{"id":"a02a96e8-48b6-47f4-91c1-fc1329171627","arxiv_id":"2505.13038","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The N-particle regularized Coulomb system converges in relative entropy and in L1 marginals to Vlasov-Poisson(-Fokker-Planck) with explicit N and noise-strength rates.","lead":"This paper proves that a large system of particles interacting through a smeared-out Coulomb force is well approximated by the Vlasov-Poisson-Fokker-Planck (or Vlasov-Poisson) equations, with control in a strong sense (L1) rather than only weak convergence. The authors combine two existing techniques, trajectory comparison and relative entropy, to get explicit error bounds that vanish as the number of particles grows.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.3's stated bound does not match its proof: the proof yields an extra N^{-δ} exp(C√log N) term and needs σ_N→0, so the VP convergence statement is misstated.","rationale":"The reader identified imported trajectory estimates and the missing σ_N→0 hypothesis as the weakest assumptions. I agree those imports need checking, but the more immediate, checkable problem is internal: Theorem 1.3's statement does not follow from the proof's own final inequality. The proof's integrated RHS contains a N^{-δ} exp(C√log N) term absent from the theorem and a linear σ_N term, while the theorem statement uses √σ(√σ+1) exp(C√log N) and omits the N^{-δ} exponential term. This is not a failure of the overall strategy: for σ_N=N^{-δ} or any σ_N→0 slowly enough, every term in the integrated proof bound still tends to zero, so the qualitative L1 propagation of chaos toward Vlasov-Poisson is plausible with a corrected statement. The main theorems for VPFP, Theorems 1.1 and 1.2, appear internally consistent, aside from a likely kernel typo in Proposition 2.2 where (1.7) is stated although Theorem 1.1 uses (1.4) and cites [10] for that kernel. Because the central quantitative claim needs a corrected statement but the mathematical core is salvageable, the appropriate verdict remains conditional rather than rejection.","tokens_in":26926,"tokens_out":17402,"duration_ms":165531,"concrete_test":"Recompute the final estimate in the proof of Theorem 1.3 by integrating the last displayed inequality of §3.2 from t=0 to T and comparing termwise with the RHS of Theorem 1.3. A decisive asymptotic check: set σ_N=N^{-100}, δ=1/4, T=1; the integrated proof bound contains C N^{-δ} exp(C√log N), while the printed RHS contains only C√σ_N exp(C√log N), and N^{-δ} is much larger than √σ_N. A second check: take σ_N=σ>0 fixed; the printed RHS tends to infinity in N, so the theorem as written cannot describe convergence to the VP solution. Either comparison settles whether Theorem 1.3 needs correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In §3.2 the proof of Theorem 1.3 ends with the differential inequality d/dt H_N(f_N^t | \\tilde f_t^{⊗N}) + ... ≤ Cσ_N + C(N^{-δ}+√σ_N) exp(C√log N) + C N^{-δ}\\log N. Since H_N(0)=0, integrating over [0,T] gives sup_{0≤t≤T} H_N ≤ C T σ_N + C T (N^{-δ}+√σ_N) exp(C√log N)+ C T N^{-δ}\\log N. The printed Theorem 1.3 instead asserts sup_{0≤t≤T} H_N ≤ C√σ(√σ+1) exp(C√log N)+ C N^{-δ}\\log N, with no N^{-δ} exp(C√log N) term and no explicit σ_N→0 hypothesis. These two bounds are not equivalent. If σ_N is fixed positive, the printed right-hand side grows like exp(C√log N) and cannot vanish; if σ_N=N^{-100} while δ=1/4, the proof's N^{-δ} exp(C√log N) term is much larger than the printed √σ_N exp(...) term, so the theorem claims a rate not established. The proof also has a σ_N term, and the 'in addition to Theorem 1.1' hypothesis inherits σ_N≡σ>0, which is incompatible with passage to σ=0. The quantitative VP result is therefore misstated; the corrected statement should include σ_N→0 and the full integrated inequality from §3.2.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the mean-field limit of an N-particle Newtonian system with a regularized Coulomb interaction (a cut-off kernel or a smoothed kernel) and with white noise of strength σ_N. The effective equations are the Vlasov-Poisson-Fokker-Planck (VPFP) equation when σ_N converges to σ>0 and the Vlasov-Poisson (VP) equation when σ_N tends to zero. The authors prove quantitative propagation of chaos in relative entropy: for fixed positive noise, the scaled relative entropy between the N-particle law and the N-fold tensor product of the VPFP solution is bounded by C exp(C√log N)/(σ N^{2δ}) under a polynomial cut-off (Theorem 1.1), and an improved rate is given for a smoothed kernel in three dimensions (Theorem 1.2). For the vanishing-noise regime, Theorem 1.3 claims a relative entropy bound against the VP solution. The method combines the Jabin-Wang relative entropy approach with trajectory closeness estimates imported from [10, 27, 35], and then converts relative entropy into L1 convergence of marginals via Csiszár-Kullback-Pinsker and subadditivity.","tokens_in":27197,"tokens_out":16086,"duration_ms":158406,"significance":"If the claims are correct, the paper yields quantitative strong (L1) propagation of chaos for singular kinetic equations with Coulomb-type interactions, going beyond the Wasserstein or convergence-in-probability statements previously available for these systems. The entropy computations are explicit and the rate balancing in Theorems 1.1 and 1.2 checks out under the stated inequalities. The paper also credits and uses published trajectory-control results rather than reproving them, which is appropriate. The principal weakness is that Theorem 1.3 as printed does not match the estimate actually proved in Section 3.2, so the VP part of the paper needs a substantive correction before the central claim is reliable.","major_comments":[{"comment":"The bound stated in Theorem 1.3 is not the bound obtained by the proof. The proof ends with the differential inequality d/dt H_N(f_N^t|\\tilde f_t^{⊗N}) + dissipation ≤ Cσ_N + C(N^{-δ}+√σ_N) exp(C√log N) + C N^{-δ} log N. Integrating over [0,T] and using H_N(0)=0 gives sup_{0≤t≤T} H_N ≤ C T σ_N + C T (N^{-δ}+√σ_N) exp(C√log N) + C T N^{-δ} log N. The theorem instead asserts C√σ(√σ+1) exp(C√log N) + C N^{-δ} log N and contains no hypothesis that σ_N tends to zero. Since Theorem 1.1, whose assumptions Theorem 1.3 inherits, fixes σ_N ≡ σ > 0, the passage to σ=0 is not justified as stated. For example, if σ_N = N^{-100} and δ = 1/4, the proof produces an N^{-δ} exp(C√log N) term that is much larger than the printed √σ_N exp(C√log N) term, so the claimed rate is unsupported. The theorem should be restated with the explicit hypothesis σ_N→0 and with the full integrated right-hand side (or an equivalent bound from which convergence follows).","section":"§3.2, Theorem 1.3"},{"comment":"The displayed chain in the alternative L1 proof is incorrect. It bounds ||f_t^{⊗N} - \\tilde f_t^{⊗N}||_{L1}^2 by C H_1(f_t|\\tilde f_t). By the same convexity/Csiszár-Kullback-Pinsker argument used in Remark 1.1, H(f_t^{⊗N}|\\tilde f_t^{⊗N}) = N H_1(f_t|\\tilde f_t), so the factor is N, not 1. In addition, the expression \"σ_N := o(N^{-2δ} exp(C√log N))\" is not a well-formed definition of a sequence. Since this subsection is presented as an alternative proof of L1 convergence to VP, it should be corrected or removed; the main proof of Theorem 1.3 does not depend on this remark.","section":"§3.3"}],"minor_comments":[{"comment":"The statement writes H_N(f_N^t | f_t^{⊗k}); since H_N is defined between N-particle densities, the reference measure should be f_t^{⊗N}. In the proof, the notation also switches to \\bar f_t^{⊗N} in the final estimate; please make the notation consistent.","section":"Theorem 1.2 statement"},{"comment":"Proposition 2.2 is stated for the regularized kernel (1.7), but it is invoked in the proof of Theorem 1.1, which uses the kernel (1.4). Please clarify which kernel the proposition concerns (presumably (1.4)) and adjust the cross-references.","section":"Proposition 2.2"},{"comment":"The displayed consequence for the first marginal appears to invert the N- and σ-dependence: Remark 1.1 gives ||f^{N,1}_t - f_t||_{L1} = O(exp(C√log N)/(√σ N^δ)), not C√σ N^{δ-} as printed.","section":"Equations (1.5)–(1.6)"},{"comment":"The assumption \"for m>3\" is unused in the statement; either remove it or state explicitly the intended polynomial moment/regularity condition on f0.","section":"Theorem 1.3 hypothesis"},{"comment":"There are several typographical inconsistencies in the notation for the reference solution (f_t, \\bar f_t, \\tilde f_t) in the entropy estimates; a careful proofreading pass would improve readability.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The VPFP results in Theorems 1.1 and 1.2 appear sound, and the combination of the relative entropy method with the imported trajectory estimates is a useful contribution. The advertised VP result, Theorem 1.3, is misstated relative to its proof and needs a genuine correction, including the condition σ_N→0. I would also ask the authors to verify that the cited trajectory estimates apply to the exact kernel used in each theorem, since Proposition 2.2 currently points to the wrong regularization."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look, especially for the VPFP part. The paper does what the abstract promises: it combines the Jabin–Wang relative entropy machinery with the trajectory closeness estimates of [10], [27], and [35] to get L1 propagation of chaos for regularized Coulomb systems. Theorems 1.1 and 1.2 are the real contribution. The entropy evolution computation is careful, the rate balancing checks out, and the authors are upfront that the trajectory controls are imported. That part is honest and the combination is new.\n\nThe soft spot is Theorem 1.3. The stress-test note is right. The proof in §3.2 ends with\n\nd/dt H_N ≤ Cσ_N + C(N^{-δ}+√σ_N) exp(C√log N) + C N^{-δ}\\log N,\n\nso after integration the bound contains an N^{-δ} exp(C√log N) term and a plain σ_N term. The printed theorem instead says\n\nsup H_N ≤ C√σ(√σ+1) exp(C√log N) + C N^{-δ}\\log N,\n\nno N^{-δ} exp term and no hypothesis that σ_N → 0. Worse, Theorem 1.3 inherits σ_N ≡ σ > 0 from Theorem 1.1. With σ fixed positive, the displayed bound doesn't go to zero, so the theorem as stated does not give convergence to Vlasov–Poisson. That's not a subtle gap; it's a statement/proof mismatch.\n\nThe fix is immediate: allow σ_N → 0 and print the integrated inequality from §3.2. With δ>0, N^{-δ} e^{C√log N} → 0, so the corrected bound does vanish. There's also a small sign/order issue in §3.3: the text says 'choosing σ_N := o(N^{-2δ} exp(C√log N))' but the preceding explicit bound needs σ_N large enough to kill the 1/(σ_N N^{2δ}) term, i.e. σ_N ≫ N^{-2δ} e^{C√log N}. Probably just a typo, but it should be fixed because someone will run into it.\n\nI did not check every inequality in the entropy sections line by line, but the structure is coherent and I see no sign of a load-bearing error in the VPFP part. The reliance on [10], [27], [35] is a debt, not a flaw, since those are published results with stated assumptions; the paper acknowledges it.\n\nIn short: the VPFP result deserves a serious referee. The VP result as written does not. I'd send it to review with clear instruction that Theorem 1.3 must be restated with σ_N→0 and the correct integrated bound, and the §3.3 asymptotics cleaned up. After that it should be a solid paper.","headline":"Solid VPFP combination, but Theorem 1.3 is misstated as printed and the VP claim needs a proper σ_N→0 hypothesis.","tokens_in":27793,"tokens_out":4165,"would_cite":true,"duration_ms":35927,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q83","82C22","82C40","60H30"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves quantitative propagation of chaos in L1 for Vlasov-Poisson-Fokker-Planck and Vlasov-Poisson limits, with explicit relative-entropy rates.","keywords":["propagation of chaos","mean-field limit","Vlasov-Poisson-Fokker-Planck equation","Vlasov-Poisson equation","relative entropy","Coulomb interaction","L1 convergence","McKean-Vlasov process"],"falsifier":"Run the N-particle system and the McKean-Vlasov flow for a smooth, compactly supported $f_0$ in $d=3$ with $\\sigma=1$ and the truncated kernel (1.4), and measure the probability of the event $\\sup_{t\\in[0,T]} \\|\\Phi_N^t - \\Psi_N^t\\|_\\infty > N^{-\\delta}$. If this probability does not decay faster than any power of $N$ for some $\\delta\\in(0,1/3)$, Theorem 1.1's bound cannot hold as stated.","tokens_in":2113,"feed_emoji":"⚛️","tokens_out":3269,"duration_ms":87128,"temperature":0.7,"pith_summary":"This paper proves a quantitative propagation of chaos result for systems of N Newtonian particles with a regularized Coulomb interaction, with or without velocity noise. It shows that the k-th marginal of the N-body distribution converges in L1 to the k-fold tensor product of the solution of the Vlasov-Poisson-Fokker-Planck equation when noise is present, or the Vlasov-Poisson equation when noise is absent or vanishing, at explicit rates. The main mechanism is controlling the relative entropy of the N-body density against the tensorized mean-field solution, using previously established probabilistic trajectory closeness between the true and mean-field flows. A reader should care because this gives strong, quantitative validation of effective kinetic equations from underlying many-body dynamics.","feed_headline":"Coulomb particle system converges to Vlasov-Poisson limit","feed_subtitle":"Explicit L1 convergence of all marginals proves Vlasov-Poisson-Fokker-Planck is the right effective equation.","key_machinery":"The central object is the per-particle relative entropy $H_N(f_N^t \\mid f_t^{\\otimes N}) = \\frac{1}{N}\\int f_N^t \\log\\frac{f_N^t}{f_t^{\\otimes N}}\\,dZ_N$. Its time derivative is bounded by a sum of five error terms, each estimating a different source of discrepancy: the difference between true and mean-field trajectories, the difference between the true and regularized force fields, the regularization error, the law-of-large-numbers fluctuation of the empirical force, and the Lipschitz sensitivity of the kernel. Subadditivity of the scaled relative entropy plus the classical entropy-to-L1 inequality convert the entropy bound into L1 convergence of all marginals. The noise strength $\\sigma$ appears in the denominator of the main bound, reflecting that the entropy-production term $\\sigma \\int |\\nabla_v \\log(f_N^t/f_t^{\\otimes N})|^2 f_N^t$ is what absorbs force errors.","core_discovery":"The central claim is that trajectory-level control is enough to upgrade weak propagation of chaos to strong L1 convergence for Vlasov-Poisson-type systems. Concretely, Theorem 1.1 states that for $\\sigma>0$, $\\sup_{0\\le t\\le T} H_N(f_N^t \\mid f_t^{\\otimes N}) \\le C\\exp(C\\sqrt{\\log N})/(\\sigma N^{2\\delta})$, and by the classical relative-entropy-to-L1 inequality and subadditivity this implies $\\|f_N^{t,k} - f_t^{\\otimes k}\\|_{L^1}^2 \\le 2k H_N(f_N^t \\mid f_t^{\\otimes N})$, so marginals converge in L1. Theorem 1.2 improves the rate in $d=3$ to $C(\\log N)^{3/2}/(\\sigma N^{2\\lambda_2})$ using a mollified kernel. Theorem 1.3 extends the argument to the Vlasov-Poisson equation, giving $\\sup_{0\\le t\\le T} H_N(f_N^t \\mid \\tilde f_t^{\\otimes N}) \\le C\\sqrt{\\sigma}(\\sqrt{\\sigma}+1)\\exp(C\\sqrt{\\log N}) + C N^{-\\delta}\\log N$ against the $\\sigma=0$ solution. The proof derives a differential inequality for the relative entropy whose error terms are controlled by the probability that true and effective trajectories stay $N^{-\\delta}$-close.","pith_inferences":["If analogous trajectory-closeness estimates become available for first-order systems or for other singular kernels such as Riesz potentials, the entropy-transport argument in Sections 2 and 3 should transfer directly, giving L1 propagation of chaos there as well.","The $1/\\sigma$ dependence in Theorem 1.1 suggests a quantitative transition: as noise vanishes, the Vlasov-Poisson-Fokker-Planck approximation degrades before the Vlasov-Poisson result takes over, with a crossover governed by $\\sigma_N \\gg N^{-2\\delta}\\exp(-C\\sqrt{\\log N})$.","The polylogarithmic improvement in Theorem 1.2 indicates that the mollified kernel mainly reduces the force-fluctuation error; one could test whether a different regularization achieves a rate closer to $N^{-2/3}$ in three dimensions."],"forward_implications":["For fixed positive noise level, the one-particle marginal tends to the Vlasov-Poisson-Fokker-Planck solution in $L^1(\\mathbb{R}^{2d})$ with rate $C/(\\sqrt{\\sigma}N^{\\delta-})$, so the spatial density also converges in $L^1(\\mathbb{R}^d)$.","In three dimensions with a mollified kernel, the convergence rate improves to polylogarithmic over $N^{2\\lambda_2}$, with $\\lambda_2 \\in (3/10, 1/3)$.","When the noise strength tends to zero, the system can still converge to Vlasov-Poisson: Theorem 1.3 gives closeness to the $\\sigma=0$ solution with error $C\\sqrt{\\sigma}\\exp(C\\sqrt{\\log N}) + C N^{-\\delta}\\log N$.","For any fixed $k$, the $k$-marginal converges in $L^1$ with a constant linear in $k$, so the chaos is quantitative and uniform on the time interval $[0,T]$."],"supporting_citations":[{"why":"Supplies the relative entropy method for second-order systems that the paper adapts to the VPFP setting.","marker":"[28]"},{"why":"Provides the trajectory closeness estimate in probability for the VPFP case, used to control the main error term in Theorem 1.1.","marker":"[10, Lemma 3.2]"},{"why":"Provides the improved trajectory closeness estimate with the mollified kernel, used for Theorem 1.2.","marker":"[27, Theorem 1.2]"},{"why":"Supplies the refined event controlling the difference of empirical forces in the proof of Theorem 1.2.","marker":"[27, Proposition 3.2]"},{"why":"Supplies the original trajectory convergence in probability for the Vlasov-Poisson case, underpinning the VP result and several auxiliary lemmas.","marker":"[35]"},{"why":"Gives the entropy-to-L1 inequality used to convert relative entropy bounds into L1 convergence of marginals.","marker":"[55]"},{"why":"Provides the subadditivity property of scaled relative entropy needed to pass from the N-body entropy to k-marginals.","marker":"[11, Lemma 21]"},{"why":"Provides the Wasserstein stability estimate used to bound the force difference between true and regularized density fields.","marker":"[38, Theorem 2.9]"}],"fun_headline_variants":["Strong chaos propagation for Vlasov-Poisson-Fokker-Planck","Trajectory control upgrades mean-field convergence","Relative entropy proves Vlasov-Poisson mean-field limit","Coulomb particles converge to Vlasov-Poisson-Fokker-Planck","L1 convergence for Vlasov-Poisson mean-field systems"],"cache_read_input_tokens":29824,"weakest_assumption_plain":"The load-bearing premise is that the imported trajectory-comparison estimates, that true and mean-field trajectories stay $N^{-\\delta}$- or $N^{-\\lambda_2}$-close on $[0,T]$ with probability at least $1 - N^{-\\alpha}$ for the stated ranges of $\\delta$, $\\lambda_1$, and $\\lambda_2$, are valid; this paper takes those bounds as input and does not prove them, and without them the entropy inequalities break.","fun_headline_variants_meta":{"raw":{"variants":["Strong chaos propagation for Vlasov-Poisson-Fokker-Planck","Trajectory control upgrades mean-field convergence","Relative entropy proves Vlasov-Poisson mean-field limit","Coulomb particles converge to Vlasov-Poisson-Fokker-Planck","L1 convergence for Vlasov-Poisson mean-field systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000625,"raw_usage":{"total_tokens":2945,"prompt_tokens":1052,"completion_tokens":1893,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":668,"completion_tokens_details":{"reasoning_tokens":1801}},"tokens_in":668,"tokens_out":1893,"duration_ms":15639,"temperature":1.0,"reasoning_tokens":1801,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:22:24.529396+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the N-particle system and the McKean-Vlasov flow for a smooth, compactly supported $f_0$ in $d=3$ with $\\sigma=1$ and the truncated kernel (1.4), and measure the probability of the event $\\sup_{t\\in[0,T]} \\|\\Phi_N^t - \\Psi_N^t\\|_\\infty > N^{-\\delta}$. If this probability does not decay faster than any power of $N$ for some $\\delta\\in(0,1/3)$, Theorem 1.1's bound cannot hold as stated.","supporting_citations":[{"cited_title":"Jabin, and Z","cited_arxiv_id":null,"evidence_quote":"Supplies the relative entropy method for second-order systems that the paper adapts to the VPFP setting."},{"cited_title":"Lazarovici and P","cited_arxiv_id":null,"evidence_quote":"Supplies the original trajectory convergence in probability for the Vlasov-Poisson case, underpinning the VP result and several auxiliary lemmas."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the entropy-to-L1 inequality used to convert relative entropy bounds into L1 convergence of marginals."}],"review_version":1}