{"id":"f924e4b4-4943-46af-a205-ed78256036f9","arxiv_id":"2505.20519","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For Bell scenarios where N-1 parties have two binary measurements, all quantum correlations can be realized with local dimensions 2 for those parties and 2^(N-1) for the last, up to a Carathéodory factor.","lead":"This paper bounds the local Hilbert space dimensions needed to reproduce all quantum correlations in Bell experiments. It shows that when all but one party have binary measurements, the last party needs at most 2^(N-1) dimensions, and it extends the bound to non-extremal and communication-assisted scenarios.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4's purification argument skips the reference system for mixed states, and the gap propagates to Lemma 5 and Theorem 7, leaving the proof dependent on Ref [20].","rationale":"The reader's weakest-assumption identification is accurate. I re-derived Lemma 4's proof and confirmed the purification issue: for a mixed state, purification introduces a reference system, so the Schmidt projector on the B subsystem alone cannot leave the state invariant. This invalidates the displayed trace manipulation for mixed states. The same flaw enters Lemma 5 and, through the Carathéodory step, Theorem 7. I also checked whether the central claims could be recovered: for convexly extremal correlations, extremality implies a pure-state realization (any pure-state decomposition of a mixed realization must yield the same extreme correlation), so the Schmidt compression is valid for the extremes that feed Theorem 6. Thus Theorem 6 is likely correct, and Theorem 7 likely correct if the mixed-state compression from Ref [20] is imported. The paper's proof, however, does not make this argument and instead asserts 'without loss of generality' purity in a way that is not valid for the stated lemmas. This is a genuine gap in rigor, not a demonstrated falsehood. The citation to Ref [20] may fill the gap, but the paper should state that explicitly and prove Lemma 5 from it. I therefore see no reason to change the reader's conditional verdict; the result is promising but needs a repaired proof.","tokens_in":11135,"tokens_out":45010,"duration_ms":498628,"concrete_test":"Run the dimension-bounded NPA hierarchy (Navascués–Vértesi method) for the bipartite scenario where Alice has 2 inputs/2 outputs and Bob has 3 inputs/3 outputs, and compare the set Q_(2,2) against the full quantum set Q∞. If any correlation exists in Q∞ but not in Q_(2,2), Lemma 4 is false; otherwise the lemma is supported and the proof gap can be closed by citing Ref [20].","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing problem is the proof of Lemma 4. For a mixed state ρ_AB, the proof says 'without loss of generality' work with a pure state and then applies the Schmidt decomposition inside H_A⊗H_B. But purifying ρ_AB produces |ψ⟩ in H_A⊗H_B⊗H_R; the Schmidt decomposition between A and BR gives vectors |φ_i⟩∈H_B⊗H_R, so the projector Π=Σ|u_i⟩⟨u_i|⊗|v_i⟩⟨v_i| with |v_i⟩∈H_B cannot leave |ψ⟩ invariant unless R is trivial. Thus the dimension-reduction step is not justified for mixed states. Lemma 5 repeats this by grouping the first N−1 parties and again purifying without a reference. Theorem 7 inherits the issue because it uses Q∞=ConvexHull(Q_d), and the size of Q_d depends on Lemma 5. The claim may be true—Ref [20] plausibly proves the needed compression—but the paper's derivation does not; the lemma is carried by citation. If the mixed-state step cannot be repaired, the upper bound for non-extremal correlations is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims finite upper bounds on the local Hilbert space dimensions sufficient to reproduce quantum correlations in certain Bell scenarios. For scenarios in which the first N-1 parties have two settings and two outcomes, it asserts (Theorem 6) that the maximum of any convex function over all quantum correlations is attained on a pure state with local dimensions 2 for the first N-1 parties and 2^(N-1) for the last party. For non-extremal correlations, Theorem 7 multiplies these dimensions by the affine dimension of the quantum set, using a Carathéodory/Fenchel argument. The authors also translate these bounds to Bell scenarios with communication through a maximal-interruption projection, yielding Corollaries 9 and 10.","tokens_in":11365,"tokens_out":21484,"duration_ms":220980,"significance":"The question of sufficient Hilbert space dimension for quantum correlations is important for numerical optimization, dimension witnesses, and the foundational classification of quantum resources. The paper addresses a meaningful problem and builds on solid ingredients: Masanes' qubit-reduction theorem, Schmidt decomposition, and the maximal-interruption construction for causal scenarios. If the proof gaps were repaired, the results would provide explicit and reasonably general finite-dimensional sufficiency bounds, and the Bell+ extension is a useful conceptual transfer. However, the current manuscript has a purification gap in Lemmas 4-5 and a convexity inconsistency in Section II.D, so the central claims are not established by the arguments as written.","major_comments":[{"comment":"The proof's first step, 'Without loss of generality we will consider pure states,' is not valid for mixed states. A purification of a mixed state on H_A⊗H_B is a pure state |ψ⟩ in H_A⊗H_B⊗H_R, not in H_A⊗H_B. The Schmidt decomposition of Lemma 3 and the projector Π of Eq. (3) are therefore not directly applicable to |ψ⟩, and the trace identities in Eq. (4) do not follow. The lemma may be true and is attributed to Ref. [20], but the proof as written leaves this step unjustified for non-pure correlations.","section":"II.B, Lemma 4, Eqs. (3)-(4)"},{"comment":"The proof of Lemma 5 invokes 'the same reasoning as in the bipartite case' to justify working with pure states. Since Lemma 4's purification step is unsupported for mixed states, Lemma 5's bound d_N ≤ d_1×...×d_{N-1} is likewise unsupported for general non-extremal correlations. This bound is load-bearing for Theorems 6 and 7, so the gap must be repaired or the proof must explicitly rely on a prior result.","section":"II.C, Lemma 5"},{"comment":"The set Q_⃗d is defined as the set of all correlations realizable with local Hilbert space dimensions at most ⃗d. This set is convex, since a convex combination of density operators with local dimensions at most ⃗d is again a density operator with the same local dimensions. Therefore ConvexHull(Q_⃗d)=Q_⃗d and CathNum(Q_⃗d)=1, and the Carathéodory/Fenchel multiplication leading to Eq. (7) and Theorem 7 is not justified. If the authors intended Q_⃗d to denote a smaller nonconvex set, such as correlations generated by pure states, that definition must be stated explicitly, and Eq. (5b) would then require the additional nontrivial claim that Q∞ is the convex hull of its extreme points, which is not established and is delicate in light of the known non-closure of Q∞ cited in the Introduction.","section":"II.D, Eqs. (5)-(7)"},{"comment":"The theorem compares a maximum over Q∞ with a maximum over a finite-dimensional set. Because Q∞ is not assumed closed, and is in general not closed, the text must argue that the maximum is attained and that it is attained at a convexly extremal point. No such argument is provided; if the maximizer lies in the closure of Q∞ but outside Q∞, the stated equality can fail. This is not merely a technicality, since the Introduction itself cites examples requiring infinite Hilbert space dimension for extremal correlations.","section":"II.C, Theorem 6"}],"minor_comments":[{"comment":"The word 'Schimdt' appears in the sentence 'Combining the use of the Schimdt decomposition'; it should read 'Schmidt'.","section":"II.B"},{"comment":"The notation CathNum(S) is nonstandard and should be defined in a way that makes clear whether it refers to the Carathéodory number of the set S itself or of its convex hull; the current usage is ambiguous.","section":"II.D"},{"comment":"The proof of Proposition 8 is essentially a citation plus a geometric statement; a short explicit argument that the maximal-interruption projection preserves the required dimension bound would improve readability and self-containment.","section":"III, Proposition 8"},{"comment":"The phrase 'p admits some quantum realization if and only if p admits a quantum realization where...' is logically redundant; the nontrivial content is an upper bound on sufficient dimensions, and the 'only if' direction holds by definition of Q∞.","section":"II.D, Theorem 7"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is not self-contained in its current form: Lemma 4 is essentially quoted from Ref. [20] but with a proof that has a genuine purification gap, and Section II.D appears to treat the convex set Q_⃗d as if it were nonconvex. These issues affect Theorems 6 and 7, which are the paper's main results. I would encourage the authors to either supply a correct proof or restructure the paper to explicitly rely on the prior literature, and to reconsider whether the Carathéodory argument adds anything beyond the convexity already present in the definition of Q_⃗d."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper gives explicit Hilbert-space dimension upper bounds for a useful family of Bell scenarios: where the first N−1 parties have two dichotomic settings, the last party can be bounded by 2^(N−1) for extremal correlations, and a Carathéodory-based bound handles all (including non-extremal) correlations. The transfer to Bell+ communication scenarios via maximal interruption is clean and practical. The writing is clear and the prior literature is handled honestly, with Lemma 4 explicitly credited to Sikora et al. [20]. That is real credit where due.\n\nThe main new content is Theorem 6 (multipartite 2^(N−1)) and Theorem 7 (Carathéodory extension). These are plausible and likely correct, but they inherit a real gap in the proof of Lemma 4. The proof says \"without loss of generality\" work with a pure state, then applies Schmidt projectors inside H_A⊗H_B. For a mixed state, purification requires a reference system R, and the Schmidt decomposition between A and BR does not give vectors in H_B alone. So the projector Π as defined cannot leave the purified state invariant unless R is trivial. This is a genuine reasoning gap, not a stylistic quibble. It propagates to Lemma 5 and therefore to Theorems 6 and 7. The result may still be true—Ref [20] likely proves the needed compression, and the paper even notes the lemma was previously observed there—but the derivation as written does not.\n\nA second soft spot is more minor: Theorem 7 uses AffineDimension(Q_d) = AffineDimension(Q_∞), citing Ref [21, Theorem 1]. That reference gives the affine dimension for the full quantum set, not for the restricted set Q_d. The equality needs justification for the specific bounded dimensions, since for small d the affine span can shrink. This is likely fixable, but it is asserted rather than shown.\n\nThe bipartite qubit result (both parties qubits when one is 2×2) is already contained in Masanes plus Sikora; the paper does not overclaim this, but the novelty rests on the multipartite and Carathéodory parts.\n\nWho this is for: people working on dimension bounds, convex optimization over the quantum set, and Bell+ scenarios. It deserves a serious referee. The gaps are repairable and the central claims are probably true, so I would accept for peer review with a clear request to fix Lemma 4's purification step and to prove or properly cite the affine-dimension equality.","headline":"Useful dimension bounds for a class of Bell scenarios, but the key bipartite lemma has a purification gap that propagates; worth a serious referee with revision.","tokens_in":11915,"tokens_out":1372,"would_cite":true,"duration_ms":16806,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","52A35"],"pacs":["03.65.Ud","03.67.-a"],"model":"deepseek-v4-flash","headline":"In N-partite Bell scenarios where the first N-1 parties have two settings and two outcomes, all convexly extremal quantum correlations are realizable with qubits for those parties and dimension $2^{N-1}$ for the last.","keywords":["Bell scenarios","quantum correlations","Hilbert space dimension","qubit bounds","Schmidt decomposition","Carathéodory's theorem","Bell scenarios with communication","convex extremal points"],"falsifier":"In the bipartite scenario where Alice has two settings and two outcomes and Bob has, say, three outcomes per setting, run a standard hierarchy of semidefinite relaxations to test whether every convexly extremal quantum correlation is attainable by a two-qubit state; a single extremal correlation requiring Bob's local dimension greater than 2 would disprove the central bound for N=2.","tokens_in":10974,"feed_emoji":"⚛️","tokens_out":10693,"duration_ms":108073,"temperature":0.7,"pith_summary":"This paper establishes finite upper bounds on the Hilbert space dimensions needed to reproduce quantum correlations in Bell experiments. Its central result is that in any N-partite Bell scenario where the first N-1 parties each have two measurement settings and two outcomes, every convexly extremal quantum correlation can be realized with qubits for those N-1 parties and a local Hilbert space of dimension $2^{N-1}$ for the remaining party. The paper then uses Carathéodory's theorem to convert this extremal-point bound into a bound valid for every quantum correlation, multiplying each local dimension by the affine dimension of the quantum set. The same bounds are shown to transfer to Bell scenarios with communication, because such scenarios are projections of ordinary Bell scenarios. A sympathetic reader should care because these are among the few multi-setting scenarios where finite-dimensional sufficiency is known, and the bounds make convex optimization over quantum correlations and the membership problem tractable.","feed_headline":"Qubits suffice for all two-setting parties in Bell tests","feed_subtitle":"And the remaining party needs no more than 2^(N-1) dimensions to reach every extremal correlation.","key_machinery":"The machinery has three parts. First, a known proposition compresses any party with two settings and two outcomes to a qubit, regardless of the rest of the scenario. Second, a Schmidt-decomposition argument compares dimensions across parties: in the bipartite case the second party's needed dimension never exceeds the first's, and the multipartite generalization bounds the last party's dimension by the product of the preceding parties' dimensions. Third, convex geometry converts these extremal-point bounds into all-correlation bounds: because the convex hull of the finite-dimensional quantum set is the full quantum set, Carathéodory's theorem multiplies each local dimension by the affine dimension of the quantum set. For Bell+ scenarios, the maximal-interruption projection carries these bounds over unchanged.","core_discovery":"The paper claims two theorems. Theorem 6 states that in an N-partite Bell scenario where the first N-1 parties have two settings and two outcomes, the maximum of any convex function over all quantum correlations equals the maximum over correlations generated by a pure state whose first N-1 local Hilbert space dimensions are 2 and whose final local Hilbert space dimension is $2^{N-1}$. Theorem 7 extends this to every quantum correlation: a correlation is quantum if and only if it can be realized with the first N-1 local dimensions equal to $2 \\cdot \\mathrm{AffineDimension}(Q_\\infty)$ and the last equal to $2^{N-1} \\cdot \\mathrm{AffineDimension}(Q_\\infty)$, where the affine dimension of the quantum set is given by a closed formula in terms of the settings and outcomes. The same dimension bounds are shown to carry over to Bell+ scenarios, where measurement settings may depend on other parties' outputs.","pith_inferences":["The theorems leave open whether $2^{N-1}$ is tight for large N; for N=2 the bound matches known qubit results, but for larger N the true minimum for specific functionals may be smaller.","The affine-dimension multiplier grows quickly with the number of settings and outcomes, so the finite bounds for non-extremal correlations may be too large for direct numerics; a likely fruitful direction is finding smaller multipliers by exploiting additional structure of the finite-dimensional quantum set.","The proof technique suggests a general recipe: whenever a scenario's extremal quantum correlations are finitely realizable, physically mixing those realizations automatically yields finite realizations for all correlations, with the Carathéodory number of the finite set as the overhead.","Because the Bell+ step uses only the projection property, the results should transfer to any causal scenario obtained by interrupting communication links between parties, not just the examples discussed."],"forward_implications":["In any bipartite Bell scenario where one party has two settings and two outcomes, all convexly extremal quantum correlations can be produced by two-qubit states, no matter how many settings or outcomes the other party has.","Maximizing any convex function over quantum correlations in the covered scenarios, such as computing maximal Bell violations, reduces to an optimization over a finite-dimensional Hilbert space of known dimension.","Every quantum correlation in the covered scenarios, not just extremal ones, has an explicit finite-dimensional realization; the dimension formula gives a route to a finite algorithm for the quantum-membership problem.","The same finite-dimension bounds apply to nonstandard Bell scenarios with communication, including instrumental scenarios, since projection cannot increase the required dimension."],"supporting_citations":[{"why":"Proves that a party with two settings and two outcomes can always be described by a qubit; this is the starting point for the dimension bounds.","marker":"[8]"},{"why":"States the bipartite dimension-comparison lemma (the second party's needed dimension does not exceed the first's) that the paper's Schmidt-projection argument relies on.","marker":"[20]"},{"why":"Provides the affine-dimension formula for the quantum correlation set used in Theorem 7 to scale the extremal-point bounds.","marker":"[21]"},{"why":"Shows that instrumental-scenario correlations are projections of Bell-scenario correlations, which lets the dimension bounds carry over to Bell+ scenarios.","marker":"[24]"},{"why":"Supplies the consistency constraint for maximal interruption, the construction that links any Bell+ scenario to a standard Bell scenario.","marker":"[23]"}],"fun_headline_variants":["Qubits for every two-setting party in Bell scenarios","Extremal Bell correlations: qubits for all but last party","Hilbert dimensions shrink: first N-1 parties become qubits","Bell scenarios: qubits for N-1 parties, 2^(N-1) for last","All two-setting parties reduce to qubit dimensions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument's load-bearing premise is that a mixed bipartite quantum correlation can be analyzed as a pure state on the same two local Hilbert spaces without introducing a third purification register; if that compression fails for mixed states, the Schmidt-projection proof of the dimension comparison does not go through.","fun_headline_variants_meta":{"raw":{"variants":["Qubits for every two-setting party in Bell scenarios","Extremal Bell correlations: qubits for all but last party","Hilbert dimensions shrink: first N-1 parties become qubits","Bell scenarios: qubits for N-1 parties, 2^(N-1) for last","All two-setting parties reduce to qubit dimensions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00175,"raw_usage":{"total_tokens":6896,"prompt_tokens":918,"completion_tokens":5978,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":5885}},"tokens_in":534,"tokens_out":5978,"duration_ms":44881,"temperature":1.0,"reasoning_tokens":5885,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:54:31.760826+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In the bipartite scenario where Alice has two settings and two outcomes and Bob has, say, three outcomes per setting, run a standard hierarchy of semidefinite relaxations to test whether every convexly extremal quantum correlation is attainable by a two-qubit state; a single extremal correlation requiring Bob's local dimension greater than 2 would disprove the central bound for N=2.","supporting_citations":[{"cited_title":"MinimumDimensionofaHilbertSpaceNeededtoGener- ateaQuantumCorrelation,","cited_arxiv_id":null,"evidence_quote":"States the bipartite dimension-comparison lemma (the second party's needed dimension does not exceed the first's) that the paper's Schmidt-projection argument relies on."},{"cited_title":"Lifting Bell inequalities,","cited_arxiv_id":null,"evidence_quote":"Provides the affine-dimension formula for the quantum correlation set used in Theorem 7 to scale the extremal-point bounds."},{"cited_title":"QuantumviolationsintheInstrumentalscenario andtheirrelationstotheBellscenario,","cited_arxiv_id":null,"evidence_quote":"Shows that instrumental-scenario correlations are projections of Bell-scenario correlations, which lets the dimension bounds carry over to Bell+ scenarios."},{"cited_title":"Quantuminflation:Ageneralapproachtoquantumcausal compatibility,","cited_arxiv_id":null,"evidence_quote":"Supplies the consistency constraint for maximal interruption, the construction that links any Bell+ scenario to a standard Bell scenario."}],"review_version":1}