pith. sign in

arxiv: 2508.12876 · v2 · submitted 2025-08-18 · 🧮 math.NA · cs.NA

Comparison of random field discretizations for high-resolution Bayesian parameter identification in finite element elasticity

Pieter Vanmechelen , Geert Lombaert , Giovanni Samaey This is my paper

Pith reviewed 2026-05-18 22:42 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords random field discretizationBayesian parameter identificationfinite element elasticitymultilevel Markov chain Monte CarloKarhunen-Loève expansionwavelet expansionlocal average subdivisionuncertainty quantification
0
0 comments X

The pith

Local average subdivision yields better mixing and lower cost-to-error ratios than Karhunen-Loève or wavelet expansions in high-resolution Bayesian elasticity parameter estimation

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

The paper compares three strategies for discretizing random fields in probabilistic identification of spatially varying material parameters within high-resolution finite element models. These are the Karhunen-Loève expansion, a wavelet expansion, and local average subdivision. The comparison is performed using multilevel Markov chain Monte Carlo on a plane stress elasticity problem with high-resolution displacement observations. All methods give similar posterior estimates, but local average subdivision demonstrates improved mixing and lower cost-to-error ratios at finer resolutions despite having more parameters. This offers practical guidance on choosing stochastic representations for uncertainty quantification in simulations of heterogeneous materials.

Core claim

In the context of multilevel Markov chain Monte Carlo for Bayesian identification of material parameters in plane stress elasticity, the local average subdivision discretization of the random field achieves improved Markov chain mixing and lower cost-to-error ratios at high resolutions compared to Karhunen-Loève and wavelet expansions, while all three approaches produce comparable posterior estimates for the parameters.

What carries the argument

Comparison of Karhunen-Loève expansion, wavelet expansion, and local average subdivision as random field discretizations within multilevel MCMC sampling for finite element based Bayesian inverse problems

Load-bearing premise

The plane stress elasticity test problem with high-resolution displacement observations and the particular multilevel MCMC implementation are representative of other finite element models and observation types

What would settle it

Observing whether the efficiency ranking changes when applying the same methods to a different finite element model such as one with three-dimensional geometry or with boundary condition observations instead of displacements

Figures

Figures reproduced from arXiv: 2508.12876 by Geert Lombaert, Giovanni Samaey, Pieter Vanmechelen.

Figure 1
Figure 1. Figure 1: Clamped beam setup, with a load placed on the center third of the top edge. Observations [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Gaussian random field generated using the KL expansion, for different values of truncation [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: One basis function each from KL (left) and wavelet (right) expansions. Note that the KL [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Gaussian random field generated using the wavelet expansion, for different values of trun [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Gaussian random field generated using the LAS method, for different number of refinement [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: shows the convergence for each of the methods. We first generate the field shown earlier in this Section using the wavelet approach for m = 5000. We then project the field onto the KL basis functions and calculate its local means for the LAS terms. Each of the three methods is then truncated after m terms and the L 2 distance to the true field is calculated. As this field has a Mat´ern covariance with smoo… view at source ↗
Figure 7
Figure 7. Figure 7: Top: Ground truth profile used for the first experiment. Generated by sampling 5000 [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Posterior mean values of the Young’s modulus field, in Pa. Top to bottom: KL, Wavelet [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Pointwise posterior standard deviations of the Young’s modulus field, in Pa. Top to [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Posterior mean values of the Young’s modulus field for the second experiment, in Pa. Top [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Pointwise posterior standard deviations of the Young’s modulus field for the second [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Left: Initialisation cost to setup each of the sampling methods prior to MCMC routine. [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Left: rejection rates at each level in the multilevel MCMC algorithm for all three methods [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Left: convergence of rejection rate in the multilevel MCMC algorithm for all three methods [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
read the original abstract

We compare three random field discretization strategies for probabilistic identification of spatially varying material parameters in high-resolution finite element models. These strategies are (i) the Karhunen-Lo\`eve expansion, (ii) a wavelet expansion, and (iii) local average subdivision. The methods are assessed in the context of multilevel Markov chain Monte Carlo applied to plane stress elasticity with high-resolution displacement observations. Emphasis is placed on numerical efficiency, initialization cost, Markov chain mixing, and cost-to-error behaviour as the discretization resolution increases. While all approaches yield comparable posterior estimates, significant differences are observed in multilevel variance reduction and sampling efficiency. In particular, local average subdivision exhibits improved mixing and lower cost-to-error ratios at fine resolutions, despite its higher nominal parameter dimension. The results provide practical guidance for selecting stochastic field representations in uncertainty quantification in finite element simulations of heterogeneous materials.

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 manuscript compares three random field discretization approaches—Karhunen-Loève expansion, wavelet expansion, and local average subdivision—for Bayesian identification of spatially varying material parameters within high-resolution finite element models. The comparison is performed in the setting of multilevel Markov chain Monte Carlo applied to a plane stress elasticity problem with dense displacement observations. The central claims are that all three methods produce comparable posterior estimates while local average subdivision exhibits superior mixing behavior and lower cost-to-error ratios at fine resolutions, despite its higher nominal dimension; these observations are used to offer practical guidance on discretization choice for uncertainty quantification in finite element simulations.

Significance. If the reported efficiency ordering proves robust, the work supplies concrete numerical evidence that can inform discretization selection in Bayesian inverse problems for heterogeneous materials. The emphasis on multilevel variance reduction and cost-to-error scaling with resolution directly addresses a practical bottleneck in high-resolution uncertainty quantification. The study also supplies a direct head-to-head evaluation against external performance metrics rather than self-referential fitting, which strengthens its utility for practitioners.

major comments (2)
  1. [§4] §4 (Numerical experiments): The abstract and results assert 'significant differences' in multilevel variance reduction, mixing, and cost-to-error ratios, yet no quantitative error bars, effective sample sizes, autocorrelation times, or convergence diagnostics are reported. Without these, the efficiency ranking cannot be verified and the central claim that LAS is preferable at fine resolutions remains unsupported by the presented data.
  2. [§5] §5 (Discussion): The practical guidance for random-field selection rests on a single model (plane stress elasticity) and a single observation type (high-resolution displacements). The observed LAS advantages could arise from interaction between local averaging and the dense observation operator or from the specific multilevel hierarchy, rather than intrinsic properties; additional test cases with varying observation sparsity or material models are required before the efficiency ordering can be treated as general.
minor comments (2)
  1. [§4] The manuscript should explicitly state the mesh resolutions, number of multilevel levels, and observation noise variance used in all experiments so that the cost-to-error curves can be reproduced.
  2. [§3] Notation for the random-field expansions (e.g., truncation level for KL and wavelet bases) should be unified across tables and figures to avoid ambiguity when comparing nominal dimensions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We respond to each major point below and indicate where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§4] §4 (Numerical experiments): The abstract and results assert 'significant differences' in multilevel variance reduction, mixing, and cost-to-error ratios, yet no quantitative error bars, effective sample sizes, autocorrelation times, or convergence diagnostics are reported. Without these, the efficiency ranking cannot be verified and the central claim that LAS is preferable at fine resolutions remains unsupported by the presented data.

    Authors: We agree that the presentation would be strengthened by additional quantitative diagnostics. In the revised version we will report effective sample sizes, integrated autocorrelation times, and Gelman-Rubin statistics for the multilevel chains under each discretization. Cost-to-error ratios will be accompanied by standard deviations obtained from independent replicate runs. These additions will allow direct verification of the reported differences in mixing and efficiency. revision: yes

  2. Referee: [§5] §5 (Discussion): The practical guidance for random-field selection rests on a single model (plane stress elasticity) and a single observation type (high-resolution displacements). The observed LAS advantages could arise from interaction between local averaging and the dense observation operator or from the specific multilevel hierarchy, rather than intrinsic properties; additional test cases with varying observation sparsity or material models are required before the efficiency ordering can be treated as general.

    Authors: The chosen test case is a standard high-resolution plane-stress problem with dense displacement data, representative of many engineering inverse problems. We will expand the discussion to explicitly note that the observed LAS advantages may interact with the dense observation operator and the particular multilevel hierarchy. We will also outline how the same comparison framework could be applied to other observation densities or constitutive models. Full additional experiments lie outside the present computational budget but are identified as valuable future work. revision: partial

Circularity Check

0 steps flagged

No circularity: direct numerical comparison of discretization methods

full rationale

The paper conducts an empirical numerical study comparing three random field discretization approaches (Karhunen-Loève expansion, wavelet expansion, and local average subdivision) within multilevel MCMC for Bayesian identification in plane stress elasticity. Reported outcomes on comparable posteriors, mixing behavior, variance reduction, and cost-to-error ratios are obtained from direct simulation against external metrics such as chain mixing time and computational cost. No equations or claims reduce by construction to fitted inputs, self-definitions, or self-citation chains; the central results derive from independent numerical experiments on the chosen test problem rather than from any internal redefinition or renaming of prior results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The comparison relies on standard assumptions of Gaussian random fields and linear elasticity but introduces no new free parameters, axioms, or invented entities beyond the three named discretization techniques.

pith-pipeline@v0.9.0 · 5680 in / 1120 out tokens · 38506 ms · 2026-05-18T22:42:55.162199+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

31 extracted references · 31 canonical work pages · 3 internal anchors

  1. [1]

    Angelikopoulos, C

    P. Angelikopoulos, C. Papadimitriou, and P. Koumoutsakos. X-TMCMC: Adaptive kriging for Bayesian inverse modeling. Computer Methods in Applied Mechanics and Engineering , 289:409–428, June 2015

    work page 2015

  2. [2]

    Representations of Gaussian random fields and approximation of elliptic PDEs with lognormal coefficients

    M. Bachmayr, A. Cohen, and G. Migliorati. Representations of Gaussian random fields and approximation of elliptic PDEs with lognormal coefficients, Mar. 2016. arXiv:1603.05559 [math]

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  3. [3]

    Blondeel, P

    P. Blondeel, P. Robbe, C. Van hoorickx, S. Francois, G. Lombaert, and S. Vandewalle. p-Refined Multilevel Quasi-Monte Carlo for Galerkin Finite Element methods with applications in civil engineering. Algorithms, 13(5):110–139, 2020

    work page 2020

  4. [4]

    S. C. Brenner and L. R. Scott. The mathematical theory of finite element methods . Number 15 in Texts in applied mathematics. Springer, New York, NY, 3. ed., [nachdr.] edition, 2010

    work page 2010

  5. [5]

    Charrier, R

    J. Charrier, R. Scheichl, and A. L. Teckentrup. Finite Element Error Analysis of Elliptic PDEs with Random Coefficients and Its Application to Multilevel Monte Carlo Methods. SIAM Journal on Numerical Analysis , 51(1):322–352, Jan. 2013

    work page 2013

  6. [6]

    Ching and Y.-C

    J. Ching and Y.-C. Chen. Transitional Markov Chain Monte Carlo Method for Bayesian Model Updating, Model Class Selection, and Model Averaging. Journal of Engineering Mechanics, 133(7):816–832, 2007. 18 IOP Publishing Inverse Problems vv (yyyy) aaaaaa P Vanmechelen et al

    work page 2007

  7. [7]

    J. A. Christen and C. Fox. Markov chain Monte Carlo Using an Approximation. Journal of Computational and Graphical Statistics , 14(4):795–810, Dec. 2005

    work page 2005

  8. [8]

    K. A. Cliffe, M. B. Giles, R. Scheichl, and A. L. Teckentrup. Multilevel Monte Carlo methods and applications to elliptic PDEs with random coefficients. Computing and Visualization in Science, 14(1):3–15, Jan. 2011

    work page 2011

  9. [9]

    Cotter, G

    S. Cotter, G. Roberts, A. M. Stuart, and D. White. Mcmc methods for functions: Modifying old algorithms to make them faster. Statistical Science, 28(3):424–446, 2013

    work page 2013

  10. [10]

    T. Cui, G. Detommaso, and R. Scheichl. Multilevel dimension-independent likelihood-informed MCMC for large-scale inverse problems. Inverse Problems, 40(3):035005, Mar. 2024

    work page 2024

  11. [11]

    T. Cui, J. Martin, Y. M. Marzouk, A. Solonen, and A. Spantini. Likelihood-informed dimension reduction for nonlinear inverse problems. Inverse Problems, 30(11):114015, Nov

    work page

  12. [12]

    Publisher: IOP Publishing

    work page

  13. [13]

    T. J. Dodwell, C. Ketelsen, R. Scheichl, and A. L. Teckentrup. A Hierarchical Multilevel Markov Chain Monte Carlo Algorithm with Applications to Uncertainty Quantification in Subsurface Flow. SIAM/ASA Journal on Uncertainty Quantification , 3(1):1075–1108, Jan. 2015

    work page 2015

  14. [14]

    T. J. Dodwell, C. Ketelsen, R. Scheichl, and A. L. Teckentrup. Multilevel Markov Chain Monte Carlo. SIAM Review, 61(3):509–545, Jan. 2019

    work page 2019

  15. [15]

    G. A. Fenton and E. H. Vanmarcke. Simulation of Random Fields via Local Average Subdivision. Journal of Engineering Mechanics , 116(8):1733–1749, Aug. 1990

    work page 1990

  16. [16]

    emcee: The MCMC Hammer

    D. Foreman-Mackey, D. W. Hogg, D. Lang, and J. Goodman. emcee: The MCMC Hammer. Publications of the Astronomical Society of the Pacific , 125(925):306–312, Mar. 2013. arXiv: 1202.3665

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  17. [17]

    R. G. Ghanem and P. D. Spanos. Stochastic Finite Element Method: Response Statistics. In R. G. Ghanem and P. D. Spanos, editors, Stochastic Finite Elements: A Spectral Approach . Springer, New York, NY, 1991

    work page 1991

  18. [18]

    M. B. Giles. Multilevel Monte Carlo Path Simulation. Operations Research, 56(3):607–617, June 2008

    work page 2008

  19. [19]

    M. B. Giles. Multilevel Monte Carlo methods. Acta Numerica, 24:259–328, May 2015

    work page 2015

  20. [20]

    Hastings

    W. Hastings. Monte-Carlo sampling methods using Markov chains and their applications. Biometrika, 57(1):97–109, 1970

    work page 1970

  21. [21]

    Helfrick, C

    M. Helfrick, C. Niezrecki, P. Avitabile, and T. Schmidt. 3D digital image correlation methods for full-field vibration measurement. Mechanical Systems and Signal Processing , 25:917–927, 2011

    work page 2011

  22. [22]

    V. H. Hoang, J. H. Quek, and C. Schwab. Analysis of a multilevel Markov chain Monte Carlo finite element method for Bayesian inversion of log-normal diffusions. Inverse Problems, 36(3):035021, Mar. 2020

    work page 2020

  23. [23]

    J. Latz, I. Papaioannou, and E. Ullmann. Multilevel Sequential2 Monte Carlo for Bayesian Inverse Problems. Journal of Computational Physics , 368:154–178, Sept. 2018. arXiv: 1709.09763

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    K. Law, A. Stuart, and K. Zygalakis. Data Assimilation: A Mathematical Introduction , volume 62 of Texts in Applied Mathematics . Springer International Publishing, Cham, 2015

    work page 2015

  25. [25]

    Y. Liu, J. Li, S. Sun, and B. Yu. Advances in gaussian random field generation: a review. Computational Geosciences, 23:1011–1047, Aug. 2019

    work page 2019

  26. [26]

    G. J. Lord, C. E. Powell, and T. Shardlow. An Introduction to Computational Stochastic PDEs. Cambridge Texts in Applied Mathematics. Cambridge University Press, New York, NY, 2014. 19 IOP Publishing Inverse Problems vv (yyyy) aaaaaa P Vanmechelen et al

    work page 2014

  27. [27]

    M. B. Lykkegaard, G. Mingas, R. Scheichl, C. Fox, and T. J. Dodwell. Multilevel Delayed Acceptance MCMC with an Adaptive Error Model in PyMC3. arXiv:2012.05668 [stat], Dec

    work page arXiv 2012

  28. [28]

    Nuttall, D

    J. Nuttall, D. Parallel Implementation and Application of the Random Finite Element Method . The University of Manchester (UK), 2011

    work page 2011

  29. [29]

    Schwab and R

    C. Schwab and R. A. Todor. Karhunen Loeve approximation of random fields by generalized fast multipole methods. Journal of Computational Physics , 217(1):100–122, Sept. 2006

    work page 2006

  30. [30]

    Uribe, I

    F. Uribe, I. Papaioannou, W. Betz, and D. Straub. Bayesian inference of random fields represented with the Karhunen Loeve expansion. Comput. Methods Appl. Mech. Engrg. , 358, 2020

    work page 2020

  31. [31]

    Vanmechelen, G

    P. Vanmechelen, G. Lombaert, and G. Samaey. Multilevel Markov Chain Monte Carlo with Likelihood Scaling for Bayesian Inversion with High-resolution Observations. SIAM/ASA Journal on Uncertainty Quantification , 13(2):375–399, June 2025. 20

    work page 2025