Pith. sign in

REVIEW 4 major objections 5 minor 40 references

Data assimilation in 2D nonlinear coupled sound and heat flow, using a stabilized explicit finite difference scheme marched backward in time

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A stabilized explicit finite difference scheme marched backward in time can recover initial states in a nonlinear coupled sound-and-heat system, but only within a short time window.

desk verdict A useful illustrative application of a previously developed backward-marching scheme to a nonlinear coupled sound-heat system, but the nonlinear experiments are too weak to separate the stabilizing filter from genuine data assimilation. read the letter →

arxiv 2501.14895 v1 pith:B2C22YLP submitted 2025-01-24 math.NA cs.NA

Alfred S. Carasso This is my paper · ORCID
classification math.NAcs.NA MSC 35L1535K1535R2565N1265N21
keywords dataassimilationbackwardintimeill-posedproblemstabilizedexplicitschemecoupledsoundandheatflowFFTLaplaciansmoothingnonlinearPDEimage
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

This paper tries to establish that data assimilation for a nonlinear 2D coupled sound-and-heat system can be performed without iteration, by marching a stabilized explicit finite difference scheme backward from hypothetical final-time data that need not correspond to an actual solution. The data are 512×512 pixel images, deliberately non-smooth, and the computation applies an FFT-based Laplacian smoother at every backward step. The paper derives a closed-form error bound, Eq. (2.18), showing that a backward-then-forward round trip succeeds only when $\delta(1+K_1^2)+K_5(1+K_1)$ is small, with $K_1=e^{\zeta_J T_{max}}$. In the demonstrated successful case at $T_{max}=1.6\times 10^{-4}$, the evolved images match the desired images with $L^1$ relative errors between 2.69% and 7.46%; at $T_{max}=8.0\times 10^{-4}$, five times larger, the same method fails with errors of 18% to 49%. A sympathetic reader should care because this offers a direct, non-iterative route to initial-value reconstruction in ill-posed dissipative systems, potentially seeding or cross-checking iterative and machine-learning assimilation methods.

What carries the argument

The carrying mechanism is the stabilized explicit marching operator $R = S + \Delta t\, S G$, where $G$ is the 3×3 spatial differential matrix for the coupled system (temperature $u$, wave velocity $v$, wave displacement $w$), and $S$ is a per-step smoothing operator $S = \exp(-\omega|\Delta t|\Lambda^p)$ built from $\Lambda=\rho(I+L)$. For the nonlinear and non-rectangular experiments, the smoothing is done with an FFT-synthesized Laplacian operator $Q_\Delta = \exp(-\epsilon|\Delta t|\Gamma^q)$ after zero-extension of the solution to an enclosing square. Marching backward from $T_{max}$, the nonlinearity is lagged at the previous time step, and the error analysis transfers from the linear selfadjoint case through Lemmas 1–3 and Theorems 1–2, with the key quantity $K_1=e^{\zeta_J T_{max}}$ controlling the exponential growth of initial-data errors. The paper's threshold statement is that assimilation succeeds only when the inexact computed initial values make the right-hand side of Eq. (2.18) sufficiently small.

What would settle it

Run the same nonlinear system at $T_{max}=1.6\times 10^{-4}$ but start from exact final-time data generated by a known smooth forward solution and set the hypothetical data to be exactly that solution; if the recovered initial state does not reproduce the known initial data to within truncation-level error, then the stabilization parameters, not the data discrepancy $\delta$, are the limiting factor. Separately, check whether the chosen $(\epsilon,q)=(8.0\times 10^{-11},3.35)$ satisfies $\|Q_\Delta g\|_2 \le \|Q g\|_2$ on the quarter-circle domain for the actual computed solution; if the bound fails, the error analysis of Theorems 1–2 does not apply to the nonlinear experiment.

Watch

Extended reading notes

Core claim

The paper's central claim is that the success or failure of backward-time data assimilation is governed by the size of the right-hand side in Eq. (2.18), $\delta(1+K_1^2)+K_5(1+K_1)$, where $K_1=e^{\zeta_J T_{max}}$ is the exponential error-amplification factor and $K_5$ collects stabilization and truncation terms. If the computed initial values $U(\cdot,0)$ round-trip through the forward problem to a final state sufficiently close to the desired hypothetical data $W^*(\cdot,T_{max})$, assimilation is successful; otherwise it is not. The paper demonstrates this in a nonlinear quarter-circular domain, with an exponential nonlinearity $s(z)=\exp(0.005z)$, advection, and three 512×512 images as hypothetical final data: at $T_{max}=1.6\times 10^{-4}$ the round trip is successful, while at $T_{max}=8.0\times 10^{-4}$ it fails. The failure is less catastrophic than the linear theory predicts, but the linear analysis is still a useful guide for parameter selection.

Load-bearing premise

The nonlinear operator $L$, with exponential nonlinearity $s(z)=\exp(0.005z)$ and advection term, is assumed to stay in a regime where the linear selfadjoint error analysis of Section 2 still guides the hand-tuned stabilization parameters ($p=3.35$, $\zeta_J=19800$, $\omega=8\times 10^{-11}$) and where the FFT Laplacian smoothing bound $\|Q_\Delta g\| \le \|Q g\|$ from [31, Eq. (6.2)] holds; the paper itself notes that the nonlinear experiment lies outside the scope of its linear theory.

Editorial extensions

If this is right

  • For time windows with $\zeta_J T_{max}$ not too large (here $K_1 < 23.8$), stabilized backward marching provides a direct, non-iterative data assimilation method for nonlinear sound–heat systems, with round-trip $L^1$ relative errors below about 8%.
  • Increasing $T_{max}$ by a factor of five changes $K_1$ from below 23.8 to above $7.5\times 10^6$, flipping the method from success to failure; this quantifies why backward recovery in dissipative systems is inherently limited.
  • Enclosing a non-rectangular domain in a square permits FFT-based Laplacian smoothing and makes 512×512 image-scale computations feasible.
  • The computed initial values can be used as starting points for iterative or machine-learning assimilation procedures, and as a check on unexpected results from such procedures.
  • The failure example is as informative as the success example: it shows the method's limits and the value of exhibiting both outcomes.

Reading between the lines

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

  • The paper's success criterion is essentially a round-trip test: since $\delta$ is unknown in practice, the method's usefulness is judged by whether the evolved images match the desired images; a rigorous a priori bound for $\delta$ would be needed to predict success before computing.
  • The parameter pair $(\epsilon,q)=(8.0\times 10^{-11},3.35)$ is found by interactive trial; an automatic rule for choosing $(\epsilon,q)$ from estimates of $|||PW|||$ or the data's smoothness would make the method more portable.
  • Because the failed experiment also changes the target data (a resolution chart replaces one of the test images), the comparison conflates a fivefold larger $T_{max}$ with different image content; a controlled test keeping identical images at both $T_{max}$ values would isolate the time-window effect.
  • If the linear theory's qualitative prediction survives in harder nonlinear regimes, the same stabilized backward-marching template may apply to other coupled hyperbolic-parabolic systems, including geophysical model problems where direct solvers could validate machine-learning assimilation.
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 paper develops a stabilized explicit finite difference scheme for solving an ill-posed backward-in-time data assimilation problem for a 2D nonlinear coupled sound and heat flow system. The scheme applies a compensating smoothing operator at every time step to quench instability; the analysis is carried out for a linear selfadjoint setting in Section 2, while the nonlinear problem of Section 3 is treated by lagging the nonlinearity and using FFT-based Laplacian smoothing. The central computational claims are based on two numerical experiments with 512x512 images: a 'successful' assimilation from Tmax=1.6e-4 (Table 2) and an 'unsuccessful' one from Tmax=8.0e-4 (Table 3). The paper explicitly acknowledges that the nonlinear experiments lie outside the scope of the linear theory, and it interprets the results using error bounds from the linear analysis (Eq. (2.18)).

Significance. If the central claim were established, the paper would offer a direct, non-iterative alternative to iterative data assimilation methods for a challenging nonlinear hyperbolic/parabolic system, with potential value as an initialization tool for geophysical or machine-learning pipelines. The paper is clearly written, the linear error analysis is quoted from prior work and is internally consistent, and the algorithm description is complete enough to reproduce the experiments. However, the significance is currently undermined by a validation gap: the nonlinear experiments only demonstrate round-trip consistency through the same stabilized scheme, with no ground-truth initial state and no independent forward solver, so the reported successes may reflect the smoothing operator rather than genuine data assimilation.

major comments (4)
  1. [Section 3, Figure 2, Tables 2–3] The central experimental claim rests on a round-trip consistency check: the backward scheme (Eq. (2.7)) produces U(·,0), and the forward leg is marched with the same stabilized scheme, after which the evolved U(·,Tmax) is compared to the desired data. No ground-truth initial state is available, and no independent, stabilization-free forward solver is used. Consequently, the experiment cannot distinguish genuine backward recovery from the effect of the repeated smoothing operator Q_delta applied in both legs; the success/failure pattern in Tables 2 and 3 is also consistent with cumulative Q_delta^N blur increasing with N. A definitive test would use a synthetic case with a known initial condition and a stabilization-free reference forward solution, or a run with G=0 to quantify the smoothing-only effect.
  2. [Section 3, Eq. (3.2), and Table 2] The physical evolution over the reported assimilation window is sub-grid. With a=6, b=5, d=0.95 and L defined in Eq. (3.2), the diffusion coefficient is about 0.001 and the advection speed about 2.75, so over Tmax=1.6e-4 the true solution changes by distances of order 4e-4, smaller than the grid spacing 1/512 ≈ 2e-3. Thus the computed U(·,0) would be nearly identical to the target images even if the scheme only copied the data backward without solving the PDE. The paper should report the actual change produced by the forward operator, or compare with a case where the evolution is resolved on the grid (e.g., larger Tmax or coarser spatial grid).
  3. [Section 2.2 and Figure 2 caption] The nonlinear experiment is explicitly stated to lie outside the scope of the linear theory of Section 2 (Figure 2 caption), yet the stabilization parameters (p, ζ_J, ω) = (3.35, 19800, 8e-11) are taken from Table 1, and the interpretation of success and failure uses Eq. (2.18) and the constants K1–K5 from linear selfadjoint theory. The transfer requires, at minimum, that the bound ||Q_delta g|| ≤ ||Q g|| from [31, Eq. (6.2)] continues to hold for the nonlinear operator L of Eq. (3.2) and for the chosen (ε,q) = (8e-11, 3.35). No justification or numerical verification of this bound for the nonlinear case is provided. The authors should either verify this inequality numerically or restrict the claims of parameter transfer.
  4. [Section 3, Tables 2–3] The failure experiment at Tmax=8e-4 is not a controlled comparison with the success experiment: in addition to changing Tmax fivefold, the third image is changed from Elizabeth Taylor to the USAF 1951 resolution chart (Section 3). The observed degradation in Tables 2 and 3 is therefore not attributable solely to the increased Tmax and the predicted factor 5.8e13 in Eq. (2.18). A controlled experiment keeping all three images identical while varying Tmax would isolate the effect of the time horizon.
minor comments (5)
  1. [References] Reference [2] spells 'backward' as 'backwrd'.
  2. [After Eq. (2.16)] In the sentence preceding Eq. (2.16), 'negligibe' should be 'negligible'.
  3. [Figure 2 caption] The caption contains a duplicated word: 'As explained in in the discussion'.
  4. [Reference [37]] Reference [37] has a typo: 'continuum echanics' should be 'continuum mechanics'.
  5. [Theorem 2] Theorem 2 refers to 'Lemma 6' when stating the conditions on p, ζ_J, ω; within this paper the relevant lemma is Lemma 1 (or the citation should be given as in [31]).

Circularity Check

2 steps flagged · score 6.0 of 10

Empirical 'success' is a tuned forward-backward round-trip through the same stabilized scheme, with no ground-truth or independent forward solver; the nonlinear validation reduces to an in-sample consistency check.

  1. fitted input called prediction [Section 3, Eq. (3.4), Figure 2 discussion and Table 2]
    "Interactive trials are needed to locate appropriate values for (ε, q). Here, a parameter pair ε = 8.0 × 10−11, q = 3.35, was arrived at after very few trials. ... The actual computed data at t = 0 involve negative values. These values are not used in forming and displaying the middle column images, but are nevertheless retained as necessary to enable computation of the images in the rightmost column in Figure 2. ... At time Tmax, the evolved L1 norms closely match the desired L1 norms, and the resulting L1 relative errors are small."

    The stabilization parameters (ε,q) and the very small Tmax are hand-tuned for this experiment, and the reported success metric is the round-trip error of the same stabilized operator: the backward march produces U(·,0), and the rightmost-column images are obtained by marching those same values forward with the same S_Δ(I+ΔtG) scheme. Because no ground-truth initial condition and no independent, stabilization-free forward solution of the nonlinear PDE are compared, the small L1 errors measure the identity defect of [S_Δ(I−|Δt|G)]^N[S_Δ(I+|Δt|G)]^N applied to the desired images, not agreement with an actual solution.

  2. other [Section 2.1, Eq. (2.18) and Section 3 interpretation of Figures 2-3]
    "Therefore, data assimilation is successful only if the inexact computed initial values U(·,0) at t=0, lead to a sufficiently small right hand side in Eq.(2.18). ... While the failure in the above nonlinear experiment with Tmax = 8.0×10−4 is less catastrophic than was predicted in the linear selfadjoint case with the same Tmax value, the linear analysis in Table 1 and Eq.(2.18), was a useful guide."

    The stated success criterion is literally the smallness of the bound’s right-hand side, built from K1=e^{ζ_J Tmax}, K5, and the same stabilization parameters used in the nonlinear run. The nonlinear experiment is admitted in the Figure 2 caption to lie outside the scope of the linear selfadjoint theory that produced Eq. (2.18), yet the success/failure verdict is imported from that bound. Thus the interpretation is not an independent test of the nonlinear scheme; it applies a linear-selfadjoint tautology to a regime where the theorem’s assumptions are not verified.

full rationale

The linear selfadjoint analysis in Section 2 is not itself circular: it cites the author’s prior work [31] for proofs of Lemmas 1-3 and Theorems 1-2, and those cited statements are parameter-free results for linear operators, so the mathematical bound in Eq. (2.18) is a legitimate (if imported) derivation. The circularity arises when this theory is used to validate the nonlinear experiments. The only empirical evidence for the central claim is a backward-then-forward round trip computed with the same stabilized explicit scheme, using interactively fitted parameters (ε,q)=(8e-11,3.35) and an extremely short time window Tmax=1.6e-4. In that regime the physical evolution is below grid scale, and the forward and backward stabilized operators are near-inverses on the resolved modes, so the reported L1 errors mostly reflect the round-trip identity defect of the smoothed scheme rather than successful recovery of an unknown initial state. The failed case at Tmax=8.0e-4 is consistent with cumulative smoothing, but it does not rescue the success claim as an independent validation. Under the review rules, this is partial circularity: the empirical 'prediction' reduces to an in-sample, parameter-fitted consistency check, while the supporting linear theory is self-cited and explicitly not applicable to the nonlinear experiment. Score 6 reflects that the central claim is substantially compromised, though not fully forced by definition, because the round-trip errors are nontrivial computed quantities and a failure case is reported.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a small number of hand-chosen stabilization parameters and on the transfer of linear selfadjoint error analysis to a nonlinear, non-rectangular problem. The paper introduces no new physical entities. The key unproven element is that the nonlinear operator remains within the regime where the linear bounds and the Laplacian smoothing domination hold.

free parameters (6)
  • p (smoothing exponent) = 3.35
    Hand-selected exponent in the stabilizing operator Q = exp(-omega|dt|Lambda^p); appears in Table 1 and in the nonlinear experiments. No fitting procedure is used.
  • zeta_J (eigenvalue bound) = 19800
    Hand-selected bound controlling the exponential growth factor e^{zeta_J Tmax}; chosen so that zeta_J Tmax is moderate while omega = zeta_J^{1-p} is small.
  • omega = 8e-11
    Determined by the equality omega = zeta_J^{1-p} = 19800^{-2.35}; effectively chosen once p and zeta_J are picked, but load-bearing for the smoothing strength.
  • epsilon (Laplacian smoothing amplitude) = 8e-11
    Tuned interactively in the nonlinear experiments ('arrived at after very few trials', Section 3). Critically controls the amount of smoothing per step.
  • q (Laplacian smoothing exponent) = 3.35
    Also tuned interactively in Section 3; equals p in this run, but is an independent choice for the FFT Laplacian smoother Q_delta.
  • Tmax = 1.6e-4
    Chosen in the successful regime; the paper shows Tmax five times larger fails. Tmax is a physical problem parameter, but in this test it is selected so that the error bound in Eq. (2.18) predicts success.
assumptions (4)
  • standard math Spectral theory of selfadjoint elliptic operators: L has a complete orthonormal eigenfunction basis and eigenvalues lambda_m → infinity, enabling the modal analysis in Section 2.
    Used implicitly in Eqs. (2.1)-(2.5) and in the lemmas quoted from [31].
  • domain assumption The lemmas and theorems from [31] (Lemmas 1-3, Theorems 1-2, and Eq. (6.2)) are correct and apply to the linear system in Eq. (2.2).
    The paper builds its entire convergence analysis on results quoted from reference [31] without reproducing the proofs.
  • ad hoc to paper For the nonlinear operator L in Eq. (3.2), the same parameter choices and the Laplacian smoothing bound ||Q_delta g|| <= ||Q g|| from [31, Eq. (6.2)] remain valid.
    The paper explicitly says the nonlinear experiment lies outside the scope of the linear theory, yet transfers the stabilization parameters and the Laplacian-domination assumption to it.
  • domain assumption The given hypothetical data at time T may differ from an actual solution by an unknown delta; the success metric is the round-trip error in L1 norm, not agreement with a known ground-truth initial state.
    This setup is stated in the Introduction and Section 2.1; the absence of ground truth makes the success criterion a consistency check.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data assimilation in 2D nonlinear coupled sound and heat flow, using a stabilized explicit finite difference scheme marched backward in time." pith.science (2026). https://pith.science/paper/B2C22YLP

@misc{pith2026250114895,
  author       = {Pith},
  title        = {Pith review of: Data assimilation in 2D nonlinear coupled sound and heat flow, using a stabilized explicit finite difference scheme marched backward in time},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B2C22YLP}},
  note         = {Machine review of arXiv:2501.14895}
}
read the original abstract

This paper considers the ill-posed data assimilation problem associated with hyperbolic/parabolic systems describing 2D coupled sound and heat flow. Given hypothetical data at time T > 0, that may not correspond to an actual solution of the dissipative system at time T, initial data at time t = 0 are sought that can evolve, through the dissipative system, into a useful approximation to the desired data at time T. That may not always be possible. A stabilized explicit finite difference scheme, marching backward in time, is developed and applied to nonlinear examples in non rectangular regions. Stabilization is achieved by applying a compensating smoothing operator at each time step, to quench the instability. Analysis of convergence is restricted to the transparent case of linear, autonomous, selfadjoint spatial differential operators. However, the actual computational scheme can be applied to more general problems. Data assimilation is illustrated using 512x512 pixel images. Such images are associated with highly irregular non smooth intensity data that severely challenge ill-posed reconstruction procedures. Successful and unsuccessful examples are presented.

Figures

Figures reproduced from arXiv: 2501.14895 by the authors.

Figure 1
Figure 1. FIGURE 1. Non smooth intensity data plot, associated with Ab [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2. Successful data assimilation experiment. See sum [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIGURE 3. Failure of data assimilation with significantly la [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 2
Figure 2. Figure 2: The actual computed data at t = 0 involve negative values. These values are not used in forming and displaying the middle column images, but are nevertheless retained as necessary to enable computation of the images in the rightmost column in [PITH_FULL_IMAGE:figures/…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 35 canonical work pages

  1. [2]

    Data assimilation in 2D nonlinear advection diffusion equations, using an explicit 10 stabilized leapfrog scheme run backwrd in time

    Carasso AS. Data assimilation in 2D nonlinear advection diffusion equations, using an explicit 10 stabilized leapfrog scheme run backwrd in time. NIST Techni cal Note 2227, July 12 2022. DOI:10.6028/NIST.TN2227

    work page doi:10.6028/nist.tn2227 2022

  2. [3]

    Data assimilation in 2D hyperbolic/parabol ic systems using a stabilized explicit finite difference scheme run backward in time

    Carasso AS. Data assimilation in 2D hyperbolic/parabol ic systems using a stabilized explicit finite difference scheme run backward in time. Applied Mathem atics in Science and Engi- neering, 2024;32:1,228641. DOI:10.1080/27690911.2023. 228641

    arXiv 2024

  3. [4]

    Data assimilation in 2D incompressible Navi er-Stokes equations, using a stabilized explicit O(∆ t)2 leapfrog finite difference scheme run backward in time

    Carasso AS. Data assimilation in 2D incompressible Navi er-Stokes equations, using a stabilized explicit O(∆ t)2 leapfrog finite difference scheme run backward in time. NIST T echnical Note TN2299, August 2024. DOI:10.6028/NIST.TN2299

    work page doi:10.6028/nist.tn2299 2024

  4. [5]

    Difference Methods for Initial V alue Problems

    Richtmyer RD, Morton KW. Difference Methods for Initial V alue Problems. 2nd ed. New York (NY): Wiley; 1967

    work page 1967

  5. [6]

    M´ ethode de Quasi-R´ eversibilit´ e et Applications [The Method of Quasi- Reversibility and Applications]

    Latt` es R, Lions JL. M´ ethode de Quasi-R´ eversibilit´ e et Applications [The Method of Quasi- Reversibility and Applications]. Paris: Dunod; 1967

    work page 1967

  6. [7]

    Sur l’approximation de la solution d’´ equatio ns d’´ evolution coupl´ ees

    Lions JL. Sur l’approximation de la solution d’´ equatio ns d’´ evolution coupl´ ees. Rend. Mat. 1968;1:141–176

    work page 1968

  7. [8]

    Coupled sound and heat flow and the method of le ast squares

    Carasso AS. Coupled sound and heat flow and the method of le ast squares. Math. Comp. 1975;29:447–463

    work page 1975

  8. [9]

    Tracking the model: data assimilation by artifi- cial neural network

    Cintra R, de Campos Velho HF, Cocke S. Tracking the model: data assimilation by artifi- cial neural network. 2016 International Joint Conference o n Neural Networks (IJCNN), Vancouver, BC, Canada. 2016:403-410. DOI:10.1109/IJCNN. 2016.7727227

    arXiv 2016

Show all 40 references
  1. [10]

    A machine learning a ugmented data assimilation method for high resolution observations

    Howard LJ, Subramanian A, Hoteit I. A machine learning a ugmented data assimilation method for high resolution observations. Journal of Advanc es in Modeling Earth Systems. 2024;16,e2023MS003774. https://doi.org/10.1029/2023MS003774

    2024 doi

  2. [11]

    Data Assimilation for Ge ophysical Fluids

    Blum J, Le Dimet F-X, Navon I-M. Data Assimilation for Ge ophysical Fluids. Handbook of Numerical Analysis. 2009;14:385-441. https://doi.org/1 0.1016/S1570-8659(08)00209-3

    2009

  3. [12]

    Physics-i nformed neural networks for mul- tiphysics data assimilation with application to subsurfac e transport

    He Q, Barajas-Solano D, Tartakovsky G, et al. Physics-i nformed neural networks for mul- tiphysics data assimilation with application to subsurfac e transport. Advances in W ater Resources 2020; https://doi.org/10.1016/j.advwatres.2 020.103610

    2020 doi

  4. [13]

    Deep data assimilation: int egrating deep learning with data assimilation

    Arcucci R, Zhu J, Hu S, et al. Deep data assimilation: int egrating deep learning with data assimilation. Appl. Sci. 2021;11:1114. https://doi.org/ 10.3390/app11031114

    2021 doi

  5. [14]

    Data assimilation with deep n eural nets informed by nudging

    Antil H, Lohner R, Price R. Data assimilation with deep n eural nets informed by nudging. https://arxiv.org/abs/2111.11505. November 2021

    2021 arXiv

  6. [15]

    A novel neural network traini ng framework with data assimi- lation

    Chen C, Dou Y, Chen J, et al. A novel neural network traini ng framework with data assimi- lation. Journal of Supercomputing. 2022;78:19020–19045. https://doi.org/10.1007/s11227- 04629-7

    2022 doi

  7. [16]

    Iterative methods f or data assimilation for Burgers’ equa- tion

    Lundvall J, Kozlov V, W einerfelt P. Iterative methods f or data assimilation for Burgers’ equa- tion. J. Inverse Ill-Posed Probl. 2006;14:505–535

    2006

  8. [17]

    A nudging-based data assimilation met hod for oceanographuc problems: the back and forth nudging (BFN) algorithm

    Auroux D, Blum J. A nudging-based data assimilation met hod for oceanographuc problems: the back and forth nudging (BFN) algorithm. Proc. Geophys. 2 008;15:305–319

  9. [18]

    Unsteady adjoint method for the optimal control of advection and Burg- ers’ equation using high order spectral difference method

    Ou K, Jameson A. Unsteady adjoint method for the optimal control of advection and Burg- ers’ equation using high order spectral difference method. 4 9th AIAA Aerospace Science Meeting, 4-7 January 2011. Orlando, Florida

    2011

  10. [19]

    The back and forth nudging algorithm f or data assimilation problems: theoretical results on transport equations

    Auroux D, Nodet M. The back and forth nudging algorithm f or data assimilation problems: theoretical results on transport equations. ESAIM:COCV 20 12;18:318–342

  11. [20]

    An evolution of the back and f orth nudging for geophysical data assimilation: application to Burgers equation and compari son

    Auroux D, Bansart P, Blum J. An evolution of the back and f orth nudging for geophysical data assimilation: application to Burgers equation and compari son. Inverse Probl. Sci. Eng. 2013;21:399-419

    2013

  12. [21]

    Numerical aspects of lar ge-time optimal control of Burgers’ equation

    Allahverdi N, Pozo A, Zuazua E. Numerical aspects of lar ge-time optimal control of Burgers’ equation. ESAIM Mathematical Modeling and Numerical Analy sis 2016;50:1371–1401

    2016

  13. [22]

    Filtered gradient algorithms for inv erse design problems of one-dimensional Burgers’ equation

    Gosse L, Zuazua E. Filtered gradient algorithms for inv erse design problems of one-dimensional Burgers’ equation. Innovative Algorithms and Analysis 201 7;197–227

  14. [23]

    Special issue o n inverse problems in geosciences

    de Campos Velho HF, Barbosa VCF, Cocke S. Special issue o n inverse problems in geosciences. Inverse Probl. Sci. Eng. 2013;21:355-356. DOI:10.1080/17 415977.2012.712532

    2013

  15. [24]

    Contaminant source identific ation in acquifers: a critical view

    Gomez-Hernandez JJ, Xu T. Contaminant source identific ation in acquifers: a critical view. Math Geosci 2022;54:437–458

    2022

  16. [25]

    Non-point cont aminant source identification in an acquifer using the ensemble smoother with multiple data ass imilation

    Xu T, Zhang W, Gomez-Hernandez JJ, et al. Non-point cont aminant source identification in an acquifer using the ensemble smoother with multiple data ass imilation. Journal of Hydrology 2022; 606:127405

    2022

  17. [26]

    Properties of advecti on algorithms in the context of varia- tional data assimilation

    Vukicevic T, Steyskal M, Hecht M. Properties of advecti on algorithms in the context of varia- tional data assimilation. Monthly W eather Review 2001;129 :1221–1231

    2001

  18. [27]

    Compensating operators and stable backwar d in time marching in nonlinear parabolic equations

    Carasso AS. Compensating operators and stable backwar d in time marching in nonlinear parabolic equations. Int J Geomath 2014;5:1–16

    2014

  19. [28]

    Stable explicit time-marching in well-pos ed or ill-posed nonlinear parabolic equa- 11 tions

    Carasso AS. Stable explicit time-marching in well-pos ed or ill-posed nonlinear parabolic equa- 11 tions. Inverse Probl. Sci. Eng. 2016;24:1364–1384

    2016

  20. [29]

    Stable explicit marching scheme in ill-pos ed time-reversed viscous wave equations

    Carasso AS. Stable explicit marching scheme in ill-pos ed time-reversed viscous wave equations. Inverse Probl. Sci. Eng. 2016;24:1454–1474

    2016

  21. [30]

    Stabilized Richardson leapfrog scheme in e xplicit stepwise computation of forward or backward nonlinear parabolic equations

    Carasso AS. Stabilized Richardson leapfrog scheme in e xplicit stepwise computation of forward or backward nonlinear parabolic equations. Inverse Probl. Sci. Eng. 2017;25:1–24

    2017

  22. [32]

    Stable explicit stepwise marching scheme i n ill- posed time-reversed 2D Burgers’ equation

    Carasso AS. Stable explicit stepwise marching scheme i n ill- posed time-reversed 2D Burgers’ equation. Inverse Probl. Sci. Eng. 2018;27(12):1-17. DOI: 10.1080/17415977.2018.1523905

    2018

  23. [33]

    Computing ill-posed time-reversed 2D Navi er-Stokes equations, using a stabilized explicit finite difference scheme marching backward in time

    Carasso AS. Computing ill-posed time-reversed 2D Navi er-Stokes equations, using a stabilized explicit finite difference scheme marching backward in time. Inverse Probl. Sci. Eng. 2019; DOI:10.1080/17415977.2019.1698564

    2019

  24. [34]

    Stabilized leapfrog scheme run backward in time, and the explicit O(∆ t)2 stepwise computation of ill-posed time-reversed 2D Navier-Stokes e quations

    Carasso AS. Stabilized leapfrog scheme run backward in time, and the explicit O(∆ t)2 stepwise computation of ill-posed time-reversed 2D Navier-Stokes e quations. Inverse Probl. Sci. Eng. 2021; DOI:10.1080/17415977.2021.1972997

    2021

  25. [35]

    Non-Standard and Improperly Pose d Problems

    Ames KA, Straughan B. Non-Standard and Improperly Pose d Problems. New York (NY): Academic Press; 1997

    1997

  26. [36]

    Some remarks on ill-posed problems for viscou s fluids

    Payne LE. Some remarks on ill-posed problems for viscou s fluids. Int. J. Engng Sci. 1992;30:1341–1347

    1992

  27. [37]

    Logarithmic convexity and other techniques a pplied to problems in continuum echan- ics

    Knops RJ. Logarithmic convexity and other techniques a pplied to problems in continuum echan- ics. In: Knops RJ, editor. Symposium on non-well-posed prob lems and logarithmic con- vexity. Vol. 316, Lecture notes in mathematics. New York (NY ): Springer-Verlag; 1973

    1973

  28. [38]

    On the stability of solutions of the N avier-Stokes equations backward in time

    Knops RJ, Payne LE. On the stability of solutions of the N avier-Stokes equations backward in time. Arch. Rat. Mech. Anal. 1968;29:331–335

    1968

  29. [39]

    Uniqueness and continuous dependence criter ia for the Navier-Stokes equations

    Payne LE. Uniqueness and continuous dependence criter ia for the Navier-Stokes equations. Rocky Mountain J. Math. 1971;2:641–660

    1971

  30. [40]

    Reconstructing the past from imprecise kno wledge of the present: Effective non- uniqueness in solving parabolic equations backward in time

    Carasso AS. Reconstructing the past from imprecise kno wledge of the present: Effective non- uniqueness in solving parabolic equations backward in time . Math. Methods Appl. Sci. 2012;36:249-–261

    2012

  31. [41]

    Computing small solutions of Burgers’ equat ion backwards in time

    Carasso A. Computing small solutions of Burgers’ equat ion backwards in time. J. Math. Anal. App. 1977;59:169–209

    1977

  32. [42]

    Stability estimates for B urgers-type equations backward in time

    H` ao DN, Nguyen VD, Nguyen VT. Stability estimates for B urgers-type equations backward in time. J. Inverse Ill Posed Probl. 2015;23:41–49. 12

    2015

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.