{"id":"754c98d0-0aa2-4a28-b987-19e45c14e675","arxiv_id":"2411.14777","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The N-particle 2D log gas on R^2 is shown to be entropically chaotic with rate e^{C t^eps}(E_N(0)+1/N), the first quantitative whole-space estimate.","lead":"Quantitative propagation of chaos is proved for the 2D Coulomb (log) gas on the whole plane, giving an explicit relative entropy bound in terms of N and time. The result adapts the modulated free energy method to the whole space and introduces new logarithmic growth estimates for the mean-field Poisson-Nernst-Planck equation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4.1 is delegated to the torus proof; whole-space boundary terms and integrability of the commutator are not established, so the Gronwall input for Theorem 1.1 is unverified.","rationale":"The reader's weakest assumption identifies the same point: Lemma 4.1 is the bridge between the modulated free-energy method developed on T^2 and the whole-space statement that is the paper's advertised novelty. Everything downstream, including the symmetrization (4.5), the large-deviation bound (4.8), and the Gronwall conclusion (4.11), assumes (4.2). The concern is not that (4.2) is known to be false; it is that the proof currently consists of a citation to a torus computation, and the whole-space modifications are exactly where the argument could fail. The Schwartz-data mismatch in Lemma 2.2 is a separate but real gap: the decay estimates (2.8) are used in Lemma 4.2 to bound terms such as ∇^2 g * ρbar by C(1+t)^{-1}, while the theorem's assumptions do not imply Schwartz regularity. I do not see a reason to reject the paper; the result is plausible and the missing details are likely fillable. The appropriate disposition remains conditional acceptance, with Lemma 4.1 and the regularity mismatch as the conditions to be verified.","tokens_in":30039,"tokens_out":27581,"duration_ms":275448,"concrete_test":"Give a complete proof of Lemma 4.1 on R^2: regularize ρ_N and ρbar_N, compute dE_N/dt on a truncated domain B_R^{2N} with smooth cutoffs, track every boundary term, and pass R → ∞ using the Gaussian upper bound on ρbar_N and the entropy dissipation inequality (1.10). If any boundary term survives or any integral in (4.2) is infinite, state its value and verify whether it can be absorbed in the Gronwall inequality; if no boundary terms survive, the transfer from [BJW19a] is justified and the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.1 states Lemma 4.1 (Eq. 4.2) as \"almost the same as the torus case\" and refers the proof to [BJW19a, Proposition 2.1]. This lemma is the only derivation of the time evolution of the modulated free energy and is the input to the Gibbs-inequality/Gronwall estimate (4.8)-(4.11). On T^2 the integrations by parts are boundary-free and w = ∇ log ρbar + β ∇g * ρbar is bounded. On R^2, w grows linearly in x, and deriving (4.2) requires integrating by parts over R^{2N} in each particle variable. The boundary terms at infinity vanish only if ρ_N has enough decay relative to ρbar_N; the entropy-solution hypothesis and (1.10) give uniform entropy dissipation but no explicit weighted decay of ρ_N. Likewise, the commutator integral in (4.2) contains (w(x)-w(y))·∇g(x-y) with w of linear growth and |∇g| ~ 1/|x-y|; its finiteness is not shown before taking expectations. Since the torus proof cannot be transferred without addressing these whole-space issues, the central Gronwall bound is presently unsupported. A secondary gap is that Lemma 2.2 (2.8) assumes Schwartz initial data while Theorem 1.1 assumes only W^{2,1} ∩ W^{2,∞}; Lemma 4.2 uses the resulting decay of ∇^2 g * ρbar, so the regularity assumptions do not match the proof as written.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims the first quantitative entropic propagation of chaos for the 2D log gas (weakly interacting particle system with 2D Coulomb interaction) on the whole space. The authors adapt the modulated free energy method of Bresch-Jabin-Wang to R^2 and prove logarithmic gradient and Hessian estimates for the mean-field PNP equation via a parabolic maximum principle. The main theorem states that, under initial bounds (1.11)-(1.12), the normalized relative entropy H_N(rho_N | rho_bar_N)(t) is bounded by C exp(C/eps t^eps)(E_N(rho_N|rho_bar_N)(0)+1/N) for any eps>0 and any beta>0. The proof combines a time evolution lemma for the modulated free energy, a large deviation estimate, and a Gronwall argument.","tokens_in":30338,"tokens_out":15274,"duration_ms":155609,"significance":"If the proof is completed, the result would be a substantial advance: it would close the gap between the torus result of BJW19a and the qualitative whole-space result of Liu-Yang, and it would provide one of the first quantitative propagation-of-chaos estimates for singular Coulomb interaction on R^2. The logarithmic growth estimates in Section 3, especially the Gaussian lower bound with the sharp exp(-|x|^2/t) scaling, are a meaningful technical contribution and improve on the earlier work by Feng-Wang for the vortex model. The paper also uses the large deviation theorem with test functions of quadratic growth, which is an interesting extension of the original JW18 framework. However, the whole-space adaptation of the modulated free energy evolution is not carried out in the manuscript, and there is a regularity mismatch between the assumptions of Theorem 1.1 and the lemmas used in the proof.","major_comments":[{"comment":"The time evolution of the modulated free energy is stated on the whole space, but its proof is delegated to [BJW19a, Proposition 2.1], which is a torus computation. This adaptation is not automatic. On T^2 the vector field w(x)=nabla log rho_bar(x)+beta nabla g*rho_bar(x) is bounded and every integration by parts is boundary-free; on R^2, w grows linearly in x and the commutator integrand (w(x)-w(y))·nabla g(x-y) is singular at x=y and only conditionally integrable as a function of the signed measure (d mu_N - d rho_bar)^(x)2. The entropy-solution hypothesis (1.10) gives a uniform entropy dissipation but no weighted spatial decay of rho_N, so the boundary terms at infinity in the integration by parts over R^{2N} are not shown to vanish, and no estimate is provided for the absolute value of the commutator integral before taking expectations. Since the Gronwall argument in (4.8)-(4.11) and the final bound in Theorem 1.1 rest entirely on (4.2), the central estimate is unsupported as written. A self-contained proof of Lemma 4.1 on R^2, or a localization/truncation argument with explicit control of the truncation error, is required.","section":"Section 4.1, Lemma 4.1 (Eq. (4.2))"},{"comment":"Lemma 2.2 assumes Schwartz initial data rho_bar_0 in S(R^2) and uses weighted L^2 moment estimates and Gagliardo-Nirenberg inequalities to obtain the decay rates (2.8). Theorem 1.1, however, assumes only rho_bar_0 in W^{2,1} cap W^{2,infty}(R^2), and the statement of Theorem 1.1 does not even include the hypothesis rho_bar in C([0,T], W^{2,1} cap W^{2,infty}) that Lemma 4.1 explicitly requires. The proofs of Theorem 3.1, Theorem 3.2, and Lemma 4.2 cite Lemma 2.2 for the decay of nabla rho_bar, nabla^2 rho_bar, and nabla^2 g*rho_bar. Consequently, the proof as written does not cover the stated regularity class. Either Theorem 1.1 must be strengthened to Schwartz (or sufficiently decaying C^2) initial data, or Lemma 2.2 and its consequences must be proved under the W^{2,1} cap W^{2,infty} assumptions with constants depending only on those norms.","section":"Section 2.2 / Theorem 1.1 / Lemma 4.2"}],"minor_comments":[{"comment":"The Gronwall argument yields an exponential factor of the form exp(C eps^{-2} ((1+t)^eps -1)), not exp(C eps^{-1} t^eps) as stated in Theorem 1.1; the statement should be adjusted, or the constant should be allowed to depend on eps in the stated way.","section":"Section 4.3, after Eq. (4.11)"},{"comment":"The repeated assertion that f = g*rho_bar is bounded is false on R^2, since f(x) behaves like -(1/(2 pi)) log|x| at infinity. The growth condition (3.6) can still be verified using the Gaussian decay of the auxiliary functions, but the justification given in the proofs of Theorem 3.1 and Theorem 3.2 should be corrected.","section":"Sections 3.1 and 3.4"},{"comment":"The bound for w_2(x) uses the factor sup_z |z| rho_bar(z) without an explicit proof; it follows from the Gaussian upper bound Lemma 2.3, but this should be stated for the reader.","section":"Lemma 4.2, Eq. (4.10)"},{"comment":"There are numerous typos and minor notation issues: 'logathimic' in the proof of Theorem 3.1, 'convinence' in Section 2.1, 'Risze' and 'Rosezweig' in the references, and the notation '1s=0' in Theorem 1.2 should be clarified as an indicator function.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an important problem and the overall strategy is plausible, but the two issues in the major comments are load-bearing and must be resolved before the result can be accepted. I see no evidence of circularity or of the authors assuming the target result. The manuscript is within the scope of the journal and, if the gaps are fixed, would be a significant contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper's claim is substantial: first quantitative entropic propagation of chaos for the 2D Coulomb/log gas on the whole space, with rate C e^{C/ε t^ε}(E_N(0)+1/N) for any ε>0 and any β>0. If it were fully proven, it would be a real advance over Liu-Yang's qualitative result and over the torus-only results of BJW19a/BJW23. The new logarithmic gradient and Hessian estimates (Theorems 3.1, 3.2) and the Gaussian lower bound (Lemma 3.1) are nontrivial and the maximum principle arguments look coherent; those sections are worth keeping.\n\nThe soft spot is exactly where the stress-test points. Lemma 4.1 (time evolution of the modulated free energy, Eq. (4.2)) is stated as \"almost the same as the torus case\" and the proof is referred to BJW19a, Proposition 2.1. On the torus, integration by parts has no boundary and w is bounded. Here w(x)=∇log ρbar + β∇g*ρbar grows like |x| by Theorem 3.1, the integral is over R^{2N}, and nothing in the entropy-solution definition (1.10) gives the weighted decay of ρN needed to kill boundary terms or even guarantee that the commutator integrand is finite. This is not cosmetic: (4.2) is the only input to the Gronwall step (4.8)-(4.11). The paper does not address these questions, so the proof of Theorem 1.1 is incomplete as written.\n\nThere is a second, smaller mismatch: Lemma 2.2, used for the higher-derivative decay in Lemma 4.2, assumes Schwartz initial data, while Theorem 1.1 assumes only W^{2,1}∩W^{2,∞} with log growth. This can probably be fixed by approximation, but the statements don't match.\n\nI don't think the gaps are insurmountable; they look addressable. But they are load-bearing until fixed, so the paper should be conditional. The authors need to either prove Lemma 4.1 on R^2, with explicit integrability and boundary checks, or state it as an open problem and soften the main theorem. The Section 3 log-growth estimates can stand alone and are likely correct.\n\nThis is a serious paper for the mean-field community and deserves refereeing, but I would not cite the main theorem in its current form.\n\nRecommendation: send to peer review, insist on a full proof of Lemma 4.1 and a cleanup of the regularity assumptions.","headline":"The first quantitative entropic propagation of chaos for the 2D log gas on R^2 is plausible and the log-growth estimates are genuinely new, but the whole-space proof of Lemma 4.1 is delegated, not done, so Theorem 1.1 currently rests on an unverified Gronwall input.","tokens_in":30897,"tokens_out":4549,"would_cite":false,"duration_ms":59477,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82C22","35K55","35B40"],"pacs":[],"model":"deepseek-v4-flash","headline":"For $N$ Brownian particles with 2D Coulomb repulsion on the whole plane, relative entropy to the mean-field law stays within $C e^{\\frac{C}{\\varepsilon}t^\\varepsilon}(E_0+1/N)$, for every $\\beta>0$.","keywords":["propagation of chaos","mean field limit","2D log gas","Coulomb gas","relative entropy","modulated free energy","Poisson-Nernst-Planck equation","maximum principle"],"falsifier":"A direct computation of the commutator integral in formula (4.2) for a two-particle Gaussian test law on $\\mathbb{R}^2$ at small $t$ should show all boundary terms at infinity vanish and the integral is finite; any divergence would falsify Lemma 4.1. Alternatively, a Monte Carlo simulation of the $N$-particle system initialized as a product Gaussian should show normalized relative entropy staying below the stated $O(1/N)$ rate up to the time prefactor, and a clearly faster growth would contradict Theorem 1.1.","tokens_in":29838,"feed_emoji":"","tokens_out":8153,"duration_ms":72099,"temperature":0.7,"pith_summary":"The paper proves that the $N$-particle stochastic system with 2D Coulomb repulsion on the whole plane converges, as $N\\to\\infty$, to the mean-field Poisson-Nernst-Planck density, in normalized relative entropy, with a quantitative rate. The rate combines a $1/N$ statistical error with a prefactor that grows at most like $e^{\\frac{C}{\\varepsilon}t^\\varepsilon}$ in time, for any $\\varepsilon>0$, and it holds for every inverse temperature $\\beta>0$. This is the first quantitative propagation of chaos for the 2D log gas on the whole space; earlier whole-space results were qualitative, and quantitative results on the torus did not cover the unbounded domain. The proof works by transplanting the modulated free energy method to $\\mathbb{R}^2$, using new logarithmic growth estimates for the mean-field density.","feed_headline":"First quantitative chaos bound for 2D log gas on the whole plane","feed_subtitle":"For every inverse temperature, the N-particle Coulomb system converges to the mean-field density at rate 1/N, up to a time prefactor.","key_machinery":"The central object is the modulated free energy $E_N(\\rho_N|\\bar\\rho_N)=H_N(\\rho_N|\\bar\\rho_N)+\\frac{1}{\\beta}F_N(\\mu_N,\\bar\\rho)$, which combines the normalized relative entropy with the modulated Coulomb energy between the empirical measure and the mean-field density. Its time derivative cancels the singular $\\delta_0$ term coming from $\\operatorname{div}K=-\\Delta g$, leaving a commutator term that is controlled by the new logarithmic growth estimates $|\\nabla\\log\\bar\\rho|^2\\lesssim \\frac{1}{1+t}(1+\\log(1+t)+\\frac{|x|^2}{1+t})$ and a corresponding Hessian bound, obtained through a parabolic maximum principle and a Gaussian lower bound for $\\bar\\rho$. These estimates replace the positive lower bound that is available on the torus but not on the whole space.","core_discovery":"The central claim is Theorem 1.1: if $\\rho_N$ is an entropy solution to the $N$-particle Liouville equation and $\\bar\\rho$ solves the mean-field PNP equation with initial data satisfying the logarithmic growth conditions and Gaussian upper bound, then for any $\\varepsilon>0$ and any $t\\ge 0$, $$H_N(\\rho_N|\\bar\\$rho^{{\\otimes N}}$)(t)\\le C\\,$e^{{\\frac{C}}${\\varepsilon}t^\\varepsilon}\\left(E_N(\\rho_N|\\bar\\$rho^{{\\otimes N}}$)(0)+\\frac{1}{N}\\right).$$ The constant $C$ depends only on $\\beta$, the initial mean-field density, and the Gaussian bound. This gives the first quantitative propagation of chaos for the 2D Coulomb gas on $\\mathbb{R}^2$, upgrading the earlier qualitative compactness argument to a rate and extending the torus result to the whole space for every $\\beta>0$.","pith_inferences":["Editorial extension: the $e^{\\frac{C}{\\varepsilon}t^\\varepsilon}$ prefactor is likely non-optimal. The paper itself notes that the bottleneck is the suboptimal $\\frac{\\log(1+t)}{1+t}$ decay in the maximum-principle estimates, so sharper maximum-principle bounds would probably yield a polynomial time growth instead.","Editorial extension: the same machinery may extend to the attractive 2D log gas or the Patlak-Keller-Segel system on the whole space, provided matching Gaussian lower and upper bounds hold; the paper states the logarithmic estimates are valid for such models on finite time intervals.","Editorial extension: a numerical test with moderate $N$ comparing the $N$-particle Coulomb system to the PNP solution could check whether the entropy difference actually follows the predicted $1/N$ rate up to a mild time factor; a rate worse than $1/N$ or a prefactor growing faster than $e^{\\frac{C}{\\varepsilon}t^\\varepsilon}$ would indicate a missing whole-space boundary effect."],"forward_implications":["For fixed time $t$, the normalized relative entropy is $O(1/N)$ up to a constant depending on $t$, giving the optimal statistical rate of mean-field convergence.","Quantitative propagation of chaos holds on $\\mathbb{R}^2$ for every inverse temperature $\\beta>0$, removing the earlier restriction to qualitative, rate-free convergence.","The same maximum-principle estimates yield a quantitative propagation-of-chaos result for the 2D viscous vortex model on the whole space, refining the previous time-growth estimate.","The equilibrium Gibbs measure of the $N$-particle system concentrates around the mean-field equilibrium in relative entropy with rate $(\\log N)/N$ in two dimensions, as stated in Theorem 1.2.","The sufficient hypotheses on the initial data are the Gaussian upper bound and logarithmic growth of $\\nabla\\log\\bar\\rho_0$ and $\\nabla^2\\log\\bar\\rho_0$; no positive lower bound on $\\bar\\rho$ is required."],"supporting_citations":[{"why":"It supplies the modulated free energy method and the time-evolution formula for the torus case that Lemma 4.1 adapts to the whole space.","marker":"[BJW19a]"},{"why":"It extends the method to singular attractive kernels and is referenced as the origin of the modulated free energy framework used here.","marker":"[BJW23]"},{"why":"It introduces the whole-space strategy with logarithmic growth estimates for the 2D viscous vortex model, whose maximum-principle argument is adapted here.","marker":"[FW23]"},{"why":"It provides the large deviation theorem used to control the exponential integral of the test function in the commutator term.","marker":"[JW18]"},{"why":"It establishes existence of entropy solutions and the qualitative propagation of chaos on the whole space that this paper upgrades to quantitative.","marker":"[LY16]"},{"why":"It gives the asymptotic lower bound for modulated energy used in Theorem 1.2's concentration estimate.","marker":"[Ser20]"},{"why":"It supplies the general maximum principle on weighted manifolds used for the logarithmic gradient and Hessian estimates.","marker":"[Gri09]"},{"why":"It provides the generalized Gronwall inequality used to close the relative entropy bound.","marker":"[YGD07]"}],"fun_headline_variants":["This is the first 1/N entropy-chaos rate for 2D log gas","The whole-space 2D Coulomb gas now has a chaos rate","Quantitative chaos for 2D log gas on R^2 is proven","Entropy propagation of chaos for 2D log gas is now quantitative","First entropy-chaos bound for 2D log gas has 1/N rate"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the whole-space version of the modulated-free-energy time evolution (Lemma 4.1) is valid with all boundary terms at infinity vanishing, even though its proof is delegated to the torus case; if that adaptation fails, the Gronwall bound collapses.","fun_headline_variants_meta":{"raw":{"variants":["This is the first 1/N entropy-chaos rate for 2D log gas","The whole-space 2D Coulomb gas now has a chaos rate","Quantitative chaos for 2D log gas on R^2 is proven","Entropy propagation of chaos for 2D log gas is now quantitative","First entropy-chaos bound for 2D log gas has 1/N rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001428,"raw_usage":{"total_tokens":5694,"prompt_tokens":809,"completion_tokens":4885,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":4782}},"tokens_in":425,"tokens_out":4885,"duration_ms":32504,"temperature":1.0,"reasoning_tokens":4782,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:56:02.272840+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct computation of the commutator integral in formula (4.2) for a two-particle Gaussian test law on $\\mathbb{R}^2$ at small $t$ should show all boundary terms at infinity vanish and the integral is finite; any divergence would falsify Lemma 4.1. Alternatively, a Monte Carlo simulation of the $N$-particle system initialized as a product Gaussian should show normalized relative entropy staying below the stated $O(1/N)$ rate up to the time prefactor, and a clearly faster growth would contradict Theorem 1.1.","supporting_citations":[],"review_version":1}