{"id":"d5109f3f-2b28-4747-a79f-eca492afc9a2","arxiv_id":"2411.16103","paper_version":2,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper claims a new free Stein method with total-variation and Wasserstein Berry-Esseen rates for weakly dependent sums, but a load-bearing Cauchy integral step is not valid for the test functions used.","lead":"This paper develops a free, non-commutative version of Stein's method that aims to bound how far sums of weakly dependent non-commutative variables are from the semicircular law. Its headline results are Berry-Esseen rates in total variation and Wasserstein distance, but the proof contains an invalid analytic-continuation step.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 6.1 applies the Cauchy integral formula to arbitrary C^1 test functions in the d_TV dual class; the formula is valid only for holomorphic h, so the main cancellation estimates for Theorems 3 and 4 are not derived.","rationale":"The central claims are Theorems 3 and 4. The proof's only mechanism for converting the semigroup expression (6.1) into the resolvent estimates is the Cauchy formula applied to the test function h. Since the admissible h are not holomorphic, the conversion is invalid. This is not a matter of a missing technical hypothesis that can be added without changing the theorem: the d_TV and Wasserstein distances are defined by suprema over non-holomorphic functions, so the contour step cannot be repaired by adding smoothness of finite order. The dual free Stein equation and the semigroup interpolation in §4 are genuinely novel, and several intermediate lemmas (Propositions 2–4, Lemma 4) are plausible; Lemma 3 and the triangle inequality in (3.3) are independent of the contour step. But the main theorem proofs do not go through. The paper contains no code, data, or machine-checked formalization, and the statements of Theorems 3 and 4 omit the normalization and block-variance hypotheses that the proofs use. The reader's REJECT verdict is therefore appropriate; my concern matches the reader's primary objection, with the variance gap in Theorem 4 as a second, independent reason.","tokens_in":25663,"tokens_out":8231,"duration_ms":79849,"concrete_test":"Take h(x)=exp(1-1/(1-(x/4)^2)) for |x|<4 and 0 otherwise, a C^∞ bump in the unit ball of C^1 used for d_TV, and check the identity claimed after (6.1) with the rectangle R having vertices (±6,±1). Because h is compactly supported and non-analytic, ∮_R h(z)/(z-x) dz cannot equal h(x); for the extension h(z)=0 off the real axis the integral is zero while h(x) is nonzero on (-4,4). Re-running the argument in §6.1 with this h would therefore fail at the first Cauchy step; a repaired proof must obtain the same bound from (6.1) without the contour representation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive step is in §6.1, immediately after (6.1): for h in the class defining d_TV, the proof asserts h(x)=(2πi)^{-1}∮_R h(z)/(z-x) dz and Dh(x)=(2πi)^{-1}∮_R h(z)/(z-x)^2 dz over a rectangle enclosing [-5,5], and then uses this to rewrite ⟨S*[νn], L⊞[Dh]⟩ as integrals involving the resolvent g(x)=(z-x)^{-2}. This is the Cauchy integral formula and holds only when h is holomorphic in a neighborhood of the contour. But d_TV is a supremum over measurable |h|≤1, and even the paper's own restriction to C^1 bounded h includes compactly supported smooth bumps that have no holomorphic extension. The representation is therefore not available for the test functions over which the final supremum is taken, and the later symmetrization step, the introduction of g, and the semigroup estimates all depend on it. A separate gap in Theorem 4 is that Proposition 4 characterizes the standard semicircle s, while the proof compares µV with g[µV] of variance σ_V^2; no rescaling of L⊞ is supplied. Both defects are internal to the proof, not disagreements with prior results.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a non-commutative analogue of Stein's method for the semicircular distribution, built on the free Ornstein-Uhlenbeck semigroup and a 'dual free Stein equation' acting on measures. The main applications are quantitative free Berry-Esseen theorems: Theorem 3 claims a total-variation bound for sums of weakly dependent (in the free sense) centered variables, with rate n^{-1/2} in the homogeneous case, and Theorem 4 claims an improved Wasserstein bound under moment-matching conditions, with rate n^{-(q-1)/2}. The proofs interpolate between the summands and semicircular variables via the free Ornstein-Uhlenbeck semigroup, reduce the test function to a resolvent via the Cauchy integral formula, and then perform cancellation estimates using free independence and moment-matching assumptions.","tokens_in":26005,"tokens_out":7710,"duration_ms":69758,"significance":"If the results were correct, they would constitute a substantial contribution: a Stein-method proof of a free Berry-Esseen theorem in total variation for weakly dependent summands, and a Wasserstein refinement under moment matching, are both natural and potentially influential results. The paper is clearly organized and the dual free Stein equation idea is elegant. However, the central proof step—the application of the Cauchy integral formula to arbitrary C^1 test functions—is invalid, and the Wasserstein proof contains a variance mismatch that is not merely cosmetic. The paper also relies on an unproved expansion (4.13) and on an unpublished reference for a key Wasserstein subadditivity lemma. These are load-bearing gaps, so the central claims are not established in the present manuscript.","major_comments":[{"comment":"The proof applies the Cauchy integral formula to an arbitrary bounded C^1 test function h, writing h(x)=(2πi)^{-1}∮_R h(z)/(z-x) dz and Dh(x)=(2πi)^{-1}∮_R h(z)/(z-x)^2 dz over the rectangle with vertices (±6,±1), and then uses this representation to replace L⊞[Dh] by expressions involving the resolvent g(x)=(z-x)^{-2}. This formula is valid only for h holomorphic in a neighborhood of the contour; the class of C^1 bounded functions defining the total-variation dual includes compactly supported smooth bumps that have no holomorphic extension, and the supremum defining d_TV is over measurable functions. Since the subsequent symmetrization identity, the introduction of g, and all cancellation estimates for Y_{θ,z} depend on this representation, the proof of Theorem 3 does not establish the claimed bound.","section":"Section 6.1, after Eq. (6.1)"},{"comment":"The theorem statements omit hypotheses that the proof uses and that appear only in the corollaries. Theorem 3 asserts the bound for any sequence ξ_n with dependency graph, but the proof uses that the summands are centered (to obtain τ[η_θ^V g(F^V_{θ,n})]=0), that the total variance is one (to identify the limit as the standard semicircle s), and that the supports are uniformly bounded (to obtain Supp ν_n ⊂ [-3,3] and Supp P_θ[ν_n] ⊂ [-5,5] via superconvergence). Without these assumptions the statement is false as written, for example for uncentered summands or for summands whose variances do not sum to one. Theorem 4 is stated without defining q and without listing the normalization and moment-matching hypotheses used in its proof.","section":"Theorems 3 and 4; Corollaries 1 and 2"},{"comment":"Proposition 4 and the operator L⊞ characterize the standard semicircular distribution s with variance one. In the proof of Theorem 4, the same operator is applied to ⟨µV,h⟩ - ⟨g[µV],h⟩, where g[µV] is the semicircular distribution with the variance of µV, which is not equal to one in general. The needed rescaling of L⊞ and P_θ for non-unit variance is not supplied, so equation (6.6) does not follow from Proposition 4. Moreover, the decomposition s = ⊞_{V∈J} g[µV] used in (6.4) requires the variances of the blocks to sum to one, a normalization that is absent from Theorem 4.","section":"Section 6.2, Eqs. (6.4)-(6.6)"},{"comment":"The expansion (a Δ(a,r))^j = ∑_{α∈I} (f^1_{j,α} Q^1_{j,α}(z) Υ^1_{j,α}(a,r) + f^2_{j,α} Q^2_{j,α}(z) Υ^2_{j,α}(a,r)) is asserted with 'universal constants' and polynomials Q that are never defined, and it is used in the proof of Theorem 4 to separate moments of ξ_θ^V from terms involving g(ψ_θ^V) and to show independence of θ. Lemma 4, on which the proof of Theorem 3 also relies, is not actually proved: the displayed argument assumes the identity for q and reduces the induction to the q=1 case, which is only asserted to follow from a direct computation and is not carried out. These are load-bearing gaps for the moment-matching Wasserstein result.","section":"Section 4.5, Eq. (4.13); Section 7, Lemma 4"}],"minor_comments":[{"comment":"Lemma 2 states that S*_⊞[µ] is a signed measure, but the sentence 'The fact that S*_⊞[µ] is a well-defined probability measure' conflicts with that; moreover, the proof only bounds the pairing with Lipschitz functions and does not construct the signed measure.","section":"Section 4.4, Lemma 2"},{"comment":"The Wasserstein subadditivity for free convolution in (3.3) is cited to the unpublished preprint [20]; since it is used in (6.4) in the proof of Theorem 4, the authors should either include a proof or provide a published reference.","section":"Section 3.2, Lemma 3.2"},{"comment":"The moment matching rank q is defined using 'mj[ν] = mj[g[ν]]', but no measure ν has been defined in the statement of the corollary; the intended statement is presumably about the µ_{k,n} or about the whole convolution, and this needs to be corrected.","section":"Corollary 2"},{"comment":"There are numerous typos and notational inconsistencies, including 'In oder' (p.6), 'Lipchitz' (p.19), 'well-possednes' (p.15), the missing bracket in '⟨P∞µn,h⟩' (p.20), and the inconsistent use of P_θ versus P*_θ in Proposition 3.","section":"Throughout"},{"comment":"The proof introduces q, but the theorem statement does not define q; in addition, the hypothesis lim_n D(E_n)^2 Σ m_3[µ_{k,n}] = 0 is not used in the argument, and the relation between the matching rank of individual blocks and the rank of the whole convolution should be stated explicitly.","section":"Section 6.2, proof of Theorem 4"}],"recommendation":"reject","confidential_remarks":"To the editor: the main proof hinges on an invalid use of the Cauchy integral formula, so the total-variation result is not established; the Wasserstein result has an additional variance mismatch that appears to be a genuine obstruction. The reliance on [20] for Lemma 3.2 and the unproved expansion (4.13) also make verification difficult. I do not see a short local fix within the present framework, and therefore I recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: the core idea—a dual free Stein equation built on the free Ornstein-Uhlenbeck semigroup—is genuinely new, and the paper is worth engaging. The main proofs, however, are not valid as written. In Section 6.1, right after (6.1), the proof invokes the Cauchy integral formula for an arbitrary bounded C^1 test function h, writing h(x) = (2πi)^{-1} ∮ h(z)/(z-x) dz over a rectangle. That formula holds only for holomorphic h. The TV dual class contains compactly supported smooth bumps, which have no holomorphic extension, so the representation is unavailable for the functions over which the final supremum is taken. All subsequent cancellation estimates, the introduction of g(x) = (z-x)^{-2}, and the semigroup bounds depend on this step. This is load-bearing, not a minor gap.\n\nWhat the paper does well: the dual Stein equation avoids solving the free Stein equation directly, which is a real departure from the Cauchy/cumulant transform literature. The semigroup interpolation is natural and clearly explained. The literature review is honest and gives credit, including to the unpublished [20] used for Lemma 3.2. The ambition of the theorem statements is clear.\n\nOther soft spots are real but secondary. Theorems 3 and 4 as stated omit centering, bounded support, and total variance normalization; the corollaries have them, but the theorems don't, and the proofs silently assume them. The bound in Theorem 4 cannot hold without total variance 1, since the target is the standard semicircle. In the proof of Theorem 4, the Stein operator L⊃ for the standard semicircle is applied to blocks μV whose semicircular targets have variance σ_V^2; no rescaling is supplied. The expansion (4.13) is asserted without proof. None of these are as serious as the Cauchy gap.\n\nThe argument is not circular, and the citation pattern is acceptable overall; the reliance on an unpublished manuscript is a minor concern. No code or data is involved, and none is needed.\n\nThis paper deserves a serious referee—the idea is novel and the flaw, while central, is identifiable and possibly repairable. If the Cauchy step can be replaced by a valid representation or approximation argument, the results would be a substantial advance. As it stands, it should not be accepted. Who this is for: researchers working on free Stein's method, quantitative free CLT, and free probability. I would not cite it yet, but I would keep an eye on a revised version.","headline":"A genuinely novel dual free Stein equation built on the free Ornstein-Uhlenbeck semigroup, but the main proofs rest on an invalid application of the Cauchy integral formula to non-holomorphic test functions.","tokens_in":26459,"tokens_out":4643,"would_cite":false,"duration_ms":42204,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60F05","46L53","60G50","60B10"],"pacs":[],"model":"deepseek-v4-flash","headline":"A non-commutative Stein method bounds semicircular approximation error in total variation and Wasserstein distance.","keywords":["free probability","Stein's method","semicircular distribution","free Berry-Esseen theorem","total variation distance","Wasserstein distance","free Ornstein-Uhlenbeck semigroup","dependency graphs"],"falsifier":"Choose a smooth, compactly supported, non-holomorphic test function, such as a bump function, and compute both sides of the identity $\\langle s,h\\rangle - \\langle\\nu_n,h\\rangle = \\int_0^\\infty \\langle P^*_\\theta[\\nu_n]\\otimes P^*_\\theta[\\nu_n] - s\\otimes s, L_{\\boxplus}[Dh]\\rangle\\,d\\theta$ for an explicit inhomogeneous family $\\nu_n$; if the right-hand side is not the value given by the Cauchy substitution used in the paper, then the derivation of the bound (1.2) does not go through for this $h$. Equivalently, the bound's universal constant can be checked against a direct numerical evaluation of $d_{TV}$ for simple summands.","tokens_in":25422,"feed_emoji":"🎲","tokens_out":9586,"duration_ms":78940,"temperature":0.7,"pith_summary":"This paper develops a Stein-type method for the semicircular distribution in free probability, using the free Ornstein-Uhlenbeck semigroup to interpolate between a candidate measure and the semicircle law. The method yields quantitative estimates for two distances: total variation and the non-commutative Wasserstein (Kantorovich-Rubinstein) distance, between the free convolution of many small centered measures and the semicircle. In the homogeneous, freely independent case the estimates are of order $n^{-1/2}$ in total variation and $n^{-(q-1)/2}$ in Wasserstein distance when the summands match the semicircle's moments up to order $q$. The bounds also accommodate weak dependence, encoded by dependency graphs, with the maximum degree of the graph entering polynomially into the error.","feed_headline":"Free Stein method gives total-variation Berry-Esseen rates","feed_subtitle":"Total variation error drops as n^{-1/2}; matching more moments accelerates Wasserstein convergence.","key_machinery":"The central object is the operator $L_{\\boxplus}[g](x,y) = -xg(x) + (g(y)-g(x))/(y-x)$, which plays the role of the classical Stein operator for the semicircle, together with the dual free Stein equation built from the free Ornstein-Uhlenbeck semigroup $P^*_\\theta[\\mu] = D_{e^{-\\theta}}[\\mu] \\boxplus D_{\\sqrt{1-e^{-2\\theta}}}[s]$. The semigroup interpolates between $\\mu$ at $\\theta=0$ and the semicircle at $\\theta=\\infty$, and the difference $\\langle s,h\\rangle-\\langle\\mu,h\\rangle$ is written as an integral over $\\theta$ of $\\langle P^*_\\theta[\\mu]\\otimes P^*_\\theta[\\mu], L_{\\boxplus}[Dh]\\rangle$. The cancellations then rely on a non-commutative Taylor expansion (Lemma 4) with $\\Delta(a,r)=2s[(z-a)r]-r^2$, applied to the resolvent $g(x)=(z-x)^{-2}$. This decomposition is what converts the third moments of the summands into the error bound.","core_discovery":"The paper's central claim is that the semicircular Stein equation can be replaced by a dual free Stein equation: for a probability measure $\\mu$, $\\langle s,h\\rangle - \\langle \\mu,h\\rangle = \\langle S^*_{\\boxplus}[\\mu], L_{\\boxplus}[Dh]\\rangle$, where $S^*_{\\boxplus}[\\mu] = \\int_0^\\infty (P^*_\\theta[\\mu]\\otimes P^*_\\theta[\\mu] - s\\otimes s)\\,d\\theta$ and $P^*_\\theta[\\mu] = D_{e^{-\\theta}}[\\mu] \\boxplus D_{\\sqrt{1-e^{-2\\theta}}}[s]$ is the free Ornstein-Uhlenbeck semigroup. The operator $L_{\\boxplus}[g](x,y) = -xg(x) + (g(y)-g(x))/(y-x)$ characterizes the standard semicircle in the sense that $\\mu=s$ if and only if $\\langle \\mu\\otimes\\mu, L_{\\boxplus}[f]\\rangle=0$ for all $f\\in C^1$. Using this identity, the paper proves that centered, uniformly bounded summands with total variance one and dependency graph of maximum degree $D(E_n)$ satisfy $d_{TV}(\\mu_{1,n}\\boxplus\\cdots\\boxplus\\mu_{n,n}, s) \\le C D(E_n)^2 \\sum_k m_3[\\mu_{k,n}]$, and with moment-matching rank $q$, $d_W(\\nu_n,s)\\le C D(E_n)^{q+1}\\sum_k m_{q+1}[\\mu_{k,n}]$. The proof transfers the classical Stein cancellation to the resolvent $g(x)=(z-x)^{-2}$ through a Cauchy integral representation of the test function $h$, which is the step that carries the main technical assumption.","pith_inferences":["Editorial inference: the same interpolation should apply to non-unit-variance semicircle targets by conjugating $L_{\\boxplus}$ with a dilation; this would mend the comparison in Theorem 4 between blocks $\\mu_V$ and $g[\\mu_V]$ when the block variance is not one.","Editorial inference: the dependency-graph setup suggests the rate estimates should transfer to random-matrix ensembles with sparse interaction structures, where asymptotic freeness replaces exact freeness; the paper does not state this application.","Editorial inference: one could test the sharpness of the exponents by constructing summands with near-zero third moment but large fourth moment, to see whether the Wasserstein bound can be improved beyond the stated power of $D(E_n)$."],"forward_implications":["The free Berry-Esseen theorem holds under total variation with rate $n^{-1/2}$ in the homogeneous, freely independent case, matching the classical Stein-method rate.","Moment matching up to order $q$ accelerates the Wasserstein convergence to $n^{-(q-1)/2}$, showing that the third-moment improvement is a special case of a general hierarchy.","Weak dependence enters the error only through the maximum degree $D(E_n)$, so the same bounds cover partial freeness, including sparse dependence structures among the summands.","The dual free Stein equation gives a probabilistic route to free limit theorems that does not rely on analytic Cauchy-transform or R-transform computations."],"supporting_citations":[{"why":"Supplies the superconvergence result that confines the support of the free convolution and of its semigroup evolution to fixed intervals, a step used before applying the Cauchy formula.","marker":"[3]"},{"why":"Provides the baseline free Berry-Esseen bound in Kolmogorov distance against which the paper compares its total-variation rates.","marker":"[16]"},{"why":"The classical Stein method whose semigroup perspective the paper adapts to the non-commutative setting.","marker":"[30]"},{"why":"Establishes the link between the free Ornstein-Uhlenbeck semigroup and the non-commutative derivative that underlies the dual Stein equation.","marker":"[21]"},{"why":"Uses semigroup interpolation for a quantitative fourth moment theorem and is the closest technical precedent for the paper's computations.","marker":"[13]"},{"why":"Defines the non-commutative Wasserstein distance used in Theorem 4.","marker":"[6]"},{"why":"Supplies the free cumulant formula for mixed moments used in the moment-matching proof of Theorem 4.","marker":"[25]"}],"fun_headline_variants":["Free Stein equation gives TV Berry-Esseen and faster Wasserstein rates","Noncommutative Stein: Berry-Esseen under total variation","Free Stein method: moment-matching boosts Wasserstein decay","Semicircular approximation: free Stein yields optimal rates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The total-variation proof assumes that every $C^1$ test function $h$, and its derivative, can be written through the Cauchy integral formula on a rectangle around the support of the measures; this representation is valid only for holomorphic functions, and the paper does not state or prove that the $C^1$ functions defining the distance are holomorphic.","fun_headline_variants_meta":{"raw":{"variants":["Free Stein equation gives TV Berry-Esseen and faster Wasserstein rates","Noncommutative Stein: Berry-Esseen under total variation","Free Stein method: moment-matching boosts Wasserstein decay","Semicircular approximation: free Stein yields optimal rates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000469,"raw_usage":{"total_tokens":2400,"prompt_tokens":1071,"completion_tokens":1329,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":687,"completion_tokens_details":{"reasoning_tokens":1258}},"tokens_in":687,"tokens_out":1329,"duration_ms":9689,"temperature":1.0,"reasoning_tokens":1258,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:34:25.198728+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Choose a smooth, compactly supported, non-holomorphic test function, such as a bump function, and compute both sides of the identity $\\langle s,h\\rangle - \\langle\\nu_n,h\\rangle = \\int_0^\\infty \\langle P^*_\\theta[\\nu_n]\\otimes P^*_\\theta[\\nu_n] - s\\otimes s, L_{\\boxplus}[Dh]\\rangle\\,d\\theta$ for an explicit inhomogeneous family $\\nu_n$; if the right-hand side is not the value given by the Cauchy substitution used in the paper, then the derivation of the bound (1.2) does not go through for this $h$. Equivalently, the bound's universal constant can be checked against a direct numerical evaluation of $d_{TV}$ for simple summands.","supporting_citations":[{"cited_title":"Bercovici and D","cited_arxiv_id":null,"evidence_quote":"Supplies the superconvergence result that confines the support of the free convolution and of its semigroup evolution to fixed intervals, a step used before applying the Cauchy formula."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the baseline free Berry-Esseen bound in Kolmogorov distance against which the paper compares its total-variation rates."},{"cited_title":"A bound for the error in the normal approximatio n to the distribution of a sum of dependent random variables","cited_arxiv_id":null,"evidence_quote":"The classical Stein method whose semigroup perspective the paper adapts to the non-commutative setting."},{"cited_title":"Wigner chaos and the fourth mo- ment","cited_arxiv_id":null,"evidence_quote":"Establishes the link between the free Ornstein-Uhlenbeck semigroup and the non-commutative derivative that underlies the dual Stein equation."},{"cited_title":"A quantitative fourth moment theorem in fre e probability theory","cited_arxiv_id":null,"evidence_quote":"Uses semigroup interpolation for a quantitative fourth moment theorem and is the closest technical precedent for the paper's computations."},{"cited_title":"Biane and D","cited_arxiv_id":null,"evidence_quote":"Defines the non-commutative Wasserstein distance used in Theorem 4."},{"cited_title":"Lectures on the combinatorics of free probability , volume 335 of London Mathematical Society Lecture Note Series","cited_arxiv_id":null,"evidence_quote":"Supplies the free cumulant formula for mixed moments used in the moment-matching proof of Theorem 4."}],"review_version":1}