{"id":"e543152d-43dd-4063-ae08-c29d0a4bb859","arxiv_id":"2608.02590","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Optimal quantum de Finetti error bounds are proven by sum-of-squares/argmax rounding, yielding subexponential separability algorithms and a counterexample to the exponential disentangler conjecture.","lead":"This paper proves tight finite quantum de Finetti bounds: bosonic two-site marginals are O(√d/N) close in trace norm to mixtures of pure tensor powers, and exchangeable marginals are O(d/N) close. It uses a new 'argmax rounding' technique to derive subexponential algorithms for separability problems and to refute Watrous's disentangler conjecture.","discovery_kind":"new_method","skeptic_critique":null,"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves optimal finite quantum de Finetti upper bounds by a new 'argmax rounding' method. The central result (Theorem 1.1) gives a trace-norm bound sqrt(d-1)/(2(N-1)) for two-site marginals of N-boson states against Hartree mixtures; a symmetric purification argument yields the optimal O(d/N) bound for exchangeable states. The method is extended to t-site marginals (O(t sqrt(d)/N) bosonic, O(t d/N) exchangeable), to two-sided Bose-symmetric extendible states (O(sqrt(min{dA,dB})/sqrt(nm))), and to applications: subexponential disentanglers, subexponential Best Separable State without perfect completeness, subexponential trace-norm separability testing, and a dimension-free bosonic Hilbert-Schmidt theorem with rate O(N^{-1/2}). All key upper-bound lemmas are proved in the text with explicit constants; Appendix B supplies matching lower bounds using rectangular Werner states.","tokens_in":1213,"tokens_out":1157,"duration_ms":814153,"significance":"This is a major advance in the quantitative theory of quantum de Finetti theorems. The proof idea--casting de Finetti approximation as an SDP integrality-gap problem and using first- and second-order optimality at an argmax tensor--is original and likely to be influential. The main bounds settle the dimension dependence left open by Christandl-Konig-Mitchison-Renner and improve the bosonic two-site bound from O(d/N) to O(sqrt(d)/N). The applications are substantial: a subexponential disentangler refuting Watrous's conjecture, the first subexponential explicit BSS algorithm without perfect completeness, the first subexponential trace-norm separability algorithm, and the first dimension-free bosonic de Finetti theorem in Hilbert-Schmidt distance. The manuscript is careful: the central derivations in Sections 3, 4, and 6 are self-contained, constants are explicit, and the lower-bound appendix makes the optimality claims checkable. The main external ingredients are the standard CKMR symmetric-purification lemma and standard representation-theoretic facts used in the lower bounds.","major_comments":[],"minor_comments":[{"comment":"The identification 'Sym^n(HA) ⊗ Sym^n(HB) identifies with the summand of Sym^{2n}(H⊕) containing n sites of each label' is only correct via the isometric embedding that symmetrizes across the A/B label boundary with the normalization sqrt((2n)!/(n!n!)). As written, a reader could mistakenly think one simply relabels the first n and last n sites. Please define this isometry explicitly; the subsequent cross-label marginal computation in (6.23) depends on it.","section":"§6.2, Lemma 6.6"},{"comment":"The symmetric purification lemma is quoted from CKMR and is load-bearing for all exchangeable results. It is true and short to prove: the canonical purification (I⊗sqrt(σ_N))|Ω⟩ is invariant under simultaneous permutations of the N composite sites. Please include the one-paragraph proof or a more precise statement so the exchangeable theorems are self-contained.","section":"§3.1.4, Lemma 3.8"},{"comment":"Several reference labels have spacing artifacts: [JL W26], [L W26], and [JWX26] contain stray spaces. These should be normalized before publication.","section":"References"},{"comment":"The symbol R_{a,r,t}(μ) is used both for the ratio of weights in (B.3) and later for the central element denoted with a hat in the proof of Lemma B.4. Consider renaming one of them to avoid confusion, e.g., use L_{a,r,t}(μ) for the likelihood ratio.","section":"Appendix B, Eq. (B.3)"},{"comment":"The sentence 'the component in u∨u⊥ is orthogonal to ηu' is slightly misleading: first-order optimality shows ηu has no mixed component, and then orthogonality of the decomposition does the rest. Rephrase for clarity.","section":"§1.3.1"},{"comment":"The notation O_ε(√d log d) hides the 1/ε dependence. This is standard but should be stated explicitly in the theorem or its proof, since the disentangler conjecture depends on the fixed-ε regime.","section":"§5.1, Theorem 5.1"}],"recommendation":"minor_revision","confidential_remarks":"This is a strong paper with a new technique and several high-impact consequences. I found no correctness risk in the central derivations. The main points to address in revision are expository: make the isometric embedding in §6.2 explicit, provide the short proof of the CKMR symmetric purification, and fix the reference/notation glitches. The lower-bound appendix is long and uses external representation-theoretic facts, but the arguments are visible and the claims are consistent with the upper bounds."},"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","90C22","15A69"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"The paper proves optimal finite quantum de Finetti upper bounds: every N-boson state's two-site marginal is within √(d−1)/(2(N−1)) of a Hartree mixture, and via symmetric purification every exchangeable state's two-site marginal is within √","keywords":["quantum de Finetti theorem","bosonic states","exchangeable states","argmax rounding","symmetric extension SDP","trace distance","disentangler","separability testing"],"falsifier":"Take a small instance, say d = 2, N = 3, and run the symmetric-tensor SDP: for a random Hermitian witness M on Sym^2(C^d) with ||M||_∞ ≤ 1, compute λ_max(M^[N]) − h(M). A single witness exceeding √(d−1)/(N−1) would directly falsify the two-site bosonic theorem; exhaustive maximization over low-dimensional witnesses would settle whether the claimed gap holds numerically.","tokens_in":50261,"feed_emoji":"⚛️","tokens_out":6752,"duration_ms":62555,"temperature":0.7,"pith_summary":"Every N-boson state on (C^d)^⊗N has a two-site marginal that is, in trace distance, at most (1/2)√(d−1)/(N−1) away from a mixture of pure tensor powers |u><u|^⊗2; for arbitrary exchangeable states, the same mechanism gives distance at most (1/2)√(d²−1)/(N−1) to mixtures of σ^⊗2. These bounds match the known lower bounds, so they settle the dimension dependence left open since 2007. The proof is a rounding argument: de Finetti approximation is cast as the integrality gap of a symmetric-extension semidefinite program, and an eigenvector is rounded by the tensor power that has largest overlap with it. If correct, the square-root improvement moves several quantum-complexity problems from exponential to subexponential dimension dependence: it refutes the disentangler conjecture, and yields subexponential algorithms for Best Separable State and trace-norm separability testing. A spectral-truncation corollary gives the first dimension-free bosonic de Finetti theorem in Hilbert–Schmidt distance at the optimal N^{−1/2} rate.","feed_headline":"Bosonic de Finetti bound improves to sqrt(d-1)/(N-1)","feed_subtitle":"Optimal dimension dependence settles a 2007 open problem and refutes the disentangler conjecture.","key_machinery":"Argmax rounding: given a top eigenvector of a symmetric-tensor or rectangular SDP lift, choose the tensor power (or product vector) with maximum overlap; first-order optimality eliminates mixed components and second-order optimality bounds the orthogonal component. This replaces universal averaging with an extremal, witness-dependent rounding choice, and the total dimension cost appears only through the trace norm of a (d−1)×(d−1) matrix. The relevant sets are Hartree mixtures SEP_H (mixtures of |u><u|^⊗t), tensor-power mixtures SEP_P, bosonic marginals BOS_N, exchangeable marginals EXCH_N, and two-sided Bose-symmetric extendible states BEXT_{n,m}.","core_discovery":"On the paper's own terms, the central discovery is that the worst-case trace-norm error from an N-boson state's two-site marginal to the closest Hartree mixture equals the integrality gap of a symmetric-tensor SDP relaxation, and that argmax rounding bounds this gap by √(d−1)/(N−1). For a Hermitian witness M, one takes a top eigenvector ψ of the N-site lift M^[N]; choosing the unit vector u with maximal overlap |<u^⊗N, ψ>|, the partial inner product satisfies η_u = c u^⊗2 + ζ_u, where ζ_u lies in Sym^2(u^⊥) and its overlap with every orthogonal square v^⊗2 is at most c/(N−1). A Takagi decomposition and Schatten duality on the resulting (d−1)-dimensional matrix pair give the square-root dimen","pith_inferences":["The witness-dependent argmax rounding scheme may be reusable in other symmetric-extension hierarchies, where choosing an extremal rounding direction instead of averaging could yield similar square-root improvements in integrality-gap bounds beyond de Finetti.","The (ε,0) disentangler refutes the original conjecture but does not imply QMA(2) = QMA; a testable next step is whether any stronger notion of disentangling can also be constructed with subexponential input dimension.","A dimension-free Hilbert–Schmidt theorem for arbitrary exchangeable states remains open, because partial trace does not contract Hilbert–Schmidt norm; the paper's spectral-truncation method would need a different purification mechanism to extend.","The t-site bound's linear factor in t is shown necessary in the paper's joint high-dimensional regime; the exact Schur-label calculations may also pinpoint the optimal constants for small t and fixed d."],"forward_implications":["For every N, d ≥ 2, the bosonic two-site de Finetti error is at most √(d−1)/(2(N−1)), and matching lower bounds make this dimension dependence optimal.","For exchangeable states, symmetric purification gives the two-site error bound √(d²−1)/(2(N−1)) to mixtures of tensor powers, matching the known lower-order behavior.","For every fixed ε ∈ (0,1), there exists an (ε,0)-disentangler with input dimension exp(O_ε(√d log d)); the disentangler conjecture is false.","General explicit Best Separable State and trace-norm separability testing admit deterministic exp(Õ(√d/ε))-time algorithms without a perfect-completeness assumption.","Spectral truncation of the optimal trace-norm theorem gives a dimension-free bosonic Hilbert–Schmidt de Finetti theorem with the optimal O(N^{−1/2}) rate when dimension may grow."],"fun_headline_variants":["De Finetti bound hits sqrt(d-1)/(N-1), settles 2007 question","Optimal de Finetti via argmax rounding: sqrt(d-1)/(N-1) error","Argmax rounding solves de Finetti dimension dependence","Bosonic de Finetti: optimal sqrt(d-1)/(N-1) and disentangler refuted"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The exchangeable theorem and the consequences built on it rely on a cited symmetric-purification lemma: every exchangeable state on (C^d)^⊗N can be purified by adding an environment of the same dimension so that the whole N-pair system remains permutation-symmetric; if that lemma failed, the O(d/N) exchangeable bounds and the purification-based algorithms would not follow from the bosonic theorem alone.","fun_headline_variants_meta":{"raw":{"variants":["De Finetti bound hits sqrt(d-1)/(N-1), settles 2007 question","Optimal de Finetti via argmax rounding: sqrt(d-1)/(N-1) error","Argmax rounding solves de Finetti dimension dependence","Bosonic de Finetti: optimal sqrt(d-1)/(N-1) and disentangler refuted"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000665,"raw_usage":{"total_tokens":2982,"prompt_tokens":961,"completion_tokens":2021,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":705,"completion_tokens_details":{"reasoning_tokens":1928}},"tokens_in":705,"tokens_out":2021,"duration_ms":12362,"temperature":1.0,"reasoning_tokens":1928,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T03:32:05.192231+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small instance, say d = 2, N = 3, and run the symmetric-tensor SDP: for a random Hermitian witness M on Sym^2(C^d) with ||M||_∞ ≤ 1, compute λ_max(M^[N]) − h(M). A single witness exceeding √(d−1)/(N−1) would directly falsify the two-site bosonic theorem; exhaustive maximization over low-dimensional witnesses would settle whether the claimed gap holds numerically.","supporting_citations":[],"review_version":1}