{"id":"8ea14926-7e17-4dc4-9702-0abcb24b2cfb","arxiv_id":"2607.20956","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Exact uniform and local moduli of continuity are established for Gaussian fields and for a critical linear SPDE, using pairwise correlation bounds in place of strong local nondeterminism.","lead":"Exact fluctuation rates for Gaussian random fields are proved using pairwise correlation bounds instead of the usual strong local nondeterminism assumption. The results settle an open problem on the sharp spatial modulus of continuity for a critical class of linear stochastic PDEs.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the SPDE decorrelation/localization argument survives scaling checks; only the non-central reliance on preprint [7] for temporal moduli remains.","rationale":"The reader correctly identified condition (1.4)/Assumption 2.3 and the localization Proposition 7.4 as the load-bearing step for the central SPDE result. My stress-test checked that step in detail and found the scaling works: Lemma 7.2 produces the required ρ log(1/h) factor after δ=h^{ρα}, Lemma 7.3 gives a constant h^2 term that is absorbed by the log factor for small h, and the subsequent covariance estimate has the needed √ρ factor. Thus the lower bound in Theorem 1.1/2.4 is supported. The reader's additional concerns about [7] are legitimate for the temporal moduli, but those are secondary to the paper's central claim, which is the spatial modulus in the critical case where SLND is unavailable. I therefore see no reason to strengthen the verdict; the conditional recommendation remains appropriate because of the external dependency and terse estimates, but no fatal or load-bearing flaw was identified in the central argument.","tokens_in":32905,"tokens_out":34939,"duration_ms":293512,"concrete_test":"Evaluate the exact L2 localization error for the fractional heat kernel with a standard numerical quadrature, for α∈{1.6,1.8,2.0}, δ=h^{ρα}, ε=h^ρ, ρ∈{0.2,0.5,0.8}, h∈{10^{-2},10^{-4},10^{-6}}, and verify numerically that ∥∆^h_{B1}u∥^2+∥∆^h_{B2}u∥^2 ≤ Kρ h^2 log(1/h) for h below a ρ-dependent threshold, confirming the scaling used to verify condition (1.4).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central new claim is Theorem 1.5(1.10), whose proof hinges on verifying condition (1.4) via Proposition 7.4. I checked the critical algebra. With δ=h^{ρα}, ε=h^ρ, Lemma 7.2 gives ∥∆^h_{B1}u∥^2 ≤ K h^2(ρ log(1/h)+1+e^{-δ/h^α}) and Lemma 7.3 gives ∥∆^h_{B2}u∥^2 ≤ K h^2. For h≤r_ρ the h^2 term is absorbed into ρ h^2 log(1/h), so the total localization error is ≤ Kρ h^2 log(1/h). The covariance decomposition then yields a factor √ρ after Cauchy–Schwarz, which can be made ≤ √(1−C0) C1^2 by choosing ρ small, exactly as (1.4) requires. The packing and Slepian lower bound in Theorem 2.4 go through with this verified condition. No internal inconsistency was found. The only outstanding dependency is the temporal SLND statement imported from the co-authored preprint [7], needed for (1.11), (1.14), and hence for K5>0 in (1.12); this does not touch the main spatial critical-case result (1.10) or the local spatial result (1.13).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a general framework for exact uniform and local moduli of continuity of Gaussian random fields, replacing the usual strong local nondeterminism (SLND) assumption by correlation bounds on pairwise increments (Assumptions 2.3 and 2.6). The main general results are Theorems 2.4 and 2.7 for anisotropic fields, with scalar corollaries Theorems 1.1 and 1.2. The central application is to the linear SPDE (1.9) in the critical case 2Hα=3, where the solution is spatially C^{1-} and SLND is not available. The authors prove sharp spatial, temporal, and joint moduli of continuity (Theorem 1.5) and sharp local moduli (Theorem 1.6). The key new ingredient is a set of localization estimates for spatial increments of the solution (Proposition 7.4), which yield decorrelation of sufficiently separated increments (Proposition 7.5). A short final section applies the framework to Gaussian Volterra processes with slowly varying variance functions.","tokens_in":33147,"tokens_out":37542,"duration_ms":313384,"significance":"If the main results are correct, this is a substantial contribution. The framework provides a route to sharp moduli of continuity that does not require SLND, stationarity, or spectral representations, and it solves an open problem for the critical SPDE case 2Hα=3. The localization and decorrelation estimates in Section 7 are likely to be of independent interest, and the sharp limit statements (1.10) and (1.13) are crisp falsifiable predictions. The proofs are detailed and self-contained for the central spatial claims; constants are not fitted but are shown to be positive almost-sure limits. The dependence of the temporal parts on the preprint [7] is a caveat, but it does not affect the main spatial critical-case result.","major_comments":[],"minor_comments":[{"comment":"The displayed lower bound for K5 says the temporal restriction gives a limit equal to αK4, but the denominator reads |t-s|^{1/α}√log(1/|t-s|^{1/α}) = α^{-1/2}|t-s|^{1/α}√log(1/|t-s|), so the limit is α^{1/2}K4, not αK4. Positivity is unaffected, but the displayed equality should be corrected.","section":"§7.2, proof of (1.12)"},{"comment":"The step Var(X(t_n)-X(t_0) | X(t_m)-X(t_0)) ≥ φ²(t_n-t_m) is justified by referring to one-sided SLND, but Lemma 8.2(ii) applies only when the conditioning points precede the evaluation point t_n; here t_m > t_n. The statement can be repaired by applying Lemma 8.2(ii) to X(t_m) conditional on X(t_n), X(t_0) and then using the Gaussian identity Var(Y | Y+W) = Var(Y) Var(W | Y)/Var(Y+W) with Y = X(t_n)-X(t_0) and W = X(t_m)-X(t_n); this gives the claimed lower bound up to an absolute constant. As written, the proof is incomplete.","section":"§8, Lemma 8.5(iii)"},{"comment":"The proofs of (1.11), (1.14), and hence the lower bounds for K5 and K8, invoke Theorem 1.3 of the unpublished e-print [7] for temporal SLND. This does not affect the spatial results (1.10) or (1.13), but the authors should either provide a proof of the needed temporal SLND statement or confirm that [7] is publicly available and acceptable for citation.","section":"§7.2 and §7.3"},{"comment":"In the scaling argument after the Fourier representation, the text says only that ξ is changed to δ^{-1/α}ξ, but the displayed integral also requires the simultaneous scaling τ ↦ δ^{-1}τ in order for the denominator to become |τ|²+|ξ|^{2α} and for the exponential factor to become exp(-(iτ+|ξ|^α)(t-δ)/δ). The final estimates are correct, but the presentation should state both scalings.","section":"§7.1, Lemma 7.2 proof"},{"comment":"The phrase 'C^{1-} in space' in the abstract and introduction is informal; since the paper elsewhere uses θ_2^- notation for Hölder exponents, it would be clearer to say that the spatial sample paths are Hölder continuous of every order below 1.","section":"§1 and §2, notation"}],"recommendation":"minor_revision","confidential_remarks":"The reliance on the e-print [7] for temporal SLND deserves attention: the reference shares a co-author with this paper, so the editor may wish to check its status and whether the temporal claims could be made independent. This does not affect the central spatial critical-case result, which appears sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nTwo things to know before you read this. First, the paper kills the open problem for the spatial modulus in the critical SPDE case (2Hα=3, d=1), where SLND is unavailable. Second, the general framework really is new: conditions (1.4) and (1.7) are pairwise correlation bounds, not SLND, and they suffice for exact uniform and local moduli. That is a genuine advance over Lee–Xiao and Meerschaert–Wang–Xiao.\n\nWhat is actually new: the correlation-bound framework (Theorems 2.4, 2.7), the localization estimates (Prop. 7.4) that give spatial decorrelation (Prop. 7.5), and the local spatial modulus (1.13) with its triple-log factor. That last result uses a harmonizable representation, a carefully chosen double-exponential sequence, and an approximate derivative; I have not seen that combination before. The constants K are shown to be finite and positive, not fitted, which is exactly right for this genre. I checked the critical algebra in Prop. 7.4: with δ=h^{ρα}, ε=h^ρ, the localization error is controlled by √ρ after Cauchy–Schwarz, so condition (1.4) can indeed be forced. The Slepian lower bound and the packing argument go through. No circularity: the central spatial decorrelation is proved in this paper.\n\nSoft spots, in proportion. The temporal moduli (1.11), (1.14), and the K5>0 part of (1.12) depend on the time process satisfying SLND, and that is imported from a co-authored preprint [7] whose proof is not included. That is not a flaw in the spatial results, but it makes those parts conditional. I would want the authors to either include the preprint's proof or state clearly that the temporal claims are joint-work results. Second, some estimates are terse: Lemma 7.6's angular integrals are written with a string of ≲ steps, and the right of the annulus cutoffs is left to the reader. Not a fatal gap, but a referee should ask for expansion there.\n\nWho this is for: anyone working on Gaussian sample paths or SPDE regularity. It deserves serious peer review, not a desk reject. My recommendation: send it to a specialist referee, ask them to verify the preprint dependency and expand Lemma 7.6, and the spatial core will stand.","headline":"Strong new framework proves the critical SPDE spatial modulus that SLND couldn't reach; the only real caveat is temporal results leaning on an unpublished preprint.","tokens_in":33718,"tokens_out":1635,"would_cite":true,"duration_ms":16675,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G15","60G60","60G17","60H15"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves exact moduli of continuity for Gaussian fields from correlation bounds alone, and settles the spatial modulus of a critical linear SPDE at $2H\\alpha=3$.","keywords":["modulus of continuity","law of the iterated logarithm","Gaussian random fields","stochastic PDEs","correlation bounds","localization estimates","decorrelation","local nondeterminism"],"falsifier":"For the SPDE, compute directly the covariance of $\\Delta_h u(t_0,x)$ and $\\Delta_h u(t_0,x')$ at separation $|x-x'|=2r^\\rho$ using the spectral formula (7.18); if for some $\\rho<1$ this covariance exceeds $\\sqrt{1-C_0}\\,\\sigma(h)^2$ for a fixed $C_0>0$ as $h\\to0$, then the decorrelation Proposition 7.5 fails and the positivity of $K_3$ would be in doubt. For the general theorem, construct a Gaussian field satisfying Assumptions 2.1 and 2.2 but with increments at separation $\\psi^{-1}(A_0(\\psi(h))^\\rho)$ having covariance bounded only by $\\sqrt{1-C_0+\\delta}$ for a fixed $\\delta>0$; then the lower-bound argument in Theorem 2.4 gives $C\\ge C_1[2\\rho(1-\\sqrt{\\delta})]^{1/2}$, and for $\\delta$ close to $1$ this can fail to be positive even though the upper bound holds.","tokens_in":32690,"feed_emoji":"🎲","tokens_out":9339,"duration_ms":72981,"temperature":0.7,"pith_summary":"This paper develops a general method for exact moduli of continuity of Gaussian random fields, requiring only variance two-sided bounds and a correlation bound on pairwise increments instead of strong local nondeterminism, stationarity, or spectral representations. It claims that, under these bounds, every anisotropic Gaussian field has an exact uniform modulus of continuity of the form $\\phi(t-s)\\sqrt{\\log[1/\\psi(\\phi(t-s))]}$ with a finite positive constant, and an analogous local modulus with $\\ell(1/\\psi(\\phi(t-s)))$ for $\\ell=\\log\\log$ or $\\ell=\\log$. As an application, it solves an open problem for linear stochastic PDEs in the critical case $2H\\alpha=3$, where the solution is $C^{1-}$ in space and strong local nondeterminism is unavailable, proving that the spatial uniform modulus is $|x-y|\\log(1/|x-y|)$ up to a constant and that the local spatial modulus carries an extra $\\sqrt{\\log\\log\\log(1/|x-x_0|)}$ factor. The proofs rest on new localization estimates for the SPDE increments that force decorrelation at small power-of-$h$ separations.","feed_headline":"Exact spatial modulus settled for critical stochastic PDE","feed_subtitle":"New correlation bounds bypass strong local nondeterminism and pin down the open 2Hα=3 case.","key_machinery":"The central object is the scale function $\\psi(r)=\\prod_{i=1}^N\\phi_i^{-1}(r)$, which measures the 'number of increments' at scale $r$ through metric entropy; the key hypothesis is a correlation bound on pairwise increments, equivalently $\\mathrm{Var}(X(t)-X(s)\\mid X(t')-X(s'))\\ge C_0\\,\\mathrm{Var}(X(t)-X(s))$ whenever the two increments are separated by at least $\\psi^{-1}(A_0(\\psi(h))^\\rho)$ (Assumption 2.3, Lemma 4.4). This correlation bound lets Slepian's lemma compare the normalized increments with a one-factor Gaussian model $Z_i=(1-C_0)^{1/4}\\xi_0+(1-\\sqrt{1-C_0})^{1/2}\\xi_i$, producing the lower bound on the modulus constant from the Gaussian tail of $\\max_i\\xi_i$; the upper bound comes from a Dudley-type entropy integral over $\\psi$. For the SPDE, the same correlation condition is verified by localizing each spatial increment to the parabolic window $[t_0-h^{\\rho\\alpha},t_0]\\times[x-h^\\rho,x+h^\\rho]$; Proposition 7.4 shows the localization error is at most $K\\rho\\,h^2\\log(1/h)$ using the fractional heat-kernel gradient bound $|\\partial_yG(s,y)|\\lesssim s|y|^{-2-\\alpha}$, and Proposition 7.5 converts this into decorrelation of increments separated by $2r^\\rho$.","core_discovery":"For a centered Gaussian field $X$ on a compact rectangle, suppose $\\|X(t)-X(s)\\|_2\\asymp\\phi(t-s)$ with $\\phi_i$ regularly varying of index $\\theta_i\\in[0,1]$, and suppose increments separated by $\\psi^{-1}(A_0(\\psi(h))^\\rho)$ are effectively independent in the sense that their covariance is at most $\\sqrt{1-C_0}$ times the product of their $\\mathrm{L}^2$ norms (Assumption 2.3, equivalent to a conditional-variance lower bound). Then Theorem 2.4 gives $\\lim_{h\\to0+}\\sup_{t,s:0<\\phi(t-s)\\le h}|X(t)-X(s)|/(\\phi(t-s)\\sqrt{\\log[1/\\psi(\\phi(t-s))]})=C$ almost surely for a finite positive constant $C$, with matching upper and lower bounds in terms of $C_0,\\rho$ and the metric-entropy constants. For the SPDE $\\partial_t u=-(-\\Delta)^{\\alpha/2}u+\\dot W$ with $\\dot W$ fractional in time of Hurst index $H$ and white in space, in the critical regime $\\alpha\\in(3/2,2], 2H\\alpha=3$ on $\\mathbb{R}$, Theorem 1.5 proves $\\lim_{h\\to0+}\\sup_{x,y\\in J:0<|x-y|\\le h}|u(t_0,x)-u(t_0,y)|/(|x-y|\\log(1/|x-y|))=K_3$ a.s., and Theorem 1.6 proves the local modulus with denominator $|x-x_0|\\sqrt{\\log(1/|x-x_0|)\\log\\log\\log(1/|x-x_0|)}=K_6$.","pith_inferences":["The same correlation-bound route could identify exact phase transitions for other SPDEs where SLND is open, such as stochastic wave equations or fractional-noise equations in higher dimensions, by verifying that the localization error is proportional to $\\rho$.","The need for the doubly-exponential scale $h_n=\\exp(-e^{n^\\gamma})$ in the local spatial modulus suggests that fixed-point spatial increments of the critical SPDE decorrelate only after enormous rescaling; this may reflect an 'approximate derivative' whose variance grows like $\\sqrt{n\\log\\log n}$, connecting to the harmonizable derivative $D(n)$.","For non-critical cases $\\theta_2\\in(0,1)$ with SLND, the same framework should recover the same constants, suggesting the correlation-bound condition is the 'right' hypothesis and SLND is only a sufficient tool.","Testing the covariance numerically at separation $h^\\rho$ for small $\\rho$ would provide a direct verification of Proposition 7.5, and could suggest the optimal $\\rho(\\varepsilon)$ tradeoff."],"forward_implications":["The uniform spatial modulus of the critical SPDE solution at a fixed time is exactly $K_3|x-y|\\log(1/|x-y|)$ for some finite positive $K_3$, so the solution's spatial sample paths are as rough as the borderline between continuous and differentiable at every scale.","At a fixed space-time point, spatial increments satisfy a local law of the iterated logarithm with denominator $|x-x_0|\\sqrt{\\log(1/|x-x_0|)\\log\\log\\log(1/|x-x_0|)}$, a rate not captured by earlier methods.","Temporal and joint moduli follow: temporal increments at fixed $x_0$ have modulus $|t-s|^{1/\\alpha}\\sqrt{\\log(1/|t-s|)}$ uniformly and $\\sqrt{\\log\\log}$ locally, and the joint field has modulus $\\phi(z,z')\\sqrt{\\log(1/\\phi(z,z'))}$.","Any Gaussian field whose increments satisfy the correlation bound has an exact modulus constant that is finite and positive; the framework therefore gives the same conclusions as SLND-based results whenever SLND holds.","Gaussian Volterra processes with slowly varying variance function $\\phi(r)=L(\\log(1/r))/(\\log(1/r))^a$, $a>1/2$, have the same uniform and local modulus functions, with a common constant."],"supporting_citations":[{"why":"Supplies the SLND-based framework the present correlation-bound conditions extend, and the known SPDE moduli in the subcritical case.","marker":"[29]"},{"why":"Provides the variance asymptotics $\\|u(t_0,x+h)-u(t_0,x)\\|_2\\asymp h\\sqrt{\\log_+(1/h)}$ and documents the open SLND/modulus problem the paper resolves.","marker":"[16]"},{"why":"Supplies the temporal SLND used for the temporal moduli (1.11) and (1.14), and records the open question of spatial SLND.","marker":"[7]"},{"why":"Gives the localization of spatial increments to a small parabolic window whose error estimates the paper adapts and sharpens.","marker":"[14]"},{"why":"Gives the gradient bound $|\\partial_yG(s,y)|\\lesssim s|y|^{-2-\\alpha}$ that controls the spatial localization error.","marker":"[8]"},{"why":"Supplies the zero-one law (Lemma 4.2), Slepian's lemma (Lemma 4.3), and the m-modulus equivalence (Lemma 4.1) used in both main proofs.","marker":"[35]"},{"why":"Provides the entropy upper bound for Gaussian suprema used to bound the modulus constant from above.","marker":"[12]"},{"why":"Gives the Gaussian concentration bound used to control tails in the local modulus proof.","marker":"[6]"},{"why":"Supplies the covering-number bound $N(\\varepsilon)\\le c_0|I|/\\psi(\\varepsilon/C_2)$ used in the entropy estimate for the upper bound.","marker":"[1]"}],"fun_headline_variants":["Correlation bounds settle open SPDE modulus problem","Sharp spatial modulus for critical SPDE without SLND","Exact modulus for critical SPDE, no SLND required","New correlation bounds give sharp SPDE spatial modulus","Bypassing SLND: sharp spatial modulus for critical SPDE"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the correlation bound that increments at distance $h$ become nearly independent as soon as their starting points are separated by a small power of $h$; if that bound fails or the localization error in the SPDE is not proportional to $\\rho$, the lower bound on the modulus constant collapses.","fun_headline_variants_meta":{"raw":{"variants":["Correlation bounds settle open SPDE modulus problem","Sharp spatial modulus for critical SPDE without SLND","Exact modulus for critical SPDE, no SLND required","New correlation bounds give sharp SPDE spatial modulus","Bypassing SLND: sharp spatial modulus for critical SPDE"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000902,"raw_usage":{"total_tokens":3926,"prompt_tokens":1036,"completion_tokens":2890,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":652,"completion_tokens_details":{"reasoning_tokens":2811}},"tokens_in":652,"tokens_out":2890,"duration_ms":15918,"temperature":1.0,"reasoning_tokens":2811,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:34:23.185651+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the SPDE, compute directly the covariance of $\\Delta_h u(t_0,x)$ and $\\Delta_h u(t_0,x')$ at separation $|x-x'|=2r^\\rho$ using the spectral formula (7.18); if for some $\\rho<1$ this covariance exceeds $\\sqrt{1-C_0}\\,\\sigma(h)^2$ for a fixed $C_0>0$ as $h\\to0$, then the decorrelation Proposition 7.5 fails and the positivity of $K_3$ would be in doubt. For the general theorem, construct a Gaussian field satisfying Assumptions 2.1 and 2.2 but with increments at separation $\\psi^{-1}(A_0(\\psi(h))^\\rho)$ having covariance bounded only by $\\sqrt{1-C_0+\\delta}$ for a fixed $\\delta>0$; then the lower-bound argument in Theorem 2.4 gives $C\\ge C_1[2\\rho(1-\\sqrt{\\delta})]^{1/2}$, and for $\\delta$ close to $1$ this can fail to be positive even though the upper bound holds.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the SLND-based framework the present correlation-bound conditions extend, and the known SPDE moduli in the subcritical case."},{"cited_title":"Herrell, R","cited_arxiv_id":null,"evidence_quote":"Provides the variance asymptotics $\\|u(t_0,x+h)-u(t_0,x)\\|_2\\asymp h\\sqrt{\\log_+(1/h)}$ and documents the open SLND/modulus problem the paper resolves."},{"cited_title":"Foondun, D","cited_arxiv_id":null,"evidence_quote":"Gives the localization of spatial increments to a small parabolic window whose error estimates the paper adapts and sharpens."},{"cited_title":"Chen and X","cited_arxiv_id":null,"evidence_quote":"Gives the gradient bound $|\\partial_yG(s,y)|\\lesssim s|y|^{-2-\\alpha}$ that controls the spatial localization error."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the zero-one law (Lemma 4.2), Slepian's lemma (Lemma 4.3), and the m-modulus equivalence (Lemma 4.1) used in both main proofs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the entropy upper bound for Gaussian suprema used to bound the modulus constant from above."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Gaussian concentration bound used to control tails in the local modulus proof."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the covering-number bound $N(\\varepsilon)\\le c_0|I|/\\psi(\\varepsilon/C_2)$ used in the entropy estimate for the upper bound."}],"review_version":2}