{"id":"5bfe2217-d101-434e-9043-d83b8e6d498e","arxiv_id":"2611.16344","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Strong existence and path-by-path uniqueness for fractional SDEs with integrable drifts are proved under conditions that are optimal up to endpoint cases and recover the Krylov-Röckner condition for Brownian motion.","lead":"This paper proves that stochastic differential equations driven by fractional Brownian motion with Hurst parameter less than 1/2 have unique strong solutions even when the drift is only integrable in space and time, under conditions that match the classical Krylov-Röckner condition for Brownian motion.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Condition (2.3d) is used in Lemma 4.10(iii) to obtain the Gagliardo–Nirenberg interpolation that feeds the exponential integrability bound (4.35); the paper itself flags it as improvable (line b127).","rationale":"The reader's weakest_assumption correctly identifies condition (2.3d) as the most fragile part of the assumptions. The paper itself acknowledges it as potentially improvable (line b127). The concern is that the proof of Lemma 4.10(iii) relies on the Gagliardo–Nirenberg interpolation (Lemma 3.9), which requires (2.3d). If (2.3d) fails for some parameters that still satisfy the subcritical condition (2.3b) and the Girsanov condition (2.3c), then the exponential integrability bound (4.35) is not proved, and the subsequent Girsanov argument (Lemma 4.12) and the strong-existence proof (Proposition 5.1) would be incomplete for those parameters. The paper does not provide an alternative argument, so the central claim is established only under the full set (2.3). This is a genuine limitation, but it does not invalidate the proof under the stated conditions. The paper is honest about the limitation. The verdict should remain ACCEPT because the proof is rigorous under the assumptions given. The concrete test would clarify whether (2.3d) is actually necessary or can be relaxed, but that is a direction for future work, not a flaw in the current paper. The reader's verdict is unchanged.","tokens_in":68816,"tokens_out":1304,"duration_ms":10853,"concrete_test":"Check whether Lemma 4.10(iii) can be proved without the Gagliardo–Nirenberg interpolation (Lemma 3.9) for parameters (p,q) that satisfy (2.3b) and (2.3c) but violate (2.3d). For example, take d=2, H=0.4, p=1.8, q=3. Then (2.3b): 0.4*2/1.8 + 1/3 ≈ 0.444+0.333=0.777 < 0.6? No, 0.777 > 0.6, so that example fails. Find a concrete triple (d,H,p,q) that satisfies (2.3b) and (2.3c) but violates (2.3d). Then attempt to derive the exponential bound (4.35) by an alternative route (e.g., using a different interpolation or a direct estimate). If the bound fails, then (2.3d) is necessary for the proof; if it can be recovered, then (2.3d) is redundant and the paper's condition is stronger than needed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Theorem 2.1) requires the full set of conditions (2.3). Condition (2.3d) is the most fragile: it is a technical inequality that arises from the Gagliardo–Nirenberg interpolation in Lemma 3.9, which is used in Lemma 4.10(iii) to bound ∥Uf∥_{˚W^{H+1/2+ε}_2} in terms of ∥Uf∥_{˚W^{1−Hd/p−ε'}_q} and ∥Uf∥_{˚W^1_p}. The paper itself notes (line b127) that it is an interesting question whether (2.3d) can be replaced by a weaker condition. The concern is not that (2.3d) is false, but that it is a non-trivial restriction whose necessity is not demonstrated: there may exist parameters (p,q) satisfying the subcritical condition (2.3b) and the Girsanov condition (2.3c) but violating (2.3d), for which the proof of the exponential integrability (4.35) would not go through. Since (4.35) is the key input for the Novikov-type bound (4.39) that justifies the Girsanov change of measure in Lemma 4.12, and Lemma 4.12 is used in the strong-existence argument (Proposition 5.1) and in the stability estimates (Theorem 2.3), a gap in (4.35) would propagate to the main results. The paper acknowledges this as an open question, which is honest, but the reader must assess whether the claimed results are proved under the stated conditions. The proof of Lemma 4.10(iii) relies on Lemma 3.9, which itself assumes condition (3.32) that is equivalent to (2.3d). If (2.3d) is not satisfied, the interpolation (3.33) does not hold, and the exponential bound (4.35) is not established. The paper does not provide an alternative argument for those parameters. Hence the central claim is proved only under the full set (2.3), but the necessity of (2.3d) is not justified, and the paper does not rule out that some parameters satisfying (2.3b) and (2.3c) but not (2.3d) might still satisfy the conclusion. This is a load-bearing concern because it identifies a condition that is both restrictive and not obviously necessary, and the proof of the main result depends on it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper studies SDEs driven by additive fractional Brownian motion with Hurst parameter H<1/2 and a time-dependent drift b in L^q([0,T]; L^p(R^d)). Under conditions (2.3) (subcriticality, p≥2Hd, and an additional technical condition (2.3d) for certain parameter ranges), the authors prove strong existence and path-by-path uniqueness of solutions. They also establish stability estimates (pathwise and in moment norms) and construct a Hölder continuous semiflow. The methods combine stochastic sewing, rough paths, Girsanov's theorem, and John–Nirenberg inequalities for VMO processes. The results recover the Krylov–Röckner condition in the Brownian limit H→1/2 up to an endpoint case.","tokens_in":69364,"tokens_out":4091,"duration_ms":27442,"significance":"If correct, this is a significant contribution to the theory of regularization by noise for fractional SDEs. The conditions (2.3) are the first to recover the classical Krylov–Röckner condition in the Brownian limit, improving on earlier works (Lê20, GG25) that had a gap. The proof techniques are sophisticated and combine several modern tools (stochastic sewing, rough paths, VMO estimates). The paper is well-structured and the arguments appear rigorous. The results are likely to be influential for further work on singular SDEs driven by fractional noise.","major_comments":[{"comment":"Condition (2.3d) is the most fragile part of the assumptions. The authors themselves note (line b127) that it is an interesting question whether it can be replaced by a weaker condition. The condition is used in Lemma 4.10(iii) via the Gagliardo–Nirenberg interpolation (Lemma 3.9) to obtain the exponential integrability bound (4.35), which is then used in the Girsanov argument (Lemma 4.12) and consequently in the strong existence proof (Proposition 5.1) and stability estimates. The paper does not provide any example showing that (2.3d) is necessary, nor does it discuss whether there exist parameters (p,q) satisfying (2.3b) and (2.3c) but violating (2.3d) for which the proof would break. While the authors are honest about this limitation, the reader must assess whether the claimed results are proved under the stated conditions. The proof of Lemma 4.10(iii) relies on Lemma 3.9, which itself assumes condition (3.32) equivalent to (2.3d). If (2.3d) is not satisfied, the interpolation (3.33) does not hold, and the exponential bound (4.35) is not established. Since (4.35) is a key input for the Novikov-type bound (4.39) that justifies the Girsanov change of measure in Lemma 4.12, and Lemma 4.12 is used in the strong-existence argument (Proposition 5.1) and in the stability estimates (Theorem 2.3), a gap in (4.35) would propagate to the main results. The paper acknowledges this as an open question, which is honest, but the reader must assess whether the claimed results are proved under the stated conditions. The proof of Lemma 4.10(iii) relies on Lemma 3.9, which itself assumes condition (3.32) that is equivalent to (2.3d). If (2.3d) is not satisfied, the interpolation (3.33) does not hold, and the exponential bound (4.35) is not established. The paper does not provide any alternative argument or relaxation. This is a load-bearing issue because the central claim (Theorem 2.1) requires the full set of conditions (2.3). The authors should either prove that (2.3d) is actually necessary for the result, or provide a modified argument that avoids it, or at least discuss the range of parameters where (2.3d) is automatically satisfied given (2.3b) and (2.3c). As it stands, the proof is conditional on (2.3d) and the paper does not demonstrate that this condition is non-vacuous or that it can be removed.","section":"§2.1, condition (2.3d)","authors_acknowledge":"Forp< 2, the latter only requiresHdp +H < 12 + Hd2 , that is,(2.3d) with q =∞; it is an interesting question whether(2.3d) can be replaced by this weaker condition in our main results.","authors_acknowledge_p":0.91},{"comment":"The Gagliardo–Nirenberg inequality in Lemma 3.9 is stated for functions ϕ: [0,T]→R^d with ϕ_0=0. The proof of Lemma 4.10(iii) applies this to Uf = ∫_0^· f(r,G_r)dr, which indeed vanishes at 0. However, the inequality (3.33) involves norms ∥ϕ∥_{˚W^{1-γ}_{p_1}} and ∥ϕ∥_{˚W^1_{p_2}}. For the application in Lemma 4.10(iii), the authors set p_1 = q, p_2 = p, and γ = Hd/p + ε'. The condition (3.32) becomes H + (Hd/p)θ < 1/2, where θ = (1/p - 1/2)/(1/p - 1/q). This is exactly (2.3d) when ε'→0. The authors note that (2.3d) is a strict inequality, so (3.32) holds for small ε'>0. However, the embedding ˚W^{1-γ}_{p_1} requires that 1-γ > 1/p_1, i.e., 1 - Hd/p - ε' > 1/q, which is equivalent to (2.3b) with a margin. This is fine. But the use of the Gagliardo–Nirenberg inequality from [BM18, Theorem 1(A)] requires that the exponents are such that the exceptional case [BM18, (1.9)] does not occur. The authors mention that 1-ε is not an integer, so the exceptional case does not occur. However, they also need to ensure that the inequality holds for functions with ϕ_0=0, which is fine because the inequality is for inhomogeneous Sobolev norms and they later convert to ˚W norms using Poincaré. The proof seems correct, but the reliance on a rather technical interpolation inequality with a condition (3.32) that is not obviously satisfied for all parameters satisfying (2.3b) and (2.3c) is a concern. The authors should provide more explanation of why (2.3d) is the correct condition for the interpolation to work, and perhaps give examples of parameters where (2.3b) and (2.3c) hold but (2.3d) fails, to illustrate the gap.","section":"§4.2, Lemma 4.10(iii) and Lemma 3.9","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of strong existence (Proposition 5.1) relies on Proposition 5.7, which shows that the limit of solutions of approximating equations is a solution. The proof of Proposition 5.7 uses Lemma 5.6 (a priori bound) and Corollary 5.5, which in turn rely on Lemma 5.4 and Lemma 5.3. Lemma 5.3 uses the multiparameter Garsia–Rodemich–Rumsey inequality [HL13, Theorem 3.1] to obtain the pathwise estimates (5.11) and (5.12). The application of this inequality requires that the process (t,x) ↦️ U^{f(·+x)}_t is continuous. For bounded continuous f, this is fine. For general f in L^qL^p, the authors use an approximation argument (Step 3 and Step 4 in the proof of Lemma 5.3). The argument seems plausible, but it is quite involved and relies on the fact that L^qL^p embeds into L^qB^{-η}_p for η>0, and that the Besov norm controls the Ξ functional. The authors should check that the multiparameter Garsia–Rodemich–Rumsey inequality applies to the function (s,t,x,y) with the given exponents, and that the constant C_{T,K} is chosen appropriately. The proof is sketchy and refers to [ALL23] for details. Since this is a crucial step for the path-by-path regularization estimates, the authors should provide a more self-contained argument or at least clearly state the assumptions of the Garsia–Rodemich–Rumsey inequality and verify them.","section":"§5.3, Proposition 5.7 and the proof of strong existence","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of moment stability (Proposition 6.1) uses a rough path lift of V and R and a Grönwall argument in L^m(Ω). The key estimate is Lemma 6.7, which gives a bound on ∥Y_t - Y_s∥_{L^m(Ω)}. The proof uses Lemma 6.6 to control the remainder of the rough path expansion. However, the definition of JY K_{s,t} involves suprema over subintervals and differences of Y, R, and iterated integrals. The proof that JY K_{s,t} satisfies the inequality (6.18) under the condition (6.17) relies on the estimates from Lemma 6.5 and Lemma 6.6. The authors then apply a Grönwall lemma (Lemma 3.11) to obtain the final bound. The argument is quite technical and it is not entirely clear that all the constants are uniform in the way claimed. In particular, the constant C in Lemma 6.6 depends on ∥b_1∥_{L^qL^p} and ∥b_2∥_{L^qL^p} (or Λ_{b_i}(η)), and the exponential factor e^{C(1∨Γ)^{n/γ}} in Proposition 6.1 seems to come from the Grönwall argument. The authors should verify that the Grönwall lemma (Lemma 3.11) is applicable with the given control w(s,t)=t-s and the function G_t = ∥Y_t∥_{L^m(Ω)}. The condition (3.39) requires that G_t - G_s ≤ C sup_{r≤t} G_r w(s,t)^{1/κ} + Υ(s,t) for all s<t with w(s,t) ≤ L. The authors obtain (6.18) which is of this form with κ = 1/γ and Υ(s,t) = 2K(1∨Γ^n)∥b_1-b_2∥_{L^qB^{-1}_p} (t-s)^γ. However, (6.18) is an inequality for ∥Y_t - Y_s∥_{L^m(Ω)}, not for G_t - G_s = ∥Y_t∥_{L^m(Ω)} - ∥Y_s∥_{L^m(Ω)}. The authors seem to use the fact that ∥Y_t∥_{L^m(Ω)} - ∥Y_s∥_{L^m(Ω)} ≤ ∥Y_t - Y_s∥_{L^m(Ω)} by the reverse triangle inequality, so (6.18) implies G_t - G_s ≤ 2K(1∨Γ^n) { sup_{r∈[s,t]} G_r + ∥b_1-b_2∥_{L^qB^{-1}_p} } (t-s)^γ. This is indeed of the form (3.39) with C = 2K(1∨Γ^n) and w(s,t)=t-s, κ=1/γ, and Υ(s,t)=2K(1∨Γ^n)∥b_1-b_2∥_{L^qB^{-1}_p} (t-s)^γ. The condition w(s,t) ≤ L is satisfied if t-s ≤ L^γ. The Grönwall lemma then gives sup_{t∈[0,T]} G_t ≤ 2 exp(w(0,T)/(λL)) (G_0 + Υ(0,T)) with λ = 1∧[L(2Ce^2)^κ]^{-1}. Choosing L small enough (e.g., L = (2C)^{-1/γ}) ensures that the condition holds on intervals of length ≤ L^γ, and then a chaining argument extends to the whole interval [0,T]. This seems plausible, but the authors should provide more details on the application of Lemma 3.11, especially the choice of L and the chaining argument.","section":"§6.3, Lemma 6.7 and the moment stability proof","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The authors state that the endpoint p=2 is not recovered when H=1/2. This is a minor limitation, but it would be helpful to mention whether this is an artifact of the proof or a genuine obstruction.","section":"§2.1, line b117","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The Gagliardo–Nirenberg inequality is stated for functions ϕ: [0,T]→R^d with ϕ_0=0. The reference [BM18, Theorem 1(A)] is for functions on R^n, not on an interval. The authors should explain how the inequality is adapted to the interval [0,T] (e.g., by extension or by using the periodic setting).","section":"§3.2, Lemma 3.9","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The statement of Lemma 4.2 includes a term 1_{H<1/2} in the bound (4.4). Since the paper focuses on H<1/2, this indicator is always 1. It would be clearer to state the bound without the indicator and note that for H≥1/2 the term may be absent.","section":"§4.1, Lemma 4.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"In the statement of Lemma 4.10, the condition (2.3d) is only needed for part (iii). The authors should make this explicit in the statement to avoid confusion.","section":"§4.2, Lemma 4.10","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of Lemma 5.3 uses the multiparameter Garsia–Rodemich–Rumsey inequality [HL13, Theorem 3.1]. The authors should verify that the conditions of that theorem are satisfied, in particular that the function (t,x) ↦️ U^{f(·+x)}_t is continuous and that the exponents are chosen appropriately. The argument is sketchy and relies on [ALL23] for details; a brief verification would improve readability.","section":"§5.2, Lemma 5.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The authors state that the ℓ-variation norm of V is not exponentially integrable in general. This is used to justify the need for the rough path lift. However, they do not provide a counterexample or a reference. A brief explanation or citation would be helpful.","section":"§6.2, line b601","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of Proposition 7.3 uses Young's inequality multiple times. The constants are not tracked explicitly, but the final estimate (7.6) is plausible. The authors should ensure that the exponents are consistent: β = min(1-ε+γ, γ/ε) > 1 is used later. The derivation of (7.6) from the estimates is a bit terse; a few more intermediate steps would improve clarity.","section":"§7.2, Proposition 7.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The mollifying sequence is defined as f_n(t,x) = 1_{∥f(t,·)∥_{L^p} ≤ n} P_{1/n} f(t,·)(x). The truncation at level n ensures boundedness. However, the authors should check that the truncation does not affect the convergence in L^qB^{-ε}_p. The proof uses the estimate ∥P_δ g - g∥_{B^{-ε}_p} ≤ C δ^{ε/2} ∥g∥_{B^0_p} ≤ C δ^{ε/2} ∥g∥_{L^p}. This is valid for ε∈(0,2]. The truncation introduces an error term 1_{∥f(t,·)∥_{L^p}>n} f(t,·), which tends to 0 in L^qL^p by dominated convergence. The argument is standard and seems correct.","section":"Appendix A, Lemma A.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper is well-written and the proofs are detailed. However, the length (65 pages) makes it challenging to review thoroughly. The authors have done a good job of structuring the paper and providing references to the literature. The use of the stochastic sewing lemma and VMO estimates is innovative.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"This paper proves strong existence, path-by-path uniqueness, stability, and a Hölder semiflow for SDEs driven by fractional Brownian motion with H<1/2 and drifts in L^qL^p. The main advance is that the conditions recover the Krylov–Röckner threshold as H→1/2, which earlier work (Lê 2020, Galeati–Gerencsér 2025) did not achieve. The combination of stochastic sewing, rough paths, and VMO estimates is genuinely novel and the proofs are detailed and rigorous. The paper is a serious piece of work and deserves a careful referee.\n\nThe soft spot is condition (2.3d). The authors themselves flag it as improvable (line b127). It is a technical inequality arising from a Gagliardo–Nirenberg interpolation in Lemma 3.9, used to obtain the exponential integrability bound (4.35) that feeds the Girsanov argument. The paper does not show that (2.3d) is necessary, nor does it provide an alternative argument for parameters that satisfy (2.3b) and (2.3c) but violate (2.3d). This is a load-bearing condition: if it fails, the proof of Lemma 4.10(iii) does not go through, and the central claim (Theorem 2.1) is not established for those parameters. The authors are honest about this, but the reader should be aware that the claimed optimality is conditional on a condition whose necessity is open.\n\nThat said, the paper is mathematically correct under the stated assumptions. The proofs are long but well-organized, and the reliance on established theorems (Girsanov, stochastic sewing, rough paths) is appropriate. The citation pattern is fair and the limitations are acknowledged. I would send this to a strong referee; it will require careful checking of the interpolation estimates and the VMO arguments, but the core is sound.\n\nFor a reading group: yes, if the group works on regularization by noise or rough paths. I would cite it in my own work on singular SDEs. The paper deserves peer review.","headline":"Strong well-posedness for fractional SDEs with integrable drifts under conditions that recover the Krylov–Röckner threshold as H→1/2; the proof is technically deep and the main results are solid, but condition (2.3d) is a genuine restriction whose necessity is not demonstrated.","tokens_in":69962,"tokens_out":564,"duration_ms":7474},"pith_extraction":{"msc":["60H10","60H50","60G22","60L20"],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Fractional noise with H < 1/2 regularizes SDEs with integrable drifts to the optimal Krylov–Röckner condition","keywords":["stochastic differential equations","fractional Brownian motion","regularization by noise","path-by-path uniqueness","strong solutions","stochastic sewing lemma","Besov spaces","Krylov–Röckner condition"],"falsifier":"Construct a drift b in L^q([0,T]; L^p(R^d)) that satisfies Hd/p + 1/q < 1 – H and p ≥ 2Hd but violates (2.3d), and show that either strong existence or path-by-path uniqueness fails for the SDE dX_t = b(t,X_t) dt + dB^H_t with H < 1/2. If such a counterexample exists, the condition (2.3d) is necessary; if not, the authors' conjecture that it can be removed would be confirmed.","tokens_in":70356,"feed_emoji":"🔀","tokens_out":1634,"duration_ms":13568,"temperature":0.7,"pith_summary":"This paper shows that fractional Brownian motion with Hurst parameter H < 1/2 can regularize a stochastic differential equation whose drift is merely p-integrable in space and q-integrable in time, all the way down to the subcritical condition Hd/p + 1/q < 1 – H. Under that condition, plus two technical conditions needed for Girsanov's theorem, the authors prove that a strong solution exists and is unique in the strongest sense: path-by-path uniqueness, meaning that for almost every driving path the deterministic integral equation has exactly one continuous solution. They also obtain pathwise and moment stability estimates and construct a Hölder continuous semiflow. When H approaches 1/2 the conditions reduce, up to an endpoint case, to the classical Krylov–Röckner condition for Brownian motion, showing that the result is the natural fractional extension.","feed_headline":"Fractional noise tames integrable drifts down to the optimal condition","feed_subtitle":"Strong well-posedness and path-by-path uniqueness for SDEs driven by fBm with H < 1/2, recovering the Krylov–Röckner limit.","key_machinery":"The proof combines three main tools: (i) the stochastic sewing lemma with conditioning, used to obtain sharp estimates on the ℓ-variation of additive functionals of Gaussian processes satisfying a local nondeterminism condition; (ii) the quantitative John–Nirenberg inequality for processes of vanishing mean oscillation, which yields exponential integrability of Sobolev norms of the drift integral; and (iii) a rough path lift of the linearized equation, allowing moment stability estimates even when the ℓ-variation of the linearisation is not exponentially integrable. The central estimate is Proposition 4.4, which shows that for f in L^q([0,T]; B^α_ρ(R^d)) the additive functional U^f has finite (1/(1+αH – Hd(1/ρ – 1/2)_+)+ε)-variation, improving previous bounds and making the Young–Grönwall argument work under the optimal condition.","core_discovery":"Theorem 2.1: For a d-dimensional SDE dX_t = b(t,X_t) dt + dB^H_t driven by additive fractional Brownian motion with Hurst parameter H < 1/2, if b belongs to L^q([0,T]; L^p(R^d)) and the parameters satisfy Hd/p + 1/q < 1 – H, p ≥ 2Hd, and an additional technical condition (2.3d) when p ≤ (1–H)^{-1} and q > 2, then there exists a strong solution and path-by-path uniqueness holds. This is the first result for multi-dimensional fractional SDEs with integrable drifts that recovers the Krylov–Röckner condition in the Brownian limit H → 1/2.","pith_inferences":["The fact that the subcritical condition Hd/p + 1/q < 1 – H is necessary for the scaling argument (2.11) suggests it is also necessary for any form of well-posedness; proving necessity would be a natural next step.","The VMO-based moment estimates (Section 6) circumvent the failure of exponential integrability of the ℓ-variation norm, a technique that may be transferable to other singular SDEs where the linearisation has poor tail behaviour.","The linearization identity (Proposition 4.15) for integrable drifts via Girsanov and Besov norms provides a clean substitute for the mean value theorem and could be used in other contexts where the drift is not differentiable."],"forward_implications":["The result provides the strongest form of uniqueness (path-by-path) for fractional SDEs with integrable drifts, matching the Brownian case in the limit H → 1/2.","The stability estimates (Theorems 2.3 and 2.6) open the door to numerical approximation schemes with sharp rates for singular fractional SDEs.","The existence of a Hölder continuous semiflow (Theorem 2.6) is a prerequisite for studying random dynamical systems generated by fractional SDEs with singular drifts.","The technical conditions (2.3c) and (2.3d) are identified as artefacts of the Girsanov argument; the authors explicitly ask whether (2.3d) can be weakened, pointing to a clean open problem."],"supporting_citations":[{"why":"Establishes the Brownian-motion benchmark (Krylov–Röckner condition) that the present paper extends to fractional noise.","marker":"[KR05]"},{"why":"Introduces the stochastic sewing lemma and proves strong well-posedness under the non-optimal condition Hd/p + 1/q < 1/2 – H, which the present paper improves.","marker":"[Lê20]"},{"why":"Obtains strong well-posedness and path-by-path uniqueness under the condition Hd/p + 1/min(q,2) < 1 – H, which is also non-optimal and fails to recover the Brownian limit.","marker":"[GG25]"},{"why":"Proves path-by-path uniqueness under the Krylov–Röckner condition for Brownian motion; the strategy (regularization estimates + semiflow) is adapted here to the fractional setting.","marker":"[ALL23]"},{"why":"The first work on fractional SDEs with integrable drifts, but only in dimension one and relying on the comparison principle; the present paper extends to multi-dimensions.","marker":"[NO03]"},{"why":"Provides the quantitative John–Nirenberg inequality for VMO processes, used here to obtain exponential integrability of Sobolev norms of the drift integral.","marker":"[Lê22]"}],"fun_headline_variants":[],"cache_read_input_tokens":0,"weakest_assumption_plain":"The condition (2.3d) – a technical inequality involving H, d, p and q that is needed for the Gagliardo–Nirenberg interpolation in the exponential integrability estimates – is the most fragile part of the assumptions; the authors themselves note that it might be improvable.","fun_headline_variants_meta":{"raw":{"variants":[]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.00011,"raw_usage":{"cost":0.00011011,"is_byok":false,"cost_details":{"upstream_inference_cost":0.00011011,"upstream_inference_prompt_cost":9.2638e-05,"upstream_inference_completions_cost":1.7472e-05},"total_tokens":1114,"prompt_tokens":1018,"completion_tokens":96,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":0}},"tokens_in":1018,"tokens_out":96,"cost_source":"openrouter_usage","duration_ms":1958,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":0,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-13T06:38:30.246188+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Construct a drift b in L^q([0,T]; L^p(R^d)) that satisfies Hd/p + 1/q < 1 – H and p ≥ 2Hd but violates (2.3d), and show that either strong existence or path-by-path uniqueness fails for the SDE dX_t = b(t,X_t) dt + dB^H_t with H < 1/2. If such a counterexample exists, the condition (2.3d) is necessary; if not, the authors' conjecture that it can be removed would be confirmed.","supporting_citations":[],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}