{"id":"f971812c-e63e-4996-8059-3bbd074ccd15","arxiv_id":"2607.21477","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For every n >= 2, the product test's worst-case acceptance probability equals (1 + mω^2 + (1−mω)^2)/2 with m = floor(1/ω), where ω is the maximum squared overlap with a product state.","lead":"This paper pins down the exact worst-case acceptance probability of the product test, a fundamental quantum circuit for detecting entanglement across many registers. The result closes a long-open low-overlap regime and improves the soundness constant in the QMA(k) to QMA(2) reduction.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: proof chain is sound and the fixed-dimension caveat is explicitly outside the theorem's scope.","rationale":"I agree with the reader that the proof is elementary and self-contained; every step can be traced. The reader's weakest assumption (fixed dimensions) is real but explicitly excluded by the definition of PT_n(ω), so it does not move the verdict. I found no hidden assumption in the induction: the branch overlap constraint and capped refinement inequality are exactly matched, and the lossy off-diagonal estimate is in the safe direction. The QMA application uses only the dimension-upper-bound direction, so it is unaffected by the fixed-dimension caveat. Verdict remains ACCEPT.","tokens_in":15828,"tokens_out":25544,"duration_ms":231588,"concrete_test":"Independently re-derive the upper-bound chain for n=3 with a Schmidt rank-3 state: compute the exact norm of each off-diagonal branch after applying Π_rest and verify (i) distinct branches remain orthogonal and (ii) inequality (6.1) holds; then evaluate the formula at ω=1/3 and check that the maximally entangled qutrit (or its product-tensor extension) achieves PT=2/3. This confirms both the safe direction of the lossy estimate and the equality case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The central claim is the exact formula PT_n(ω) = (1+s(ω))/2. The lower bound (Prop. 4.1) realizes the capped-simplex spectrum with a bipartite state; the upper bound (Prop. 6.1) is a clean induction: Lemma 5.1 keeps all diagonal branches and bounds off-diagonal branches by 1 in the safe direction, Lemma 5.3 transfers the overlap promise to each branch, and Lemma 3.4 converts the branch data into a single feasible capped vector. I checked the orthogonality/cross-term step in Lemma 5.1: the first-register factors vanish between distinct branches, so the norm decomposition is valid. The only lossy estimate is the off-diagonal bound, but it is an upper bound and equality is supplied by the bipartite extremizer tensored with product states. Section 8 explicitly flags the fixed-dimension caveat: the matching lower bound may need dimension about 1/ω, so a fixed-dimension exact curve is not claimed. That is a scope limitation, not a flaw in Theorem 1.1. No internal inconsistency, circularity, or parametric freedom found.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper determines the exact worst-case acceptance probability of the product test for pure multipartite states as a function of the maximum squared overlap with a product state. It proves that for every n ≥ 2 and every ω ∈ (0,1], PT_n(ω) = (1+s(ω))/2, where s(ω) = mω^2 + (1−mω)^2 and m = ⌊1/ω⌋. The proof consists of a matching lower bound via a bipartite state whose Schmidt spectrum is the capped-simplex optimizer, and an upper bound by induction that uses a first-SWAP reduction retaining all diagonal Schmidt branches, a branch-overlap constraint, and a capped refinement inequality. The result recovers the known ω ≥ 1/2 branch, resolves the low-overlap regime including the limit PT_n(ω)→1/2 as ω→0, and improves the one-shot soundness parameter in the Harrow–Montanaro QMA(k)→QMA(2) reduction from 1−(1−σ)^2/100 to 1−(1−σ)^2/4.","tokens_in":16067,"tokens_out":26389,"duration_ms":208908,"significance":"If the result holds, it resolves open problems posed by Soleimanifar–Wright (SODA 2022) and Lovitz–Lowe, showing that the bipartite extremal curve is universal for every number of parties. The proof is elementary, self-contained, and fully rigorous, with explicit extremal states and a sharp linear rejection bound ε/2 for the complexity-theoretic application. The fixed-dimension caveat is explicitly scoped out in Section 8 and does not affect the dimension-free theorem. The paper also gives a clean structural explanation of why the low-overlap regime requires retaining all diagonal branches, and it provides a concrete improvement in a well-studied QMA collapse reduction. Overall this is a significant contribution to quantum property testing and unentangled quantum proofs.","major_comments":[],"minor_comments":[{"comment":"When Lemma 3.4 is invoked to conclude Σ_i λ_i^2 s(φ_i) ≤ s(ω), the text is mathematically correct but terse: the lemma is applied with weights w_i = λ_i and α_i = φ_i, using Σ_i λ_i = 1 and λ_i φ_i ≤ ω from Lemma 5.3. It would help the reader to state this explicitly.","section":"§6, Proof of Proposition 6.1"},{"comment":"Equation (7.5) is stated as the two-state analogue of Lemma 2.3 without derivation. A one-line expansion of the accepting projector Π_Prod,k = 2^{−k} Σ_S SWAP_S would make the section self-contained.","section":"§7, Eq. (7.5)"},{"comment":"The in-text citation for the survey by Jeronimo, Leigh, and Wu appears as '[JL W26]' with an extra space, and the bibliography entry has a similar spacing issue in the label. This is a minor formatting glitch.","section":"References / §1.5"}],"recommendation":"accept","confidential_remarks":"The paper is a strong, self-contained resolution of a well-motivated open problem. The central theorem is correct, the proof is elementary and verifiable, and the complexity-theoretic application is clearly presented. The minor comments are optional polish and do not affect the validity of the results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline is simple: Theorem 1.1 is right, and it closes a long-open problem. The exact worst-case acceptance curve for the product test is now known for every number of parties and every overlap: PT_n(ω) = (1 + s(ω))/2, with s(ω) the capped-simplex collision probability. I read the proof carefully and it holds up. The lower bound is the natural bipartite construction; the upper bound is a clean induction that keeps all diagonal Schmidt branches and only bounds off-diagonal terms by 1, which is enough because the extremizers are bipartite. The capped refinement inequality (Lemma 3.4) is the key step, and it is correct: subdividing masses cannot beat the global cap. I checked the n=3 GHZ example against the purity formula and it matches. No circularity, no fitted parameters, no hidden assumptions. The paper is honest that the lower bound may need local dimension about 1/ω, so the exact curve for fixed bounded dimensions is not claimed; that is a scope limitation, not a flaw. The QMA(k) to QMA(2) soundness improvement from 1 − (1−σ)^2/100 to 1 − (1−σ)^2/4 is a clean corollary of the linear rejection bound ε/2, and the proof of the two-state Cauchy–Schwarz step is sound. Nothing here is overtouted. The exposition is unusually clear; even the AI-assistant acknowledgment is transparent and the math is straightforwardly human-verifiable. The main soft spot is minor: the fixed-dimension behavior remains open, and the paper says so. Also, the novelty is incremental over Soleimanifar–Wright and Lovitz–Lowe in the sense that the high-overlap branch and the bipartite case were known, but the full multipartite low-overlap curve is a real theorem, not a routine extension. I would recommend accepting the paper for peer review without hesitation; a good referee will find the proof correct and the presentation ready for publication after light polishing. This deserves to be in the literature and cited.","headline":"Exact product-test curve resolved for all n and ω; proof is elementary, correct, and a genuine advance, with only a minor fixed-dimension caveat.","tokens_in":16617,"tokens_out":1042,"would_cite":true,"duration_ms":12759,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P68","68Q12"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every number of parties, the product test's exact worst-case acceptance curve is (1 + mω² + (1−mω)²)/2, closing the open low-overlap regime.","keywords":["product test","swap test","quantum property testing","product state fidelity","unentangled quantum proofs","QMA(k) to QMA(2) reduction","capped collision probability","multipartite entanglement"],"falsifier":"Numerically maximize Σ_i λ_i² s(φ_i) over probability vectors λ and branch overlaps φ_i satisfying λ_i φ_i ≤ ω; Lemma 3.4 asserts the maximum is s(ω). A single instance, say near a reciprocal cap ω = 1/(m+1)+ε, where this sum exceeds s(ω) would break the inductive upper bound and falsify the curve. Equivalently, an explicit n-partite state with product overlap ω and acceptance probability above (1+s(ω))/2 would refute Theorem 1.1.","tokens_in":15712,"feed_emoji":"⚛️","tokens_out":9582,"duration_ms":81693,"temperature":0.7,"pith_summary":"This paper proves an exact formula for the worst-case acceptance probability of the product test, the standard circuit that checks whether an n-partite quantum state is fully unentangled by comparing corresponding registers of two copies. With ω denoting the largest squared overlap of the input with any product state and m = floor(1/ω), the formula reads PT_n(ω) = (1 + mω² + (1−mω)²)/2 for every n ≥ 2 and every ω in (0,1]. It recovers the previously known tight branch ω ≥ 1/2, supplies every low-overlap piece ω < 1/2, and implies that the acceptance probability approaches 1/2 as ω → 0, resolving an open question. The result also sharpens the one-shot soundness loss in the reduction from many unentangled quantum witnesses to two, from a constant 100 to a constant 4. A reader should care because this settles the exact soundness curve of a basic primitive in quantum property testing and unentangled-prover complexity.","feed_headline":"Product-test worst-case curve is now exact for every overlap","feed_subtitle":"One quadratic formula closes the low-overlap gap and sharpens unentangled-witness soundness.","key_machinery":"The load-bearing object is the capped collision probability s(ω): the maximum of Σ_j p_j² over probability vectors with every entry at most ω, solved greedily by filling m = ⌊1/ω⌋ entries to height ω and putting the residual mass 1−mω on one more entry. Two quantum identities connect this classical quantity to the test: the product-test acceptance probability equals 2^{−n} Σ_{S⊆[n]} Tr(ρ_S²), and for bipartite inputs it reduces to (1 + Σ_j λ_j²)/2 with λ_j the squared Schmidt coefficients. The inductive upper bound rests on a first-swap reduction that keeps every diagonal Schmidt branch after the first local swap test, bounding only the off-diagonal branches by the norm of the remaining proj","core_discovery":"The central claim is Theorem 1.1: for every n ≥ 2 and every ω ∈ (0,1], the largest possible acceptance probability of the product test over all n-partite pure states with closest-product overlap ω is PT_n(ω) = (1 + s(ω))/2, where s(ω) = mω² + (1−mω)² and m = ⌊1/ω⌋. Equivalently, on each interval 1/(m+1) < ω ≤ 1/m the curve is the quadratic 1 − mω + m(m+1)ω²/2. The supremum is attained by a bipartite state whose squared Schmidt coefficients are the capped-simplex optimizer (ω, …, ω, 1−mω), tensored with arbitrary product states on the remaining registers; this is why the curve is independent of the number of parties. The matching upper bound is dimension-free and proceeds by a sharp first-swa","pith_inferences":["The same capped-refinement accounting should transfer to any two-copy test whose acceptance depends only on branch collision probabilities, suggesting exact curves for related local-symmetric measurements could be derived by the same induction.","Because the matching construction needs local dimension ⌊1/ω⌋+1, the dimension-free curve may not be the exact worst case for fixed small local dimensions; a natural conjecture is that the fixed-dimension extremal curve is the same capped-simplex value truncated to the available Schmidt rank.","The paper notes but does not optimize the sharper soundness expression obtained from the full curve; a numerical or analytic optimization over the two branch overlaps could yield conversion parameters better than the uniform 1/4 constant for specific soundness ranges.","The piecewise-quadratic structure mirrors extremal density curves in graph theory; reading the proof as a template, one might expect exact property-testing soundness curves in other settings to be encoded by greedy concentration problems with a cap, indexed by how many active coefficients the cap permits."],"forward_implications":["The low-overlap limit is resolved: PT_n(ω) → 1/2 as ω → 0, with the quantitative bounds 1/2 ≤ PT_n(ω) ≤ 1/2 + ω/2.","The exact curve gives a tight trace-distance soundness statement: if a state is δ-far from every product state, the product test accepts with probability at most (1 + s(1−δ²))/2, which is 1 − δ² + δ⁴ when δ ≤ 1/√2.","The one-shot soundness parameter in the reduction from k unentangled witnesses to two improves from 1 − (1−σ)²/100 to 1 − (1−σ)²/4, where σ is the original soundness; the collapse QMA(k) = QMA(2) is unchanged.","The worst case cannot be worsened by adding parties: for every n ≥ 2, a bipartite state realizing the capped Schmidt spectrum, tensored with product states, is extremal, so the curve is n-independent.","At reciprocal points ω = 1/d the d-dimensional maximally entangled state is extremal and achieves acceptance (1 + 1/d)/2, matching the quadratic pieces that meet continuously at those breakpoints."],"fun_headline_variants":["Exact product test curve found for every overlap value","Product test acceptance curve: exact quadratic for all overlaps","Low overlap product test gap closed: exact curve obtained","One quadratic settles product test acceptance for all overlaps","Product test curve closed form: works for all n and overlaps"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The theorem's worst-case curve lets the lower-bound construction choose local dimensions as large as about 1/ω; if an application fixes smaller local dimensions, the construction may not fit, and the formula could overstate the true worst-case acceptance for that fixed dimension.","fun_headline_variants_meta":{"raw":{"variants":["Exact product test curve found for every overlap value","Product test acceptance curve: exact quadratic for all overlaps","Low overlap product test gap closed: exact curve obtained","One quadratic settles product test acceptance for all overlaps","Product test curve closed form: works for all n and overlaps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000906,"raw_usage":{"total_tokens":3816,"prompt_tokens":913,"completion_tokens":2903,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":657,"completion_tokens_details":{"reasoning_tokens":2826}},"tokens_in":657,"tokens_out":2903,"duration_ms":19757,"temperature":1.0,"reasoning_tokens":2826,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T07:20:40.610792+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically maximize Σ_i λ_i² s(φ_i) over probability vectors λ and branch overlaps φ_i satisfying λ_i φ_i ≤ ω; Lemma 3.4 asserts the maximum is s(ω). A single instance, say near a reciprocal cap ω = 1/(m+1)+ε, where this sum exceeds s(ω) would break the inductive upper bound and falsify the curve. Equivalently, an explicit n-partite state with product overlap ω and acceptance probability above (1+s(ω))/2 would refute Theorem 1.1.","supporting_citations":[],"review_version":1}