pith. sign in

arxiv: 2606.00480 · v1 · pith:56MYFRYRnew · submitted 2026-05-30 · 🧮 math.DS · cs.NA· math.NA· math.OC· stat.ML

Continuous Data Assimilation with Learned Surrogate Dynamics

Wenwen Li , Daniel Sanz-Alonso This is my paper

Pith reviewed 2026-06-28 18:31 UTC · model grok-4.3

classification 🧮 math.DS cs.NAmath.NAmath.OCstat.ML
keywords continuous data assimilationnudging algorithmssurrogate modelsexponential convergencemachine learningmodel errorobservation noisedynamical systems
0
0 comments X

The pith

Nudging algorithms using learned surrogate models converge exponentially to an explicit error floor set by approximation error and observation noise.

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

The paper develops a unified finite-dimensional analysis for nudging algorithms in continuous data assimilation when the true dynamics are replaced by learned surrogates. It first establishes general conditions on the dynamics and observations that guarantee accurate tracking with the true model, in both noise-free and noisy settings. It then shows that surrogate-based nudging retains exponential convergence but only up to an explicit error floor that accounts for the surrogate approximation error and the observation noise. The analysis further treats surrogates learned from the vector field or from the short-time solution map and gives bounds on the training data needed to keep the error floor small in the noise-free case. Numerical experiments are used to support the theoretical results.

Core claim

Under general conditions on the dynamics and observations that guarantee accurate tracking for nudging with the true dynamics model in both the noise-free and noisy settings, nudging algorithms that employ surrogate models retain exponential convergence up to an explicit error floor that quantifies the effects of surrogate approximation error and observation noise. The paper analyzes surrogate models obtained by learning either the vector field or the short-time solution map of the system and quantifies the amount of training data needed to ensure accurate nudging in the noise-free setting.

What carries the argument

Nudging algorithm driven by a learned surrogate dynamics model, which evolves the state estimate according to the surrogate while continuously nudging it toward partial observations, producing convergence to a quantifiable error floor.

If this is right

  • Exponential convergence to an error floor holds whenever the surrogate is substituted into a nudging scheme that works with the true model.
  • The size of the error floor is controlled explicitly by the surrogate approximation error and the observation noise level.
  • Quantifiable amounts of training data suffice to make the error floor arbitrarily small when learning either the vector field or the short-time solution map in the noise-free case.
  • The same finite-dimensional framework covers both noise-free and noisy observation regimes.
  • Numerical experiments are expected to reproduce the predicted convergence rates and error floors.

Where Pith is reading between the lines

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

  • The explicit error floor supplies a practical criterion for deciding when a surrogate is accurate enough for a given application without needing the full model.
  • The training-data bounds could be used to compare the data efficiency of vector-field learning versus solution-map learning before deployment.
  • Similar error-floor analysis might apply to other assimilation schemes that replace the model with a learned component.
  • In chaotic systems the size of the floor relative to the attractor diameter would indicate whether the surrogate is usable for long-term state estimation.

Load-bearing premise

There exist general conditions on the dynamics and observations that guarantee accurate tracking for nudging with the true dynamics model, both in the noise-free and noisy settings.

What would settle it

A numerical test in which the true dynamics and observations satisfy the stated general conditions yet nudging with a surrogate model fails to exhibit exponential convergence or produces an observed error floor substantially larger than the explicit bound given by the surrogate approximation error plus observation noise.

read the original abstract

Continuous data assimilation seeks to estimate the state of a dynamical system from partial observations. In many applications, however, the state dynamics are unknown or prohibitively expensive to simulate at the required resolution, leading to model error. Motivated by this challenge and the increasing adoption of machine learning surrogates in data assimilation, this paper develops a unified finite-dimensional analysis of nudging algorithms that employ learned surrogate models of the dynamics. We first establish general conditions on the dynamics and observations that guarantee accurate tracking for nudging with the true dynamics model, both in the noise-free and noisy settings. We then show that nudging algorithms that employ surrogate models retain exponential convergence up to an explicit error floor that quantifies the effects of surrogate approximation error and observation noise. Finally, we analyze surrogate models obtained by learning either the vector field or the short-time solution map of the system, and quantify the amount of training data needed to ensure accurate nudging in the noise-free setting. Numerical experiments support the theory.

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

0 major / 3 minor

Summary. The manuscript develops a unified finite-dimensional analysis of nudging-based continuous data assimilation that incorporates learned surrogate models of the dynamics. It first derives general conditions on the dynamics and observations guaranteeing accurate tracking when the true model is used, both noise-free and with observation noise. It then extends the analysis to show that surrogate-based nudging retains exponential convergence up to an explicit error floor that accounts for surrogate approximation error and noise. The paper further analyzes two surrogate constructions (learned vector field and learned short-time solution map), quantifies the training data volume needed to control the error floor in the noise-free case, and presents numerical experiments supporting the theory.

Significance. If the derivations hold, the explicit error-floor construction and the training-data quantification constitute a concrete advance: they supply falsifiable, parameter-dependent bounds that separate the effects of surrogate error from observation noise. The finite-dimensional setting and the two surrogate types (vector field vs. solution map) are treated uniformly, which is a strength for reproducibility and comparison with existing nudging analyses.

minor comments (3)
  1. [Abstract] The abstract states that numerical experiments support the theory but supplies no information on the test systems, the observed convergence rates, or the measured error floors; adding a brief quantitative summary (e.g., “on the Lorenz-63 system the observed floor matched the predicted O(ε+δ) bound within 10 %”) would strengthen the claim without lengthening the abstract.
  2. Notation for the surrogate approximation error (presumably denoted ε or similar) and the observation-noise level should be introduced once in §2 or §3 and used consistently thereafter; occasional re-definition of these quantities interrupts readability.
  3. [Introduction] The statement that the analysis is “unified” would be clearer if a short comparison table (or paragraph) contrasted the new error-floor result with the classical nudging bounds that assume a perfect model.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The derivation begins with independently stated general conditions on dynamics and observations that guarantee tracking for nudging with the true model (noise-free and noisy cases). It then extends this to surrogates by deriving an explicit error floor incorporating approximation error and observation noise, without reducing any prediction or convergence claim to a fitted parameter or self-referential definition. Analysis of vector-field or solution-map learning quantifies training data needs within the noise-free setting as part of the same framework. No load-bearing self-citations, uniqueness theorems imported from prior author work, or ansatzes smuggled via citation are present in the described chain; the finite-dimensional setting and error-floor construction remain externally verifiable against the stated assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Analysis rests on unspecified general conditions for the true-dynamics case that are then extended to surrogates; no free parameters, new entities, or ad-hoc axioms are mentioned in the abstract.

axioms (1)
  • domain assumption General conditions on the dynamics and observations guarantee accurate tracking for nudging with the true dynamics model
    Invoked as the baseline before proving the surrogate extension and error floor.

pith-pipeline@v0.9.1-grok · 5705 in / 1216 out tokens · 23637 ms · 2026-06-28T18:31:13.478752+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

58 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    Abur and A

    A. Abur and A. G. Expósito,Power System State Estimation: Theory and Implementation, CRC Press, Boca Raton, FL, 2004

    2004

  2. [2]

    Adrian, D

    M. Adrian, D. Sanz-Alonso, and R. Willett,Data assimilation with machine learning surrogate models: A case study with FourCastNet, Artificial Intelligence for the Earth Systems, 4 (2025), p. e240050

    2025

  3. [3]

    D. A. F. Albanez and M. J. Benvenutti,Continuous data assimilation algorithm for simplified Bardina model, Evolution Equations and Control Theory, 7 (2018), pp. 33–52

    2018

  4. [4]

    D. A. F. Albanez, H. J. Nussenzveig Lopes, and E. S. Titi,Continuous data assimilation for the three-dimensional Navier–Stokes-𝛼model, Asymptotic Analysis, 97 (2016), pp. 139–164

    2016

  5. [5]

    Anthony and P

    M. Anthony and P. Bartlett,Neural Network Learning: Theoretical Foundations, Cam- bridge University Press, 1999

    1999

  6. [6]

    Azouani, E

    H. Azouani, E. Olson, and E. S. Titi,Continuous data assimilation using general inter- polant observables, Journal of Nonlinear Science, 24 (2014), pp. 277–304

    2014

  7. [7]

    E. Bach, R. Baptista, D. Sanz-Alonso, and A. Stuart,Machine learning for inverse problems and data assimilation, arXiv preprint arXiv:2410.10523, (2024)

    work page arXiv 2024

  8. [8]

    Balakrishna and A

    A. Balakrishna and A. Biswas,Determining map, data assimilation and an observable regularity criterion for the three-dimensional Boussinesq system, Applied Mathematics & Optimization, 86 (2022), p. 28

    2022

  9. [9]

    Bessaih, E

    H. Bessaih, E. Olson, and E. S. Titi,Continuous data assimilation with stochastically noisy data, Nonlinearity, 28 (2015), p. 729–753

    2015

  10. [10]

    Biswas and R

    A. Biswas and R. Price,Continuous data assimilation for the three-dimensional Navier– Stokes equations, SIAM Journal on Mathematical Analysis, 53 (2021), pp. 6697–6723

    2021

  11. [11]

    Brajard, A

    J. Brajard, A. Carrassi, M. Bocquet, and L. Bertino,Combining data assimilation and machine learning to emulate a dynamical model from sparse and noisy observations: a case study with the Lorenz 96 model, Journal of Computational Science, 44 (2020), p. 101171

    2020

  12. [12]

    Continuous Data Assimilation for Semilinear Parabolic Equations with Multiplicative Observation Noise

    J. Bröcker, G. Del Sarto, M. Hieber, F. Palma, and T. Zöchling,Continuous data assimilation for semilinear parabolic equations with multiplicative observation noise, arXiv preprint arXiv:2605.10322, (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  13. [13]

    Y. Cao, A. Giorgini, M. S. Jolly, and A. Pakzad,Continuous data assimilation for the 3D Ladyzhenskaya model: analysis and computations, Nonlinear Analysis: Real World Applications, 68 (2022), p. 103659

    2022

  14. [14]

    Carlson, J

    E. Carlson, J. Hudson, and A. Larios,Parameter recovery for the 2 dimensional Navier– Stokes equations via continuous data assimilation, SIAM Journal on Scientific Computing, 42 (2020), pp. A250–A270. 40

    2020

  15. [15]

    Chen,Stochastic Methods for Modeling and Predicting Complex Dynamical Systems, Springer, 2023

    N. Chen,Stochastic Methods for Modeling and Predicting Complex Dynamical Systems, Springer, 2023

    2023

  16. [16]

    N. Chen, A. Farhat, and E. Lunasin,Data assimilation with model error: Analytical and computational study for Sabra shell model, Physica D: Nonlinear Phenomena, 443 (2023), p. 133552

    2023

  17. [17]

    Cheng, C

    S. Cheng, C. Quilodrán-Casas, S. Ouala, A. Farchi, C. Liu, P. Tandeo, R. Fablet, D. Lucor, B. Iooss, J. Brajard, D. Xiao, T. Janjic, W. Ding, Y. Guo, A. Carrassi, M. Bocquet, and R. Arcucci,Machine learning with data assimilation and uncertainty quantification for dynamical systems: A review, IEEE/CAA Journal of Automatica Sinica, 10 (2023), pp. 1361–1387

    2023

  18. [18]

    Çıbık, R

    A. Çıbık, R. Fang, W. Layton, and F. Siddiqua,Data assimilation with model errors, arXiv preprint arXiv:2504.16291, (2025)

    work page arXiv 2025

  19. [19]

    W. M. Czarnecki, S. Osindero, M. Jaderberg, G. Swirszcz, and R. Pascanu,Sobolev training for neural networks, in Advances in Neural Information Processing Systems 30 (NeurIPS 2017), 2017

    2017

  20. [20]

    Del Sarto, M

    G. Del Sarto, M. Hieber, F. Palma, and T. Zöchling,Continuous data assimilation for semilinear parabolic equations: A general approach by evolution equations, arXiv preprint arXiv:2602.20769, (2026)

    work page arXiv 2026

  21. [21]

    Evensen,Data Assimilation: The Ensemble Kalman Filter, Springer, Berlin, Heidelberg, 2 ed., 2009

    G. Evensen,Data Assimilation: The Ensemble Kalman Filter, Springer, Berlin, Heidelberg, 2 ed., 2009

    2009

  22. [22]

    Farhat, M

    A. Farhat, M. S. Jolly, and E. S. Titi,Continuous data assimilation for the 2D Bénard convection through velocity measurements alone, Physica D, 303 (2015), pp. 59–66

    2015

  23. [23]

    FitzHugh,Impulses and physiological states in theoretical models of nerve membrane, Biophysical Journal, 1 (1961), pp

    R. FitzHugh,Impulses and physiological states in theoretical models of nerve membrane, Biophysical Journal, 1 (1961), pp. 445–466

    1961

  24. [24]

    R. J. Gladstone, M. A. Nabian, N. Sukumar, A. Srivastava, and H. Meidani, FO-PINNs: A first-order formulation for physics-informed neural networks, arXiv preprint arXiv:2210.14320, (2022)

    work page arXiv 2022

  25. [25]

    M. S. Hansen and P. Kellman,Image reconstruction: An overview for clinicians, Journal of Magnetic Resonance Imaging, 41 (2015), pp. 573–585

    2015

  26. [26]

    D. Hsu, S. M. Kakade, and T. Zhang,Random design analysis of ridge regression, Foundations of Computational Mathematics, 14 (2014), pp. 569–600

    2014

  27. [27]

    H. A. Ibdah, C. F. Mondaini, and E. S. Titi,Fully discrete numerical schemes of a data assimilation algorithm: uniform-in-time error estimates, IMA Journal of Numerical Analysis, 40 (2020), pp. 2584–2625

    2020

  28. [28]

    M. S. Jolly, V. R. Martinez, E. J. Olson, and E. S. Titi,Continuous data assimilation with blurred-in-time measurements of the surface quasi-geostrophic equation, Chinese Annals of Mathematics, Series B, 40 (2019), pp. 721–764

    2019

  29. [29]

    M. S. Jolly, V. R. Martinez, and E. S. Titi,A data assimilation algorithm for the subcritical surface quasi-geostrophic equation, Advances in Nonlinear Studies, 17 (2017), pp. 167– 192

    2017

  30. [30]

    M. S. Jolly and A. Pakzad,Data assimilation with higher order finite element interpolants, International Journal for Numerical Methods in Fluids, 95 (2023), pp. 472–490

    2023

  31. [31]

    Kalnay,Atmospheric Modeling, Data Assimilation and Predictability, Cambridge University Press, Cambridge, 2003

    E. Kalnay,Atmospheric Modeling, Data Assimilation and Predictability, Cambridge University Press, Cambridge, 2003. 41

    2003

  32. [32]

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

    2015

  33. [33]

    K. J. Law, D. Sanz-Alonso, A. Shukla, and A. M. Stuart,Filter accuracy for the Lorenz 96 model: Fixed versus adaptive observation operators, Physica D: Nonlinear Phenomena, 325 (2016), pp. 1–13

    2016

  34. [34]

    Li and D

    W. Li and D. Sanz-Alonso,Code and numerical results for continuous data assimilation with learned surrogate dynamics. https://github.com/WenwenLi2002/ continuous-data-assimilation-with-surrogate-dynamics, 2026. GitHub repository

    2026

  35. [35]

    Loiseau, B

    J.-C. Loiseau, B. R. Noack, and S. L. Brunton,Sparse reduced-order modelling: sensor- based dynamics to full-state estimation, Journal of Fluid Mechanics, 844 (2018), pp. 459–490

    2018

  36. [36]

    E. N. Lorenz,Deterministic nonperiodic flow, Journal of the Atmospheric Sciences, 20 (1963), pp. 130–141

    1963

  37. [37]

    I, Reading, 1996, ECMWF, pp

    ,Predictability: A problem partly solved, in Proceedings of the Seminar on Predictability, Vol. I, Reading, 1996, ECMWF, pp. 1–18

    1996

  38. [38]

    A. J. Moodey, A. S. Lawless, R. W. Potthast, and P. J. Van Leeuwen,Nonlinear error dynamics for cycled data assimilation methods, Inverse Problems, 29 (2013), p. 025002

    2013

  39. [39]

    Nagumo, S

    J. Nagumo, S. Arimoto, and S. Yoshizawa,An active pulse transmission line simulating nerve axon, Proceedings of the IRE, 50 (1962), pp. 2061–2070

    1962

  40. [40]

    N. J. Nair and A. Goza,Leveraging reduced-order models for state estimation using deep learning, Journal of Fluid Mechanics, 897 (2020), p. R1

    2020

  41. [41]

    Newey, J

    J. Newey, J. P. Whitehead, and E. Carlson,Model discovery on the fly using continuous data assimilation, Journal of Computational Physics, 537 (2025), p. 114121

    2025

  42. [42]

    Ou and H

    W. Ou and H. Bölcskei,Covering numbers for deep ReLU networks with applications to func- tion approximation and nonparametric regression, Foundations of Computational Mathematics, (2026). To appear

    2026

  43. [43]

    Pollard,Empirical Processes: Theory and Applications, vol

    D. Pollard,Empirical Processes: Theory and Applications, vol. 2 of NSF-CBMS Regional Conference Series in Probability and Statistics, Institute of Mathematical Statistics, 1990

    1990

  44. [44]

    Reich and C

    S. Reich and C. Cotter,Probabilistic Forecasting and Bayesian Data Assimilation, Cam- bridge University Press, Cambridge, 2015

    2015

  45. [45]

    Sanz-Alonso, A

    D. Sanz-Alonso, A. Stuart, and A. Taeb,Inverse Problems and Data Assimilation, vol. 107, Cambridge University Press, 2023

    2023

  46. [46]

    Sanz-Alonso and A

    D. Sanz-Alonso and A. M. Stuart,Long-time asymptotics of the filtering distribution for partially observed chaotic dynamical systems, SIAM/ASA Journal on Uncertainty Quantification, 3 (2015), pp. 1200–1220

    2015

  47. [47]

    Sanz-Alonso and N

    D. Sanz-Alonso and N. Waniorek,Long-time accuracy of ensemble Kalman filters for chaotic dynamical systems and machine-learned dynamical systems, SIAM Journal on Applied Dynamical Systems, 24 (2025), pp. 2246–2286

    2025

  48. [48]

    G. Shen, Y. Jiao, Y. Lin, J. L. Horowitz, and J. Huang,Nonparametric estimation of non-crossing quantile regression process with deep ReQU neural networks, Journal of Machine Learning Research, 25 (2024), pp. 1–75

    2024

  49. [49]

    Simon,Optimal State Estimation: Kalman, 𝐻∞, and Nonlinear Approaches, Wiley- Interscience, Hoboken, NJ, 2006

    D. Simon,Optimal State Estimation: Kalman, 𝐻∞, and Nonlinear Approaches, Wiley- Interscience, Hoboken, NJ, 2006. 42

    2006

  50. [50]

    H. Son, J. W. Jang, W. J. Han, and H. J. Hwang,Sobolev training for physics-informed neural networks, Communications in Mathematical Sciences, 21 (2023), pp. 1679–1705

    2023

  51. [51]

    J. T. Stuart,On the non-linear mechanics of hydrodynamic stability, Journal of Fluid Mechanics, 4 (1958), pp. 1–21

    1958

  52. [52]

    Takeshima, K

    A. Takeshima, K. Shiraishi, A. Okazaki, T. Tsuyuki, and S. Kotsuki,Bridging artificial intelligence and data assimilation: the data-driven ensemble forecasting system ClimaX-LETKF, arXiv preprint arXiv:2512.14444, (2025)

    work page arXiv 2025

  53. [53]

    M. Tang, Y. Liu, and L. J. Durlofsky,A deep-learning-based surrogate model for data assimilation in dynamic subsurface flow problems, Journal of Computational Physics, 413 (2020), p. 109456

    2020

  54. [54]

    A. R. Thompson, J. M. Moran, and G. W. J. Swenson,Interferometry and Synthesis in Radio Astronomy, Springer, 3 ed., 2017

    2017

  55. [55]

    Willcox,Unsteady flow sensing and estimation via the gappy proper orthogonal decompo- sition, Computers & Fluids, 35 (2006), pp

    K. Willcox,Unsteady flow sensing and estimation via the gappy proper orthogonal decompo- sition, Computers & Fluids, 35 (2006), pp. 208–226

    2006

  56. [56]

    Y. Yang, Y. Wu, H. Yang, and Y. Xiang,Nearly optimal approximation rates for deep super ReLU networks on Sobolev spaces, arXiv preprint arXiv:2310.10766, (2023)

    work page arXiv 2023

  57. [57]

    Y. Yang, H. Yang, and Y. Xiang,Nearly optimal VC-dimension and pseudo-dimension bounds for deep neural network derivatives, in Advances in Neural Information Processing Systems 36, 2023, pp. 21721–21756

    2023

  58. [58]

    J. Yu, L. Lu, X. Meng, and G. E. Karniadakis,Gradient-enhanced physics-informed neural networks for forward and inverse PDE problems, Computer Methods in Applied Mechanics and Engineering, 393 (2022), p. 114823. 43 A Additional example verifications Verification for the Stuart–Landau oscillator.For (2.7), we take 𝐻=𝑉=R 2, 𝐴=𝐼, 𝜈= 1, 𝑓= 0, and 𝑁(𝑢):=−(𝜆+ 1...

    2022