{"id":"fcd710cb-a38d-4df7-a8ec-dcf8d75dd5c9","arxiv_id":"2601.00383","paper_version":3,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Probabilistic entanglement distillation under approximate (dually) non-entangling instruments has error exponent equal to the regularized Hilbert projective metric to separable states.","lead":"This paper claims a formula for the error exponent of probabilistic entanglement distillation under approximate non-entangling operations: the exponent equals the regularized Hilbert projective metric to separable states. The proof connects probabilistic distillation to postselected hypothesis testing, but several load-bearing lemmas are incomplete or garbled.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2's converse bound is false: for m=2, δ=0, the identity DNE channel gives ratio 2/3, while the printed bound evaluates to 1/3 (or 1), invalidating the postselected-test parameter.","rationale":"The reader's verdict is REJECT, and I agree that the current manuscript does not establish its central claims. However, the specific weakest assumption identified by the reader — the m²−1 normalization in Lemma 3 — appears to be a false alarm: the separability condition p≤q/(m−1) is correct with τ_m normalized to trace 1. The load-bearing defect I find is different and more concrete: Theorem 2's converse relies on a quantitative bound that is either false (for δ=0, m=2, the identity channel yields 2/3, while the bound with 2δ−1 gives 1/3) or vacuous (with 2^δ−1, the error parameter becomes 1, making the postselected testing exponent infinite and the converse useless). Because Theorem 2 is one of the two central exponent characterizations, this invalidates the proof as written. The underlying formulas may still be correct and fixable, but the present derivation does not justify them. Hence the reader's REJECT verdict should stand, without endorsing the specific normalization objection.","tokens_in":23592,"tokens_out":35881,"duration_ms":340804,"concrete_test":"Set m=2, δ=0, and take the dually nonentangling subchannel E1=id on C²⊗C². Define Ψ=|Φ+⟩⟨Φ+|, τ=(I−Ψ)/3, M2=(I+2Ψ)/3, M1=2(I−Ψ)/3, so M1+M2=I. For σ=|00⟩⟨00|, compute tr(M2σ)=1/2 and tr((M1+M2)σ)=1, giving conditional error 1/2? Wait recompute: tr(M2σ)=tr((I+2Ψ)/3 · |00⟩⟨00|) = (1 + 2·(1/2))/3 = 2/3? Actually tr(Ψ|00⟩⟨00|)=1/2, so tr((I+2Ψ)/3 σ)= (1 + 2/2)/3 = 2/3. Thus ratio=2/3. Compare with the paper's bound: if 2δ−1=−1, bound=2/2+(2/3)(−1)=1/3, false; if 2^δ−1=0, bound=1, so ϵ=1. Run this evaluation for any m=2 DNE_0 protocol to confirm the claimed inequality fails or becomes vacuous.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The core bridge in Theorem 2 is the converse step (Eq. 27) that converts an arbitrary DNE_δ distillation protocol into a feasible postselected test with separable measurements. For a twirled subchannel E1(X)=tr(MX)Ψ_m+tr(NX)τ_m, the proof sets M2=E1†((I+mΨ_m)/(m+1)) and M1=E1†(m/(m+1)(I−Ψ_m)), and claims that for every separable σ,\n\ntrM2σ / tr((M1+M2)σ) ≤ 2/m + m/(m+1) min[(2δ−1), 2].\n\nThis inequality is quantitatively wrong. Take m=2, δ=0, and E1=id, which is dually nonentangling. Then M2=(I+2Ψ)/3 and M1=2(I−Ψ)/3, so M1+M2=I. For the separable state σ=|00⟩⟨00|, f=tr(σΨ)=1/2, and\n\ntrM2σ / tr((M1+M2)σ) = (1+2f)/3 = 2/3.\n\nThe printed bound with δ=0 gives either 1/3 (if '2δ−1' is read literally) — false — or 1 (if '2^δ−1' was intended) — so the β parameter is ϵ=1, which makes β_SEP_1 infinite and the converse vacuous. Since the argument depends on this bound to identify a valid postselected test with ϵ<1, Theorem 2 is not established as written.\n\nBy contrast, I do not find the reader's specific m²−1 normalization objection to Lemma 3 decisive: τ_m=(I−Ψ_m)/(m²−1) has trace 1, and the condition p≤q/(m−1) for separability of isotropic substates is correct. The genuine obstruction is the unjustified and incorrect bound in Theorem 2's converse.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies probabilistic (postselected) entanglement distillation and dilution under \\delta-approximately non-entangling (NE_\\delta) and dually non-entangling (DNE_\\delta) instruments. The central claims are Theorem 1, asserting that the asymptotic NE_\\delta distillation error exponent equals \\hat{D}^{\\mathrm{reg}}_{\\Omega,\\mathrm{Sep}}(\\rho), and Theorem 2, asserting that the DNE_\\delta exponent equals \\hat{D}^{\\mathrm{reg},\\mathrm{SEP}}_{\\Omega,\\mathrm{Sep}}(\\rho), a regularized Hilbert projective metric to separable states restricted to separable measurements. The proofs proceed by converting distillation protocols into three-outcome postselected hypothesis tests and by using a closed form for the reversed composite testing exponent. Additional results compare probabilistic entanglement costs under NE and DNE operations and give a lower bound on the non-entangling cost.","tokens_in":24135,"tokens_out":31452,"duration_ms":301072,"significance":"If correct, these would be the first exact asymptotic error exponents for probabilistic entanglement distillation under these operationally motivated relaxations, complementing prior work on postselected testing and non-entangling operations. A notable strength is that the results are derived from operational definitions with no fitted parameters, and Example 1 gives an explicit closed-form value for Werner states. The claimed \\delta- and m-independence of the exponents is a strong, falsifiable statement. However, several load-bearing proofs in the current manuscript are incomplete or contain incorrect inequalities, so the central claims are not yet supported as written.","major_comments":[{"comment":"The displayed bound in the converse part of Theorem 2, trM2σ / tr(M1+M2)σ ≤ 2/m + m/(m+1) min[(2δ−1),2], is quantitatively false. Take m=2, δ=0, and E1 to be the twirling channel, which is dually non-entangling. For σ=|00⟩⟨00|, one has M2=(I+2Ψ)/3, M1=2(I−Ψ)/3, M1+M2=I, and the left side equals (1+2⟨σ|Ψ|σ⟩)/3 = 2/3. The printed right side is 1/3 if '2δ−1' is read literally, and 1 if '2^δ−1' was intended; in the latter case the postselected-test parameter ε becomes 1, making the β_SEP_1 bound vacuous. Thus the upper bound (29) does not follow and Theorem 2 is not established. A corrected bound can be derived from the NE_δ condition alone, namely trM2σ/tr(M1+M2)σ ≤ 1/(m+1) + m 2^δ / [(m+1)(2^δ + m −1)], but the proof must be rewritten with this or an equivalent argument.","section":"Appendix VI E, Eq. (27), Theorem 2 proof"},{"comment":"Lemma 3 is the bridge used in both Theorems 1 and 2, but its proof is a sketch rather than a derivation. For Λ_{M,N}(X)=tr(MX)Ψ_m + tr(NX)τ_m, the claimed equivalence with sup_{X∈Sep} tr(MX)/tr(NX) ≤ 2^ε/(m−1) requires showing that D_{Ω,Sep}(Λ(X)) = max(0, log(tr(MX)(m−1)/tr(NX))) for isotropic substates. The proof instead states an inequality involving undefined p,q and uses '2ε' where an exponent is needed. The DNE part is also asserted with a one-line justification. Because every conversion between distillation protocols and postselected tests passes through this lemma, a complete proof is essential.","section":"Appendix VI D, Lemma 3"},{"comment":"The proof of Eq. (9), the closed form of the reversed composite postselected testing exponent, contains unjustified algebraic steps. After the substitution M′_2 = t/(1−t)M_2, the constraint involves an undefined variable t′; the step from the optimization over M_1,M′_2 to 1 + ε/(1−ε) trM_1σ/trM′_2σ is not derived; and the infimum over σ and the supremum over measurements are interchanged without justification. Since Corollary 3 is used to identify the distillation exponent with the regularized projective metric, this proof cannot remain as is. If the formula is meant to be imported from Ref. [29], that should be stated explicitly.","section":"Appendix C, Corollary 3"},{"comment":"In the meta-converse part of Theorem 1, Lemma 3 is applied to an arbitrary NE_δ subchannel E_i without first replacing it by its twirled version. Lemma 3 applies to subchannels of the form Λ_{M,N}(X)=tr(MX)Ψ_m + tr(NX)τ_m, and the required bound on tr(E_i(σ)Ψ_m)/tr(E_i(σ)) does not follow for a general E_i. One must use that T∘E_i has the same fidelity to Ψ_m and also lies in NE_δ. Lemma 5 supplies this reduction, but Theorem 1 does not cite it; the proof should state this step explicitly.","section":"Appendix VI E, Theorem 1 proof"}],"minor_comments":[{"comment":"The manuscript frequently writes '2δ' or '2ε' where an exponent '2^δ' or '2^ε' is clearly intended (e.g., Lemma 3, Theorem 2 proof). This must be corrected throughout.","section":"Definitions, Lemma 3"},{"comment":"The final limit in the proof of Theorem 2 writes lim_{m→∞} E^{NE_δ}_{d,err,p}(ρ), but it should be E^{DNE_δ}_{d,err,p}(ρ).","section":"Theorem 2, final line"},{"comment":"Lemma 5's statement says 'probabilistic distillation exponent' but then writes E^{(m),Fδ}_{c,err,p}; the subscript should be d,err,p.","section":"Lemma 5 statement"},{"comment":"In Example 1, 'As p 1/2 is separable' should read 'As ρ_{1/2} is separable'.","section":"Example 1 proof"},{"comment":"In the proof of Lemma 8, λ_min(σ^{⊗n}) is written as nλ_min(σ); this should be λ_min(σ)^n. This is in a technical appendix and not used in the main theorems, but it should be fixed.","section":"Appendix, Lemma 8"}],"recommendation":"major_revision","confidential_remarks":"The concrete false inequality in Theorem 2's converse is the main obstacle. The central characterization may well be correct and repairable by replacing the faulty bound with the NE_δ-derived bound and by completing the proofs of Lemma 3 and Corollary 3. However, as the manuscript stands, the central claim is not supported. I recommend major revision rather than rejection because the errors appear local and fixable within the paper's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — quick take: the claimed exponents are genuinely new and worth caring about, but only the NE case looks salvageable as written. The stress-test note is right: Theorem 2's converse is load-bearing and breaks on a simple example.\n\nWhat is new: an exact error exponent for probabilistic entanglement distillation under δ-approximately nonentangling instruments, and a separable-measurement-restricted version under dually nonentangling instruments. The high-level bridge from probabilistic distillation to postselected hypothesis testing is the right move, and the final formula is clean and plausible. The dilution/cost results, including the one-shot NE/DNE gap of one ebit, are also useful. There is no curve-fitting or invented data; everything is derived from operational definitions, and the self-citation is not load-bearing.\n\nWhat is soft: Theorem 2's converse uses the bound\n trM2σ / tr(M1+M2)σ ≤ 2/m + m/(m+1) · min[(2δ−1),2].\nThat bound is simply false. Take m=2, δ=0, and E1 = identity, which is dually nonentangling. For σ=|00⟩⟨00|, the ratio is 2/3, while the printed bound gives 1/3 (or 1 if 2^δ was intended). So the constructed M1,M2 need not be a feasible postselected test with ε<1, and the reverse inequality in Theorem 2 collapses. This is not a cosmetic issue: the whole DNE characterization depends on identifying that postselected-test parameter.\n\nThe reader's objection about the m²−1 normalization in Lemma 3 is not the real problem; the separability condition for isotropic substates is fine. Lemma 3 is still asserted rather than proved, and Corollary 3's derivation is compressed past the point of reliability — the minimax swap and the algebra leading to β = log[1 + ε/(1−ε)Ω] are not actually shown. The manuscript also systematically loses superscripts on 2^δ, which makes the notation untrustworthy even where the underlying claim may be true.\n\nWho this is for: people working on entanglement manipulation under LOCC relaxations will want to know these formulas even if the proofs need repair. The paper deserves a serious referee because the question is real and the approach is promising, but I would not accept it in this form. A referee should ask for a corrected converse for Theorem 2, a real proof of Lemma 3, and a cleaner derivation of Corollary 3. I would not cite the DNE formula until the proof is fixed.","headline":"Genuinely new and plausible formulas for probabilistic distillation exponents under NE/DNE operations, but Theorem 2's converse proof contains a concrete false inequality; the DNE result is not established as written.","tokens_in":24442,"tokens_out":4647,"would_cite":false,"duration_ms":48314,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45"],"pacs":["03.65.Ud","03.67.Mn"],"model":"deepseek-v4-flash","headline":"Probabilistic entanglement distillation under nearly free operations has exact error exponents given by a single projective-metric quantity.","keywords":["entanglement distillation","error exponents","probabilistic protocols","postselected hypothesis testing","non-entangling operations","dually non-entangling operations","Hilbert projective metric"],"falsifier":"Construct an explicit bipartite state ρ, a slack δ, and a dimension m for which either (a) a distillation protocol achieves an exponent strictly larger than D̂^{reg}_{Ω,Sep}(ρ), or (b) a postselected test with the required conditional errors cannot be converted into a feasible δ-approximately non-entangling subchannel. More directly, for a specific pair of positive operators M,N on a small bipartite system, compute sup_{X∈Sep} tr(MX)/tr(NX) and compare it with the true Hilbert projective metric D_{Ω,Sep}(Λ_{M,N}(σ)) over all separable σ; a mismatch disproves Lemma 3.","tokens_in":1502,"feed_emoji":"⚛️","tokens_out":2146,"duration_ms":69270,"temperature":0.7,"pith_summary":"The paper proves that probabilistic (postselected) entanglement distillation under operations that can generate only a small amount of entanglement has a closed-form asymptotic error exponent. The exponent equals the regularized Hilbert projective metric to the set of separable states, and under dually non-entangling operations it becomes the same metric restricted to separable measurements. The proof establishes an exact operational equivalence: any distillation protocol can be recast as a postselected hypothesis test against separable states, and any feasible test can be converted back into a distillation subchannel with matching error scaling. This matters because it reduces a complex operational question to a single analytic quantity and shows that small slack in the free-operation condition and arbitrarily large target dimension do not change the asymptotic rate.","feed_headline":"Distillation error exponent equals distance to separable states","feed_subtitle":"Probabilistic protocols under nearly free operations reduce to a postselected test; slack and target size do not matter.","key_machinery":"The central objects are three-outcome postselected quantum hypothesis tests (with outcomes 'ρ', 'σ', and 'abstain') and their reversed form, where the conditional type-II error is controlled against all separable states. The bridge to distillation is a twirled subchannel of the form Λ_{M,N}(X)=tr(MX)Ψ_m + tr(NX)τ_m, with τ_m=(I-Ψ_m)/(m²−1). Lemma 3 asserts that such a subchannel is δ-approximately non-entangling if and only if sup_{X∈Sep} tr(MX)/tr(NX) ≤ 2^δ/(m−1), and the dual version adds the cone(Sep) constraints. This lemma converts fidelity constraints of distillation into conditional-error constraints of postselected tests, and the exponent formula follows from the closed form of the r","core_discovery":"Theorems 1 and 2 give exact equalities. For any bipartite state ρ, under δ-approximately non-entangling instruments the asymptotic distillation error exponent is E_{d,err,p}^{NEδ}(ρ) = D̂^{reg}_{Ω,Sep}(ρ), for every finite target dimension m, independent of δ. Under δ-approximately dually non-entangling instruments, E_{d,err,p}^{DNEδ}(ρ) = D̂^{reg,SEP}_{Ω,Sep}(ρ), where the superscript SEP means the postselected test is restricted to separable measurements. The paper further shows that probabilistic entanglement costs under NE and DNE instruments coincide asymptotically, with at most a one-ebit gap in one shot, and gives a lower bound on the cost under NE instruments in terms of the regulari","pith_inferences":["Because the exponent is δ-independent, the same formula may survive other natural definitions of 'approximately free' operations, provided the threshold in the twirled subchannel characterization scales similarly.","The one-shot NE-versus-DNE cost gap of at most one ebit suggests the dual constraint is asymptotically negligible; a direct test would be to check whether distillation rates under NE and DNE also converge when δ is fixed and m grows.","The equivalence with postselected testing may generalize to other convex resource theories with a canonical maximally resourceful state, such as coherence or athermality—directions the paper itself lists as open.","Since the Hilbert projective metric admits a semidefinite programming formulation, the exponents are computable in principle for small states; numerical experiments on random states could reveal how quickly the regularized quantity converges to its limit."],"forward_implications":["For Werner states, the distillation error exponent is exactly max(0, log((1−p)/p)), with a sharp transition at p=1/2.","The NE and DNE exponents are ordered: because separable measurements form a subset of all measurements, the DNE exponent is never larger than the NE exponent.","The asymptotic exponent is independent of both the slack δ and the target dimension m; only the input state and the class of operations matter.","Probabilistic entanglement costs under NE and DNE operations are asymptotically equal, and the one-shot cost differs by at most one ebit.","The probabilistic cost under NE operations is at least the regularized max-relative entropy to separable states, providing a quantitative lower bound on dilution difficulty."],"fun_headline_variants":["Postselected tests make distillation error exponent exact","Error exponent for probabilistic distillation: independent of δ","Distillation error exponent: exact distance to separability","Exact error exponent for distillation via postselected hypothesis testing","Probabilistic distillation error exponent: no slack dependence"],"cache_read_input_tokens":25728,"weakest_assumption_plain":"The load-bearing premise is Lemma 3's characterization (Appendix VI D): a twirled subchannel Λ_{M,N} is δ-approximately non-entangling exactly when sup_{X∈Sep} tr(MX)/tr(NX) ≤ 2^δ/(m−1), with the dual version requiring N and (1/m)M+(1−1/m)N in cone(Sep); if this threshold condition is not exactly right, both exponent equalities collapse.","fun_headline_variants_meta":{"raw":{"variants":["Postselected tests make distillation error exponent exact","Error exponent for probabilistic distillation: independent of δ","Distillation error exponent: exact distance to separability","Exact error exponent for distillation via postselected hypothesis testing","Probabilistic distillation error exponent: no slack dependence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001233,"raw_usage":{"total_tokens":4874,"prompt_tokens":692,"completion_tokens":4182,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":4122}},"tokens_in":436,"tokens_out":4182,"duration_ms":28074,"temperature":1.0,"reasoning_tokens":4122,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T13:05:16.270671+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct an explicit bipartite state ρ, a slack δ, and a dimension m for which either (a) a distillation protocol achieves an exponent strictly larger than D̂^{reg}_{Ω,Sep}(ρ), or (b) a postselected test with the required conditional errors cannot be converted into a feasible δ-approximately non-entangling subchannel. More directly, for a specific pair of positive operators M,N on a small bipartite system, compute sup_{X∈Sep} tr(MX)/tr(NX) and compare it with the true Hilbert projective metric D_{Ω,Sep}(Λ_{M,N}(σ)) over all separable σ; a mismatch disproves Lemma 3.","supporting_citations":[],"review_version":1}