pith. sign in

arxiv: 2605.17168 · v1 · pith:73CGM75Hnew · submitted 2026-05-16 · 🧮 math.NA · cs.NA

Rational approximation and intrinsic Gaussian processes

Christopher Beattie , David Higdon , Leanna House , Colby Stakun-Pickering , Jared Clark This is my paper

Pith reviewed 2026-05-20 14:11 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords intrinsic Gaussian processesrational approximationspatial modelingvariogram modelsinference algorithmsnon-stationary processesGaussian processes
0
0 comments X

The pith

Intrinsic Gaussian processes can be modeled via rational approximation to yield practical algorithms and variogram models that parallel those for stationary GPs.

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

The paper develops a systematic framework for intrinsic Gaussian processes by linking them to rational approximation methods. This connection clarifies the underlying structure and supports the creation of practical algorithms for modeling, inference, and computation. It also supplies dependence and variogram models that function in ways comparable to those in traditional covariance-based approaches. A sympathetic reader would care because the framework addresses long-standing barriers to adopting intrinsic GPs for spatial data, potentially improving robustness and efficiency where stationarity assumptions do not hold.

Core claim

Gaussian processes defined through intrinsic random fields provide a flexible framework for modeling spatial phenomena but have lacked accompanying computational methods and dependence specifications. A close connection between intrinsic GP models and rational approximation clarifies the problem structure, enabling practical algorithms and variogram models for modeling, inference, and computation that parallel those of traditional stationary GPs. Numerical examples show deployment in practice and highlight advantages in robustness, interpretability, and computational efficiency.

What carries the argument

The connection between intrinsic GP models and rational approximation, which clarifies the problem structure to support development of algorithms and dependence specifications.

If this is right

  • Practical algorithms become available for fitting and prediction with intrinsic GPs.
  • Dependence and variogram models can be specified to work in parallel with stationary cases.
  • Intrinsic-field modeling gains advantages in robustness and computational efficiency for spatial applications.
  • The framework facilitates handling of spatial phenomena where traditional covariance approaches are less suitable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This approach may extend GP use to larger or more irregular spatial domains by reducing reliance on stationarity.
  • Rational approximation techniques could transfer to other classes of non-stationary random fields.
  • Efficiency gains might support real-time updating in dynamic spatial monitoring tasks.

Load-bearing premise

The link between intrinsic GP models and rational approximation is sufficiently clear to allow development of practical algorithms and dependence specifications that perform as described.

What would settle it

A spatial dataset where the rational-approximation-based variogram models produce invalid covariance structures or yield predictions less accurate than those from standard stationary GP methods.

read the original abstract

Gaussian processes (GPs) defined through intrinsic random fields provide a flexible framework for modeling spatial phenomena, and have been advocated in a variety of applications over the past several decades. Nevertheless, their adoption has lagged behind traditional, covariance-based approaches, in part because the intrinsic formulation has lacked an accompanying toolkit of computational methods and dependence specifications that facilitate fitting and prediction. We develop here a systematic framework for modeling intrinsic GPs and introduce practical algorithms and dependence/variogram models for modeling, inference and computation that parallel those of traditional, stationary GPs. We explore a close connection between intrinsic GP models and rational approximation, which clarifies the underlying problem structure. Numerical examples illustrate how the new tools can be deployed in practice, highlighting the advantages of intrinsic-field modeling in terms of robustness, interpretability, and computational efficiency.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper develops a systematic framework for intrinsic Gaussian processes by establishing a connection to rational approximation of covariance functions or spectral densities. It introduces practical algorithms, dependence/variogram models, and computational methods for modeling, inference, and prediction that are designed to parallel those available for stationary GPs while respecting the intrinsic polynomial null space of order k.

Significance. If the rational-approximation link is shown to yield exactly intrinsic kernels (i.e., kernels that annihilate polynomials up to the required degree), the work supplies a missing practical toolkit for intrinsic random fields. This could improve robustness and interpretability for spatial data exhibiting polynomial trends, with potential gains in computational efficiency over ad-hoc detrending approaches.

major comments (2)
  1. [§3.2, Eq. (8)] §3.2 and Eq. (8): the rational approximation of the covariance is performed without an explicit constraint (e.g., on pole locations or numerator/denominator degrees) that guarantees the resulting kernel lies in the orthogonal complement of polynomials of degree ≤ k. The manuscript must demonstrate that the constructed kernels are exactly intrinsic rather than approximately so; otherwise the claimed robustness and interpretability advantages rest on an unverified property.
  2. [§5, Algorithm 1] §5, Algorithm 1: the inference procedure for the variogram parameters appears to treat the intrinsic null-space coefficients as fixed rather than jointly estimated; if the rational approximation does not automatically enforce the null-space condition, this step may introduce bias in the dependence estimates that is not quantified in the numerical examples.
minor comments (2)
  1. [Figure 2] Figure 2: the legend does not distinguish the rational-approximation variogram from the empirical one; add a clear marker or line style.
  2. Notation: the symbol k is used both for the intrinsic order and for the number of poles in the rational approximant; a distinct symbol for the latter would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments, which have helped us identify areas where the manuscript can be strengthened. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§3.2, Eq. (8)] §3.2 and Eq. (8): the rational approximation of the covariance is performed without an explicit constraint (e.g., on pole locations or numerator/denominator degrees) that guarantees the resulting kernel lies in the orthogonal complement of polynomials of degree ≤ k. The manuscript must demonstrate that the constructed kernels are exactly intrinsic rather than approximately so; otherwise the claimed robustness and interpretability advantages rest on an unverified property.

    Authors: We agree that an explicit demonstration is necessary to fully substantiate the claim that the kernels are exactly intrinsic. The rational approximations in our framework are constructed via the spectral density of the variogram, which by design annihilates the polynomial null space of order k for the models considered; this follows from the fact that the approximants preserve the singularity structure at the origin that corresponds to the intrinsic property. To make this rigorous and address the referee's concern, we will add a new proposition in Section 3.2 (with proof) establishing that the resulting kernels lie exactly in the orthogonal complement of polynomials up to degree k. We will also include a brief numerical check confirming exact annihilation in the revised manuscript. This revision will be made. revision: yes

  2. Referee: [§5, Algorithm 1] §5, Algorithm 1: the inference procedure for the variogram parameters appears to treat the intrinsic null-space coefficients as fixed rather than jointly estimated; if the rational approximation does not automatically enforce the null-space condition, this step may introduce bias in the dependence estimates that is not quantified in the numerical examples.

    Authors: We thank the referee for this observation. In the intrinsic GP formulation, the null-space coefficients correspond to the deterministic polynomial trend and are estimated separately via generalized least squares (or equivalent) to maintain identifiability with the stochastic component; this is standard in the literature on intrinsic random fields and does not rely on the rational approximation enforcing the condition. Nevertheless, we acknowledge the value of quantifying any potential effects. In the revision we will clarify this separation in the description of Algorithm 1, add a short discussion of why joint estimation is typically avoided, and include additional numerical results that compare fixed versus jointly estimated coefficients to assess bias in the dependence parameters. revision: partial

Circularity Check

0 steps flagged

No significant circularity; framework extends intrinsic random field concepts with independent algorithmic contributions

full rationale

The paper develops a systematic framework for intrinsic GPs by exploring their connection to rational approximation and introducing parallel algorithms, dependence/variogram models, and computational methods. No load-bearing step reduces by construction to fitted inputs, self-definitions, or self-citation chains; the rational approximation link is presented as a clarifying tool rather than a tautological reparameterization, and the new models are derived to respect intrinsic null spaces while paralleling stationary GP practice. The derivation remains self-contained against external benchmarks of intrinsic random fields, with no evidence that central claims equate to their own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the contribution is described at the level of framework development and algorithms without detailing specific fitted quantities or new postulates.

pith-pipeline@v0.9.0 · 5667 in / 1121 out tokens · 46880 ms · 2026-05-20T14:11:18.641465+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    Advances in applied probability5(3), 439–468 (1973)

    Matheron, G.: The intrinsic random functions and their applications. Advances in applied probability5(3), 439–468 (1973)

    work page 1973

  2. [2]

    Journal of the Southern African Institute of Mining and Metallurgy76(9), 383–391 (1976)

    Krige, D.G.: Some basic considerations in the application of geostatistics to the valuation of ore in south african gold mines. Journal of the Southern African Institute of Mining and Metallurgy76(9), 383–391 (1976)

    work page 1976

  3. [3]

    Academic Press, New York (1978)

    Journel, A.G., Huijbregts, C.J.: Mining Geostatistics. Academic Press, New York (1978)

    work page 1978

  4. [4]

    John Wiley & Sons, ??? (2015)

    Cressie, N.: Statistics for Spatial Data. John Wiley & Sons, ??? (2015)

    work page 2015

  5. [5]

    Williams, C.K., Rasmussen, C.E.: Gaussian Processes for Machine Learning vol

    work page

  6. [6]

    MIT Press, Cambridge, MA (2006)

    work page 2006

  7. [7]

    Chapman and Hall/CRC, ??? (2020)

    Gramacy, R.B.: Surrogates: Gaussian Process Modeling, Design, and Optimiza- tion for the Applied Sciences. Chapman and Hall/CRC, ??? (2020)

    work page 2020

  8. [8]

    Technometrics48(4), 458–466 (2006)

    Joseph, V.R.: Limit kriging. Technometrics48(4), 458–466 (2006)

    work page 2006

  9. [9]

    Statistical science, 3–41 (1995) 23

    Besag, J., Green, P., Higdon, D., Mengersen, K.: Bayesian computation and stochastic systems. Statistical science, 3–41 (1995) 23

    work page 1995

  10. [10]

    International Statis- tical Review71(2), 181–199 (2003)

    Wikle, C.K.: Hierarchical models in environmental science. International Statis- tical Review71(2), 181–199 (2003)

    work page 2003

  11. [11]

    Journal of the American Statistical Association 120(549), 548–558 (2025)

    Joseph, V.R.: Rational kriging. Journal of the American Statistical Association 120(549), 548–558 (2025)

    work page 2025

  12. [12]

    Technometrics53(3), 254–265 (2011)

    Joseph, V.R., Kang, L.: Regression-based inverse distance weighting with appli- cations to computer experiments. Technometrics53(3), 254–265 (2011)

    work page 2011

  13. [13]

    In: Proceedings of the 1968 23rd ACM National Conference, pp

    Shepard, D.: A two-dimensional interpolation function for irregularly-spaced data. In: Proceedings of the 1968 23rd ACM National Conference, pp. 517–524 (1968)

    work page 1968

  14. [14]

    Cambridge University Press, New York (2013)

    Horn, R.A., Johnson, C.R.: Matrix Analysis, 2nd edn. Cambridge University Press, New York (2013)

    work page 2013

  15. [15]

    Electron

    Beattie, C.: Harmonic Ritz and Lehmann bounds. Electron. Trans. Numer. Anal 7, 18–39 (1998)

    work page 1998

  16. [16]

    SIAM Journal on Matrix Analysis and Applications 19(4), 1097–1110 (1998)

    Cheng, S.H., Higham, N.J.: A modified cholesky algorithm based on a symmet- ric indefinite factorization. SIAM Journal on Matrix Analysis and Applications 19(4), 1097–1110 (1998)

    work page 1998

  17. [17]

    In: 2018 IEEE Aerospace Conference, pp

    Vaze, P., Kaki, S., Limonadi, D., Esteban-Fernandez, D., Zohar, G.: The surface water and ocean topography mission. In: 2018 IEEE Aerospace Conference, pp. 1–9 (2018). IEEE

    work page 2018

  18. [18]

    Ubelmann, L.G., Fu, L.-L.: SWOT Simulator for Ocean Science (2021)

    C. Ubelmann, L.G., Fu, L.-L.: SWOT Simulator for Ocean Science (2021). https://swot-simulator.readthedocs.io/en/doc-initiate doc/science.html Accessed 2026-02-28

    work page 2021

  19. [19]

    Springer, New York (2006)

    Nocedal, J., Wright, S.J.: Numerical Optimization. Springer, New York (2006)

    work page 2006

  20. [20]

    Handbook of Markov Chain Monte Carlo, 449–478 (2011)

    Haran, M.: Gaussian random field models for spatial data. Handbook of Markov Chain Monte Carlo, 449–478 (2011)

    work page 2011

  21. [21]

    SIAM journal on scientific and statistical computing 8(3), 210–221 (1987)

    Bojanczyk, A.W., Brent, R., Van Dooren, P., De Hoog, F.: A note on downdating the cholesky factorization. SIAM journal on scientific and statistical computing 8(3), 210–221 (1987)

    work page 1987

  22. [22]

    SIAM, Philadelphia (1998)

    Stewart, G.W.: Matrix Algorithms: Volume 1: Basic Decompositions. SIAM, Philadelphia (1998)

    work page 1998

  23. [23]

    Remote Sensing15(17), 4277 (2023)

    Yaremchuk, M., Beattie, C., Panteleev, G., D’Addezio, J.M., Smith, S.: The effect of spatially correlated errors on sea surface height retrieval from SWOT altimetry. Remote Sensing15(17), 4277 (2023)

    work page 2023

  24. [24]

    Remote 24 Sensing16(11), 1954 (2024)

    Yaremchuk, M., Beattie, C., Panteleev, G., D’Addezio, J.: Block-circulant approx- imation of the precision matrix for assimilating SWOT altimetry data. Remote 24 Sensing16(11), 1954 (2024)

    work page 1954

  25. [25]

    John Wiley & Sons, New York (2012)

    Chiles, J.-P., Delfiner, P.: Geostatistics: Modeling Spatial Uncertainty. John Wiley & Sons, New York (2012)

    work page 2012

  26. [26]

    Computers & Geosciences32(10), 1615–1628 (2006)

    Emery, X., Lantu´ ejoul, C.: Tbsim: A computer program for conditional simula- tion of three-dimensional gaussian random fields via the turning bands method. Computers & Geosciences32(10), 1615–1628 (2006)

    work page 2006

  27. [27]

    Journal of the Royal Statistical Society Series B: Statistical Methodology 50(2), 297–312 (1988)

    Vecchia, A.V.: Estimation and model identification for continuous spatial pro- cesses. Journal of the Royal Statistical Society Series B: Statistical Methodology 50(2), 297–312 (1988)

    work page 1988

  28. [28]

    Statistical Science36(1), 124–141 (2021)

    Katzfuss, M., Guinness, J.: A general framework for vecchia approximations of gaussian processes. Statistical Science36(1), 124–141 (2021)

    work page 2021

  29. [29]

    In: Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, pp

    Murray, I., Adams, R., MacKay, D.: Elliptical slice sampling. In: Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, pp. 541–548 (2010). JMLR Workshop and Conference Proceedings 25

    work page 2010