{"id":"769d30fe-ebd9-41e1-9804-12f1fbf923de","arxiv_id":"1909.01705","paper_version":4,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The complex Reznick Positivstellensatz is proved with improved bounds via an explicit inversion of the Chiribella identity, yielding better exponential de Finetti error estimates; the real-case version is not yet established because a lemma has a wrong constant.","lead":"This paper uses quantum information tools, such as partial traces and cloning maps, to give new proofs and improved bounds for Reznick's Positivstellensatz, a classical result about writing positive polynomials as sums of squares. The improved bounds also sharpen exponential quantum de Finetti theorems, which describe how symmetric quantum states approach independent product states.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4.1 is false as stated: the real-case proof of Theorem 4.2 uses (2k)^{2(k−t)} where the correct factor is ∏_{j=t+1}^{k} 2j(2j−1), so the claimed real improvement over Reznick is not established.","rationale":"The reader's weakest assumption points to precisely the same flaw: Lemma 4.1 asserts (2n)^{2(n−k)} tr = Δ^{n−k}p, while the displayed computation yields the constant 2n(2n−1) per Laplacian step, and the product over j is not (2n)^{2(n−k)}. Following the proof for small cases confirms the discrepancy. This is load-bearing because Theorem 4.2, the real Positivstellensatz, explicitly relies on the lemma at the step where the partial trace is replaced by an iterated Laplacian, and the rest of the proof is omitted. The claimed real bound (19) and the improvement in Remark 4.4 are therefore not established as written. I checked the complex proof independently: its analogous Lemma 2.1 is correct, because the partial trace of |v⟩⟨v|^{⊗k} is ||v||²|v⟩⟨v|^{⊗(k−1)} and Δp = k²||v||²|⟨x|v⟩|^{2k−2}, so the complex theorem and the exponential de Finetti application appear unaffected. The issue is repairable if the corrected product constants still yield (19), but the text currently contains a false lemma and an omitted computation, so the reader's CONDITIONAL verdict remains appropriate.","tokens_in":21877,"tokens_out":22071,"duration_ms":199429,"concrete_test":"Re-derive Lemma 4.1 for k=2, t=1 in d=1: p=x⁴ has Δp=12x² and tr_{2→1}(p)=x², so the correct identity is 12 tr = Δ, not 16 tr = Δ; the general factor is ∏_{j=t+1}^{k} 2j(2j−1). Then recompute equation (22) and the exponential estimate in Theorem 4.2 with this product in place of (2k)^{2(k−t)}, and check whether the resulting sufficient exponent still satisfies (19) and (24). If the corrected bound is weaker than (19), the claimed real improvement over Reznick in Remark 4.4 is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 4.1 (Section 4) states (2n)^{2(n−k)} tr_{n→k}(p) = Δ^{n−k}p for every p ∈ H^{2n}(R^d). The proof itself computes, on the spanning set p_v(x) = ⟨x|v⟩^{2n}, that Δp_v = 2n(2n−1)||v||²⟨x|v⟩^{2n−2} and tr_{n→n−1}(v^{⊗2n}) = ||v||²v^{⊗(2n−2)}. These two equations give 2n(2n−1) tr = Δ, not (2n)² tr = Δ; iterating gives Δ^{n−k}p_v = [∏_{j=k+1}^{n} 2j(2j−1)] · ||v||^{2(n−k)}⟨x|v⟩^{2k}. For example, in d=1, n=2, k=1, p(x)=x⁴ has Δp=12x² and tr_{2→1}(p)=x², so the constant is 12, not 16. Theorem 4.2's proof then substitutes p_{(tr_{k→t}⊗id)(v)} = ((2k)^{2(k−t)})^{-1}Δ^{k−t}p_v, and the omitted computation leading to (19) and (24) inherits the wrong coefficient; equations (22) and the numerical comparisons in Example 4.5 are therefore not justified. Because Section 4 explicitly leaves the final calculation to the reader, the claimed improvement over Reznick (Remark 4.4) is unsupported. The complex Theorem 3.1 and Theorem 5.1 do not rely on Lemma 4.1.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops quantum-information-theoretic proofs of Reznick-type Positivstellensätze. In the complex case (Theorem 3.1) it proves an explicit integral representation of ||x||^{2(n-k)} p_W(x,y) as ∫ p_{\\widetilde W}(φ,y)|⟨φ|x⟩|^{2n} dφ with p_{\\widetilde W} ≥ 0, for n satisfying (7), and shows how this representation yields sum-of-squares decompositions supported on complex spherical designs. In the real case (Theorem 4.2) it claims an analogous bound that improves on Reznick's classical bound. Section 5 applies the inverse of the Chiribella identity to derive an exponential quantum de Finetti theorem in diamond norm. The appendices construct Gaussian-based Hilbert identities and an elementary family of complex spherical designs.","tokens_in":22264,"tokens_out":9732,"duration_ms":84639,"significance":"If the real-case claims were repaired, the paper would be a solid contribution: Theorem 3.1 is a complete, self-contained proof with explicit constants and constructively computable SOS decompositions, and the de Finetti application in Theorem 5.1 gives an explicit exponential error rate. The complex spherical design construction in Appendix B is elementary and potentially useful. The main additional advertised value, however, is the improvement over Reznick in the real setting, and that part is currently unsupported because of the error in Lemma 4.1. The complex results do not rely on the faulty lemma, so the paper's central complex contribution remains intact.","major_comments":[{"comment":"Lemma 4.1 asserts (2n)^{2(n-k)} tr_{n→k}(p) = Δ_R^{n-k} p for p ∈ H^{2n}(R^d). The proof's own computation gives Δ_R p_{v^{⊗2n}}(x) = 2n(2n-1)||v||²⟨x|v⟩^{2n-2} and tr_{n→n-1}(v^{⊗2n}) = ||v||² v^{⊗(2n-2)}, so the correct comparison is 2n(2n-1) tr_{n→n-1} = Δ_R, not (2n)² tr = Δ_R. Iterating gives Δ_R^{n-k} p_v = [∏_{j=k+1}^{n} 2j(2j-1)] ||v||^{2(n-k)} ⟨x|v⟩^{2k}, not (2n)^{2(n-k)} tr_{n→k}(p). For a concrete check in d=1, n=2, k=1, p(x)=x⁴ satisfies Δ_R p = 12x² and tr_{2→1}(p) = x², so the constant is 12, not 16. The lemma as stated is false.","section":"Section 4, Lemma 4.1"},{"comment":"Theorem 4.2's proof uses the incorrect identity p_{(tr_{k→t}⊗id)(v)} = ((2k)^{2(k-t)})^{-1} Δ_R^{k-t} p_v, which is exactly the relation that Lemma 4.1 was supposed to establish. Since the estimates leading to (19), (22), (24) and the numerical comparison in Example 4.5 all depend on this coefficient, the claimed improvement over Reznick in Remark 4.4 is not established. The proof also explicitly leaves the final computation to the reader, so the corrected constants have not been propagated; a revised version must supply this calculation and verify that the advertised improvement survives. The complex Theorem 3.1 and the complex de Finetti theorem in Theorem 5.1 do not depend on Lemma 4.1.","section":"Section 4, Theorem 4.2, Eqs. (19), (22), (24); Example 4.5"}],"minor_comments":[{"comment":"The printed abstract says the denominator can be chosen as an 'N-th power of a linear form', but Reznick's denominator is an N-th power of the squared norm, ||x||^{2N}; the arXiv metadata abstract states this correctly, so the printed abstract should be corrected.","section":"Abstract"},{"comment":"In the paragraph after (26), 'Not that this latter bound is necessarily better' should read 'Note that this latter bound is necessarily better'.","section":"Example 4.5"},{"comment":"There is a typo in the phrase 'even by roughly a factor of 2 for the case k = 1 in Eq. (9)': the comparison is with To--Yeung's bound, and the wording should make clear which bound is being compared; also 'Renzick' is misspelled in Remark 4.4.","section":"Remark 3.4"},{"comment":"The captions of Figures 3 and 4 are verbose and would be clearer if they identified which curves correspond to which equations in the text rather than only in prose; the typo 'Not' in the right-panel description of Figure 3 also occurs there.","section":"Section 4, Figure captions"}],"recommendation":"major_revision","confidential_remarks":"The real-case defect is concentrated in Section 4 and can in principle be fixed by a recomputation with the corrected coefficient ∏ 2j(2j-1). If the corrected constants no longer yield (19) or (24), the authors should either replace the real-case claims with a statement that matches the computation or withdraw the claimed improvement. The complex theorem and the complex de Finetti application appear sound and can stand independently; I do not see grounds for rejection at this stage, but the advertised real-case improvement should not be published without a complete corrected calculation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Things you should know: this paper gives an explicit inversion of Chiribella's identity and uses it to prove a complex Positivstellensatz with better constants than To–Yeung, plus an exponential de Finetti bound. That part is real, new, and carefully derived. But the real-case extension is not sound as written: Lemma 4.1 mis-states the constant connecting the real Laplacian to the partial trace, and every real-case bound that relies on it (Theorem 4.2, Remark 4.4, Example 4.5, Remark 5.2) is unsupported until the factor is corrected and the omitted computation supplied.\n\nThe technical heart is Lemma 3.2. The coefficients q(n,k,t) are explicit, and the proof that Φ∘Ψ = id is clean. The derivation of Theorem 3.1 via measure-and-prepare maps and the Bernstein inequality is a nice unification of Reznick and To–Yeung; the k=1 bound is genuinely better than TY06. The de Finetti application is a canonical use of the inverse, and the error bound with δ is a clear improvement over Harrow's unspecified constants. The Laguerre-root construction of complex spherical designs in Appendix B is elementary and clearly presented.\n\nNow the soft spot. Lemma 4.1 states (2n)^{2(n−k)} tr_{n→k}(p) = Δ^{n−k}p, but on the spanning set v^{⊗2n} the proof itself computes Δp = 2n(2n−1)||v||² p and tr = ||v||²p, so the constant per step is 2n(2n−1), not (2n)². Iterating gives a product of 2j(2j−1) from j=k+1 to n. The d=1, n=2, k=1 example of x⁴ makes the failure concrete: 12x² versus 16x². Theorem 4.2's proof then uses the wrong constant and leaves the final computation to the reader, so the claimed improvement over Reznick (Remark 4.4) is not established. The complex results do not depend on this lemma, so a corrected real section could plausibly restore the claims; the product is just a different constant. Still, the paper as submitted should not claim the real improvement, and the real de Finetti remark should be checked as well.\n\nCitation pattern looks fine: the external benchmarks are genuinely prior work and are credited. The appendices are review material but are useful and carefully done.\n\nWho should read this: anyone working with effective Positivstellensätze, symmetric subspace methods, or explicit de Finetti rates. The complex part alone deserves a serious referee. For the real part, the authors need to fix Lemma 4.1 and carry out the calculation.\n\nMy recommendation: send it to peer review. The complex part is substantive and correct; the real error is local and probably repairable. A referee can check whether the corrected constants still improve on Reznick. Desk rejection would be wrong.","headline":"Explicit Chiribella inversion yields a solid complex Positivstellensatz and de Finetti bounds; the real section has a concrete arithmetic error in Lemma 4.1 and its claimed improvement over Reznick is currently unsupported.","tokens_in":22813,"tokens_out":4029,"would_cite":false,"duration_ms":33552,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims explicit, improved exponents for Reznick's Positivstellensatz by inverting the Chiribella identity from quantum cloning.","keywords":["Positivstellensatz","sum of squares","Reznick's theorem","Chiribella identity","measure-and-prepare map","exponential de Finetti theorem","spherical designs","quantum cloning"],"falsifier":"Apply Lemma 4.1 to $p(x) = (v_1 x_1 + \\cdots + v_d x_d)^{2n}$ with $\\|v\\|=1$. The proof's own computation gives $\\Delta p = 2n(2n-1)\\|v\\|^2\\langle x|v\\rangle^{2n-2}$, while the claimed partial trace is $\\|v\\|^2 v^{\\otimes(2n-2)}$, so each Laplacian step contributes the factor $2n(2n-1)$, not $(2n)^2$. Iterating, the ratio $\\Delta^{n-k} p / \\mathrm{tr}_{n\\to k}(p)$ is $(2n(2n-1))^{n-k}$; if this calculation stands, the real-case bound (19) does not follow from the proof as written.","tokens_in":21671,"feed_emoji":"🧮","tokens_out":9916,"duration_ms":85551,"temperature":0.7,"pith_summary":"This paper claims that Reznick's Positivstellensatz---a strictly positive homogeneous polynomial becomes a sum of squares once multiplied by a large enough power of the squared norm---can be reproved and sharpened by translating it into the language of quantum information theory. The translation represents polynomials by Hermitian operators on symmetric tensor powers, so that multiplication by $\\|x\\|^2$ and Laplacians become partial traces and their adjoints. The main result is an explicit bound on the required exponent $n$ in both the complex and the real case, together with a constructive sum-of-squares decomposition based on finite spherical designs. If the bounds are correct, they improve previously known constants and yield sharper exponential de Finetti theorems for quantum states.","feed_headline":"Quantum maps tighten Reznick's sum-of-squares exponents","feed_subtitle":"Explicit bounds turn positive polynomials into constructive sums of squares, with sharper de Finetti errors.","key_machinery":"The load-bearing mechanism is the Chiribella identity, which expresses the measure-and-prepare map $\\mathrm{MP}_{n\\to k}$ as a weighted sum of partial traces followed by their adjoints, i.e. by approximate-cloning maps. The paper's main technical move is an explicit inverse $\\Psi^{(n)}_{k\\to k}$ of the map $\\Phi^{(n)}_{k\\to k}$ appearing in that identity, with coefficients $q(n,k,t)$ given in closed form. Applying this inverse inside the adjoint of the measure-and-prepare map rewrites $p_W$ as an integral of a new form $p_{\\widetilde W}$; positivity of $p_{\\widetilde W}$ is then controlled by Bernstein-type inequalities for the Laplacian. Finite complex spherical designs convert the continuous integral into an explicit finite sum, yielding the sum-of-squares certificate.","core_discovery":"The paper's central claim is Theorem 3.1: for a Hermitian operator $W$ on $\\vee^k \\mathbb{C}^d \\otimes \\mathbb{C}^D$ with $m(W)>0$, whenever $n \\ge d k(2k-1)/\\ln(1 + m(W)/M(W)) - d - k + 1$, the identity $\\|x\\|^{2(n-k)} p_W(x,y) = \\int p_{\\widetilde{W}}(\\varphi,y) |\\langle\\varphi|x\\rangle|^{2n} d\\varphi$ holds with $p_{\\widetilde{W}} \\ge 0$, so the left-hand side is a sum of squares. For $k=1$ the bound improves to $n \\ge d M(W)/m(W) - d$. In the real case, Theorem 4.2 claims the analogous bound $2n \\ge d k(2k-1)/\\ln(1 + m(v)/M(v)) + 2 - 2k - d$, with $2n \\ge d M(v)/m(v) - d$ when $k=1$; the authors state that this improves Reznick's bound by shrinking the leading constant. The same inversion of the Chiribella identity gives Theorem 5.1, an exponential de Finetti theorem with error at most $\\delta^{r+1}/(1-3\\delta)$.","pith_inferences":["The explicit inverse of the Chiribella identity is a standalone operator identity: it expands partial traces into approximate-cloning channels, and de Finetti-type bounds are only the first place it is likely to be useful.","If the real-case constant in Lemma 4.1 is corrected, the qualitative structure of Theorem 4.2 should survive, but the exponent may grow by a constant factor; the complex theorem does not depend on that lemma.","A numerical scan over random low-dimensional Hermitian $W$ could test how close the analytic bounds (7) and (9) are to the minimal $n$ for which (8) holds, extending the Motzkin example the paper computes in the real case."],"forward_implications":["For $k=1$, a strictly positive complex bi-homogeneous form admits the representation as soon as $n \\ge d M(W)/m(W) - d$, a bound the paper compares against earlier complex-case results.","For general $k$, the required exponent grows like $d k(2k-1)/\\ln(1 + m(W)/M(W))$, giving a logarithmic rather than linear dependence on the ratio $M/m$.","Using finite complex spherical designs, the representation becomes an explicit sum of squares with at most $(n+k+1)^{2d}$ terms of the form $|\\langle\\varphi|x\\rangle|^{2n}|\\langle w_\\varphi^{(i)}|y\\rangle|^2$.","In the real case, the paper claims the leading constant in Reznick's bound is reduced; the same operator inversion yields an exponential de Finetti bound with error $\\delta^{r+1}/(1-3\\delta)$ and real parameter $\\delta_R = k(2k+d-2)/(2n+2k+d-2)$."],"supporting_citations":[{"why":"Supplies the real Positivstellensatz and the Bernstein inequality that the paper refines, and serves as the baseline for the claimed improvement.","marker":"[Rez95]"},{"why":"Supplies the measure-and-prepare identity that the paper inverts to obtain its main decomposition.","marker":"[Chi10]"},{"why":"Supplies the symmetric-subspace dictionary between polynomials and operators, the exposition of the Chiribella identity, and the connection to de Finetti theorems.","marker":"[Har13]"},{"why":"Supplies the earlier complex-case Positivstellensatz whose bounds are compared and improved.","marker":"[TY06]"},{"why":"Supplies Hobson's identity used in the real case as the analogue of the Chiribella identity.","marker":"[Hob31]"},{"why":"Supplies the optimal cloning maps that appear when partial traces are decomposed into cloning channels.","marker":"[KW99]"}],"fun_headline_variants":["Quantum tricks shrink Reznick's sum-of-squares bound","Chiribella identity tightens Positivstellensatz exponents","Quantum de Finetti bounds improved via partial traces","Sharper real and complex sum-of-squares exponents"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The real-case bound in Theorem 4.2 rests on Lemma 4.1, which asserts that the iterated real Laplacian equals $(2n)^{2(n-k)}$ times the partial trace; the displayed calculation in the lemma's proof appears to yield $2n(2n-1)$ per iteration instead, so unless that constant is corrected or the calculation reconciled, the claimed real-case improvement over Reznick is not established.","fun_headline_variants_meta":{"raw":{"variants":["Quantum tricks shrink Reznick's sum-of-squares bound","Chiribella identity tightens Positivstellensatz exponents","Quantum de Finetti bounds improved via partial traces","Sharper real and complex sum-of-squares exponents"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000862,"raw_usage":{"total_tokens":3753,"prompt_tokens":969,"completion_tokens":2784,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":585,"completion_tokens_details":{"reasoning_tokens":2717}},"tokens_in":585,"tokens_out":2784,"duration_ms":20975,"temperature":1.0,"reasoning_tokens":2717,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:12:11.691319+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply Lemma 4.1 to $p(x) = (v_1 x_1 + \\cdots + v_d x_d)^{2n}$ with $\\|v\\|=1$. The proof's own computation gives $\\Delta p = 2n(2n-1)\\|v\\|^2\\langle x|v\\rangle^{2n-2}$, while the claimed partial trace is $\\|v\\|^2 v^{\\otimes(2n-2)}$, so each Laplacian step contributes the factor $2n(2n-1)$, not $(2n)^2$. Iterating, the ratio $\\Delta^{n-k} p / \\mathrm{tr}_{n\\to k}(p)$ is $(2n(2n-1))^{n-k}$; if this calculation stands, the real-case bound (19) does not follow from the proof as written.","supporting_citations":[],"review_version":1}