{"id":"e62ee5cd-c7b8-4dda-886a-48808085ff12","arxiv_id":"2412.11323","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An inductive scaling procedure gives functional LIL and distributional limits at time zero for degenerate additive-noise diffusions, plus a sufficient condition for boundary regularity.","lead":"This mathematics paper computes the correct small-time scaling for diffusions with polynomial drift and degenerate additive noise, giving both a functional law of the iterated logarithm and a distributional limit. The results provide a practical, checkable sufficient condition for a boundary point to be regular, applied to models such as the stochastic Lorenz '96 system and second-order Langevin dynamics.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 8.1 hinges on the explicit geometric hypothesis O* ⊂ S_{ε,D}(O_ε); cusps or fine boundary oscillations fall outside it, so the criterion is sufficient but not necessary.","rationale":"I re-checked the main arguments. The scaling induction in Sections 4–5 is consistent: proj1 a_j = proj1 b_j for all j, and the remainders R_{ε,L} and R_{ε,D} vanish uniformly on compacts by Lemma 4.5 and Lemma 5.2. The proof of Corollary 5.4 via a Gronwall comparison is valid because P_D has the triangular structure that guarantees nonexplosion and pathwise convergence. In Theorem 8.1, the chain P{ξ≤εt} ≥ P{y_{ε,t}∈O*} is correct because O*⊂S_{ε,D}(O_ε), and Portmanteau gives liminf_{ε} P{y_{ε,t}∈O*} ≥ P{y_t∈O*} > 0 by the support theorem and the reachability assumption. Blumenthal's 0–1 law then yields regularity. The only non-classical input is the geometric O* condition; it is explicitly stated, and the cusped-boundary example shows the criterion is not necessary, but the paper does not claim necessity. The reader's weakest-assumption analysis accurately identifies this restriction, and I agree with the ACCEPT verdict at moderate confidence.","tokens_in":36168,"tokens_out":31796,"duration_ms":272933,"concrete_test":"Take Brownian motion on R^2 (so b_1=b_2=1/2) and the cusped domain O = {(x,y) : y > |x|^{1/2}}. Compute S_{ε,D}(O_ε) = {(u,v)∈(-1,1)^2 : v > ε^{-1/4}|u|^{1/2}}. For every ε∈(0,1) the right-hand side exceeds 1 whenever |u| > ε^{1/2}, so any fixed nonempty open O* would have to lie inside the degenerate set {u=0, v>0}, which has empty interior in R^2. Hence no O* exists, and Theorem 8.1 does not apply, even though 0 is known to be regular for this domain by the classical cone condition. This confirms the geometric hypothesis is the limiting factor and that the criterion is only sufficient.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The single most load-bearing step in Theorem 8.1 is the containment O* ⊂ S_{ε,D}(O_ε) for all sufficiently small ε. This is where the boundary geometry enters: it converts a positive hitting probability for the rescaled process y_{ε,t} into a positive probability that the original process x_{ε t} lies in O. Every other step in the proof is standard: Corollary 5.4 supplies weak convergence of y_ε to y, Lemma 5.5 supplies the support description, and Blumenthal's 0–1 law converts any positive lower bound into P{ξ=0}=1. If the boundary has structure at a finer scale than the scaling exponents b_j — for instance a cusp x_j = |x_i|^{γ} with γ b_i < b_j — then S_{ε,D}(O_ε) degenerates (it becomes a thin set near the lower-dimensional subspace) and no nonempty open O* can be contained for all small ε. The theorem is then silent even though the origin may be perfectly regular. This is not an internal inconsistency, but it means the 'practical criterion' is confined to boundaries whose local shape is resolved by the polynomial scaling exponents; it does not cover the full class of domains relevant to the regular point problem.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an inductive scaling procedure for diffusions of the form dx_t = P(x_t) dt + σ dB_t with polynomial drift and diagonal, possibly degenerate, additive noise. It identifies two small-time rescalings: a functional law-of-the-iterated-logarithm rescaling and a distributional rescaling. For each rescaling the paper derives a limiting control problem, solves the control problem in several nontrivial examples (Kolmogorov-type chains, Lorenz'96, Sabra shell model, Langevin dynamics) using geometric control theory, and then uses the distributional control data plus a geometric boundary condition to give sufficient criteria for the origin to be a regular boundary point (Theorem 8.1). The main structural theorems are Theorem 4.4 (LIL scaling of the SDE), Theorem 4.7 (functional LIL limit set, deferred to the authors' prior work [7]), Theorem 5.3 and Corollary 5.4 (distributional convergence), Theorem 6.8 (openness of the time-1 reachable set under component homogeneity), and Theorem 8.4 (regularity of all points on level sets of the Hamiltonian for second-order Langevin dynamics).","tokens_in":36342,"tokens_out":29705,"duration_ms":249308,"significance":"If the results are accepted, the paper provides a genuinely algorithmic method for the small-time behavior of a broad class of degenerate diffusions with polynomial drift, going beyond the strongly hypoelliptic setting; the examples of Lorenz'96 and Sabra-type shell models are nontrivial and are worked out in detail. The regular-point criterion is a sufficient condition that converts the regular-point problem into a scaling computation plus a control reachability check, and the paper is careful to state that the criterion is not necessary (the boundary must be well-behaved under the computed scaling). The paper is also honest about its limitations: it explicitly flags the unproven general assertion that noise propagation is weaker than hypoellipticity, it acknowledges that the iterated-Kolmogorov example is covered only under extra geometric hypotheses compared to the known necessary-and-sufficient results of Lachal, and it relies on a published support theorem and a prior LIL theorem. No free parameters are fitted to the answer; the scaling exponents are derived from the drift and noise structure.","major_comments":[{"comment":"Theorem 4.7 is the paper's functional law of the iterated logarithm, but its proof is entirely deferred to [7, Theorem 2.6] with the single sentence: 'The fact that the noise in equation (4.18) is additive allows us to check Assumption 2 in the statement of [7, Theorem 2.6] by way of [7, Corollary 2.10].' The assumptions of [7, Theorem 2.6] are not stated in the manuscript, and the paper does not demonstrate how the remainder R_{ε,L} in (4.18) satisfies the required uniformity on the time interval, nor how explosive trajectories in the space E are handled. Since Theorem 4.7 is one of the two advertised asymptotic results, the verification of the cited theorem's hypotheses is load-bearing. Please either state the hypotheses of [7, Theorem 2.6] and verify them explicitly for (4.18), or provide a self-contained proof.","section":"Section 4, Theorem 4.7"},{"comment":"The paper defines x* ∈ ∂U to be regular for (x_t(x*), U) if P{ξ_{x*} < ∞} = 1, where ξ_{x*} = inf{t > 0 : x_t(x*) ∈ U}. This is inconsistent with the definition used for the origin in (1.2), where 0 is regular only if P{ξ = 0} = 1, i.e., the process enters O immediately. The proof of Theorem 8.4 establishes the immediate-entry property (via Proposition 8.3 and Theorem 8.1), so the displayed definition should read P{ξ_{x*} = 0} = 1 rather than P{ξ_{x*} < ∞} = 1. As written, the statement of Theorem 8.4 asserts a weaker and nonstandard property, and the phrase 'irregular otherwise' is then ambiguous.","section":"Section 8.1, definition of regularity for general boundary points"}],"minor_comments":[{"comment":"The boundary graph condition writes x_j = b(x_1, ..., x_{j-1}, x_j, ..., x_n), which is circular because x_j appears on both sides; the argument of b should presumably be (x_1, ..., x_{j-1}, x_{j+1}, ..., x_n).","section":"Section 8, (8.2)-(8.3)"},{"comment":"The text states that P^j_L depends only on π_{∪_{k=1}^ℓ I_k}(x); this should be π_{∪_{k=0}^ℓ I_k}(x), since for ℓ=0 the union starting at k=1 would be empty, yet P^j_L for j ∈ I_1 depends on the noise directions in I_0.","section":"Section 4.1, after (4.4)"},{"comment":"There are two notation slips: 'Let F = F(V0;V1,V0)' should be 'F = F(V0;V1,V2)', and 'V3 ∈ Scale(V3)' should be 'V3 ∈ Scale(F3)'.","section":"Example 7.6"},{"comment":"In the line 'Set b_{ℓ+1,j} = (m_{ℓ+1}, 0) + (1, 0) if j ∈ I_{k+1}', the index k+1 appears to be a typo for ℓ+1.","section":"Section 5.1, (5.2)"},{"comment":"The support theorem for the SDE (5.7) is quoted from [30,31] and the proof is only a paragraph. It would be helpful to state the precise version of the Stroock–Varadhan support theorem used and to justify the continuity of the map f ↦ φ_·(PD, f)0 on C with the same level of detail as the integrability arguments in (4.21)-(4.24).","section":"Lemma 5.5"},{"comment":"There are minor typographical errors: 'hueristics' should be 'heuristics' in Remark 8.2, and 'Browian' should be 'Brownian' in Example 8.1.","section":"Remark 8.2 and Example 8.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a serious and well-written contribution, and the reader's report recommending acceptance is understandable. My main reservation is the lack of a detailed check of the hypotheses of [7, Theorem 2.6] in the proof of Theorem 4.7; given that this theorem is a central advertized result, I would like to see the verification written out or an appendix added. The second major point is a definitional error in Section 8.1 that should be corrected before publication. The remaining comments are local and should not require extensive new mathematics."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:2412.11323. The core contribution is a general inductive algorithm that computes the correct time-zero scales (both distributional and LIL) for diffusions with polynomial drift and degenerate additive noise, under a \"noise propagating\" condition that is explicit and checkable. That is genuinely new. Prior work covered specific examples (iterated primitives, linear hypoelliptic diffusions) or strongly hypoelliptic systems; this handles a broad polynomial class. The examples are not ornaments—Lorenz'96, Sabra, and the Langevin level-set theorem (8.4) all require real work and are among the best parts.\n\nThe proofs of the main scaling theorems (4.4, 5.3, 8.1) are detailed and I found no errors. The reliance on [7] for Theorem 4.7 is transparent, and since [7] is published and peer-reviewed, that is acceptable. The authors also openly flag that \"noise propagating\" is only conjecturally weaker than hypoellipticity, and they give a concrete example.\n\nThe soft spot is exactly what the stress-test note says: Theorem 8.1 needs a fixed open set O* that survives the scaling. For boundaries with cusps or oscillations at a finer scale than the exponents b_j, the rescaled domain degenerates and the criterion is silent. That does not make the theorem wrong, but it does mean the \"practical criteria\" are practical only for scaling-compatible boundaries. The paper does not claim necessity, so this is a limitation in scope, not an error. Still, I would have liked a more prominent statement that regular points can exist outside the conditions.\n\nBottom line: this deserves a serious referee. I would send it out and, if I were the editor, accept after a minor revision that tightens the framing of Theorem 8.1 and perhaps adds a remark about cusp boundaries. The scaling algorithm alone is worth the paper. I'd bring it to reading group and would cite it.","headline":"A genuinely new scaling algorithm for polynomial-drift degenerate diffusions, with sound proofs and a regular point criterion that is useful but only sufficient.","tokens_in":36958,"tokens_out":4733,"would_cite":true,"duration_ms":42625,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J60","60F15","93B05","35H10","60H07"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper gives a scaling recipe that decides when a degenerate diffusion with polynomial drift enters a domain immediately from a boundary point, reducing the question to a control-reachability check.","keywords":["small-time asymptotics","hypoelliptic diffusion","degenerate additive noise","polynomial drift","law of the iterated logarithm","regular boundary points","geometric control theory","scaling limits"],"falsifier":"Take the iterated Kolmogorov diffusion in \\(\\mathbb{R}^3\\) (Example 8.2) with domain \\(O=\\{x_3>|x_2|^{1/3}+|x_1|^{1/5}\\}\\). The paper's criterion declares 0 regular because the boundary is aligned with \\(b=(1/2,3/2,5/2)\\) and \\(A_{D,t}(0)=\\mathbb{R}^3\\). Compare this with the independent necessary-and-sufficient regular point criterion cited as [20] for the same example; a disagreement would expose an error in the scaling or control-reachability step. Concretely, simulate the exit time for \\(\\varepsilon=$10^{{-3}}$,$10^{{-4}}$,$10^{{-5}}$\\) and check whether \\(P\\{\\xi\\le\\varepsilon\\}\\to1\\).","tokens_in":35889,"feed_emoji":"🎲","tokens_out":8321,"duration_ms":70397,"temperature":0.7,"pith_summary":"This paper develops an inductive scaling procedure that determines, direction by direction, how fast the solution of a degenerate diffusion with polynomial drift and additive noise moves away from its starting point at time zero. Starting from the coordinates where Brownian noise acts directly, each remaining coordinate's characteristic scale is read off from the polynomial drift and the scales already found. The procedure terminates and yields two rescaled limits: a functional law of the iterated logarithm and a weaker distributional limit, each governed by an explicit control system. The distributional control system is used for boundary regularity: if the domain near the origin survives the rescaling and the control trajectories can reach it, then the origin is a regular boundary point, meaning the process enters the domain immediately. The payoff is a practical, checkable criterion for regular points for the whole class of noise-propagating polynomial-drift diffusions, applied here to iterated integrator dynamics, the stochastically perturbed Lorenz'96 model, and second-order Langevin dynamics.","feed_headline":"Scaling algorithm predicts instant entry for degenerate diffusions","feed_subtitle":"Polynomial-drift diffusions with silent coordinates get a control-theoretic test for boundary regularity.","key_machinery":"The load-bearing mechanism is the inductive scaling procedure on the coordinate index set. Initialize \\(I_0=\\{j:\\sigma_j>0\\}\\) with scale \\((1/2,1/2)\\); at each stage, for each remaining coordinate \\(j\\) compute the scaling \\(P_j(a_{\\ell,1},\\ldots,a_{\\ell,n})\\) in the ordered semigroup \\(\\mathcal{S}=(\\tfrac12\\mathbb{Z}_{\\ge0})^2\\) with lexicographic order, select the coordinates whose first component is minimal, define \\(a_{\\ell+1,j}=P_j(a_\\ell)+(1,0)\\), and keep the leading homogeneous polynomial \\(P^j_L\\). If the union of the \\(I_\\ell\\) reaches all coordinates, the system is called noise propagating. The distributional variant uses power scalings with second component 0 and yields the vector field \\(P_D\\). The control problem \\(\\dot{x}=P_D(x)+\\$\\sigma$\\dot{f}\\) has accessible sets \\(A_{D,t}(0)\\); geometric controllability is analyzed through parameterized continuous local semigroups and the uniform saturate, producing simplified ray trajectories. Theorem 8.1 combines the scaling map \\(S_{\\varepsilon,D}\\), the control reachability, and Blumenthal's 0\\u20131 law to conclude regularity.","core_discovery":"On the paper's own terms, the central discovery is the Regular Point Criteria (Theorem 8.1). For a diffusion (1.1) that is noise propagating, the paper constructs a distributional scaling map S_{\\varepsilon,D} and an associated control system \\(\\dot{x}=P_D(x)+\\$\\sigma$\\dot{f}\\), with accessible set \\(A_{D,t}(0)\\). If there is a fixed non-empty open set \\(O_\\ast\\) that stays inside the rescaled domain \\(S_{\\varepsilon,D}(O_\\varepsilon)\\) for all small \\(\\varepsilon\\), and the union over \\(t\\in(0,1]\\) of the closures of \\(A_{D,t}(0)\\) meets \\(O_\\ast\\), then 0 is regular for the domain \\(O\\): the process hits \\(O\\) at arbitrarily small positive times with probability one. The proof runs through convergence of the rescaled process in law to the SDE \\(dy_t=P_D(y_t)dt+\\$\\sigma$ dB_t\\), identification of the support of that limit with the control trajectories via support theory, and Blumenthal's 0\\u20131 law. This is meant as a practical algorithm: compute scales inductively, solve a control problem, check one geometric alignment condition.","pith_inferences":["Beyond the paper, the criterion suggests that boundary regularity for this class of diffusions can be automated: input the polynomial drift, the noise matrix, and the domain boundary, and output the scales \\(b_j\\) plus a verdict on control reachability.","The geometric alignment condition \\(O_\\ast\\subset S_{\\varepsilon,D}(O_\\varepsilon)\\) is the real bottleneck; boundaries with log-periodic oscillations or cusps finer than any power \\(\\varepsilon^{b_j}\\) fall outside it. A natural extension would use the finer LIL rescaling, which carries logarithmic corrections, to treat some of those boundaries.","Because noise propagation is strictly weaker than hypoellipticity (Example 4.5), the scaling method may apply to diffusions supported on lower-dimensional manifolds; a testable question is whether the regular point conclusion survives when the limiting control system is only approximately controllable rather than exactly controllable."],"forward_implications":["For any noise-propagating polynomial-drift system whose domain satisfies the geometric alignment condition, boundary regularity of the origin reduces to a computable scaling computation plus a control reachability check.","The distributional rescaling yields an explicit limiting SDE, and whenever densities exist, the rescaled densities converge pointwise to the limiting density (Corollary 5.6).","The criterion produces concrete regular point theorems for iterated primitives of Brownian motion, the stochastic Lorenz'96 model, and Langevin dynamics on Hamiltonian energy shells (Theorem 8.4).","For the Lorenz'96 model, the control problem is exactly controllable, so any sufficiently aligned domain has 0 regular (Example 8.3)."],"supporting_citations":[{"why":"Supplies the functional LIL theorem used to prove Theorem 4.7, which characterizes the pathwise limit points of the LIL-scaled process.","marker":"[7]"},{"why":"Provides the uniform saturate and parameterized semigroup machinery used in Section 7 to solve the control problems and prove Corollary 7.5.","marker":"[10]"},{"why":"Gives the inductive procedure for saturating trajectories that is adapted in the geometric control section to produce new reachable directions.","marker":"[11]"},{"why":"Establishes the Perron-Wiener cone-type criterion for hypoelliptic diffusions that the present regular point criteria generalize.","marker":"[15]"},{"why":"Provides laws of the iterated logarithm for the iterated Kolmogorov diffusion, used in Example 8.2 as a benchmark.","marker":"[19]"},{"why":"Gives a necessary and sufficient regular point condition for the iterated Kolmogorov diffusion, the comparison case for Example 8.2.","marker":"[20]"},{"why":"Supplies the noise-as-control idea used in the proof of Theorem 6.3 to connect Malliavin covariance to Gramian invertibility.","marker":"[23]"},{"why":"Supports identification of the support of the limiting diffusion with the closure of control trajectories in Lemma 5.5.","marker":"[30]"},{"why":"Together with [30], provides the support theorem used in Lemma 5.5 to identify the support of the limiting process.","marker":"[31]"}],"fun_headline_variants":["Control test reveals boundary regularity for degenerate diffusions","Hypoelliptic diffusions: control criterion predicts instant boundary hits","Scaling plus control solves boundary regularity for silent-coordinate diffusions","Regular point criteria via control geometry for degenerate diffusions","Predict instant boundary entry for hypoelliptic diffusions with control"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument hinges on the existence of one fixed open set \\(O_\\ast\\) that remains inside the rescaled domain \\(S_{\\varepsilon,D}(O_\\varepsilon)\\) for every sufficiently small \\(\\varepsilon\\); if the boundary oscillates or has cusps at scales finer than the polynomial exponents \\(b_j\\), this alignment fails and the criterion has nothing to say.","fun_headline_variants_meta":{"raw":{"variants":["Control test reveals boundary regularity for degenerate diffusions","Hypoelliptic diffusions: control criterion predicts instant boundary hits","Scaling plus control solves boundary regularity for silent-coordinate diffusions","Regular point criteria via control geometry for degenerate diffusions","Predict instant boundary entry for hypoelliptic diffusions with control"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1320,"prompt_tokens":882,"completion_tokens":438,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":498,"completion_tokens_details":{"reasoning_tokens":355}},"tokens_in":498,"tokens_out":438,"duration_ms":4306,"temperature":1.0,"reasoning_tokens":355,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:03:44.845354+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the iterated Kolmogorov diffusion in \\(\\mathbb{R}^3\\) (Example 8.2) with domain \\(O=\\{x_3>|x_2|^{1/3}+|x_1|^{1/5}\\}\\). The paper's criterion declares 0 regular because the boundary is aligned with \\(b=(1/2,3/2,5/2)\\) and \\(A_{D,t}(0)=\\mathbb{R}^3\\). Compare this with the independent necessary-and-sufficient regular point criterion cited as [20] for the same example; a disagreement would expose an error in the scaling or control-reachability step. Concretely, simulate the exit time for \\(\\varepsilon=$10^{{-3}}$,$10^{{-4}}$,$10^{{-5}}$\\) and check whether \\(P\\{\\xi\\le\\varepsilon\\}\\to1\\).","supporting_citations":[{"cited_title":"Carfagnini, J","cited_arxiv_id":null,"evidence_quote":"Supplies the functional LIL theorem used to prove Theorem 4.7, which characterizes the pathwise limit points of the LIL-scaled process."},{"cited_title":"Glatt-Holtz, D.P","cited_arxiv_id":null,"evidence_quote":"Provides the uniform saturate and parameterized semigroup machinery used in Section 7 to solve the control problems and prove Corollary 7.5."},{"cited_title":"Herzog and J.C","cited_arxiv_id":null,"evidence_quote":"Gives the inductive procedure for saturating trajectories that is adapted in the geometric control section to produce new reachable directions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the Perron-Wiener cone-type criterion for hypoelliptic diffusions that the present regular point criteria generalize."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides laws of the iterated logarithm for the iterated Kolmogorov diffusion, used in Example 8.2 as a benchmark."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives a necessary and sufficient regular point condition for the iterated Kolmogorov diffusion, the comparison case for Example 8.2."},{"cited_title":"Mattingly and ´E","cited_arxiv_id":null,"evidence_quote":"Supplies the noise-as-control idea used in the proof of Theorem 6.3 to connect Malliavin covariance to Gramian invertibility."},{"cited_title":"Stroock and S.R.S","cited_arxiv_id":null,"evidence_quote":"Supports identification of the support of the limiting diffusion with the closure of control trajectories in Lemma 5.5."},{"cited_title":"Stroock and S.R.S","cited_arxiv_id":null,"evidence_quote":"Together with [30], provides the support theorem used in Lemma 5.5 to identify the support of the limiting process."}],"review_version":1}