Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

The BdryMat\'ern GP: Reliable incorporation of boundary information on irregular domains for Gaussian process modeling

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The BdryMatérn GP embeds boundary conditions directly into the covariance kernel, so Gaussian-process surrogates can exploit known physics on irregular domains without sacrificing smoothness control or error guarantees.

desk verdict The Dirichlet kernel and FEM approximation are genuinely useful, but the Robin-boundary construction is not justified on the cornered domains it tests, and the coupled MC estimator is biased on the diagonal. read the letter →

arxiv 2507.09178 v1 pith:NS75Z2EW submitted 2025-07-12 stat.ME math.STstat.MLstat.TH

Liang Ding , Simon Mak , C. F. Jeff Wu This is my paper
classification stat.MEmath.STstat.MLstat.TH MSC 62M4062G08
keywords GaussianprocessesMatérnkernelboundaryconditionsstochasticpartialdifferentialequationsFeynman-Kacformulafiniteelementmethodirregulardomainscomputerexperiments
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

This paper introduces a Gaussian process (GP) model whose covariance kernel itself encodes Dirichlet, Neumann, or Robin boundary conditions, so that the model can be trained with limited simulator data on arbitrary connected domains. The central claim is that the kernel defined in path-integral form via the Feynman-Kac formula is the covariance of a GP whose sample paths satisfy the prescribed boundaries almost surely while retaining the Matérn smoothness parameter ν. The paper also provides a finite-element approximation procedure for the kernel on irregular domains, with a proven L2 error rate of O($m^{{-ν/(2ν+d)}}$), matching the minimax rate for GP regression. A tensor version gives closed-form kernels on high-dimensional unit hypercubes. If correct, this gives surrogate modelers a principled way to inject known physical boundary information into emulators and to control both smoothness and approximation error.

What carries the argument

The load-bearing construction is the Feynman-Kac path-integral representation of the Green's function for the elliptic operator (κ²−Δ) under boundary conditions. It turns the unknown Green's function into expectations of the Matérn kernel evaluated at Brownian hitting points and hitting times, so the kernel can be approximated by simulating Brownian motions; positivity is preserved by coupling the Monte Carlo draws across rows of the kernel matrix, and B-spline finite elements give the approximation error rate.

What would settle it

On a square domain with Dirichlet boundary conditions, simulate the BdryMatérn kernel of (12) by Brownian hitting locations and compare it against a direct FEM solution of (8); if the kernel matrix fails positive definiteness or sample paths visibly fail the zero boundary at corners, the path-integral claim on non-smooth boundaries would be contradicted.

Watch

Extended reading notes

Core claim

The paper claims to derive, in Equations (12) and (27), a BdryMatérn kernel as the covariance of the solution to the SPDE (κ²−Δ)f = Wν with zero Dirichlet or Robin boundary conditions on a connected domain whose boundary is twice-differentiable almost everywhere. The kernel is expressed as the base Matérn kernel k_{ν+2} corrected by expectations over Brownian hitting locations and hitting times (or boundary local times for Robin boundaries). Theorems 3.2 and 3.4 state that sample paths satisfy the boundary condition almost surely and are (⌈ν⌉−1)-differentiable. For computation, the paper proposes a coupled Monte Carlo estimator of the kernel matrix that is positive definite almost surely, then uses finite-element (B-spline) projection to obtain a kernel approximation with the error bound above.

Load-bearing premise

The load-bearing premise is that the domain boundary is twice-differentiable almost everywhere, which the Feynman-Kac and boundary-local-time representations require; the paper does not show this holds for the polygonal domains used in its experiments.

Editorial extensions

If this is right

  • Surrogate models can incorporate boundary information on non-hypercube domains without projection tricks, so emulators of PDE-based simulators on complex geometries become more accurate with small n.
  • The smoothness parameter ν carries over from the Matérn GP, giving modelers control over differentiability of the fitted surface while boundaries are satisfied.
  • The FEM/B-spline approximation rate matches the minimax L2 rate for GP regression, meaning kernel approximation need not slow the overall statistical convergence.
  • The tensor form gives closed-form boundary-integrated kernels on [0,1]^d, which can be used directly in high-dimensional GP prediction.
  • The rule-of-thumb m = n^{(2ν+d)/d} for the number of inducing points keeps kernel approximation error at the same order as the GP contraction rate.

Reading between the lines

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

  • The same SPDE-plus-Feynman-Kac strategy may extend to other linear PDE constraints (e.g., partial observations of gradients) by substituting the appropriate boundary functional, though the paper does not prove this.
  • If the boundary-regularity assumption fails on polygonal domains, the path-integral representation may still hold in a limiting sense; a direct check on a square could resolve whether the polygon experiments are fully justified.
  • The coupled MC estimator's positive-definiteness guarantee suggests a general recipe for making Monte-Carlo kernel approximations stable in other settings, not only boundary-constrained ones.
  • For very high-dimensional hypercube domains, the tensor product may suffer from the curse of dimensionality despite the closed form; comparing it with the FEM approach would be a natural stress test.
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

4 major / 5 minor

Summary. The paper proposes the BdryMatérn GP, a Gaussian process model that incorporates Dirichlet, Neumann, or Robin boundary information on connected, possibly irregular domains. The covariance kernel is derived from the SPDE representation of Matérn fields by solving (κ²−Δ)f = W on the domain with boundary conditions, and it is expressed in path-integral form via Feynman-Kac formulas (Eqs. (12) and (27)). The authors prove sample-path boundary satisfaction and differentiability control (Theorems 3.2 and 3.4), propose a coupled Monte Carlo estimator and an FEM/B-spline approximation with an L2 error rate (Theorems 4.1–4.4), and give a closed-form tensor kernel on hypercubes (Theorem 5.1). Numerical experiments on T-shaped, ring, disk, and holed-rectangle domains, as well as on 30-dimensional hypercube test functions, show improved predictive performance over standard Matérn GP baselines.

Significance. If the technical gaps are repaired, this would be a meaningful advance: it moves boundary-integrated GPs from distance-function heuristics to an SPDE-based derivation with explicit sample-path smoothness and an explicit finite-element approximation rate, and the tensor form provides a practical closed-form alternative on hypercubes. The Dirichlet path-integral representation is on solid ground, and the error-rate formula in Theorem 4.4 is a concrete, falsifiable target. However, several load-bearing points—the Robin-boundary representation on nonsmooth domains, the bias of the coupled Monte Carlo estimator, and the boundary satisfaction of the FEM-based kernel—are currently unresolved and need to be addressed before the main claims can be accepted.

major comments (4)
  1. [§3.2, Theorem 3.3, Eqs. (27)–(28)] The Robin-boundary Feynman-Kac representation is a load-bearing step, but it is used under the assumption that the boundary is 'twice-differentiable almost everywhere,' and the paper does not show that this suffices for the cited machinery. Reference [50, Corollary 4.4] and the boundary-local-time bounds of [26,27] are standardly stated for C² or at least Lipschitz domains, and no extension is provided for domains with corners or reentrant corners. The experimental domains in §6.1 (T-shaped, holed rectangle) are precisely such domains, where the normal derivative in (24) is not classically defined at corner points. Until this gap is closed, Theorem 3.4(a) and the Robin parts of Theorems 4.3–4.4 are unsupported for the domains used in the experiments.
  2. [§4.1, Eq. (36); Appendix C] The coupled Monte Carlo estimator is biased on diagonal entries. For i=j, Eq. (36) uses the same Brownian path in the fourth term, producing σ²e^{-2κ²τ_i}, whereas the exact path-integral fourth term in (12) requires independent Brownian motions and has expectation E[kν(Bγ,B'_τ)e^{-κ²(τ+γ)}]. Since kν ≤ σ² and E[e^{-2κ²τ_i}] ≥ (E[e^{-κ²τ_i}])², the diagonal bias is at least σ²Var(e^{-κ²τ_i}) > 0, and it does not vanish as m grows. The proof of Theorem 4.3 in Appendix C does not account for this dependence: the decompositions K31 and A31 treat the noise in the two factors of P^m[G_i](x)P^m[G_i](x') as independent. Consequently, the L2 error bound (50) and the rate in Theorem 4.4 do not follow for the proposed estimator. The estimator should be modified to use independent paths for diagonal entries, or the error analysis must explicitly include the diagonal bias.
  3. [§4.2, Corollary 1] The claim that the FEM-based kernel satisfies the boundary conditions almost surely does not follow from Theorems 4.1 and 4.2, as asserted. The kernel \hat{k}_{m,Q}(x,x') = Φ(x)^T M Φ(x') in (46) vanishes on ∂\hat{X} only if the basis functions or the coefficient matrix enforce this; positive definiteness of \hat{K}^C(U,U) does not constrain values of the kernel outside the inducing set. For the B-spline basis described in §4.3, the basis functions do not all vanish on the boundary (e.g., N_s is nonzero at 1 for s>1), so Corollary 1 is not valid for the recommended implementation unless boundary-modified basis functions are introduced. This affects the claimed boundary satisfaction of the approximated posterior.
  4. [§3.1, Theorem 3.2; Appendix A] The smoothness statement is internally inconsistent in its parameter indexing. The SPDE (8) with Wν yields the kernel k_{ν+2,B} in (11), but after (12) the paper renames this kernel as k_{ν,B}. Theorem 3.2(b) then claims (⌈ν⌉−1)-differentiability, while the proof in Appendix A concludes (⌈ν+2⌉−1)-differentiability and, for integer ν, 'ν+1'-differentiability. These statements cannot all describe the same kernel. The paper must either consistently index the smoothness parameter or reparameterize the SPDE to W_{ν−2} and adjust all formulas accordingly; otherwise, the smoothing guarantee is ambiguous.
minor comments (5)
  1. [§3.1, after Eq. (12)] The notation k_{ν,B} in (12) conflicts with the use of k_{ν+2,B} in (11)–(12); please standardize the subscript or explicitly state the reparameterization.
  2. [Appendix B] The first line of Appendix B says the proof is identical to the proof of Theorem 3.4, but it should refer to Theorem 3.2.
  3. [§5, Eq. (60)] The closed-form expressions in (60) have denominators (1−cκ) and (1+cκ), so the formula is undefined at cκ=1; either assume cκ≠1 or supply the limiting form.
  4. [§6.1] The text states that n=50 training points are sampled, while Figure 3 plots results over a range of numbers of design points; please clarify the protocol, e.g., whether subsampling was used.
  5. [§4.1, Theorem 4.1 proof] In the proof of Theorem 4.1, the vector δ_i is introduced without a formal definition, which makes the matrix inequality in (37)–(38) harder to follow.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the BdryMatérn kernel is derived from the SPDE/Feynman–Kac representation, and the experimental validation is a conventional simulation check.

full rationale

The central derivation is self-contained: the BdryMatérn kernel is defined in Eq. (11) as the covariance of the solution f of the boundary-constrained SPDE (8)/(25), and Theorems 3.1 and 3.3 derive the path-integral expressions (12)/(27) from the Green's function representation plus external Feynman–Kac formulas (Dirichlet: [7]; Robin: [50, Cor. 4.4]). No target kernel or boundary-satisfaction property is substituted into the derivation: the Green's function is characterized by Eq. (9)/(26), and the path-integral form is obtained by solving the homogeneous boundary-value problem for v_ω, not by assuming it. Theorems 3.2 and 3.4 verify, rather than presuppose, that the GP with this covariance has the desired boundary behavior; the proofs proceed from the PDE eigenfunction decomposition and standard regularity results. The FEM approximation in Eq. (46) is compared with the same kernel k_{ν,B} in Theorems 4.3 and 4.4, and the error bound is an honest approximation-theoretic statement; convergence rates come from B-spline theory (de Boor's conjecture [63]), not from the target result. No parameter is fitted to a subset of the boundary data and then relabeled a prediction; the hyperparameters are standard GP parameters estimated by maximum likelihood. The numerical experiments generate test functions from the BdryMatérn GP itself, which is a standard simulation check and does not make the derivation circular. Self-citations (e.g., [13], [14], [15], [32], [37]) appear in related-work or implementation contexts and are not load-bearing for the main mathematical claims. The Robin-boundary regularity gap flagged by the skeptic is a possible correctness or assumption issue—the cited Feynman–Kac result [50] may require more boundary regularity than 'twice-differentiable almost everywhere'—but that is not circularity: the paper relies on an external theorem, not on its own conclusion, and the remedy would be additional regularity conditions or a proof of the local-time representation on the stated domains.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The derivation relies entirely on standard PDE and stochastic calculus tools. No free parameters are fitted to make the derivation work; the Gaussian process hyperparameters ν, κ, and σ² are standard model parameters. No new physical entities are postulated. The main burden is the regularity assumption on the domain boundary and the associated existence of local times.

assumptions (6)
  • standard math The Feynman-Kac formula provides a path-integral representation for solutions of elliptic PDEs with Dirichlet and Robin boundary conditions.
    Used in Theorems 3.1 and 3.3 to convert the Green's function representation into expectations over boundary hitting times and local times.
  • domain assumption The domain boundary is twice-differentiable almost everywhere.
    Stated as the standing assumption on the domain in Theorems 3.1, 3.2, 3.3, and 3.4. The paper does not prove that this condition is sufficient for the cited Feynman-Kac theorems, which typically require C² regularity.
  • domain assumption Boundary local time for reflected Brownian motion exists and is integrable under the assumed boundary regularity.
    Invoked in Theorem 3.3 and the definition (28) of local time; the paper's weak boundary regularity may not guarantee this.
  • domain assumption The SPDE (κ² - Δ)f = W_ν with homogeneous Dirichlet or Robin boundary conditions has a unique solution that is a Gaussian process.
    Assumed in Section 3; the paper cites Lifshits for the Gaussian property but does not provide existence machinery on general bounded domains.
  • standard math The Matérn kernel admits an eigendecomposition on a compact domain with eigenvalues λ_i = O(i^{-(2ν/d+1)}).
    Used in Appendix A for sample path regularity and in Appendix D for the B-spline approximation rate.
  • domain assumption The coefficient c(x) in the Robin boundary condition is nonnegative.
    Required in Theorem 3.3 and Theorem 4.2 for the Feynman-Kac representation and the positivity of the coupled estimator.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The BdryMat\'ern GP: Reliable incorporation of boundary information on irregular domains for Gaussian process modeling." pith.science (2026). https://pith.science/paper/NS75Z2EW

@misc{pith2026250709178,
  author       = {Pith},
  title        = {Pith review of: The BdryMat\'ern GP: Reliable incorporation of boundary information on irregular domains for Gaussian process modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NS75Z2EW}},
  note         = {Machine review of arXiv:2507.09178}
}
read the original abstract

Gaussian processes (GPs) are broadly used as surrogate models for expensive computer simulators of complex phenomena. However, a key bottleneck is that its training data are generated from this expensive simulator and thus can be highly limited. A promising solution is to supplement the learning model with boundary information from scientific knowledge. However, despite recent work on boundary-integrated GPs, such models largely cannot accommodate boundary information on irregular (i.e., non-hypercube) domains, and do not provide sample path smoothness control or approximation error analysis, both of which are important for reliable surrogate modeling. We thus propose a novel BdryMat\'ern GP modeling framework, which can reliably integrate Dirichlet, Neumann and Robin boundaries on an irregular connected domain with a boundary set that is twice-differentiable almost everywhere. Our model leverages a new BdryMat\'ern covariance kernel derived in path integral form via a stochastic partial differential equation formulation. Similar to the GP with Mat\'ern kernel, we prove that sample paths from the BdryMat\'ern GP satisfy the desired boundaries with smoothness control on its derivatives. We further present an efficient approximation procedure for the BdryMat\'ern kernel using finite element modeling with rigorous error analysis. Finally, we demonstrate the effectiveness of the BdryMat\'ern GP in a suite of numerical experiments on incorporating broad boundaries on irregular domains.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Neural Low-Discrepancy Sequences

    cs.LG 2025-10 conditional novelty 6.0 of 10

    A trained index-to-point network generates low-discrepancy sequences whose prefixes beat Sobol, Halton, and scrambled Sobol discrepancy in four dimensions.

Reference graph

Works this paper leans on

86 extracted references · 75 canonical work pages · cited by 1 Pith paper

  1. [11]

    , GAO, H

    DALTON , D., LAZARUS , A. , GAO, H. and HUSMEIER , D. (2024). Boundary constrained Gaussian processes for robust physics-informed machine learning of linear partial differential equations. Journal of Machine Learning Research 25 1–61

    work page 2024

  2. [1]

    and STEGUN , I

    ABRAMOWITZ , M. and STEGUN , I. A. (1965). Handbook of Mathematical Functions: with Formulas, Graphs, and Mathematical Tables. Courier Corporation

    1965

  3. [2]

    BRENNER , S. C. and SCOTT, L. R. (2008). The Mathematical Theory of Finite Element Methods , 3nd ed. Springer. BDRYMATÉRN GP 25

    2008

  4. [3]

    and G RIEBEL , M

    B UNGARTZ , H.-J. and G RIEBEL , M. (2004). Sparse Grids. Acta Numerica 13 1-123

    2004

  5. [4]

    and B ERGER , R

    C ASELLA , G. and B ERGER , R. L. (2002). Statistical Inference, 2 ed. Cengage Learning

    2002

  6. [5]

    , MAK, S

    CHEN , Z. , MAK, S. and WU, C. J. (2024). A hierarchical expected improvement method for Bayesian optimization. Journal of the American Statistical Association 119 1619–1632

    2024

  7. [6]

    , CRUMP, C

    CHOI , H. , CRUMP, C. , DURIEZ , C. , ELMQUIST , A. , HAGER , G. , HAN, D., HEARL , F ., HODGINS , J. , JAIN , A. , LEVE , F .et al. (2021). On the use of simulation in robotics: Opportunities, challenges, and suggestions for moving forward. Proceedings of the National Academy of Sciences 118 e1907856118

    2021

  8. [7]

    CHUNG , K. L. and RAO, K. M. (1981). Feynman-Kac Functional and the Schrödinger Equation In Semi- nar on Stochastic Processes, 1981 1–29. Birkhäuser Boston, Boston, MA. https://doi.org/10.1007/ 978-1-4612-3938-3_1

    work page 1981

Show all 86 references
  1. [8]

    D ˜UNG , D. (2011). Optimal adaptive sampling recovery. Advances in Computational Mathematics 34 1–41

    2011

  2. [9]

    D ˜UNG , D. (2011). Adaptive algorithms in sampling recovery. arXiv preprint arXiv:1102.3540

    2011 arXiv

  3. [10]

    , PFÖRTNER , M

    DA COSTA , N. , PFÖRTNER , M. , DA COSTA , L. and HENNIG , P .(2023). Sample path regularity of Gaussian processes from the covariance kernel. arXiv preprint arXiv:2312.14886

    2023

  4. [12]

    D ANG , C. (2012). Triangulations and Simplicial Methods 421. Springer Science & Business Media

    2012

  5. [13]

    , MAK, S

    DING , L. , MAK, S. and WU, C. (2019). BdryGP: a new Gaussian process model for incorporating boundary information. arXiv preprint arXiv:1908.08868

    2019 arXiv

  6. [14]

    , TUO, R

    DING , L. , TUO, R. and SHAHRAMPOUR , S. (2020). Generalization guarantees for sparse kernel approxima- tion with entropic optimal features. In International Conference on Machine Learning 2545–2555. PMLR

    2020

  7. [15]

    and ZHANG , X

    DING , L. and ZHANG , X. (2024). Sample and computationally efficient stochastic kriging in high dimen- sions. Operations Research 72 660–683

    2024

  8. [16]

    , CHEN , Y., MULLIGAN , J

    EHLERS , R. , CHEN , Y., MULLIGAN , J. , JI, Y., KUMAR , A. , MAK, S. , JACOBS , P ., MAJUMDER , A. , ANGERAMI , A. , ARORA , R. et al. (2024). Bayesian Inference analysis of jet quenching using inclusive jet and hadron suppression measurements. arXiv preprint arXiv:2408.08247

    2024

  9. [17]

    E VANS , L. C. (2015). Partial Differential Equations, 2nd ed. AMS

    2015

  10. [18]

    F OLLAND , G. B. (2009). Fourier analysis and its applications 4. American Mathematical Soc

    2009

  11. [19]

    GAGLIARDO , E. (1959). Ulteriori proprietà di alcune classi di funzioni in più variabili. Ricerche Mat. 8 24–51

    1959

  12. [20]

    GOLCHI , S., BINGHAM , D. R. , CHIPMAN , H. and CAMPBELL , D. A. (2015). Monotone emulation of computer experiments. SIAM/ASA Journal on Uncertainty Quantification 3 370-392

    2015

  13. [21]

    GRAMACY , R. B. (2020). Surrogates: Gaussian Process Modeling, Design, and Optimization for the Applied Sciences. CRC Press

    2020

  14. [22]

    GREBENKOV , D. S. (2019). Probability distribution of the boundary local time of reflected Brownian motion in Euclidean domains. Physical Review E 100 062110

    2019

  15. [23]

    , FRANKEL , A

    GULIAN , M. , FRANKEL , A. and SWILER , L. (2022). Gaussian process regression constrained by boundary value problems. Computer Methods in Applied Mechanics and Engineering 388 114117

    2022

  16. [24]

    HARRISON , J. M. and REIMAN , M. I. (1981). Reflected Brownian motion on an orthant. The Annals of Probability 9 302–308

    1981

  17. [25]

    H ÖLLIG , K. (2003). Finite Element Methods with B-splines. SIAM

    2003

  18. [26]

    Reflecting Brownian motion, boundary local time and the Neumann problem

    HSU, P .(1984). Reflecting Brownian motion, boundary local time and the Neumann problem. Stanford University

    1984

  19. [27]

    Probabilistic approach to the Neumann problem

    HSU, P .(1985). Probabilistic approach to the Neumann problem. Communications on pure and applied mathematics 38 445–472

    1985

  20. [28]

    HUEBNER , K. H. , DEWHIRST , D. L. , SMITH , D. E. and BYROM , T. G. (2001). The Finite Element Method for Engineers. John Wiley & Sons

    2001

  21. [29]

    and D ELANGE , S

    H UMPHREY , J. and D ELANGE , S. L. (2016). Introduction to Biomechanics. Springer

    2016

  22. [30]

    ITÔ, S. (1957). Fundamental solutions of parabolic differential equations and boundary value problems. In Japanese journal of mathematics: transactions and abstracts 27 55–102. The Mathematical Society of Japan

    1957

  23. [31]

    JACKSON , S. E. and VERNON , I. (2023). Efficient emulation of computer models utilising multiple known boundaries of differing dimension. Bayesian Analysis 18 165–191

    2023

  24. [32]

    , MAK, S

    JI, Y. , MAK, S. , SOEDER , D. , PAQUET , J.-F . and BASS , S. A. (2024). A graphical multi-fidelity Gaussian process model, with application to emulation of heavy-ion collisions. Technometrics 66 267–281

    2024

  25. [33]

    JI, Y., YUCHI , H. S. , SOEDER , D., PAQUET , J.-F ., BASS , S. A. , JOSEPH , V. R., WU, C. J. and MAK, S. (2024). Conglomerate multi-fidelity Gaussian process modeling, with application to heavy-ion collisions. SIAM/ASA Journal on Uncertainty Quantification 12 473–502. 26

    2024

  26. [34]

    JOHNSON , M. E. , MOORE , L. M. and YLVISAKER , D. (1990). Minimax and maximin distance designs.Journal of Statistical Planning and Inference 26 131–148

    1990

  27. [35]

    KIEHN , R. (2001). Some closed form solutions to the Navier Stokes equations. arXiv preprint. https: //arxiv.org/abs/physics/0102002

    2001 arXiv

  28. [36]

    , SHARMA , A

    LEIMKUHLER , B. , SHARMA , A. and TRETYAKOV , M. V. (2023). Simplest random walk for approximating Robin boundary value problems and ergodic limits of reflected diffusions. The Annals of Applied Probability 33 1904–1960

    2023

  29. [37]

    and MAK, S

    LI, K. and MAK, S. (2025). ProSpar-GP: scalable Gaussian process modeling with massive non-stationary datasets. Journal of Computational and Graphical Statistics just-accepted 1–28

    2025

  30. [38]

    , MAK, S

    LI, K. , MAK, S. , PAQUET , J.-F . and BASS , S. A. (2023). Additive multi-index Gaussian process modeling, with application to multi-physics surrogate modeling of the quark-gluon plasma. arXiv preprint arXiv:2306.07299

    2023 arXiv

  31. [39]

    and TAN, M

    LI, Z. and TAN, M. H. Y. (2022). Improving Gaussian Process Emulators with Boundary Information. In Artificial Intelligence, Big Data and Data Science in Statistics: Challenges and Solutions in Environ- metrics, the Natural Sciences and Technology 171–192. Springer

    2022

  32. [40]

    L IFSHITS , M. (2012). Lectures on Gaussian processes. In Lectures on Gaussian Processes 1–117. Springer

    2012

  33. [41]

    and LINDSTRÖM , J

    LINDGREN , F ., RUE, H. and LINDSTRÖM , J. (2011). An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach.J. R. Statist. Soc. B 73 423-498

    2011

  34. [42]

    and MAK, S

    LIU, Y.-C. and MAK, S. (2025). QuIP: Experimental design for expensive simulators with many Qualitative factors via Integer Programming. arXiv preprint arXiv:2501.14616

    2025 arXiv

  35. [43]

    , HEINZ , U

    LIYANAGE , D., JI, Y., EVERETT , D., HEFFERNAN , M. , HEINZ , U. , MAK, S. and PAQUET , J.-F . (2022). Efficient emulation of relativistic heavy ion collisions with transfer learning. Physical Review C 105 034910

    2022

  36. [44]

    and JOSEPH , V

    MAK, S. and JOSEPH , V. R. (2018). Minimax and minimax projection designs using clustering. Journal of Computational and Graphical Statistics 27 166–178

    2018

  37. [45]

    , SUNG , C.-L

    MAK, S. , SUNG , C.-L. , WANG , X. , YEH, S.-T. , CHANG , Y.-H. , JOSEPH , V. R., YANG , V. and WU, C. J. (2018). An efficient surrogate model for emulation and physics extraction of large eddy simulations. Journal of the American Statistical Association 113 1443–1456

    2018

  38. [46]

    MILLER , J. J. , MAK, S. , SUN, B., NARAYANAN , S. R. , YANG , S. , SUN, Z. , KIM, K. S. and KWEON , C.-B. M. (2024). Expected diverse utility (EDU): Diverse Bayesian optimization of expensive computer simula- tors. arXiv preprint arXiv:2410.01196

    2024 arXiv

  39. [47]

    MILSTEIN , G. N. and TRETYAKOV , M. V. (2004). Stochastic numerics for mathematical physics 39. Springer

    2004

  40. [48]

    NARAYANAN , S. R. , JI, Y., SAPRA , H. D. , KWEON , C.-B. M. , KIM, K. S. , SUN, Z. , KOKJOHN , S. , MAK, S. and YANG , S. (2024). A misfire-integrated Gaussian process (MInt-GP) emulator for energy-assisted compression ignition (EACI) engines with varying cetane number jet fu...

    2024

  41. [49]

    NIRENBERG , L. (1959). On elliptic partial differential equations. Annali della Scuola Normale Superiore di Pisa-Scienze Fisiche e Matematiche 13 115–162

    1959

  42. [50]

    PAPANICOLAOU , V. G. (1990). The probabilistic solution of the third boundary value problem for second order elliptic equations. Probability Theory and related fields 87 27–77

    1990

  43. [51]

    and CHICA -O LMO , M

    PARDO -I GUZQUIZA , E. and CHICA -O LMO , M. (2008). Geostatistics with the Matern semivariogram model: A library of computer programs for inference, kriging and simulation. Computers & Geosciences 34 1073–1079

    2008

  44. [52]

    PAULSEN , V. I. and RAGHUPATHI , M. (2016). An introduction to the theory of reproducing kernel Hilbert spaces 152. Cambridge University Press

    2016

  45. [53]

    and WU, C

    PENG , C.-Y. and WU, C. F . J. (2014). On the choice of nugget in kriging modeling for deterministic computer experiments. Journal of Computational and Graphical Statistics 23 151–168

    2014

  46. [54]

    PETTERSSON , R. (1997). Penalization Schemes for Reflecting Stochastic Differential Equations. Bernoulli 3 403–414

    1997

  47. [55]

    PLUMLEE , M. (2014). Fast prediction of deterministic functions using sparse grid experimental designs. Journal of the American Statistical Association 109 1581-1591

    2014

  48. [56]

    PRATOLA , M. T. , HARARI , O., BINGHAM , D. and FLOWERS , G. E. (2017). Design and analysis of experiments on nonconvex regions. Technometrics 59 36–47

    2017

  49. [57]

    , MARTINO , S

    RUE, H. , MARTINO , S. and CHOPIN , N. (2009). Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society Series B: Statistical Methodology 71 319–392

    2009

  50. [58]

    RVACHEV , V. L. and SHEIKO , T. I. (1995). R-functions in boundary value problems in mechanics. Applied Mechanics Reviews 48 151—188

    1995

  51. [59]

    SAKELLARIS , G. (2021). On scale-invariant bounds for the Green’ s function for second-order elliptic equations with lower-order coefficients and applications. Analysis & PDE 14 251–299. BDRYMATÉRN GP 27

    2021

  52. [60]

    SANTNER , T. J. , WILLIAMS , B. J. and NOTZ , W. I. (2018). The Design and Analysis of Computer Experiments. Springer

    2018

  53. [61]

    , KIM, B., ANGERAMI , A., ARORA , R., BASS , S., CHEN , Y., DATTA , R., DU, L

    SENGUPTA , A., FRIES , R., KORDELL II, M. , KIM, B., ANGERAMI , A., ARORA , R., BASS , S., CHEN , Y., DATTA , R., DU, L. et al. (2025). Hybrid Hadronization–A Study of In-Medium Hadronization of Jets. arXiv preprint arXiv:2501.16482

    2025 arXiv

  54. [62]

    and PARKS , G

    SESHADRI , P ., YUCHI , S. and PARKS , G. T. (2019). Dimension reduction via Gaussian ridge functions. SIAM/ASA Journal on Uncertainty Quantification 7 1301–1322

    2019

  55. [63]

    SHADRIN , A. Y. (2001). The L∞-norm of the L2-spline projector is bounded independently of the knot sequence: A proof of de Boor’ s conjecture.Acta Math 187 59–137

    2001

  56. [64]

    SHU, C. (2012). Differential quadrature and its application in engineering. Springer Science & Business Media

    2012

  57. [65]

    and GHAHRAMANI , Z

    SNELSON , E. and GHAHRAMANI , Z. (2005). Sparse Gaussian processes using pseudo-inputs. Advances in neural information processing systems 18

    2005

  58. [66]

    and KOK, M

    SOLIN , A. and KOK, M. (2019). Know your boundaries: Constraining Gaussian processes by variational harmonic features. In The 22nd International Conference on Artificial Intelligence and Statistics 2193–2202. PMLR

    2019

  59. [67]

    and SÄRKKÄ , S

    SOLIN , A. and SÄRKKÄ , S. (2020). Hilbert space methods for reduced-rank Gaussian process regression. Statistics and Computing 30 419–446

    2020

  60. [68]

    S TEIN , M. L. (1999). Interpolation of Spatial Data: Some Theory for Kriging. Spring

    1999

  61. [69]

    STROOCK , D. W. and VARADHAN , S. S. (1971). Diffusion processes with boundary conditions. Communica- tions on Pure and Applied Mathematics 24 147–225

    1971

  62. [70]

    and B INGHAM , D

    S URJANOVIC , S. and B INGHAM , D. (2016). Virtual Library of Simulation Experiments: Test Functions and Datasets

    2016

  63. [71]

    T AN, M. H. Y. (2018). Gaussian process modeling with boundary information. Statistica Sinica 621–648

    2018

  64. [72]

    TEMAM , R. (2001). Navier-Stokes Equations: Theory and Numerical Analysis 343. American Mathematical Society

    2001

  65. [73]

    TROPP, J. A. (2012). User-friendly tail bounds for sums of random matrices. Foundations of computational mathematics 12 389–434

    2012

  66. [74]

    and WU, C

    TUO, R. and WU, C. F . J. (2016). A theoretical framework for calibration in computer models: parametriza- tion, estimation and convergence properties. SIAM/ASA JUQ 767-795

    2016

  67. [75]

    , JACKSON , S

    VERNON , I. , JACKSON , S. E. and CUMMING , J. A. (2019). Known boundary emulation of complex computer models. SIAM/ASA Journal on Uncertainty Quantification 7 838–876

    2019

  68. [76]

    and JING , B.-Y

    WANG , W. and JING , B.-Y. (2022). Gaussian process regression: Optimality, robustness, and relationship with kernel ridge regression. Journal of machine learning research 23 1–67

    2022

  69. [77]

    and BERGER , J

    WANG , X. and BERGER , J. O. (2016). Estimating shape constrained functions using Gaussian processes. SIAM/ASA Journal on Uncertainty Quantification 4 1-25

    2016

  70. [78]

    W ENDLAND , H. (2010). Scattered Data Approximation, 2nd ed. Cambridge University Press

    2010

  71. [79]

    WHEELER , M. W. , DUNSON , D. B. , PANDALAI , S. P ., BAKER , B. A. and HERRING , A. H. (2014). Mechanistic hierarchical Gaussian processes. Journal of the American Statistical Association 109 894–904

    2014

  72. [80]

    W HITE , F . M. and CORFIELD , I. (2006). Viscous Fluid Flow. McGraw-Hill New York

    2006

  73. [81]

    W HITTLE , P . (1954). On stationary processes in the plane.Biometrika 434–449

    1954

  74. [82]

    Stochastic-processes in several dimensions

    WHITTLE , P .(1963). Stochastic-processes in several dimensions. Bulletin of the International Statistical Institute 40 974–994

    1963

  75. [83]

    , NYCHKA , D

    WIENS , A. , NYCHKA , D. and KLEIBER , W. (2020). Modeling spatial data using local likelihood estimation and a Matérn to spatial autoregressive translation. Environmetrics 31 e2652

    2020

  76. [84]

    , JIA, X

    WILLARD , J. , JIA, X. , XU, S. , STEINBACH , M. and KUMAR , V. (2020). Integrating physics-based modeling with machine learning: A survey. arXiv preprint arXiv:2003.04919 1 1–34

    2020 arXiv

  77. [85]

    YANG , R. (2022). Graph Matérn Fields with Applications in Inverse Problems and Machine Learning, PhD thesis, The University of Chicago

    2022

  78. [86]

    1 m X l =1 (I − P ) ¯k (I − P )T (ul , ul ) # + Z X (I − P ) ¯k (I − P )T (x, x)d x ! + Ã λSφ λ2m

    ZWILLINGER , D. and DOBRUSHKIN , V. (2021). Handbook of Differential Equations. Chapman and Hall/CRC. 28 APPENDIX A: PROOF OF THEOREM 3.2 PROOF OF (a). Given our assumption that X is compact and connected with boundary set twice-differentiable almost everywhere, the GP with Ma...

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.