{"id":"ceb38431-63c0-414f-96d4-56f7b8869bf8","arxiv_id":"2504.18370","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A well-posedness theory is established for the inhomogeneous Dean-Kawasaki equation with square-root Stratonovich noise, including a new logarithmic regularization effect.","lead":"This paper proves existence, uniqueness, and stability of nonnegative solutions to the inhomogeneous Dean-Kawasaki stochastic PDE with square-root Stratonovich noise. It also shows the logarithm of the solution is integrable, a new regularizing effect that fails for the plain heat equation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Boundary non-vanishing estimate (79) is dimensionally inconsistent; it underpins the boundary part of the dichotomy (15) used in uniqueness step (49).","rationale":"The reader correctly identified the non-vanishing dichotomy (15) as load-bearing and as ultimately depending on the strictly positive quadratic variation ⟨ξ⟩₁. My stress-test focused on how the boundary part of that dichotomy is proved. The displayed estimate (79) is not valid as written, and yet it is the only place where the boundary zero set is controlled. Since step (49) of the uniqueness proof explicitly uses that 1{ρ>0}=1 in the sense of traces on ∂U×[0,T], a gap here touches the central uniqueness claim. However, the gap appears repairable: replacing (79) by the pointwise inequality with a single |log M|^{-1} factor and combining it with the H¹-in-time regularity of (76) should suffice, so I do not see a fatal flaw in the overall strategy. No other internal inconsistency stood out after checking the kinetic formulation, the cutoff-removal order δ→0, M→∞, β→0, and the L¹-initial-data extension via monotone approximation. Because the written proof needs a correction at a load-bearing point, the appropriate verdict is conditional acceptance rather than outright rejection.","tokens_in":47969,"tokens_out":28628,"duration_ms":310967,"concrete_test":"Test inequality (79) directly on U=(0,1), T=1/2, with a boundary set A of positive measure where ρ≡1/M in time and ρ≡1 on the complement: the displayed inequality fails for T<1. Then verify the alternative bound |A_x|≤|log(1/M)|^{-1}|∫_0^T log ρ_M dt| and check, using (76) and the L¹-integrability of log ρ, that ‖Tr∫_0^T log ρ_M‖_{L²(∂U)} grows at most like C|log M|^{1/2}; if this holds, the zero-measure conclusion for {ρ=0} on ∂U×[0,T] is restored.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Definition 2.4(iii) requires {ρ=0} to have zero Lebesgue measure in ∂U×[0,T], and the proof of uniqueness explicitly uses this boundary property in step (49) to let φβ(ρ_i)→1{ρ_i>0}=1 and discard the integrated boundary term. The only argument supplied for the boundary half of (15) is inequality (79) in Theorem 2.18. As written, (79) cannot be correct: the left-hand side is a measure on ∂U×[0,T], while the right-hand side is |log(1/M)|^{-2} times the square of an L2(∂U)-norm of a time integral. Dimensionally the right side carries an extra factor of time. A concrete counterexample to the displayed inequality is obtained by taking, on a set A⊂∂U of surface measure a, ρ=1/M for all t∈[0,T], and ρ=1 elsewhere; then the left side is aT while the right side is aT², which fails whenever T<1. The correct pointwise bound is |A_x|≤|log(1/M)|^{-1}|∫_0^T log ρ_M dt|, losing only one inverse power of |log M|, and this can likely be combined with the H¹-in-time regularity of (76) to restore the boundary non-vanishing conclusion. But that repair is not what the paper displays, and without some such argument the boundary part of the dichotomy is not established by the written proof.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves well-posedness for a conservative stochastic PDE (the inhomogeneous Dean–Kawasaki equation) with correlated, non-stationary Stratonovich noise and square-root noise coefficient, on a smooth bounded domain with Neumann boundary conditions. The solution concept is a stochastic kinetic solution (Definition 2.4), and the main results are uniqueness and pathwise L1-contraction (Theorem 2.8) and existence for L1(Ω;L1(U)) initial data (Theorem 2.18). The novelty is a logarithmic regularization: the Stratonovich-to-Itô correction produces estimates that make log ρ integrable and give time-averaged H1 regularity, which in turn yields a dichotomy: on the event that the initial data are nonzero, the solution is almost everywhere positive both in U×[0,T] and, in trace sense, on ∂U×[0,T]. This dichotomy is used to handle the discontinuous coefficient 1_{ρ>0} in the uniqueness proof.","tokens_in":48255,"tokens_out":5401,"duration_ms":59143,"significance":"If the proof is completed as intended, this is a substantial contribution: it resolves a case explicitly left open by Fehrman and Gess [27], namely conservative square-root Stratonovich noise with non-stationary correlated noise and Neumann boundary conditions. The logarithmic estimates are new even in previously covered settings; they are quantitative, with explicit constants, and they yield a strong pathwise L1 contraction. The paper also gives a clean conceptual explanation of why the Stratonovich correction is needed. However, the manuscript currently contains a load-bearing gap in the proof of the boundary half of the non-vanishing dichotomy, specifically in inequality (79); until that estimate is repaired, the written proof of uniqueness is incomplete because step (49) of Theorem 2.8 relies on the boundary non-vanishing. I view the gap as repairable and the central approach as sound, but the repair is not merely cosmetic.","major_comments":[{"comment":"The displayed inequality (79) is dimensionally inconsistent and, as written, false. The left-hand side is the measure of a subset of ∂U×[0,T] and carries units of surface area multiplied by time, while the right-hand side is |log(1/M)|^{-2} times the square of an L2(∂U)-norm of a time-integrated function and therefore carries units of surface area only. A concrete counterexample is obtained by taking A⊂∂U of surface measure a and setting ρ~=1/M on A for all t∈[0,T] and ρ~=1 elsewhere; then the left-hand side is aT while the right-hand side is aT², which fails whenever T<1. This inequality is the only argument supplied for the boundary half of the non-vanishing condition (15) in Definition 2.4(iii), and the uniqueness proof in Theorem 2.8 explicitly uses the boundary half of (15) in step (49) to discard the boundary term after sending β→0. Thus the boundary non-vanishing is not established by the written proof. The natural repair—use the pointwise bound |A_x| ≤ |log(1/M)|^{-1} |∫_0^T log ρ~_M dt| and then control the resulting integral over ∂U using the H1-in-time regularity from (76)—is plausible, but it is not what the paper displays. This point must be fixed before the main theorem can be accepted.","section":"Theorem 2.18, inequality (79)"}],"minor_comments":[{"comment":"The proof of Proposition 2.11 is a one-sentence sketch citing a Galerkin approximation, the estimates of Proposition 2.12, and Aubin–Lions–Simon compactness. Since this is the existence result for the regularized Neumann problem that underpins all subsequent estimates, a short indication of the Galerkin basis and the passage to the limit would significantly improve verifiability.","section":"Proposition 2.11"},{"comment":"In (84) the displayed equality is missing the final '= 0', and the second integrand contains '√ρ~_k' where '√ρ~' is clearly intended; these errors occur in a key identification of the martingale terms and should be corrected.","section":"Theorem 2.18, near (84)"},{"comment":"The sentence introducing the Skorokhod representation says the conclusion follows 'heuristically' from the Skorokhod theorem and then cites Jakubowski's theorem; since this step deals with non-metrizable weak topologies, the invocation should be stated as a rigorous application rather than as a heuristic.","section":"Theorem 2.18, after (72)"},{"comment":"In the last paragraph of the proof, the passage from L2p(Ω;L2(U)) initial data to L1(Ω;L1(U)) initial data by truncation is compressed: the text says 'a repetition of the above arguments' is used to transfer properties (14), (15), and (17) to the limit, but it should spell out how the monotone limit interacts with the trace-based boundary non-vanishing and with the kinetic measures q_n.","section":"Theorem 2.18, final L1 truncation step"}],"recommendation":"major_revision","confidential_remarks":"The paper is strong in conception and execution: the logarithmic regularization estimates are new, the constants are tracked, and the L1 contraction is a convincing central result. The only serious obstacle is the boundary non-vanishing argument based on inequality (79), which is dimensionally wrong. I believe a repair is within reach along the lines indicated in the major comment, but it is a genuine gap in the written proof rather than a presentation issue. If the author supplies a correct boundary estimate and integrates it with the H1-in-time regularity of (76), I would be inclined to accept."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this paper does real work: it extends the Fehrman–Gess framework to the square-root noise case with non-stationary noise and Neumann boundary conditions, and the log-integrability and time-averaged H1-regularity estimates in Propositions 2.14–2.16 are new even in settings already covered by [27]. The main estimates are written with enough detail that a patient referee can check them, and the uniqueness proof is genuinely careful about the discontinuous coefficient 1{ρ>0}. The self-citation to [27] is legitimate: the new estimates are proven here, not assumed.\n\nSecond, there is a genuine error in the boundary half of the non-vanishing argument. Inequality (79) is dimensionally inconsistent: the left side is a measure on ∂U×[0,T], while the right side is |log(1/M)|^{-2} times the square of an L2(∂U)-norm of a time integral. The right side carries an extra factor of time. The stress-test example is correct: take ρ=1/M on A×[0,T] and ρ=1 elsewhere; then (79) claims aT ≤ aT², which fails for T<1. Since (79) is the only argument supplied for the boundary part of (15), and uniqueness step (49) explicitly uses that boundary non-vanishing to discard the integrated boundary term, the written proof has a gap.\n\nThat gap is repairable, and not a fatal one. Because ρ_M≤1, one has log ρ_M≤0, so pointwise |{t:ρ_M=1/M}| ≤ |log(1/M)|^{-1} |∫ log ρ_M|; integrating over ∂U and using Cauchy–Schwarz gives a bound with only one inverse power of |log M|, which still tends to zero when the trace of ∫ log ρ is in L2(∂U). The paper’s displayed estimate would need to be corrected along these lines, and the trace argument made precise. As it stands, the proof of Theorem 2.18 is incomplete at exactly the point where the boundary dichotomy is established.\n\nOther soft spots are minor. Proposition 2.11 is a one-sentence Galerkin proof, but that is a credible simplification of the surrounding arguments. Some subsequence passages are compressed but not misleading. No quantities are fitted, and the central claims do not reduce to earlier results.\n\nWho is this for: researchers in conservative SPDEs and fluctuating hydrodynamics, especially those working on Dean–Kawasaki-type equations with non-stationary noise. It deserves a serious referee: the advance is substantive and the flaw is isolated and fixable. My recommendation is to send it to peer review and ask the author to repair (79) and the ensuing boundary trace argument.","headline":"Solid extension of the kinetic-solution framework to square-root Dean–Kawasaki noise, with genuinely new log-regularity estimates, but the boundary non-vanishing proof contains a real—and repairable—error in inequality (79).","tokens_in":48755,"tokens_out":8008,"would_cite":true,"duration_ms":89688,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q84","60F10","60H15","60K35","82B21"],"pacs":[],"model":"deepseek-v4-flash","headline":"Dean–Kawasaki with square-root noise now has unique solutions","keywords":["conservative SPDE","Dean–Kawasaki equation","stochastic kinetic solution","Stratonovich noise","square-root noise","non-stationary noise","logarithmic regularization","Neumann boundary condition"],"falsifier":"Take the deterministic heat equation (the case $\\langle\\xi\\rangle_1=0$) in a ball with initial data $1_{B_{1/2}}$ and check whether $\\log\\rho$ is integrable over space and time; direct calculation shows it is not, so a single numerical or analytic verification that the same estimate holds only when $\\langle\\xi\\rangle_1>0$ would settle the role of the noise. Concretely, one could compute the left side of estimate (10) for a family of regularized square-root approximations with $\\langle\\xi\\rangle_1$ approaching zero and watch the logarithmic term diverge.","tokens_in":47761,"feed_emoji":"🌊","tokens_out":8183,"duration_ms":71853,"temperature":0.7,"pith_summary":"One case resisted the author's earlier theory with Gess: conservative stochastic PDEs combining non-stationary Stratonovich noise with square-root nonlinearities, such as the inhomogeneous Dean–Kawasaki equation $\\partial_t\\rho = \\nabla\\cdot(a(x)\\nabla\\varphi(\\rho)) - \\nabla\\cdot(\\sqrt{\\rho}\\circ s(x)\\,d\\xi)$ with Neumann boundary conditions. The paper proves that this equation is well posed: for every nonnegative initial density $\\rho_0\\in L^1(\\Omega;L^1(U))$ there exists a unique stochastic kinetic solution, and any two solutions satisfy an $L^1(U)$ contraction in time. The engine is a Stratonovich-to-Itô correction term proportional to $\\langle\\xi\\rangle_1 \\nabla\\cdot(a\\nabla\\log\\rho)$, whose logarithmic structure is shown to give a genuine time-averaged regularization: on the event that the initial mass is positive, $\\log\\rho$ is space-time integrable and the set where $\\rho=0$ has zero measure. A careful reader would care because this is the natural SPDE for density fluctuations of diffusive particle systems in inhomogeneous media, and the well-posedness closes the square-root gap in the existing theory.","feed_headline":"Dean–Kawasaki with square-root noise now has unique solutions","feed_subtitle":"A Stratonovich correction makes log density integrable, stopping solutions from touching zero and closing a gap in the existing theory.","key_machinery":"The key object is the stochastic kinetic solution: a nonnegative density $\\rho$ whose kinetic function $\\chi_t(x,\\eta)=1_{\\{0<\\eta<\\rho(x,t)\\}}$ satisfies an entropy inequality quantified by a nonnegative kinetic defect measure $q$. The argument is carried by the Stratonovich-to-Itô correction, which in Itô form replaces the naive log term with $\\frac{\\langle\\xi\\rangle_1}{8}\\nabla\\cdot(a\\nabla\\log\\rho) + \\frac{\\langle\\xi\\rangle_1}{4}\\nabla\\cdot(1_{\\{\\rho>0\\}}s(\\nabla\\cdot s^t))$. The paper's load-bearing mechanism is the regularization estimate of Proposition 2.14, which proves that $\\log\\rho$ is space-time integrable even when the initial data is only $L^1$; that fact yields the non-vanishing dichotomy and lets the uniqueness proof remove cutoffs near $\\rho=0$ and near $\\rho=\\infty$ using Proposition 2.6 and the vanishing-at-infinity of the kinetic measure.","core_discovery":"On the paper's own terms, the central discovery is a dichotomy that tames the discontinuous coefficient $1_{\\{\\rho>0\\}}$ in the Itô form (4). The paper shows that any stochastic kinetic solution is either identically zero (on the event $\\rho_0=0$) or positive almost everywhere, in the interior and on the boundary in the sense of trace. This follows from new estimates—entropy, $H^{-1}$ regularity, and Hölder regularity of the time-averaged logarithmic flux—that imply $\\log\\rho$ is integrable and that the set where $\\rho=0$ has zero $(d+1)$-dimensional and boundary $d$-dimensional measure. With the dichotomy in hand, the uniqueness proof can pass through the singularity $\\rho=0$ by treating $1_{\\{\\rho>0\\}}$ as the constant function, and existence follows by taking smooth approximations $\\sigma_n(\\rho)\\to\\sqrt{\\rho}$ and passing to the kinetic limit. The paper states the result as: under Assumptions 2.1 and 2.2, for every $\\rho_0\\in L^1(\\Omega;L^1(U))$ there exists a unique stochastic kinetic solution, and $\\max_{t\\in[0,T]}\\|\\rho_1(t)-\\rho_2(t)\\|_{L^1(U)} \\leq \\|\\rho_{0,1}-\\rho_{0,2}\\|_{L^1(U)}$.","pith_inferences":["I would expect the same logarithmic regularization to operate for other conservative SPDEs whose noise coefficient vanishes like $\\sqrt{\\rho}$ at zero, including equations with rough or white-in-space noise after renormalization, because the mechanism is the sign of the correction term rather than the smoothness of the noise.","The dichotomy suggests a practical simplification for simulation: once positivity is established from the estimates, numerical schemes can drop the indicator function in the Itô correction and track $\\log\\rho$ directly, avoiding the discontinuous coefficient that complicates discretizations.","A testable extension is whether the time-averaged regularity of $L(\\rho)$ holds in the Klimontovich regime (the $\\theta=1$ case) and fails in the Itô regime ($\\theta=0$), since the paper shows the equation is not uniformly parabolic there; checking this numerically would separate the role of the Stratonovich correction from mere noise smoothing."],"forward_implications":["If the theorem is right, the inhomogeneous Dean–Kawasaki equation with square-root noise and Neumann boundary conditions has a complete well-posedness theory, including the $L^1(U)$ contraction that makes solutions depend continuously on their initial data.","The same methods extend, as the paper states in Remark 2.19, to more general noise coefficients satisfying the standing assumptions, and to degenerate diffusions such as porous media nonlinearities $\\varphi(\\rho)=\\rho^m$ for $m\\in(0,\\infty)$.","The quantitative estimates of Propositions 2.14–2.16 become available; they give explicit control of $\\log\\rho$ and of the Hölder regularity of the time-averaged flux $L(\\rho)_t=\\int_0^t(\\varphi(\\rho)+\\frac{\\langle\\xi\\rangle_1}{8}\\log\\rho-\\mathrm{mean})\\,dr$ in $H^1(U)$.","The non-vanishing dichotomy implies that solutions built from positive initial mass never develop a macroscopic vacuum, so the coefficient $1_{\\{\\rho>0\\}}$ is almost surely constant and the equation's Itô form is almost surely smooth in that respect.","Mass is preserved exactly: $\\|\\rho(\\cdot,t)\\|_{L^1(U)}=\\|\\rho_0\\|_{L^1(U)}$ almost surely, which is the property needed if $\\rho$ is to describe empirical densities of diffusing particles."],"supporting_citations":[{"why":"Supplies the stochastic kinetic solution concept and the general well-posedness framework that this paper extends; the square-root non-stationary case is the gap it closes.","marker":"[27]"},{"why":"Provides the kinetic formulation of degenerate parabolic-hyperbolic equations, from which the kinetic function and defect measure here are drawn.","marker":"[6]"},{"why":"Supplies the Itô formula for SPDEs used to derive the Itô form, the entropy estimates, and the Itô step in uniqueness.","marker":"[43]"},{"why":"Gives the convergence lemma used to pass from tightness to a probabilistically strong solution.","marker":"[34]"},{"why":"Gives the quantitative Kolmogorov continuity theorem used to obtain Hölder regularity of the solution and of the time-averaged flux.","marker":"[29]"},{"why":"Supplies the Aubin–Lions compactness criterion used for the Galerkin approximations and tightness.","marker":"[1]"}],"fun_headline_variants":["Unique solutions for Dean–Kawasaki SPDEs with square-root noise","Log-regularization proves uniqueness for Dean–Kawasaki type SPDEs","Stochastic kinetic solutions close gap for square-root noise","Unique kinetic solutions for singular Dean–Kawasaki SPDEs","Now unique: Dean–Kawasaki SPDEs with non-stationary noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the noise has strictly positive, spatially constant quadratic variation $\\langle\\xi\\rangle_1>0$ and that the diffusion $\\varphi$ is uniformly nondegenerate; if $\\langle\\xi\\rangle_1=0$, the logarithmic correction term vanishes and with it the regularization that keeps solutions away from the vacuum set $\\{\\rho=0\\}$.","fun_headline_variants_meta":{"raw":{"variants":["Unique solutions for Dean–Kawasaki SPDEs with square-root noise","Log-regularization proves uniqueness for Dean–Kawasaki type SPDEs","Stochastic kinetic solutions close gap for square-root noise","Unique kinetic solutions for singular Dean–Kawasaki SPDEs","Now unique: Dean–Kawasaki SPDEs with non-stationary noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000778,"raw_usage":{"total_tokens":3462,"prompt_tokens":993,"completion_tokens":2469,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":2372}},"tokens_in":609,"tokens_out":2469,"duration_ms":18753,"temperature":1.0,"reasoning_tokens":2372,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:18:18.076129+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the deterministic heat equation (the case $\\langle\\xi\\rangle_1=0$) in a ball with initial data $1_{B_{1/2}}$ and check whether $\\log\\rho$ is integrable over space and time; direct calculation shows it is not, so a single numerical or analytic verification that the same estimate holds only when $\\langle\\xi\\rangle_1>0$ would settle the role of the noise. Concretely, one could compute the left side of estimate (10) for a family of regularized square-root approximations with $\\langle\\xi\\rangle_1$ approaching zero and watch the logarithmic term diverge.","supporting_citations":[{"cited_title":"Fehrman and B","cited_arxiv_id":null,"evidence_quote":"Supplies the stochastic kinetic solution concept and the general well-posedness framework that this paper extends; the square-root non-stationary case is the gap it closes."},{"cited_title":"Chen and B","cited_arxiv_id":null,"evidence_quote":"Provides the kinetic formulation of degenerate parabolic-hyperbolic equations, from which the kinetic function and defect measure here are drawn."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Itô formula for SPDEs used to derive the Itô form, the entropy estimates, and the Itô step in uniqueness."},{"cited_title":"Gy¨ ongy and N","cited_arxiv_id":null,"evidence_quote":"Gives the convergence lemma used to pass from tightness to a probabilistically strong solution."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the quantitative Kolmogorov continuity theorem used to obtain Hölder regularity of the solution and of the time-averaged flux."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Aubin–Lions compactness criterion used for the Galerkin approximations and tightness."}],"review_version":1}