{"id":"7c6cd084-4aca-4116-b404-e4ac558ecd28","arxiv_id":"2602.10841","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"McKean-Vlasov SDEs with distributional local negative-Sobolev kernels have global well-posedness from smoothed initial laws and entropy-cost estimates.","lead":"This paper proves when a stochastic particle equation whose drift depends on the whole crowd's distribution has a unique solution, even when the interaction is a very rough object rather than a smooth function. It matters for mean-field models behind Burgers-, Navier-Stokes-, and p-Laplacian-type equations, where such singular interactions appear.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.1 assumes only (2.1), but its proof via Lemmas 4.1–4.2 requires the stronger conditions η<κ+1 or η<κ+3/2; for κ>1/2 the theorem overclaims.","rationale":"The reader identified Assumption (A) as the weakest premise, but A is not the soft spot I found. For convolution kernels, the pointwise bound in (A) follows from the translation-invariant local negative Sobolev norm, so A is natural. The actual load-bearing issue is an internal inconsistency between the theorem's stated condition and the proofs of the lemmas used to prove it. Theorem 2.1(1) claims global well-posedness for arbitrary initial laws whenever η<1+2κ. The proof of Proposition 4.3, however, uses Lemma 4.1(3), which requires η<κ+1, and Lemma 4.2, which requires η<κ+3/2. These extra conditions are not consequences of (2.1) when κ>1/2. The divergence of the key convolution integral at s=0 is not a minor technicality: it means the invariant subset ^C is not shown to be mapped into itself, so the contraction fixed-point argument has no domain. This is especially consequential because Example 2.6 leverages the advertised Theorem 2.1(1) to claim global solvability for δ up to 1+2κ, which includes parameter values excluded by the actual hypotheses of the proof lemmas. I am not asserting the theorem is false; it may be salvageable by a more delicate argument with a different time weight or a revised fixed-point space. But as written, the central claim is not supported in the stated generality. Therefore the reader's ACCEPT verdict should be adjusted to CONDITIONAL: the paper's main theorem and examples need either a corrected condition or a completed proof in the missing parameter region.","tokens_in":31509,"tokens_out":24060,"duration_ms":205859,"concrete_test":"Analytically verify the gap: fix κ=1, ε=0, p=∞, η=2.8. Check whether the hypotheses of Lemma 4.1(3) and Lemma 4.2 are satisfied under (2.1); they are not, since η>κ+1 and η>κ+3/2. Then examine the key estimate in Lemma 4.1(3): the integral ∫_0^t s^{κ-η}(t-s)^{-1/2}ds diverges for η=2.8, so the claimed Φ:^C→^C invariant-set argument collapses. If the theorem is to stand, the authors must either supply a proof of the invariant-set/contraction estimates under only (2.1), or restrict Theorem 2.1 and Example 2.6 to η<κ+3/2 (and η<κ+1 for the global arbitrary-initial claim).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The global well-posedness claim in Theorem 2.1(1) for ε=0, p=∞ is stated under (2.1), i.e. η=δ+d/k<1+2κ. But the proof of Proposition 4.3 passes through Lemma 4.1 and Lemma 4.2, and those lemmas contain extra hypotheses that are not implied by (2.1). Lemma 4.1(3) explicitly requires η<κ+1, and Lemma 4.1(1)/Lemma 4.2 require η<κ+3/2. For κ>1/2 these are strictly stronger than (2.1). For example, when κ=1, (2.1) permits η<3, while Lemma 4.1(3) requires η<2 and Lemma 4.2 requires η<5/2. The gap is concrete: in Lemma 4.1(3), the invariant-set estimate for Φ controls terms of the form ∫_0^t s^{κ-η}(t-s)^{-1/2}ds, which diverges at s=0 for η≥1+κ. Lemma 4.2's contraction estimate similarly needs rη<1 and (2-r)η<2+2κ with r=2/(3+2κ), which forces η<κ+3/2. Since Theorem 2.1 and Example 2.6 explicitly claim results in the region η∈[κ+1,1+2κ) (e.g. δ=2.8, κ=1, n=2, d=1), the central theorem is not established in exactly the parameter range advertised.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":31910,"tokens_out":13730,"duration_ms":130351,"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":[{"comment":"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.","section":"§4, Lemma 4.1 and Proposition 4.3"},{"comment":"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.","section":"Example 2.6"},{"comment":"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.","section":"Theorem 2.3"}],"minor_comments":[{"comment":"Typo: 'first assrtion' should be 'first assertion'.","section":"§4, proof of Proposition 4.3"},{"comment":"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.","section":"Equation (2.18)"},{"comment":"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).","section":"Lemma 3.4(2)"},{"comment":"The symbol c_2(t) is used for two different constants in the same proof; please rename one of them to avoid ambiguity.","section":"Proof of Theorem 2.3(2)"}],"recommendation":"major_revision","confidential_remarks":"The reader's ACCEPT appears too optimistic in light of the concrete gap identified in §4. The paper's central framework is coherent, but Theorem 2.1 as stated is not proved in the full parameter range, and this propagates to the advertised examples. A major revision is appropriate: either supply the missing estimates for the range η∈[κ+1,1+2κ) or restrict the statements to the hypotheses actually used. The latter would still preserve the abstract's main global result δ+d/k<1 and the local existence for arbitrary singular indices, so the paper is not beyond repair."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","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.\n\nThe 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.\n\nThe 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.\n\nThe 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.\n\nProportionally: 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.\n\nI'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.","headline":"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.","tokens_in":32422,"tokens_out":7207,"would_cite":false,"duration_ms":60237,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60H50"],"pacs":[],"model":"deepseek-v4-flash","headline":"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.","keywords":["McKean-Vlasov SDE","local distributional interaction","negative Sobolev space","well-posedness","entropy-cost inequality","Nemytskii-type SDE","singular kernel","Wasserstein distance"],"falsifier":"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).","tokens_in":31398,"feed_emoji":"🎲","tokens_out":7135,"duration_ms":67817,"temperature":0.7,"pith_summary":"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.","feed_headline":"Heat-kernel smoothing tames singular McKean–Vlasov SDEs","feed_subtitle":"Unique global solutions and entropy-cost bounds hold even for kernels more singular than Riesz potentials.","key_machinery":"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","core_discovery":"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","pith_inferences":["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."],"forward_implications":["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."],"fun_headline_variants":["Heat-kernel smoothing yields global solutions for singular McKean-Vlasov SDEs","Taming singular McKean-Vlasov SDEs via heat-kernel smoothing","Global well-posedness for singular McKean-Vlasov SDEs with entropy-cost bounds","Entropy-cost estimates for McKean-Vlasov SDEs with singular kernels","Heat-kernel tames singular McKean-Vlasov SDEs: global well-posedness"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Heat-kernel smoothing yields global solutions for singular McKean-Vlasov SDEs","Taming singular McKean-Vlasov SDEs via heat-kernel smoothing","Global well-posedness for singular McKean-Vlasov SDEs with entropy-cost bounds","Entropy-cost estimates for McKean-Vlasov SDEs with singular kernels","Heat-kernel tames singular McKean-Vlasov SDEs: global well-posedness"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000788,"raw_usage":{"total_tokens":3318,"prompt_tokens":755,"completion_tokens":2563,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":2458}},"tokens_in":499,"tokens_out":2563,"duration_ms":16651,"temperature":1.0,"reasoning_tokens":2458,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T00:57:40.906140+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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).","supporting_citations":[],"review_version":1}