Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Neural ensemble Kalman filter: Data assimilation for compressible flows with shocks

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Pith's one-line read The paper claims that the standard ensemble Kalman filter fails near uncertain shocks because forecast distributions turn bimodal, and that updating in neural-network weight space, with weights aligned by nearest-neighbor chain training, re

desk verdict A genuinely new neural-space EnKF for shock-dominated DA, with a real but explicitly conceded gap: the weight-space Gaussianity/linearity the method rests on is only visually supported. read the letter →

arxiv 2602.23461 v2 pith:7FY3VU25 submitted 2026-02-26 physics.flu-dyn cs.LG

classification physics.flu-dyncs.LG
keywords dataassimilationensembleKalmanfiltercompressibleflowsshocksneuralnetworkstransferlearningbimodaldistributionsstructure-preserving
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 sets out to explain why the ensemble Kalman filter (EnKF) produces spurious oscillations and nonphysical states when assimilating data into compressible flows with shocks, and to offer a cure. Its diagnosis: uncertainty in the shock location makes the forecast distribution at points near the shock bimodal, violating the near-Gaussian assumption that the EnKF's linear update relies on. Its proposed cure is the neural EnKF, which moves the update out of physical state space and into the weight space of deep neural networks representing the flow fields. To make that parameter-space update meaningful, the networks are trained along a nearest-neighbor chain so that the weights vary smoothly across the ensemble. Numerical experiments on Burgers' equation, the Sod shock tube, and a 2D blast wave show the analysis ensembles preserving sharp fronts without oscillations, where the standard EnKF fails.

What carries the argument

The central object is the neural-space EnKF update: an ensemble Kalman update applied to the weights and biases of a shared-architecture network that maps spatial coordinates to flow variables, instead of to the physical state vector itself. The load-bearing mechanism that makes this work is nearest-neighbor chain training, which orders the ensemble by physical-space similarity and initializes each network from its nearest previously trained neighbor, preventing independently trained networks from landing in unrelated minima of the non-convex loss landscape. The bimodal forecast distribution identified in physical space—diagnosed with a hyperbolic-tangent surrogate shock—is replaced by a smo

What would settle it

Run the neural EnKF on a shock tube with deliberately large shock-location uncertainty, collect the trained weight ensemble, and apply a multivariate normality test to the parameter perturbations; if the weight ensemble is not approximately Gaussian, or if random re-initialization of the chain's first member changes the analysis ensemble substantially, the central claim would be falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the key discovery is that the standard EnKF's failure in shocked flows is caused by bimodal forecast distributions near uncertain shock locations. Because shock-position uncertainty makes some ensemble members place the jump to one side of a grid point and others to the other side, the forecast marginal at that point becomes two-humped, and the EnKF's linear, Gaussian-motivated update mixes those modes into oscillatory and sometimes negative density or pressure fields. The proposed neural EnKF sidesteps this by encoding each forecast member as a neural network, running the EnKF update on the network weights and biases, and reconstructing the flow from the updated ne

Load-bearing premise

The load-bearing premise is that nearest-neighbor chain training makes the ensemble of network parameters behave like a smooth, near-Gaussian cloud whose covariance is meaningful for the EnKF update—something the paper says cannot be guaranteed in a non-convex setting and supports only with visual evidence.

Editorial extensions

If this is right

  • If the neural EnKF is right, EnKF-style data assimilation can be applied to shocked compressible flows without the spurious oscillations and nonphysical states that plague the standard EnKF.
  • The diagnosis implies that any linear-in-state assimilation method will struggle near uncertain discontinuities whenever the forecast distribution is multimodal, so the fix is to find a representation in which ensemble statistics are smoother.
  • The method can recover flow structures even when the ensemble is structurally heterogeneous—members initially missing features present in the truth—as demonstrated by the Burgers' equation experiment.
  • The experiments indicate that directly observed variables (pressure) converge faster than unobserved ones (density), a property that should be expected in practice.
  • The framework does not depend specifically on neural networks: the paper notes that other function representations capturing smooth variation across ensemble members could serve the same role.

Reading between the lines

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

  • The nearest-neighbor chain is a heuristic; a more rigorous alignment criterion—for example, matching members by shock and contact-interface positions before training—could be tested and would likely strengthen or replace the current procedure.
  • The core testable assumption is that the aligned weight ensemble is approximately Gaussian; applying a multivariate normality test to the trained parameter perturbations would either validate the method as broadly applicable or reveal the conditions under which it breaks.
  • Because the neural EnKF is already a parameter-space update, the same construction may extend naturally to estimating uncertain model or initial/boundary parameters, not just states.
  • The observed density lag suggests that adding thermodynamic observations or assimilating multiple state variables jointly would improve density recovery; this is a concrete experimentable consequence of the paper's pressure-only setup.
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. This paper studies ensemble Kalman filtering (EnKF) for data assimilation in compressible shocked flows. It diagnoses the failure of the standard EnKF as a consequence of bimodal forecast distributions induced by uncertain shock locations, which violates the near-Gaussianity assumption underpinning the EnKF. To mitigate this, the authors propose a 'neural EnKF' in which each ensemble member is represented by a fully connected neural network mapping spatial coordinates to flow variables; the analysis step is performed on the network weights and biases via the standard EnKF update (Eq. (9)). To make the weight-space ensemble statistically meaningful, they introduce nearest-neighbor chain training: ensemble members are ordered by physical-space similarity and trained sequentially with transfer-learning initialization. The method is demonstrated on the inviscid Burgers equation, the Sod shock tube (with pressure-only observations), and a 2D blast wave, including a parametric study varying observation noise, density, and frequency. The authors report that the neural EnKF avoids spurious oscillations and nonphysical states that the standard EnKF produces.

Significance. The diagnosis of EnKF failure via shock-location-induced bimodality is clearly presented, and the parameter-space reformulation is an original construction that is conceptually appealing: it maps discontinuous physical fields into a smoother representation before applying the EnKF. The numerical demonstrations are internally consistent, use synthetic observations generated independently of the forecast ensemble, and include a useful parametric study (Appendix D). The principal gap is that the core mechanism—approximate Gaussianity and linearity of the weight-space ensemble after chain training—is asserted and illustrated visually but not statistically validated in the regimes where it matters most. In addition, the experiments do not compare against the closest existing shock-aware methods (normal-score EnKF [28], feature-preserving ETPF [31]), so the incremental contribution over the state of the art is not fully established. If the mechanism is validated and comparisons added, the method could be a valuable addition to the DA toolbox for shocked flows.

major comments (4)
  1. [§3.2, Eq. (9)] The neural EnKF update is an EnKF in weight space and is therefore a linear-Gaussian regression. Its validity hinges on the forecast ensemble in parameter space being approximately Gaussian and on the parameter-to-observation map being near-linear over the ensemble spread. The paper explicitly concedes that chain-training alignment 'cannot be guaranteed in a non-convex setting' (§3.2), and the only evidence is Fig. 5(c2), which is a surrogate hyperbolic-tangent case with modest shock-location uncertainty (std 0.1). In the Sod shock tube (§4.2) the diaphragm position has std 0.2 on [0,1], and in the 2D blast wave (§4.3) the center has std 0.1—regimes where, by the paper's own argument, physical-space bimodality is strongest. No parameter-space diagnostics (multivariate normality tests, PCA/spectral analysis, total variation along the chain, or linearity checks) are reported for these case
  2. [§4.2, §4.3] The numerical comparison is only against the standard EnKF (Appendix B). The manuscript itself cites the normal-score EnKF [28] and the feature-preserving ETPF [31] as methods developed for the same problem, with reported success on shock-tube and blast-wave cases, yet no quantitative comparison against either is provided. The abstract claims that the neural EnKF 'avoids the spurious oscillations and nonphysical features that plague the standard EnKF,' but the standard EnKF is a baseline, not the state of the art. To establish the contribution, the authors should add comparisons (e.g., RMSE, ensemble spread, physical-realizability violation counts) against at least one of [28] or [31] on the Sod and blast-wave test cases, ideally using the same observation configurations.
  3. [Figs. 7, 11, 16; Appendix D] All RMSE/spread traces and the parametric study are based on a single synthetic-noise realization and a single ensemble draw. The observation noise and the initial-condition sampling are stochastic, so the reported curves are single samples whose variability is unknown. Claims of 'rapid reduction' and 'sustained comparably low levels' would be more convincing with statistics over repeated trials, or at least error bars/percentile bands. This is especially important because the ensemble is small (40 or 30 members) and the observation noise realizations directly affect the analysis in Eq. (9).
  4. [§5] The concluding section lists a 'rigorous theoretical characterization of the induced ensemble geometry in the neural space' as future work. This is the crux of the method, not a peripheral issue. The limitation should be stated prominently in Section 3 where the chain-training assumption is introduced, and the main text should be explicit that the numerical results are empirical evidence for, not a validation of, the Gaussian/linear mechanism.
minor comments (5)
  1. [§1] Typo: 'keep atmopsheric models on track' should read 'keep atmospheric models on track'.
  2. [Fig. 1(c)] The caption says the PDFs are 'estimated using 10,000 forecast ensemble members,' but the ensemble described in Section 2.2 has 50 members. Clarify whether the PDFs are computed from 10,000 Monte Carlo samples drawn from the same parameter distributions or from some other procedure.
  3. [Figs. 10, 13] The phrase 'farthest ensemble member' is misleading: the member is defined as the last element in the nearest-neighbor chain, which is not necessarily the farthest from the truth in physical distance. Either verify that this member indeed has the largest structural mismatch or rename it (e.g., 'last chain member').
  4. [Appendix A] The statement that 'similar DA performance is observed with larger networks (not shown)' is not verifiable. Either show this sensitivity check or remove the claim, since the method's sensitivity to network architecture is otherwise unquantified.
  5. [§3.3, §4] The paper states that training is terminated at a prescribed MSE threshold of order 10^-6, but it does not report the actual fitting errors achieved for the forecast ensembles. Reporting the achieved reconstruction errors would support the important assumption that the neural representation is accurate enough that H(z_f) can be used in place of H(F_NN(θ_f)).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the neural EnKF is an empirical construction whose claims are evaluated against independent synthetic observations, with no fitted constant or self-citation chain doing the work.

full rationale

The paper's derivation chain is self-contained and non-circular. The claimed failure mechanism for the standard EnKF is not an output derived from the proposed method; it is demonstrated on a surrogate example where the forecast ensemble is explicitly sampled from Gaussian physical parameters (Eq. 5), and the resulting bimodality at the shock location (Fig. 1c) is a direct consequence of that sampling, not a quantity the paper later 'predicts'. The neural EnKF is a construction: it maps each forecast member to NN weights, performs the standard EnKF regression in weight space (Eq. 9), and reconstructs physical states (Eq. 10). The reported error reductions (RMSE/spread, Figs. 7, 11, 16) are genuine outcomes of applying this procedure to synthetic observations generated independently from the reference solution and the numerical solver (Eqs. 13, 23, 27), not quantities fitted into the method. The central enabling assumption — that chain training produces a smooth, approximately Gaussian parameter ensemble — is explicitly flagged by the authors as unguaranteed in non-convex settings (Sec. 3.2: 'this behavior cannot be guaranteed in a non-convex setting') and is listed as future work (Sec. 5: 'a rigorous theoretical characterization of the induced ensemble geometry in the neural space remains to be developed'). This is an honest limitation on the strength of the empirical claim, not a circular step. The self-citations (e.g., refs. [15], [18-23], [33], [34]) are used for background, standard subspace/Gaussian-approximation results, or prior applications of EnKF; none replaces an argument or supplies the central claim. No equation in the paper is equal to its input by construction, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No physical constants are fitted and no new physical entities are introduced. The method depends on hand-set network architectures, training thresholds, and the unproven alignment of weights by chain-based transfer learning.

free parameters (3)
  • Neural-network architecture per test case = 4x50 (Burgers), 4x64 (Sod), 6x100 (blast); Table A.1
    Hand-chosen; the paper says larger networks behave similarly but does not show this.
  • Training termination MSE threshold = ~1e-6
    Prescribed in Sec. 3.3; affects weight alignment and computational cost.
  • Transfer-learning 'reduced learning rate' = not reported
    Sec. 3.3 states chain members are trained with a reduced learning rate, but the value and scheduler are omitted.
assumptions (5)
  • standard math The ensemble Kalman update (Eq. 3) correctly estimates the analysis when forecast and observation distributions are approximately Gaussian.
    Inherited from standard EnKF theory; used without proof in Sec. 2.1 and transferred to parameter space in Sec. 3.1.
  • domain assumption Bimodality of the physical-space forecast distribution near an uncertain shock is the cause of standard EnKF failure.
    Established only through the hyperbolic-tangent diagnostic in Sec. 2.2; plausible but not proven for general shocked flows.
  • domain assumption Neural-network fitting errors are negligible at the assimilation level, so physical-space predicted observations can be used in the parameter-space update.
    Stated in Sec. 3.1: 'given that the neural parameterization is sufficiently accurate and the fitting errors are negligible.'
  • ad hoc to paper Nearest-neighbor chain training produces smooth parameter variation and an approximately Gaussian parameter ensemble.
    Core enabler of the method; the paper itself says in Sec. 3.2 this 'cannot be guaranteed in a non-convex setting,' and no Gaussianity test is provided.
  • domain assumption Reconstructed analysis states are suitable initial conditions for forward integration by the compressible solver.
    The analysis states are decoded network outputs; positivity of rho and p is enforced by softplus, but discrete conservation and thermodynamic realizability are not analyzed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural ensemble Kalman filter: Data assimilation for compressible flows with shocks." pith.science (2026). https://pith.science/paper/7FY3VU25

@misc{pith2026260223461,
  author       = {Pith},
  title        = {Pith review of: Neural ensemble Kalman filter: Data assimilation for compressible flows with shocks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7FY3VU25}},
  note         = {Machine review of arXiv:2602.23461}
}
read the original abstract

Data assimilation (DA) for compressible flows with shocks is challenging because many classical DA methods generate spurious oscillations and nonphysical features near uncertain shocks. We focus here on the ensemble Kalman filter (EnKF). We show that the poor performance of the EnKF may be attributed to the bimodal forecast distribution that can arise in the vicinity of an uncertain shock location; this violates the assumptions underpinning the EnKF, which assume a forecast which is close to Gaussian. To address this issue we introduce the new neural EnKF. The basic idea is to systematically embed neural function approximations within ensemble DA by mapping the forecast ensemble of shocked flows to the parameter space (weights and biases) of a deep neural network (NN) and to subsequently perform DA in that space. The nonlinear mapping encodes sharp and smooth flow features in an ensemble of NN parameters. Neural EnKF updates are therefore well-behaved only if the NN parameters vary smoothly within the neural representation of the forecast ensemble. We show that such a smooth variation of network parameters can be enforced via physics-informed transfer learning, and demonstrate that in so-doing the neural EnKF avoids the spurious oscillations and nonphysical features that plague the EnKF. The applicability of the neural EnKF is demonstrated through a series of systematic numerical experiments with the inviscid Burgers' equation, the Sod shock tube, and a two-dimensional blast wave.

Figures

Figures reproduced from arXiv: 2602.23461 by the authors.

Figure 1
Figure 1. Behavior of the standard EnKF for a shock-like transition, illustrated using a hyperbolic-tangent [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Schematic of the neural EnKF framework: at each DA cycle, forward simulations generate a [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Schematic illustration of neural-network training across ensemble members on a non-convex loss [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Schematic illustration of the nearest-neighbor chain construction: (a) ensemble members in the [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Comparison of neural EnKF updates under two training strategies for the hyperbolic-tangent [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: DA results for the inviscid Burgers’ equation with a structurally heterogeneous initial ensemble [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Evolution of RMSE and ensemble spread over DA cycles for the inviscid Burgers’ equation with [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Schematic of the reference shock tube configuration. The diaphragm at [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: The large initial uncertainty is progressively reduced through sequential DA using [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 9
Figure 9. Figure 9: Analysis ensemble for the shock tube problem at representative DA times, showing increasing [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Comparison of the space–time evolution under DA for the shock tube problem, with neural [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: RMSE and ensemble spread over successive DA cycles for the shock tube problem under the [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Schematic illustration of the 2D blast-wave initial configuration and ensemble sampling. (a) [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Pressure fields for the 2D blast wave over the full assimilation window, demonstrating rapid [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Velocity fields (x-component) for the 2D blast wave over the full assimilation window, demonstrating effective correction after the first neural EnKF update and sustained agreement with the reference solution thereafter. Rows show the reference solution, the ensemble …
Figure 15
Figure 15. Figure 15: Density fields for the 2D blast wave over the full assimilation window, demonstrating effective [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Evolution of RMSE and ensemble spread over successive DA cycles for the 2D blast wave. [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Feature-preserving Latent-EnKF for Data Assimilation of Flows with Shocks

    physics.comp-ph 2026-06 unverdicted novelty 7.0 of 10

    A feature-preserving latent-EnKF performs ensemble updates in a learned latent space to recover shocks and discontinuities in compressible flows from sparse noisy observations without spurious oscillations.

  2. Multifidelity-Augmented Gaussian Process Inputs for Surrogate Modeling from Scarce Data

    stat.ML 2026-03 conditional novelty 6.0 of 10

    Augmenting a high-fidelity GP's inputs with predictions from all low-fidelity surrogates improves accuracy and cuts cost versus cokriging and autoregressive multifidelity GPs on scarce-data problems.

Reference graph

Works this paper leans on

41 extracted references · 1 linked inside Pith · cited by 2 Pith papers

  1. [28]

    J. J. Hansen, D. Brouzet, M. Ihme, A normal-score ensemble Kalman filter for 1D shock waves, in: AIAA SCITECH 2024 Forum, 2024, p. 1022. 32

  2. [31]

    A. N. Subrahmanya, A. Sandu, Feature preserving data assimilation via feature align- ment, Computer Methods in Applied Mechanics and Engineering 447 (2025) 118345

  3. [1]

    Raman, S

    V. Raman, S. Prakash, M. Gamba, Nonidealities in rotating detonation engines, Annual Review of Fluid Mechanics 55 (2023) 639–674

  4. [2]

    J. D. Anderson, Hypersonic and high temperature gas dynamics, AIAA, 1989

  5. [3]

    Schuabb, L

    M. Schuabb, L. Duan, K. M. Casper, R. M. Wagnild, M. M. Choudhari, P. Paredes, Hypersonic boundary-layer transition over a circular cone in a Mach 8 digital wind tunnel, Journal of Fluid Mechanics 1017 (2025) A33

  6. [4]

    S. Cao, G. Wang, O. Coutier-Delgosha, K. Wang, Shock-induced bubble collapse near solid materials: Effect of acoustic impedance, Journal of Fluid Mechanics 907 (2021) A17

  7. [5]

    X. Zhao, W. Ma, J. Chen, G. Xiang, P. Zhong, K. Wang, Vapour bubbles produced by long-pulsed laser: a race between advection and phase transition, Journal of Fluid Mechanics 999 (2024) A103

  8. [6]

    S. Cao, Y. Zhang, D. Liao, P. Zhong, K. G. Wang, Shock-induced damage and dynamic fracture in cylindrical bodies submerged in liquid, International journal of solids and structures 169 (2019) 55–71

Show all 41 references
  1. [7]

    W. Ma, X. Zhao, C. Gilbert, K. Wang, Computational analysis of bubble–structure interactions in near-field underwater explosion, International Journal of Solids and Structures 242 (2022) 111527

  2. [8]

    Narkhede, S

    A. Narkhede, S. Islam, X. Sun, K. Wang, Fluid–structure coupled simulation framework for lightweight explosion containment structures under large deformations, International Journal of Impact Engineering 199 (2025) 105238

  3. [9]

    Evensen, Data assimilation: the ensemble Kalman filter, Springer, 2009

    G. Evensen, Data assimilation: the ensemble Kalman filter, Springer, 2009

  4. [10]

    M. Asch, M. Bocquet, M. Nodet, Data assimilation: methods, algorithms, and applica- tions, SIAM, 2016

  5. [11]

    Evensen, F

    G. Evensen, F. C. Vossepoel, P. J. Van Leeuwen, Data assimilation fundamentals: A unified formulation of the state and parameter estimation problem, Springer Nature, 2022

  6. [12]

    Kalnay, Atmospheric modeling, data assimilation and predictability, Cambridge uni- versity press, 2003

    E. Kalnay, Atmospheric modeling, data assimilation and predictability, Cambridge uni- versity press, 2003

  7. [13]

    Carrassi, M

    A. Carrassi, M. Bocquet, L. Bertino, G. Evensen, Data assimilation in the geosciences: An overview of methods, issues, and perspectives, Wiley Interdisciplinary Reviews: Climate Change 9 (2018) e535. 31

  8. [14]

    Zhang, H

    X.-L. Zhang, H. Xiao, X. Luo, G. He, Ensemble Kalman method for learning turbulence models from indirect observation data, Journal of Fluid Mechanics 949 (2022) A26

  9. [15]

    X.-H. Zhou, H. Wang, J. McClure, C. Chen, H. Xiao, Inference of relative permeability curvesinreservoirrockswithensembleKalmanmethod, TheEuropeanPhysicalJournal E 46 (2023) 44

  10. [16]

    M. I. Zafar, X. Zhou, C. J. Roy, D. Stelter, H. Xiao, Data-driven turbulence modeling approach for cold-wall hypersonic boundary layers, Journal of Thermophysics and Heat Transfer (2025) 1–17

  11. [17]

    Z.-R.Liu, H.-C.Wang, Z.-L.Zhao, H.Xiao, Towardsaunifiedturbulencemodelthrough multi-objective learning, arXiv preprint arXiv:2509.17189 (2025)

  12. [18]

    M. Wang, T. A. Zaki, State estimation in turbulent channel flow from limited observa- tions, Journal of Fluid Mechanics 917 (2021) A9

  13. [19]

    T. A. Zaki, Turbulence from an observer perspective, Annual Review of Fluid Mechanics 57 (2025) 311–334

  14. [20]

    D. A. Buchta, T. A. Zaki, Observation-infused simulations of high-speed boundary-layer transition, Journal of Fluid Mechanics 916 (2021) A44

  15. [21]

    D. A. Buchta, S. J. Laurence, T. A. Zaki, Assimilation of wall-pressure measurements in high-speed flow over a cone, Journal of Fluid Mechanics 947 (2022) R2

  16. [22]

    Morra, C

    P. Morra, C. Meneveau, T. A. Zaki, ML for fast assimilation of wall-pressure measure- ments from hypersonic flow over a cone, Scientific Reports 14 (2024) 12853

  17. [23]

    Q. Wang, T. A. Zaki, Domain of dependence for wall-pressure measurements in high- speed boundary layers, Journal of Fluid Mechanics 1009 (2025) A67

  18. [24]

    E. J. West, J. F. MacArt, R. Munipalli, Variational data assimilation in shock tube flows, in: AIAA SCITECH 2025 Forum, 2025, p. 1166

  19. [25]

    Houba, A

    T. Houba, A. Edoh, R. Munipalli, M. E. Harvazinski, Sequential data assimilation in flows with shocks, in: AIAA SCITECH 2024 Forum, 2024, p. 0587

  20. [26]

    A. Edoh, T. Houba, R. Munipalli, M. E. Harvazinski, Sequential ensemble Kalman filtering of compressible flows with shocks: Enforcing positivity, in: AIAA SCITECH 2025 Forum, 2025, p. 0918

  21. [27]

    A. K. Edoh, E. J. West, T. Houba, R. Munipalli, M. E. Harvazinski, W. Kang, Data assimilation of compressible flows with discontinuities: Evaluating algorithms on Sod’s shock tube, International Journal for Numerical Methods in Fluids (2026)

  22. [29]

    H.Zhou, J.J.Gomez-Hernandez, H.-J.H.Franssen, L.Li, Anapproachtohandlingnon- gaussianity of parameters and state variables in ensemble Kalman filtering, Advances in water Resources 34 (2011) 844–864

  23. [30]

    T. Li, A. Gelb, Y. Lee, A structurally informed data assimilation approach for nonlinear partial differential equations, Journal of Computational Physics 519 (2024) 113396

  24. [32]

    Srivastava, W

    A. Srivastava, W. Kang, D. M. Tartakovsky, Feature-informed data assimilation, Jour- nal of Computational Physics 494 (2023) 112499

  25. [33]

    M.A.Iglesias, K.J.Law, A.M.Stuart, EnsembleKalmanmethodsforinverseproblems, Inverse Problems 29 (2013) 045001

  26. [34]

    Morzfeld, D

    M. Morzfeld, D. Hodyss, Gaussian approximations in filters and smoothers for data assimilation, Tellus A: Dynamic Meteorology and Oceanography 71 (2019) 1600344

  27. [35]

    D. I. Ketcheson, K. T. Mandli, A. J. Ahmadia, A. Alghamdi, M. Quezada de Luna, M. Parsani, M. G. Knepley, M. Emmett, PyClaw: Accessible, Extensible, Scalable Tools for Wave Propagation Problems, SIAM Journal on Scientific Computing 34 (2012) C210–C231

  28. [36]

    G. A. Sod, A survey of several finite difference methods for systems of nonlinear hyper- bolic conservation laws, Journal of Computational Physics 27 (1978) 1–31

  29. [37]

    X. Zhao, W. Ma, S. Islam, A. Narkhede, K. Wang, M2C: An open-source software for multiphysics simulation of compressible multi-material flows and fluid–structure inter- actions, Computer Physics Communications (2026) 110023

  30. [38]

    Wang, M2C: Multiphysics Modeling and Computation,https://github.com/ kevinwgy/m2c, 2026

    K. Wang, M2C: Multiphysics Modeling and Computation,https://github.com/ kevinwgy/m2c, 2026. Accessed: 2026-01-25

  31. [39]

    E. Ott, B. R. Hunt, I. Szunyogh, A. V. Zimin, E. J. Kostelich, M. Corazza, E. Kalnay, D. Patil, J. A. Yorke, A local ensemble Kalman filter for atmospheric data assimilation, Tellus A: Dynamic Meteorology and Oceanography 56 (2004) 415–428

  32. [40]

    B. R. Hunt, E. J. Kostelich, I. Szunyogh, Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter, Physica D: Nonlinear Phenomena 230 (2007) 112–126

  33. [41]

    X. T. Tong, M. Morzfeld, Localized ensemble Kalman inversion, Inverse Problems 39 (2023) 064002. 33

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.