McKean-Vlasov SDEs with Local Distributional Interactions: Well-Posedness and Entropy-Cost Estimates

Feng-Yu Wang; Panpan Ren; Xing Huang

REVIEW 3 major objections 4 minor 21 references

McKean-Vlasov SDEs with Local Distributional Interactions: Well-Posedness and Entropy-Cost Estimates

T0 review · 3 major / 4 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read McKean–Vlasov SDEs whose drift is a singular distributional kernel remain uniquely solvable for arbitrary singularity indices, with explicit entropy-cost estimates for their time-marginals.

desk verdict Theorem 2.1 states well-posedness for η<1+2κ, but the proof only works for η<κ+1 (or η<κ+3/2); that gap needs closing before I'd trust the advertised range. read the letter →

arxiv 2602.10841 v3 pith:IJ64TWPK submitted 2026-02-11 math.PR

Xing Huang , Panpan Ren , Feng-Yu Wang This is my paper · ORCID

classification math.PR MSC 60H1060H50

keywords McKean-VlasovSDElocaldistributionalinteractionnegativeSobolevspacewell-posednessentropy-costinequalityNemytskii-typesingularkernelWassersteindistance

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

This paper proves that McKean–Vlasov SDEs with interaction kernels lying in a local negative Sobolev space—kernels that may be true distributions, not functions—admit unique global solutions even when the kernel is more singular than Riesz or Coulomb potentials. The key is a heat-semigroup smoothing estimate that controls the negative-Sobolev norm of the time-marginal distribution and makes the singularity harmless near time zero, enabling a contraction fixed-point argument on a path space of measures. The paper also establishes quantitative regularity of the solution map: both the negative-Sobolev distance and the relative entropy of two time-marginals are bounded by Wasserstein distances of the initial laws, with explicit time decay. These results apply to Nemytskii-type SDEs depending on higher derivatives of the density and to kernels with arbitrary singular order, where prior well-posedness results were unavailable.

What carries the argument

The central object is the local negative Sobolev space W̃^{−δ,k}, defined as the closure of bounded measurable functions under the norm sup_{z∈R^d} ||1_{B(z,1)}(1−Δ)^{−δ/2} f||_{L^k}. The load-bearing estimate is the heat-semigroup smoothing bound ||∇^i P^0_t||_{W̃^{−δ,k} → W̃^{−ε,p}} ≤ B t^{−(i+δ−ε)/2 − d(p−k)/(2pk)}, which converts Brownian regularization into a time-decay factor that cancels the kernel's singularity near t=0. This defines a weighted path space C^T_{ε,p;δ,k} of measure-valued paths on which the drift map is a contraction; the fixed point is the unique solution. A time-shift argument extends the contraction to heat-kernel-convolved initial laws, and a bi-coupling argument p

What would settle it

For the density-derivative example, compute ||∇^{n−1} δ_0||_{W̃^{−δ,∞}}: the paper's own condition requires δ > d+n−1. If this norm turned out finite for δ ≤ d+n−1 and yielded a counterexample to global well-posedness, the scope claim would fail. More directly, check numerically whether sup_{γ∈P̂_r, s≤t} s^{δ/2+d/(2k)} ||P*_s γ||_{δ,k*} stays finite for a Dirac-type initial law at the boundary δ + d/k = 1 + 2κ; any blow-up would violate the paper's bound (2.5).

Watch

Extended reading notes

Core claim

The central claim is that, under a Lipschitz condition on the drift in the local negative-Sobolev norm with a time weight t^κ, the SDE has a unique maximal weak and strong solution for any singular indices (δ,k) when the initial law lies in the appropriate dual space. If the initial law is a convolution of any probability measure with a heat kernel at positive time, the solution is global for every δ and k, with a uniform bound on the negative-Sobolev norm of the time-marginals over the whole class of such initial laws. For arbitrary initial laws, global well-posedness holds whenever δ + d/k < 1. The companion regularity theorem gives explicit inequalities of the form ||P*_t γ − P*_t γ̃||_{δ

Load-bearing premise

Assumption (A), which requires the drift to be bounded and Lipschitz in the measure variable under the local negative-Sobolev norm with a time factor t^κ; if that Lipschitz condition or the finiteness of the duality pairing does not hold for the relevant laws, the contraction fixed-point argument and every bound in the paper collapse.

Editorial extensions

If this is right

For any initial law that is a heat-kernel convolution, the SDE has a unique global weak and strong solution for every singularity index (δ,k), covering kernels with pointwise growth like c z/|z|^{d+2n_0+ε_0}, which are more singular than Riesz kernels.
For completely arbitrary initial laws, global well-posedness holds whenever δ + d/k < 1; outside this range, unique maximal solutions exist with explicit life-time lower bounds depending on the initial law's norm.
The solution map γ ↦ P*_t γ is locally Lipschitz with respect to a Wasserstein distance: the negative-Sobolev distance between two time-marginals decays with a fixed power of t and is controlled by W_q(γ,γ̃), so small changes in the initial law propagate at a controlled rate.
In the basic ε=0, p=∞ case, the relative entropy satisfies Ent(P*_t γ|P*_t γ̃) ≤ (β_t/t) W_2(γ,γ̃)^2, giving a log-Harnack-type estimate uniform over all initial distributions.
Nemytskii-type density-derivative SDEs, where the drift depends on derivatives of the density up to order n−1, are globally well-posed for δ > d+n−1, with the same entropy-cost estimate for heat-kernel-convolved initial laws.

Reading between the lines

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

A natural next step is to probe numerically whether δ + d/k = 1 + 2κ marks a true phase transition: the paper's estimates degenerate exactly there, and the condition (2.1) suggests the time-marginal norm may blow up for initial laws not in the heat-convolution class.
The paper notes that propagation of chaos remains open for local distributional kernels with δ > 0 and k ≥ 1; the quantitative entropy-cost estimates proved here are a plausible ingredient for such a propagation-of-chaos argument, though the paper does not take that step.
The time-shift argument that secures global well-posedness for heat-kernel-convolved initial laws might extend to other singular initial classes if an analogue of the heat-semigroup smoothing estimate exists under fractional Brownian or stable noise; this is a testable extension the paper leaves implicit.

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.

Desk Editor's Note

read the letter

Here's the short version: the paper has a real new idea, but its main theorem is not proved in the parameter range it claims. For κ=0 the headline result works; for κ>1/2 the proof and statement diverge.

The genuine novelty is treating interactions in local negative Sobolev spaces W̃^{-δ,k} rather than L^k or Riesz/Besov classes, and deriving entropy-cost estimates for distributional kernels. The time-shift trick for initial laws in P̂_0 is clever. The estimates in Theorem 2.3 are carefully derived once the semigroup bounds are accepted.

The problem is in the well-posedness proof. Theorem 2.1(1) assumes only (2.1), η<1+2κ. But Lemma 4.1, which does the heavy lifting, is stated with an extra hypothesis η<κ+3/2, and part (3) of that lemma requires η<κ+1. Proposition 4.3 applies these lemmas without checking the extra conditions. For κ>1/2, these are strictly stronger than (2.1). The stress-test example is on point: κ=1, η=2.8 satisfies (2.1) (η<3) but violates both η<2.5 and η<2. Example 2.6 with n=2,d=1,δ=2.8 falls right in that gap. So the paper proves global well-posedness for a narrower range than it advertises.

The rest of the paper is solid in the sense that no circularity or data fitting appears. There is heavy reliance on prior papers by the same group ([10], [14], [20]) for key propositions; that is not disqualifying, but it means a referee must be willing to check those imports too.

Proportionally: this is a major, but localized, flaw. The abstract's own headline case δ+d/k<1 is the κ=0 case, where the lemma conditions are satisfied, so the paper's most prominent claim probably survives. The overreach is in the more general time-weighted drift statement.

I'd send this to a serious referee, but not with an acceptance recommendation as is. The gap is concrete; either the theorem needs restriction or the lemmas need strengthening. If the authors can close it, this will be a useful paper. As it stands, I would not cite the theorem in its advertised form.

Referee Report

3 major / 4 minor

Summary. The paper studies McKean-Vlasov SDEs on R^d whose drift is given by a kernel h_t in a local negative Sobolev space \tilde W^{-\delta,k}. It introduces a notion of C^{\varepsilon,p;\delta,k}-solution and proves, under a Lipschitz condition (A) with respect to the dual negative-Sobolev norm and a condition (2.1) on \eta=\delta-\varepsilon+d(p-k)/(pk), the existence and uniqueness of maximal and global solutions (Theorem 2.1), together with Wasserstein and relative-entropy estimates for the time-marginal laws (Theorem 2.3). Applications are given to kernels more singular than Riesz kernels and to Nemytskii-type SDEs depending on density derivatives.

Significance. If the results were established in full, the paper would provide a substantial extension of the existing theory of McKean-Vlasov SDEs with distributional interactions, going beyond Besov-space and Riesz-kernel settings. The fixed-point framework, the heat-semigroup estimates in Lemmas 3.1 and 3.4, and the entropy-cost estimates are valuable and carefully organized. The claimed applications to arbitrary singular indices and to density-derivative SDEs are attractive. However, a concrete parameter-range gap in the proof of Theorem 2.1 means that some of the stated results, including parts of Example 2.6, are not currently justified.

major comments (3)

[§4, Lemma 4.1 and Proposition 4.3] Theorem 2.1 is stated under only (2.1), but its proof via Proposition 4.3 uses Lemma 4.1, which explicitly assumes η<κ+3/2, and Lemma 4.1(3) even assumes η<κ+1. Lemma 4.2's proof also uses (4.5), which is exactly η<κ+3/2. These hypotheses are strictly stronger than (2.1) when κ>1/2. For example, with κ=1, ε=0, p=∞ and δ=2.5, (2.1) holds because 2.5<3, but Lemma 4.1(3) requires η<2 and Lemma 4.1(1)/Lemma 4.2 require η<2.5. The divergence of the integral ∫_0^t s^{κ−η}(t−s)^{-1/2} ds at s=0 when η≥1+κ explains why the stronger condition is not merely cosmetic. Consequently the maximal/global well-posedness claim is not established in the full parameter range stated in Theorem 2.1, and the range δ∈(d+n−1,1+2κ) in Example 2.6 is not justified when δ≥κ+1.
[Example 2.6] The Nemytskii drift b_t(x, ℓ_X^{<n}(x)) is only defined for measures with a density, but condition (A) is imposed on all μ,ν∈P^{δ,k*}, and the path space C^T_{ε,p;δ,k} in Definition 1.1 contains arbitrary weakly continuous probability paths without a density requirement. The verification of (A) in Example 2.6 computes ||∇^iδ_0||_{\tilde W^{-δ,∞}} but does not show that the drift is defined, bounded, or Lipschitz on the full domain P^{δ,k*}. The proof needs either a restricted state space or a uniform density argument for the marginals appearing in the fixed-point construction. Without this, the density-derivative application is not justified by the abstract theorem as written.
[Theorem 2.3] The regularity estimates in Theorem 2.3 are derived for P_t^*γ once well-posedness is available from Theorem 2.1. Since the proof of Theorem 2.1 has the parameter gap described above, the entropy-cost and Wasserstein estimates inherit the same gap in the range η∈[κ+1,1+2κ) for ε=0, p=∞. The authors should either close that range with additional estimates or state Theorem 2.1 and Theorem 2.3 under the stronger hypotheses actually used, e.g. η<κ+1 or η<κ+3/2 as needed. The abstract's headline global result δ+d/k<1 and the local well-posedness for arbitrary singular indices would survive such a restriction, but the broader claims would not.

minor comments (4)

[§4, proof of Proposition 4.3] Typo: 'first assrtion' should be 'first assertion'.
[Equation (2.18)] The subordination identity (2.18) is introduced inside Example 2.5 but is used earlier in Lemma 3.1 and Lemma 3.2. It should be stated as a standalone preparation in Section 3, with a clear reference or proof and a precise description of the domain of the identity.
[Lemma 3.4(2)] The condition 'ξ<1∨(2−i−(η−2κ)+)' is typographically hard to parse. Please add explicit parentheses, e.g. ξ<1∨(2−i−(η−2κ)_+), and similarly in part (1).
[Proof of Theorem 2.3(2)] The symbol c_2(t) is used for two different constants in the same proof; please rename one of them to avoid ambiguity.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular reduction identified; the main estimates are derived from Assumption (A), and the flagged parameter-range gap is a correctness issue, not circularity.

full rationale

Walked the derivation chain. The paper's central object is Assumption (A), a Lipschitz condition on the drift in the negative-Sobolev norm ||·||_{δ,k*}. From (A) and the heat-semigroup estimate (1.3), Lemma 3.4 proves semigroup estimates, Lemmas 4.1–4.2 give invariance and contraction for the fixed-point map Φ, and Proposition 4.3/Theorem 2.1 assemble the fixed point. Theorem 2.3 estimates ||P*_t γ − P*_t γ̃||_{δ,k*} and relative entropy via Duhamel, log-Harnack, and bi-coupling; the target quantities are estimated, not assumed. No step redefines a fitted parameter or predicts a quantity that is an input by construction. The proof does import technical tools from the same authors' earlier works ([10, Prop. 5.1, 5.4, 5.5, Lemma 5.3, Thm 2.1/2.3], [14, Lemma 2.1], [20, Thm 1.3.1]); these are prior results on related but less singular McKean–Vlasov and heat-semigroup problems, and they are used as lemmas rather than as a uniqueness theorem forbidding alternatives. The manuscript explicitly contrasts its setting with [6], [2], [11] and claims newness for the local distributional space \tilde W^{-δ,k}; no renaming of a known result is apparent. One flagged issue: the proof of Theorem 2.1 via Lemmas 4.1–4.2 requires extra hypotheses η<κ+1 or η<κ+3/2 that are not stated in (2.1), so the theorem appears to overclaim in the range κ>1/2, η∈[κ+1,1+2κ). This is a correctness/parameter-range gap, not circularity: the missing hypotheses are stronger conditions, not restatements of the conclusion. No circular step was found.

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

The central claim carries no fitted data; the cost is measured by auxiliary Sobolev and Wasserstein norms. The proof leans on the authors' prior Riesz-kernel results [10] and bi-coupling [14], so independence of those internal citations should be checked, but no circular fit of a target quantity is present.

assumptions (5)

domain assumption Assumption (A): |bt(x,nu)| <= K_t t^kappa ||nu||_{delta,k*} and |bt(x,mu)-bt(x,nu)| <= K_t t^kappa ||mu-nu||_{delta,k*}
Main structural input used throughout Theorems 2.1 and 2.3, Section 2.1. Verified for examples, not derived.
domain assumption Operator-norm estimates for the heat semigroup and perturbed semigroups from [10, Props. 5.1, 5.2, 5.4, 5.5] and [10, Lemma 5.3]
Black-box results from the authors' previous paper; the new proofs reduce to these estimates, especially in Lemma 3.4 and Section 5.
domain assumption Bi-coupling inequality [14, Lemma 2.1]
Used in the proof of Theorem 2.3(2) to split relative entropy into two terms; cited without reproof.
standard math Sobolev embedding with local norms: ||.||_{W-tilde^{-epsilon,p}} <= c0 ||.||_{L-tilde^{p0}}, 1/p0 = 1/p + epsilon/d
Used in Section 5, around Eq. (5.4), to connect negative Sobolev norms to local L^{p0} norms.
standard math Subordination formula (1-Delta)^{-r} = 1/Gamma(r) integral s^{r-1} e^{-s} P_s^0 ds, asserted on union of local Sobolev spaces
Used to prove Lemma 3.1 and to verify Example 2.5, near Eq. (2.18).

how reviews work

0 comments

Cite this review

Pith. "Pith review of McKean-Vlasov SDEs with Local Distributional Interactions: Well-Posedness and Entropy-Cost Estimates." pith.science (2026). https://pith.science/paper/IJ64TWPK

@misc{pith2026260210841,
  author       = {Pith},
  title        = {Pith review of: McKean-Vlasov SDEs with Local Distributional Interactions: Well-Posedness and Entropy-Cost Estimates},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJ64TWPK}},
  note         = {Machine review of arXiv:2602.10841}
}

abstract

We study McKean-Vlasov SDEs with interaction kernels in $\tt W^{-\dd,k},$ the local negative Sobolev space on $\R^d$ with indexes $\dd \in [0,\infty)$ and $k\in [1,\infty].$ We derive the local well-posedness for any singular indexes $(\dd,k)\in [0,\infty)\times [1,\infty],$ and prove the global well-posedness for any initial distributions provided $\dd+\ff d k<1$. Moreover, the relative entropy and the $\|\cdot\|_{\dd,k*}$-distance induced by $ \tt W^{-\dd,k}$ are estimated for the time-marginal distributions of solutions by using the Wasserstein distance of initial distributions, which describe the regularity of the solution in initial distribution. In particular, the main results apply to Nemytskii-type SDEs which depend on higher order derivatives of the density functions, as well as McKean-Vlasov SDEs with interactions more singular than Riesz kernels.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 1 canonical work pages

[10]

Huang, P

X. Huang, P. Ren, F.-Y. Wang, Entropy-cost inequality for McKean- Vlasov SDEs with singular interactions,arXiv:2505.19787. 37

arXiv
[14]

Ren, F.-Y

P. Ren, F.-Y. Wang, Bi-coupling method and applications,Probab. Theory Relat. Fields(2025). https://doi.org/10.1007/s00440-025- 01394-5

work page doi:10.1007/s00440-025- 2025
[20]

F.-Y. Wang, P. Ren, Distribution Dependent Stochastic Differential Equations,World Scientific, 2025, Singapore

2025
[1]

Barbu, M

V. Barbu, M. Rehmeier, M. R¨ ockner,p-Brownian motion and the p-Laplacian,arXiv:2409.18744

arXiv
[2]

Chaudru de Raynal, J.-F

P.-E. Chaudru de Raynal, J.-F. Jabir, S. Menozzi, Multidimensional stable driven McKean-Vlasov SDEs with distributional interaction kernel: a regularization by noise perspective,Stoch. Partial Differ. Equ. Anal. Comput.13(2025), 367-420

2025
[3]

Gr¨ afner, Energy solutions to SDEs with supercritical distributional drift: An extension and weak convergence rates,arXiv:2407.09222

L. Gr¨ afner, Energy solutions to SDEs with supercritical distributional drift: An extension and weak convergence rates,arXiv:2407.09222

arXiv
[4]

Gr¨ afner, N

L. Gr¨ afner, N. Perkowski, Weak well-posedness of energy so- lutions to singular SDEs with supercritical distributional drift, arXiv:2407.09046

arXiv
[5]

Guillin, P

A. Guillin, P. Le Bris, P. Monmarch´ e, On systems of particles in singular repulsive interaction in dimension one: log and Riesz gas,J. ´Ec. polytech. Math.10(2023), 867-916

2023

Show all 21 references

[6]

Z. Hao, M. R¨ ockner, X. Zhang, Second order fractional mean-field SDEs with singular kernels and measure initial data,Ann. Probab. 54(2026), 1-62

2026
[7]

Z. Hao, X. Zhang, SDEs with supercritical distributional drifts, Comm. Math. Phys.406(2025), Paper No. 250, 56 pp

2025
[8]

Huang, F.-Y

X. Huang, F.-Y. Wang, Log-Harnack inequality and Bismut formula for McKean-Vlasov SDEs with singularities in all Variables,Math. Ann.393(2025), 241-269

2025
[9]

Huang, P

X. Huang, P. Ren, F.-Y. Wang, Probability distance estimates between diffusion processes and applications to singular McKean- Vlasov SDEs,J. Differential Equations420(2025) 376-399

2025
[11]

Issoglio, S

E. Issoglio, S. Pagliarani, F. Russo, D. Trevisani, Degenerate McKean-Vlasov equations with drift in anisotropic negative Besov spaces,arXiv:2401.09165
[12]

Jabin, Z

P.-E. Jabin, Z. Wang, Quantitative estimates of propagation of chaos for stochastic systems withW −1,∞ kernels,Invent. Math.214 (2018), 523-591

2018
[13]

Lewin, Coulomb and Riesz gases: the known and the unknown, J

M. Lewin, Coulomb and Riesz gases: the known and the unknown, J. Math. Phys.63(2022), No. 061101

2022
[15]

Rosenzweig, S

M. Rosenzweig, S. Serfaty, Global-in-time mean-field convergence for singular Riesz type diffusive flows,Ann. Appl. Probab.33(2023), 954- 998

2023
[16]

Serfaty, Lectures on Coulomb and Riesz Gases,arXiv:2407.21194

S. Serfaty, Lectures on Coulomb and Riesz Gases,arXiv:2407.21194

arXiv
[17]

A. S. Sznitman, Topics in propagation of chaos,Lecture Notes in Math.,1464 Springer-Verlag, Berlin, 1991, 165-251

1991
[18]

Wang, Harnack inequalities on manifolds with boundary and applications,J

F.-Y. Wang, Harnack inequalities on manifolds with boundary and applications,J. Math. Pures Appl.94(2010), 304-321

2010
[19]

Wang, Harnack Inequality and Applications for Stochastic Par- tial Differential Equations,Springer, New York,2013

F.-Y. Wang, Harnack Inequality and Applications for Stochastic Par- tial Differential Equations,Springer, New York,2013

2013
[21]

P. Xia, L. Xie, X. Zhang, G. Zhao,L q(Lp)-theory of stochastic dif- ferential equations,Stochastic Process. Appl.130(2020), 5188-5211. 38

2020

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [10]

Huang, P

X. Huang, P. Ren, F.-Y. Wang, Entropy-cost inequality for McKean- Vlasov SDEs with singular interactions,arXiv:2505.19787. 37

arXiv

[2] [14]

Ren, F.-Y

P. Ren, F.-Y. Wang, Bi-coupling method and applications,Probab. Theory Relat. Fields(2025). https://doi.org/10.1007/s00440-025- 01394-5

work page doi:10.1007/s00440-025- 2025

[3] [20]

F.-Y. Wang, P. Ren, Distribution Dependent Stochastic Differential Equations,World Scientific, 2025, Singapore

2025

[4] [1]

Barbu, M

V. Barbu, M. Rehmeier, M. R¨ ockner,p-Brownian motion and the p-Laplacian,arXiv:2409.18744

arXiv

[5] [2]

Chaudru de Raynal, J.-F

P.-E. Chaudru de Raynal, J.-F. Jabir, S. Menozzi, Multidimensional stable driven McKean-Vlasov SDEs with distributional interaction kernel: a regularization by noise perspective,Stoch. Partial Differ. Equ. Anal. Comput.13(2025), 367-420

2025

[6] [3]

Gr¨ afner, Energy solutions to SDEs with supercritical distributional drift: An extension and weak convergence rates,arXiv:2407.09222

L. Gr¨ afner, Energy solutions to SDEs with supercritical distributional drift: An extension and weak convergence rates,arXiv:2407.09222

arXiv

[7] [4]

Gr¨ afner, N

L. Gr¨ afner, N. Perkowski, Weak well-posedness of energy so- lutions to singular SDEs with supercritical distributional drift, arXiv:2407.09046

arXiv

[8] [5]

Guillin, P

A. Guillin, P. Le Bris, P. Monmarch´ e, On systems of particles in singular repulsive interaction in dimension one: log and Riesz gas,J. ´Ec. polytech. Math.10(2023), 867-916

2023