{"id":"d0d44f2e-8550-4893-b282-0c0a8beb5fa7","arxiv_id":"2507.17435","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"Conditional gradient methods can certify multipartite entanglement heuristically and rigorously, with improved noise robustness bounds for Horodecki states.","lead":"This paper presents a unified toolbox based on conditional gradient optimization for detecting and certifying multipartite quantum entanglement in systems up to ten qubits. It combines fast heuristic detection with rigorous witness construction and tests it on bound entangled states and realistic noise channels.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Appendix D sign error invalidates Proposition 3: Cauchy–Schwarz gives tr(Λτ) ≥ −||Λ||, not the ≤ used to prove the Eq. (7) witness, so the rigorous-certification claim is unsupported.","rationale":"The reader's weakest assumption is exactly the load-bearing point. The proof of Proposition 3 is the only rigorous bridge from the ε-net search to a certified witness; if the inequality in Appendix D fails, the main advertised contribution of rigorous entanglement certification with numerical error control is unsupported. I re-derived the Cauchy–Schwarz step: with the Hilbert–Schmidt inner product, |tr(Λτ)| ≤ ||Λ||·||τ|| ≤ ||Λ||, so tr(Λτ) is bounded below by −||Λ||, not above. The proof's final inequality has the reverse sign, so the conclusion does not follow. I also note a normalization mismatch between Eq. (5) and Eq. (7), although the sign error alone is sufficient. I am not objecting to the heuristic detection or the numerical benchmarks; those are useful, and the public code is a real asset. But a paper whose central claim is a unified toolbox with rigorous certification cannot support that claim with a witness whose only proof is invalid. The concrete test above would settle whether the witness can be salvaged by a corrected derivation or whether the numerical witness routine works only by accident. Because this agrees with the reader's verdict, no change is needed.","tokens_in":14520,"tokens_out":8424,"duration_ms":89508,"concrete_test":"Take a 2-qubit example, e.g. ρ = 0.95|Φ+⟩⟨Φ+| + 0.05·I/4, with a closest separable σ obtained from the CG output. Compute Λ = σ−ρ, build a coarse ε-net S_ε^k as in Appendix C, set β = min_{φ∈S_ε} tr(Λφ), and form W from Eq. (7). Evaluate tr(Wτ) for τ = σ, for all computational-basis product states, and for a product state aligned with the most positive eigenvector of Λ. If any separable τ gives tr(Wτ)<0, Proposition 3 is false. As a direct check, print (1−η)tr(Λτ) and −(1−η)||Λ|| for these τ; the former should be ≥ the latter, contradicting Appendix D's use of ≤.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline claim includes 'rigorous witness construction with numerical error control.' That claim rests on Proposition 3 / Eq. (7). In Appendix D, for τ∈S_k the proof writes tr(Λτ) = tr(Λητ)+(1−η)tr(Λτ) ≤ β+(1−η)tr(Λτ) ≤ β−(1−η)||Λ||, where the last step is justified by 'Cauchy–Schwarz.' This is the wrong direction: |tr(Λτ)| ≤ ||Λ|| implies tr(Λτ) ≥ −||Λ||, not ≤ −||Λ||. For example, any τ with positive expectation on Λ, including many product states and sometimes τ=σ itself, violates the claimed inequality. Consequently the proof does not establish tr(Wτ)≥0 for all separable τ, and W in Eq. (7) is not shown to be an entanglement witness. A corrected argument would also need to reconcile the coefficient (β−ϵ)/||Λ|| in Eq. (7) with the unnormalized Λ used in the witness construction around Eq. (5). The heuristic detection part and the numerical benchmarks may still be useful, but the advertised rigorous component is not proven.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a conditional-gradient (CG) framework for multipartite entanglement certification. It combines a heuristic detector based on the dual-gap efficiency r_t with a purported rigorous entanglement witness built from an epsilon-net of separable states, and it reports fast detection in systems of up to ten qubits, near-optimal white-noise robustness thresholds for Horodecki bound entangled states, and robustness under several quantum noise channels. The central advertised contributions are thus a scalable heuristic and a rigorous certification subroutine with controlled numerical error.","tokens_in":1596,"tokens_out":1680,"duration_ms":128786,"significance":"If the rigorous components were correct, the framework would be a genuinely useful advance: it targets arbitrary k-separability structures, scales to ten qubits, provides an open-source Julia implementation, and benchmarks against analytical and SDP-based thresholds. The heuristic detection and numerical benchmarks may indeed be useful. However, the rigorous entanglement witness is the paper's core differentiator, and its proof in Appendix D contains a sign error that cannot be repaired by a local correction. The heuristic criterion's 'if and only if' statement is also false in general. The advertised rigorous certification claim is therefore unsupported, although the numerical/algorithmic exploration could form the basis of a revised manuscript.","major_comments":[{"comment":"The proof of Proposition 3 rests on the inequality (1−η)tr(Λτ) ≤ −(1−η)||Λ||, justified by 'Cauchy-Schwarz'. That is the wrong direction: Cauchy–Schwarz gives tr(Λτ) ≥ −||Λ|| for trace-one states, not the displayed upper bound. The asserted inequality fails for any separable τ with tr(Λτ) > 0, for example τ = σ when tr(σ²) > tr(ρσ). Consequently the chain in Eq. (D3) does not establish tr(Wτ) ≥ 0 for all τ ∈ S_k, and Proposition 3 is unproved. This is load-bearing because the 'rigorous witness construction' advertised in the abstract and conclusion depends entirely on this argument.","section":"Appendix D, Eq. (D3)"},{"comment":"Even if the positivity proof were repaired, the statement of Proposition 3 does not prove that W is a witness for ρ. An entanglement witness must also satisfy tr(Wρ) < 0. The proposition imposes no condition on the separable state σ, and the expression tr(Wρ) = tr(Λρ) − (β − ε) can be nonnegative for generic σ. The proof in Appendix D only addresses the separable-side inequality; the detection condition is never shown. Thus the operator in Eq. (7) is not demonstrated to be an entanglement witness for the target state ρ.","section":"Proposition 3 / Eq. (7)"},{"comment":"The statement 'A quantum state ρ is entangled if and only if the inequality f(ρ,σ_t) > g_t holds' is false for a fixed iteration t. Sufficiency follows from Proposition 1 by taking τ = ψ_t, but necessity would require that the current CG iterate σ_t, obtained with a heuristic LMO, satisfies the inequality for every entangled ρ. This is not guaranteed, and the text later concedes that 'there is no guarantee of global optimality' and that the computed quantities 'cannot be directly used to certify entanglement or separability'. Corollary 2 should be weakened to a sufficient detection criterion.","section":"Corollary 2"},{"comment":"The 'certified' entanglement and separability intervals reported in Fig. 3 and Fig. 4 rely on Proposition 3 and Proposition 4. Since Proposition 3 is unproved, those intervals are not rigorously certified as claimed. Proposition 4 and its proof in Appendix E also need clarification: the parameter ε is introduced without definition, the construction ρ_t = (1−ε)ρ − ε 11/d is not connected to the noisy state in Eq. (E1), and the step leading to the separable state at noise (p+ε)/(1+ε) should be stated with explicit hypotheses. These issues matter because the 'closing the gap' claim is a central numerical result.","section":"Applications for closing gap / Appendix E"}],"minor_comments":[{"comment":"The norm ||Λ|| is never defined; the proof implicitly uses the Hilbert–Schmidt norm in the Cauchy–Schwarz step, while Appendix C works with a general norm on the ambient space. Please specify the norm conventions consistently.","section":"Eq. (7) and Appendix D"},{"comment":"The phrase 'normalization omitted' is too terse: the witness formula depends on the trace convention and on whether σ and ρe are normalized or unnormalized operators. Please state the conventions explicitly.","section":"Eq. (5)"},{"comment":"The sentence 'A entanglement witness W is a necessary and sufficient criteria' contains a grammatical error ('A' should be 'An', and 'criteria' should be 'criterion').","section":"Introduction"},{"comment":"The dashed threshold line is labeled r = 1, but the text says the three-qubit GHZ state with 80% noise lies 'exactly on the boundary'; the figure's apparent crossing of the threshold by the 80% curve should be clarified, since the curve is used to motivate the detection criterion.","section":"Figure 2"},{"comment":"The entries in Table II report 'certified interval' error bars, but the certification mechanism is not described in Appendix F2; please explain how the reported upper and lower bounds are obtained from the CG iterations and which of them rely on the unproved witness construction.","section":"Appendix F2 / Table II"}],"recommendation":"reject","confidential_remarks":"The heuristic CG pipeline and the numerical benchmarks may be salvageable as a methods/software contribution, but the rigorous certification claim is invalid as written: the core witness proof has a sign error, the witness detection condition is not established, and the heuristic criterion is stated with a false equivalence. These are not local presentation issues. I recommend rejection, while noting that a substantially reworked construction (e.g., a corrected witness geometry and a precise proof of both witness inequalities) could justify a fresh submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is a mixed bag. The CG-based toolbox for multipartite entanglement detection is a genuinely useful idea: the dual-gap ratio r_t as an early stopping criterion is simple and works well numerically, and the authors demonstrate detection up to ten qubits with noisy GHZ and Dicke states. The explicit ε-net construction in Appendix C is also a real contribution, with better cardinality scaling than the standard packing bound. If all the paper did was present this as a heuristic toolbox, I'd be fairly positive.\n\nThe problem is the rigorous witness. Proposition 3 claims to construct a provable entanglement witness with numerical error control, and that is the advertised headline. The proof in Appendix D uses the inequality tr(Λτ) ≤ −||Λ||, purportedly from Cauchy–Schwarz. That is the wrong direction: Cauchy–Schwarz gives |tr(Λτ)| ≤ ||Λ||, so tr(Λτ) ≥ −||Λ||, not ≤. The step as written is false for most τ, and it is load-bearing: without it, the operator in Eq. (7) is not shown to be a witness. I checked the surrounding argument, and the error is not a typo that can be patched by flipping a sign; the needed lower bound on tr(Wτ) would require a different argument, and the authors do not provide one.\n\nThis doesn't kill the entire paper. The heuristic detection criterion is explicitly non-rigorous, and that part stands on its own as a numerical tool. The Horodecki-state thresholds and the noise-channel benchmarks are plausible, and the code is public. But the central claim — rigorous certification with controlled error — is unsupported. The \"closing the gap\" results lose their certified status until the proof is fixed.\n\nThere are also a few smaller issues: the definition of ε in Eq. (7) is confusing (it's set to (1−η)||Λ|| but also used as the net error), and the paper is thin on how the ε-net is actually used computationally in the benchmarks, so it's hard to tell how much of the reported tightness comes from the rigorous subroutine.\n\nIn sum: this deserves a serious referee, but the referee should reject the current version. The right move is major revision — fix the Appendix D proof, or demote the witness construction to a heuristic with a clear caveat. As it stands, I would not cite it for the rigor claim, and I'd be wary of the numerical intervals that depend on it.\n\nRecommendation: send to peer review, but the referee should go straight to Appendix D.","headline":"Useful heuristic toolbox, but the rigorous witness proof has a wrong-direction Cauchy-Schwarz bound that invalidates the main claim.","tokens_in":15301,"tokens_out":5305,"would_cite":false,"duration_ms":53359,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","90C25"],"pacs":["03.65.Ud","03.67.-a"],"model":"deepseek-v4-flash","headline":"A conditional-gradient toolbox reduces multipartite entanglement certification to a single convex optimization, certifying entanglement in up to ten qubits and nearly closing the white-noise gap for Horodecki bound entangled states.","keywords":["multipartite entanglement","entanglement witness","conditional gradient","Frank-Wolfe","bound entangled states","white noise robustness","quantum noise channels","separable states"],"falsifier":"Compute the operator W from Eq. (7) for a concrete target state and direction Λ, then search over a fine net of separable states for any τ with tr(Wτ) < 0; a single such τ would disprove the claim that W is always a valid witness.","tokens_in":14365,"feed_emoji":"⚛️","tokens_out":6537,"duration_ms":65010,"temperature":0.7,"pith_summary":"This paper argues that a single optimization method, the conditional gradient (Frank-Wolfe) algorithm, can serve both as a fast heuristic and as a rigorous certifier for multipartite entanglement. The method decides whether a noisy quantum state is separable or entangled by measuring a dual-gap efficiency ratio, and it constructs an explicit entanglement witness with controlled numerical error whenever a rigorous certificate is needed. The authors demonstrate that this toolbox certifies entanglement in systems of up to ten qubits, and that it sharply locates the white-noise robustness thresholds of a family of bound entangled states, closing most of the gap between the best known entanglement and separability bounds. If correct, this makes entanglement certification practical for larger multipartite systems and for realistic noise channels that semidefinite-programming methods cannot handle.","feed_headline":"One toolbox certifies entanglement up to ten qubits","feed_subtitle":"Fast heuristic detection plus rigorous witnesses narrows the gap to separability bounds for bound entangled states.","key_machinery":"The central object is the dual-gap efficiency r_t = g_t / f(ρ,σ_t), where g_t = tr[(σ_t − ρ)(σ_t − ψ_t)] is the Frank-Wolfe gap and f is half the squared Hilbert-Schmidt distance to the k-separable set. Corollary 2 states that r_t < 1 holds exactly when ρ is entangled, turning the optimization trajectory into an entanglement test. For rigorous certificates the machinery is an ε-net S_ε^k of pure separable boundary states, which gives a finite search with provable error bound, and the associated witness operator W = Λ − (β − ε)1/||Λ||. The same CG output feeds a geometric reconstruction (Proposition 4) that certifies separability, so the whole workflow shares one engine.","core_discovery":"The central discovery is that entanglement certification can be reduced to a single projection-free optimization problem: minimize the squared Hilbert-Schmidt distance to the convex set of k-separable states, and then read the answer off from two quantities produced by the same algorithm. For fast detection, the paper shows that a state is entangled exactly when the dual gap g_t exceeds the current distance f(ρ,σ_t), so the ratio r_t = g_t / f(ρ,σ_t) falling below 1 certifies entanglement; in practice, thresholds like r_t < 1/5 are used to guard against heuristic error. For rigorous certification, the algorithm searches an ε-net of pure separable states, bounds the error ε in the optimal distance, and constructs the operator W = Λ − (β − ε)1/||Λ||, which is claimed to be a valid entanglement witness for the target state. The paper uses this toolbox to compute, for Horodecki bound entangled states, white-noise thresholds that lie an order of magnitude closer to the previously known separability bounds than earlier entanglement bounds, and it gives certified robustness thresholds for a Bell state under several realistic quantum noise channels.","pith_inferences":["If the dual-gap efficiency r_t can be given a rigorous stopping rule with a finite iteration bound, the heuristic phase of the toolbox would become a certified detector as well; the paper only uses conservative thresholds without such a guarantee.","The ε-net construction described in the appendix for the unit sphere could be reused to certify membership in other convex sets of quantum states, such as sets defined by PPT or by restricted entanglement structures.","The observed correlation between entanglement strength and number of CG iterations suggests that iteration counts could serve as a cheap, quantitative proxy for white-noise robustness in larger systems."],"forward_implications":["Entanglement certification becomes practical for systems of ten or more qubits, where SDP-based methods currently break down; the paper demonstrates detection within dozens of iterations for 10-qubit GHZ and Dicke states at 70% white noise.","Rigorous witnesses with explicit numerical error control can be fed directly into experimental verification protocols, since the certified interval is stated with error bars.","The white-noise robustness gap for Horodecki states is reduced by roughly an order of magnitude, and the previously inconsistent bounds in the regime 0 < a < 0.02 are reconciled.","The same framework handles nonlinear quantum noise channels such as bit-flip, phase-flip, amplitude damping, and phase damping, giving certified thresholds that match known analytical values."],"supporting_citations":[{"why":"Defines the family of Horodecki bound entangled states used as the benchmark for the white-noise robustness analysis.","marker":"[4]"},{"why":"First proposed a CG-type algorithm for entanglement characterization, providing the starting point for the method.","marker":"[17]"},{"why":"Introduced the geometric reconstruction approach for separability certification that the paper extends in Proposition 4.","marker":"[18]"},{"why":"Establishes NP-hardness of the separability problem, motivating the projection-free CG approach.","marker":"[22]"},{"why":"The original Frank-Wolfe algorithm, the foundation of the conditional gradient method used throughout.","marker":"[24]"},{"why":"Review of conditional gradient methods that supplies the dual-gap convergence machinery used for the heuristic detection criterion.","marker":"[34]"},{"why":"Establishes the radius of the separable ball around the maximally mixed state, used in the separability certification of Proposition 4.","marker":"[48]"},{"why":"Provides the generalized Wootters formula, the previous entanglement bound that the paper's thresholds improve upon.","marker":"[49]"}],"fun_headline_variants":["Ten qubits, one algorithm: certified entanglement","One toolbox closes the entanglement-separability gap","Fast heuristic plus rigorous witness in one CG step","Entanglement certification scales to ten qubits with CG","Unified method certifies bound entangled states under noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"For the constructed operator W to be a valid entanglement witness, every separable state τ must satisfy a trace inequality that the paper derives in Appendix D using a step that does not follow from Cauchy-Schwarz and is not generally true.","fun_headline_variants_meta":{"raw":{"variants":["Ten qubits, one algorithm: certified entanglement","One toolbox closes the entanglement-separability gap","Fast heuristic plus rigorous witness in one CG step","Entanglement certification scales to ten qubits with CG","Unified method certifies bound entangled states under noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00035,"raw_usage":{"total_tokens":1872,"prompt_tokens":870,"completion_tokens":1002,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":486,"completion_tokens_details":{"reasoning_tokens":929}},"tokens_in":486,"tokens_out":1002,"duration_ms":10233,"temperature":1.0,"reasoning_tokens":929,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T14:48:19.359425+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the operator W from Eq. (7) for a concrete target state and direction Λ, then search over a fine net of separable states for any τ with tr(Wτ) < 0; a single such τ would disprove the claim that W is always a valid witness.","supporting_citations":[{"cited_title":"Horodecki, Separability criterion and inseparable mixed states with positive partial transposition, Phys","cited_arxiv_id":null,"evidence_quote":"Defines the family of Horodecki bound entangled states used as the benchmark for the white-noise robustness analysis."},{"cited_title":"Kampermann, O","cited_arxiv_id":null,"evidence_quote":"First proposed a CG-type algorithm for entanglement characterization, providing the starting point for the method."},{"cited_title":"Shang and O","cited_arxiv_id":null,"evidence_quote":"Introduced the geometric reconstruction approach for separability certification that the paper extends in Proposition 4."},{"cited_title":"STOC’03, STOC ’03 (Association for Computing Machinery, New York, NY, USA, 2003) pp","cited_arxiv_id":null,"evidence_quote":"Establishes NP-hardness of the separability problem, motivating the projection-free CG approach."},{"cited_title":"Frank and P","cited_arxiv_id":null,"evidence_quote":"The original Frank-Wolfe algorithm, the foundation of the conditional gradient method used throughout."},{"cited_title":"Gurvits and H","cited_arxiv_id":null,"evidence_quote":"Establishes the radius of the separable ball around the maximally mixed state, used in the separability certification of Proposition 4."},{"cited_title":"Chen, Z.-H","cited_arxiv_id":null,"evidence_quote":"Provides the generalized Wootters formula, the previous entanglement bound that the paper's thresholds improve upon."}],"review_version":1}