{"id":"4bca1469-d514-439b-a63c-1e228946239a","arxiv_id":"2501.18745","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For admissible smoothing kernels, the diffusion-velocity approximation to the porous media equation converges in Wasserstein-2 at rate ε^r with r<1/[d(4k+2)], and at rate ε^{1/2} in one dimension.","lead":"This paper proves a quantitative rate at which a smoothed, particle-friendly version of the porous media equation converges to the original equation in the Wasserstein-2 distance. It is the first such rate for diffusion-velocity particle methods, obtained with a new commutator estimate on optimal transport geodesics.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 3.4's C^(3) bound needs a uniform L1 bound on the deconvolution kernel L_{epsilon,eta} that Definition 1.1(v) does not state; without it the proof of the claimed rate does not close.","rationale":"The paper's central claim is a quantitative W2 convergence rate for the diffusion-velocity particle method. The proof architecture is coherent: energy estimates in Lemma 3.1 give uniform H1-type control on the mollified solution, the kernel lemmas control the intermediate scale, and the commutator decomposition in Section 3.3 reduces the rate to three terms. The reader identified the weakest point precisely: the C^(3) bound needs a uniform L1 bound on L_{epsilon,eta} that Definition 1.1(v) does not include. This is load-bearing because C^(3) supplies the epsilon^(1/p) factor that produces the final rate; if it cannot be bounded independently of epsilon, no rate follows. The issue is not a contradiction in the paper but an unstated hypothesis or missing proof step. There are secondary concerns, such as the unproved regularity lemma for optimal transport velocities and the typo in Proposition 2.3, but they do not replace the L1-bound issue as the central obstruction. The concern is fixable by strengthening Definition 1.1 and verifying it for admissible kernels, so the reader's CONDITIONAL verdict remains the right one; my stress-test does not change it.","tokens_in":1246,"tokens_out":1430,"duration_ms":177865,"concrete_test":"For a concrete admissible kernel, e.g. R with Fourier transform (1+|xi|^2)^(-k/2) for k > d/2 + 1 on the torus, compute L_{epsilon,eta} as the inverse Fourier transform of the ratio of Fourier transforms of R_epsilon and R_eta for a sequence epsilon = 2^(-m), eta = 2^(-m-1). Numerically evaluate the L1 norm of L_{epsilon,eta} and the fractional moment integral of |y|^(1/p) |L_{epsilon,eta}(y)|. If the L1 norm grows with epsilon/eta, the proof of C^(3) in Section 3.3 fails for this admissible kernel; if it stays uniformly bounded, the missing step is to state and prove that bound. Also compare the C^(3) integral with L replaced by |L| on a signed example: if the sign changes the estimate, the manuscript must use absolute values throughout.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The rate in Theorem 1.2 rests on Proposition 3.4, whose proof bounds the commutator C^(3) in Section 3.3 by an expression containing the integral of |y|^(1/p) L_{epsilon,eta}(y), which is then estimated as C epsilon^(1/p). This step needs the fractional-moment bound on L_{epsilon,eta}. Definition 1.1(v) supplies only that L_{epsilon,eta} is in L1 and that its first absolute moment is bounded by C epsilon. By Holder's inequality, the fractional moment is bounded by the first moment to the power 1/p times the L1 norm to the power 1/q. Therefore the proof needs a uniform L1 bound on L_{epsilon,eta} in epsilon and eta, but no such bound is stated or proved. The integral of L_{epsilon,eta} equals 1, but that does not imply uniform L1 boundedness because L need not be nonnegative. Moreover, the displayed estimate in the C^(3) bound writes L without absolute values even though L may be signed, unless the kernel is known to be nonnegative. If the uniform L1 bound fails, or if the sign matters, the C^(3) estimate and Proposition 3.4 do not follow, and Theorem 1.2 is not established. This is a genuine gap in the written proof, but it appears fixable by adding the uniform L1 bound as an explicit condition on admissible kernels and proving it for concrete kernels, so the appropriate verdict remains conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a deterministic particle method for the porous media equation (1.1) on the torus T^d, d=1,2,3. It establishes a convergence rate in the Wasserstein-2 metric between the solution u of (1.1) and the solution ũ = u_ε of the smoothed transport equation (1.2) with kernel R_ε, proving W2(u,ũ) ≲ ε^r for any r < 1/[d(4k+2)], where k is a decay exponent of the kernel. A separate one-dimensional result gives the rate ε^{1/2} under a convexity condition on the kernel. The proof uses the gradient-flow structure of the porous media equation in Wasserstein space and introduces a three-term commutator estimate involving an intermediate scale η = ε^γ. The introduction is transparent that only the continuum smoothed equation is analyzed, not the N-particle discretization.","tokens_in":16485,"tokens_out":24307,"duration_ms":207197,"significance":"If the proof were complete, this would be a valuable contribution: it would give the first quantitative rate for diffusion-velocity/regularized-transport approximations of the porous media equation, and the commutator structure is potentially reusable. The paper does not tune parameters against the target rate; the rate emerges from explicit estimates. However, the proof as written rests on several unproved or under-specified regularity statements. In particular, the uniform BV/L∞ bound for optimal-transport velocities (Lemma 2.4) and the integrability of the deconvolution kernel L_{ε,η} (Definition 1.1(v)) are load-bearing and need to be fixed before the claim is established.","major_comments":[{"comment":"Lemma 2.4 claims a uniform bound sup_s (‖v_s‖_{L∞(T^d)} + ‖v_s‖_{BV(T^d)}) ≤ C for the Benamou-Brenier velocity between arbitrary absolutely continuous probability measures, with C independent of the measures. No proof or reference is supplied. This bound is used in every commutator bound in Section 3.3 (through v_1) and in the proof of Theorem 1.4. For measures with densities that are only L∞ and not bounded below, the optimal transport map need not be Lipschitz, and the velocity can have arbitrarily large BV norm; hence the claimed uniformity is not a standard consequence of being on a compact manifold. The authors need to either prove the lemma under the specific L∞-density assumptions or add a correct hypothesis and track the resulting dependence.","section":"2.2 (Lemma 2.4)"},{"comment":"The bound of C^(3) in Section 3.3 requires the fractional-moment estimate ∫ |y|^{1/p}|L_{ε,η}(y)| dy ≤ C ε^{1/p}. By Hölder's inequality, this estimate also requires a uniform bound on ‖L_{ε,η}‖_{L1} in ε,η, which is not part of Definition 1.1(v); the integral of L_{ε,η} equals 1, but L_{ε,η} is generally signed, so this does not control the L1 norm. Moreover, for kernels satisfying the Fourier bounds in Definition 1.1(iii), the ratio R̂(εξ)/R̂(ηξ) typically does not go to 0 as |ξ|→∞ (for a power-law tail it tends to a positive constant), so by the Riemann-Lebesgue lemma L_{ε,η} cannot be an L1 function at all; the deconvolution kernel should be treated as a finite signed measure and a uniform total-variation bound added to the definition. Until this is fixed, Proposition 3.4 and Theorem 1.2 are not proved.","section":"3.3 and Definition 1.1(v)"},{"comment":"The proof of Lemma 3.2 for f ∈ H^{-1} is incomplete. The sentence 'WLOG assume f = ∂_{x1} φ for some φ ∈ L2' does not cover a general H^{-1} element, and the subsequent approximation argument is not written out; the limit passage in the displayed inequalities is not justified. Since Lemma 3.2 is applied to f = ∇ũ ∈ H^{-1} in the bounds of C^(2) and C^(3), a complete proof is needed.","section":"3.3 (Lemma 3.2)"},{"comment":"Proposition 2.3 states only that u ∈ L∞([0,T], L∞(T^d)), which is already contained in Proposition 2.1 and is not the regularity used in the proof of Theorem 1.4. The proof of Theorem 1.4 needs u ∈ L∞([0,T], W^{1,∞}(T)) to justify the estimate |C_{u,ũ}| ≤ C ε. The statement should be corrected or the proof amended.","section":"2.1 (Proposition 2.3)"}],"minor_comments":[{"comment":"In the displayed estimates for C^(2) and C^(3), the integrals involving L_{ε,η}(y) should have absolute values around L_{ε,η}; the text writes e.g. '∫ |y| L_{ε,η}(y) dy' without the absolute value on L.","section":"3.3 (C^(2), C^(3))"},{"comment":"In the proof of Proposition 3.4, the summary term omits the C^(2) contribution ε(ε/η)^{2k}; it is dominated by ε^{1/p}(ε/η)^{2k} for the chosen γ, but the omission should be explained.","section":"3.3 (Proposition 3.4)"},{"comment":"The abstract's phrase 'first quantitative rate for diffusion-velocity particle methods' is stronger than the statements in the paper, since the theorems concern the smoothed transport equation (1.2) and not the N-particle ODE system (1.3). The introduction is transparent about this, but the abstract should be qualified.","section":"Abstract"},{"comment":"There are several LaTeX/OCR artifacts in the displayed text, e.g., '/BD' in the Fourier integrals in Lemma 3.2 and 'P ARTICLE' in the title; these should be cleaned up.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The central idea is promising, and the explicit rate derivation is a strength. However, the two main gaps—Lemma 2.4 and the L_{ε,η} integrability—are serious enough that I cannot recommend acceptance until they are resolved. I would encourage the editor to seek a report from an expert in optimal transport regularity as well as one in Fourier analysis of deconvolution kernels, since the feasibility of the admissibility condition is a key question. The authors should also compare their rate with those in [8,9,10] to substantiate the 'first quantitative rate' claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth knowing: this is the first quantitative Wasserstein-2 rate for the diffusion-velocity particle method applied to the porous medium equation, and the main mechanism—the commutator estimate on the Wasserstein transport map—is a real idea. The paper deserves a serious referee, but the written proof has two gaps that should be closed before I'd trust the theorem exactly as stated.\n\nThe setup is clean. They view (1.1) as the gradient flow of ∫u²/2, replace ∇u by ∇Rε*u, and compare u and uε through the W2 metric. The commutator estimate splits the error into three pieces and uses an intermediate scale Rη to convert the insufficient regularity of uε into L2 bounds on Rε^{1/2}*uε. The resulting rate r < 1/[d(4k+2)] is explicit, and the fact that it degrades as the kernel becomes smoother matches intuition. The one-dimensional theorem with rate ε^{1/2} under a convexity condition is a nice bonus. The energy estimates in Lemma 3.1 are standard but correctly handle the convolution structure. The citations to [8,9,10,18] are appropriate, and the paper does not assume the target result.\n\nNow the soft spots, in order of seriousness. The stress-test note is right: the C^(3) bound in Section 3.3 needs ∫|y|^{1/p}|L_{ε,η}(y)|dy ≤ Cε^{1/p}, but Definition 1.1(v) only gives the first absolute moment and leaves the L1 norm of L_{ε,η} uncontrolled. Hölder gives the fractional moment only if ||L_{ε,η}||_1 is uniformly bounded in ε and η, which is neither stated nor proved. The display also drops the absolute value around L in the integrand; since L can be signed, that matters. This is a genuine gap in the proof of Proposition 3.4, but it looks fixable: add a uniform L1 condition to the admissibility definition and verify it for concrete kernels satisfying the Fourier bounds in iii.\n\nSecond, Lemma 2.4 asserts a universal L∞∩BV bound on optimal transport velocities with no proof and no stated dependence on the measures. On the torus this may be true, but as written it is doing a lot of work. They should either prove it or cite a precise statement.\n\nThird, Proposition 2.3 is misstated—it should be a W^{1,∞} bound, not L∞—and the proof of Theorem 1.4 relies on that missing regularity. Minor, but worth fixing.\n\nAlso keep in mind what the theorem does and does not cover: the rate is for the smoothed PDE (1.2) to (1.1), not for the finite-particle system with N particles. The authors say this explicitly, so I don't count it as a flaw, just a boundary condition.\n\nOverall: the central claim is likely correct, and the gaps are patchable rather than fatal. I'd send it to peer review, and I'd cite it once the final version appears. For a reading group, yes—this is exactly the kind of paper worth pulling apart.","headline":"A genuine first quantitative W2 rate for the diffusion-velocity particle method, built on a plausible commutator estimate; two regularity gaps in the written proof should be fixed before the theorem is fully rigorous.","tokens_in":17016,"tokens_out":3984,"would_cite":true,"duration_ms":37651,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35K65","65M75","49Q22"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves a quantitative convergence rate for a deterministic particle method solving the porous media equation.","keywords":["porous media equation","diffusion-velocity particle method","Wasserstein distance","commutator estimate","aggregation equation","gradient flow","convergence rate","deterministic particle method"],"falsifier":"Compute $L_{\\varepsilon,\\eta}$ for an admissible kernel with polynomial Fourier decay, such as $\\widehat{R}(\\xi)\\sim|\\xi|^{-k}$ at infinity, and measure $\\|L_{\\varepsilon,\\eta}\\|_{L^1}$ and $\\int|y|^{1/p}|L_{\\varepsilon,\\eta}(y)|dy$ as $\\eta/\\varepsilon\\to0$; if the $L^1$ norm grows without bound or the fractional moment grows faster than $\\varepsilon^{1/p}$, the $C^{(3)}$ estimate in Proposition 3.4 collapses. A direct numerical check of the claimed rate in dimensions $d=2$ or $3$ for several admissible kernels would also settle the issue.","tokens_in":15953,"feed_emoji":"💧","tokens_out":4839,"duration_ms":45908,"temperature":0.7,"pith_summary":"The paper establishes the first quantitative convergence rate for a diffusion-velocity particle method applied to the porous media equation. It proves that the weak solution of the smoothed transport equation, obtained by convolving the density with a kernel $R_\\varepsilon$, stays within Wasserstein-2 distance $O(\\varepsilon^r)$ of the true weak solution, where $r<\\frac{1}{d(4k+2)}$ and $k$ is a decay exponent of the kernel's Fourier transform. The argument compares both solutions to an intermediate scale $R_\\eta \\star u_\\varepsilon$ and controls the difference through a new commutator estimate for the optimal-transport velocity field. A separate one-dimensional result gives the better rate $\\varepsilon^{1/2}$ under a convexity condition on the kernel.","feed_headline":"First quantitative rate for diffusion-velocity particle methods","feed_subtitle":"A commutator estimate ties smoothed particle flow to the true porous-media solution, with error decaying as $\\varepsilon^r$.","key_machinery":"The engine is a commutator estimate for the Wasserstein transport map. For two densities, Lemma 2.5 produces action-minimizing velocity fields $v_0,v_1$ with uniform $L^\\infty \\cap BV$ bounds, and rearrangement of the $W_2$ derivative yields one term governed by displacement convexity of the functional $E(\\mu)=\\int \\mu^2/2\\,dx$ plus three commutators $C^{(1)},C^{(2)},C^{(3)}$. These are bounded using the kernel decomposition $R_\\varepsilon=R_\\varepsilon^{1/2}\\star R_\\varepsilon^{1/2}$, the intermediate-scale kernel $L_{\\varepsilon,\\eta}$ defined by $\\widehat{L}_{\\varepsilon,\\eta}=\\widehat{R}_\\varepsilon/\\widehat{R}_\\eta$, the first-moment bound $\\int|x||L_{\\varepsilon,\\eta}(x)|dx\\le C\\varepsilon$, and the energy estimates of Lemma 3.1.","core_discovery":"The central claim is Theorem 1.2: for admissible kernels $R$ satisfying Definition 1.1, the estimate $\\sup_{t\\in[0,T]} W_2(u,u_\\varepsilon) \\le C\\varepsilon^r$ holds for every $r<\\frac{1}{d(4k+2)}$, where $u$ is the weak solution of the porous media equation, $u_\\varepsilon$ is the weak solution of the smoothed transport equation, and $C$ is independent of $\\varepsilon$. If true, this gives the first quantitative convergence rate for diffusion-velocity particle methods solving diffusive equations. The proof uses the Wasserstein gradient-flow structure: the true solution and a suitably mollified version of the approximate solution are both convected by velocity fields, and the evolution inequality for the $W_2$ distance reduces the problem to three commutator terms controlled by kernel moments and energy estimates.","pith_inferences":["The missing uniform $L^1$ bound on $L_{\\varepsilon,\\eta}$ may be provable for natural kernels with algebraically decaying Fourier transforms; if so, the $C^{(3)}$ commutator estimate closes unconditionally.","The same Wasserstein commutator strategy likely transfers to other diffusion equations with displacement-convex energies, such as the heat equation or porous media equations with general exponents $m\\neq 2$, possibly with modified rates.","The predicted tradeoff between kernel smoothness and convergence speed could be tested numerically by comparing kernels with different Fourier decay exponents and measuring the empirical $W_2$ error decay.","A direct numerical check of the rate for $d=2,3$ with a few admissible kernels would confirm whether the constant $C$ is genuinely independent of $\\varepsilon$ in practice."],"forward_implications":["For any admissible kernel, the diffusion-velocity particle method converges to the porous media solution with an explicit algebraic rate in the kernel width $\\varepsilon$.","The Wasserstein-1 error is also controlled, and Corollary 1.3 transfers the rate to an $L^2([0,T]\\times\\mathbb{T}^d)$ estimate for $u-R_\\varepsilon^{1/2}\\star u_\\varepsilon$.","In one spatial dimension, under the kernel convexity condition, the rate improves to $\\varepsilon^{1/2}$.","Because the rate becomes worse as the kernel gets smoother (larger $k$), the result suggests that less regular kernels are preferable for practical particle approximations.","Combined with the known many-particle limit for fixed $\\varepsilon$, the result indicates how to choose $\\varepsilon\\sim N^{-\\alpha}$ to balance particle number and smoothing error."],"supporting_citations":[{"why":"Establishes convergence of the smoothed transport equation to the porous media equation as $\\varepsilon\\to0$ for periodic solutions, the baseline this paper quantifies.","marker":"[18]"},{"why":"Supplies the Benamou-Brenier action-minimizing formulation of $W_2$ used in Lemma 2.5.","marker":"[4]"},{"why":"Provides the Wasserstein gradient-flow structure of the porous media equation and the displacement convexity of $E(\\mu)$.","marker":"[19]"},{"why":"Gives the rigorous metric-space gradient-flow theory underlying the evolution inequality and velocity regularity estimates.","marker":"[1]"},{"why":"Supports existence and uniqueness of weak solutions of the porous media equation used as baseline in Proposition 2.1.","marker":"[5]"},{"why":"Provides the classical porous medium equation theory for well-posedness and regularity used throughout the argument.","marker":"[20]"},{"why":"Supplies background on optimal transport and Wasserstein distances, including the Kantorovich, Monge, and Benamou-Brenier formulations.","marker":"[21]"},{"why":"Introduces the diffusion-velocity particle method and the ODE system (1.3) that the paper analyzes.","marker":"[14]"},{"why":"Establishes the aggregation equation as a gradient flow and its displacement convexity in dimension one, used in the proof of Theorem 1.4.","marker":"[12]"}],"fun_headline_variants":["First quantitative rate for particle methods on porous media","Porous media equation solved with deterministic particles, rate proven","New commutator estimate yields convergence rate for porous media","Particle method for porous media gets first convergence guarantee","Wasserstein-2 rate proven for diffusion-velocity particle methods"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the $C^{(3)}$ bound requires the fractional moment estimate $\\int|y|^{1/p}|L_{\\varepsilon,\\eta}(y)|dy\\le C\\varepsilon^{1/p}$, but Definition 1.1 only states $L_{\\varepsilon,\\eta}\\in L^1$ and the first-moment bound $\\int|x||L_{\\varepsilon,\\eta}(x)|dx\\le C\\varepsilon$, so the fractional bound is guaranteed only if $\\|L_{\\varepsilon,\\eta}\\|_{L^1}$ is uniformly bounded in $\\varepsilon,\\eta$, a condition neither stated nor proved.","fun_headline_variants_meta":{"raw":{"variants":["First quantitative rate for particle methods on porous media","Porous media equation solved with deterministic particles, rate proven","New commutator estimate yields convergence rate for porous media","Particle method for porous media gets first convergence guarantee","Wasserstein-2 rate proven for diffusion-velocity particle methods"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000537,"raw_usage":{"total_tokens":2495,"prompt_tokens":778,"completion_tokens":1717,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":394,"completion_tokens_details":{"reasoning_tokens":1652}},"tokens_in":394,"tokens_out":1717,"duration_ms":10437,"temperature":1.0,"reasoning_tokens":1652,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T22:43:55.322442+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $L_{\\varepsilon,\\eta}$ for an admissible kernel with polynomial Fourier decay, such as $\\widehat{R}(\\xi)\\sim|\\xi|^{-k}$ at infinity, and measure $\\|L_{\\varepsilon,\\eta}\\|_{L^1}$ and $\\int|y|^{1/p}|L_{\\varepsilon,\\eta}(y)|dy$ as $\\eta/\\varepsilon\\to0$; if the $L^1$ norm grows without bound or the fractional moment grows faster than $\\varepsilon^{1/p}$, the $C^{(3)}$ estimate in Proposition 3.4 collapses. A direct numerical check of the claimed rate in dimensions $d=2$ or $3$ for several admissible kernels would also settle the issue.","supporting_citations":[{"cited_title":"Lions and S","cited_arxiv_id":null,"evidence_quote":"Establishes convergence of the smoothed transport equation to the porous media equation as $\\varepsilon\\to0$ for periodic solutions, the baseline this paper quantifies."},{"cited_title":"Benamou and Y","cited_arxiv_id":null,"evidence_quote":"Supplies the Benamou-Brenier action-minimizing formulation of $W_2$ used in Lemma 2.5."},{"cited_title":"Otto , The geometry of dissipative evolution equations: the porou s medium equation , Comm","cited_arxiv_id":null,"evidence_quote":"Provides the Wasserstein gradient-flow structure of the porous media equation and the displacement convexity of $E(\\mu)$."},{"cited_title":"Ambrosio, N","cited_arxiv_id":null,"evidence_quote":"Gives the rigorous metric-space gradient-flow theory underlying the evolution inequality and velocity regularity estimates."},{"cited_title":"B ´enilan, M","cited_arxiv_id":null,"evidence_quote":"Supports existence and uniqueness of weak solutions of the porous media equation used as baseline in Proposition 2.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the classical porous medium equation theory for well-posedness and regularity used throughout the argument."},{"cited_title":"Villani , Optimal Transport: Old and New , vol","cited_arxiv_id":null,"evidence_quote":"Supplies background on optimal transport and Wasserstein distances, including the Kantorovich, Monge, and Benamou-Brenier formulations."},{"cited_title":"Degond and F.-J","cited_arxiv_id":null,"evidence_quote":"Introduces the diffusion-velocity particle method and the ODE system (1.3) that the paper analyzes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the aggregation equation as a gradient flow and its displacement convexity in dimension one, used in the proof of Theorem 1.4."}],"review_version":1}