{"id":"2851e666-c236-4bfd-b475-810acf84195f","arxiv_id":"1909.01142","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The scaled fluctuations of the Circular beta-Ensemble at inverse temperature beta/N converge to a Gaussian process with variance <psi, L^{-1} psi>_H, interpolating from the L2 norm to the H^{1/2} norm.","lead":"This paper proves that at inverse temperature beta/N, the density fluctuations of the Circular beta-Ensemble around its equilibrium converge to a Gaussian field with an explicit covariance operator that interpolates between the independent-particle and random-matrix regimes. It provides concentration estimates and a W2 convergence rate, making the high-temperature crossover quantitative.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The advertised W2 rate in Theorem 1.2 is not obtained by the proof: the displayed estimates give sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}, not N^{-(gamma-1)/(gamma+1)}.","rationale":"The reader's verdict is CONDITIONAL and its rationale already flags the rate mismatch between the displayed estimates and the theorem statement. My stress-test confirms that this is a concrete, load-bearing internal inconsistency: the proof, taken at face value, does not yield the advertised W2 rate, although it does yield a slower rate that still tends to zero and thus preserves the CLT convergence. I agree with the conditional verdict. I do not elevate to REJECT because the main probabilistic statement -- the Gaussian fluctuation with the stated covariance -- is supported by the proof chain up to this rate error, and the rate may be repairable by a sharper tail estimate for the projected test function. I chose not to make the spectral self-adjointness issue the headline concern: it is a genuine gap in the written proof, but it is likely fixable by standard form methods, whereas the rate mismatch is an explicit discrepancy between the theorem statement and the estimates proved in the same section. The agreement is 'partial' because the reader's stated weakest_assumption is the spectral structure of L, while my headline concern is the quantitative rate, though the reader's rationale also mentions the rate issue.","tokens_in":29477,"tokens_out":13546,"duration_ms":135842,"concrete_test":"Independently re-derive the exponent by substituting kappa_j ~ alpha j^2 into (58) and (60) and minimizing over d ~ N^a. For gamma=2, the optimal a is 1/12, yielding sqrt(log N) N^{-1/6}, not the N^{-1/3} stated in Theorem 1.2. If the same computation is repeated for general gamma, the minimal exponent is (gamma-1)/(2(gamma+1)); this settles that the claimed rate requires a genuinely new estimate rather than a different choice of truncation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 1.2 (Section 4), the three bounds (58), (59) and (60) are combined by taking d to be the integer part of N^{1/(4(gamma+1))}. Substituting the Weyl asymptotics kappa_j ~ alpha j^2 from Proposition 4.3(c), estimate (58) gives W2(nu_N(psi), nu_N(psi[d])) <= C sqrt(log N) d^{-2(gamma-1)} = C sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}. The dominant term in (60) is the second one, sqrt(log N) d^4 / sqrt(N), which at this choice of d is also sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}. Hence the proof establishes only W2 <= C sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}, whereas Theorem 1.2 states the faster rate N^{-(gamma-1)/(gamma+1)}. For gamma=2 the stated exponent is 1/3, but the proof gives only 1/6. The weak convergence nu_N(psi) => N(0, sigma^2) is unaffected, so the CLT itself may still be correct; however the quantitative W2 rate, which is an advertised contribution of the paper, is not supported by the argument as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Circular beta-Ensemble on the torus in the high-temperature regime where the inverse temperature scales as beta/N, with an external potential V. The main object is the signed fluctuation field nu_N = sqrt(N)(mu_N - mu^V_beta) around the equilibrium measure mu^V_beta, which minimizes the functional beta E + relative entropy. Theorem 1.2 states a CLT for linear statistics nu_N(psi) with explicit variance sigma^V_beta(psi)^2 = <psi, L^{-1} psi>_H, where L is the Sturm-Liouville-type operator in (15), together with a quantitative Wasserstein-2 rate sqrt(log N) N^{-(gamma-1)/(gamma+1)}. The proof combines a concentration estimate in W1 (Theorem 1.4), obtained from a Coulomb transport inequality and an energy regularization, with the Stein-type normal approximation of Lambert-Ledoux-Webb and a spectral decomposition of L. Sections 2-8 develop the equilibrium measure properties, the concentration proof, the eigenbasis regularity, and the beta -> 0/infty behavior of the variance.","tokens_in":29725,"tokens_out":11794,"duration_ms":119458,"significance":"The CLT at high temperature is a natural and interesting result: it gives an explicit crossover covariance structure interpolating between the L2 noise of independent particles at beta = 0 and the H^{1/2} noise of the usual beta-ensemble at beta = infinity, and it provides quantitative convergence in W2. The overall strategy is coherent and builds on solid external tools, and the spectral analysis of the operator L is a genuine contribution. The concentration inequality of Theorem 1.4 is also of independent interest. However, the advertised W2 rate in Theorem 1.2 is not supported by the estimates as written, and there are gaps in the spectral self-adjointness argument; these affect a stated quantitative result, though the CLT itself appears salvageable.","major_comments":[{"comment":"The stated W2 rate in Theorem 1.2 is not what the proof delivers. Taking d to be the integer part of N^{1/(4(gamma+1))}, estimate (58) gives W2(nu_N(psi), nu_N(psi[d])) <= C sqrt(log N) d^{-2(gamma-1)} = C sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}. The dominant term in (60) is of the same order, or slower if the displayed Proposition 4.4 bound is used literally. Hence the proof establishes only W2 <= C sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}, not the claimed N^{-(gamma-1)/(gamma+1)}. The weak convergence is unaffected, but the quantitative rate in the theorem and the related claim in Remark 1.1 must be corrected.","section":"Section 4, proof of Theorem 1.2, Eqs. (58)-(60)"},{"comment":"Self-adjointness of W on H is asserted without proof. The text states 'Since W is non-negative and self-adjoint on H', but only positivity of the quadratic form is shown in Lemma 6.1. Positivity of a quadratic form does not imply symmetry or self-adjointness for an unbounded operator, and this property is load-bearing for the spectral theorem and the min-max comparison that yield the eigenvalue asymptotics. Please provide a domain and a proof of self-adjointness, for instance by an integration by parts using H* = -H and the equilibrium equation, or cite a precise reference.","section":"Section 6, Proposition 6.3"},{"comment":"There is an inconsistency in the d-dependence of the second error term. Proposition 4.4 states the second contribution as sqrt(log N) sqrt(sum kappa_j^2) sum kappa_j, which for kappa_j ~ alpha j^2 is of order sqrt(log N) d^{11/2}. Equation (60), however, writes sqrt(log N) d^4. The square-root of the bound in (55) naturally gives sqrt(log N) sqrt(sum kappa_j^2) sqrt(sum kappa_j), which is d^4; the displayed Proposition 4.4 is weaker. The two forms are not interchangeable as written, and this discrepancy affects the resulting rate in Theorem 1.2. Please reconcile the statement of Proposition 4.4 with its proof and with equation (60).","section":"Section 4, Proposition 4.4 and Eq. (60)"}],"minor_comments":[{"comment":"There is a stray period in the statement: 'kappa_j ~ alpha j^2 as j -> infinity. for every j >= 1' should be 'kappa_j ~ alpha j^2 as j -> infinity, for every j >= 1'.","section":"Proposition 4.3(c)"},{"comment":"There is a typo in the abstract: 'equilibrium mesure' should be 'equilibrium measure'. Similar small typos appear elsewhere and should be cleaned up.","section":"Abstract and Introduction"},{"comment":"The remark that the rate for C^infty test functions is sqrt(log N)/N is based on the currently stated exponent; if the correct rate is the weaker one identified above, this remark must be updated accordingly.","section":"Remark 1.1"},{"comment":"The citation to [Chafai, Hardy, Maida, Theorem 1.1] and 'the discussion below' could be made more precise by specifying the exact form of the Coulomb transport inequality used in (41).","section":"Section 3, proof of Theorem 1.4"}],"recommendation":"major_revision","confidential_remarks":"The central CLT appears sound and the paper has a coherent line of proof; the main issue is that the advertised quantitative W2 rate is not obtained by the displayed estimates. This is a load-bearing statement but is locally fixable by correcting the rate and reconciling Proposition 4.4 with (60). I also recommend that the authors add the missing self-adjointness argument for W, since the spectral decomposition is central to the proof."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I gave the paper a careful read. The headline is that this is the first explicit CLT for the circular beta-ensemble in the beta/N high-temperature regime, with a covariance operator that interpolates between L2 and H^{1/2}. The proof is a real chain: Coulomb transport inequality plus concentration (Thm 1.4), Stein-type normal approximation from Lambert-Ledoux-Webb, and a detailed spectral analysis of L = A + 2πβW. The V=0 case gives the transparent formula sigma^2 = 2 sum_k |hat psi_k|^2 / (1 + beta/k), and the general variance is given as <psi, L^{-1} psi>_H. That is genuinely new; the earlier Gaussian-beta results did not have an explicit variance.\n\nThe soft spot is exactly where the stress-test note lands. In the proof of Thm 1.2, estimates (58)–(60) are combined with d = N^{1/(4(gamma+1))}. Substituting this d into (58) gives sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}, and the second term of (60) gives the same order. So the proof establishes only W2 <= C sqrt(log N) N^{-(gamma-1)/(2(gamma+1))}. The stated rate N^{-(gamma-1)/(gamma+1)} is not obtained. For gamma=2 this is N^{-1/6} instead of N^{-1/3}. The CLT convergence is unaffected; this is a quantitative claim that the argument does not support. The authors can either correct the statement to match the proof or find a sharper argument. This is material and should be fixed before publication.\n\nOne smaller point: in Prop 6.3, the self-adjointness of W on H is asserted rather than shown; given the explicit form W phi = -H(phi' mu) and the regularity of mu, it is likely true, but the proof should include the one-line symmetry computation.\n\nAside from that, the paper is careful, the references are appropriate (the self-citations are the actual tools being used), and there are no fitted parameters. The concentration inequality with explicit constant for V=0 is a nice bonus.\n\nWho is this for? Random matrix theorists and anyone working on high-temperature Coulomb gases or Stein-method rates. It deserves a serious referee: the main result is important and the overall structure is sound, with one quantitative claim that needs repair. My recommendation: send to peer review, ask for the rate statement to be reconciled with the proof.","headline":"First explicit CLT at the beta/N crossover for circular beta ensembles, with an honest proof chain; but the advertised W2 rate is not obtained by the argument as written.","tokens_in":30314,"tokens_out":3246,"would_cite":true,"duration_ms":29887,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60F05","60B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"At high temperature, the circular beta-ensemble's linear statistics satisfy a central limit theorem with an explicit covariance that interpolates between white noise and random-matrix noise.","keywords":["circular beta-ensemble","high temperature","central limit theorem","linear statistics","Gaussian fluctuations","equilibrium measure","Sturm-Liouville spectrum","Wasserstein metric"],"falsifier":"For $V=0$, simulate the particle system at fixed $\\beta$ and measure the variance of $N^{-1/2}\\sum_i\\cos(x_i)$; Theorem 1.2 predicts it approaches $1/(2(1+\\beta))$, so a clear mismatch at large $N$ would refute the CLT. More directly, one can numerically diagonalize a discretization of $L$ for a non-constant smooth $V$ and check whether the eigenvalue ratios $\\kappa_j/j^2$ converge to a positive constant and whether the stated eigenfunction Lipschitz bounds hold.","tokens_in":29216,"feed_emoji":"🎲","tokens_out":12045,"duration_ms":111521,"temperature":0.7,"pith_summary":"This paper proves a central limit theorem for the linear statistics of the circular $\\beta$-ensemble when the inverse temperature is scaled as $\\beta/N$, the regime where the system crosses over from independent particles to random-matrix behavior. It shows that for a smooth potential $V$ and a smooth mean-zero test function $\\psi$, the normalized fluctuation $N^{-1/2}\\sum_{i=1}^N \\psi(x_i)$ converges in distribution to a Gaussian with variance $\\langle\\psi, L^{-1}\\psi\\rangle_H$, where $L$ is an explicit differential operator built from the equilibrium measure. The variance interpolates between the $L^2$ norm at $\\beta=0$ and the Sobolev $H^{1/2}$ semi-norm at $\\beta=\\infty$, so one family of formulas covers both extremes. The paper also supplies a rate of convergence in the Wasserstein $W_2$ metric that improves with the smoothness of the test function. If correct, this gives a quantitative description of the full temperature crossover for fluctuations of this particle system.","feed_headline":"Hot circular beta-ensembles fluctuate like a Gaussian field","feed_subtitle":"The limiting variance interpolates between white-noise statistics and random-matrix statistics as temperature drops.","key_machinery":"The load-bearing object is the operator $L=A+2\\pi\\beta W$ acting on the Hilbert space $H=\\{\\psi\\in L^2(\\mathbb{T}):\\psi'\\in L^2(\\mathbb{T}),\\int\\psi\\,d\\mu_\\beta^V=0\\}$ with inner product $\\langle\\phi,\\psi\\rangle_H=\\int \\phi'\\psi'\\,d\\mu_\\beta^V$. Here $A\\phi=-\\phi''-(\\log\\mu_\\beta^V)'\\phi'$ is a Sturm-Liouville operator and $W\\phi=-H(\\phi'\\mu_\\beta^V)$ involves the Hilbert transform $H$. Proposition 4.3 supplies a complete orthonormal eigenbasis $(\\varphi_j)$ of $H$ with eigenvalues $\\kappa_j\\sim\\alpha j^2$ and Lipschitz bounds $\\|\\varphi_j^{(k)}\\|_{\\mathrm{Lip}}\\le C_k\\kappa_j^{(k+1)/2}$. This spectral structure is what lets the proof decompose an arbitrary test function into eigenmodes, truncate the expansion with controllable error, and apply a normal-approximation theorem to the finite-dimensional projections.","core_discovery":"The central discovery is that, in the high-temperature scaling where the inverse temperature is $2\\beta/N$, the random measure $\\nu_N=\\sqrt{N}(\\mu_N-\\mu_\\beta^V)$ has Gaussian macroscopic fluctuations with an explicit covariance operator. For $\\beta>0$, $V\\in C^{3,1}(\\mathbb{T})$ and $\\psi\\in C^{2\\gamma+1}(\\mathbb{T})$, $\\gamma\\ge 2$, with $\\int\\psi\\,d\\mu_\\beta^V=0$, Theorem 1.2 states that $\\nu_N(\\psi)=N^{-1/2}\\sum_i\\psi(x_i)$ converges in law to $\\mathcal{N}(0,\\sigma_\\beta^V(\\psi)^2)$ with $\\sigma_\\beta^V(\\psi)^2=\\langle\\psi,L^{-1}\\psi\\rangle_H=\\int \\psi'(L^{-1}\\psi)'\\,d\\mu_\\beta^V$, and $W_2(\\nu_N(\\psi),\\mathcal{N}(0,\\sigma^2))\\le C\\sqrt{\\log N}\\,N^{-(\\gamma-1)/(\\gamma+1)}$. Here $\\mu_\\beta^V$ is the unique minimizer of $\\beta E(\\mu)+K(\\mu|\\mu_0^V)$, the operator $L$ is defined by $-L\\phi=\\phi''+2\\pi\\beta H(\\mu_\\beta^V\\phi')+(\\log\\mu_\\beta^V)'\\phi'$, and $H$ is the Hilbert transform. In the unweighted case $V=0$, the variance is $\\sigma_\\beta^0(\\psi)^2=2\\sum_{k\\ge1}\\frac{1}{1+\\beta/k}|\\widehat{\\psi}_k|^2$, which visibly interpolates between the $L^2$ norm and the $H^{1/2}$ semi-norm.","pith_inferences":["The same operator-level strategy would likely extend to Gaussian beta-ensembles on the real line and to general potentials, giving an explicit limiting variance in settings where only implicit central limit theorems were previously available.","The variance formula suggests a two-parameter family of Gaussian fields parametrized by $\\beta$; at finite $\\beta$ the fluctuations are neither pure white noise nor pure random-matrix noise, which could be tested numerically as a systematic interpolation between the two known regimes.","The paper leaves open whether the condition $\\beta\\inf\\mu_\\beta^V\\to\\infty$ can fail for some smooth non-constant potential; if it does fail, the $H^{1/2}$ universality at $\\beta=\\infty$ would break and the limiting variance would retain a dependence on $V$.","A direct numerical check of the unweighted formula $\\sigma_\\beta^0(\\psi)^2=2\\sum_{k\\ge1}\\frac{1}{1+\\beta/k}|\\widehat{\\psi}_k|^2$, mode by mode, would isolate the operator's spectrum from other sources of error and sharpen confidence in the whole proof."],"forward_implications":["At high temperature, the macroscopic fluctuations of the circular beta-ensemble are Gaussian with a covariance operator that interpolates continuously between the $L^2$ white-noise covariance at $\\beta=0$ and the $H^{1/2}$ random-matrix covariance at $\\beta=\\infty$.","The rate of convergence in the Wasserstein $W_2$ metric is $O(\\sqrt{\\log N}\\,N^{-(\\gamma-1)/(\\gamma+1)})$ for $\\psi\\in C^{2\\gamma+1}$, so smoother test functions yield faster convergence; for $C^\\infty$ test functions the rate is $O(\\sqrt{\\log N}/N)$.","For $V=0$, the limiting variance has the explicit Fourier formula $2\\sum_{k\\ge1}\\frac{1}{1+\\beta/k}|\\widehat{\\psi}_k|^2$, reducing to the classical formulas in the limits $\\beta\\to0$ and $\\beta\\to\\infty$.","The equilibrium measure $\\mu_\\beta^V$ is the unique minimizer of $\\beta E(\\mu)+K(\\mu|\\mu_0^V)$, has a density bounded above and below by positive constants, and is as smooth as the potential $V$.","The concentration estimate of Theorem 1.4 gives exponential tail bounds for $W_1(\\mu_N,\\mu_\\beta^V)$, which imply almost sure convergence of the empirical measure whenever $\\beta\\gg N^{-1}$."],"supporting_citations":[{"why":"Supplies the quantitative normal-approximation theorem (Theorem 4.5) that turns spectral estimates into the $W_2$ bound.","marker":"[Lambert, Ledoux, and Webb, 2017]"},{"why":"Supplies the Coulomb transport inequality $W_1^2\\le 4\\pi E$ used to prove the concentration estimate Theorem 1.4.","marker":"[Chafaï, Hardy, and Maïda, 2018]"},{"why":"Provides the fixed-temperature circular beta-ensemble CLT whose $H^{1/2}$ variance is the $\\beta\\to\\infty$ endpoint.","marker":"[Johansson, 1988]"},{"why":"Provides the Sturm-Liouville spectral theory and Weyl asymptotic used to get the eigenbasis and eigenvalue growth of $A$.","marker":"[Brown et al., 2013]"},{"why":"Supplies logarithmic potential theory facts: compact level sets, strict convexity, and uniqueness of the equilibrium measure.","marker":"[Saff and Totik, 1997]"},{"why":"Supplies the perturbation and self-adjoint compact operator arguments used to transfer spectral properties of $A$ to $L=A+2\\pi\\beta W$.","marker":"[Kato, 1995]"},{"why":"Supplies the min-max principle used to compare eigenvalues of $L$ and $A$ and get $\\kappa_j\\sim\\alpha j^2$.","marker":"[Reed and Simon, 1978]"}],"fun_headline_variants":["Hot circular beta-ensembles: Gaussian fluctuations with explicit covariance","High-temp circular beta-ensembles: CLT with interpolating covariance","Beta-ensemble fluctuations at high temperature are Gaussian","Gaussian fluctuations for hot circular beta-ensembles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the operator $L$, including the Hilbert-transform part $W$, is self-adjoint on the Sobolev space $H$ and has a complete eigenbasis with eigenvalues growing like the square of their index and with derivatives of eigenfunctions controlled by powers of those eigenvalues. The rate proof depends on this spectral picture, so if any of it fails, the stated CLT and convergence rate do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Hot circular beta-ensembles: Gaussian fluctuations with explicit covariance","High-temp circular beta-ensembles: CLT with interpolating covariance","Beta-ensemble fluctuations at high temperature are Gaussian","Gaussian fluctuations for hot circular beta-ensembles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000833,"raw_usage":{"total_tokens":3729,"prompt_tokens":1132,"completion_tokens":2597,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":748,"completion_tokens_details":{"reasoning_tokens":2528}},"tokens_in":748,"tokens_out":2597,"duration_ms":19746,"temperature":1.0,"reasoning_tokens":2528,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:29:43.897235+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For $V=0$, simulate the particle system at fixed $\\beta$ and measure the variance of $N^{-1/2}\\sum_i\\cos(x_i)$; Theorem 1.2 predicts it approaches $1/(2(1+\\beta))$, so a clear mismatch at large $N$ would refute the CLT. More directly, one can numerically diagonalize a discretization of $L$ for a non-constant smooth $V$ and check whether the eigenvalue ratios $\\kappa_j/j^2$ converge to a positive constant and whether the stated eigenfunction Lipschitz bounds hold.","supporting_citations":[],"review_version":1}