Pith. sign in

REVIEW 4 major objections 8 minor 55 references

Domain-decomposed neural surrogates cut the forecast cost of ensemble Kalman filters for high-dimensional PDE parameter ID while still matching full-model and MCMC posteriors.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 18:38 UTC pith:Y6HDRBPO

load-bearing objection Solid DDM-NN forecast surrogate for 3D elastic parameter ID; the decentralized Kalman analysis that carries the scalability claim is derived but never actually run. the 4 major comments →

arxiv 2607.24305 v1 pith:Y6HDRBPO submitted 2026-07-27 cs.CE

Domain-Decomposition Neural Surrogates for Scalable Decentralized Ensemble Kalman Filter Based Parameter Identification in High-Dimensional Stochastic PDEs

Timm G\"odde , Bojana Rosi\'c This is my paper
classification cs.CE
keywords ensemble Kalman filterdomain decompositionneural network surrogateparameter identificationstochastic PDEsdecentralized estimationaugmented Lagrangianfinite element method
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that you can replace the expensive sample-based forecast step inside an ensemble Kalman filter with local neural networks trained on subdomains of a high-dimensional spatial domain, then couple those networks with an augmented-Lagrange interface scheme so the global field stays continuous. The same partition is used to turn the Kalman update itself into local gains plus a cheap iterative correction for cross-subdomain correlations, so neither the surrogate nor the covariance inversion has to act on the full observation dimension. On a three-dimensional linear-elastic cylinder whose bulk and shear moduli are inferred from noisy surface displacements, the resulting DDM-NN Kalman filter recovers posterior means and spreads that approach both a pure finite-element EnKF and a long MCMC reference, provided the surrogate modelling error is folded into the forecast covariance. A sympathetic reader cares because full-field material identification and digital-twin updating routinely hit the wall of repeated high-fidelity solves; the method claims a practical route to keep uncertainty quantification while making both the offline surrogate and the online filter scale with subdomain size rather than global mesh size.

Core claim

A domain-decomposition neural surrogate, trained subdomain-wise under augmented-Lagrange continuity constraints and then embedded in a block-diagonal preconditioned ensemble Kalman update, captures the posterior of material parameters in a three-dimensional stochastic elasticity problem and approaches both pure EnKF and MCMC reference solutions while substantially reducing forecast-model cost.

What carries the argument

Augmented-Lagrange DDM-NN surrogate plus decentralized EnKF: local networks map (coordinates, parameters) to local fields; interface networks and dual ascent enforce C1 continuity; the Kalman correction is rewritten as a preconditioned Richardson iteration whose block-diagonal preconditioner is the set of local Kalman gains, recovering global coupling without inverting the full observation covariance.

Load-bearing premise

Every measurement is assumed to depend only on the solution inside one subdomain, so local observation operators and block-diagonal gains stay well-defined without sensors that mix distant regions.

What would settle it

Replace the surface point sensors with a single integral or long-range observation that couples both cylinder halves; if the decentralized update then fails to recover the MCMC/EnKF posterior means and spreads (or the claimed complexity reduction disappears), the central scalability claim does not hold as derived.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Offline training cost and number of network parameters scale with subdomain size rather than full mesh size, enabling larger three-dimensional identification problems on modest hardware.
  • Online Kalman updates avoid O(Ny³) global covariance inversion, replacing it by M independent local inversions plus a few Richardson iterations.
  • Including the empirical surrogate-error covariance prevents over-confident, too-narrow posteriors and keeps means within roughly half a percent of truth on the cylinder benchmark.
  • The same partition can be reused for sequential or time-dependent EnKF once local time-dependent surrogates exist.
  • Regions of near-zero displacement can be dropped from assimilation without harming the inferred material-parameter posteriors.

Where Pith is reading between the lines

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

  • If the locality assumption is relaxed by a sparse long-range observation graph, the same Richardson skeleton could still work with a few extra off-diagonal blocks rather than a full dense gain.
  • The method’s reliance on an offline FEM training set suggests a natural hybrid: start with a coarse global network, then refine only subdomains whose local residual or KLD stays high.
  • Adaptive estimation of the model-error covariance inside the filter, rather than a fixed validation estimate, is a direct next experiment the architecture already supports.
  • The same DDM-NN forecast could feed other ensemble smoothers or variational schemes that also bottleneck on repeated forward solves.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 8 minor

Summary. The manuscript addresses parameter identification for stochastic PDE-based models by coupling (i) a neural-network surrogate of the parameter-to-observation map, decomposed over spatial subdomains via an augmented-Lagrangian DDM with C^1 interface conditions, with (ii) an ensemble Gauss-Markov-Kalman update in which the surrogate model error is estimated empirically from validation residuals and added to the measurement-noise covariance. Section 5 further proposes a decentralized analysis step: the global observation-covariance solve is reformulated as a linear system (Eq. 56) and attacked with a Richardson iteration preconditioned by the block-diagonal (per-subdomain) covariance (Eqs. 58–59), for which an O(N_y^3/M^2) complexity reduction is claimed. The method is demonstrated on a 3D linear-elastic cylinder (1,800 hex elements) with two unknowns (bulk and shear moduli), trained on 10,000 FEM samples. Posteriors from the DDM-NN filter (with and without model-error covariance) are compared against a pure FEM-EnKF (10,000 samples) and a 100,000-sample Metropolis-Hastings MCMC reference, with ablations over measurement noise (1–10%), ensemble size (100–10,000), and sensor count (100–600). With model error included, posterior means are within ~0.5% of truth and spreads match the FEM-EnKF; without it, the posterior is overconfident.

Significance. If the central claims hold, the paper offers a practically useful combination: a DDM-trained NN forecast surrogate with an empirically calibrated model-error covariance plugged into an ensemble Kalman update, validated on a 3D problem against both an independent MCMC reference (100k Metropolis-Hastings samples on the FEM model) and a surrogate-free FEM-EnKF — a validation standard above the norm for surrogate-filter papers. The ablation over measurement noise (1–10%), ensemble size (100–10k), and sensor count (100–600) gives the empirical model-error treatment real evidential weight: with the error covariance included, posterior means stay within ~0.5% of truth and spreads track the FEM-EnKF; without it, the filter is visibly overconfident (truth in the posterior tail, Fig. 18). The result is not parameter-free — architectures, penalty rho, noise levels, and the empirical error covariance are all tuned or estimated — but the tuning is disclosed. The significance is, however, capped by the gap between what is demonstrated (a 2-parameter, 600-observation problem solved by the direct global update) and what is claimed in the title and abstract (scalable, decentralized, high-dimensiona

major comments (4)
  1. [§5, Eqs. (56)–(61); §6.3] The preconditioned Richardson update of §5 (Eqs. 58–61) is never exercised numerically. Section 6.3 states twice that the filter uses 'Eq. (57)', the direct auxiliary-variable form, which still requires solving Eq. (56) with the full 600x600 observation covariance. No Richardson iteration counts, relaxation parameter alpha, residual decay curves, or local-vs-global timing comparisons are reported anywhere. The paper's title and abstract claim a 'scalable decentralized' Kalman filter and §5 claims an O(N_y^3/M^2) complexity reduction, but every posterior in Figs. 17-21 and Tables 6-7 is produced by the direct global solve. The decentralized analysis step is one of the two stated contributions ('addresses both computational bottlenecks', §1) and it is currently asserted, not demonstrated. This must either be demonstrated numerically (iteration counts, alpha, accuracy vs. the direct solve,
  2. [§5, Eqs. (58)–(59)] No convergence condition is given for the iteration in Eq. (59). Richardson iteration converges only if the spectral radius of I - alpha*Ctilde^{-1}C is below 1, and for this problem class that is a substantive question, not a formality: the unknowns are two global scalar parameters (kappa, mu) shared by both subdomains, so the forecast observation covariance is dominated by a low-rank term in which displacements in the upper and lower halves correlate strongly through the same parameters. The off-diagonal blocks Cov(y^{f,i}, y^{f,j}) dropped in Eq. (58) therefore carry the same physical signal as the retained diagonal blocks, and the block-diagonal preconditioner is structurally a coarse approximation of C here. A small-alpha, many-iteration regime would also erode the M^2 saving, since each iteration applies the full covariance in the residual. The authors should give a convergence con
  3. [§5 (complexity paragraph after Eq. 61); §6] The demonstrated problem size does not support the 'high-dimensional' and 'scalable' framing, and the stated motivation for §5 is partially self-defeating in the demonstrated regime. The inverse problem has two unknown parameters and 600 point observations; the O(N_y^3) inversion cited as the bottleneck is a 600x600 matrix, which is trivially cheap, and for parameter-space updates the standard ensemble/Woodbury formulation reduces the gain computation to an inversion in parameter dimension (2x2) regardless of N_y. The decentralized machinery of §5 would only pay off for high-dimensional field inversion, which is not shown. Relatedly, while the surrogate's forecast speed-up is plausible (10,000 NN evaluations vs. 10,000 LS-DYNA solves), no wall-clock comparison of the forecast step within the filter is reported — only hyperparameter-search training times (Tables 3-5). The complexity and c
  4. [§3, Eq. (30); §6.3] The model-error covariance entering the gain in Eq. (52) is estimated per spatial point (Eq. (30) defines Cov_eps(x_i) as a scalar variance), which suggests spatial correlations of the model error are neglected; this should be stated explicitly and justified, since surrogate errors of a smooth displacement field are typically spatially correlated, and correlated model error would change the posterior spread. There is also an internal inconsistency: §3 prescribes estimating epsilon_model from validation data, while §6.3 states the error is quantified 'between the training data and the surrogate model predictions'. Training residuals systematically underestimate generalization error; if training data were used, the experiment should be repeated on the held-out test set, or the text corrected.
minor comments (8)
  1. [§6.2, Fig. 16] Fig. 16 caption: the interface point is given as '(x,y,z) = (5,-8,66, 35)' — presumably (5, -8.66, 35).
  2. [§4, Eq. (41); §6, Eq. (67)] Eq. (41) contains malformed notation ('u^int_{f,ij}((ω)'); Eq. (67), the definition of the symmetric gradient operator, is garbled as written.
  3. [Various] Typos: 'posteriorors' (§6.3, paragraph describing Fig. 17), 'atleast' (§6.2), 'enforceC^k' (§1), 'a augmented Lagrange multiplier' (Abstract), 'bonded' for 'bounded' (§6, description of the elasticity tensor). Decimal separators are inconsistent ('10.000 test samples' vs '10,000').
  4. [§6, Table 1] Table 1: the parameter variances are listed as '10' with units GPa in the header; please state the variance units (GPa^2) and give the prior distribution parameters explicitly (the lognormal mean/variance used to generate training samples).
  5. [§6.2–6.3] §6.2: the Kullback-Leibler criterion is used to exclude near-zero-displacement measurement points from the filter. Please state quantitatively how many of the candidate points were excluded and how the 600/300/100 sensor locations were selected; the posterior results could be sensitive to this choice.
  6. [§6.1] §6.1: the 'combined validation loss' of the two local NNs (9.34e-5) is compared against the global NN (8.24e-5); clarify how the combination is computed (sum or weighted average over subdomains) to make the comparison well-defined.
  7. [§6.3, Tables 6–7] §6.3: the large gap between the EnKF posterior spread and the MCMC spread (e.g., sigma(mu) = 1,146 vs. 25 in Table 7) is attributed to problem nonlinearity; a short discussion of why the Gaussian/linear-regression approximation inflates the spread by an order of magnitude here would strengthen the interpretation, since the forward map is nearly linear in the parameters.
  8. [§4, Eq. (52); §3, Eq. (29)] Notation: Cov(eps_model + eps_f) in Eq. (52) should be Cov(eps_model + eps_f, eps_model + eps_f) for consistency with the definitions in §2; similarly check Eq. (29).

Circularity Check

0 steps flagged

No significant circularity: posteriors are validated against independent MCMC and pure FEM-EnKF; self-citation to prior DDM-NN training is methodological reuse, not a load-bearing uniqueness claim.

full rationale

The paper's central empirical claim—that a DDM-NN forecast surrogate plus EnKF recovers bulk/shear posterior means and spreads comparable to pure FEM-EnKF and approaching MCMC—is checked against external benchmarks that never see the surrogate (MCMC on the FEM model; sampling EnKF on FEM). Training data and model-error covariances come from the same FEM code, which is ordinary surrogate practice, not fitting the target posterior by construction. The GMKF/EnKF update (Eqs. 11–14, 28–29, 52–57) is the standard conditional-expectation derivation; the DDM-NN construction (augmented Lagrangian, local NNs, interface C1 constraints) is an approximation method whose accuracy is measured by relative error, KLD, and held-out samples, not redefined as the answer. Self-citation [22] supplies the authors' prior DDM-NN training algorithm used in §4; that is reuse of a method, not a uniqueness theorem or ansatz that forces the parameter-ID result. The skeptic concern that §5's Richardson/block-diagonal analysis step is under-demonstrated is a completeness/correctness issue, not circularity. No equation reduces the reported posterior to a fitted input by definition.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The claim rests on classical Kalman/conditional-expectation theory, standard FEM well-posedness for linear elasticity, the authors’ prior ALM-DDM training algorithm, and several modeling choices (local observations, lognormal independent priors, Gaussian additive model+measurement error, block-diagonal neglect of cross-subdomain observation covariances inside the preconditioner). No new physical entities are postulated. Free parameters are architectural and algorithmic knobs selected by validation loss and stability, not physical constants fitted to the posterior target.

free parameters (6)
  • Local NN architectures (width×depth) and global NN reference = NN1: 20×6 (2283 params); NN2: 20×7 (2703); global ref 40×6
    Chosen by grid search on validation MSE then manual pick of ‘third-best’ configs (20×6 and 20×7 swish) to leave capacity for interface constraints; directly controls surrogate bias that enters the filter.
  • Augmented-Lagrangian penalty ρ = 1e-3
    Hand-set weight on interface constraint violation in the local Lagrangians; affects continuity quality and thus observation residual statistics.
  • Richardson relaxation α = not numerically reported in results
    Step size in the preconditioned iteration that recovers cross-subdomain covariance coupling; convergence of the decentralized analysis depends on it.
  • Number of subdomains M and interface treatment = M=2
    Problem is split into M=2 stacked halves; complexity claims and continuity burden scale with M and interface size.
  • Empirical model-error covariance Cov(ε_model) = per-sensor sample variance from Nm validation realizations
    Sample covariance of FEM−surrogate residuals on validation draws; injected into Kalman gain. Without it posteriors collapse; with it they match FEM-EnKF. Estimated from data, not derived.
  • Prior means/variances of κ, μ and measurement noise level c_f = E[κ]=175 GPa, Var=10; E[μ]=81 GPa, Var=10; c_f=1% of u_true
    Lognormal prior moments (Table 1) and 1% (baseline) additive Gaussian noise define the Bayesian problem; not fitted to the posterior but chosen by the experimental design.
axioms (6)
  • standard math Gauss–Markov–Kalman / conditional-expectation update is a valid approximate Bayesian posterior map for the (mildly) nonlinear parameter-to-observation map when second moments exist.
    Invoked from §2 Eqs. 7–15 and the ensemble form Eqs. 19–21; standard in the cited Rosić/Matthies line.
  • domain assumption Linear isotropic elasticity with lognormal independent bulk and shear moduli yields a well-posed stochastic BVP whose FEM discretization errors are controllable.
    §6 Eqs. 62–67 and Table 1; classical continuum mechanics + FEM.
  • domain assumption Observation operator is spatially local: each measurement depends only on the solution in a limited region / its subdomain.
    Stated explicitly after Eq. 3 and before Eq. 49; required for decomposing Y and the Kalman gain into subdomain blocks.
  • ad hoc to paper Total modeling error (FEM + NN + DDM + observation-net) can be treated as zero-mean Gaussian with covariance estimated from finite validation residuals, independent enough to add to measurement noise in the gain.
    §3 Eqs. 30 and following; §5–6.3. Authors note AR/joint-state alternatives as future work; Gaussian plug-in is a modeling choice that materially changes posterior width.
  • ad hoc to paper Block-diagonal observation-covariance preconditioner (neglecting cross-subdomain Cov(y_i,y_j)) plus Richardson iteration converges to the global Kalman update at acceptable iteration cost.
    §5 Eqs. 58–61; complexity claim O(N_y^3/M^2) assumes this. Not proved; supported only indirectly by final posterior agreement for M=2.
  • domain assumption C^1 interface constraints enforced by augmented Lagrange multipliers produce a globally consistent surrogate whose interface error is negligible for filtering after discarding near-zero displacement sensors.
    §4 Eqs. 34–46 and §6.2 KLD/error maps; near-zero BC regions are manually excluded from assimilation.

pith-pipeline@v1.2.0-grok45-kimik3 · 34375 in / 4373 out tokens · 99497 ms · 2026-07-31T18:38:29.252960+00:00 · methodology

0 comments
read the original abstract

Ensemble Kalman filters (EnKF) provide an efficient framework for parameter identification of physics based laws from spatially distributed measurements. Their forecast models require a large number of samples to accurately represent uncertainties, leading to high computational costs. A NN-based surrogate model is introduced to replace the sample-based forecast model. The proposed NN surrogate maps spatial coordinates and physics-based parameters to the forecasted observation. Such maps require a large number of parameters for high-dimensional spatial domains. To overcome this limitation, a augmented Lagrange multiplier domain decomposition method (DDM) is developed, where local NN models are optimized independently before global communication and coupling. This reduces the number of NN parameters while improving local approximation accuracy. Furthermore, a distributed and decentralized ensemble Kalman filter approach based on DDM-NN surrogate model is investigated, where the parameter identification problem is decomposed into local subproblems. Each local estimator updates the material parameters using locally available information, while communication between neighboring subdomains enables the reconstruction of a consistent global estimate to reduce the computational cost. The proposed method is evaluated on a three-dimensional material parameter identification problem and compared with an EnKF and a MCMC reference solution. The results show that the proposed DDM NN-based KF captures the posterior parameter distribution and approaches the solutions obtained with both EnKF and MCMC. While MCMC provides the most accurate representation of the posterior distribution, the proposed approach achieves comparable parameter estimates with reduced computational requirements for the forecast model.

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

55 extracted references · 10 canonical work pages · 2 internal anchors

  1. [1]

    Springer, New York, NY, USA (2006)

    Nocedal, J., Wright, S.J.: Numerical Optimization, 2e edn. Springer, New York, NY, USA (2006)

    2006

  2. [2]

    Acta Numerica19, 451– 559 (2010) https://doi.org/10.1017/S0962492910000061

    Stuart, A.M.: Inverse problems: A bayesian perspective. Acta Numerica19, 451– 559 (2010) https://doi.org/10.1017/S0962492910000061

    doi:10.1017/s0962492910000061 2010

  3. [3]

    Journal of Basic Engineering82(1), 35–45 (1960) https://doi.org/10.1115/1.3662552 46

    Kalman, R.E.: A new approach to linear filtering and prediction problems. Journal of Basic Engineering82(1), 35–45 (1960) https://doi.org/10.1115/1.3662552 46

    doi:10.1115/1.3662552 1960

  4. [4]

    The Journal of Chemical Physics21(6), 1087–1092 (1953) https://doi.org/10.1063/1.1699114

    Metropolis, N., Rosenbluth, A.W., Rosenbluth, M.N., Teller, A.H., Teller, E.: Equation of state calculations by fast computing machines. The Journal of Chemical Physics21(6), 1087–1092 (1953) https://doi.org/10.1063/1.1699114

    doi:10.1063/1.1699114 1953

  5. [5]

    Journal of Ocean Dynamics53, 343–367 (2003) https://doi.org/ 10.1007/s10236-003-0036-9

    Evensen, G.: The ensemble kalman filter: theoretical formulation and practical implementation. Journal of Ocean Dynamics53, 343–367 (2003) https://doi.org/ 10.1007/s10236-003-0036-9

    doi:10.1007/s10236-003-0036-9 2003

  6. [6]

    Quarterly Journal of the Royal Meteorological Society148(743), 620–640 (2021) https://doi.org/10.1002/qj.4221

    Nerger, L.: Data assimilation for nonlinear systems with a hybrid nonlinear kalman ensemble transform filter. Quarterly Journal of the Royal Meteorological Society148(743), 620–640 (2021) https://doi.org/10.1002/qj.4221

    doi:10.1002/qj.4221 2021

  7. [7]

    American Journal of Mathematics60(4), 897–936 (1938)

    Wiener, N.: The homogeneous chaos. American Journal of Mathematics60(4), 897–936 (1938). Accessed 2024-08-14

    1938

  8. [8]

    In: Pro- ceedings of the 9th International Conference on Neural Information Processing Systems

    Williams, C.K.I., Rasmussen, C.E.: Gaussian processes for regression. In: Pro- ceedings of the 9th International Conference on Neural Information Processing Systems. NIPS’95, pp. 514–520. MIT Press, Cambridge, MA, USA (1995)

    1995

  9. [9]

    Current Science78(7), 808–817 (2000)

    Chatterjee, A.: An introduction to the proper orthogonal decomposition. Current Science78(7), 808–817 (2000)

    2000

  10. [10]

    Math- ematics of Control, Signals and Systems2, 303–314 (1989) https://doi.org/10

    Cybenko, G.: Approximation by superpositions of a sigmoidal function. Math- ematics of Control, Signals and Systems2, 303–314 (1989) https://doi.org/10. 1007/BF02551274

    1989

  11. [11]

    Advances in Neural Information Processing Systems, 1097–1105 (2012) https://doi.org/10.1145/3065386

    Hinton, G., Krizhevsky, A., Sutskever, I.: Imagenet classification with deep con- volutional neural networks. Advances in Neural Information Processing Systems, 1097–1105 (2012) https://doi.org/10.1145/3065386

    doi:10.1145/3065386 2012

  12. [12]

    Foun- dations and Trends®in Machine Learning12(4), 307–392 (2019) https://doi

    Kingma, D.P., Welling, M.: An introduction to variational autoencoders. Foun- dations and Trends®in Machine Learning12(4), 307–392 (2019) https://doi. org/10.1561/2200000056

    doi:10.1561/2200000056 2019

  13. [13]

    Journal of Computational Physics378, 686–707 (2019) https://doi.org/10.1016/j.jcp.2018.10.045

    Raissi, M., Perdikaris, P., Karniadakis, G.E.: Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics378, 686–707 (2019) https://doi.org/10.1016/j.jcp.2018.10.045

    doi:10.1016/j.jcp.2018.10.045 2019

  14. [14]

    Jour- nal of Computational Physics463, 111301 (2022) https://doi.org/10.1016/j.jcp

    Basir, S., Senocak, I.: Physics and equality constrained artificial neural networks: Application to forward and inverse problems with multi-fidelity data fusion. Jour- nal of Computational Physics463, 111301 (2022) https://doi.org/10.1016/j.jcp. 2022.111301

    arXiv 2022

  15. [15]

    Computer Methods in Applied Mechanics and Engineering415, 116229 (2023) https://doi.org/10.1016/j.cma.2023.116229 47

    Liu, M., Cai, Z., Ramani, K.: Deep ritz method with adaptive quadrature for linear elasticity. Computer Methods in Applied Mechanics and Engineering415, 116229 (2023) https://doi.org/10.1016/j.cma.2023.116229 47

    arXiv 2023

  16. [16]

    IEEE Transactions on Signal Processing70, 1532–1547 (2022) https://doi.org/ 10.1109/TSP.2022.3158588

    Revach, G., Shlezinger, N., Ni, X., Escoriza, A.L., Sloun, R.J.G., Eldar, Y.C.: Kalmannet: Neural network aided kalman filtering for partially known dynamics. IEEE Transactions on Signal Processing70, 1532–1547 (2022) https://doi.org/ 10.1109/TSP.2022.3158588

    arXiv 2022

  17. [17]

    Journal of Computational Physics477, 111918 (2023) https: //doi.org/10.1016/j.jcp.2023.111918

    Chattopadhyay, A., Nabizadeh, E., Bach, E., Hassanzadeh, P.: Deep learning- enhanced ensemble-based data assimilation for high-dimensional nonlinear dynamical systems. Journal of Computational Physics477, 111918 (2023) https: //doi.org/10.1016/j.jcp.2023.111918

    arXiv 2023

  18. [18]

    Springer Series in Computational Mathematics, vol

    Toselli, A., Widlund, O.: Domain Decomposition Methods - Algorithms and Theory. Springer Series in Computational Mathematics, vol. 34. Springer, Berlin, Heidelberg (2005). https://doi.org/10.1007/b137868 . https://doi.org/10.1007/b137868

    doi:10.1007/b137868 2005

  19. [19]

    Journal f¨ ur die reine und ange- wandte Mathematik1869(70), 105–120 (1869) https://doi.org/10.1515/crll.1869

    Schwarz, H.A.: ¨Uber einige abbildungsaufgaben. Journal f¨ ur die reine und ange- wandte Mathematik1869(70), 105–120 (1869) https://doi.org/10.1515/crll.1869. 70.105

    work page doi:10.1515/crll.1869

  20. [20]

    In: Glowinski, R., Golub, G.H., Meurant, G.A., Periaux, J

    Lions, P.L.: On the Schwarz alternating method I. In: Glowinski, R., Golub, G.H., Meurant, G.A., Periaux, J. (eds.) First Proceedings of Domain Decomposition Methods for Partial Differential Equations, pp. 1–42. SIAM, Philadelphia, USA (1988). Chap. 1

    1988

  21. [21]

    In: Chan, T.F., Glowinski, R., P´ eriaux, J., Widlund, O.B

    Lions, P.L.: On the Schwarz alternating method III: A variant for nonoverlapping subdomains. In: Chan, T.F., Glowinski, R., P´ eriaux, J., Widlund, O.B. (eds.) Third International Symposium on Domain Decomposition Methods for Partial Differential Equations, pp. 202–223. SIAM, Philadelphia, USA (1990). Chap. 11

    1990

  22. [22]

    Mathematical and Computational Applications31(3) (2026) https://doi.org/10.3390/mca31030073

    G¨ odde, T., Atzema, E.H., Rosi´ c, B.: Domain decomposition of large neural net- work surrogate models. Mathematical and Computational Applications31(3) (2026) https://doi.org/10.3390/mca31030073

    work page doi:10.3390/mca31030073 2026

  23. [23]

    Computer Methods in Applied Mechanics and Engineering354, 307–330 (2019) https://doi.org/10.1016/j.cma.2019.05.039

    Xiao, D., Fang, F., Heaney, C.E., Navon, I.M., Pain, C.C.: A domain decom- position method for the non-intrusive reduced order modelling of fluid flow. Computer Methods in Applied Mechanics and Engineering354, 307–330 (2019) https://doi.org/10.1016/j.cma.2019.05.039

    doi:10.1016/j.cma.2019.05.039 2019

  24. [24]

    Physics of Fluids34(5), 055111 (2022) https://doi.org/10.1063/5.0088070

    Heaney, C.E., Wolffs, Z., T´ omasson, J.A., Kahouadji, L., Salinas, P., Nicolle, A., Navon, I.M., Matar, O.K., Srinil, N., Pain, C.C.: An ai-based non-intrusive reduced-order model for extended domains applied to multiphase flow in pipes. Physics of Fluids34(5), 055111 (2022) https://doi.org/10.1063/5.0088070

    work page doi:10.1063/5.0088070 2022

  25. [25]

    IEEE Access8, 5283–5294 (2019) https://doi

    Li, K., Tang, K., Wu, T., Liao, Q.: D3m: A deep domain decomposition method for partial differential equations. IEEE Access8, 5283–5294 (2019) https://doi. org/10.1109/ACCESS.2019.2957200 48

    arXiv 2019

  26. [26]

    In: Jianfeng, L., Rachel, W

    Li, W., Xiang, X., Xu, Y.: Deep domain decomposition method: Elliptic problems. In: Jianfeng, L., Rachel, W. (eds.) Proceedings of The First Mathematical and Scientific Machine Learning Conference, vol. 107, pp. 269–286. PMLR, ??? (2020)

    2020

  27. [27]

    Adv Comput Math49, 1–39 (2023) https://doi.org/10.1007/ s10444-023-10065-9

    Moseley, B., Markham, A., Nissen-Meyer, T.: Finite basis physics-informed neural networks (fbpinns): a scalable domain decomposition approach for solving differ- ential equations. Adv Comput Math49, 1–39 (2023) https://doi.org/10.1007/ s10444-023-10065-9

    2023

  28. [28]

    CiCP28(5), 2002–2041 (2020) https://doi.org/10.4208/cicp.OA-2020-0164

    Jagtap, A.D., Karniadakis, G.E.: Extended physics-informed neural networks (xpinns): A generalized space-time domain decomposition based deep learning framework for nonlinear partial differential equations. CiCP28(5), 2002–2041 (2020) https://doi.org/10.4208/cicp.OA-2020-0164

    doi:10.4208/cicp.oa-2020-0164 2002

  29. [29]

    Computer Methods in Applied Mechanics and Engineering365, 113028 (2020) https://doi.org/10.1016/j.cma.2020.113028

    Jagtap, A.D., Kharazmi, E., Karniadakis, G.E.: Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems. Computer Methods in Applied Mechanics and Engineering365, 113028 (2020) https://doi.org/10.1016/j.cma.2020.113028

    arXiv 2020

  30. [30]

    Partitioned neural network approximation for partial differential equations enhanced with Lagrange multipliers and localized loss functions

    Jang, K., Kim, K., Kim, H.H.: Partitioned neural network approximation for partial differential equations enhanced with lagrange multipliers and localized loss functions. CoRR (2023) https://doi.org/10.48550/arxiv.2312.14370

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2312.14370 2023

  31. [31]

    Computer Methods in Applied Mechanics and Engineering387, 114129 (2021) https://doi.org/10.1016/ j.cma.2021.114129

    Dong, S., Li, Z.: Local extreme learning machines and domain decomposition for solving linear and nonlinear partial differential equations. Computer Methods in Applied Mechanics and Engineering387, 114129 (2021) https://doi.org/10.1016/ j.cma.2021.114129

    arXiv 2021

  32. [32]

    Computers and Mathematics with Applications189, 109–128 (2025) https://doi.org/10.1016/j.camwa.2025.04.001

    Lee, C., Lee, Y., Ryoo, B.: A nonoverlapping domain decomposition method for extreme learning machines: Elliptic problems. Computers and Mathematics with Applications189, 109–128 (2025) https://doi.org/10.1016/j.camwa.2025.04.001

    work page doi:10.1016/j.camwa.2025.04.001 2025

  33. [33]

    In: 2014 IEEE 8th Sensor Array and Multichannel Signal Processing Workshop (SAM), pp

    Shahid, A., ¨Ustebay, D., Coates, M.: Distributed ensemble kalman filtering. In: 2014 IEEE 8th Sensor Array and Multichannel Signal Processing Workshop (SAM), pp. 217–220 (2014). https://doi.org/10.1109/SAM.2014.6882379

    arXiv 2014

  34. [34]

    Proceedings of the IEEE98(11), 1847–1864 (2010) https://doi.org/10.1109/JPROC.2010.2052531

    Dimakis, A.G., Kar, S., Moura, J.M.F., Rabbat, M.G., Scaglione, A.: Gossip algorithms for distributed signal processing. Proceedings of the IEEE98(11), 1847–1864 (2010) https://doi.org/10.1109/JPROC.2010.2052531

    arXiv 2010

  35. [35]

    Computers and Chemical Engineering 156, 107544 (2022) https://doi.org/10.1016/j.compchemeng.2021.107544

    Masooleh, L.S., Arbogast, J.E., Seider, W.D., Oktem, U., Soroush, M.: Dis- tributed state estimation in large-scale processes decomposed into observable subsystems using community detection. Computers and Chemical Engineering 156, 107544 (2022) https://doi.org/10.1016/j.compchemeng.2021.107544

    arXiv 2022

  36. [36]

    Monthly Weather Review 49 139(7), 2046–2060 (2011) https://doi.org/10.1175/2011MWR3552.1

    Janji´ c, T., Nerger, L., Albertella, A., Schr¨ oter, J., Skachko, S.: On domain local- ization in ensemble-based kalman filter algorithms. Monthly Weather Review 49 139(7), 2046–2060 (2011) https://doi.org/10.1175/2011MWR3552.1

    work page doi:10.1175/2011mwr3552.1 2046

  37. [37]

    Journal of Computa- tional Science36, 100654 (2019) https://doi.org/10.1016/j.jocs.2019.100654

    Nino-Ruiz, E.D., Sandu, A., Deng, X.: A parallel implementation of the ensemble kalman filter based on modified cholesky decomposition. Journal of Computa- tional Science36, 100654 (2019) https://doi.org/10.1016/j.jocs.2019.100654

    arXiv 2019

  38. [38]

    Journal of Computational Physics509, 113059 (2024) https://doi.org/10.1016/j.jcp.2024.113059

    Zhang, X.-L., Zhang, L., He, G.: Parallel ensemble kalman method with total variation regularization for large-scale field inversion. Journal of Computational Physics509, 113059 (2024) https://doi.org/10.1016/j.jcp.2024.113059

    arXiv 2024

  39. [39]

    Signal Processing233, 110785 (2026) https://doi.org/10.1016/j.sigpro.2026.110785

    Li, G., Li, W., Gao, Z., Gao, S.: High-dimensional kalman filtering: From sparse modeling to ai-augmented architectures. Signal Processing233, 110785 (2026) https://doi.org/10.1016/j.sigpro.2026.110785

    arXiv 2026

  40. [40]

    Yale University Press, ??? (1923)

    Hadamard, J.: Lectures on Cauchy’s Problem in Linear Partial Differential Equations. Yale University Press, ??? (1923)

    1923

  41. [41]

    Engineering Structures50, 179–196 (2013) https://doi.org/10.1016/j.engstruct.2012.12.029

    Rosi´ c, B.V., Kuˇ cerov´ a, A., S´ ykora, J., Pajonk, O., Litvinenko, A., Matthies, H.G.: Parameter identification in a probabilistic setting. Engineering Structures50, 179–196 (2013) https://doi.org/10.1016/j.engstruct.2012.12.029

    work page doi:10.1016/j.engstruct.2012.12.029 2013

  42. [42]

    Cambridge University Press, ??? (2005)

    Bobrowski, A.: Functional Analysis for Probability and Stochastic Processes: An Introduction. Cambridge University Press, ??? (2005)

    2005

  43. [43]

    Advanced Modeling and Simulation in Engineering Sciences3(1), 24 (2016) https://doi.org/10.1186/ s40323-016-0075-7

    Matthies, H.G., Zander, E., Rosi´ c, B.V., Litvinenko, A.: Parameter estima- tion via conditional expectation: a bayesian inversion. Advanced Modeling and Simulation in Engineering Sciences3(1), 24 (2016) https://doi.org/10.1186/ s40323-016-0075-7

    2016

  44. [44]

    Stochastic state estimation via incremental iterative sparse polynomial chaos based Bayesian-Gauss-Newton-Markov-Kalman filter

    Rosi´ c, B.: Stochastic state estimation via incremental iterative sparse polyno- mial chaos based bayesian-gauss-newton-markov-kalman filter. arXiv preprint arXiv:1909.07209 (2019) https://doi.org/10.48550/arXiv.1909.07209

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1909.07209 1909

  45. [45]

    In: Franzke, C.L.E., O’Kane, T.J

    Harlim, J.: Model error in data assimilation. In: Franzke, C.L.E., O’Kane, T.J. (eds.) Nonlinear and Stochastic Climate Dynamics, pp. 276–317. Cam- bridge University Press, Cambridge (2017). Chap. 10. https://doi.org/10.1017/ 9781316339251.010

    2017

  46. [46]

    IntechOpen, Rijeka (2013)

    Chan, V.: Theory and Applications of Monte Carlo Simulations. IntechOpen, Rijeka (2013). https://doi.org/10.5772/45892

    work page doi:10.5772/45892 2013

  47. [47]

    CoRRabs/2506.11639(2025)

    Mortada, H., Falcon, C., Kahil, Y., Clavaud, M., Michel, J.: Recursive kalmannet: Deep learning-augmented kalman filtering for state estimation with consistent uncertainty quantification. CoRRabs/2506.11639(2025)

    Pith/arXiv arXiv 2025

  48. [48]

    50 Journal of Machine Learning for Modeling and Computing1(2), 119–156 (2020) https://doi.org/10.1615/JMachLearnModelComput.2020035155

    Swiler, L.P., Gulian, M., Frankel, A.L., Safta, C., Jakeman, J.D.: A survey of con- strained gaussian process regression: Approaches and implementation challenges. 50 Journal of Machine Learning for Modeling and Computing1(2), 119–156 (2020) https://doi.org/10.1615/JMachLearnModelComput.2020035155

    doi:10.1615/jmachlearnmodelcomput.2020035155 2020

  49. [49]

    Foundations and Trends in Machine Learning3, 1–122 (2011) https://doi.org/ 10.1561/2200000016

    Boyd, S., Parikh, N., Chu, E., Peleato, B., Eckstein, J.: Distributed optimiza- tion and statistical learning via the alternating direction method of multipliers. Foundations and Trends in Machine Learning3, 1–122 (2011) https://doi.org/ 10.1561/2200000016

    doi:10.1561/2200000016 2011

  50. [50]

    SIAM Journal on Scientific Computing16(5), 1190– 1208 (1995) https://doi.org/10.1137/0916069

    Byrd, R.H., Lu, P., Nocedal, J., Zhu, C.: A limited memory algorithm for bound constrained optimization. SIAM Journal on Scientific Computing16(5), 1190– 1208 (1995) https://doi.org/10.1137/0916069

    doi:10.1137/0916069 1995

  51. [51]

    IF AC-PapersOnLine53(2), 8199–8204 (2020) https:// doi.org/10.1016/j.ifacol.2020.12.1996

    Zocco, F., McLoone, S.: An adaptive memory multi-batch l-bfgs algorithm for neural network training. IF AC-PapersOnLine53(2), 8199–8204 (2020) https:// doi.org/10.1016/j.ifacol.2020.12.1996 . 21st IF AC World Congress

    work page doi:10.1016/j.ifacol.2020.12.1996 2020

  52. [52]

    CoRRabs/2307.13744(2023)

    Niu, Y., Fabian, Z., Lee, S., Soltanolkotabi, M., Avestimehr, S.: An overview of gradient descent optimization algorithms. CoRRabs/2307.13744(2023)

    Pith/arXiv arXiv 2023

  53. [53]

    Philosophical Transactions of the Royal Society A 210(459–470), 307–357 (1911) https://doi.org/10.1098/rsta.1911.0009

    Richardson, L.F.: The approximate arithmetical solution by finite differences of physical problems involving differential equations, with an application to the stresses in a masonry dam. Philosophical Transactions of the Royal Society A 210(459–470), 307–357 (1911) https://doi.org/10.1098/rsta.1911.0009

    arXiv 1911

  54. [54]

    arXiv preprint arXiv:2302.08952 (2023)

    Stotsky, A.: Richardson approach or direct methods? what to apply in the ill- conditioned least squares problem. arXiv preprint arXiv:2302.08952 (2023)

    Pith/arXiv arXiv 2023

  55. [55]

    Ansys, Inc., Canonsburg, PA (2024)

    Ansys, Inc.: LS-DYNA User’s Manual, R15.0 edn. Ansys, Inc., Canonsburg, PA (2024). Ansys, Inc. 51

    2024