pith. sign in

arxiv: 2605.20639 · v1 · pith:RCRSNGEBnew · submitted 2026-05-20 · 🧮 math.OC · cs.LG· math.DS

Time-Dependent PDE-Constrained Optimization via Weak-Form Latent Dynamics

April Tran , Terry Haut , David Bortz , Youngsoo Choi This is my paper

Pith reviewed 2026-05-21 04:24 UTC · model grok-4.3

classification 🧮 math.OC cs.LGmath.DS
keywords PDE-constrained optimizationreduced-order modelinglatent dynamicsweak-form system identificationtime-dependent PDEgradient-based optimizationsurrogate modelingnoise robustness
0
0 comments X

The pith

Weak-form latent dynamics accelerate gradient-based optimization of time-dependent PDE systems with speedups of up to five orders of magnitude.

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

The paper establishes that weak-form latent-space reduced-order models can serve as efficient surrogates for full-order PDE solves during gradient-based optimization. This matters because repeated forward and sensitivity calculations for high-dimensional time-dependent PDEs make direct optimization computationally prohibitive in many design and control settings. The method compresses trajectories into a low-dimensional latent representation and identifies parametric dynamics via weak-form system identification, which avoids explicit differentiation and improves robustness to noise. Direct-sensitivity and adjoint gradient expressions are then derived in the latent space, and the approach is shown to recover accurate optimal designs on three benchmark problems while delivering large computational savings.

Core claim

The framework builds on Weak-form Latent Space Dynamics Identification (WLaSDI) to compress high-dimensional solution trajectories into a low-dimensional latent representation and identify parametric latent dynamics using weak-form system identification. This avoids explicit numerical differentiation of training trajectories, improving robustness to noisy data and yielding more reliable surrogate dynamics. The resulting reduced PDE-constrained optimization problem is formulated with derived direct-sensitivity and adjoint-based gradient expressions for the learned latent dynamics, enabling scalable gradient evaluation with respect to design parameters. Demonstrations on thermal radiative hohl

What carries the argument

Weak-form Latent Space Dynamics Identification (WLaSDI), which compresses high-dimensional trajectories into a latent representation and identifies parametric dynamics via weak-form system identification to create a noise-robust surrogate for deriving optimization gradients.

If this is right

  • The method produces accurate optimal designs for thermal radiative transfer, Vlasov-Poisson, and Burgers equation problems.
  • Performance holds under noisy training data.
  • Gradient evaluation scales efficiently through the latent-space formulation using either direct or adjoint methods.
  • Computational cost drops by up to five orders of magnitude relative to full-order optimization.

Where Pith is reading between the lines

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

  • The same surrogate construction could be tested on other time-dependent systems such as optimal control of fluid flows or structural dynamics where full-order solves dominate the cost.
  • Accuracy for parameter values far outside the training range would need separate verification to bound the domain of reliable use.
  • Pairing the weak-form identification step with adaptive sampling of training trajectories could further reduce the number of full-order solves required upfront.

Load-bearing premise

The latent dynamics identified from weak-form system identification on training trajectories remain sufficiently accurate approximations of the full-order PDE behavior throughout the optimization search, including for unseen design parameters and under noisy data.

What would settle it

Simulate the full-order PDE at the design parameters returned by the latent optimization and compare the resulting objective value to the value predicted by the surrogate model; a large discrepancy would show that the latent approximation has failed to capture the true optimum.

Figures

Figures reproduced from arXiv: 2605.20639 by April Tran, David Bortz, Terry Haut, Youngsoo Choi.

Figure 1
Figure 1. Figure 1: Overview of the WLaSDI framework, featuring 4 key stages: Data Sampling, Compression, Weak-form Dynamics Identification, and [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Burgers’ Equation: High-fidelity solution of Eq. (20) with 40% noise used for training (µ = [0.7, 1.1, 0.9, 0.9]). For compression, we employ Proper Orthogonal Decomposition (POD) with latent dimension Nz = 15. The dimension is selected such that the first 15 singular values capture 99.99% of the total singular value energy (defined as the cumulative sum of squared singular values divided by the total sum)… view at source ↗
Figure 3
Figure 3. Figure 3: Burgers’ Equation: Comparison between the target state uN(µ ⋆), (orange) and the recovered solution u˜ N(µˆ) (blue) obtained using WLaSDI model trained with 40% noise. u˜ N(µˆ) closely matches the target state uN(µ ⋆). Since the true solution of the inverse problem is known, we can directly quantify the accuracy of the recovered parameter µˆ. We evaluate performance using two metrics: (i) the relative para… view at source ↗
Figure 4
Figure 4. Figure 4: Vlasov–Poisson: Phase-space evolution of the distribution function u(x, v, t; µ) for µ = [k, T] = [1, 0.9], including 5% additive Gaussian noise. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Vlasov-Poisson: Parameter space D = [1.0, 1.2] × [0.9, 1.1], with training parameters shown in blue. 4.2.2. WLaSDI Performance in PDE-Constrained Optimization To demonstrate the performance of the WLaSDI surrogate in a PDE-constrained optimization setting, we consider an inverse problem in which the goal is to identify the parameter vector µ defining the initial condition such that the solution at the fina… view at source ↗
Figure 6
Figure 6. Figure 6: Vlasov-Poisson: Target solution uN(µ ⋆ = [1.16, 1.03]); reconstructed solution u˜ N(µˆ = [1.159, 1.023]) from WLaSDI surrogate with 5% noise, and the absolute error |uN(µ ⋆) − u˜ N(µˆ)|. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Thermal Radiative Transfer: Computational setup for the hohlraum simulation: (a) Idealized hohlraum geometry and material regions; (b) DG spatial mesh. To model thermal radiation transport within this configuration, we consider the frequency-averaged, grey TRT equations. These equations describe the coupled evolution of the specific intensity (angular flux) ψ(x, Ω, t) and the material temperature T(x, t). … view at source ↗
Figure 8
Figure 8. Figure 8: Thermal Radiative Transfer: Time evolution of the temperature field T(x, t; µ) for µ = [Touter, Tinner] = [0.3, 0.15] keV. Corner and central hot spots are initialized at Touter and Tinner, respectively, while the remainder of the domain is initialized at 0.05 keV. of the local compression and decompression operators, reduces computational cost, and better captures variations in local solution complexity a… view at source ↗
Figure 9
Figure 9. Figure 9: Thermal Radiative Transfer: Decomposition of the training parameter space. Each checkerboard cell defines a local WLaSDI submodel constructed from the enclosed training parameter (blue). The latent dimension Nz of each submodel is shown. For each local surrogate, Proper Orthogonal Decomposition (POD) is employed as the compression step. The solution snapshots are assembled into a data matrix, and singular … view at source ↗
Figure 10
Figure 10. Figure 10: illustrates the time evolution of σ(IΓun) for two parameter choices: µ = [0.3, 0.15] (blue) and µ = [0.05, 0.05] (orange). The latter yields an almost negligible standard deviation, indicating nearly uniform tem￾perature throughout the simulation. This behavior is expected, as the initial temperature is 0.05 keV and the hot-spot temperatures are set to the same value, resulting in minimal thermal gradient… view at source ↗
Figure 11
Figure 11. Figure 11: illustrates the time evolution of the temperature field corresponding to the optimized parameter µˆ. Figure 12a compares the standard deviation of the temperature along the capsule boundary for a non-optimized parameter (µ = [0.3, 0.15]) and for the optimized parameter µˆ. The optimized configuration produces a clear reduction in temperature variation along the capsule surface. To further assess the robus… view at source ↗
Figure 12
Figure 12. Figure 12: Thermal Radiative Transfer: Quantity of interest (QoI): standard deviation of the capsule surface temperature. (a) QoI from FOM with µ = [0.3, 0.15] (blue) and from surrogate using noise-free global optimized parameter µˆ = [0.215, 0.134] in Eq. (32) (black), showing reduced temperature variation. (b) Comparison of QoI predictions from surrogates trained with 5% noisy data: noise-free reference (black), W… view at source ↗
Figure 13
Figure 13. Figure 13: Thermal Radiative Transfer: Direct gradient-based optimization results using surrogates trained with 5% noisy data. Each point corresponds to one of the 50 runs. The dashed line indicates the noise-free global optimum µˆ = [0.215, 0.134]. WLaSDI (green) consistently recovers values close to the optimum, whereas LaSDI (red) exhibits significantly larger scatter. Finally, [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
read the original abstract

Optimization problems constrained by high-dimensional, time-dependent partial differential equations require repeated forward and sensitivity solves, making high-fidelity optimization computationally prohibitive in many-query design and control settings. We present a weak-form latent-space reduced-order modeling framework for accelerating gradient-based PDE-constrained optimization. The proposed approach builds on Weak-form Latent Space Dynamics Identification (WLaSDI), which compresses high-dimensional solution trajectories into a low-dimensional latent representation and identifies parametric latent dynamics using weak-form system identification. By avoiding explicit numerical differentiation of training trajectories, the weak-form improves robustness to noisy data and yields more reliable surrogate dynamics for optimization. We formulate the resulting reduced PDE-constrained optimization problem and derive both direct-sensitivity and adjoint-based gradient expressions for the learned latent dynamics, enabling scalable gradient evaluation with respect to design parameters. The framework is demonstrated on three time-dependent benchmark problems: thermal radiative transfer for optimal hohlraum design, the two-stream instability Vlasov-Poisson system, and the inviscid Burgers equation. Across these examples, WLaSDI produces accurate optimal designs, remains robust under noisy training data, and delivers substantial computational savings, including speedups of up to five orders of magnitude relative to full-order optimization. These results demonstrate that weak-form latent dynamics provide an efficient and noise-robust surrogate foundation for gradient-based optimization of complex time-dependent PDE systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces a weak-form latent-space reduced-order modeling framework based on Weak-form Latent Space Dynamics Identification (WLaSDI) to accelerate gradient-based optimization of time-dependent PDE-constrained problems. High-dimensional solution trajectories are compressed into a low-dimensional latent representation, parametric latent dynamics are identified via weak-form system identification to improve noise robustness by avoiding explicit differentiation, the reduced optimization problem is formulated, and both direct-sensitivity and adjoint-based gradient expressions are derived for the learned dynamics. The approach is demonstrated on three benchmarks—thermal radiative transfer for optimal hohlraum design, the two-stream instability Vlasov-Poisson system, and the inviscid Burgers equation—reporting accurate optimal designs, robustness under noisy training data, and speedups of up to five orders of magnitude relative to full-order optimization.

Significance. If the latent dynamics remain faithful approximations when the optimizer explores design parameters outside the training distribution, the framework would represent a meaningful advance for making otherwise prohibitive many-query PDE optimization problems tractable. Notable strengths include the weak-form identification step for noise robustness and the explicit derivation of scalable gradients in the reduced space. The empirical results across diverse time-dependent benchmarks provide concrete evidence of practical computational savings when the surrogate assumptions hold.

major comments (2)
  1. [Numerical experiments] Numerical experiments: The central claim that WLaSDI surrogates enable reliable gradient-based optimization with large speedups requires that the identified parametric latent dynamics accurately represent full-order PDE behavior for design parameters encountered during the search, including those outside the training trajectories. The reported results on the three benchmarks show final design accuracy and noise robustness but provide no explicit out-of-distribution validation, such as prediction-error monitoring on intermediate optimizer iterates or ablation studies on parameter ranges beyond the training set. This omission leaves open whether the observed performance stems from faithful surrogate behavior or from limited parameter exploration in the tested cases.
  2. [Gradient derivation] Gradient derivation for latent dynamics: The manuscript derives direct and adjoint gradients with respect to design parameters for the learned latent system, but it is unclear how these expressions handle the dependence on the parameters fitted during the weak-form system identification step itself. If the latent model coefficients are treated as fixed after training, the gradients may miss additional sensitivities that affect the optimization trajectory; a concrete statement of this assumption and its impact on the reported results would strengthen the derivation.
minor comments (1)
  1. [Abstract] The abstract states speedups of 'up to five orders of magnitude'; specifying which benchmark attains this figure and the precise baseline (full-order forward/adjoint solves) in the main text would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive assessment of the framework's strengths in noise robustness and gradient derivations. We address each major comment below and will incorporate revisions to strengthen the presentation of the results.

read point-by-point responses

  1. Referee: [Numerical experiments] Numerical experiments: The central claim that WLaSDI surrogates enable reliable gradient-based optimization with large speedups requires that the identified parametric latent dynamics accurately represent full-order PDE behavior for design parameters encountered during the search, including those outside the training trajectories. The reported results on the three benchmarks show final design accuracy and noise robustness but provide no explicit out-of-distribution validation, such as prediction-error monitoring on intermediate optimizer iterates or ablation studies on parameter ranges beyond the training set. This omission leaves open whether the observed performance stems from faithful surrogate behavior or from limited parameter exploration in the tested cases.

    Authors: We agree that explicit out-of-distribution validation would provide stronger support for the reliability of the surrogates. In the reported benchmarks the optimization trajectories remained within or near the parameter ranges used to generate training data, yielding accurate final designs. To address the concern we will add monitoring of latent prediction errors on intermediate optimizer iterates for each benchmark and include an ablation study with design parameters extended beyond the original training set. These additions will appear in the revised manuscript. revision: yes

  2. Referee: [Gradient derivation] Gradient derivation for latent dynamics: The manuscript derives direct and adjoint gradients with respect to design parameters for the learned latent system, but it is unclear how these expressions handle the dependence on the parameters fitted during the weak-form system identification step itself. If the latent model coefficients are treated as fixed after training, the gradients may miss additional sensitivities that affect the optimization trajectory; a concrete statement of this assumption and its impact on the reported results would strengthen the derivation.

    Authors: We thank the referee for highlighting this point. The weak-form system identification is performed once during the offline training stage to obtain fixed latent dynamics coefficients; these coefficients are held constant during optimization. The derived direct and adjoint gradients are therefore taken with respect to the design parameters while treating the identified latent model as a fixed surrogate. This is the standard assumption in reduced-order surrogate optimization. We will add an explicit statement of the assumption and a short discussion of its implications in the gradient derivation section of the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper identifies a latent dynamics model via WLaSDI on training trajectories, then derives direct-sensitivity and adjoint gradient expressions for the reduced optimization problem with respect to design parameters. These gradient derivations operate on the already-identified surrogate and do not reduce by construction to the training data or to the optimization objective itself. No self-definitional steps, fitted-input predictions, or load-bearing self-citation chains appear in the central claims. The accuracy of the surrogate on unseen parameters is treated as an assumption to be validated empirically rather than enforced by definition. The framework is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides insufficient detail to enumerate specific free parameters, axioms, or invented entities; the central claim rests on the unstated assumption that the learned latent dynamics generalize to the optimization task.

pith-pipeline@v0.9.0 · 5781 in / 1169 out tokens · 38297 ms · 2026-05-21T04:24:40.668730+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

73 extracted references · 73 canonical work pages

  1. [1]

    N. V . Queipo, R. T. Haftka, W. Shyy, T. Goel, R. Vaidyanathan, P. Kevin Tucker, Surrogate-based analysis and optimization, Progress in Aerospace Sciences 41 (1) (2005) 1–28.doi:10.1016/j.paerosci.2005.02.001

    work page doi:10.1016/j.paerosci.2005.02.001 2005

  2. [2]

    A. I. Forrester, A. J. Keane, Recent advances in surrogate-based optimization, Progress in Aerospace Sciences 45 (1-3) (2009) 50–79.doi:10.1016/j.paerosci.2008.11.001

    work page doi:10.1016/j.paerosci.2008.11.001 2009

  3. [3]

    Benner, S

    P. Benner, S. Gugercin, K. Willcox, A Survey of Projection-Based Model Reduction Methods for Parametric Dynamical Systems, SIAM Rev. 57 (4) (2015) 483–531.doi:10.1137/130932715

    work page doi:10.1137/130932715 2015

  4. [4]

    Berkooz, P

    G. Berkooz, P. Holmes, J. L. Lumley, The Proper Orthogonal Decomposition in the Analysis of Turbulent Flows, Annu. Rev. Fluid Mech. 25 (1) (1993) 539–575.doi:10.1146/annurev.fl.25.010193.002543

    work page doi:10.1146/annurev.fl.25.010193.002543 1993

  5. [5]

    Rozza, D

    G. Rozza, D. B. P. Huynh, A. T. Patera, Reduced Basis Approximation and a Posteriori Error Estimation for Affinely Parametrized Elliptic Coercive Partial Differential Equations: Application to Transport and Continuum Mechanics, Arch Computat Methods Eng 15 (3) (2008) 229–275.doi:10.1007/s11831-008-9019-9

    work page doi:10.1007/s11831-008-9019-9 2008

  6. [6]

    Safonov, R

    M. Safonov, R. Chiang, A Schur method for balanced-truncation model reduction, IEEE Trans. Automat. Contr. 34 (7) (1989) 729–733.doi:10.1109/9.29399

    work page doi:10.1109/9.29399 1989

  7. [7]

    G. E. Hinton, R. R. Salakhutdinov, Reducing the Dimensionality of Data with Neural Networks, Science 313 (5786) (2006) 504–507.doi:10.1126/science.1127647

    work page doi:10.1126/science.1127647 2006

  8. [8]

    X. Xie, G. Zhang, C. G. Webster, Non-Intrusive Inference Reduced Order Model for Fluids Using Deep Multi- step Neural Network, Mathematics 7 (8) (2019) 757.doi:10.3390/math7080757

    work page doi:10.3390/math7080757 2019

  9. [9]

    Kadeethum, F

    T. Kadeethum, F. Ballarin, Y . Choi, D. O’Malley, H. Yoon, N. Bouklas, Non-intrusive reduced order model- ing of natural convection in porous media using convolutional autoencoders: Comparison with linear subspace techniques, Advances in Water Resources 160 (2022) 104098.doi:10.1016/j.advwatres.2021.104098

    work page doi:10.1016/j.advwatres.2021.104098 2022

  10. [10]

    D. M. Copeland, S. W. Cheung, K. Huynh, Y . Choi, Reduced order models for Lagrangian hydrodynamics, Computer Methods in Applied Mechanics and Engineering 388 (2022) 114259.doi:10.1016/j.cma.2021. 114259

    work page doi:10.1016/j.cma.2021 2022

  11. [11]

    S. W. Cheung, Y . Choi, D. M. Copeland, K. Huynh, Local Lagrangian reduced-order modeling for the Rayleigh- Taylor instability by solution manifold decomposition, Journal of Computational Physics 472 (2023) 111655. doi:10.1016/j.jcp.2022.111655

    work page doi:10.1016/j.jcp.2022.111655 2023

  12. [12]

    Maulik, B

    R. Maulik, B. Lusch, P. Balaprakash, Non-autoregressive time-series methods for stable parametric reduced- order models, Phys. Fluids 32 (8) (2020) 087115.doi:10.1063/5.0019884. 25

    work page doi:10.1063/5.0019884 2020

  13. [13]

    Amsallem, C

    D. Amsallem, C. Farhat, Interpolation Method for Adapting Reduced-Order Models and Application to Aeroe- lasticity, AIAA Journal 46 (7) (2008) 1803–1813.doi:10.2514/1.35374

    work page doi:10.2514/1.35374 2008

  14. [14]

    Bui-Thanh, M

    T. Bui-Thanh, M. Damodaran, K. Willcox, Aerodynamic Data Reconstruction and Inverse Design Using Proper Orthogonal Decomposition, AIAA Journal 42 (8) (2004) 1505–1516.doi:10.2514/1.2159

    work page doi:10.2514/1.2159 2004

  15. [15]

    S. A. McQuarrie, C. Huang, K. E. Willcox, Data-driven reduced-order models via regularised Operator Inference for a single-injector combustion process, Journal of the Royal Society of New Zealand 51 (2) (2021) 194–211. doi:10.1080/03036758.2020.1863237

    work page doi:10.1080/03036758.2020.1863237 2021

  16. [16]

    Y . Choi, G. Boncoraglio, S. Anderson, D. Amsallem, C. Farhat, Gradient-based constrained optimization using a database of linear reduced-order models, Journal of Computational Physics 423 (2020) 109787.doi:10.1016/ j.jcp.2020.109787

    work page arXiv 2020

  17. [17]

    Amsallem, M

    D. Amsallem, M. Zahr, Y . Choi, C. Farhat, Design optimization using hyper-reduced-order models, Struct Mul- tidisc Optim 51 (4) (2015) 919–940.doi:10.1007/s00158-014-1183-y

    work page doi:10.1007/s00158-014-1183-y 2015

  18. [18]

    W. D. Fries, X. He, Y . Choi, LaSDI: Parametric Latent Space Dynamics Identification, Computer Methods in Applied Mechanics and Engineering 399 (2022) 115436.doi:10.1016/j.cma.2022.115436

    work page doi:10.1016/j.cma.2022.115436 2022

  19. [19]

    Champion, B

    K. Champion, B. Lusch, J. N. Kutz, S. L. Brunton, Data-driven discovery of coordinates and governing equa- tions, Proc. Natl. Acad. Sci. U.S.A. 116 (45) (2019) 22445–22451.doi:10.1073/pnas.1906995116

    work page doi:10.1073/pnas.1906995116 2019

  20. [20]

    S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proc. Natl. Acad. Sci. U.S.A. 113 (15) (2016) 3932–3937.doi:10.1073/pnas. 1517384113

    work page doi:10.1073/pnas 2016

  21. [21]

    X. He, Y . Choi, W. D. Fries, J. L. Belof, J.-S. Chen, gLaSDI: Parametric physics-informed greedy latent space dynamics identification, Journal of Computational Physics 489 (2023) 112267.doi:10.1016/j.jcp.2023. 112267

    work page doi:10.1016/j.jcp.2023 2023

  22. [22]

    Bonneville, Y

    C. Bonneville, Y . Choi, D. Ghosh, J. L. Belof, GPLaSDI: Gaussian Process-based interpretable Latent Space Dynamics Identification through deep autoencoder, Computer Methods in Applied Mechanics and Engineering 418 (2024) 116535.doi:10.1016/j.cma.2023.116535

    work page doi:10.1016/j.cma.2023.116535 2024

  23. [23]

    J. S. R. Park, S. W. Cheung, Y . Choi, Y . Shin, tLaSDI: Thermodynamics-informed latent space dynamics identi- fication, Comput. Methods Appl. Mech. Eng. 429 (2024) 117144.doi:10.1016/j.cma.2024.117144

    work page doi:10.1016/j.cma.2024.117144 2024

  24. [24]

    Stephany, W

    R. Stephany, W. Anderson, Y . Choi, Higher-order lasdi: Reduced order modeling with multiple time derivatives, Computer Methods in Applied Mechanics and Engineering 455 (2026) 118890

    work page 2026

  25. [25]

    Rollout-lasdi: Enhancing the long-term accuracy of latent space dynamics.arXiv preprint arXiv:2509.08191, 2025

    R. Stephany, Y . Choi, Rollout-LaSDI: Enhancing the long-term accuracy of Latent Space Dynamics (2025). doi:10.48550/ARXIV.2509.08191

    work page doi:10.48550/arxiv.2509.08191 2025

  26. [26]

    Anderson, S

    W. Anderson, S. W. Chung, Y . Choi, Sequential decoder training for improved latent space dynamics identifica- tion (2025).doi:10.48550/ARXIV.2510.03535

    work page doi:10.48550/arxiv.2510.03535 2025

  27. [27]

    S. W. Chung, C. Miller, Y . Choi, P. Tranquilli, H. K. Springer, K. Sullivan, Latent space dynamics identification for interface tracking with application to shock-induced pore collapse, Journal of Computational Physics (2026) 114978

    work page 2026

  28. [28]

    A. Tran, X. He, D. A. Messenger, Y . Choi, D. M. Bortz, Weak-form latent space dynamics identification, Computer Methods in Applied Mechanics and Engineering 427 (2024) 116998.doi:10.1016/j.cma.2024. 116998

    work page doi:10.1016/j.cma.2024 2024

  29. [29]

    X. He, A. Tran, D. M. Bortz, Y . Choi, Physics-Informed Active Learning With Simultaneous Weak-Form Latent Space Dynamics Identification, Numerical Meth Engineering 126 (1) (2025) e7634.doi:10.1002/nme.7634. 26

    work page doi:10.1002/nme.7634 2025

  30. [30]

    Bonneville, X

    C. Bonneville, X. He, A. Tran, J. S. Park, W. Fries, D. A. Messenger, S. W. Cheung, Y . Shin, D. M. Bortz, D. Ghosh, J.-S. Chen, J. Belof, Y . Choi, A Comprehensive Review of Latent Space Dynamics Identification Algorithms for Intrusive and Non-Intrusive Reduced-Order-Modeling (2024).doi:10.48550/ARXIV.2403. 10748

    work page doi:10.48550/arxiv.2403 2024

  31. [31]

    D. M. Bortz, D. A. Messenger, V . Dukic, Direct Estimation of Parameters in ODE Models Using WENDy: Weak-Form Estimation of Nonlinear Dynamics, Bull Math Biol 85 (11) (2023) 110.doi:10.1007/ s11538-023-01208-6

    work page 2023

  32. [32]

    Yamaleev, B

    N. Yamaleev, B. Diskin, E. Nielsen, Adjoint-Based Methodology for Time-Dependent Optimization, in: 12th AIAAISSMO Multidiscip. Anal. Optim. Conf., American Institute of Aeronautics and Astronautics, Victoria, British Columbia, Canada, 2008.doi:10.2514/6.2008-5857

    work page doi:10.2514/6.2008-5857 2008

  33. [33]

    Zahr, P.-O

    M. Zahr, P.-O. Persson, An adjoint method for a high-order discretization of deforming domain conservation laws for optimization of flow problems, Journal of Computational Physics 326 (2016) 516–543.doi:10.1016/ j.jcp.2016.09.012

    work page 2016

  34. [34]

    Shinbrot, On the Analysis of Linear and Nonlinear Systems, J

    M. Shinbrot, On the Analysis of Linear and Nonlinear Systems, J. Fluids Eng. 79 (3) (1957) 547–551.doi: 10.1115/1.4013092

    work page doi:10.1115/1.4013092 1957

  35. [35]

    J. Loeb, G. Cahen, More about process identification, IEEE Trans. Automat. Contr. 10 (3) (1965) 359–361. doi:10.1109/TAC.1965.1098172

    work page doi:10.1109/tac.1965.1098172 1965

  36. [36]

    Preisig, D

    H. Preisig, D. Rippin, Theory and application of the modulating function method—I. Review and theory of the method and theory of the spline-type modulating functions, Computers & Chemical Engineering 17 (1) (1993) 1–16.doi:10.1016/0098-1354(93)80001-4

    work page doi:10.1016/0098-1354(93)80001-4 1993

  37. [37]

    S. H. Rudy, S. L. Brunton, J. L. Proctor, J. N. Kutz, Data-driven discovery of partial differential equations, Sci. Adv. 3 (4) (2017) e1602614.doi:10.1126/sciadv.1602614

    work page doi:10.1126/sciadv.1602614 2017

  38. [38]

    Pantazis, I

    Y . Pantazis, I. Tsamardinos, A unified approach for sparse dynamical system inference from temporal measure- ments, Bioinformatics 35 (18) (2019) 3387–3396.doi:10.1093/bioinformatics/btz065

    work page doi:10.1093/bioinformatics/btz065 2019

  39. [39]

    D. R. Gurevich, P. A. K. Reinbold, R. O. Grigoriev, Robust and optimal sparse regression for nonlinear PDE models, Chaos Interdiscip. J. Nonlinear Sci. 29 (10) (2019) 103113.doi:10.1063/1.5120861

    work page doi:10.1063/1.5120861 2019

  40. [40]

    Schaeffer, S

    H. Schaeffer, S. G. McCalla, Sparse model selection via integral terms, Phys. Rev. E 96 (2) (2017) 023302. doi:10.1103/PhysRevE.96.023302

    work page doi:10.1103/physreve.96.023302 2017

  41. [41]

    Z. Wang, X. Huan, K. Garikipati, Variational system identification of the partial differential equations governing the physics of pattern-formation: Inference under varying fidelity and noise, Computer Methods in Applied Mechanics and Engineering 356 (2019) 44–74.doi:10.1016/j.cma.2019.07.007

    work page doi:10.1016/j.cma.2019.07.007 2019

  42. [42]

    P. A. K. Reinbold, D. R. Gurevich, R. O. Grigoriev, Using noisy or incomplete data to discover models of spatiotemporal dynamics, Phys. Rev. E 101 (1) (2020) 010203.doi:10.1103/PhysRevE.101.010203

    work page doi:10.1103/physreve.101.010203 2020

  43. [43]

    D. A. Messenger, A. Tran, V . Dukic, D. M. Bortz, The Weak Form Is Stronger Than You Think, SIAM News 57 (8) (Oct. 2024)

    work page 2024

  44. [44]

    D. A. Messenger, D. M. Bortz, Asymptotic consistency of the WSINDy algorithm in the limit of continuum data, IMA J. Numer. Anal. (2024) drae086doi:10.1093/imanum/drae086

    work page doi:10.1093/imanum/drae086 2024

  45. [45]

    A. Tran, D. M. Bortz, Weak Form Scientific Machine Learning: Test Function Construction for System Identifi- cation, SIAM J. Sci. Comput. (accepted) (2026).doi:10.48550/ARXIV.2507.03206

    work page doi:10.48550/arxiv.2507.03206 2026

  46. [46]

    D. A. Messenger, D. M. Bortz, Weak SINDy: Galerkin-Based Data-Driven Model Selection, Multiscale Model. Simul. 19 (3) (2021) 1474–1497.doi:10.1137/20M1343166. 27

    work page doi:10.1137/20m1343166 2021

  47. [47]

    D. A. Messenger, D. M. Bortz, Weak SINDy for partial differential equations, Journal of Computational Physics 443 (2021) 110525.doi:10.1016/j.jcp.2021.110525

    work page doi:10.1016/j.jcp.2021.110525 2021

  48. [48]

    Rummel, D

    N. Rummel, D. A. Messenger, S. Becker, V . Dukic, D. M. Bortz, WENDy for Nonlinear-in-Parameters ODEs (2025).doi:10.48550/ARXIV.2502.08881

    work page doi:10.48550/arxiv.2502.08881 2025

  49. [49]

    Chawla, D

    A. Chawla, D. M. Bortz, V . Dukic, Bias and Coverage Properties of the WENDy-IRLS Algorithm, in: B. Sri- raman (Ed.), Handbook of Visual, Experimental and Computational Mathematics, Springer Nature Switzerland, Cham, 2026, pp. 1–108.doi:10.1007/978-3-030-93954-0_86-1

    work page doi:10.1007/978-3-030-93954-0_86-1 2026

  50. [50]

    D. A. Messenger, G. E. Wheeler, X. Liu, D. M. Bortz, Learning anisotropic interaction rules from individual trajectories in a heterogeneous cellular population, J. R. Soc. Interface. 19 (195) (2022) 20220412.doi:10. 1098/rsif.2022.0412

    work page arXiv 2022

  51. [51]

    Messenger, G

    D. Messenger, G. Dwyer, V . Dukic, Weak-form inference for hybrid dynamical systems in ecology, J. R. Soc. Interface. 21 (221) (2024) 20240376.doi:10.1098/rsif.2024.0376

    work page doi:10.1098/rsif.2024.0376 2024

  52. [52]

    Heitzman-Breen, V

    N. Heitzman-Breen, V . Dukic, D. M. Bortz, A Practical Identifiability Criterion Leveraging Weak-Form Param- eter Estimation, Bull Math Biol 88 (5) (2026) 70.doi:10.1007/s11538-026-01639-x

    work page doi:10.1007/s11538-026-01639-x 2026

  53. [53]

    Lyons, V

    R. Lyons, V . Dukic, D. M. Bortz, Learning structured population models from data with WSINDy, PLoS Comput Biol 21 (12) (2025) e1013742.doi:10.1371/journal.pcbi.1013742

    work page doi:10.1371/journal.pcbi.1013742 2025

  54. [54]

    Minor, B

    S. Minor, B. D. Elderd, B. Van Allen, D. M. Bortz, V . Dukic, Weak Form Learning for Mean-Field Partial Differential Equations: An Application to Insect Movement (2025).doi:10.48550/ARXIV.2510.07786

    work page doi:10.48550/arxiv.2510.07786 2025

  55. [55]

    Minor, D

    S. Minor, D. A. Messenger, V . Dukic, D. M. Bortz, Learning Physically Interpretable Atmospheric Models From Data With WSINDy, Journal of Geophysical Research: Machine Learning and Computation 2 (3) (2025) e2025JH000602.doi:10.1029/2025JH000602

    work page doi:10.1029/2025jh000602 2025

  56. [56]

    Vasey, D

    G. Vasey, D. A. Messenger, D. M. Bortz, A. J. Christlieb, B. O’Shea, Influence of initial conditions on data- driven model identification and information entropy for ideal mhd problems, Journal of Computational Physics 524 (2025) 113719.doi:10.1016/j.jcp.2025.113719

    work page doi:10.1016/j.jcp.2025.113719 2025

  57. [57]

    Messenger, B

    D. Messenger, B. Southworth, H. Hammer, L. Chacon, Learning interpretable closures for thermal radiation transport in optically-thin media using WSINDy (2025).doi:10.48550/ARXIV.2510.11840

    work page doi:10.48550/arxiv.2510.11840 2025

  58. [58]

    A. R. Conn, N. I. M. Gould, P. L. Toint, Trust Region Methods, Society for Industrial and Applied Mathematics, 2000.doi:10.1137/1.9780898719857

    work page doi:10.1137/1.9780898719857 2000

  59. [59]

    On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming,

    A. Wächter, L. T. Biegler, On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming, Math. Program. 106 (1) (2006) 25–57.doi:10.1007/s10107-004-0559-y

    work page doi:10.1007/s10107-004-0559-y 2006

  60. [60]

    P. E. Gill, W. Murray, M. A. Saunders, SNOPT: An SQP Algorithm for Large-Scale Constrained Optimization, SIAM Rev. 47 (1) (2005) 99–131.doi:10.1137/S0036144504446096

    work page doi:10.1137/s0036144504446096 2005

  61. [61]

    E., et al

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. Van Der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey, ˙I. Polat, Y . Feng, E. W. Moore, J. VanderPlas, D. Laxalde, J. Perk- told, R. Cimrman, I. Henriks...

    work page doi:10.1038/s41592-019-0686-2 2020

  62. [62]

    T. M. Ragonneau, Model-Based Derivative-Free Optimization Methods and Software, Ph.D. thesis, The Hong Kong Polytechnic University, China (2022)

    work page 2022

  63. [63]

    T. M. Ragonneau, Z. Zhang, COBYQA Version 1.1.3. 2025 (2025)

    work page 2025

  64. [64]

    Nocedal, S

    J. Nocedal, S. J. Wright, Numerical Optimization, 2nd Edition, Springer Series in Operations Research, Springer, New York, 2006

    work page 2006

  65. [65]

    Ghosh, J

    D. Ghosh, J. Loffeld, Y . Kim, HyPar: Finite-Difference Hyperbolic-Parabolic PDE Solver on Cartesian Grids

    work page

  66. [66]

    Jiang, C.-W

    G.-S. Jiang, C.-W. Shu, Efficient Implementation of Weighted ENO Schemes, Journal of Computational Physics 126 (1) (1996) 202–228.doi:10.1006/jcph.1996.0130

    work page doi:10.1006/jcph.1996.0130 1996

  67. [67]

    Carlson, K

    B. Carlson, K. Lathrop, L. A. S. Laboratory, Transport Theory: The Method of Discrete Ordinates, Los Alamos Scientific Laboratory of the University of California, 1965

    work page 1965

  68. [68]

    E. W. Larsen, J. E. Morel, Advances in Discrete-Ordinates Methodology, Springer Netherlands, Dordrecht, 2010, pp. 1–84.doi:10.1007/978-90-481-3411-3_1

    work page doi:10.1007/978-90-481-3411-3_1 2010

  69. [69]

    Olivier, W

    S. Olivier, W. Pazner, T. S. Haut, B. C. Yee, A family of independent Variable Eddington Factor methods with efficient preconditioned iterative solvers, Journal of Computational Physics 473 (2023) 111747.doi: 10.1016/j.jcp.2022.111747

    work page doi:10.1016/j.jcp.2022.111747 2023

  70. [70]

    Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces,

    R. Storn, K. Price, Differential Evolution – A Simple and Efficient Heuristic for global Optimization over Con- tinuous Spaces, Journal of Global Optimization 11 (4) (1997) 341–359.doi:10.1023/A:1008202821328

    work page doi:10.1023/a:1008202821328 1997

  71. [71]

    C. L. Lawson, R. J. Hanson, Solving Least Squares Problems, no. 15 in Classics in Applied Mathematics, SIAM, Philadelphia, 1995

    work page 1995

  72. [72]

    R. H. Byrd, M. E. Hribar, J. Nocedal, An interior point algorithm for large-scale nonlinear programming, SIAM J. Optim. 9 (1999) 877–900

    work page 1999

  73. [73]

    ∂˜r0 ∂z0 #−1 ∂˜r0 ∂µi , ∂zn ∂µi =−

    A. Iserles, A First Course in the Numerical Analysis of Differential Equations, 2nd Edition, Cambridge Univer- sity Press, 2008.doi:10.1017/CBO9780511995569. Appendix A. Interpolation of Coefficients Within the LaSDI framework, latent dynamics are governed by an ODE whose coefficient matrix may depend on the physical parameterµ∈ D. Given training dataS= n...

    work page doi:10.1017/cbo9780511995569 2008