{"id":"8303c11c-a9a9-4e17-a04a-fd9185ede0ee","arxiv_id":"2508.12234","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Kinetic SDEs dX=V dt, dV=b dt+sqrt(2)dW with distributional drift b of anisotropic Holder order alpha in (-1,0) and bounded velocity divergence admit unique weak solutions, with Krylov and moment estimates.","lead":"This paper proves that a large class of kinetic stochastic differential equations with rough, distribution-valued drift coefficients still have unique weak solutions, as long as the rough part of the drift satisfies a subcritical scaling condition and its velocity divergence is controlled. It extends previous well-posedness results to a wider range of drift regularity and provides explicit Gaussian random field examples.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Remark 4.1 reduces weighted drifts to (Hsub) with regularity α_b−κ, not α_b; Theorem 1.4's Krylov range α∈(−1,α_b] is unsupported for κ>0.","rationale":"The reader's weakest_assumption correctly highlights the div_v b regularity condition as a key fragility of the argument; this condition is indeed essential in Lemma 2.8, Lemma 3.3, and Lemma 4.10. However, the most load-bearing issue I find is different: the reduction from the weighted hypothesis of Theorem 1.4 to the unweighted assumption (Hsub), via Remark 4.1, appears to lower the available Hölder regularity from α_b to α_b−κ. Since Theorem 4.19's Krylov estimate is stated in terms of the same α_b appearing in (Hsub), this reduction cannot deliver the full range α∈(−1,α_b] claimed in Theorem 1.4. If κ=0 the issue disappears; for κ>0, and especially in the Gaussian example where κ>3d/p, the claimed range is not justified by the provided proof. Because the concern is about a parameter range rather than a complete breakdown of the method, I would not reject the paper outright; instead, the authors should either prove that the decomposition preserves unweighted regularity α_b, or restrict the statement of Theorem 1.4 to α≤α_b−κ (or to κ below the threshold where (Hsub_w) applies). This is a concrete, checkable point rather than a vague worry about technical detail, and it changes the central claim's quantitative content, so the reader's ACCEPT verdict should be made conditional on resolving it.","tokens_in":30541,"tokens_out":23756,"duration_ms":245546,"concrete_test":"For a model weighted drift b as in Theorem 1.4 with κ>0, compute the unweighted C^{α_b}_a norm of the rough component b1 produced by the decomposition of Remark 4.1. If ‖b1‖_{C^{α_b}_a} is infinite but ‖b1‖_{C^{α_b−κ}_a} is finite, then Theorem 4.19 can only be applied with exponent α_b−κ, and the Krylov range in Theorem 1.4 must be adjusted accordingly. Alternatively, inspect the proof of Lemma 4.5(i) under (Hsub) with b1 of regularity α_b−κ and check whether any estimate recovers α_b; if not, the claimed α≤α_b is unsupported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 1.4 reduces the weighted drift assumption b∈L^{q_b}_T C^{α_b}_a(ρ_κ) to the unweighted assumption (Hsub) by invoking Remark 4.1, which states that b decomposes as b=b1+b2 with b1∈C^{α_b−κ}_a unweighted and b2∈C^{α_b+1−κ}_a(ρ_1) satisfying linear growth. But (Hsub), as used in Theorem 4.19, requires the singular component to lie in L^{q_b}_T C^{α_b}_a, with the same α_b appearing in the Krylov estimate. If κ>0, then α_b−κ<α_b, so the decomposition only verifies (Hsub) with the reduced exponent α_b−κ. Consequently Theorem 4.19 delivers the Krylov estimate (4.17) only for α∈(−1,α_b−κ], not for the claimed range α∈(−1,α_b]. This is not a cosmetic issue: in the Gaussian example, κ>3d/p is typically positive, and the range κ∈((1+α_b−2/q_b)/(3+α_b−2/q_b),1+α_b) is precisely the regime where the weighted assumption (Hsub_w) is not available, so the proof of the full α_b range relies entirely on the deficient reduction. Additionally, the divergence condition in (Hsub) may also be affected: it requires div_v b1∈L^{q_b}_T C^{α_b}_a, but if div_v b2 contributes unbounded unweighted terms, even this condition is not automatic from the stated hypotheses. The central claim as stated therefore appears to overreach its proof for κ>0, unless the decomposition of Remark 4.1 can be sharpened to preserve the original regularity α_b in the unweighted rough component.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies weak well-posedness of the kinetic SDE (1.3) with a distributional drift b in weighted anisotropic Hölder spaces L^{q_b}_T C^{α_b}_a(ρ_κ), where α_b∈(-1,0), κ∈[0,1+α_b), and with a bounded velocity divergence. Weak solutions are defined via mollified drifts and an L²-limit of the drift integrals (Definition 1.1). The authors prove Schauder-type estimates for the associated kinetic PDE using paraproducts (Theorem 3.6), derive uniform Krylov estimates for the approximating SDEs (Lemma 4.5), obtain existence by tightness and Skorokhod representation (Theorem 4.9), and prove uniqueness via a generalized Itô formula and martingale-problem localization (Theorems 4.10--4.19). Theorem 1.4 states existence, uniqueness, Krylov estimates, and moment bounds for all κ∈[0,1+α_b), with an additional weighted Krylov estimate for smaller κ. Section 5 applies the result to divergence-free Gaussian random fields.","tokens_in":30887,"tokens_out":9476,"duration_ms":101017,"significance":"If the central theorem is correct, this is a substantial extension: it pushes the admissible negative regularity for kinetic SDE drifts from the paracontrolled range α∈(-2/3,-1/2) in [17] to all α∈(-1,0), while allowing polynomial weights in the drift. The proof combines several nontrivial tools: weighted anisotropic Hölder spaces, Bony paraproducts for distributional drifts, localization in PDE Schauder theory, Krylov estimates for approximating SDEs, and martingale-problem uniqueness. The Gaussian example (Example 5.3) gives a concrete family of admissible drifts. The main shortcoming is that the proof of Theorem 1.4 relies on a reduction that loses regularity in κ, so the stated full range of α for κ>0 is not established by the arguments given.","major_comments":[{"comment":"The reduction of the weighted hypothesis b∈L^{q_b}_T C^{α_b}_a(ρ_κ) to assumption (Hsub) via Remark 4.1 produces a rough component b1 in C^{α_b−κ}_a, not in C^{α_b}_a. Since Theorem 4.19 requires b1∈L^{q_b}_T C^{α_b}_a and returns Krylov's estimate only for α∈(-1,α_b], the actual chain of implications gives α∈(-1,α_b−κ] for κ>0. Thus the stated range α∈(-1,α_b] in Theorem 1.4 is unsupported for κ>0, including the regime κ∈((1+α_b−2/q_b)/(3+α_b−2/q_b),1+α_b) where the weighted assumption (Hsub_w) is unavailable. The theorem should either be restated with the reduced exponent α_b−κ, or Remark 4.1 must be strengthened to preserve the original regularity α_b in the unweighted rough component.","section":"4"},{"comment":"The reduction in Remark 4.1 also does not verify the divergence condition required by (Hsub). To apply Theorem 4.19 one needs div_v b1∈L^{q_b}_T C^{α_b}_a, but the hypotheses of Theorem 1.4 give at most a bounded, or L^{q_b}_T C^{α_b}_a, divergence for b, and differentiating the component b1∈C^{α_b−κ}_a loses one more derivative of v-regularity. The statement of Theorem 1.4 uses the phrase 'bounded divergence in v' while Theorem 1.2 and (Hsub_w) use div_v b∈L^{q_b}_T C^{α_b}_a; this ambiguity matters because Lemma 2.8 and the generalized Itô formula in Lemma 4.10 require paraproduct estimates for the divergence term. Please state the exact divergence assumption and prove that a decomposition satisfying (4.1) exists under it.","section":"4"}],"minor_comments":[{"comment":"The phrase 'bounded divergence in v' should be made precise; in Theorem 1.2 and in (Hsub_w) the condition is div_v b∈L^{q_b}_T C^{α_b}_a, whereas in Theorem 1.4 it could be read as merely div_v b∈L∞.","section":"1.1"},{"comment":"When κ equals the upper endpoint (1+α_b−2/q_b)/(3+α_b−2/q_b), the lower endpoint of the α-interval in (1.12) coincides with α_b, so the stated interval is empty; either exclude the endpoint or give a limiting interpretation.","section":"1.1"},{"comment":"After deriving U∈L^p(Ω;C^β_a(ρ_κ)) for β<3(γ−d)/2−4d/p, the paper should explicitly state that one chooses β∈(-1,0) so that the hypotheses α_b∈(-1,0) of Theorem 1.4 are met, and should verify κ∈[0,1+β) for the chosen p.","section":"5"},{"comment":"There are several typographical issues: 'Schwarz space' should be 'Schwartz space', 'Prohorov' is usually spelled 'Prokhorov', and the displayed formula in the proof of Lemma 3.3 contains a stray period in '∥I^λ_ . (f)∥'.","section":"2"},{"comment":"The lower bound in the α-range in Theorem 1.2, α>2/q_b+(3κ−1)/(1−κ), deserves a brief explanation of its provenance, since it is not immediately evident why this combination of κ and q_b appears.","section":"1.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is sophisticated and the proof framework is coherent, but the main theorem as stated overreaches its proof for κ>0 because the reduction in Remark 4.1 lowers the regularity of the rough component from α_b to α_b−κ. A revision should either prove a sharper decomposition or restrict the statements accordingly; the current version cannot be accepted as-is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is a real step forward: it extends weak well-posedness for kinetic SDEs with distributional drift into the previously open range alpha in (-1,-2/3], under a velocity-divergence condition. The proof chain is coherent, and the PDE estimates (Theorems 3.4 and 3.6) plus the generalized Ito formula (Lemma 4.10) are the genuine new tools. The Gaussian example is a nice application, and the authors are upfront about the div_v restriction. The reliance on the same group's earlier paracontrolled framework is appropriate; the new range is the contribution.\n\nThe main soft spot is in the passage from the weighted assumption in Theorem 1.4 to the unweighted condition (Hsub). Remark 4.1 decomposes b in C^{alpha_b}(rho_kappa) into b1 in C^{alpha_b-kappa} plus a smooth part, which only verifies (Hsub) with the lowered exponent alpha_b-kappa, not alpha_b. Then Theorem 4.19 can only give the Krylov estimate for alpha in (-1, alpha_b-kappa]. The full range alpha in (-1, alpha_b] claimed in Theorem 1.4 for every kappa in [0,1+alpha_b) is not supported by the proof. This is not cosmetic: the Gaussian example needs kappa > 3d/p > 0, so the issue is live. The smaller-kappa weighted estimate (1.12) does work when kappa is below the threshold, and that is likely enough for the example, but the theorem statement overreaches.\n\nThere is also some vagueness in the \"bounded divergence\" hypothesis of Theorem 1.4. The proof uses the stronger condition div_v b in L^{q_b}_T C^{alpha_b}_a, as in (Hsub_w). If you only have L^infty divergence, the paraproduct estimates (2.16) do not go through.\n\nOverall, the core machinery is sound. The fix is probably a corrected statement: either restrict the unweighted Krylov range to alpha_b-kappa, or find a sharper decomposition that preserves the exponent. I would not desk-reject this; I would send it out and let the referees push on the parameter range.","headline":"Solid new step into the subcritical range for kinetic SDEs, but the main theorem overclaims the Krylov range for weighted drifts because the reduction to the unweighted condition loses regularity.","tokens_in":31515,"tokens_out":6362,"would_cite":true,"duration_ms":66931,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60J60","35K70","60G60"],"pacs":[],"model":"deepseek-v4-flash","headline":"Kinetic SDEs with subcritical distribution-valued drifts have unique weak solutions whenever the drift's velocity divergence is as regular as the drift itself, with Krylov estimates and moment bounds.","keywords":["kinetic SDEs","distributional drift","anisotropic Hölder space","paraproduct","Krylov estimate","weak well-posedness","Gaussian random fields","martingale problem"],"falsifier":"Construct a drift from the Gaussian field of Example 5.3 multiplied by a $v$-profile $\\eta$ that is only $\\alpha_b$-Hölder in $v$, so that $b\\in C_a^{\\alpha_b}$ but $\\operatorname{div}_v b$ is one order rougher, and check whether the paraproduct estimate (2.16) and then the mollified drift integral in Definition 1.1 still behave as claimed. If (2.16) fails and the $L^2$ limit of $\\int_0^t b_n(s,Z_s)\\,ds$ is not finite, the divergence regularity is genuinely load-bearing; if the limit exists and is unique anyway, the theorem's assumption is stronger than needed.","tokens_in":30283,"feed_emoji":"🎲","tokens_out":17620,"duration_ms":178303,"temperature":0.7,"pith_summary":"The paper proves that a degenerate kinetic SDE - position driven by velocity, velocity driven by Brownian noise plus a distribution-valued drift - is well posed whenever the drift is subcritical (Hölder exponent $\\alpha_b\\in(-1,0)$ with integrability $q_b>2/(1+\\alpha_b)$) and its velocity divergence has the same Hölder regularity as the drift itself. The weak solution is defined through a mollification limit: integrals of the mollified drifts against the path converge in $L^2$, and the limiting process satisfies the Krylov estimate (1.11), which controls occupation-time integrals by the anisotropic Hölder norm of the test function, together with a moment bound in weighted distance. Along the way the paper builds a weighted Schauder theory for the associated kinetic Kolmogorov equation, and the main well-posedness theorem also covers drifts decomposed into a singular Hölder piece plus a regular piece with linear growth. This matters because distribution-valued drifts arise naturally as Gaussian random fields modelling particles in a random environment, and earlier frequency-splitting treatments only reached Hölder exponents between $-2/3$ and $-1/2$; the present result spans the whole subcritical interval $(-1,0)$ at the price of the divergence condition.","feed_headline":"Unique weak solutions for kinetic SDEs with subcritical rough drifts","feed_subtitle":"Velocity divergence matching the drift's Hölder regularity unlocks the full subcritical range.","key_machinery":"The load-bearing construction is a paraproduct (frequency-splitting) calculus on weighted anisotropic Hölder spaces $C_a^\\alpha(\\rho_\\kappa)$, whose scaling vector $a=(3,1)$ encodes the kinetic relation that position scales like the cube of velocity. The product $b\\cdot\\nabla_v u$, meaningless for distributional $b$, is redefined as $b\\odot\\nabla_v u-(\\operatorname{div}_v b)\\prec\\!\\!u$ with $\\prec$ the low-frequency paraproduct; the key estimate (2.16) shows the second term is controlled exactly when $\\operatorname{div}_v b$ lies in the same Hölder space as $b$. With this product, the kinetic Kolmogorov equation $\\partial_t u=\\Delta_v u-v\\cdot\\nabla_x u-\\lambda u+b\\cdot\\nabla_v u+f$ is solved by localization to anisotropic balls and sharp Schauder estimates, giving the weighted regularity of $u$ used to derive Krylov bounds for the approximating SDEs. Uniqueness rests on a generalized Itô formula (Lemma 4.10) plus a martingale-problem argument with stopping-time localization.","core_discovery":"The central claim (Theorem 1.4) is that if $b\\in L_T^{q_b}C_a^{\\alpha_b}(\\rho_\\kappa)$ with $\\alpha_b\\in(-1,0)$, $q_b\\in(2/(1+\\alpha_b),\\infty]$, $\\kappa\\in[0,1+\\alpha_b)$ and $\\operatorname{div}_v b$ belongs to the same weighted anisotropic Hölder space, then for every starting point $z_0$ and every $p\\ge2$ there is a unique weak solution to $dX_t=V_t\\,dt$, $dV_t=b(t,X_t,V_t)\\,dt+\\sqrt{2}\\,dW_t$. Weak solution means the mollified drift integrals $\\int_0^t b_n(s,Z_s)\\,ds$ converge in $L^2$ and the limiting path is well defined; the solution satisfies Krylov's estimate (1.11) and the moment bound $\\mathbb{E}\\sup_t\\rho_\\delta(Z_t)\\le C\\rho_\\delta(z_0)$. The theorem also holds for $b=b_1+b_2$ with $b_1$ singular and $b_2$ a regular drift of at most linear growth (Theorem 4.19). The route is to solve the kinetic Kolmogorov equation with the distributional drift interpreted through paraproducts, use the resulting regularity to get uniform Krylov estimates for approximating diffusions, and prove uniqueness through a generalized Itô formula and a martingale-problem argument.","pith_inferences":["The paper explicitly leaves the supercritical regime $2/q>1+\\alpha$ open; since the Schauder scale gives $\\nabla_x u$ no regularity there, crossing that threshold would require a qualitatively different energy or renormalization argument.","Because the sole extra condition is on $\\operatorname{div}_v b$, the method suggests that divergence-free random drifts — the physically natural class for velocity fields — are exactly where the subcritical threshold is the real frontier, and one could try to push $\\alpha_b$ toward $-1$ by exploiting the vanishing of the dangerous paraproduct term.","The weighted estimates (1.12) open a duality route to quantitative density bounds: the transition density lies in weighted Besov spaces by Remark 1.6, so extracting explicit heat-kernel upper bounds from the same localization machinery is a natural next step."],"forward_implications":["Every subcritical Hölder exponent $\\alpha_b\\in(-1,0)$ is now covered for weak well-posedness, so the obstruction left by earlier treatments is the divergence condition rather than the Hölder exponent.","The Krylov estimate (1.11) and its weighted version (1.12) give quantitative control of occupation-time integrals; as the paper notes, this implies the transition law admits a density with weighted anisotropic Besov regularity.","The moment bound $\\mathbb{E}\\sup_t\\rho_\\delta(Z_t)\\le C\\rho_\\delta(z_0)$ means the solution inherits polynomial growth or decay from the weight, so both confined and heavy-tailed initial data are within scope.","For $b=b_1+b_2$, adding a regular drift of at most linear growth preserves uniqueness, so rough random-field terms can be combined with smooth confining or forcing terms without leaving the theory."],"supporting_citations":[{"why":"Supplies the anisotropic Hölder framework, the localized paraproduct estimates, and the Gaussian-field regularity results that this paper extends.","marker":"[17]"},{"why":"Provides the mollified weak-solution definition, Krylov-estimate framework, and Young-integral substitution used in Section 4.","marker":"[16]"},{"why":"Gives the sharp kinetic Schauder estimate for the unperturbed Kolmogorov equation that anchors Lemma 3.3 and Theorem 3.4.","marker":"[14]"},{"why":"Establishes the sharp Schauder estimates and maximum principle for smooth-coefficient kinetic PDEs used for the approximating equations.","marker":"[3]"},{"why":"Provides the Besov-space interpolation and duality facts underlying the weighted anisotropic Hölder spaces.","marker":"[1]"},{"why":"Supplies the martingale-problem framework and the stopping-time gluing lemmas behind the uniqueness argument for b=b1+b2.","marker":"[26]"},{"why":"Contains the generalized Itô formula for singular kinetic SDEs that links the PDE solution to the martingale problem.","marker":"[33]"},{"why":"Gives the stochastic calculus without probability measures on which the generalized Itô formula is based.","marker":"[9]"}],"fun_headline_variants":["Kinetic SDEs: unique weak solutions for subcritical distributional drifts","Subcritical rough drifts: kinetic SDEs get unique weak solutions","Well-posed kinetic SDEs with subcritical Hölder drifts","Unique weak solutions for kinetic SDEs under subcritical drift conditions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument requires the velocity divergence of the singular part of the drift to be exactly as regular as the drift itself (the same Hölder exponent in the same weighted space); if $\\operatorname{div}_v b$ is a degree rougher, the paraproduct estimate that makes $b\\cdot\\nabla_v u$ meaningful fails and the construction cannot start.","fun_headline_variants_meta":{"raw":{"variants":["Kinetic SDEs: unique weak solutions for subcritical distributional drifts","Subcritical rough drifts: kinetic SDEs get unique weak solutions","Well-posed kinetic SDEs with subcritical Hölder drifts","Unique weak solutions for kinetic SDEs under subcritical drift conditions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000324,"raw_usage":{"total_tokens":1911,"prompt_tokens":1131,"completion_tokens":780,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":747,"completion_tokens_details":{"reasoning_tokens":700}},"tokens_in":747,"tokens_out":780,"duration_ms":8387,"temperature":1.0,"reasoning_tokens":700,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:26:40.668795+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a drift from the Gaussian field of Example 5.3 multiplied by a $v$-profile $\\eta$ that is only $\\alpha_b$-Hölder in $v$, so that $b\\in C_a^{\\alpha_b}$ but $\\operatorname{div}_v b$ is one order rougher, and check whether the paraproduct estimate (2.16) and then the mollified drift integral in Definition 1.1 still behave as claimed. If (2.16) fails and the $L^2$ limit of $\\int_0^t b_n(s,Z_s)\\,ds$ is not finite, the divergence regularity is genuinely load-bearing; if the limit exists and is unique anyway, the theorem's assumption is stronger than needed.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the anisotropic Hölder framework, the localized paraproduct estimates, and the Gaussian-field regularity results that this paper extends."},{"cited_title":"Chaudru de Raynal, I","cited_arxiv_id":null,"evidence_quote":"Establishes the sharp Schauder estimates and maximum principle for smooth-coefficient kinetic PDEs used for the approximating equations."},{"cited_title":"Bahouri, J.-Y","cited_arxiv_id":null,"evidence_quote":"Provides the Besov-space interpolation and duality facts underlying the weighted anisotropic Hölder spaces."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the martingale-problem framework and the stopping-time gluing lemmas behind the uniqueness argument for b=b1+b2."},{"cited_title":"F¨ ollmer, Calcul d’Itˆ o sans probabilit´ es.Seminar on Probability, XV (1981), 143–150","cited_arxiv_id":null,"evidence_quote":"Gives the stochastic calculus without probability measures on which the generalized Itô formula is based."}],"review_version":2}