{"id":"ea9dedac-00d2-487b-bba2-8bdffc988d37","arxiv_id":"2507.21787","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"For Haar-random high-dimensional NPT states, the paper proposes a rank- and dimension-dependent order among entanglement detection criteria, but the analytic rank thresholds it claims are only ensemble-average arguments and are contradicted by small nonzero detection fractions.","lead":"This paper tests four standard ways of detecting entanglement on thousands of random high-dimensional quantum states and ranks them by how often they work. It claims to prove when two of the methods stop working, but the proof relies on averaged states rather than individual ones, and its own simulations show the claimed cutoffs are not exact.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Props. 1 and 2 replace state-specific entropies and purities with Haar averages, yet conclude every state is undetected; the paper's own Table I shows nonzero detection above the claimed rank thresholds.","rationale":"I read the paper in good faith: the numerical study is extensive and the qualitative conclusion that entropy is weak and realignment degrades with rank is plausible. But the two analytic propositions are the advertised main results, and both fail at a logical level. The proof of Proposition 1 proves statements about averages and then switches to all states. The proof of Proposition 2 proves an upper bound using the average purity, but the bound is only useful if the purity of every state is at most the average (or if a concentration bound shows violations are negligible to the point of zero probability; no such bound is given or plausible since the purity distribution has positive variance). The internal contradiction with Tables I and II, where FE and FRl are nonzero above the claimed thresholds, is decisive. One might rescue the statements as typicality claims with explicit probability bounds and corrected thresholds, but the universal 'fail to detect' language is unsupported. Since the reader's weakest_assumption already targeted this same ensemble-average-to-every-state inference and their verdict is REJECT, my independent stress-test agrees: the central claim is not established as stated.","tokens_in":23631,"tokens_out":3630,"duration_ms":41655,"concrete_test":"Run the authors' own sampling (Haar random |psi_123>, trace out subsystem 3) for 2x5 at ranks k=6 and k=7 with at least 1e5 samples. For each sample compute (i) the two conditional entropies S12-S1 and S12-S2 and (ii) the realignment trace norm ||rho_Rl||_1. Count the fraction of states with S12-S1<0 or S12-S2<0 (entropy detection) and with ||rho_Rl||_1>1 (realignment detection). If either fraction is nonzero, Propositions 1 and 2 are false as stated. As a second check, record the maximum purity and maximum ||rho_Rl||_1 among samples; for k=7, if any sample has Tr rho^2 > 1/4, the average-purity substitution in Eq. (8) is invalid for a universal claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on two proofs that infer a universal statement from ensemble averages. In Proposition 1 (Eq. 3), the authors show the averaged conditional entropies satisfy <S12>-<S2> >= 0 and <S1> <= <S2> for k > max(d1,d2), and then conclude E 'fails to detect' for all states. But the entropic criterion is per state: entanglement is certified when S12-S1<0 or S12-S2<0. Nonnegativity of the averages does not preclude a subset of states with negative conditional entropy; without concentration or an extreme-value bound, the conclusion is a quantifier fallacy. The simulations themselves contradict the universal statement: Table I reports FE=0.054 at k=6 for 2x5, while Proposition 1 predicts zero for all k>5. In Proposition 2 (Eq. 8), the realignment trace norm is bounded via ||rho_Rl||_1 <= d1 sqrt(Tr rho^2), and the average purity (d1d2+k)/(d1d2k+1) is inserted as if it were the purity of each individual state. The inequality requires an upper bound on the actual purity of every state; the average being below 1/d1^2 says nothing about the maximum of the purity distribution. For 2x5 at k=7, the average purity is 17/71 ~ 0.239 < 1/4, but individual samples with purity above 1/4 would give ||rho_Rl||_1 > 1 and hence detection; Table I indeed shows FRl=0.021 at k=7, above the claimed cutoff k0=6.5. Thus the load-bearing step of both propositions is the unjustified replacement of a state-dependent quantity by its Haar expectation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes four entanglement detection criteria (majorization, reduction, realignment, and entropic) for Haar-random bipartite mixed states of fixed rank, benchmarking them against the partial transposition criterion. The headline contribution is a pair of analytic propositions (Propositions 1 and 2) claiming to establish lower bounds on the rank beyond which the entropic and realignment criteria fail to detect entanglement. The paper also reports extensive numerical simulations (5×10^5 states per parameter set), introducing normalized fractions, mean detectable entanglement, and minimum detectable entanglement as figures of merit, and uses these to propose a hierarchy of criteria in qubit-qudit and qudit-qudit systems. A separate proposition gives an independent proof of the equivalence of the partial transposition and reduction criteria in 2⊗d systems.","tokens_in":23970,"tokens_out":6651,"duration_ms":71796,"significance":"If the analytic propositions were correct, the paper would provide a useful and surprising limitation on two standard entanglement criteria. The numerical study is extensive and the three figures of merit are natural and clearly presented; the observed dimension- and rank-dependence of the criteria is informative. The paper also gives a clean proof of the known P–R_d equivalence for qubit-qudit systems. However, the central claims rest on the two propositions, and those proofs are not logically sound. The paper's own simulation tables contradict the universal reading of the propositions. The remaining numerical content is of interest, but it does not justify the headline theoretical statements.","major_comments":[{"comment":"The proof establishes inequalities only for Haar-averaged conditional entropies: ⟨S12⟩−⟨S2⟩ ≥ 0 and ⟨S1⟩ ≤ ⟨S2⟩ for k > d2. The entropic criterion, however, is state-dependent: a state is detected when S12−S1 < 0 or S12−S2 < 0. Nonnegativity of the averages does not preclude a subset of states with negative conditional entropy, and no concentration or extreme-value bound is supplied. The paper's own Table I reports FE = 0.054 at k = 6 for 2⊗5 (with max(d1,d2) = 5), and Remark 1 reports FE = 0.055 at k = 9 for 2⊗8, both above the claimed threshold. These numerical results directly contradict the universal claim in Proposition 1. The proposition is therefore unsupported as stated and would require either a proof of concentration or a reformulation as a statement about typical states.","section":"Sec. II A, Proposition 1 (Eqs. (3)–(6))"},{"comment":"The proof bounds ||ρ_Rl||_1 ≤ d1 sqrt(Tr ρ^2) and then substitutes the ensemble-average purity (d1d2+k)/(d1d2k+1) as if it were the purity of each individual state. The inequality used in the proof requires an upper bound on Tr ρ^2 that holds for every state in the ensemble; the average purity being below a threshold says nothing about the maximum of the purity distribution. For 2⊗5, the claimed cutoff is k0 = (d1^3 d2 − 1)/(d1(d2−d1)) = 6.5, so Proposition 2 predicts F_Rl = 0 for k ≥ 7. Table I instead reports F_Rl = 0.021 at k = 7. The proposed universal threshold is thus both unproven and contradicted by the authors' own numerics. Without a concentration bound or a per-state argument, the proposition cannot be accepted.","section":"Sec. II A, Proposition 2 (Eqs. (7)–(8))"},{"comment":"The abstract and Section II A present the central claim as 'we prove lower bounds on the rank of mixed quantum states beyond which the realignment and entropic criteria fail to detect entanglement.' This is a universal statement about all states of a given rank. The proofs in Propositions 1 and 2 are ensemble-average arguments and cannot support such a universal conclusion. Even if reinterpreted as average or typical-state statements, the simulations in Tables I and II show nonzero detection fractions above the claimed thresholds, so the quantitative predictions of the propositions are not consistent with the numerical evidence. The headline result of the paper is therefore not established by the material presented.","section":"Abstract and Sec. II A"}],"minor_comments":[{"comment":"The derivative of ⟨S12⟩−⟨S2⟩ is miscomputed: differentiating ⟨S2⟩ ≈ log d2 − d2/(2d1k) contributes +d2/(2d1k^2), so the numerator should be 2d1d2k − k^2 + d2^2 rather than 2d1d2k − k^2 − d2^2. The monotonicity conclusion is unaffected, but the printed formula is incorrect.","section":"Eq. (4)"},{"comment":"The definition of F_E is ambiguous: E^{(i)} is described as 'the number of states that can be detected by a fixed criterion E', but the formula requires an indicator variable (0 or 1) for each state i. Please clarify the notation, e.g., by writing the indicator as 1_E(i).","section":"Sec. II B 3, Eq. (9)"},{"comment":"The normalized fractions are reported without statistical uncertainties. Given that the hierarchy is based on small differences at moderate ranks (e.g., F_E vs F_Rl at k = 5 in 2⊗5), a confidence interval or a statement about the number of detected states would be helpful.","section":"Tables I and II"},{"comment":"The wording 'the entropy criterion fails to detect the entanglement of Haar uniformly generated states' is too strong and is inconsistent with Remark 1 and Table I, which report nonzero detection fractions above the claimed thresholds. If the intended statement is an average or typical-state statement, the wording should be changed accordingly.","section":"Sec. II A and Remark 1"}],"recommendation":"reject","confidential_remarks":"The numerical study is extensive and the paper contains a useful collection of observations about how entanglement criteria behave for random states. However, the two analytic propositions that form the paper's main claim are invalid as stated, and the paper's own tables contradict them. A revision could potentially turn the work into a purely numerical study of typical-state behavior, or add rigorous concentration bounds and reframe the propositions as typicality results, but this would be a substantial reworking. I found no indication of misconduct; the issue is a genuine logical gap in the proofs."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know up front.\n\nThe numerical study is genuinely useful and the empirical hierarchy it reports looks real: majorization beats realignment in 2×d systems, realignment overtakes majorization in higher dimensions at low rank, and detection fractions drop with rank. The paper is clearly written and the generation of rank-k random states via partial trace is handled properly.\n\nBut the two analytic propositions at the center of the title do not hold as stated. Proposition 1 shows that the Haar-averaged conditional entropies are non-negative for k > max(d1,d2) and then concludes the entropy criterion fails for every state. Proposition 2 inserts the average purity (d1d2+k)/(d1d2k+1) into the inequality ||ρ_Rl||_1 ≤ d1√Trρ² and concludes realignment fails for k ≥ k0. In both cases the step from average to universal is a quantifier shift: non-negativity of an average says nothing about the minimum of the conditional entropy, and the average purity being below 1/d1² does not bound the purity of individual samples. The paper's own Table I supplies direct counterexamples—FE=0.054 at k=6 in 2×5, and FRl=0.021 at k=7, both above the claimed cutoffs. The authors note a 0.055 fraction in 2×8 at k=9 in Remark 1 without noticing it contradicts Proposition 1.\n\nWhat is good: the normalized fraction FE and the mean/minimum detectable entanglement figures of merit are sensible; the observation that realignment stays comparatively effective when |d1−d2| is small is new and numerically well supported; and the independent proof of the P/Rd equivalence in 2⊗d is correct, if known. The citations to the relevant background are appropriate.\n\nThe soft spot is concentrated in Sec. II A. The propositions are repairable as typicality statements with concentration bounds—it is plausible that the detected fractions vanish exponentially above the cutoffs—but as written they overclaim. The numerical hierarchy itself does not depend on the false universality, so a revised version with honest 'typicality' claims could be a solid contribution.\n\nThis is a paper for people working on entanglement detection in high-dimensional states, and for anyone who wants a sharp example of why averages do not imply typicality. It deserves a serious referee—the questions are worth asking and the numerics are worth checking—but the submitted version should not be accepted. I would send it back for major revision, insisting that the analytic claims be either corrected with concentration bounds or downgraded to conjectures.","headline":"The numerics are useful and the empirical hierarchy looks plausible, but the two headline propositions infer universal failure from Haar averages and are directly contradicted by the paper's own tables.","tokens_in":24495,"tokens_out":7487,"would_cite":false,"duration_ms":79967,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","81P68"],"pacs":["03.67.Mn","03.65.Ud"],"model":"deepseek-v4-flash","headline":"For Haar-random bipartite states, entropic and realignment tests provably stop working above rank thresholds set by the local dimensions.","keywords":["entanglement detection","Haar random states","rank thresholds","realignment criterion","entropic criterion","majorization criterion","reduction criterion","subsystem asymmetry"],"falsifier":"Generate any single Haar-random rank-$k$ state in $2 \\otimes 5$ with $k = 6$ (above $\\max(d_1, d_2) = 5$) and compute $S_{12} - S_1$ and $S_{12} - S_2$; if either is negative for at least one sample, Proposition 1's universal failure claim is disproved. The paper's own Table I, reporting $F_E = 0.054$ at $k = 6$, already provides such evidence.","tokens_in":23431,"feed_emoji":"🎲","tokens_out":4536,"duration_ms":44848,"temperature":0.7,"pith_summary":"This paper asks how well four standard entanglement detection criteria—majorization, reduction, realignment, and entropy—perform on Haar-random bipartite mixed states of a given rank in $d_1 \\otimes d_2$. It proves rank thresholds beyond which the entropic and realignment criteria cannot detect entanglement for these states: entropy fails when rank $k > \\max(d_1, d_2)$, and realignment fails when $k \\geq (d_1^3 d_2 - 1)/(d_1(d_2 - d_1))$. It then establishes an ordering $P = R_d > M > R_l > E$ in qubit-qudit systems and a different ordering in higher dimensions, where realignment overtakes majorization and can even beat reduction at moderate ranks. The paper also shows that the spread between subsystem dimensions controls how well any criterion works, with realignment staying effective at high ranks when $d_2 - d_1$ is small.","feed_headline":"Entropy and realignment miss entanglement past proven ranks","feed_subtitle":"Two standard tests fail above rank max(d1,d2) and (d1^3 d2-1)/(d1(d2-d1)), and dimension asymmetry shifts the whole order.","key_machinery":"The argument rests on two Haar-average formulas plus a norm inequality. First, the standard Haar-average entropies for a rank-$k$ state, $\\langle S\\rangle \\approx \\log k - k/(2d_1d_2)$ and $\\langle S_1\\rangle \\approx \\log d_1 - d_1/(2d_2 k)$, turn the entropic criterion into a calculus problem on the rank $k$. Second, the average purity $\\mathrm{Tr}(\\rho^2) = (d_1d_2 + k)/(d_1d_2 k + 1)$, combined with $\\|\\rho^{R_l}\\|_1 \\leq d_1 \\|\\rho\\|_F$, converts the realignment criterion into a monotonically decreasing function of $k$ with an explicit zero at $k_0$. The equivalence of partial transposition and reduction in $2 \\otimes d$ is proven separately by noting that the reduction operator $r = I_1 \\otimes \\rho_2 - \\rho_{12}$ is unitarily related to $\\rho^{T_1}$ via $U = I_1 \\otimes \\sigma_y$.","core_discovery":"The central claim is that for Haar-uniform random states, two widely used entanglement tests have provable dead zones in rank. Proposition 1 uses ensemble-averaged von Neumann entropies, $\\langle S_{12}\\rangle \\approx \\log k - k/(2d_1d_2)$ and $\\langle S_1\\rangle \\approx \\log d_1 - d_1/(2d_2 k)$, to show that when $k > \\max(d_1, d_2)$, the averaged conditional entropies are non-negative, so the entropic criterion never declares such states entangled on average. Proposition 2 bounds the realignment trace norm by $\\|\\rho^{R_l}\\|_1 \\leq d_1 \\sqrt{\\mathrm{Tr}(\\rho^2)}$ using the Frobenius norm and the Haar-average purity $\\mathrm{Tr}(\\rho^2) = (d_1d_2 + k)/(d_1d_2 k + 1)$, giving the threshold $k_0 = (d_1^3 d_2 - 1)/(d_1(d_2 - d_1))$ above which realignment cannot certify entanglement. Simulations of NPT states then yield a hierarchy of detection power that depends on rank and on the difference $|d_1 - d_2|$: qubit-qudit systems obey $P \\equiv R_d > M > R_l > E$, while for $d_1, d_2 \\geq 3$ realignment overtakes majorization and can surpass reduction at moderate ranks.","pith_inferences":["Concentration inequalities would turn these average statements into 'with overwhelming probability' statements; without them, the universal rank thresholds overclaim what one can say about a single random draw.","The formula for $k_0$ implies that the realignment criterion's critical rank diverges as $d_2 \\to d_1$, so near-square systems are exactly the regime where realignment should be tested experimentally at high ranks.","The same average-purity machinery could be applied to other trace-norm criteria, such as the computable cross-norm criterion or Schmidt-number witnesses, to produce analogous rank cutoffs.","The dimension-asymmetry trends suggest a benchmark design: fix total dimension $d_1 d_2$ and compare detection fractions across all factorizations of that product, as the paper's Table VI begins to do for $d_{12} = 12, 16, 18, 24, 36$."],"forward_implications":["Entropy-based tests are effectively useless for typical mixed states once the rank exceeds the larger local dimension.","Realignment tests on typical states have a hard rank ceiling fixed by $d_1$ and $d_2$, so realignment cannot serve as a general high-rank detector despite its ability to catch some bound entangled states.","In qubit-qudit systems, majorization is strictly more powerful than realignment for typical NPT states, reversing the intuition drawn from bound-entanglement detection.","In dimensions 3 and above, realignment overtakes majorization and can exceed reduction at moderate ranks, giving a dimension-dependent hierarchy.","For fixed total dimension, the most asymmetric split produces higher mean detectable entanglement, while realignment only stays alive near the full-rank regime when $|d_1 - d_2|$ is small."],"supporting_citations":[{"why":"Supplies the Haar-average entropy and purity formulas that both Propositions 1 and 2 differentiate and evaluate.","marker":"[72]"},{"why":"Defines the realignment criterion and its trace-norm bound used in Proposition 2.","marker":"[58]"},{"why":"Provides the separable-state condition $\\sum_i g_i \\leq 1$ for the realignment criterion.","marker":"[60]"},{"why":"Gives the entropic separability condition $S_{12} - S_1 \\geq 0$, $S_{12} - S_2 \\geq 0$ that Proposition 1 analyzes.","marker":"[53]"},{"why":"Establishes the equivalence of partial transposition and reduction in $2 \\otimes d$ that the paper re-proves as Proposition 3.","marker":"[101]"},{"why":"Defines partial transposition, the benchmark criterion against which all other criteria are measured.","marker":"[51]"},{"why":"States the separability condition for partial transposition and its limits beyond $2 \\otimes 3$, motivating the NPT focus.","marker":"[52]"},{"why":"Defines logarithmic negativity, the entanglement measure used for the mean and minimum detectable entanglement figures of merit.","marker":"[70]"}],"fun_headline_variants":["Entropy and realignment fail past proven ranks in random states","Rank thresholds doom entropy and realignment entanglement tests","Dimension asymmetry reshuffles entanglement detection hierarchy","Proven dead zones for entropy and realignment in high rank","Realignment beats entropy only when subsystem sizes nearly match"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proofs replace each individual random state's entropy and purity with the ensemble-averaged values, and the paper's own simulations show nonzero detection fractions above the claimed cutoffs, so the 'fails to detect' conclusion holds only if every state in the ensemble is well represented by its average.","fun_headline_variants_meta":{"raw":{"variants":["Entropy and realignment fail past proven ranks in random states","Rank thresholds doom entropy and realignment entanglement tests","Dimension asymmetry reshuffles entanglement detection hierarchy","Proven dead zones for entropy and realignment in high rank","Realignment beats entropy only when subsystem sizes nearly match"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000145,"raw_usage":{"total_tokens":1248,"prompt_tokens":1086,"completion_tokens":162,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":702,"completion_tokens_details":{"reasoning_tokens":85}},"tokens_in":702,"tokens_out":162,"duration_ms":2549,"temperature":1.0,"reasoning_tokens":85,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T12:22:56.544325+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate any single Haar-random rank-$k$ state in $2 \\otimes 5$ with $k = 6$ (above $\\max(d_1, d_2) = 5$) and compute $S_{12} - S_1$ and $S_{12} - S_2$; if either is negative for at least one sample, Proposition 1's universal failure claim is disproved. The paper's own Table I, reporting $F_E = 0.054$ at $k = 6$, already provides such evidence.","supporting_citations":[{"cited_title":"Chen and L.-A","cited_arxiv_id":null,"evidence_quote":"Supplies the Haar-average entropy and purity formulas that both Propositions 1 and 2 differentiate and evaluate."},{"cited_title":"Gupta, S","cited_arxiv_id":null,"evidence_quote":"Establishes the equivalence of partial transposition and reduction in $2 \\otimes d$ that the paper re-proves as Proposition 3."}],"review_version":1}