Pith. sign in

REVIEW 1 major objections 5 minor 42 references

Nonparametric Inference for Noise Covariance Kernels in Parabolic SPDEs using Space-Time Infill-Asymptotics

T0 review · 1 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A discretized infinite-dimensional realized covariation estimates the full noise covariance of a linear parabolic SPDE without knowing the elliptic operator, and its CLT supports omnibus goodness-of-fit tests.

desk verdict Solid extension of realized covariation to discrete sampling and unknown A, with one real but fixable gap in the local-average CLT. read the letter →

arxiv 2508.20947 v1 pith:43IZFHIY submitted 2025-08-28 math.ST stat.TH

Andreas Petersson , Dennis Schroers This is my paper · ORCID
classification math.STstat.TH MSC 60H1562M2062G0562G1035R60
keywords stochasticpartialdifferentialequationsnoisecovarianceestimationrealizedcovariationgoodness-of-fittestspace-timeinfillasymptoticsfunctionaldataanalysisHilbert-SchmidtnormMatérn
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that for a linear parabolic SPDE perturbed by spatially colored noise, the full spatial covariance operator Q of the noise can be estimated nonparametrically from discrete space-time observations, even when the differential operator A is unknown. The estimator is the normalized infinite-dimensional realized covariation RV_T^{Δ,O}/T, a sum of tensor squares of observed increments. Under an analytic-semigroup assumption on A, this estimator is consistent for Q in Hilbert–Schmidt norm, and after scaling by Δ^{-1/2} it is asymptotically normal with covariance operator ΓB = Q(B + B*)Q, for continuous observations, local spatial averages, and pointwise sampling. From this CLT the paper builds omnibus goodness-of-fit tests for H0: Q = Q0 and for parametric families such as Matérn and inverse-fractional-Laplacian covariances. A sympathetic reader should care because misspecified noise models in spatio-temporal SPDE applications propagate into prediction error, and prior global estimation of Q required knowing A and having continuous spatial observations.

What carries the argument

The central object is the infinite-dimensional realized covariation RV_T^{Δ,O} = Σ_i (OX_{iΔ} − OX_{(i−1)Δ})^{⊗2}, normalized by T, where ⊗ denotes the Hilbert-space tensor square and O is the spatial observation operator: I for continuous data, the L^2 projection P_h for local averages, and the piecewise-linear interpolant I_h for pointwise data. The estimator is compared with a semigroup-adjusted version SARCV_T^{Δ,O} = Σ_i (OX_{iΔ} − OS(Δ)X_{(i−1)Δ})^{⊗2}; the difference is shown to be negligible, so the limit theory of the latter transfers. The analytic semigroup S(t) = e^{−tA} supplies the smoothing estimates that force the bias term involving the initial value to vanish. The limiting c

What would settle it

Take the same SPDE but replace A by a non-analytic semigroup generator, such as a hyperbolic wave-type operator with imaginary spectrum, keep the same trace-class Q and X0 = 0, and compute the Hilbert–Schmidt distance between RV_T^{Δ,O}/T and Q as Δ, h → 0; if it does not converge to 0, the identifiability claim fails outside the analytic class. Within the assumptions, a Monte-Carlo check of the CLT for A = −Δ_D, Q = (−Δ_D)^{−σ}, O = I_h with h = O(Δ^{3/8}) in d = 3, comparing the empirical covariance of Δ^{−1/2}(RV/T − Q) against Γ = Q(· + ·*)Q, would falsify the normality claim if the match

Watch

Extended reading notes

Core claim

Formally, the paper proves that under Assumption 1, RV_T^{Δ,I}/T converges to Q in L^2(Ω, L2(H)) as Δ↓0, giving identifiability of the whole kernel. Under Assumption 2 with 2ι ≥ γ ≥ 1, for each of the three observation schemes—identity (continuous spatial data), orthogonal projection onto piecewise constants (local averages), and piecewise linear interpolation (pointwise data)—one has Δ^{-1/2}(RV_T^{Δ,O}/T − Q) converging in distribution to N(0, Γ) in the Hilbert–Schmidt norm, with ΓB = Q(B + B*)Q. The discrete spatial schemes require coupling the mesh size h to the time step Δ: h = o(Δ^{1/2}) for local averages, and h = O(Δ^{3/(2d+2)}) for d ≤ 2 or O(Δ^{3/8}) for d = 3 for pointwise samplin

Load-bearing premise

The load-bearing premise is that the differential operator A is self-adjoint, positive definite with compact inverse, so that its semigroup is analytic and smooths the initial state exponentially; without that smoothing, the realized covariation need not converge to Q.

Editorial extensions

If this is right

  • For any linear parabolic SPDE satisfying the assumptions, the full noise covariance kernel can be estimated consistently from space-time data, so noise-model validation can be separated from estimation of the differential operator.
  • The CLT gives asymptotic normality at rate √Δ when the smoothness exponent γ ≥ 1, matching the canonical finite-dimensional semimartingale rate, and spatial resolution may be coarser than temporal resolution.
  • An omnibus goodness-of-fit test for a fixed hypothesized kernel Q0 is asymptotically of exact level α, with a computable generalized-χ² null distribution.
  • A conservative test for parametric families such as Matérn or inverse-fractional-Laplacian covariances is asymptotically of level at most α and exhibits high power against qualitatively different kernel families.
  • When the regularity conditions fail, such as low Matérn smoothness, the empirical size of the test drifts away from α, confirming the conditions are needed and not merely technical.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: Because Γ is built from Q alone, the same CLT yields asymptotic confidence ellipsoids for the kernel by plugging the estimated Q into Γ; the paper constructs tests, but the covariance formula is ready-made for interval construction.
  • Editorial inference: The proof's reliance on polynomial-approximation bounds suggests the estimator extends to other observation schemes, such as weighted averages, finite-element projections, or irregular grids, with rates set by ∥O−I∥ and the same CLT whenever the semigroup remains analytic.
  • Editorial inference: The paper notes the bias does not vanish as T→∞ for fixed Δ; a natural next step, implicit in Remark 4.8, is to estimate and subtract the bias using long time spans, which would extend the √Δ CLT to settings where bias currently dominates.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 5 minor

Summary. The paper develops asymptotic inference for the trace-class spatial covariance operator Q of the additive noise in a linear parabolic SPDE, based on realized covariations of the solution under space-time infill asymptotics. The central results are: (i) consistency of RV_T^{Δ,O}/T for Q in Hilbert-Schmidt norm under a general self-adjoint elliptic operator A, with A unknown; (ii) convergence rates for three sampling schemes (continuous observation, local averages, pointwise interpolation) depending on regularity parameters; (iii) a CLT of the form Δ^{-1/2}(RV_T^{Δ,O}/T - Q) ⇒ N(0,Γ) with ΓB = Q(B+B*)Q under Assumption 2 with 2ι ≥ γ ≥ 1; and (iv) omnibus goodness-of-fit tests for fixed and parametric covariance hypotheses. The proofs are detailed in an appendix, and a simulation study examines rates, size, and power.

Significance. If the stated results hold, this is a substantial contribution to statistical inference for parabolic SPDEs. The main novelty—estimating the full noise covariance operator nonparametrically from realized covariations without knowing A—is of real practical value, and the paper backs it with explicit rates, a CLT whose covariance depends only on Q, a lower bound showing near-optimality, and a careful treatment of three sampling designs. The simulation study is honest, explicitly showing size distortion when the regularity conditions of the CLT fail. The proofs are detailed and the assumptions are illustrated with natural examples (Matérn kernels, fractional inverse Laplacians). These strengths make the manuscript worth publishing after revision.

major comments (1)
  1. [Theorem 4.13(B) and Lemma B.9] The proof of part (B) of Theorem 4.13 relies on Lemma B.9 to show that the local-average semigroup-adjusted covariation is asymptotically equivalent to the identity-observation version (term IV in the proof of Theorem 4.13). Lemma B.9, however, is proved only for nested subdivisions (T_h): the proof uses monotone convergence of ∥(I−P_h)v∥ and Dini's theorem to obtain sup_{r≤1}∥(P_h−I)S(r)Q^{1/2}∥_{L2(H)}→0. Theorem 4.13(B) and Section 3(B) only assume a non-degenerate family of subdivisions and h=o(Δ^{1/2}); no nestedness is stated. Without nestedness (or a different direct compactness argument), the uniform convergence is not established, and the displayed o_p(T^{1/2}Δ^{1/2}) bound for the difference between the P_h and I realized covariations does not follow. Since part (B) is one of the three sampling schemes in the paper's central CLT, this is a load-bearing gap. It is fixable by add
minor comments (5)
  1. [Lemma B.9, second bullet] The displayed condition for O=I_h, 'h = o(Δ^{(tilde-epsilon-1)/(2 min(psi,2))})', does not match the condition used in Theorem 4.13(C), and appears to contain a typo in the exponent. Please check and harmonize the statement with the proof and with Theorem 4.13(C).
  2. [Theorem 5.4 proof] In the proof, the line 'T(θ0)=0 < T(θ0)' should read '0 < T(θ)' for θ ≠ θ0.
  3. [Theorem 4.9(C)] There is a typo: 'Assumption 4 is satisfied and and that' should be 'and that'.
  4. [Algorithm 2] The instruction 'Proceed with Steps 3–4 of Algorithm 1' is ambiguous because the test statistic has already been computed in Algorithm 2; it should say 'Proceed with Steps 4–6 of Algorithm 1' or equivalent.
  5. [Section 3(B) and Section 4.3] If the nested condition is indeed required for Theorem 4.13(B), it should be introduced in Section 3(B) as part of the definition of the sampling scheme, not only inside a lemma in the appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims reduce to an independently established semigroup-adjusted CLT, not to their own definitions or fitted inputs.

full rationale

The estimation target Q is the trace-class covariance of the driving Wiener process, defined independently of the estimator RV_T^{Δ,O}/T. The central CLT (Theorem 4.13) is proved by decomposing the discretely sampled realized covariation into remainders (Lemmas B.1, B.2, B.4, B.6, B.7, B.9) plus the semigroup-adjusted estimator SARCV^{Δ,I}, whose asymptotic normality is quoted from the prior work [5, 6, 7] (Theorem D.10 in [7]). Although one present author (Schroers) is a coauthor of those prior papers, the cited results are published mathematical theorems with their own proofs and do not themselves assume the conclusion of this paper; they are not fitted to any data, and they do not define Q in terms of the statistic. The new sampling schemes (B) and (C) are handled by showing that the projection/interpolation error is asymptotically negligible, which is genuine new content. The proof of Theorem 4.13(B) as written uses a nested-subdivision condition in Lemma B.9 that is not stated in the theorem; this is a potential correctness gap in the proof, but it is not circularity, because the missing condition is a hypothesis on the mesh, not the target Q. No parameter is fitted and then renamed a prediction, and no known empirical pattern is repackaged as a derivation. Accordingly the circularity score is 0.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

No parameters are fitted to data in the theoretical derivation; the conditions involve abstract regularity indices (γ, ι, β, η, ζ) that are not estimated. The paper does introduce a technical extra assumption (nested subdivisions) in a proof, which is flagged.

assumptions (7)
  • domain assumption A is closed, densely defined, self-adjoint, positive definite on H with compact inverse (Assumption 1)
    Needed for analytic semigroup and mild solution. Used throughout; without it the realized covariation may not be consistent (cf. [6] Example 4(ii)).
  • domain assumption Q is trace-class and X0 ∈ L4(Ω,H) (Assumption 1)
    Ensures the noise covariance is Hilbert-Schmidt in kernel form and the initial condition has finite fourth moment for variance bounds.
  • domain assumption Regularity conditions of Assumption 2: ∥A^{γ/2}Q∥_{L2}<∞ and ∥A^{ι/2}X0∥_{L4}<∞
    These determine convergence rates and are needed for the CLT with γ≥1, 2ι≥γ.
  • domain assumption Assumption 3: ˙H^1 ↪ H^1 (for local averages)
    Provides the spatial approximation rate for Ph via Lemma 3.4.
  • domain assumption Assumption 4: stronger operator-norm regularity and ˙H^2 ↪ H^2 (for pointwise sampling)
    Allows point evaluation through cylindrical random variables and controls interpolation error Ih.
  • ad hoc to paper Nested subdivisions of T_h in Lemma B.9 for O=Ph
    This condition is used in the proof but is not stated in Theorem 4.13(B); it is an unflagged additional assumption.
  • standard math Standard Itô isometry, BDG inequality, spectral calculus, embedding theorems
    Used throughout the proofs without proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Nonparametric Inference for Noise Covariance Kernels in Parabolic SPDEs using Space-Time Infill-Asymptotics." pith.science (2026). https://pith.science/paper/43IZFHIY

@misc{pith2026250820947,
  author       = {Pith},
  title        = {Pith review of: Nonparametric Inference for Noise Covariance Kernels in Parabolic SPDEs using Space-Time Infill-Asymptotics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/43IZFHIY}},
  note         = {Machine review of arXiv:2508.20947}
}
read the original abstract

We develop an asymptotic limit theory for nonparametric estimation of the noise covariance kernel in linear parabolic stochastic partial differential equations (SPDEs) with additive colored noise, using space-time infill asymptotics. The method employs discretized infinite-dimensional realized covariations and requires only mild regularity assumptions on the kernel to ensure consistent estimation and asymptotic normality of the estimator. On this basis, we construct omnibus goodness-of-fit tests for the noise covariance that are independent of the SPDE's differential operator. Our framework accommodates a variety of spatial sampling schemes and allows for reliable inference even when spatial resolution is coarser than temporal resolution.

Figures

Figures reproduced from arXiv: 2508.20947 by the authors.

Figure 1
Figure 1. √ Approximation of the error of Theorem 4.9, with h = ∆ and q a Mat´ern covariance kernel. √ ∆, Theorem 4.9(B) predicts an RMSE of order O(∆min(γ,1)/2 ). For a Mat´ern smoothness parameter ν > 1/4, we have γ > 1, which yields an optimal rate of O(∆1/2 ). For ν ≤ 1/4, the rate is O(∆γ/2 ) for any γ < 2ν+1/2, so we expect a rate approaching O(∆ν+1/4 ). These theoretical rates are consistent with the numerical simulati… view at source ↗
Figure 2
Figure 2. √ Approximation of the error of Theorem 4.9, with h = ∆ and q defined in terms of the eigenbasis of A. ple, we let A be the negative Laplacian with zero Dirichlet boundary conditions and [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Rejection rates for the test of Theorem 5.2 (α = 0.05), under O = Ih with ∆ = 2−8 , h = 2−4 . true kernel is from the Mat´ern family, the test for a Mat´ern null correctly fails to reject, with rejection rates close to zero, consistent with the conservative nature of the test. Conversely, when the data is generated from a commutative kernel, the same test consistently rejects the null hypothesis with a power close t… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Left: Rejection rates for the goodness-of-fit test of The￾orem 5.2 (α = 0.05) under O = Ih with ∆ = 2−12, h = 2−6 . Right: Rejection rates for the goodness-of-fit test of Theorem 5.4 (α = 0.05) with ∆ = 2−8 , h = 2−4 . [5] F. E. Benth, D. Schroers, and A. E. D. Veraart…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 37 canonical work pages

  1. [1]

    Altmeyer, R

    I. Altmeyer, R. Cialenko and G. Pasemann. Parameter estimation for semilinear spdes from local measurements. Bernoulli, 29(3):2035–2061, 2023

    work page

  2. [2]

    Altmeyer and M

    R. Altmeyer and M. Reiß. Nonparametric estimation for linear spdes from local measure- ments. Ann. Appl. Probab., 31(1):1–38, 2021

    work page 2021

  3. [3]

    Altmeyer, A

    R. Altmeyer, A. Tiepner, and M. Wahl. Optimal parameter estimation for linear spdes from multiple measurements. Ann. Statist., 52(4):1307–1333, 2024

    work page 2024

  4. [4]

    Ben Belgacem and S

    F. Ben Belgacem and S. C. Brenner. Some nonstandard finite element estimates with ap- plications to 3 D Poisson and Signorini problems. ETNA, Electron. Trans. Numer. Anal. , 12:134–148, 2001. NONPARAMETRIC INFERENCE FOR NOISE COV ARIANCE KERNELS 23 0.136 1.000 1.000 1.000 1.000 1.000 1.000 0.044 1.000 1.000 1.000 1.000 1.000 1.000 0.051 1.000 1.000 1.000 ...

    work page 2001

  5. [5]

    F. E. Benth, D. Schroers, and A. E. D. Veraart. A weak law of large numbers for realised covariation in a Hilbert space setting. Stoch. Proc. Applic., 145:241–268, 2022

    work page 2022

  6. [6]

    F. E. Benth, D. Schroers, and A. E. D. Veraart. A feasible central limit theorem for realised covariation of spdes in the context of functional data. Ann.Appl.Prob., 2024

    work page 2024

  7. [7]

    a feasible central limit theorem for realised covariation of spdes in the context of functional data

    F. E. Benth, D. Schroers, and A. E. D. Veraart. Supplement to “a feasible central limit theorem for realised covariation of spdes in the context of functional data”. Ann.Appl.Prob., Supplementary Material, 2024

    work page 2024

  8. [8]

    Bibinger and M

    M. Bibinger and M. Trabs. On central limit theorems for power variations of the solution to the stochastic heat equation. In A. Steland, E. Rafaj lowicz, and O. Okhrin, editors,Stochastic Models, Statistics and Their Applications , pages 69–84, Cham, 2019. Springer International Publishing

    work page 2019

Show all 42 references
  1. [9]

    Bibinger and M

    M. Bibinger and M. Trabs. Volatility estimation for stochastic pdes using high-frequency observations. Stoch. Proc. Applic., 130(5):3005 – 3052, 2020

    2020

  2. [10]

    V. I. Bogachev. Gaussian Measures, volume 62 of Mathematical Surveys and Monographs . American Mathematical Society, Providence, RI, 1998

    1998

  3. [11]

    S. C. Brenner and L. R. Scott. The mathematical theory of finite element methods , volume 15 of Texts in Applied Mathematics . Springer, New York, third edition, 2008

    2008

  4. [12]

    C. Chong. High-frequency analysis of parabolic stochastic pdes. Ann. Statist. , 48(2):1143– 1167, 04 2020

    2020

  5. [13]

    Chong and R

    C. Chong and R. C. Dalang. Power variations in fractional Sobolev spaces for a class of parabolic stochastic PDEs. E-print arXiv:2006.15817 , 2020

    2006 arXiv

  6. [14]

    Cialenco

    I. Cialenco. Statistical inference for SPDEs: an overview. Statist. Inf. Stoch. Proc., 20(2):309– 329, 12 2018

    2018

  7. [15]

    Cialenco and Y

    I. Cialenco and Y. Huang. A note on parameter estimation for discretely sampled SPDEs. Stoch. Dynamics, 20(03), 2020

    2020

  8. [16]

    R. Cont. Modeling term structure dynamics: An infinite dimensional approach. Int. J. Theor. Appl. Finance, 08(03):357–380, 2005. 24 NONPARAMETRIC INFERENCE FOR NOISE COV ARIANCE KERNELS

    2005

  9. [17]

    Da Prato and J

    G. Da Prato and J. Zabczyk. Stochastic Equations in Infinite Dimensions , volume 152 of Encyclopedia of Mathematics and its Applications . Cambridge University Press, Cambridge, second edition, 2014

    2014

  10. [18]

    Dupont and R

    T. Dupont and R. Scott. Polynomial approximation of functions in Sobolev spaces. Math. Comp., 34(150):441–463, 1980

    1980

  11. [19]

    Gapaillard

    C. Gapaillard. Un r´ esultat de compacit´ e pour l’interpolation de couples hilbertiens. C. R. Acad. Sci. Paris S´ er. A, 278:681–684, 1974

    1974

  12. [20]

    Gaudlitz and M

    S. Gaudlitz and M. Reiß. Estimation for the reaction term in semi-linear spdes under small diffusivity. Bernoulli, 29(4):3033–3058, 2023

    2023

  13. [21]

    M. Hairer. An introduction to stochastic PDEs, 2009. Preprint at arXiv:0907.4178

    2009 arXiv

  14. [22]

    Hildebrandt and M

    F. Hildebrandt and M. Trabs. Parameter estimation for SPDEs based on discrete observations in time and space. Electron. J. Stat. , 15(1):2716 – 2776, 2021

    2021

  15. [23]

    Hildebrandt and M

    F. Hildebrandt and M. Trabs. Nonparametric calibration for stochastic reaction–diffusion equations based on discrete observations. Stoch. Proc. Applic., 162:171–217, 2023

    2023

  16. [24]

    Jacod and P

    J. Jacod and P. Protter. Discretization of Processes, volume 67 of Stochastic Modelling and Applied Probability. Springer, Heidelberg, 2012

    2012

  17. [25]

    Kaino and M

    Y. Kaino and M. Uchida. Parametric estimation for a parabolic linear spde model based on discrete observations. J. Statist. Plann. Inference , 2020

    2020

  18. [26]

    Kov´ acs, A

    M. Kov´ acs, A. Lang, and A. Petersson. Hilbert–Schmidt regularity of symmetric integral op- erators on bounded domains with applications to SPDE approximations. Stochastic Analysis and Applications, 41(3):564–590, 2023

    2023

  19. [27]

    R. Kruse. Strong and weak approximation of semilinear stochastic evolution equations . Springer, 2014

    2014

  20. [28]

    Lindgren, H

    F. Lindgren, H. Bakka, D. Bolin, E. Krainski, and H. Rue. A diffusion-based spatio-temporal extension of gaussian mat` ern fields.arXiv preprint arXiv:2006.04917 , 2020

    2006 arXiv

  21. [29]

    Lindgren, H

    F. Lindgren, H. Rue, and J. Lindstr¨ om. An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach. J. R. Statist. Soc. B, 73(4):423—-498, 2011

    2011

  22. [30]

    J. L. Lions and E. Magenes. Non-homogeneous boundary value problems and applications. Vol. I . Springer-Verlag, New York-Heidelberg, 1972. Translated from the French by P. Ken- neth, Die Grundlehren der mathematischen Wissenschaften, Band 181

    1972

  23. [31]

    X. Liu, K. M. Yeo, and S. Y. Lu. Statistical modeling for spatio-temporal data from stochastic convection-diffusion processes. J. Amer. Statist. Assoc. , 117:1482–1499, 2022

    2022

  24. [32]

    G. J. Lord and A. Petersson. Piecewise linear interpolation of noise in finite element approx- imations of parabolic spdes. SIAM Journal on Numerical Analysis , 63(2):542–563, 2025

    2025

  25. [33]

    M. Reiß, C. Strauch, and L. Trottner. Change point estimation for a stochastic heat equation. Ann. Statist. (to appear) , 2023

    2023

  26. [34]

    M. Riedle. Cylindrical Wiener processes. In S´ eminaire de Probabilit´ es XLIII, Poitiers, France, Juin 2009. , pages 191–214. Berlin: Springer, 2011

    2009

  27. [35]

    Schroers

    D. Schroers. Dynamically consistent analysis of realized covariations in term structure models. arXiv preprint arXiv:2406.19412 , 2024

    2024 arXiv

  28. [36]

    Schroers

    D. Schroers. Robust functional data analysis for stochastic evolution equations in infinite dimensions. arXiv preprint arXiv:2401.16286 , 2024

    2024 arXiv

  29. [37]

    Sigrist, H

    F. Sigrist, H. K¨ unsch, and W. Stahel. A dynamic nonstationary spatio-temporal model for short term prediction of precipitation. Ann. Appl. Stat. , 2012

    2012

  30. [38]

    Sigrist, H

    F. Sigrist, H. K¨ unsch, and W. Stahel. Stochastic partial differential equation based modelling of large space-time data sets. J. R. Statist. Soc. B , 77(1):3–33, 2015

    2015

  31. [39]

    M. Stein. Interpolation of spatial data: some theory for kriging . Springer Science & Business Media, 1999

    1999

  32. [40]

    A. W. van der Vaart and J. Wellner. Weak Convergence and Empirical Processes: with Applications to Statistics . Springer Science & Business Media, New York, 1996

    1996

  33. [41]

    Wendland

    H. Wendland. Scattered data approximation, volume 17 of Cambridge Monographs on Applied and Computational Mathematics . Cambridge University Press, Cambridge, 2005

    2005

  34. [42]

    A. Yagi. Abstract parabolic evolution equations and their applications . Springer monographs in mathematics. Springer-Verlag, Berlin, 2010. NONPARAMETRIC INFERENCE FOR NOISE COV ARIANCE KERNELS 25 Appendix A. Pointwise evaluation of the SPDE This section provides the technical...

    2010

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.