{"id":"34fe341f-7b01-4d22-aa24-23e6ffc204fb","arxiv_id":"2608.04644","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"At time zero, the solution of a linear stochastic fractional diffusion equation obeys sharp Khinchin and Chung laws of the iterated logarithm with explicit constants.","lead":"Random solutions of a broad family of fractional diffusion equations are shown to have exact, non-random oscillation sizes near time zero. The result gives two laws of the iterated logarithm with explicit constants, with one part depending on an imported small-ball theorem.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Chung-type LIL is conditional on an unproved transfer of [13, Thm 4.3(1)] to the Riesz density: Proposition 4.1 asserts, rather than demonstrates, the exact small-ball constant, so the support for Theorem 1.3 is incomplete.","rationale":"The paper is carefully written, and the Khinchin-type LIL proof is mostly self-contained: self-similarity (Lemma 2.1), the Hoelder-type bound (Proposition 2.1), the frequency-truncation estimate (Lemma 2.2, Proposition 2.2), and the block Borel-Cantelli argument are all present in sufficient detail. The Chung-type LIL, however, stands or falls with Proposition 4.1, because both the upper and lower bounds in Section 4.2 invoke it directly. The proof of Proposition 4.1 is a two-paragraph assertion that the exact small-ball theorem of [13] transfers to the Riesz-type density. No decomposition, remainder bound, or correlation estimate is given; the reader is asked to accept that homogeneity and the cited estimates are all that matter. Since [13] is not reproduced and the present paper does not state precisely which theorem conditions match, the most important uncertainty is exactly the one the reader's weakest_assumption identifies. An honest verdict is therefore CONDITIONAL: the claim may well be true, but the evidence as written does not yet establish the sharp constant of the Chung-type LIL. A full proof of the small-ball transfer, or a formal statement that Theorem 1.3 assumes it, would resolve the issue. No independent error in the Khinchin part was found, so no change to the reader's verdict is warranted.","tokens_in":20333,"tokens_out":16565,"duration_ms":182155,"concrete_test":"Independently derive the analog of [13, Prop. 4.2(1)] for the present harmonizable representation with spatial density |xi|^{ell-d}: exhibit the decomposition u(t,x)=kappa B_H(t)+R(t), compute the small-deviation exponent of R, and verify that the Gaussian correlation between B_H and R is negligible at the epsilon^{1/H} log scale exactly under beta+gamma<2+H. If this derivation cannot be completed, Theorem 1.3 should be restated with Proposition 4.1 as an explicit assumption rather than as a proved step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The single load-bearing concern is Proposition 4.1, the exact small-ball asymptotic used in both halves of Theorem 1.3. Its proof is a short transfer argument: the temporal spectral density is identified with that of kappa B_H, and then it is asserted that the estimates in [13, Prop. 4.2(1) and Thm. 4.3(1)] carry over to |xi|^{ell-d} because they use only homogeneity and Mittag-Leffler estimates. What is not shown is the actual decomposition of u(t,x) into kappa B_H plus a remainder, a bound on the remainder's small-deviation exponent under the full restriction beta+gamma<2+H, and the behavior of the Gaussian correlation between the B_H component and the remainder. For beta+gamma>2 the asserted control reduces to -H_0+beta ell/(2alpha)<1, equivalent to beta+gamma<2+H, but no estimate is displayed. Remark 1.1 itself acknowledges that these conditions are inherited from [13]. If the small-deviation exponent of the remainder is not strictly below 1/H, or if the correlation term contributes at the epsilon^{1/H} log scale, the exact constant kappa lambda_H^H in Theorem 1.3 is unsupported. Theorem 1.2 would survive, since its proof is largely self-contained.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the linear stochastic fractional diffusion equation (1.1) driven by a Gaussian noise that is fractional in time and has Riesz-type spatial covariance. For each fixed spatial point x, the authors prove a Khinchin-type law of the iterated logarithm at time zero (Theorem 1.2, constant \\tilde\\kappa) and, under the additional hypotheses 0≤γ<1 and β+γ<2+H, a Chung-type law of the iterated logarithm (Theorem 1.3, constant κλ_H^H). The proofs rely on self-similarity of the temporal process (Lemma 2.1), sharp frequency-truncation estimates (Proposition 2.2), a covering-number argument for the Khinchin upper bound, a block-decomposition/Borel–Cantelli argument for the Khinchin lower bound, and an exact small-ball estimate (Proposition 4.1) transferred from the authors' earlier result [13, Theorem 4.3(1)].","tokens_in":20612,"tokens_out":9495,"duration_ms":92334,"significance":"If the results are correct, they provide sharp almost-sure initial-time oscillation rates with explicit constants for a broad family of time-fractional stochastic diffusion equations, extending previously known results for stochastic heat equations. The Khinchin part is the stronger contribution: it is self-contained apart from standard Gaussian concentration and covering estimates, and its constant \\tilde\\kappa is an explicit integral involving the Mittag–Leffler function. The Chung part, by contrast, hinges on Proposition 4.1, whose proof is a one-paragraph reduction to an external small-ball theorem; that transfer is not fully demonstrated. The paper uses no fitted parameters and states all constants explicitly, which is a definite strength.","major_comments":[{"comment":"The proof of Proposition 4.1 is a reduction to [13, Theorem 4.3(1)] and does not actually demonstrate the transfer to the noise class (1.4). It asserts that the stationary-increment component has temporal spectral density I_{α,β,ℓ}/(2π)|τ|^{-2H-1}, that the smoother remainder has a strictly smaller small-deviation exponent under β+γ<2+H, and that the Gaussian correlation argument of [13] yields the same small-ball constant. However, no decomposition u(t,x)=κB_H(t)+R(t) is exhibited for the solution of (1.1), no bound on the small-deviation exponent of R is displayed, and the correlation between B_H and R is not controlled at the scale ε^{1/H}\\log ε^{-1}. Since Theorem 1.3 rests entirely on (4.1), the constant κλ_H^H is unsupported unless Proposition 4.1 is proved in detail or the hypotheses of [13, Theorem 4.3(1)] are stated and verified for the Riesz-type density |ξ|^{ℓ-d}.","section":"§4.1, Proposition 4.1 and Theorem 1.3"},{"comment":"Remark 1.1 acknowledges that the conditions 0≤γ<1 and β+γ<2+H are inherited from [13, Theorem 4.3(1)], but the paper does not state the precise hypotheses of that theorem. In particular, it is not specified which spatial covariance class [13] treats and whether the Riesz density |ξ|^{ℓ-d} with ℓ∈(0,2d∧2α) is included. Without this information the reader cannot verify that the reduction in Proposition 4.1 is legitimate, and the claim that the restriction β+γ<2+H is 'needed' remains an assertion. Please either state the referred theorem in full or provide a self-contained proof of (4.1).","section":"§1, Remark 1.1"}],"minor_comments":[{"comment":"The title and abstract contain spacing artifacts ('LA WS', 'ITERA TED', 'A T'); please correct them in the final version.","section":"Title/Abstract"},{"comment":"The constant \\tilde\\kappa is introduced in (1.8) with a tilde, but in several places (e.g., (1.7), Lemma 2.1, and the proof of Theorem 1.2) it is rendered as 'rκ'. Please use a consistent symbol.","section":"Eq. (1.7)–(1.8) and throughout"},{"comment":"In the proof of Proposition 4.1, the phrase 'the temporal Fourier factor is (2π)^{-1}' is terse; a short derivation of the temporal spectral density from (1.4) and (2.11) would help the reader verify the normalization.","section":"§4.1"},{"comment":"Reference [7] is an arXiv preprint from 2026; if a published version exists, please update the citation.","section":"References"},{"comment":"The notation E_{a,b}(-x) is used without stating explicitly that the argument is real and non-positive; a sentence clarifying the real-valued convention would avoid ambiguity.","section":"§2.2, Eq. (2.5)"}],"recommendation":"major_revision","confidential_remarks":"The main load-bearing external result, [13, Theorem 4.3(1)], is co-authored by one of the present authors (R. Wang). This is not by itself a problem, but it strengthens the need for a fully spelled-out verification of Proposition 4.1 rather than an asserted transfer. The Khinchin theorem (Theorem 1.2) appears sound and is the more novel contribution; the Chung theorem is conditional on a repaired Proposition 4.1."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Key thing to know: the paper splits cleanly in two. The Khinchin-type LIL at time zero (Theorem 1.2) is a competent, largely self-contained extension of existing techniques to a broad fractional-diffusion model. The Chung-type LIL (Theorem 1.3) is not at the same standard because its load-bearing input, Proposition 4.1, is asserted rather than proved.\n\nWhat is new and good: the parameter set is genuinely wider than prior work—time-fractional derivative beta in (0,2), fractional time noise H0 in [1/2,1), Riesz spatial covariance, and RL integral gamma >= 0. The proof of Theorem 1.2 uses a harmonizable representation, sharp frequency-truncation estimates, and a block decomposition with independent frequency bands. The covering-number and tail arguments are standard but executed carefully. The self-similarity of the temporal process at a fixed spatial point is established cleanly and used properly. This part deserves credit.\n\nThe soft spot is Theorem 1.3. Proposition 4.1 claims an exact small-ball limit, with constant kappa lambda_H^H, by reducing to [13, Theorem 4.3(1)]. The proof checks the Fourier normalization and identifies the stationary-increment component as kappa B_H. That is fine. But then it says the estimates and Gaussian correlation argument from [13] carry over to the Riesz-type density because they use only homogeneity, the Mittag-Leffler estimate, and the differentiability identity. That is not demonstrated. You need the actual decomposition u = kappa B_H + remainder, a bound on the remainder's small-deviation exponent strictly below 1/H, and control of the correlation term at the epsilon^{1/H} log scale. The condition beta+gamma<2+H is derived in one line, but the underlying estimate is not displayed. Since [13] is co-authored by R. Wang, the citation itself is not a problem; the problem is that the transfer is not shown. If the transfer is invalid, the constant in Theorem 1.3 is unsupported, although Theorem 1.2 survives.\n\nMinor point: Remark 1.1 admits the limitations, which is honest, but a remark is not a substitute for a proof.\n\nOverall: for the SPDE fine-asymptotics community, this is a useful paper with one strong theorem and one conditional one. It deserves a serious referee, but the referee should send it back with a clear request: either prove Proposition 4.1 or state precisely which theorem in [13] covers the Riesz density and why. I would not desk reject it; I would invite a revision.","headline":"The Khinchin half is solid and largely self-contained, but the Chung half rests on an asserted transfer of an exact small-ball theorem and needs referee pressure rather than desk rejection.","tokens_in":21155,"tokens_out":3295,"would_cite":true,"duration_ms":32421,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H15","60G17","60G22"],"pacs":[],"model":"deepseek-v4-flash","headline":"The temporal process of a linear stochastic fractional diffusion equation is shown to satisfy Khinchin and Chung laws of the iterated logarithm at time zero, with explicit almost-sure constants at each fixed spatial point.","keywords":["stochastic fractional diffusion equation","law of the iterated logarithm","Chung's law of the iterated logarithm","small-ball probabilities","self-similar Gaussian process","harmonizable representation","Mittag-Leffler function","temporal regularity"],"falsifier":"Evaluate Proposition 4.1 numerically for the parameter set $(\\beta,\\gamma,H_0,\\alpha,\\ell,d)=(1.5,0.6,0.5,1,1,1)$, which satisfies Condition 1.1, $H=0.85<1$, $0\\le\\gamma<1$, and $\\beta+\\gamma=2.1<2+H$: approximate $\\varepsilon^{1/H}\\log P(\\sup_{0\\le t\\le 1}|u(t,x)|\\le\\varepsilon)$ for small $\\varepsilon$ from many simulations of the harmonizable representation. If the values do not converge to $-\\kappa^{1/H}\\lambda_H$, the transferred small-ball theorem underlying Theorem 1.3 fails; if they do, the theorem's constant is corroborated.","tokens_in":20116,"feed_emoji":"⏱️","tokens_out":11646,"duration_ms":114544,"temperature":0.7,"pith_summary":"The paper proves sharp almost-sure oscillation rates at the initial time for the temporal process $t\\mapsto u(t,x)$ of a broad class of linear stochastic fractional diffusion equations. For each fixed spatial point, Theorem 1.2 identifies the exact limsup constant $\\tilde\\kappa$, and Theorem 1.3 identifies the exact Chung-type liminf constant $\\kappa\\lambda_H^H$, where $H$ is the Hurst index of the Gaussian solution. These statements extend known time-zero laws for stochastic heat equations to time-fractional equations with Mittag-Leffler kernels and Riesz-type spatial noise. The initial time is genuinely different from positive times because the smoother remainder that is negligible away from zero contributes at the same order near $t=0$, so the proof needs a frequency-localization argument. The laws provide sharp benchmarks for nonlinear stochastic heat-type equations near time zero.","feed_headline":"Two exact limit laws found for fractional diffusion at time zero","feed_subtitle":"At each fixed spatial point, the solution's temporal size near zero is pinned by explicit almost-sure constants","key_machinery":"The central object is the harmonizable representation of the solution, a Fourier-integral formula in which the value $u(t,x)$ is a stochastic integral over space-time frequencies and different frequency bands produce independent Gaussian processes. The argument decomposes the temporal process into a low-frequency and a high-frequency part, uses the self-similarity established in Lemma 2.1, and applies sharp frequency-truncation estimates from Lemma 2.2 and Proposition 2.2. For the Chung-type law, the load-bearing mechanism is the exact small-ball asymptotic of Proposition 4.1, transferred from a prior theorem after a Fourier-normalization check, together with a Gaussian comparison inequality and Borel-Cantelli localization over exponentially spaced time scales.","core_discovery":"The paper establishes two almost-sure laws for the temporal process $t\\mapsto u(t,x)$ of the linear stochastic fractional diffusion equation, at the initial time. For each fixed $x$, this process is centered, Gaussian, and self-similar with Hurst index $H$ given by (1.5). Theorem 1.2 states that $\\limsup_{t\\downarrow 0}|u(t,x)|\\big/\\big(t^H\\sqrt{2\\log\\log(1/t)}\\big)=\\tilde\\kappa$ almost surely, and Theorem 1.3 states that, under $0\\le\\gamma<1$ and $\\beta+\\gamma<2+H$, $\\liminf_{\\varepsilon\\downarrow 0}\\sup_{0\\le t\\le\\varepsilon}|u(t,x)|\\big/\\big(\\varepsilon^H(\\log\\log\\varepsilon^{-1})^{-H}\\big)=\\kappa\\lambda_H^H$ almost surely. The central point is that both statements are sharp at time zero, where the smoother remainder of the pinned-string decomposition is no longer negligible; the proof controls that remainder through a harmonizable representation, frequency truncation, and localization.","pith_inferences":["Beyond the paper, the same frequency-block decomposition appears capable of producing exact small-time moduli of continuity for the temporal process, such as statements about $\\sup_{s,t\\in[0,\\varepsilon]}|u(s)-u(t)|$, since the block estimates in Proposition 2.2 already control uniform fluctuations.","Beyond the paper, the Khinchin constant $\\tilde\\kappa$ is an explicit integral of Mittag-Leffler functions and could be evaluated numerically; Monte Carlo simulation of the harmonizable representation could independently test the value before relying on the small-ball transfer.","Beyond the paper, the boundary cases $H=1$ and $\\gamma\\ge 1$ remain open; if the exact small-ball asymptotic were extended to those regimes, the Chung-type constant would likely take a different form because the balance between the fractional-Brownian component and the smoother remainder changes.","Beyond the paper, for parameter ranges where $\\beta+\\gamma<2$, the remainder is automatically lower order and the Chung constant should be consistent with the stochastic-heat-equation limits already known for that family, providing a consistency check across parameter regimes."],"forward_implications":["For every fixed $x$, the pointwise temporal process satisfies $|u(t,x)|\\sim\\tilde\\kappa\\,t^H\\sqrt{2\\log\\log(1/t)}$ in the limsup sense as $t\\downarrow 0$, so the lowest-order growth at time zero is exactly of fractional-Brownian scale.","Under the extra conditions $0\\le\\gamma<1$ and $\\beta+\\gamma<2+H$, the running supremum over $[0,\\varepsilon]$ is almost surely asymptotic to $\\kappa\\lambda_H^H\\varepsilon^H(\\log\\log\\varepsilon^{-1})^{-H}$, pinning the deepest small-time trough of the process.","The temporal process is self-similar with Hurst index $H$ at every fixed spatial point, and the almost-sure constants do not depend on $x$.","The Khinchin-type statement holds for all parameters in Condition 1.1 with $H<1$; the additional restrictions are needed only for the Chung-type statement.","These laws offer sharp initial-time benchmarks for nonlinear stochastic heat-type equations, where temporal increments are often compared with the linearized solution near $t=0$."],"supporting_citations":[{"why":"Supplies the exact small-ball asymptotic that Proposition 4.1 transfers to the Riesz-density setting; it sets the constant in Theorem 1.3.","marker":"[13, Theorem 4.3(1)]"},{"why":"Gives existence and uniqueness of the random-field solution, justifying the stochastic-convolution representation used throughout.","marker":"[7, Theorem 1.1(i)]"},{"why":"Provides the finite constant $\\tilde\\kappa$ and the $L^2$ scaling $\\|u(t,x)\\|_2=\\tilde\\kappa t^H$ that underpin self-similarity and the Khinchin constant.","marker":"[7, Theorem 4.1(i)]"},{"why":"The Khinchin-type LIL at time zero for related processes; the proof of Theorem 1.2 follows its frequency-block strategy.","marker":"[29, Proposition 5.1(a)]"},{"why":"The general Chung-type LIL framework via harmonizable representations and exact moduli of continuity; it drives the upper-bound argument in Theorem 1.3.","marker":"[22]"},{"why":"Supplies small-ball constants and the localization method for the heat-equation Chung LIL, providing the decomposition template used here.","marker":"[19]"},{"why":"The Chung-type LIL for the linear stochastic fractional heat equation at the origin, which Theorem 1.3 extends.","marker":"[25]"},{"why":"Defines the fractional Brownian small-ball constant $\\lambda_H$ used in the statement of Theorem 1.3.","marker":"[24]"},{"why":"Provides the Gaussian covering and tail bound used in Lemma 3.1 for the Khinchin upper bound.","marker":"[30]"},{"why":"An Anderson-type comparison inequality used to transfer small-ball probabilities from $u$ to the frequency-block component in the Chung upper bound.","marker":"[11, Lemma 2.8]"}],"fun_headline_variants":["Khinchin and Chung laws established for fractional diffusion at time zero","Exact LILs at time zero for stochastic fractional diffusion","Time-zero Khinchin and Chung laws for fractional diffusion","Fractional diffusion: exact iterated logarithm laws at zero"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The Chung-type law's constant rests on a small-ball estimate taken from a prior theorem and assumed, after a Fourier-normalization check, to remain valid for the Riesz spectral density $|\\xi|^{\\ell-d}$ with the added restriction $\\beta+\\gamma<2+H$; if that transfer is wrong, Theorem 1.3 loses its constant while Theorem 1.2 would survive.","fun_headline_variants_meta":{"raw":{"variants":["Khinchin and Chung laws established for fractional diffusion at time zero","Exact LILs at time zero for stochastic fractional diffusion","Time-zero Khinchin and Chung laws for fractional diffusion","Fractional diffusion: exact iterated logarithm laws at zero"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001281,"raw_usage":{"total_tokens":5265,"prompt_tokens":1002,"completion_tokens":4263,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":618,"completion_tokens_details":{"reasoning_tokens":4193}},"tokens_in":618,"tokens_out":4263,"duration_ms":31237,"temperature":1.0,"reasoning_tokens":4193,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:55:57.295079+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate Proposition 4.1 numerically for the parameter set $(\\beta,\\gamma,H_0,\\alpha,\\ell,d)=(1.5,0.6,0.5,1,1,1)$, which satisfies Condition 1.1, $H=0.85<1$, $0\\le\\gamma<1$, and $\\beta+\\gamma=2.1<2+H$: approximate $\\varepsilon^{1/H}\\log P(\\sup_{0\\le t\\le 1}|u(t,x)|\\le\\varepsilon)$ for small $\\varepsilon$ from many simulations of the harmonizable representation. If the values do not converge to $-\\kappa^{1/H}\\lambda_H$, the transferred small-ball theorem underlying Theorem 1.3 fails; if they do, the theorem's constant is corroborated.","supporting_citations":[{"cited_title":"Lee and Y","cited_arxiv_id":null,"evidence_quote":"The general Chung-type LIL framework via harmonizable representations and exact moduli of continuity; it drives the upper-bound argument in Theorem 1.3."},{"cited_title":"Khoshnevisan, K","cited_arxiv_id":null,"evidence_quote":"Supplies small-ball constants and the localization method for the heat-equation Chung LIL, providing the decomposition template used here."},{"cited_title":"Liu and R","cited_arxiv_id":null,"evidence_quote":"The Chung-type LIL for the linear stochastic fractional heat equation at the origin, which Theorem 1.3 extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the fractional Brownian small-ball constant $\\lambda_H$ used in the statement of Theorem 1.3."},{"cited_title":"Talagrand, Sharper bounds for Gaussian and empirical processes,Ann","cited_arxiv_id":null,"evidence_quote":"Provides the Gaussian covering and tail bound used in Lemma 3.1 for the Khinchin upper bound."}],"review_version":1}