Pith. sign in

REVIEW 4 major objections 5 minor 129 references

Modeling Partially Observed Nonlinear Dynamical Systems and Efficient Data Assimilation via Discrete-Time Conditional Gaussian Koopman Network

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

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

Pith's one-line read Discrete-time CGKN learns a latent embedding of unobserved states that makes the surrogate conditionally linear, so data assimilation becomes closed-form Gaussian filtering while forecast accuracy stays competitive with neural operators…

desk verdict A credible discrete-time successor to CGKN with strong benchmarks on three PDEs, but the conditional-linear representability assumption is asserted rather than proven and the reported DA metrics cannot fully validate it. read the letter →

arxiv 2507.08749 v1 pith:WCWDJRN4 submitted 2025-07-11 cs.LG

Chuanqi Chen , Zhongrui Wang , Nan Chen , Jin-Long Wu This is my paper · ORCID
classification cs.LG MSC 68T0737M9962M20
keywords dataassimilationconditionalGaussiansystemKoopmanembeddingscientificmachinelearningpartiallyobservedsystemsspatiotemporaldynamicslatentspacemodelturbulence
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper develops a discrete-time conditional Gaussian Koopman network (CGKN), a neural surrogate for nonlinear, partially observed dynamical systems, trained for both state forecast and data assimilation (DA). The central assertion is that encoding unobserved states into a low-dimensional latent variable makes it possible to learn a model in which observed states evolve nonlinearly and latent states evolve linearly in themselves, with all coefficients depending only on the current observations. Such a model is conditionally Gaussian, so online filtering has closed-form Gaussian posterior updates instead of ensemble sampling. On the viscous Burgers, Kuramoto-Sivashinsky, and 2-D Navier-Stokes equations, the authors report forecast errors comparable to a Fourier neural operator and DA errors comparable to an ensemble Kalman filter applied to the true governing equations, at a fraction of the computational cost. The wider point is that one differentiable surrogate can be trained jointly for prediction and state estimation, opening a route to outer-loop tasks like inverse problems and control.

What carries the argument

The load-bearing object is the discrete-time conditional Gaussian Koopman network (CGKN), an encoder-decoder pair $\phi,\psi$ with sub-networks that output the coefficient maps $F_1,G_1,F_2,G_2$. The encoder maps unobserved states $u_2$ to a low-dimensional latent state $v$; the decoder maps $v$ back to $u_2$; and the sub-networks make the coefficient maps depend only on observed states $u_1$. This structure makes the modeled system conditionally Gaussian, which is the mechanism that carries the argument: the data-assimilation update is a closed-form Gaussian filter whose mean-covariance recursion uses a Kalman-gain analog. The closed form turns DA into $O(d_v^3)$ linear algebra and makes the DA loss differentiable, so assimilation accuracy is trained into the network rather than tuned as a separate ensemble procedure.

What would settle it

Train CGKN on a partially observed system whose unresolved part is known to have no finite-dimensional conditionally linear embedding (for instance, a chaotic field with a continuous spectrum), and increase the latent dimension $d_v$ while tracking DA error against an EnKF run on the true equations. If the required $d_v$ grows with the unobserved dimension, or if the filter's posterior covariance cannot be made consistent with residuals, the central representability assumption is broken.

Watch

Extended reading notes

Core claim

Formally, the paper claims that a partially observed discrete system $u^{n+1}=(G_1(u^n_1,u^n_2),G_2(u^n_1,u^n_2))$ can be replaced by a surrogate $u^{n+1}_1=F_1(u^n_1)+G_1(u^n_1)v^n+\sigma_1\epsilon^n_1$, $v^{n+1}=F_2(u^n_1)+G_2(u^n_1)v^n+\sigma_2\epsilon^n_2$, where $v^n=\phi(u^n_2)$ is a latent embedding produced by an encoder and $F_1,G_1,F_2,G_2$ are neural maps of the observed states alone. Because the latent dynamics are linear in $v$ given $u_1$, the filter posterior $p(v^n|\{u^i_1\}_{i=0}^n)$ is Gaussian, and its mean and covariance are updated by closed-form formulas with a Kalman-gain-like matrix. The numerical claim is that this single surrogate matches EnKF-based data assimilation on the true equations and approaches FNO-level forecasting on three PDEs with turbulence, shocks, and intermittency. The paper explicitly notes that decoding the filter mean is not generally a valid estimator of the unobserved-state mean; the DA loss term is what makes the decoded posterior mean trustworthy in practice.

Load-bearing premise

The load-bearing premise is that the unobserved part of the target system can be encoded into finitely many latent numbers whose next state is linear in those numbers plus Gaussian noise, with coefficients depending only on observed variables. The paper gives no proof, error bound, or dimension estimate for this representability; if it fails, the closed-form filter is exact for the surrogate but not for the true system.

Editorial extensions

If this is right

  • One trained CGKN replaces the learn-then-filter pipeline: the same network is optimized jointly for forecast accuracy and state-estimation accuracy, because the DA update is closed form and differentiable.
  • DA cost drops to $O(d_v^3)$ with $d_v$ the latent dimension; on the 2-D Navier-Stokes test the authors report 80 seconds for 20,000 assimilation steps versus about 7 hours for an EnKF on the true equations.
  • Because the filter is analytical, there is no ensemble size, covariance inflation, or localization to tune, and no need to know the prior distribution of the initial state.
  • Forecast accuracy approaches that of a Fourier neural operator on the tested PDEs while DA accuracy is comparable to an EnKF that uses the true governing equations, so a single model serves both tasks.
  • The same design principle can be carried into outer-loop applications such as inverse problems, design optimization, and optimal control, where a tractable posterior is the enabling ingredient.

Reading between the lines

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

  • Beyond the paper, the conditional-linear ansatz may be easier to satisfy than full Koopman linearization for chaotic systems, since nonlinearity is kept in the observed-state coefficients; a direct test is to measure how the one-step residual shrinks as latent dimension $d_v$ grows.
  • Beyond the paper, the explicit caveat that decoding the filter mean is not generally valid makes the DA loss load-bearing; removing only $L_{DA}$ from the total loss would isolate how much DA accuracy comes from supervision rather than from the closed-form filter.
  • Beyond the paper, the framework suggests a sensor-placement criterion: place observations where the unobserved field's conditional dynamics are closest to linear, which is checkable by comparing DA errors across candidate observation masks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a discrete-time conditional Gaussian Koopman network (CGKN), a data-driven surrogate model for partially observed nonlinear dynamical systems. The surrogate expresses the observed state u1 through a nonlinear map F1(u1)+G1(u1)v and the latent state v through F2(u1)+G2(u1)v, with Gaussian noises, so that the filtering distribution of the latent state is Gaussian and can be updated with analytical formulas. The model is trained with a combined loss involving autoencoder reconstruction, one-step/multi-step forecasts, and a data-assimilation term. The authors report experiments on viscous Burgers, Kuramoto-Sivashinsky, and 2D Navier-Stokes equations, comparing state-forecast accuracy with DNN, CNN, and FNO, and DA accuracy with EnKF and spatial interpolation. The main claims are that CGKN achieves forecast accuracy comparable to FNO and DA accuracy comparable to EnKF applied to the true equations, while being far cheaper than ensemble DA.

Significance. If the central claims hold, CGKN is a practically valuable contribution: one trained surrogate supports both open-loop forecasting and online data assimilation without ensemble integration, and the analytical filter updates are attractive for high-dimensional systems. The paper provides clear code availability, uses standard PDE benchmarks, and compares against strong baselines including FNO and EnKF on the true governing equations. The main weakness is that the core structural assumption—the existence of a low-dimensional conditionally linear latent representation—is asserted rather than established. The empirical DA results, as designed, cannot by themselves validate that assumption because the DA loss supervises the decoder output directly. This gap does not necessarily invalidate the benchmark results, but it substantially limits the confidence in the paper's generality claims and should be addressed before publication.

major comments (4)
  1. [Section 2, Eq. (2.4) and Appendix B] The framework assumes the existence of a finite-dimensional latent state v = φ(u2) such that the dynamics become conditionally linear in v, i.e., Eq. (2.4). Appendix B merely restates this assumption as a 'generalized application of Koopman theory' but provides no existence theorem, no error bound, and no estimate of the required latent dimension. Since the analytical filter (Eq. (2.6)) and the efficiency claims rely entirely on this structure, the paper's generality claim for partially observed nonlinear PDEs is not established. I recommend either proving the representation for a meaningful class of systems (e.g., systems with finite-dimensional invariant subspaces), or adding a statistical consistency diagnostic that does not depend on the decoder's flexibility, and explicitly limiting the scope if the validation remains only the three benchmarks.
  2. [Section 2.3, Eq. (2.12)] The DA loss LDA trains the decoder through μ = ψ(μv) against the true unobserved states u2⋆. A sufficiently flexible decoder can therefore make the DA mean accurate even when the conditional-linear latent dynamics are misspecified, because the decoder acts as a post-processing map. Consequently, the low DA errors in Table 3.1 do not validate the conditional Gaussian structure; they only indicate that the observation-to-state regression fits the benchmark data. The manuscript acknowledges that 'such a transformation of the estimated mean μv generally does not hold' but does not draw the implication that the filter's correctness is untested. Please add a filter-consistency check, such as standardized innovations with respect to the predicted covariance, or a comparison of the analytical posterior covariance propagated through the decoder Jacobian against the empirical residual covariance.
  3. [Table 3.1 and all numerical experiments] No repeated runs, random seeds, or standard deviations are reported. The comparisons that underlie the central claims (CGKN vs. FNO for forecast; CGKN vs. EnKF for DA) are based on single numbers, which is insufficient given the stochasticity of neural-network training. The authors should rerun each method with multiple seeds and report mean ± standard deviation, or at least state the number of runs used.
  4. [Sections 3.2 and 3.3] The evaluation protocol for the experiments with added observation noise is ambiguous. The text states that noise N(0,0.2^2) for KS and N(0,2.5^2) for NS is 'added to the true states', and the DA error is defined as the MSE between 'true unobserved states' and the posterior mean. If the reference is the noise-corrupted state, then the MSE contains an irreducible noise floor and is not a standard measure of DA performance; if the reference is the clean state, this should be stated explicitly. In addition, training LDA against noise-corrupted u2⋆ may encourage the decoder to reproduce noise. Please clarify the reference used in Table 3.1 and in Figures 3.3, 3.5, 3.7, 3.10, and 3.11, and if necessary, recompute the reported errors against the clean reference.
minor comments (5)
  1. [Section 3 and Algorithm 1] The paper does not report training hyperparameters such as learning rate, batch size, number of epochs, or the hardware used for timing comparisons. Including these details, or at least a pointer to a configuration file in the released code, would substantially improve reproducibility.
  2. [Section 3] The claim that 'a comparable number of parameters is used in all those deep-learning-based models' is not backed by parameter counts for DNN, CNN, and FNO. Please include these numbers, ideally in Table 3.1 or in a separate table.
  3. [Section 2.2.2] The statement that 'such a transformation of the estimated mean μv generally does not hold, but the DA loss term used to train the CGKN model can ensure the validity of the posterior mean μ' is conceptually confusing. It should be expanded to explain why training makes the decoder-map-of-mean an acceptable estimator, and what limitations remain for the covariance.
  4. [Abstract and Section 4] The abstract and conclusion claim that CGKN serves as an example for unifying SciML with outer-loop applications such as design optimization, inverse problems, and optimal control, but none of these applications is demonstrated. Consider rewording to 'may inspire' or removing this claim.
  5. [Throughout] There are minor typographical issues, e.g., 'latent sates' in Section 2.3 and inconsistent use of 'the sates' in Section 3.1. A careful proofread would be beneficial.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central forecast and DA claims are validated on held-out data against independent baselines, and the conditional-linear latent embedding is a learned modeling assumption rather than a conclusion derived from itself.

full rationale

The paper's derivation chain is not circular. The surrogate model in Eq. (2.4) is not claimed to be derived from the data; it is an architecture whose parameters are trained using losses in Eqs. (2.8)-(2.12), and the resulting forecast and DA errors in Table 3.1 are computed on test data. The conditional Gaussian filter formulae in Eq. (2.6) are standard external results cited to Liptser and Shiryaev [127], not to the authors' own work. The DA comparison against an ensemble Kalman filter applied to the true governing equations provides an independent benchmark, so the reported DA accuracy is not forced by construction. The conditional-linear representability of unobserved states in Appendix B is an unproven assumption and a genuine generality limitation, but the paper does not present it as a theorem derived from Koopman theory; it is learned and empirically tested, so it is not a circular reduction. Self-citations to [100] and [101] describe the prior continuous-time CGKN framework and the DA loss, but the discrete-time extension and its PDE evaluations stand on their own numerical experiments. No equation is defined in terms of the quantity it is supposed to predict, and no fitted parameter is renamed as a prediction. The absence of an existence proof or dimension estimate for the latent embedding is a correctness and applicability risk, not a circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 1 invented entities

The central structure rests on an assumed conditional-linear embedding (v), noise matrices entered into Kalman-like formulas, and hand-set latent dimensions. These are pulled from architecture choices and neural network training rather than derived from the PDEs, so the paper contributes a learned fit and a numerical demonstration, not a new physical law.

free parameters (4)
  • Latent dimension dv = 10 (Burgers), 12 (K-S), 256 (N-S)
    Chosen by hand for each example; controls representational capacity and DA complexity O(dv^3). No sensitivity analysis is reported.
  • sigma1 noise coefficient = Estimated via RMSE of one-step predictions, Eq. (2.13)
    Enter the analytical DA formulas in Eq. (2.6). Estimated post-training from the surrogate's own errors rather than derived from the true system.
  • sigma2 noise coefficient = Set manually or trainable; values not reported
    Appears in the covariance update in Eq. (2.6). The paper says results are robust to different choices but omits the actual values and tests.
  • Loss weights lambda_AE, lambda_u, lambda_v, lambda_DA = 1/d2, 1/d, 1/dv, 1/d2
    Inverse-dimensional weighting chosen by design; not tuned per problem, but affects the training objective and therefore the learned model.
assumptions (3)
  • standard math Conditional Gaussian filter formulas from Liptser-Shiryaev are valid for the surrogate system in Eq. (2.4).
    Invoked in Section 2.2.2 and Appendix C. This is a standard result under the assumed Gaussian noise model and Gaussian initial condition.
  • ad hoc to paper For each PDE, a finite-dimensional latent representation v with conditionally linear dynamics exists and is learnable by neural networks.
    Introduced in Section 2.1 and Appendix B. No theorem, convergence guarantee, or dimension estimate is provided; only numerical demonstrations.
  • domain assumption Gaussian white noise on both u1 and v sufficiently captures the surrogate model error.
    Used throughout Eq. (2.4) and the UQ procedure in Section 2.4. No normality diagnostics or tests against non-Gaussian model error are reported.
invented entities (1)
  • Latent state v = phi(u2)
    purpose: A learned compressed representation of unobserved states that makes the dynamics conditionally linear, enabling analytical DA formulas.
    The latent state is an internal modeling construct with no direct physical observable and no independent falsifiable prediction. Its validity is only assessed indirectly through forecast and DA MSE on test data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Modeling Partially Observed Nonlinear Dynamical Systems and Efficient Data Assimilation via Discrete-Time Conditional Gaussian Koopman Network." pith.science (2026). https://pith.science/paper/WCWDJRN4

@misc{pith2026250708749,
  author       = {Pith},
  title        = {Pith review of: Modeling Partially Observed Nonlinear Dynamical Systems and Efficient Data Assimilation via Discrete-Time Conditional Gaussian Koopman Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WCWDJRN4}},
  note         = {Machine review of arXiv:2507.08749}
}
read the original abstract

A discrete-time conditional Gaussian Koopman network (CGKN) is developed in this work to learn surrogate models that can perform efficient state forecast and data assimilation (DA) for high-dimensional complex dynamical systems, e.g., systems governed by nonlinear partial differential equations (PDEs). Focusing on nonlinear partially observed systems that are common in many engineering and earth science applications, this work exploits Koopman embedding to discover a proper latent representation of the unobserved system states, such that the dynamics of the latent states are conditional linear, i.e., linear with the given observed system states. The modeled system of the observed and latent states then becomes a conditional Gaussian system, for which the posterior distribution of the latent states is Gaussian and can be efficiently evaluated via analytical formulae. The analytical formulae of DA facilitate the incorporation of DA performance into the learning process of the modeled system, which leads to a framework that unifies scientific machine learning (SciML) and data assimilation. The performance of discrete-time CGKN is demonstrated on several canonical problems governed by nonlinear PDEs with intermittency and turbulent features, including the viscous Burgers' equation, the Kuramoto-Sivashinsky equation, and the 2-D Navier-Stokes equations, with which we show that the discrete-time CGKN framework achieves comparable performance as the state-of-the-art SciML methods in state forecast and provides efficient and accurate DA results. The discrete-time CGKN framework also serves as an example to illustrate unifying the development of SciML models and their other outer-loop applications such as design optimization, inverse problems, and optimal control.

Figures

Figures reproduced from arXiv: 2507.08749 by the authors.

Figure 2.1
Figure 2.1. Schematic diagram of the architecture and application of the conditional Gaussian Koopman network (CGKN). a, Overview of the transformation from discrete dynamical systems to surrogate models following condi￾tional Gaussian structure via generalized application of Koopman theory. b, The architecture and workflow of CGKN. The CGKN, consisting of an encoder φ, a decoder ψ and sub-networks η, is developed to learn the … view at source ↗
Figure 2
Figure 2. (b) are optimized by minimizing a total loss function: [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. shows the spatiotemporal plot of the true simulation of the viscous Burgers’ equation, [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (12 more)
Figure 3.1
Figure 3.1. Figure 3.1: Spatiotemporal evolution of the DA results for the viscous Burgers’ equation. The spatiotemporal plots of true simulation, DA posterior mean from CGKN, DA posterior mean from EnKF, and interpolation result are shown in each sub-figure. EnKF is applied to the true gov…
Figure 3.2
Figure 3.2. Figure 3.2: Results of state forecast for the viscous Burger’s equation. Starting from three different initial conditions in test data, the evolution of true spatial functions is compared with predictive spatial functions from various models, including CGKN, DNN, CNN, and FNO. I…
Figure 3.3
Figure 3.3. Figure 3.3: Time series of the DA results for the viscous Burgers’ equation. True signals, DA posterior means from CGKN and EnKF together with uncertainty areas, and interpolation results for three unobserved states are shown in the figure. The uncertainties of the two standard …
Figure 3.4
Figure 3.4. Figure 3.4: Spatial profiles of the DA results for the viscous Burgers’ equation. Starting from a random guess, the DA results from CGKN are compared with the true spatial functions in the first row, while those from EnKF are displayed in the second row. The interpolation result…
Figure 3
Figure 3. Figure 3: displays the spatiotemporal plot of the true simulation of the K-S equation, posterior [PITH_FULL_IMAGE:figures/full_fig_p020_3.png]
Figure 3.5
Figure 3.5. Figure 3.5: Spatiotemporal evolution of the DA results for the Kuramoto-Sivashinsky equation. The spatiotemporal plots of true simulation, DA posterior mean from CGKN, DA posterior mean from EnKF, and interpolation result are displayed in each sub-figure [PITH_FULL_IMAGE:figure…
Figure 3.6
Figure 3.6. Figure 3.6: Results of state forecast for the Kuramoto-Sivashinsky equation. The evolution of true spatial functions is compared with predictive spatial functions from various models including CGKN, DNN, CNN, and FNO [PITH_FULL_IMAGE:figures/full_fig_p022_3_6.png]
Figure 3.7
Figure 3.7. Figure 3.7: Time series of the DA results for the Kuramoto-Sivashinsky equation. True signal, DA posterior mean from CGKN and EnKF together with uncertainty area, and interpolation results for the unobserved state at spatial position 9.625 are presented. The uncertainties of the…
Figure 3.8
Figure 3.8. Figure 3.8: Spatial profiles of the DA results for the Kuramoto-Sivashinsky equation. Starting from a random guess, the DA results from CGKN are compared with the true spatial functions in the first row, while those from EnKF are presented in the second row. The interpolation re…
Figure 3.9
Figure 3.9. Figure 3.9: Results of state forecast for the Navier-Stokes equations. The evolution of the true vorticity field is compared with the predictive vorticity field from various models including CGKN, DNN, CNN, and FNO [PITH_FULL_IMAGE:figures/full_fig_p026_3_9.png]
Figure 3.10
Figure 3.10. Figure 3.10: Spatiotemporal evolution of the DA results for the Navier-Stokes equations. Heatmaps of each row display the vorticity fields for the true simulation, the DA posterior means from CGKN and EnKF, and the interpolation result [PITH_FULL_IMAGE:figures/full_fig_p027_3_10.png]
Figure 3.11
Figure 3.11. Figure 3.11: Time series of the DA results for the Navier-Stokes equations. The true signal, DA posterior means from CGKN and EnKF together with uncertainty areas of two standard deviations, and interpolation result for the unobserved state at spatial position (0.5625, 0.5625) a…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

129 extracted references · 74 canonical work pages

  1. [1]

    Dynamical systems: examples of complex behaviour

    J ¨urgen Jost. Dynamical systems: examples of complex behaviour . Springer Science & Business Media, 2005

    2005

  2. [2]

    Introduction to applied nonlinear dynamical systems and chaos , vol- ume 2

    Stephen Wiggins. Introduction to applied nonlinear dynamical systems and chaos , vol- ume 2. Springer, 2003

    2003

  3. [3]

    Nonlinear dynamics and statistical theories for basic geophysical flows

    Andrew Majda and Xiaoming Wang. Nonlinear dynamics and statistical theories for basic geophysical flows. Cambridge University Press, 2006

    2006

  4. [4]

    Stochastic Methods for Modeling and Predicting Complex Dynamical Systems: Uncertainty Quantification, State Estimation, and Reduced-Order Models

    Nan Chen. Stochastic Methods for Modeling and Predicting Complex Dynamical Systems: Uncertainty Quantification, State Estimation, and Reduced-Order Models. Springer Nature, 2023. 29

    2023

  5. [5]

    Physics-informed neural networks for parameter learning of wildfire spread- ing

    Konstantinos V ogiatzoglou, Costas Papadimitriou, Vasilis Bontozoglou, and Konstantinos Ampountolas. Physics-informed neural networks for parameter learning of wildfire spread- ing. Computer Methods in Applied Mechanics and Engineering, 434:117545, 2025

    2025

  6. [6]

    Development and application of a fluid mechanics analysis framework based on complex network theory

    Zihao Wang, Guiyong Zhang, Tiezhi Sun, and Bo Zhou. Development and application of a fluid mechanics analysis framework based on complex network theory. Computer Methods in Applied Mechanics and Engineering, 435:117677, 2025

    2025

  7. [7]

    Nonlinear climate dynamics

    Henk A Dijkstra. Nonlinear climate dynamics. Cambridge University Press, 2013

    2013

  8. [8]

    Attribution of climate ex- treme events

    Kevin E Trenberth, John T Fasullo, and Theodore G Shepherd. Attribution of climate ex- treme events. Nature Climate Change, 5(8):725–730, 2015

    2015

Show all 129 references
  1. [9]

    Extreme events in turbulent flow

    HK Moffatt. Extreme events in turbulent flow. Journal of Fluid Mechanics, 914:F1, 2021

    2021

  2. [10]

    Introduction to PDEs and Waves for the Atmosphere and Ocean, volume 9

    Andrew Majda. Introduction to PDEs and Waves for the Atmosphere and Ocean, volume 9. American Mathematical Soc., 2003

    2003

  3. [11]

    Model error, information barriers, state estimation and prediction in complex multiscale systems

    Andrew J Majda and Nan Chen. Model error, information barriers, state estimation and prediction in complex multiscale systems. Entropy, 20(9):644, 2018

    2018

  4. [12]

    Intermittency and the Lorenz model

    Paul Manneville and Yves Pomeau. Intermittency and the Lorenz model. Physics Letters A, 75(1-2):1–2, 1979

    1979

  5. [13]

    Scientific ma- chine learning for closure models in multiscale problems: A review

    Benjamin Sanderse, Panos Stinis, Romit Maulik, and Shady E Ahmed. Scientific ma- chine learning for closure models in multiscale problems: A review. arXiv preprint arXiv:2403.02913, 2024

    2024 arXiv

  6. [14]

    Learning about structural errors in models of complex dynamical systems

    Jin-Long Wu, Matthew E Levine, Tapio Schneider, and Andrew Stuart. Learning about structural errors in models of complex dynamical systems. Journal of Computational Physics, 513:113157, 2024

    2024

  7. [15]

    A physics-informed data-driven algorithm for ensemble forecast of complex turbulent systems

    Nan Chen and Di Qi. A physics-informed data-driven algorithm for ensemble forecast of complex turbulent systems. Applied Mathematics and Computation, 466:128480, 2024

    2024

  8. [16]

    Data-driven discovery of partial differential equations

    Samuel H Rudy, Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Data-driven discovery of partial differential equations. Science advances, 3(4):e1602614, 2017

    2017

  9. [17]

    Sparse dynam- ics for partial differential equations

    Hayden Schaeffer, Russel Caflisch, Cory D Hauck, and Stanley Osher. Sparse dynam- ics for partial differential equations. Proceedings of the National Academy of Sciences , 110(17):6634–6639, 2013

    2013

  10. [18]

    Strategies for reduced-order models for predicting the statistical responses and uncertainty quantification in complex turbulent dynamical systems

    Andrew J Majda and Di Qi. Strategies for reduced-order models for predicting the statistical responses and uncertainty quantification in complex turbulent dynamical systems. SIAM Review, 60(3):491–549, 2018

    2018

  11. [19]

    The gnat method for nonlinear model reduction: effective implementation and application to computational fluid dynamics and turbulent flows

    Kevin Carlberg, Charbel Farhat, Julien Cortial, and David Amsallem. The gnat method for nonlinear model reduction: effective implementation and application to computational fluid dynamics and turbulent flows. Journal of Computational Physics, 242:623–647, 2013. 30

    2013

  12. [20]

    Reduced-order modelling for flow control, volume 528

    Bernd R Noack, Marek Morzynski, and Gilead Tadmor. Reduced-order modelling for flow control, volume 528. Springer Science & Business Media, 2011

    2011

  13. [21]

    A causality-based learning approach for discovering the underlying dynamics of complex systems from partial observations with stochastic parame- terization

    Nan Chen and Yinling Zhang. A causality-based learning approach for discovering the underlying dynamics of complex systems from partial observations with stochastic parame- terization. Physica D: Nonlinear Phenomena, 449:133743, 2023

    2023

  14. [22]

    CEBoosting: Online sparse identification of dynamical systems with regime switching by causation entropy boosting

    Chuanqi Chen, Nan Chen, and Jin-Long Wu. CEBoosting: Online sparse identification of dynamical systems with regime switching by causation entropy boosting. Chaos: An Interdisciplinary Journal of Nonlinear Science, 33(8), 2023

    2023

  15. [23]

    Minimum reduced-order models via causal inference

    Nan Chen and Honghu Liu. Minimum reduced-order models via causal inference. Nonlin- ear Dynamics, pages 1–25, 2024

    2024

  16. [24]

    Solving and learning nonlinear pdes with Gaussian processes

    Yifan Chen, Bamdad Hosseini, Houman Owhadi, and Andrew M Stuart. Solving and learning nonlinear pdes with Gaussian processes. Journal of Computational Physics , 447:110668, 2021

    2021

  17. [25]

    Deep learning

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521(7553):436– 444, 2015

    2015

  18. [26]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems , 25, 2012

    2012

  19. [27]

    Learning representations by back-propagating errors

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating errors. nature, 323(6088):533–536, 1986

    1986

  20. [28]

    Convolutional networks for images, speech, and time series

    Yann LeCun, Yoshua Bengio, et al. Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks, 1995

    1995

  21. [29]

    Handwritten digit recognition with a back-propagation network

    Yann LeCun, Bernhard Boser, John Denker, Donnie Henderson, Richard Howard, Wayne Hubbard, and Lawrence Jackel. Handwritten digit recognition with a back-propagation network. Advances in Neural Information Processing Systems, 2, 1989

    1989

  22. [30]

    Physics-informed machine learning

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning. Nature Reviews Physics , 3(6):422–440, 2021

    2021

  23. [31]

    Turbulence modeling in the age of data

    Karthik Duraisamy, Gianluca Iaccarino, and Heng Xiao. Turbulence modeling in the age of data. Annual review of fluid mechanics, 51(1):357–377, 2019

    2019

  24. [32]

    Learning dynamical systems from data: An introduction to physics-guided deep learning

    Rose Yu and Rui Wang. Learning dynamical systems from data: An introduction to physics-guided deep learning. Proceedings of the National Academy of Sciences , 121(27):e2311808121, 2024

    2024

  25. [33]

    Physics-informed machine learning ap- proach for reconstructing reynolds stress modeling discrepancies based on dns data

    Jian-Xun Wang, Jin-Long Wu, and Heng Xiao. Physics-informed machine learning ap- proach for reconstructing reynolds stress modeling discrepancies based on dns data. Physi- cal Review Fluids, 2(3):034603, 2017. 31

    2017

  26. [34]

    Physics-informed machine learning approach for augmenting turbulence models: A comprehensive framework

    Jin-Long Wu, Heng Xiao, and Eric Paterson. Physics-informed machine learning approach for augmenting turbulence models: A comprehensive framework. Physical Review Fluids, 3(7):074602, 2018

    2018

  27. [35]

    Approximation by superpositions of a sigmoidal function

    George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989

    1989

  28. [36]

    Multilayer feedforward networks are universal approximators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989

    1989

  29. [37]

    Tianping Chen and Hong Chen. Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems.IEEE transactions on neural networks, 6(4):911–917, 1995

    1995

  30. [38]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  31. [39]

    Neural ordi- nary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordi- nary differential equations. Advances in neural information processing systems, 31, 2018

    2018

  32. [40]

    Physics-informed neural net- works: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. Physics-informed neural net- works: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics , 378:686–707, 2019

    2019

  33. [41]

    Learn- ing nonlinear operators via deeponet based on the universal approximation theorem of op- erators

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learn- ing nonlinear operators via deeponet based on the universal approximation theorem of op- erators. Nature machine intelligence, 3(3):218–229, 2021

    2021

  34. [42]

    Fourier neural operator for paramet- ric partial differential equations

    Zongyi Li, Nikola Borislavov Kovachki, Kamyar Azizzadenesheli, Burigede liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for paramet- ric partial differential equations. In International Conference on Learning Representations, 2021

    2021

  35. [43]

    Neural dynamical operator: Continuous spatial-temporal model with gradient-based and derivative-free optimization methods

    Chuanqi Chen and Jin-Long Wu. Neural dynamical operator: Continuous spatial-temporal model with gradient-based and derivative-free optimization methods. Journal of Computa- tional Physics, 520:113480, 2025

    2025

  36. [44]

    MODNO: Multi-operator learning with distributed neural operators

    Zecheng Zhang. MODNO: Multi-operator learning with distributed neural operators. Com- puter Methods in Applied Mechanics and Engineering, 431:117229, 2024

    2024

  37. [45]

    B-deeponet: An enhanced Bayesian deeponet for solving noisy parametric pdes using accelerated replica exchange sgld.Journal of Computational Physics, 473:111713, 2023

    Guang Lin, Christian Moya, and Zecheng Zhang. B-deeponet: An enhanced Bayesian deeponet for solving noisy parametric pdes using accelerated replica exchange sgld.Journal of Computational Physics, 473:111713, 2023

    2023

  38. [46]

    A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks

    Chenxi Wu, Min Zhu, Qinyang Tan, Yadhu Kartha, and Lu Lu. A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering, 403:115671, 2023. 32

    2023

  39. [47]

    Auto-encoding variational bayes, 2013

    Diederik P Kingma, Max Welling, et al. Auto-encoding variational bayes, 2013

    2013

  40. [48]

    Generative adversarial networks

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communi- cations of the ACM, 63(11):139–144, 2020

    2020

  41. [49]

    Enforc- ing statistical constraints in generative adversarial networks for modeling chaotic dynamical systems

    Jin-Long Wu, Karthik Kashinath, Adrian Albert, Dragos Chirila, Heng Xiao, et al. Enforc- ing statistical constraints in generative adversarial networks for modeling chaotic dynamical systems. Journal of Computational Physics, 406:109209, 2020

    2020

  42. [50]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Ad- vances in neural information processing systems, 33:6840–6851, 2020

    2020

  43. [51]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020

    2011 arXiv

  44. [52]

    Data-driven stochastic closure modeling via conditional diffusion model and neural operator

    Xinghao Dong, Chuanqi Chen, and Jin-Long Wu. Data-driven stochastic closure modeling via conditional diffusion model and neural operator. Journal of Computational Physics , 534:114005, 2025

    2025

  45. [53]

    Condi- tional neural field latent diffusion model for generating spatiotemporal turbulence

    Pan Du, Meet Hemant Parikh, Xiantao Fan, Xin-Yang Liu, and Jian-Xun Wang. Condi- tional neural field latent diffusion model for generating spatiotemporal turbulence. Nature Communications, 15(1):10416, 2024

    2024

  46. [54]

    Can diffusion models capture extreme event statistics? Computer Methods in Applied Mechanics and Engineering , 435:117589, 2025

    Stamatis Stamatelopoulos and Themistoklis P Sapsis. Can diffusion models capture extreme event statistics? Computer Methods in Applied Mechanics and Engineering , 435:117589, 2025

    2025

  47. [55]

    Generative learning of the solution of parametric partial differential equations using guided diffusion models and vir- tual observations

    Han Gao, Sebastian Kaltenbach, and Petros Koumoutsakos. Generative learning of the solution of parametric partial differential equations using guided diffusion models and vir- tual observations. Computer Methods in Applied Mechanics and Engineering, 435:117654, 2025

    2025

  48. [56]

    Active learning literature survey

    Burr Settles. Active learning literature survey. 2009

    2009

  49. [57]

    Learning active learning from data

    Ksenia Konyushkova, Raphael Sznitman, and Pascal Fua. Learning active learning from data. Advances in neural information processing systems, 30, 2017

    2017

  50. [58]

    Simulation-based optimal Bayesian experimental de- sign for nonlinear systems

    Xun Huan and Youssef M Marzouk. Simulation-based optimal Bayesian experimental de- sign for nonlinear systems. Journal of Computational Physics, 232(1):288–317, 2013

    2013

  51. [59]

    Bayesian inference in physics.Reviews of Modern Physics, 83(3):943– 999, 2011

    Udo V on Toussaint. Bayesian inference in physics.Reviews of Modern Physics, 83(3):943– 999, 2011

    2011

  52. [60]

    Active-learning-driven surrogate model- ing for efficient simulation of parametric nonlinear systems

    Harshit Kapadia, Lihong Feng, and Peter Benner. Active-learning-driven surrogate model- ing for efficient simulation of parametric nonlinear systems. Computer Methods in Applied Mechanics and Engineering, 419:116657, 2024. 33

    2024

  53. [61]

    Optimum experimental designs, with SAS, volume 34

    Anthony Atkinson, Alexander Donev, and Randall Tobias. Optimum experimental designs, with SAS, volume 34. OUP Oxford, 2007

    2007

  54. [62]

    Active learning of model discrepancy with Bayesian experimental design

    Huchen Yang, Chuanqi Chen, and Jin-Long Wu. Active learning of model discrepancy with Bayesian experimental design. arXiv preprint arXiv:2502.05372, 2025

    2025 arXiv

  55. [63]

    A new approach to linear filtering and prediction problems

    Rudolph Emil Kalman. A new approach to linear filtering and prediction problems. Trans. ASME, D, 82:35–44, 1960

    1960

  56. [64]

    New results in linear filtering and prediction theory

    Rudolph E Kalman and Richard S Bucy. New results in linear filtering and prediction theory. Journal of Fluids Engineering, 83:95–108, 1961

    1961

  57. [65]

    Filtering complex turbulent systems

    Andrew J Majda and John Harlim. Filtering complex turbulent systems. Cambridge Univer- sity Press, 2012

    2012

  58. [66]

    Data assimilation

    Kody Law, Andrew Stuart, and Kostas Zygalakis. Data assimilation. Cham, Switzerland: Springer, 214, 2015

    2015

  59. [67]

    Learning stochastic closures using ensemble Kalman inversion

    Tapio Schneider, Andrew M Stuart, and Jin-Long Wu. Learning stochastic closures using ensemble Kalman inversion. Transactions of Mathematics and Its Applications , 5(1):tnab003, 2021

    2021

  60. [68]

    Ensemble Kalman inversion for sparse learning of dynamical systems from time-averaged data

    Tapio Schneider, Andrew M Stuart, and Jin-Long Wu. Ensemble Kalman inversion for sparse learning of dynamical systems from time-averaged data. Journal of Computational Physics, 470:111559, 2022

    2022

  61. [69]

    Ensemble Kalman methods for inverse problems

    Marco A Iglesias, Kody JH Law, and Andrew M Stuart. Ensemble Kalman methods for inverse problems. Inverse Problems, 29(4):045001, 2013

    2013

  62. [70]

    An efficient continuous data assimilation al- gorithm for the sabra shell model of turbulence

    Nan Chen, Yuchen Li, and Evelyn Lunasin. An efficient continuous data assimilation al- gorithm for the sabra shell model of turbulence. Chaos: An Interdisciplinary Journal of Nonlinear Science, 31(10), 2021

    2021

  63. [71]

    Efficient statistically accurate algorithms for the fokker– planck equation in large dimensions

    Nan Chen and Andrew J Majda. Efficient statistically accurate algorithms for the fokker– planck equation in large dimensions. Journal of Computational Physics , 354:242–268, 2018

    2018

  64. [72]

    Geir Evensen. Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to forecast error statistics.Journal of Geophysical Research: Oceans, 99(C5):10143–10162, 1994

    1994

  65. [73]

    An ensemble Kalman-Bucy filter for continuous data assimilation

    Kay Bergemann and Sebastian Reich. An ensemble Kalman-Bucy filter for continuous data assimilation. Meteorologische Zeitschrift, 21(3):213, 2012

    2012

  66. [74]

    The ensemble Kalman filter: Theoretical formulation and practical imple- mentation

    Geir Evensen. The ensemble Kalman filter: Theoretical formulation and practical imple- mentation. Ocean dynamics, 53:343–367, 2003

    2003

  67. [75]

    Ensemble data assimilation without perturbed observations

    Jeffrey S Whitaker and Thomas M Hamill. Ensemble data assimilation without perturbed observations. Monthly weather review, 130(7):1913–1924, 2002. 34

    1913

  68. [76]

    Analysis scheme in the ensemble Kalman filter

    Gerrit Burgers, Peter Jan van Leeuwen, and Geir Evensen. Analysis scheme in the ensemble Kalman filter. Monthly weather review, 126(6):1719–1724, 1998

    1998

  69. [77]

    Data-driven variational multiscale reduced order models

    Changhong Mou, Birgul Koc, Omer San, Leo G Rebholz, and Traian Iliescu. Data-driven variational multiscale reduced order models. Computer Methods in Applied Mechanics and Engineering, 373:113470, 2021

    2021

  70. [78]

    Dynamic data-driven reduced-order models

    Benjamin Peherstorfer and Karen Willcox. Dynamic data-driven reduced-order models. Computer Methods in Applied Mechanics and Engineering, 291:21–41, 2015

    2015

  71. [79]

    Data-driven pod- galerkin reduced order model for turbulent flows

    Saddam Hijazi, Giovanni Stabile, Andrea Mola, and Gianluigi Rozza. Data-driven pod- galerkin reduced order model for turbulent flows. Journal of Computational Physics , 416:109513, 2020

    2020

  72. [80]

    Data-driven model reduction, Wiener projections, and the Koopman-Mori-Zwanzig formalism

    Kevin K Lin and Fei Lu. Data-driven model reduction, Wiener projections, and the Koopman-Mori-Zwanzig formalism. Journal of Computational Physics, 424:109864, 2021

    2021

  73. [81]

    Operator inference driven data assimilation for high fidelity neutron transport

    Wei Xiao, Xiaojing Liu, Jianhua Zu, Xiang Chai, Hui He, and Tengfei Zhang. Operator inference driven data assimilation for high fidelity neutron transport. Computer Methods in Applied Mechanics and Engineering, 430:117214, 2024

    2024

  74. [82]

    Combining stochastic parameterized reduced-order models with machine learning for data assimilation and uncertainty quan- tification with partial observations

    Changhong Mou, Leslie M Smith, and Nan Chen. Combining stochastic parameterized reduced-order models with machine learning for data assimilation and uncertainty quan- tification with partial observations. Journal of Advances in Modeling Earth Systems , 15(10):e2022MS003597, 2023

    2023

  75. [83]

    Stochastic parameterization: Toward a new view of weather and climate models

    Judith Berner, Ulrich Achatz, Lauriane Batte, Lisa Bengtsson, Alvaro de la C ´amara, Han- nah M Christensen, Matteo Colangeli, Danielle RB Coleman, Daan Crommelin, Stamen I Dolaptchiev, et al. Stochastic parameterization: Toward a new view of weather and climate models. Bullet...

    2017

  76. [84]

    Toward a stochastic parameterization of ocean mesoscale eddies

    PierGianLuca Porta Mana and Laure Zanna. Toward a stochastic parameterization of ocean mesoscale eddies. Ocean Modelling, 79:1–20, 2014

    2014

  77. [85]

    Simulating weather regimes: Impact of model resolution and stochastic parameterization

    Andrew Dawson and TN Palmer. Simulating weather regimes: Impact of model resolution and stochastic parameterization. Climate Dynamics, 44:2177–2193, 2015

    2015

  78. [86]

    A stochastic precipitating quasi-geostrophic model

    Nan Chen, Changhong Mou, Leslie M Smith, and Yeyu Zhang. A stochastic precipitating quasi-geostrophic model. Physics of Fluids, 36(11), 2024

    2024

  79. [87]

    Anderson

    Jeffrey L. Anderson. An ensemble adjustment Kalman filter for data assimilation. Monthly Weather Review, 129(12):2884–2903, 2001

    2001

  80. [88]

    State, global, and local parameter estimation using local ensemble Kalman filters: Applications to online machine learning of chaotic dynamics

    Quentin Malartic, Alban Farchi, and Marc Bocquet. State, global, and local parameter estimation using local ensemble Kalman filters: Applications to online machine learning of chaotic dynamics. Quarterly Journal of the Royal Meteorological Society, 148(746):2167– 2193, 2022. 35

    2022

  81. [89]

    Machine learning with data assimilation and uncertainty quantification for dynamical systems: a review.IEEE/CAA Journal of Automatica Sinica, 10(6):1361–1387, 2023

    Sibo Cheng, C ´esar Quilodr ´an-Casas, Said Ouala, Alban Farchi, Che Liu, Pierre Tandeo, Ronan Fablet, Didier Lucor, Bertrand Iooss, Julien Brajard, et al. Machine learning with data assimilation and uncertainty quantification for dynamical systems: a review.IEEE/CAA Journal o...

    2023

  82. [90]

    Deep learning to represent sub- grid processes in climate models

    Stephan Rasp, Michael S Pritchard, and Pierre Gentine. Deep learning to represent sub- grid processes in climate models. Proceedings of the national academy of sciences , 115(39):9684–9689, 2018

    2018

  83. [91]

    Machine learning for model error inference and correction

    Massimo Bonavita and Patrick Laloyaux. Machine learning for model error inference and correction. Journal of Advances in Modeling Earth Systems, 12(12):e2020MS002232, 2020

    2020

  84. [92]

    Combining data as- similation and machine learning to infer unresolved scale parametrization

    Julien Brajard, Alberto Carrassi, Marc Bocquet, and Laurent Bertino. Combining data as- similation and machine learning to infer unresolved scale parametrization. Philosophical Transactions of the Royal Society A, 379(2194):20200086, 2021

    2021

  85. [93]

    Statistical variational data assimilation

    Amina Benaceur and Barbara Verf ¨urth. Statistical variational data assimilation. Computer Methods in Applied Mechanics and Engineering, 432:117402, 2024

    2024

  86. [94]

    Multi-domain encoder–decoder neural networks for latent data assimi- lation in dynamical systems

    Sibo Cheng, Yilin Zhuang, Lyes Kahouadji, Che Liu, Jianhua Chen, Omar K Matar, and Rossella Arcucci. Multi-domain encoder–decoder neural networks for latent data assimi- lation in dynamical systems. Computer Methods in Applied Mechanics and Engineering , 430:117201, 2024

    2024

  87. [95]

    Deep learning-enhanced ensemble-based data assimilation for high-dimensional nonlinear dy- namical systems

    Ashesh Chattopadhyay, Ebrahim Nabizadeh, Eviatar Bach, and Pedram Hassanzadeh. Deep learning-enhanced ensemble-based data assimilation for high-dimensional nonlinear dy- namical systems. Journal of Computational Physics, 477:111918, 2023

    2023

  88. [96]

    Kalmannet: Neural network aided Kalman filtering for partially known dynamics

    Guy Revach, Nir Shlezinger, Xiaoyong Ni, Adria Lopez Escoriza, Ruud JG Van Sloun, and Yonina C Eldar. Kalmannet: Neural network aided Kalman filtering for partially known dynamics. IEEE Transactions on Signal Processing, 70:1532–1547, 2022

    2022

  89. [97]

    Data assim- ilation networks

    Pierre Boudier, Anthony Fillion, Serge Gratton, Selime G¨urol, and Sixin Zhang. Data assim- ilation networks. Journal of Advances in Modeling Earth Systems, 15(4):e2022MS003353, 2023

    2023

  90. [98]

    Autodifferentiable ensemble Kalman filters

    Yuming Chen, Daniel Sanz-Alonso, and Rebecca Willett. Autodifferentiable ensemble Kalman filters. SIAM Journal on Mathematics of Data Science, 4(2):801–833, 2022

    2022

  91. [99]

    Reduced-order autodifferentiable ensemble Kalman filters

    Yuming Chen, Daniel Sanz-Alonso, and Rebecca Willett. Reduced-order autodifferentiable ensemble Kalman filters. Inverse Problems, 39(12):124001, 2023

    2023

  92. [100]

    CGNSDE: Conditional Gaussian neural stochastic differential equation for modeling complex systems and data assimilation

    Chuanqi Chen, Nan Chen, and Jin-Long Wu. CGNSDE: Conditional Gaussian neural stochastic differential equation for modeling complex systems and data assimilation. Com- puter Physics Communications, 304:109302, 2024

    2024

  93. [101]

    CGKN: A deep learning framework for modeling complex dynamical systems and efficient data assimilation

    Chuanqi Chen, Nan Chen, Yinling Zhang, and Jin-Long Wu. CGKN: A deep learning framework for modeling complex dynamical systems and efficient data assimilation. Jour- nal of Computational Physics, 532:113950, 2025. 36

    2025

  94. [102]

    Hamiltonian systems and transformation in Hilbert space

    Bernard O Koopman. Hamiltonian systems and transformation in Hilbert space. Proceed- ings of the National Academy of Sciences, 17(5):315–318, 1931

    1931

  95. [103]

    Applied koopmanism

    Marko Budi ˇsi´c, Ryan Mohr, and Igor Mezi ´c. Applied koopmanism. Chaos: An Interdisci- plinary Journal of Nonlinear Science, 22(4), 2012

    2012

  96. [104]

    Analysis of fluid flows via spectral properties of the Koopman operator.Annual review of fluid mechanics, 45(1):357–378, 2013

    Igor Mezi ´c. Analysis of fluid flows via spectral properties of the Koopman operator.Annual review of fluid mechanics, 45(1):357–378, 2013

    2013

  97. [105]

    Williams, Clarence W

    Matthew O. Williams, Clarence W. Rowley, and Ioannis G. Kevrekidis. A kernel-based method for data-driven Koopman spectral analysis. Journal of Computational Dynamics , 2(2):247–265, 2015

    2015

  98. [106]

    Physics-informed probabilistic learning of linear em- beddings of nonlinear dynamics with guaranteed stability

    Shaowu Pan and Karthik Duraisamy. Physics-informed probabilistic learning of linear em- beddings of nonlinear dynamics with guaranteed stability. SIAM Journal on Applied Dy- namical Systems, 19(1):480–509, 2020

    2020

  99. [107]

    Deep learning for universal linear embeddings of nonlinear dynamics

    Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Deep learning for universal linear embeddings of nonlinear dynamics. Nature communications, 9(1):4950, 2018

    2018

  100. [108]

    On the Koopman operator of algorithms

    Felix Dietrich, Thomas N Thiem, and Ioannis G Kevrekidis. On the Koopman operator of algorithms. SIAM Journal on Applied Dynamical Systems, 19(2):860–885, 2020

    2020

  101. [109]

    Prediction accuracy of dynamic mode decomposi- tion

    Hannah Lu and Daniel M Tartakovsky. Prediction accuracy of dynamic mode decomposi- tion. SIAM Journal on Scientific Computing, 42(3):A1639–A1662, 2020

    2020

  102. [110]

    Koopman operators for estimation and control of dynamical systems

    Samuel E Otto and Clarence W Rowley. Koopman operators for estimation and control of dynamical systems. Annual Review of Control, Robotics, and Autonomous Systems , 4(1):59–87, 2021

    2021

  103. [111]

    Koopman operator dynamical mod- els: Learning, analysis and control

    Petar Bevanda, Stefan Sosnowski, and Sandra Hirche. Koopman operator dynamical mod- els: Learning, analysis and control. Annual Reviews in Control, 52:197–212, 2021

    2021

  104. [112]

    Extended dynamic mode decomposition for inho- mogeneous problems

    Hannah Lu and Daniel M Tartakovsky. Extended dynamic mode decomposition for inho- mogeneous problems. Journal of Computational Physics, 444:110550, 2021

    2021

  105. [113]

    Brunton, Marko Budi ˇsi´c, Eurika Kaiser, and J

    Steven L. Brunton, Marko Budi ˇsi´c, Eurika Kaiser, and J. Nathan Kutz. Modern koopman theory for dynamical systems. SIAM Review, 64(2):229–340, 2022

    2022

  106. [114]

    Conditional Gaussian systems for multiscale nonlinear stochastic systems: Prediction, state estimation and uncertainty quantification

    Nan Chen and Andrew J Majda. Conditional Gaussian systems for multiscale nonlinear stochastic systems: Prediction, state estimation and uncertainty quantification. Entropy, 20(7):509, 2018

    2018

  107. [115]

    Conditional Gaussian nonlinear system: A fast preconditioner and a cheap surrogate model for complex nonlinear systems

    Nan Chen, Yingda Li, and Honghu Liu. Conditional Gaussian nonlinear system: A fast preconditioner and a cheap surrogate model for complex nonlinear systems. Chaos: An Interdisciplinary Journal of Nonlinear Science, 32(5), 2022

    2022

  108. [116]

    Filtering nonlinear turbulent dynamical systems through conditional Gaussian statistics

    Nan Chen and Andrew J Majda. Filtering nonlinear turbulent dynamical systems through conditional Gaussian statistics. Monthly Weather Review, 144(12):4885–4917, 2016. 37

    2016

  109. [117]

    Information barriers for noisy lagrangian tracers in filtering random incompressible flows

    Nan Chen, Andrew J Majda, and Xin T Tong. Information barriers for noisy lagrangian tracers in filtering random incompressible flows. Nonlinearity, 27(9):2133, 2014

    2014

  110. [118]

    Noisy lagrangian tracers for filtering random rotating compressible flows

    Nan Chen, Andrew J Majda, and Xin T Tong. Noisy lagrangian tracers for filtering random rotating compressible flows. Journal of Nonlinear Science, 25(3):451–488, 2015

    2015

  111. [119]

    Beating the curse of dimension with accurate statistics for the fokker–planck equation in complex turbulent systems

    Nan Chen and Andrew J Majda. Beating the curse of dimension with accurate statistics for the fokker–planck equation in complex turbulent systems. Proceedings of the National Academy of Sciences, 114(49):12864–12869, 2017

    2017

  112. [120]

    A closed-form nonlinear data assimilation algorithm for multi-layer flow fields

    Zhongrui Wang, Nan Chen, and Di Qi. A closed-form nonlinear data assimilation algorithm for multi-layer flow fields. arXiv preprint arXiv:2412.11042, 2024

    2024

  113. [121]

    Efficient quadratures for high- dimensional Bayesian data assimilation

    Ming Cheng, Peng Wang, and Daniel M Tartakovsky. Efficient quadratures for high- dimensional Bayesian data assimilation. Journal of Computational Physics , 506:112945, 2024

    2024

  114. [122]

    Localized ensemble Kalman inversion

    Xin T Tong and Matthias Morzfeld. Localized ensemble Kalman inversion. Inverse Prob- lems, 39(6):064002, 2023

    2023

  115. [123]

    Dropout ensemble Kalman inversion for high dimensional inverse problems

    Shuigen Liu, Sebastian Reich, and Xin T Tong. Dropout ensemble Kalman inversion for high dimensional inverse problems. arXiv preprint arXiv:2308.16784, 2023

    2023 arXiv

  116. [124]

    High-dimensional covariance estimation from a small number of samples

    David Vishny, Matthias Morzfeld, Kyle Gwirtz, Eviatar Bach, Oliver RA Dunbar, and Daniel Hodyss. High-dimensional covariance estimation from a small number of samples. Journal of Advances in Modeling Earth Systems, 16(9):e2024MS004417, 2024

    2024

  117. [125]

    A Monte Carlo implementation of the non- linear filtering problem to produce ensemble assimilations and forecasts

    Jeffrey L Anderson and Stephen L Anderson. A Monte Carlo implementation of the non- linear filtering problem to produce ensemble assimilations and forecasts. Monthly weather review, 127:18, 1999

    1999

  118. [126]

    Construction of correlation functions in two and three dimensions

    Gregory Gaspari and Stephen E Cohn. Construction of correlation functions in two and three dimensions. Quarterly Journal of the Royal Meteorological Society , 125(554):723– 757, 1999

    1999

  119. [127]

    Statistics of random processes II: Applications , volume 6

    Robert S Liptser and Albert N Shiryaev. Statistics of random processes II: Applications , volume 6. Springer Science & Business Media, 2013. 38 Appendix A. Koopman Theory For nonlinear dynamical systems in the general form: un+1 = G(un), (A.1) where un ∈ Rd is the system state...

    2013

  120. [128]

    Therefore, the full system remains highly nonlinear, fundamentally contrasting with the standard Koopman theory. The approximated finite-dimensional dynamical 39 system is given by: un+1 1 = F1 un 1 + G1 un 1 vn + σ1ϵn 1 , vn+1 = F2 un 1 + G2 un 1 vn + σ2ϵn 2 , (B.2) where vn ...

  121. [129]

    Appendix C

    The ϵn 1 ∈ Rd1 and ϵn 2 ∈ Rdv are independent Gaussian white noises with σ1 ∈ Rd1×d1 and σ2 ∈ Rdv×dv as noise coefficients. Appendix C. Conditional Gaussian Nonlinear Systems The system in Eq. (B.2) belongs to a class of nonlinear stochastic dynamical systems called con- ditio...

Pith tools

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