{"id":"3346e173-c5ac-4714-92f1-4690f4e698c4","arxiv_id":"2411.14334","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For second-order Langevin particle systems, the associated Dean-Kawasaki SPDE is well posed exactly for atomic, empirical-measure initial data and provably ill posed for smooth initial data.","lead":"This paper proves a stark dichotomy for a large class of stochastic PDEs that describe particles with inertia: exact solutions exist only when the initial condition is a collection of point particles, and smoothed initial data have no solution. The result makes rigorous a family of fluctuating hydrodynamics equations used in active matter and confirms they are exact encodings of particle systems, not approximations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified","rationale":"I read the proof of Theorem 3.5 as sound under its stated hypotheses. The Laplace duality, the rigidity argument via analyticity of the moment generating function, and the Girsanov reduction are all coherent, modulo minor presentational typos in α factors that do not affect the mathematics. The reader's weakest-assumption identification is accurate: the interacting nonexistence proof genuinely depends on the gradient structure (F.1), and for general non-gradient Vlasov-Fokker-Planck interactions the dichotomy remains open. However, this is a stated assumption of the theorem, not an unacknowledged gap, so it does not change the verdict on the paper's central claim as formulated. The only concrete check I would want is an independent re-derivation of the Girsanov calculation with all scalings, since that is the single step where a sign or scaling error would invalidate the interacting case.","tokens_in":20758,"tokens_out":36505,"duration_ms":355416,"concrete_test":"As a verification of the one delicate step, re-derive Proposition 5.9 directly from Definition 3.1: apply the Itô formula with E = αG, verify the drift contribution αLδG/δµ + αF·∇δG/δµ and the covariation [M^{αG}, ⟨µ, φ⟩] = α⟨µ, (σᵀ∇δG)·(σᵀ∇φ)⟩, and confirm that the density ε(−M^{αG}) removes exactly the αF·∇φ term; if any α factor is misplaced, the interacting reduction fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central dichotomy of Theorem 3.5 is established carefully: Laplace duality gives uniqueness and rigidity for the interaction-free case, and the Girsanov reduction in Proposition 5.9 correctly transfers these to the interacting case under the stated fluctuation-dissipation condition (F.1). The proof is internally consistent, and the atomic/empirical existence direction follows by direct Itô calculation from the particle SDE. The principal scope restriction is that the interacting nonexistence argument requires the gradient form F_µ = σσᵀ∇δG/δµ; for non-gradient interactions the paper proves nothing about smooth initial data. This is explicitly stated as Assumption 3.4, so it is a limitation of scope rather than a defect in the central argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Cauchy problem for the stochastic Vlasov-Fokker-Planck / Dean-Kawasaki type SPDE ∂tµ = αL*µ + α∇·(µF_µ) + ∇·(√µ σ ˙W) over probability measures on R^k, with L a (possibly degenerate) diffusion generator and F an interaction satisfying a fluctuation-dissipation gradient condition. The main result (Theorem 3.5) asserts a dichotomy: a martingale solution exists if and only if α is a positive integer n and the initial datum is an empirical measure 1/n∑δ_{z^i}; the solution is then the empirical measure of the corresponding n-particle SDE. Consequently smooth initial data admit no solution. The argument uses a Laplace duality for the interaction-free case to derive uniqueness and rigidity of solutions, and a Girsanov reduction to transfer the result to the interacting case. Examples cover inertial Langevin dynamics, active particle models, and flocking/alignment models.","tokens_in":20791,"tokens_out":40230,"duration_ms":347431,"significance":"If correct, the theorem is a substantial extension of the known ill-posedness/triviality result for the Dean-Kawasaki equation [32,31] to second-order, hypoelliptic, and non-reversible settings. It gives a mathematically precise interpretation of fluctuating hydrodynamics as an exact finite-particle representation, and it provides a general martingale framework (Assumption 3.2) that is verified on several physically relevant examples. The paper is honest about the restrictive gradient condition (F.1) on interactions, and the nonexistence theorem for smooth data in the interacting case is conditional on it. The proof strategy—Laplace duality, moment-generating-function rigidity, and Girsanov reduction—is well suited to the problem and, once the factor errors noted below are fixed, is convincing.","major_comments":[{"comment":"The time-dependent martingale used in the duality proof is displayed with generator L rather than αL: M_t(ψ_·) := ⟨µ_t,ψ_t⟩−⟨µ_0,ψ_0⟩−∫⟨µ_s, Lψ_s+∂_sψ_s⟩ ds. Since (14) has generator αL*, this process is generally not a martingale for solutions of (14). The subsequent Itô computation correctly uses αL V_{t−s}φ, so the fixed formula should read αLψ_s; please correct the display and check the intermediate steps.","section":"Section 5.1, Proposition 5.2"},{"comment":"Proposition 5.6 states the Itô formula with drift term ⟨µ_s, αL δE/δµ + F_µs·∇δE/δµ⟩, omitting the factor α in front of F. Definition 3.1 has αLφ+αF_µs·∇φ, so the displayed formula fails already for E(µ)=⟨µ,φ⟩. The proof sketch for cylinder functions produces the αF term, so the statement has a typographical omission, but as written the formula is wrong. Since the Girsanov density in Proposition 5.9 is built from this Itô formula, the factor must be corrected for the interacting reduction to be valid.","section":"Section 5.2, Proposition 5.6"},{"comment":"The statement defines Q by the stochastic exponential E(M^G), while the proof uses E(M^{αG}). These are not interchangeable: the cross-variation of M^G with M(φ) is ⟨µ,∇φ·F⟩, so E(M^G) changes the drift by only F·∇φ, not αF·∇φ; the correct density for removing the interaction term α∇·(µF) is E(M^{αG}). The proof's use of M^{αG} is right, but the proposition statement and the subsequent proof of Theorem 3.5 for F≠0 (which writes E(M^G(µ_·))) need to be made consistent.","section":"Section 5.2, Proposition 5.9"}],"minor_comments":[{"comment":"The statement writes |H(µ,z)| instead of |F(µ,z)|, and the Fatou display in the proof should take liminf over N (with the stopping time kept) rather than over R. The argument is otherwise clear.","section":"Lemma 5.5"},{"comment":"In the verification of condition (F.1) for the interacting Langevin model, the displayed expression for δG/δµ has the second term ∫ v′·f(x,x′)dµ(x′,v′); the correct functional derivative contains ∫ v′·f(x′,x)dµ(x′,v′). This does not affect the computed ∇_v derivative, but the formula should be corrected.","section":"Section 4.3"},{"comment":"The phrase 'if and only if α=n for some n∈N' should be read together with the requirement that µ0 be an n-point empirical measure of the stated form; as written, the sentence could be misread as asserting existence for every µ0 when α is an integer. Please rephrase for clarity.","section":"Theorem 3.5"},{"comment":"In the verification of (L.4), the Girsanov display should be an inequality (probability ≥ E^0[M_t 1_{...}]) rather than an equality; the intended estimate is clear but the formula as printed is not quite accurate.","section":"Section 4.1"}],"recommendation":"major_revision","confidential_remarks":"The factor-α inconsistencies in Section 5 appear to be typographical rather than conceptual; the intended Girsanov argument with M^{αG} is standard and fixable. After a careful correction of Proposition 5.2, Proposition 5.6, and Proposition 5.9, the paper would be a solid contribution. The reader's report's accept verdict is premature given these load-bearing display errors."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:2411.14334.\n\nThe paper proves the Dean-Kawasaki existence/nonexistence dichotomy for a class of second-order, possibly hypoelliptic Vlasov-Fokker-Planck SPDEs, where previously it was known only for first-order diffusive systems. Theorem 3.5 is the core: solutions exist exactly when the noise scaling alpha is an integer n and the initial data is the empirical measure of n particles; smooth initial data gives no solution. That is a real extension, and the examples (inertial Langevin, active matter, flocking) make clear the scope.\n\nWhat the paper does well: the Laplace duality (Prop. 5.2) and rigidity (Prop. 5.4) are written carefully, with the technical Assumption 3.2 doing the work for degenerate non-reversible diffusions. The Girsanov reduction in Prop. 5.9 is correct; I checked the sign on the density, and the drift removal works out. The authors also verify Assumption 3.2 in detail for the main examples rather than citing abstract semigroup theory, which is useful. The Section 2 derivation is formal, but they are upfront about it, and Theorem 3.5 gives the rigorous post hoc justification. Good, honest paper.\n\nSoft spots, in proportion. (1) The interacting case requires the fluctuation-dissipation gradient form (F.1), F = sigma sigma^T grad(delta G/delta mu). For non-gradient interactions there is no nonexistence theorem. This is stated as an assumption, so it is a scope limitation, not a hidden flaw, but it is the main restriction. (2) Assumption 3.2 is load-bearing and not trivial to check; the paper gives verifications for the examples but no general recipe. (3) The existence direction for F not equal to 0 is dispatched in one terse converse-Girsanov sentence; it is standard, but a referee may ask for a few more lines. (4) Minor typos: 'space marginal' in (7) should be 'velocity marginal', and 'dQ = dQ' in Prop. 5.9. None of this threatens the central argument.\n\nWho this is for: anyone working on singular SPDEs, Dean-Kawasaki equations, or fluctuating hydrodynamics for active matter. I would send it to peer review; it deserves a careful referee. My own verdict is accept after minor revision, mostly for the typos and a small expansion of the Girsanov existence argument.","headline":"Solid extension of the Dean-Kawasaki existence/nonexistence dichotomy to second-order and hypoelliptic systems, with the main caveat being the gradient-structure assumption on interactions.","tokens_in":21365,"tokens_out":17664,"would_cite":true,"duration_ms":152871,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H15","35R60","82C22","35Q83","60J60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that stochastic Vlasov-Fokker-Planck (Dean-Kawasaki-type) equations for second-order Langevin systems are well-posed exactly for n-particle empirical initial data and become insoluble for smooth initial data.","keywords":["Dean-Kawasaki equation","fluctuating hydrodynamics","Vlasov-Fokker-Planck equation","martingale solutions","empirical measures","ill-posedness","hypoelliptic diffusions","Girsanov transform"],"falsifier":"Take $\\mu_0$ to be a smooth density with finite second moment, say a Gaussian on $\\mathbb{R}^{2d}$, fix $\\alpha=1$, and look for a continuous $\\mathcal{M}_1$-valued process satisfying the martingale problem of Definition 3.1; the theorem predicts none exists, so any explicit or numerical construction of such a process would refute Theorem 3.5. For the interacting case, a sharper test is to take a non-gradient velocity-alignment force $F$ and check whether smooth-data solutions appear, which would indicate that the gradient fluctuation-dissipation condition is genuinely necessary.","tokens_in":20521,"feed_emoji":"🎲","tokens_out":15110,"duration_ms":126929,"temperature":0.7,"pith_summary":"Fluctuating-hydrodynamics equations are stochastic PDEs intended to describe finite collections of particles, but the standard physical derivation replaces a particle-dependent noise by a statistically similar one. This paper gives a rigorous martingale derivation for second-order Langevin systems and proves that the resulting stochastic Vlasov-Fokker-Planck (Dean-Kawasaki-type) equations are exact encodings of the particle dynamics, not approximations. Its main theorem extends the known first-order dichotomy: a solution exists if and only if the noise scale $\\alpha$ is a positive integer $n$ and the initial measure is the empirical measure of $n$ particles, and then the solution is the empirical measure of the particle trajectories. Smoothing the initial data destroys existence, and for general total mass $m$ existence is equivalent to $\\alpha m \\in \\mathbb{N}$. The result covers inertial Langevin dynamics and active-matter or flocking models whose interaction has the fluctuation-dissipation gradient form $F_\\mu = \\sigma\\sigma^T\\nabla \\frac{\\delta G}{\\delta\\mu}$.","feed_headline":"Dean-Kawasaki SPDEs admit no smooth solutions","feed_subtitle":"For second-order Langevin systems, only n-particle empirical measures solve them, and only at noise scale n.","key_machinery":"The machinery is the weak martingale-solution formulation on the space $\\mathcal{M}_1(\\mathbb{R}^k)$ of probability measures. For $F=0$, uniqueness and atomicity follow from a Laplace duality, $\\mathbb{E}[e^{-\\langle\\mu_t,\\varphi\\rangle}] = e^{-\\langle\\mu_0,V_t\\varphi\\rangle}$, where $V_t\\varphi = -\\alpha\\ln(P_{\\alpha t} e^{-\\varphi/\\alpha})$ solves the Hamilton-Jacobi-Bellman equation $\\partial_t\\psi = \\alpha L\\psi - \\Gamma(\\psi)$, with $\\Gamma(f)=\\frac{1}{2}|\\sigma^T\\nabla f|^2$ the carré du champ. The Cole-Hopf form makes the moment generating function of $\\alpha\\mu_t(A)$ analytic near $0$, forcing $\\alpha\\mu_t(A)\\in\\mathbb{N}$ and hence atomicity, via the exhaustion condition (L.4). The interacting case is reduced to $F=0$ by the Girsanov transform of Proposition 5.9, which is why the proof needs the fluctuation-dissipation gradient condition $F_\\mu = \\sigma\\sigma^T\\nabla\\frac{\\delta G}{\\delta\\mu}$ plus the growth bounds of Assumption 3.4. Assumptions (L.1)-(L.4) supply the Feller semigroup regularity, smoothing estimates, and stability under initial data needed to run these steps for degenerate hypoelliptic generators.","core_discovery":"On the paper's own terms, the central result is Theorem 3.5. For the class of equations $\\partial_t\\mu_t = \\alpha L^*\\mu_t + \\alpha\\nabla\\cdot(\\mu_t F_{\\mu_t}) + \\nabla\\cdot(\\sqrt{\\mu_t}\\,\\sigma\\, \\dot{W}_{z,t})$ with $L = b\\cdot\\nabla + \\frac{1}{2}\\sigma\\sigma^T:\\nabla^2$ and initial data $\\mu_0$ a probability measure with finite second moment, a martingale solution exists if and only if $\\alpha = n\\in\\mathbb{N}$ and $\\mu_0 = \\frac{1}{n}\\sum_{i=1}^n \\delta_{z_i}$; the unique-in-law solution is $\\mu_t = \\frac{1}{n}\\sum_{i=1}^n\\delta_{z_i(t)}$, where the $z_i$ solve $dz_i = (\\alpha b(z_i)+\\alpha F_{\\mu_t}(z_i))\\,dt + \\sqrt{\\alpha}\\,\\sigma(z_i)\\,dW_i$. In particular, no solution exists for smooth or otherwise non-atomic initial data, so the SPDE is an exact transcription of the particle system rather than a free-standing PDE. The same dichotomy is proved for degenerate, hypoelliptic, and non-reversible generators, and for interactions of the fluctuation-dissipation gradient form.","pith_inferences":["The paper's gradient condition is sufficient; whether it is necessary is open. If a non-gradient velocity-alignment force admitted smooth-data solutions, the dichotomy would fail outside the fluctuation-dissipation class, so testing such models is a natural next step.","Because the empirical measure path determines the underlying particle trajectories up to relabelling, exactness suggests that inferring the interaction $F$ from fluctuating-hydrodynamics data is equivalent to inferring it from trajectories, which may simplify identifiability questions for active-matter models.","The same square-root multiplicative noise with fluctuation-dissipation balance appears in other conservative SPDEs; by analogy one would expect atomic-only well-posedness there as well, although the paper does not address those equations."],"forward_implications":["For every $n$-particle second-order Langevin system covered by the assumptions, the Dean-Kawasaki-type SPDE is an exact representation: its only solution is the empirical measure of the particle trajectories, so the SPDE carries no information beyond the particle system.","Mollifying the initial data, no matter how slightly, makes the SPDE unsolvable; numerical methods that assume a smooth density profile therefore cannot be used without regularising the noise.","The dichotomy extends to hypoelliptic and non-reversible generators, covering inertial Langevin dynamics, active swimmers, and flocking models with gradient-type interactions.","For initial data of total mass $m$, solutions exist exactly when $\\alpha m\\in\\mathbb{N}$ and take the form $\\frac{1}{\\alpha}\\sum_{i=1}^{\\alpha m}\\delta_{z_i(\\alpha t)}$.","The martingale derivation supplies a rigorous replacement for the noise-substitution heuristic used to write down equations of fluctuating hydrodynamics."],"supporting_citations":[{"why":"Supplies the original Langevin-equation derivation of density dynamics for interacting particles, including the noise-replacement step the present martingale argument makes rigorous.","marker":"[14]"},{"why":"Establishes the first-order Dean-Kawasaki ill-posedness-versus-triviality dichotomy whose Laplace-duality proof is extended here to degenerate non-reversible generators.","marker":"[32]"},{"why":"Gives the interacting Dean-Kawasaki case with smooth drift potential and the Girsanov reduction that Proposition 5.9 adapts to Vlasov-Fokker-Planck-type equations.","marker":"[31]"},{"why":"Provides the moment-generating-function atomicity lemma and the time-dependent martingale-solution machinery used for rigidity and for the general-mass case.","marker":"[33]"},{"why":"Supplies the semigroup smoothing estimates used to verify Assumptions (L.2) and (L.3) in the inertial Langevin example.","marker":"[13]"},{"why":"Provides the strong Feller property and Girsanov estimate used to verify the exhaustion condition (L.4).","marker":"[45]"},{"why":"Provides the martingale-problem and uniqueness-of-law framework underlying the definition of solutions.","marker":"[20]"}],"fun_headline_variants":["Dean-Kawasaki: only particles, no smooth solutions","SPDE solutions only for atomic data, never smooth","No smooth solutions for second-order Langevin SPDEs","Dean-Kawasaki SPDEs: atomic only, smooth impossible","Exact particle SPDE: no solutions for smooth data"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the interaction has the fluctuation-dissipation gradient form $F_\\mu = \\sigma\\sigma^T\\nabla\\frac{\\delta G}{\\delta\\mu}$ with the growth bounds of Assumption 3.4, together with the semigroup conditions (L.1)-(L.4); if any of these fails, the paper proves neither existence nor the smooth-data nonexistence for interacting systems.","fun_headline_variants_meta":{"raw":{"variants":["Dean-Kawasaki: only particles, no smooth solutions","SPDE solutions only for atomic data, never smooth","No smooth solutions for second-order Langevin SPDEs","Dean-Kawasaki SPDEs: atomic only, smooth impossible","Exact particle SPDE: no solutions for smooth data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000426,"raw_usage":{"total_tokens":2218,"prompt_tokens":1020,"completion_tokens":1198,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":636,"completion_tokens_details":{"reasoning_tokens":1116}},"tokens_in":636,"tokens_out":1198,"duration_ms":8472,"temperature":1.0,"reasoning_tokens":1116,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:17:59.586560+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $\\mu_0$ to be a smooth density with finite second moment, say a Gaussian on $\\mathbb{R}^{2d}$, fix $\\alpha=1$, and look for a continuous $\\mathcal{M}_1$-valued process satisfying the martingale problem of Definition 3.1; the theorem predicts none exists, so any explicit or numerical construction of such a process would refute Theorem 3.5. For the interacting case, a sharper test is to take a non-gradient velocity-alignment force $F$ and check whether smooth-data solutions appear, which would indicate that the gradient fluctuation-dissipation condition is genuinely necessary.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the original Langevin-equation derivation of density dynamics for interacting particles, including the noise-replacement step the present martingale argument makes rigorous."},{"cited_title":"von Renesse","cited_arxiv_id":null,"evidence_quote":"Establishes the first-order Dean-Kawasaki ill-posedness-versus-triviality dichotomy whose Laplace-duality proof is extended here to degenerate non-reversible generators."},{"cited_title":"On Dean-Kawasaki dynamics with smooth drift potential","cited_arxiv_id":null,"evidence_quote":"Gives the interacting Dean-Kawasaki case with smooth drift potential and the Girsanov reduction that Proposition 5.9 adapts to Vlasov-Fokker-Planck-type equations."},{"cited_title":"Dean–Kawasaki equation with initial con- dition in the space of positive distributions","cited_arxiv_id":null,"evidence_quote":"Provides the moment-generating-function atomicity lemma and the time-dependent martingale-solution machinery used for rigidity and for the general-mass case."},{"cited_title":"Alexandru Myller","cited_arxiv_id":null,"evidence_quote":"Supplies the semigroup smoothing estimates used to verify Assumptions (L.2) and (L.3) in the inertial Langevin example."},{"cited_title":"Large and moderate deviations and exponential convergence for stochastic damping Hamiltonian systems","cited_arxiv_id":null,"evidence_quote":"Provides the strong Feller property and Girsanov estimate used to verify the exhaustion condition (L.4)."},{"cited_title":"Ethier and Thomas G","cited_arxiv_id":null,"evidence_quote":"Provides the martingale-problem and uniqueness-of-law framework underlying the definition of solutions."}],"review_version":1}