Pith. sign in

REVIEW 2 major objections 4 minor 29 references

Ill-posedness of the pure-noise Dean-Kawasaki equation

T0 review · 2 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The pure-noise Dean-Kawasaki equation has no solutions for any finite initial measure when the drift is bounded.

desk verdict The result is real and nearly right; two fixable glitches — zero initial measure and a Brownian normalization typo — keep it from being claimable as stated. read the letter →

arxiv 2501.09677 v1 pith:LOCQMCZM submitted 2025-01-16 math.PR

Lorenzo Dello Schiavo , Vitalii Konarovskyi This is my paper · ORCID
classification math.PR MSC 60H1560G5782C31
keywords Dean-Kawasakiequationpure-noiseSPDEmartingalesolutionmeasure-valueddiffusionill-posednessspace-timewhitenoiseWassersteinfiniteBorelmeasures
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proves that the Dean-Kawasaki-type stochastic partial differential equation with a square-root multiplicative white noise and a bounded measurable drift term has no continuous measure-valued martingale solutions for any finite initial measure on Euclidean space or the flat torus. This includes the pure-noise case, where the drift is absent, so the failure is not caused by a particular drift choice. The proof assumes a solution exists, uses the solution's total mass and time-averaged marginals to build a test function whose gradient has absolute value one almost everywhere, and then extracts a Brownian motion that forces the solution to assign negative mass to a nonnegative function. If correct, the equation is ill-posed in the strongest possible sense: no initial condition admits a solution in this class.

What carries the argument

The central object is a bounded, nonnegative, piecewise affine function $f(x)=g(x_1)$ on $\mathbb{R}^d$ or $\mathbb{T}^d$ whose gradient satisfies $|\nabla f|=1$ almost everywhere and whose singular set $\Sigma_f$ has zero solution measure in time average, as in equation (2.11). The proof regularizes $f$ to smooth test functions $f_n$, uses Proposition 2.3 to pass the martingale property to the limit, and obtains a martingale $B_t$ with quadratic variation $ct$; L\'evy's characterization then turns $B_t/c$ into a standard Brownian motion. The negative tail of that Brownian motion makes $\mu_T f<0$ with positive probability, contradicting the nonnegativity of $f$ and of the measure-valued solution.

What would settle it

Exhibit a finite positive Borel initial measure and a bounded measurable drift $H$ for which the martingale problem in Definition 1.1 admits a continuous solution on $\mathbb{R}^d$ or $\mathbb{T}^d$. Theorem 1.2 asserts that no such pair exists, so any explicit example would refute it; short of that, checking the key identity (2.11) on a candidate solution would pinpoint where the argument must break.

Watch

Extended reading notes

Core claim

Theorem 1.2 asserts that for $\alpha=0$ and $G(\mu)=\nabla\cdot(\mu H(\mu))$ with $H:\mathcal{M}_b^+\to\mathbb{R}^d$ bounded and Borel measurable, equation (1.1) has no solutions for any initial condition $\mu_0\in\mathcal{M}_b^+$ on $\mathbb{R}^d$ or $\mathbb{T}^d$. The proof constructs a bounded nonnegative piecewise affine function $f$ with $|\nabla f|=1$ almost everywhere and with singular set that is never charged by the solution in time average; this lets the authors pass from regularized smooth test functions to $f$ in the martingale problem. The resulting process is a martingale with quadratic variation $ct$, so by L\'evy's characterization it is a Brownian motion, and its negative tail makes $\mu_T f$ negative with positive probability, contradicting $f\ge 0$ and the nonnegativity of the measure-valued solution.

Load-bearing premise

The contradiction rests on the existence of a nonnegative piecewise affine function whose gradient has absolute value one almost everywhere and whose singular set is not charged by the time-averaged solution measure; on $\mathbb{R}^d$ and $\mathbb{T}^d$ this follows because a finite measure has only countably many atoms, while on a general manifold the cut locus may make the singular set too large.

Editorial extensions

If this is right

  • The pure-noise equation ($H\equiv 0$) is not merely non-unique or unstable: no finite initial measure admits a continuous martingale solution, so the ill-posedness is total within this solution class.
  • Allowing unbounded drift is the only known route to solvability: the singular drift in equation (1.3) has solutions, so the boundedness assumption in Theorem 1.2 marks a sharp boundary.
  • The same nonexistence holds on the flat torus $\mathbb{T}^d$, ruling out explanations based on noncompactness of the spatial domain.
  • On manifolds with one chart and on standard spheres the proof extends, suggesting the obstruction is a structural feature of the equation rather than of flat space.
  • For $\alpha>0$, solutions exist exactly when the initial datum is an empirical measure of particles with mass $1/\alpha$; for $\alpha=0$ no initial datum works, so the parameter $\alpha=0$ is a singular limit.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the proof technique should transfer to other measure-valued SPDEs with the same square-root noise structure, where extracting a Brownian motion and testing against a nonnegative sawtooth function can yield nonexistence even when the drift has a different form.
  • Editorial inference: the sharp bounded-versus-unbounded drift boundary suggests that any well-posed regularisation of the pure-noise equation must either smooth the noise or allow $H$ to grow without bound; discrete-particle models correspond to the second option through singular empirical drifts.
  • Editorial inference: on manifolds whose cut locus is large, the theorem leaves open the possibility that solutions exist; if so, the obstruction is genuinely geometric rather than analytical.
  • Editorial inference: a concrete numerical check would be to approximate the equation with a vanishing viscosity and bounded $H$; the theorem predicts no limiting continuous measure-valued solution, so the approximations must show developing singularities or sign-changing approximants.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper studies the Dean-Kawasaki-type SPDE (1.1) on R^d or T^d in the case alpha = 0, with drift G(mu) = div(mu H(mu)) for a bounded Borel map H on the space of finite positive measures. Theorem 1.2 asserts that this equation admits no continuous measure-valued martingale solutions for any initial measure and any such bounded H. The proof assumes a solution, constructs a piecewise affine test function g whose singular set avoids the atoms of a time-averaged projected measure, regularizes g to obtain C^2 approximations, passes to the limit via Proposition 2.3 to obtain a continuous martingale B with quadratic variation c t (where c = mu_0(M_d)), and then obtains a contradiction by producing a positive-probability event on which mu_T f < 0 while f is nonnegative. Section 3 discusses possible extensions to manifolds and explicitly identifies the cut-locus obstruction.

Significance. For nonzero initial measures, the result is significant: it completes the picture begun in earlier work for alpha > 0 and sharply contrasts with the known existence of solutions for unbounded H, showing that the pure-noise Dean-Kawasaki equation with bounded drift is ill-posed on flat spaces and tori. The proof is largely self-contained and transparent, and Proposition 2.3, which extends the martingale formulation to non-smooth test functions under an L^1(integral mu_s ds) convergence condition, is a useful standalone tool. The authors are also honest about the limitations of the argument on general manifolds, which is a strength of the exposition.

major comments (2)
  1. [Section 1.1, Theorem 1.2 and proof of Theorem 1.2] The theorem as stated includes the zero initial measure mu_0 = 0 if M_b^+ is understood in the standard sense of nonnegative finite Borel measures. But the proof starts with c := mu_0(M_d) > 0, which excludes this case. Indeed, for mu_0 = 0 the constant process mu_t = 0 satisfies Definition 1.1 with M^f_t = 0 and quadratic variation 0 for every f in C^2_b, so the theorem's conclusion is false as stated. Please add the explicit hypothesis mu_0 != 0 (or define M_b^+ to consist of nonzero measures) and adjust the abstract accordingly.
  2. [Section 2, proof of Theorem 1.2, Step 2] The normalization in the Lévy characterization step is incorrect. From [B]_t = c t, the process W_t := B_t / sqrt(c) is a standard Brownian motion, not W_t := B_t / c, since [B_t / c]_t = t / c. The contradiction argument is easily repaired: with the correct scaling, the event {B_T < -c ||H||_0 T} equals {W_T < -sqrt(c) ||H||_0 T}, which has positive probability for a standard Brownian motion. This is a local but load-bearing fix, since the current text invokes Lévy's characterization for a process that is not a Brownian motion unless c = 1.
minor comments (4)
  1. [Proof of Theorem 1.2, after equation (2.12)] The text says '(nabla f)(x) = g'_epsilon(x_1) = 1 on Sigma_f^c', but the function being differentiated is f(x) = g(x_1), so this should read g'(x_1), not g'_epsilon(x_1).
  2. [Lemma 2.2] The notation k in Z^pm is nonstandard and slightly confusing; please define explicitly that Z^+ = {1,2,...} and Z^- = {-1,-2,...}, or use separate indices.
  3. [Abstract and Theorem 1.2] The phrase 'any initial measure' should be qualified to refer to nonzero initial measures, as discussed in the first major comment.
  4. [Step 1 of the proof of Theorem 1.2] The statement that mu_s |nabla f|^2 = mu_s 1 is 'a.s. well-defined for a.e. s' follows from (2.11), but the wording is terse; a short explanatory sentence would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the nonexistence proof derives a contradiction from the martingale definition and standard stochastic calculus; self-citations only frame prior cases and sharpness.

full rationale

The central claim (Theorem 1.2) is not an input to the proof. The proof argues by contradiction: assuming a solution in the sense of Definition 1.1, it constructs a piecewise affine function g whose singular set avoids the atoms of the finite time-averaged marginal measure, regularizes it, uses Proposition 2.3 and the Cherny-Engelbert lemma to pass to the limit, applies Levy's characterization to obtain a Brownian motion, and derives a negative value of mu_T f from the boundedness of H. Every step uses only the hypothesis that a solution exists, standard martingale calculus, and the stated properties of H. The self-citations to [21,22] supply the definition of solution and the earlier alpha>0 rigidity results, but those earlier results are not used as premises for Theorem 1.2; the paper explicitly says their technique does not apply to alpha=0. The sharpness citations [9,10,24,25] are independent existence constructions for unbounded H and are not needed to derive nonexistence. No parameter is fitted to the conclusion, and no known result is renamed or smuggled in via citation. Section 3 honestly limits the rigorous argument to flat spaces and the sphere, which is a scope limitation rather than circular reasoning. The only notable caveat is non-circular: the proof sets c := mu_0(M_d) > 0, while the theorem states 'any initial condition mu_0 in M_b^+'; if the convention for 'positive finite Borel measures' includes the zero measure, the statement needs an explicit nonzero-mass hypothesis. That is a correctness/edge-case concern, not a circularity. Overall, the derivation is self-contained and the cited self-work is not load-bearing.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The theorem has no fitted constants. It imports the martingale solution concept and standard stochastic calculus; the only substantive assumptions are boundedness of H, finiteness of the initial total mass, and the countability of atoms used to build the watching function.

assumptions (5)
  • domain assumption Martingale solution definition of Definition 1.1 is the operative notion of solution.
    Imported from [22, Def. 1]; the theorem only concerns this class of measure-valued martingale solutions.
  • standard math Levy's characterization and the martingale convergence lemma [4, Lem. B.11].
    Used in Proposition 2.3 and Step 2 to pass to limits and identify Brownian motion; not proved in the paper.
  • standard math Finite Borel measures have at most countably many atoms and their complement is dense.
    Used in Lemma 2.2 to choose a piecewise affine function whose singular set avoids the atoms of mu*_T.
  • domain assumption The initial measure is deterministic and finite, so assumption (2.1) holds.
    Theorem states any mu_0 in M_b^+; the proof assumes E[mu_0 M_d] finite, trivially satisfied for deterministic mu_0.
  • domain assumption H is bounded Borel, so the sup norm controls the drift.
    Used in (2.10) and in the event E; this is the theorem's assumption, and its sharpness is shown by existing existence results for unbounded H.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ill-posedness of the pure-noise Dean-Kawasaki equation." pith.science (2026). https://pith.science/paper/LOCQMCZM

@misc{pith2026250109677,
  author       = {Pith},
  title        = {Pith review of: Ill-posedness of the pure-noise Dean-Kawasaki equation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LOCQMCZM}},
  note         = {Machine review of arXiv:2501.09677}
}
abstract

We prove that the Dean-Kawasaki-type stochastic partial differential equation $$\partial \rho= \nabla\cdot (\sqrt{\rho\,}\, \xi) + \nabla\cdot \left(\rho\, H(\rho)\right)$$ with vector-valued space-time white noise $\xi$, does not admit solutions for any initial measure and any vector-valued bounded measurable function $H$ on the space of measures. This applies in particular to the pure-noise Dean-Kawasaki equation ($H\equiv 0$). The result is sharp, in the sense that solutions are known to exist for some unbounded $H$.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 11 canonical work pages

  1. [1]

    Adams, N

    S. Adams, N. Dirr , M. A. Peletier , and J. Zimmer . From a Lar ge-Deviations Principle to the Wasserstein Gradient Flow: A New Micro-Macro Passage. Commun. Math. Phys. , 307(3):791–815, 2011. doi:10.1007/s00220-011-1328-4

    doi:10.1007/s00220-011-1328-4 2011

  2. [2]

    Benois, C

    O. Benois, C. Kipnis, and C. Landim. Large deviations fro m the hydrodynamical limit of mean zero asymmetric zero range processes. Stoch. Proc. Appl. , 55(1):65–89, January 1995. doi:10.1016/0304-4149(95)91543-a

    doi:10.1016/0304-4149(95)91543-a 1995

  3. [3]

    Bertini, A

    L. Bertini, A. De Sole, D. Gabrielli, G. Jona-Lasinio, an d C. Landim. Macroscopic fluctuation theory .Rev . Mod. Phys., 87(2):593–636, June 2015. doi:10.1103/revmodphys.87.593

    doi:10.1103/revmodphys.87.593 2015

  4. [4]

    A. S. Cherny and H.-J. Engelbert. Singular Stochastic Differential Equations , volume 1858 of Lecture Notes in Mathematics . Springer Berlin Heidelberg, 2005. doi:10.1007/b104187

    doi:10.1007/b104187 2005

  5. [5]

    Cornalba and J

    F . Cornalba and J. Fischer . Multilevel Monte Carlo metho ds for the Dean– Kawasaki equation from Fluctuating Hydrodynamics. arXiv:2311.08872, 2023. doi:10.48550/ARXIV.2311.08872

    arXiv doi:10.48550/arxiv.2311.08872 2023

  6. [6]

    Cornalba and J

    F . Cornalba and J. Fischer . The Dean–Kawasaki Equation a nd the Structure of Density Fluc- tuations in Systems of Diffusing Particles. Arch. Rational Mech. Anal. , 247(5), August 2023. doi:10.1007/s00205-023-01903-7. Page 7/9 Ill-posedness of the pure-noise Dean–Kawasaki equation

    doi:10.1007/s00205-023-01903-7 2023

  7. [7]

    Cornalba, J

    F . Cornalba, J. Fischer , J. Ingmanns, and C. Raithel. Den sity fluctuations in weakly in- teracting particle systems via the Dean–Kawasaki equation . arXiv:2303.00429, 2023. doi:10.48550/ARXIV.2303.00429

    arXiv doi:10.48550/arxiv.2303.00429 2023

  8. [8]

    D. S. Dean. Langevin equation for the density of a system o f interacting Langevin processes. J. Phys. A: Math. Gen. , 29:L613–L617, 1996

    work page 1996

Show all 29 references
  1. [9]

    Dello Schiavo

    L. Dello Schiavo. The Dirichlet–Ferguson Diffusion on t he Space of Probability Measures over a Closed Riemannian Manifold. Ann. Probab. , 50(2):591–648, 2022. doi:10.1214/21-AOP1541

    2022 doi

  2. [10]

    Dello Schiavo

    L. Dello Schiavo. Massive Particle Systems, Wasserste in Brownian Motions, and the Dean–Kawasaki Equation. arXiv:2411.14936, pages 1–103, 2024. doi:10.48550/arXiv.2411.14936

    2024 doi arXiv

  3. [11]

    Dirr , M

    N. Dirr , M. Stamatakis, and J. Zimmer . Entropic and grad ient flow formulations for nonlinear diffusion. J. Math. Phys. , 57(8), August 2016. doi:10.1063/1.4960748

    2016 doi

  4. [12]

    Djurdjevac, H

    A. Djurdjevac, H. Kremp, and N. Perkowski. Weak error an alysis for a nonlinear SPDE approximation of the Dean–Kawasaki equation. Stoch. PDE: Anal. Comp. , March 2024. doi:10.1007/s40072-024-00324-1

    2024 doi

  5. [13]

    Embacher , N

    P . Embacher , N. Dirr , J. Zimmer , and C. Reina. Computing diffusivities from par- ticle models out of equilibrium. Proc. R. Soc. A , 474(2212):20170694, April 2018. doi:10.1098/rspa.2017.0694

    2018

  6. [14]

    Fehrman and B

    B. Fehrman and B. Gess. Non-equilibrium large deviatio ns and parabolic- hyperbolic PDE with irregular drift. Invent. math. , 234(2):573–636, July 2023. doi:10.1007/s00222-023-01207-3

    2023 doi

  7. [15]

    Fehrman and B

    B. Fehrman and B. Gess. Conservative stochastic PDEs on the whole space. arXiv:2410.00254, 2024. doi:10.48550/ARXIV.2410.00254

    2024 doi arXiv

  8. [16]

    Fehrman and B

    B. Fehrman and B. Gess. Well-Posedness of the Dean–Kawa saki and the Nonlinear Dawson– Watanabe Equation with Correlated Noise. Arch. Rational Mech. Anal. , 248(2), March 2024. doi:10.1007/s00205-024-01963-3

    2024 doi

  9. [17]

    B. Gess, R. S. Gvalani, and V . Konarovskyi. Conservativ e SPDEs as fluctuating mean field limits of stochastic gradient descent. Probab. Theor . Relat. Fields (to appear) , 2022. doi:10.48550/ARXIV.2207.05705

    2022 doi arXiv

  10. [18]

    R. L. Jack and J. Zimmer . Geometrical interpretation of fluctuating hydrodynam- ics in diffusive systems. J. Phys. A: Math. Theor . , 47(48):485001, November 2014. doi:10.1088/1751-8113/47/48/485001

    2014 doi

  11. [19]

    Karatzas and S

    I. Karatzas and S. E. Shreve. Brownian Motion and Stochastic Calculus . Springer US, 1988. doi:10.1007/978-1-4684-0302-2

    1988 doi

  12. [20]

    Kawasaki

    K. Kawasaki. Stochastic model of slow dynamics in super cooled liquids and dense colloidal suspensions. Physica A: Statist. Mech. Appl. , 208(1):35–64, July 1994. doi:10.1016/0378-4371(94)90533-9

    1994 doi

  13. [21]

    Konarovskyi, T

    V . Konarovskyi, T . Lehmann, and M.-K. von Renesse. Dean-Kawasaki dynamics: ill-posedness vs. triviality. Electron. Commun. Probab. , 24, jan 2019. doi:10.1214/19-ecp208

    2019 doi

  14. [22]

    Konarovskyi, T

    V . Konarovskyi, T . Lehmann, and M.-K. von Renesse. On De an-Kawasaki Dy- namics with Smooth Drift Potential. J. Statist. Phys. , 178(3):666–681, nov 2019. doi:10.1007/s10955-019-02449-3

    2019 doi

  15. [23]

    Konarovskyi and F

    V . Konarovskyi and F . Müller . Dean–Kawasaki equation w ith initial condition in the space of positive distributions. J. Evol. Equ. , 24(4), October 2024. doi:10.1007/s00028-024-01018-w

    2024 doi

  16. [24]

    Konarovskyi and M.-K

    V . Konarovskyi and M.-K. von Renesse. Modified Massive A rratia flow and Wasserstein diffu- sion. Comm. Pure Appl. Math. , 72(4):764–800, 2019. doi:10.1002/cpa.21758

    2019 doi

  17. [25]

    Konarovskyi and M.-K

    V . Konarovskyi and M.-K. von Renesse. Reversible coale scing-fragmentating Wasser- stein dynamics on the real line. J. Funct. Anal. , 286(8):110342, April 2024. doi:10.1016/j.jfa.2024.110342. Page 8/9 Ill-posedness of the pure-noise Dean–Kawasaki equation

    2024

  18. [26]

    Müller , M.-K

    F . Müller , M.-K. von Renesse, and J. Zimmer . Well-Posed ness for Dean– Kawasaki Models of Vlasov-Fokker-Planck Type. arXiv:2411.14334, nov 2024. doi:10.48550/arXiv.2411.14334

    2024 doi arXiv

  19. [27]

    Quastel, F

    J. Quastel, F . Rezakhanlou, and S. R. S. Varadhan. Large deviations for the symmetric simple exclusion process in dimensions d ≥ 3. Probab. Theory Relat. Fields , 113(1):1–84, February

  20. [28]

    te Vrugt, H

    M. te Vrugt, H. Löwen, and R. Wittkowski. Classical dyna mical density functional theory: from fundamentals to applications. Adv . Phys. , 69(2):121–247, April 2020. doi:10.1080/00018732.2020.1854965. Page 9/9

    2020

  21. [1999]

    doi:10.1007/s004400050202

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.