only Assumptions 4.B-4.C, Fekete subadditivity, and the Theorem 7 criterion, without invoking Theorem 5. If neither route works, search for a family satisfying Assumption 1 for which the Stein exponent is strictly below the regularized relative entropy; such a counterexample would invalidate the sufficiency direction of Theorem 10.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central iff in Theorem 10 is structurally sound conditional on two external inputs: Assumption 1.(1.C) and the generalized Stein lemma stated as Theorem 5. Assumption (1.C) is explicit and acknowledged to exclude SEP/PPT, so it is a scope restriction rather than a hidden flaw. A more load-bearing verification gap is Theorem 5 itself: the manuscript does not prove it, citing only [20, Theorem 33], an unrefereed preprint. Theorem 5 is used exactly once, but crucially: Proposition 9's superadditivity lower bound regD(⊗_j ρ_j) ≥ Σ_j regD(ρ_j) is obtained by identifying regD with the Stein exponent and then exploiting multiplicativity of the support functions under Assumption 4.C. Without that lower bound, the proof of Theorem 10's sufficiency has no route from the single-copy condition (71) to the inequality D(⊗_j ρ_j^{⊗m_j}) ≥ Σ_j m_j D(ρ_j); ordinary subadditivity gives only the reverse inequality. Thus the entire 'if' direction of the central characterization inherits the correctness of [20]. If [20]'s assumptions are insufficient, or its proof contains a gap, the single-copy criterion could be a necessary but not sufficient condition. This is a missing proof in the manuscript, not an internal contradiction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the minimization of the Umegaki relative entropy, and of the α-z-Rényi divergences, over convex compact families {F_n} of positive operators on n-fold tensor products, subject to Assumption 1: convexity/compactness, closure under tensor products, and closure of the polar sets under tensor products. The central result is a single-copy criterion for additivity: weak additivity D(ρ^{⊗n}) = n D(ρ∥σ_0) for all n holds if and only if sup_{t∈R, τ∈F_1} Tr[τ σ_0^{-(1+it)/2} ρ σ_0^{-(1-it)/2}] = 1, where σ_0 is a single-copy optimizer (Theorem 7); a multi-state version is given in Theorem 10. The proofs are built on a Fréchet-derivative optimality condition (Lemma 6), integral representations for the logarithm and power functions (Lemmas 6 and 11), and multiplicativity of support functions derived from the polar-set assumption (Lemma 1). The criterion is then applied to the Stein, Chernoff, and Hoeffding exponents, with partial results for the strong converse exponent, and illustrated on arbitrarily varying sources, the Rains set, and non-positive-mana sets.","tokens_in":47874,"tokens_out":19836,"duration_ms":239126,"significance":"If the main results are accepted, this is a conceptually valuable contribution: it reduces the question of whether regularization is needed in a broad class of generalized hypothesis-testing and resource-theory problems to a property of a single-copy optimizer. The paper also provides new additive monotones for the Rains and mana-based resource theories, and explicit qubit examples where additivity holds for an arbitrarily large finite number of copies and then fails. The internal arguments are mostly careful: Lemma 6 and Appendix A address support degeneracies in the derivative condition; Lemma 11 gives a self-contained derivation of the needed integral representation; and the tensor-test/Fekete arguments in Proposition 9 and Theorem 22 are clean. The main caveat is that the sufficiency of the central characterization inherits the status of the generalized Stein lemma from the unreviewed preprint [20].","major_comments":[{"comment":"The sufficiency direction of the central characterization depends on Theorem 5, the generalized Stein lemma quoted from [20, Theorem 33], and no proof is given in this manuscript. Theorem 5 is used exactly once but crucially in the proof of Proposition 9 to obtain the superadditivity regD(⊗_j ρ_j) ≥ Σ_j regD(ρ_j); without that lower bound, the chain D(⊗_j ρ_j^{⊗m_j}) ≥ regD(...) = Σ_j m_j regD(ρ_j) = Σ_j m_j D(ρ_j) in Theorem 10 has no route from the single-copy condition (71). Please either include a self-contained proof of Theorem 5, or of the needed superadditivity, or explicitly state Theorem 10 and Corollary 8 as conditional on [20, Theorem 33]. In the same comment, please confirm that [20, Theorem 33] applies under the normalization conventions used here, since the Rains and mana families contain unnormalized positive operators and the polar set is intersected with the positive semidefinite cone.","section":"III.C, Proposition 9 and Theorem 10"},{"comment":"The criterion (52)/(71) is called a single-copy criterion, but its verification involves a supremum over a continuum of parameters t∈R and over the whole set F_1. No algorithm or complexity bound is given for this optimization, and the problem is not obviously convex in (t,τ). The claim in the abstract that the criterion 'opens the door to an efficiently computable characterization' should therefore be qualified: the criterion removes the many-copy regularization, but it is not shown to be efficiently decidable except in the symmetric or commuting cases where the supremum over t simplifies.","section":"I and III.A, statement of the single-copy criterion"}],"minor_comments":[{"comment":"The notation D(ρ) is used for both the Umegaki relative entropy D(ρ∥σ) and its minimization over F; please introduce a distinct notation, such as D_F(ρ), to avoid confusion in statements like (2) and (15).","section":"I and II.A"},{"comment":"In the expression for f_p(n), the sums over k and ℓ appear to run from 1 to n, but the binomial expansion of (|−⟩⟨−|)^{⊗n} suggests the sums should run from 0 to n; please check the indexing and the normalization factor.","section":"III.D, Eq. (79)"},{"comment":"The notation s(ρ∥F) is used without being defined; earlier the Stein exponent is denoted s(ρ∥{F_n}_{n∈N}). Please use a consistent notation.","section":"III.B, Corollary 8"},{"comment":"The line labeled (161) in the proof of Lemma 21 is garbled: the 'standard minimax inequality' should be written as sup_α inf_σ φ(α,σ) ≤ inf_σ sup_α φ(α,σ), and the presentation should be corrected so that the chain of inequalities leading to equality is readable.","section":"V.A, Lemma 21"},{"comment":"The statement of Theorem 5 says 'A proof for this setting can be found in [20]', but the appendix does not restate the precise assumptions or the exact statement of [20, Theorem 33]. Since this theorem is load-bearing, at least a precise statement of the imported result should be included, even if the proof is not reproduced.","section":"II.D, Theorem 5"}],"recommendation":"major_revision","confidential_remarks":"The main technical risk is the dependence on the unreviewed preprint [20] for the generalized Stein lemma, which is needed for the sufficiency of Theorem 10 and for Proposition 9. I would ask the editor to require the authors either to include a proof of the needed statement or to mark those results as conditional. The rest of the proofs appear sound, and the paper is likely publishable once this dependence is resolved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Marcus—read this one. The headline: Beigi, Rubboli and Tomamichel give a single-copy criterion for when the minimized Umegaki relative entropy is additive, and it is a genuinely useful result. Theorem 7 and its multi-state extension Theorem 10 say that additivity holds iff sup_{t,τ} Tr[τ σ_0^{-(1+it)/2} ρ σ_0^{-(1-it)/2}] = 1, a condition on the single-copy optimizer. Theorem 14 extends this to α-z Rényi divergences, and the applications to Stein, Chernoff, and Hoeffding exponents are natural. The proofs of the central equivalence are in-paper: Lemma 6 (optimizer condition via Fréchet derivative and the integral representation) and Lemma 1 (support-function multiplicativity from the polar condition) are done carefully, and I checked the structure of the two directions—they are consistent. Proposition 9 (strong additivity of the regularized relative entropy) is clean given its inputs.\n\nThe honest scoping is a plus: the polar condition (1.C) deliberately excludes SEP and PPT, so the relative entropy of entanglement remains open, and they say so.\n\nNow the soft spots, in order of importance. First, the sufficiency direction of Theorem 10 inherits the generalized Stein lemma, stated as Theorem 5 and cited to the unrefereed preprint [20] (Fang–Fawzi–Fawzi). The manuscript does not prove it. This is load-bearing: Proposition 9's superadditivity uses Theorem 5 to identify the regularized relative entropy with the Stein exponent and then applies a tensor product test. Without that lower bound, the chain 'D ≥ regD = Σ m_j regD(ρ_j)' in Theorem 10's proof has no route. So if [20] has a gap or hidden hypotheses, the criterion could be necessary but not sufficient. That is a missing proof, not an internal contradiction, but it is the right thing for a referee to push on. The stress-test note is accurate.\n\nSecond, the delayed-violation qubit example is nice conceptually—additivity up to n copies and breaking beyond—but the threshold claim rests on the numerical plot in Fig. 1. The proof that additivity fails for some n (via the explicit phase t) is complete; the 'holds up to n0' part is not proven.\n\nThird, a minor point: the condition checks require finding the single-copy optimizer, which is itself nontrivial, though for SDP-representable sets it is accessible. Not a flaw, just context.\n\nWho should read this: anyone working on composite hypothesis testing or resource-theory regularization. It deserves a serious referee; I would engage despite the dependence on [20]. My recommendation: send it out, and ask the authors to either prove or properly import the Stein lemma statement and assumptions in a self-contained appendix.","headline":"Single-copy additivity criterion for optimized relative entropy is real and largely sound, but the sufficiency direction leans on an unrefereed Stein lemma and the delayed-violation example rests on numerics.","tokens_in":48493,"tokens_out":5212,"would_cite":true,"duration_ms":55163,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P68","94A17"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"A single-copy identity decides whether many-copy relative entropy is additive","keywords":["quantum relative entropy","additivity","regularized relative entropy","Stein exponent","generalized quantum hypothesis testing","Rains set","mana","Renyi divergences"],"falsifier":"Take any state in the Rains or mana family, compute a single-copy optimizer $\\sigma_0$ numerically, and evaluate $\\sup_{t\\in\\mathbb{R},\\tau\\in\\mathcal{F}_1} \\operatorname{Tr}[\\tau \\sigma_0^{-(1+it)/2}\\rho\\sigma_0^{-(1-it)/2}]$. The paper predicts additivity for all $n$ if and only if this quantity equals 1; a value strictly above 1 must produce a violation of additivity at some finite number of copies, and a value exactly 1 must not.","tokens_in":47408,"feed_emoji":"⚛️","tokens_out":8375,"duration_ms":89765,"temperature":0.7,"pith_summary":"Quantum information tasks are often governed by regularized relative entropies, limits over arbitrarily many copies that are hard to compute. This paper asks when the limit is unnecessary, i.e., when the many-copy minimum is just the single-copy minimum repeated. For a large class of alternative sets, it proves that additivity holds exactly when a single-copy optimizer satisfies the stationarity identity $\\sup_{t\\in\\mathbb{R},\\tau\\in\\mathcal{F}_1} \\operatorname{Tr}[\\tau \\sigma_0^{-(1+it)/2}\\rho \\sigma_0^{-(1-it)/2}] = 1$. Because the condition is checked on one copy, the question of whether regularization is needed becomes a finite optimization problem. This applies to generalized hypothesis testing and to resource-theoretic bounds for entanglement and magic, and it yields strong additivity of the regularized relative entropy for those families.","feed_headline":"A single-copy test decides when relative entropy is additive","feed_subtitle":"If a one-copy optimizer satisfies one sup condition, no many-copy limit is needed for Stein, Rains, and magic bounds.","key_machinery":"The load-bearing mechanism has two parts. First, the first-order condition for a convex minimizer: $\\sigma_0$ minimizes $D(\\rho\\|\\cdot)$ over $\\mathcal{F}$ if and only if $\\sup_{\\tau\\in\\mathcal{F}} \\operatorname{Tr}[\\tau \\Xi(\\rho,\\sigma_0)] = 1$, where $\\Xi(\\rho,\\sigma)=\\int_{\\mathbb{R}} \\sigma^{-(1+it)/2}\\rho\\,\\sigma^{-(1-it)/2}\\beta_0(t)\\,dt$ is obtained from the Frechet derivative of the logarithm (Lemma 6). Second, the polar-set condition $\\mathcal{F}_m^{\\circ}\\otimes \\mathcal{F}_n^{\\circ}\\subseteq \\mathcal{F}_{m+n}^{\\circ}$, which Lemma 1 shows is equivalent to submultiplicativity of the support functions $h_{\\mathcal{F}}(X)=\\sup_{\\sigma\\in\\mathcal{F}}\\operatorname{Tr}(\\sigma X)$. That multiplicativity is what turns tensor-product directions into products of single-copy suprema, driving the sufficiency direction of the main theorem.","core_discovery":"The paper establishes a single-copy criterion for additivity of the minimized Umegaki relative entropy. For a family $\\{\\mathcal{F}^{(n)}\\}_{n\\in\\mathbb{N}}$ of convex compact sets of positive operators whose polars are closed under tensor products (Assumptions 1 and 4), Theorem 10 shows that $D(\\bigotimes_{j=1}^k \\rho_j^{\\otimes m_j}) = \\sum_{j=1}^k m_j D(\\rho_j)$ for all integers $m_j$ if and only if, for every $j$, the single-copy optimizer $\\sigma_{0,j}\\in\\arg\\min_{\\sigma\\in\\mathcal{F}^{(1)}} D(\\rho_j\\|\\sigma)$ satisfies $\\sup_{t\\in\\mathbb{R},\\tau\\in\\mathcal{F}^{(1)}} \\operatorname{Tr}[\\tau \\sigma_{0,j}^{-(1+it)/2} \\rho_j \\sigma_{0,j}^{-(1-it)/2}]=1$. If the identity fails, additivity fails for some number of copies. Corollary 8 translates this into the Stein exponent: the generalized Stein exponent equals the one-copy value $D(\\rho\\|\\sigma_0)$ if and only if the same condition holds. The paper extends the criterion to Petz and sandwiched Renyi divergences, and to Chernoff and Hoeffding exponents with the condition evaluated at a single-copy saddle point.","pith_inferences":["This suggests a practical testbed for resource theories with semidefinite-representable sets: compute the single-copy optimizer once, evaluate the supremum, and classify additivity without multi-copy computation.","Because separable and PPT sets violate the polar closure condition, the long-open additivity of the regularized relative entropy of entanglement remains untouched; a polar-preserving approximation of those sets is what would settle it.","The same single-copy stationarity pattern may extend to other convex resource monotones beyond relative entropies, wherever the optimizer admits an integral representation for its derivative."],"forward_implications":["For any family satisfying the polar-tensor-product assumption, deciding whether regularization is needed becomes a finite single-copy computation.","For the Rains set and the subnormalized non-positive-mana set, the regularized relative entropy is strongly additive, giving new additive monotones that bound entanglement and magic distillation.","For Werner, isotropic, and noisy strange states, the optimizer commutes with the state, so additivity holds and the regularized value equals the single-copy value.","The Stein exponent is single-letter exactly when the single-copy optimizer condition holds, and the Chernoff and Hoeffding exponents have analogous saddle-point additivity criteria.","There are qubit arbitrarily-varying-source examples in which additivity holds for any prescribed finite number of copies but fails later, so finite-copy checks cannot certify additivity."],"supporting_citations":[{"why":"Introduces the polar-set closure condition (1.C) and proves the generalized Stein lemma identifying the Stein exponent with the regularized relative entropy.","marker":"[20]"},{"why":"Derives necessary and sufficient conditions for optimizers of Renyi relative entropies that the paper extends with integral representations.","marker":"[55]"},{"why":"Provides the integral representation of the Frechet derivative of the logarithm used to write the operator $\\Xi(\\rho,\\sigma)$.","marker":"[57]"},{"why":"Gives the first-order convex optimality condition for the minimized relative entropy used in Lemma 6.","marker":"[24]"},{"why":"Establishes the quantum Chernoff bound for simple hypotheses, the baseline that the generalized Chernoff exponent extends.","marker":"[3]"},{"why":"Supplies the composite-hypothesis error-exponent framework, including the strong-converse and Hoeffding anti-divergence expressions used in Sections VI and VII.","marker":"[43]"},{"why":"Shows non-additivity of the relative entropy of entanglement for symmetric states, providing the contrast addressed by the Rains-set results.","marker":"[63]"},{"why":"Introduces the subnormalized non-positive-mana set used as an example satisfying the polar condition and bounding magic state distillation.","marker":"[65]"}],"fun_headline_variants":["Single-copy condition settles relative entropy additivity","One-copy optimizers decide when regularization is needless","Additivity of Umegaki entropy hinges on one-copy sup","No many-copy limit if one-copy optimizer satisfies sup=1","Relative entropy additivity decided by a single-copy test"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument collapses if the allowed alternative sets do not have polar sets that are closed under tensor products (equivalently, if their support functions are not submultiplicative), which is why separable and PPT sets are outside the theorem.","fun_headline_variants_meta":{"raw":{"variants":["Single-copy condition settles relative entropy additivity","One-copy optimizers decide when regularization is needless","Additivity of Umegaki entropy hinges on one-copy sup","No many-copy limit if one-copy optimizer satisfies sup=1","Relative entropy additivity decided by a single-copy test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000579,"raw_usage":{"total_tokens":2783,"prompt_tokens":1051,"completion_tokens":1732,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":1648}},"tokens_in":667,"tokens_out":1732,"duration_ms":12921,"temperature":1.0,"reasoning_tokens":1648,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:23:19.875202+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take any state in the Rains or mana family, compute a single-copy optimizer $\\sigma_0$ numerically, and evaluate $\\sup_{t\\in\\mathbb{R},\\tau\\in\\mathcal{F}_1} \\operatorname{Tr}[\\tau \\sigma_0^{-(1+it)/2}\\rho\\sigma_0^{-(1-it)/2}]$. The paper predicts additivity for all $n$ if and only if this quantity equals 1; a value strictly above 1 must produce a violation of additivity at some finite number of copies, and a value exactly 1 must not.","supporting_citations":[{"cited_title":"Quantum resource theories","cited_arxiv_id":null,"evidence_quote":"Introduces the polar-set closure condition (1.C) and proves the generalized Stein lemma identifying the Stein exponent with the regularized relative entropy."},{"cited_title":"New additivity properties of the relative entropy of entanglement and its generalizations","cited_arxiv_id":null,"evidence_quote":"Provides the integral representation of the Frechet derivative of the logarithm used to write the operator $\\Xi(\\rho,\\sigma)$."},{"cited_title":"Potential theoretic approach to rendezvous numbers","cited_arxiv_id":null,"evidence_quote":"Gives the first-order convex optimality condition for the minimized relative entropy used in Lemma 6."},{"cited_title":"τ σ − 1 2 1− 1−α z +it 0 χα,z(ρ, σ0)σ − 1 2 1− 1−α z −it 0 # =Q α,z(ρ∥σ0).(109) Proof.We derive the equivalence between (108) and sup t∈R,τ∈F 1 Tr","cited_arxiv_id":null,"evidence_quote":"Establishes the quantum Chernoff bound for simple hypotheses, the baseline that the generalized Chernoff exponent extends."},{"cited_title":"Quantum hypothesis testing and the operational interpretation of the quantum R´ enyi relative entropies","cited_arxiv_id":null,"evidence_quote":"Supplies the composite-hypothesis error-exponent framework, including the strong-converse and Hoeffding anti-divergence expressions used in Sections VI and VII."},{"cited_title":"The resource theory of stabilizer quantum computation","cited_arxiv_id":null,"evidence_quote":"Shows non-additivity of the relative entropy of entanglement for symmetric states, providing the contrast addressed by the Rains-set results."},{"cited_title":"Entanglement measures under symmetry","cited_arxiv_id":null,"evidence_quote":"Introduces the subnormalized non-positive-mana set used as an example satisfying the polar condition and bounding magic state distillation."}],"review_version":1}