{"id":"5559e4e5-ab40-48df-bb37-2e0db76d3c5f","arxiv_id":"2412.17389","paper_version":4,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A new optimal transport argument shows that Dyson Brownian motions with beta at least 2, and their scaling limits, have Brownian-like uniform modulus of continuity bounds with constants independent of the particle layer.","lead":"This paper proves that many random curve models, including Dyson Brownian motion with beta at least 2, have the same modulus of continuity as Brownian motion, by viewing them as log-concave perturbations of Brownian paths. The proof introduces an optimal transport comparison and yields uniform estimates that also cover the Airy and KPZ line ensembles.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the critical log-concavity step (Lemma 2.11) checks out; the proof route is sound.","rationale":"The reader identified Lemma 2.11 as the weakest assumption, and I agree it is the critical input. However, I found the lemma correct: the density of the discretized Brownian bridge is log-concave in (x,y,w) because the mean is affine in the endpoints, and Prékopa–Leindler applies. The subsequent use of Hargé's inequality is legitimate, and the passage from finite-dimensional marginals to the full modulus estimates is standard. The only expositional gap is closure of LC under the affine edge-scaling transformation, but this follows by direct calculation and does not affect correctness. I therefore see no load-bearing concern that would warrant rejecting or conditioning the paper.","tokens_in":18018,"tokens_out":44522,"duration_ms":408871,"concrete_test":"Independently verify Lemma 2.11 for a 1D bridge with H(z)=∫_0^1 (z(s))^4 ds by computing log Z(x,y) via the Laplace method or exact expansion and confirming concavity of log Z; also numerically test the Hessian on a grid. Additionally, write out the affine transformation of a measure in ˚M under t ↦ c t + d, z ↦ a z + ψ(t) with a^2 c = 1, checking the transformed Hamiltonian remains in H_R.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing step is indeed Lemma 2.11: log-concavity of the Brownian-bridge partition function in its endpoints. I checked the argument: after discretization, the integrand e^{-H_n(w)} ρ(w|x,y) is log-concave in (x,y,w) because ρ(w|x,y) is the density of a Gaussian whose mean is affine in (x,y), so Prékopa–Leindler gives the result. The reduction to non-negative H via Lemma 2.2 is valid since the affine factor preserves log-concavity. I find no flaw. The only notable gap in the exposition is that the affine edge-scaling used to pass from Dyson Brownian motion to the Airy_β ensemble is asserted to preserve LC without proof; a direct calculation shows the diffusion constant after scaling is normalized by √(β/2), so the pushforward of a shifted Wiener measure is again in the class, and the transformed Hamiltonian remains of the form (2.1). This is fillable and not a correctness risk. Theorem 2.8 then follows as stated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a new method, based on Caffarelli's contraction theorem and Hargé's convex/log-concave correlation inequality, to prove sharp and uniform modulus-of-continuity estimates for a class LC of random curve collections that are log-concave perturbations of Brownian motions or Brownian bridges, together with their distributional limits. The main result, Theorem 2.8, shows that for every member of LC, the centered increments satisfy Brownian-motion-type moment bounds, Gaussian tail bounds for a suitably normalized sup-norm, and Garsia-Rodemich-Rumsey type Hölder estimates, with constants uniform in the layer. The paper then proves that β-Dyson Brownian motions for β ≥ 2 belong to LC, and, by taking distributional limits, that the Airy_β line ensemble, the O'Connell-Yor line ensemble, and the KPZ line ensemble also belong to LC. Applications include new modulus-of-continuity estimates for the KPZ line ensemble and a short proof of tightness for the scaled Dyson Brownian motions.","tokens_in":18212,"tokens_out":15988,"duration_ms":141434,"significance":"If the results are fully established, this is a substantial contribution to the study of line ensembles and random matrix diffusions. The paper provides a unified framework that yields sharp, layer-uniform estimates matching those of Brownian motion for several important models, extending earlier work that was either restricted to β = 2 or had layer-dependent constants. The method is novel and conceptually clean, reducing the infinite-dimensional comparison to a finite-dimensional log-concavity statement (Lemma 2.11) that is checked via Prékopa–Leindler. A particular strength is that the proofs use no fitted parameters: all constants are explicit or universal, and the main estimates are derived from external theorems (Caffarelli, Hargé) rather than from self-cited results. The applications to the KPZ line ensemble and the O'Connell-Yor line ensemble are new and directly address an open gap in the construction of the fundamental solution to the KPZ equation.","major_comments":[{"comment":"The claim that √(β/2) A^β(t) belongs to LC is not justified. Proposition 3.1 establishes that the β-Dyson Brownian motion X^{β,N} itself is in LC, but the edge scaling (1.2) followed by normalization √(β/2) is an affine change of both space and time. The class LC is defined with respect to unit-diffusion Wiener measures and Hamiltonians of the form (2.1); it is not immediate that the pushforward of a member of LC under this transformation remains in LC. Since Corollaries 1.2 and 1.3 depend on applying Theorem 2.8 to the scaled process, the proof should either verify that the scaled base measure is a unit-diffusion Wiener measure and the transformed Hamiltonian remains of the form (2.1), or derive the estimate by direct scaling from Theorem 2.8 for X^{β,N}.","section":"§3.1, proof of Corollary 1.3"},{"comment":"Several statements that are used in the main proofs are omitted: the bridge counterpart of Lemma 2.2 (after (2.3)), the bridge counterpart of Lemma 2.4, and the entire Lemma 2.12. The bridge counterpart of Lemma 2.2 is used in the proof of Lemma 2.11 to reduce to non-negative Hamiltonians; Lemma 2.12 is used in the proof of Lemma 3.5 to show that jλ is log-concave; and the bridge counterpart of Lemma 2.4 is needed for Proposition 2.9(2.5) and hence for the bridge part of Theorem 2.8. These results are plausible variants of the proved arguments, but since they are load-bearing, the authors should supply the proofs or at least a detailed indication of the modifications required.","section":"§2, Lemmas 2.2, 2.4, 2.12"}],"minor_comments":[{"comment":"The displayed definition of H(z) appears to miss a minus logarithm; as written, H(z) = jλ(z(t0)) exp(−∫ F), which is not a Hamiltonian of the form (2.1). The intended H is presumably −log jλ(z(t0)) + ∫ F(z(s)) ds, consistent with the subsequent discussion of log-concavity of jλ.","section":"§3.2, Lemma 3.5"},{"comment":"The symbol a is used both as a real parameter in [0,1/2) and as the left endpoint of the interval [a,b]; consider renaming the parameter (e.g., γ) to avoid confusion.","section":"§2, proof of Theorem 2.8(ii)"},{"comment":"The sentence 'It is straightforward to check that for M (t) is an non-negative P0-martingale' contains a grammatical error and a stray period; more importantly, the martingale property is a nontrivial step in the Girsanov argument and should be justified by a reference or a short argument.","section":"§3.2, Lemma 3.5"},{"comment":"The convergence e^{−Hε(z)} ↓ hβ(x∗) dµ/dγ(z) for γ-a.e. z is stated as straightforward, but a short explanation of why the approximating convex functions yield monotone convergence and why the limit vanishes on paths that leave the Weyl chamber would improve readability.","section":"§3.1, proof of Proposition 3.1"}],"recommendation":"major_revision","confidential_remarks":"The central method is sound and the key log-concavity step (Lemma 2.11) checks out. The main gaps are missing justifications for the preservation of LC under edge scaling and several omitted proofs of similar lemmas; these are repairable within the scope of the manuscript. The paper is a strong fit for the journal, and I expect that after the authors supply the missing arguments, acceptance would be appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Key take: the central proof route is sound, and the paper deserves a serious referee. The log-concavity lemma (Lemma 2.11) that everything rests on checks out: discretize the Hamiltonian, use Prékopa–Leindler on the joint density of endpoints and interior bridge values, then integrate. I read it carefully; it works. From there the paper gets Brownian-optimal modulus of continuity estimates uniformly in the layer for every element of the LC class, which is a real improvement over DV21's O(e^j) constants. The consequences for Airy_β, KPZ, and O'Connell–Yor are credible and cleanly stated. No fitted parameters, no self-cited load-bearing input; the convergence of the models is cited from the literature.\n\nWhat is actually new: the Caffarelli–Hargé comparison as a tool for line ensembles is a fresh perspective, and the log-concavity of finite-dimensional marginals for KPZ and Airy_β (β≥2) is a nice corollary. The paper is mostly very readable, and the main theorem is stated in a clean, general form.\n\nSoft spots: several minor gaps and typos. Lemma 2.12 and the bridge half of Lemma 2.2 are omitted as “similar” — fillable but should be written down. Lemma 3.5's displayed H(z) is wrong: it should be −log j_λ(z(t0)) + ∫F, not the Radon–Nikodym derivative itself. The claim that the edge scaling preserves the LC class is asserted without proof; a direct calculation fills it, but it should be included. The tightness proof in Corollary 1.2 is also a bit terse. None of these threaten the main result.\n\nVerdict: this is a solid paper with a load-bearing method that holds up. I would accept it after minor revisions. The estimates are likely to be used, and the method may propagate to other log-concave perturbations.","headline":"A genuinely new and correct method: Caffarelli–Hargé comparison for log-concave perturbations yields Brownian-sharp, layer-uniform modulus estimates for DBM, Airy_β, KPZ, and OY; minor gaps only.","tokens_in":620,"tokens_out":775,"would_cite":true,"duration_ms":79185,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","60H10","60G17","49Q22"],"pacs":[],"model":"deepseek-v4-flash","headline":"A new optimal-transport comparison proves that β-Dyson Brownian motions with β≥2 and their scaling limits have exactly the sharp modulus of continuity of Brownian motion, uniformly across all layers, for a broad class of log-concave…","keywords":["Dyson Brownian motion","log-concave perturbations","optimal transport","Caffarelli contraction","modulus of continuity","Airy line ensemble","KPZ line ensemble","O'Connell-Yor line ensemble"],"falsifier":"Compute, for a concrete convex Hamiltonian $H$, the partition function $Z_H(x,y)$ of a Brownian bridge on a fixed interval and check numerically whether $\\log Z_H$ is concave in $(x,y)$; a single counterexample would falsify Lemma 2.11 and the whole theorem chain. Alternatively, simulate a $\\beta=2$ Dyson Brownian motion with many particles and measure the variance of a centered single-particle increment over a small time step; if it exceeds $|t-s|$ by a non-negligible margin, Theorem 1.1(i) would be false.","tokens_in":17808,"feed_emoji":"📈","tokens_out":11499,"duration_ms":87599,"temperature":0.7,"pith_summary":"The paper claims that β-Dyson Brownian motions with β≥2, along with a broad class of random curve collections that are log-concave perturbations of Brownian motion, have exactly the same sharp modulus of continuity as a standard Brownian motion: their centered increments satisfy $\\mathbb{E}|\\hat L_j(t)-\\hat L_j(s)|^p \\le N_p |t-s|^{p/2}$ for every $p\\ge1$, and the supremum over an interval has a Gaussian tail with constants independent of the layer. The proof is a direct comparison with Brownian motion, based on viewing the perturbed process as a reweighting of Wiener measure by $e^{-H}$ with a convex Hamiltonian $H$, and then applying Caffarelli's contraction theorem from optimal transport. The class $LC$ includes the β-Dyson Brownian motion for β≥2, the Airy$_β$ line ensemble, the KPZ line ensemble, and the O'Connell-Yor line ensemble, so the same estimates hold for all of them uniformly across every layer. If the argument is right, these are the sharp, layer-uniform continuity estimates that earlier work lacked, and they fill a gap in the program of constructing the directed landscape and the KPZ equation from line ensembles.","feed_headline":"Dyson paths match Brownian continuity via optimal transport","feed_subtitle":"New uniform bounds cover Airy, KPZ, and O'Connell-Yor line ensembles layer by layer.","key_machinery":"The machinery is the class $LC$ of log-concave perturbations defined by convex Hamiltonians $H(z)=\\sum_i f_i(z(t_i))+\\int_a^b F(t,z(t))\\,dt$ on Wiener space, and the key identity is Lemma 2.11: for any convex $H$, the partition function $Z_H(x,y)=\\mathbb{E}_{x,y}[e^{-H(z)}]$ of a tilted Brownian bridge is log-concave in the endpoint data $(x,y)$. This log-concavity, transferred to finite-dimensional marginal densities via the Prékopa–Leindler theorem, makes the Radon–Nikodym derivative $d\\hat\\mu/d\\hat\\gamma$ log-concave (Lemma 2.4). Then Caffarelli's contraction theorem (the optimal-transport map pushing a Gaussian onto a log-concave perturbation is 1-Lipschitz), together with Hargé's convex/log-concave correlation inequality, gives the comparison $\\mathbb{E}_\\mu[g(w-\\bar w_\\mu)] \\le \\mathbb{E}_\\gamma[g(w-\\bar w_\\gamma)]$ for every convex $g$ depending on finitely many times, which is exactly what yields the Brownian-style moment and tail bounds.","core_discovery":"The central discovery is Theorem 2.8: for every random continuous function $L$ in the class $LC$, the centered increment $\\hat L_j(t)-\\hat L_j(s)$ has $p$-th moment at most $N_p |t-s|^{p/2}$ (the Brownian value), and the normalized supremum over $[a,b]$ satisfies $\\mathbb{P}\\left(\\sup_{t,s\\in[a,b]}\\frac{|\\hat L_j(t)-\\hat L_j(s)|}{\\sqrt{|t-s|\\log(2(b-a)/|t-s|)}}>K\\right) \\le C_1 e^{-C_2 K^2}$ with universal constants $C_1,C_2$ independent of the layer $j$. In particular, β-Dyson Brownian motions with $\\beta\\ge2$ satisfy the same sharp bounds as Brownian motion, uniformly in $j,t,s$, and the same holds after edge scaling for the Airy$_β$ line ensemble, for the KPZ line ensemble, and for the O'Connell-Yor line ensemble. The comparison is made by proving that finite-dimensional marginals of any $LC$ element are log-concave perturbations of Gaussian marginals, which brings the contraction theorem into play.","pith_inferences":["The same optimal-transport comparison should apply to other log-concave random-matrix-type diffusions such as $\\beta$-Laguerre and $\\beta$-Jacobi processes, once their Girsanov Hamiltonians are shown convex; the paper notes these examples but leaves them open.","The layer-uniform Brownian-quality bounds for $\\beta\\ge2$ suggest that Airy$_β$ line ensembles are as regular as Brownian motion at small scales; this regularity may be the missing ingredient for constructing directed-landscape analogues for general $\\beta$ from these ensembles.","One concrete test would be numerical: for a non-quadratic convex Hamiltonian, compute $\\log Z_H(x,y)$ and check its Hessian in $(x,y)$; a negative eigenvalue would locate the failure of Lemma 2.11 rather than merely weakening the estimates.","Log-concavity of finite-dimensional marginals implies positive-correlation (FKG-type) properties for these line ensembles, which may lead to new monotonicity results not stated in the paper."],"forward_implications":["For $\\beta\\ge2$, the $\\beta$-Dyson Brownian motion and its edge-scaling limit Airy$_β$ satisfy uniform Gaussian-tail bounds on the modulus of continuity that are independent of the layer, with constants that do not grow in $j$.","The KPZ line ensemble, with $\\hat X^T_j(t)=X^T_j(t)+2^{-1}t^2$, satisfies $\\mathbb{E}\\|\\hat X^T_j\\|^p_{\\alpha,[a,b]} \\le C(\\alpha,p)(b-a)^{p/2-\\alpha p}$, confirming Conjecture 1.12 in [Wu23b].","The finite-dimensional marginal distributions of the $\\beta$-Dyson Brownian motion, the Airy$_β$ line ensemble, both with $\\beta\\ge2$, the O'Connell-Yor line ensemble, and the KPZ line ensemble are log-concave.","Because $LC$ is closed under distributional convergence, the estimates pass automatically to any distributional limit of log-concave perturbations, yielding an alternative short proof of tightness for the Airy$_β$ scaling limits."],"supporting_citations":[{"why":"Provides Caffarelli's contraction theorem, the optimal-transport input that compares the log-concave perturbation with the Gaussian.","marker":"[Caf00]"},{"why":"Supplies the convex/log-concave correlation inequality for Gaussian measures that converts contraction into moment bounds on finite-dimensional marginals.","marker":"[Har04]"},{"why":"Provides the Prékopa–Leindler theorem used to prove log-concavity of the partition function and of finite-dimensional marginal densities.","marker":"[Sim11]"},{"why":"Supplies the Cameron–Martin theory used in Lemma 2.2 to normalise the log-concave reweighting.","marker":"[Jan97]"},{"why":"Gives existence, uniqueness, and Weyl-chamber facts for β-Dyson Brownian motion used in Proposition 3.1.","marker":"[AGZ10]"},{"why":"Introduces the O'Connell-Yor line ensemble and its Whittaker-function SDE, used in Proposition 3.4.","marker":"[O'C12]"},{"why":"Establishes the KPZ line ensemble as a distributional limit of O'Connell-Yor line ensembles, used in Corollary 1.4.","marker":"[CH16]"},{"why":"Provides the earlier modulus-of-continuity estimate for the Airy line ensemble that Theorem 2.8 improves, and supplies Lemma 3.3 used in the proof of Theorem 2.8(ii).","marker":"[DV21]"},{"why":"Supplies the convergence framework for Airy$_β$ line ensembles as limits of β-Dyson Brownian motions, used to transfer LC membership to Corollary 1.3.","marker":"[HZ24]"}],"fun_headline_variants":["Optimal transport yields Brownian modulus for Dyson and line ensembles","Caffarelli contraction gives Brownian continuity to Dyson and line ensembles","Dyson and line ensembles match Brownian modulus of continuity","Sharp uniform bounds via optimal transport for Dyson and line ensembles","Brownian-like continuity for Dyson and line ensembles via optimal transport"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the partition function of a Brownian bridge tilted by any convex Hamiltonian is log-concave in the bridge endpoints; if that failed for some convex Hamiltonian, the finite-dimensional densities would not be log-concave and the contraction comparison would break.","fun_headline_variants_meta":{"raw":{"variants":["Optimal transport yields Brownian modulus for Dyson and line ensembles","Caffarelli contraction gives Brownian continuity to Dyson and line ensembles","Dyson and line ensembles match Brownian modulus of continuity","Sharp uniform bounds via optimal transport for Dyson and line ensembles","Brownian-like continuity for Dyson and line ensembles via optimal transport"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000891,"raw_usage":{"total_tokens":3808,"prompt_tokens":877,"completion_tokens":2931,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":2842}},"tokens_in":493,"tokens_out":2931,"duration_ms":18987,"temperature":1.0,"reasoning_tokens":2842,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T05:29:40.287545+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a concrete convex Hamiltonian $H$, the partition function $Z_H(x,y)$ of a Brownian bridge on a fixed interval and check numerically whether $\\log Z_H$ is concave in $(x,y)$; a single counterexample would falsify Lemma 2.11 and the whole theorem chain. Alternatively, simulate a $\\beta=2$ Dyson Brownian motion with many particles and measure the variance of a centered single-particle increment over a small time step; if it exceeds $|t-s|$ by a non-negligible margin, Theorem 1.1(i) would be false.","supporting_citations":[],"review_version":1}