pith. sign in

arxiv: 2604.16664 · v1 · submitted 2026-04-17 · ⚛️ physics.plasm-ph · physics.comp-ph

Physics-Informed Latent Space Dynamics Identification for Time-Dependent NLTE Atomic Kinetics

Jeongwoo Nam , William Anderson , Youngsoo Choi , Hai P. Le , Mark E. Foord , Byoung Ick Cho , Haewon Jeong , Min Sang Cho This is my paper

Pith reviewed 2026-05-10 06:50 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph physics.comp-ph
keywords NLTE atomic kineticsphysics-informed machine learninglatent space dynamicsplasma modelingreduced-order modelingEUV lithography plasmastime-dependent kineticscharge-state evolution
0
0 comments X

The pith

A physics-informed latent dynamics model reproduces time-dependent NLTE atomic kinetics with under 2% error and 100,000-fold speedups while staying stable outside training data.

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

The paper presents a framework that learns an explicit reduced governing equation for the time evolution of non-local thermodynamic equilibrium atomic populations in plasmas. It adds loss terms that enforce consistency with macroscopic quantities, keep the dynamics stable over long integrations, and drive the system to the correct steady state. This matters for radiation-hydrodynamics simulations because full NLTE calculations remain too expensive to run at every time step. If the method works as described, it would replace slow kinetic solvers with a fast surrogate that still produces physically admissible charge-state histories along realistic temperature-density paths.

Core claim

The pLaSDI framework learns a latent-space dynamical model for tin NLTE population kinetics by combining dimensionality reduction with a physics-informed neural ODE; three custom loss terms enforce macroscopic consistency, dynamical stability, and convergence to the correct steady-state solution under fixed plasma conditions. When trained on hydrodynamically modeled trajectories, the reduced model reproduces charge-state evolution and mean charge with errors below 2 percent, delivers speedups of 5 times 10 to the 4 to 10 to the 5, and remains stable when integrated outside the training trajectories, converging to admissible states and the proper equilibrium.

What carries the argument

The pLaSDI model, which reduces the high-dimensional NLTE rate equations to a low-dimensional latent space whose time evolution is learned as an explicit dynamical system subject to physics-informed loss terms for consistency, stability, and steady-state convergence.

If this is right

  • The reduced model can be integrated orders of magnitude faster than the original NLTE solver while preserving key physical behaviors such as charge-state evolution.
  • Long-time simulations remain stable even when the plasma conditions move outside the original training trajectories.
  • Under fixed plasma conditions the model converges to the physically correct steady-state populations without drifting into unphysical states.
  • Charge-state distributions and mean charges match reference data closely enough for direct use inside radiation-hydrodynamics codes.

Where Pith is reading between the lines

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

  • The same physics-loss construction could be applied to other atomic species or plasma regimes beyond the tin data used here.
  • Coupling the fast surrogate directly into a radiation-hydrodynamics code would enable self-consistent, long-duration NLTE simulations that are currently intractable.
  • The approach suggests that explicit enforcement of stability and steady-state constraints can reduce the amount of training data needed for reliable extrapolation in stiff kinetic systems.

Load-bearing premise

That the chosen physics-informed loss terms for macroscopic consistency, dynamical stability, and steady-state convergence are sufficient to keep the learned latent dynamics faithful to the underlying NLTE kinetics during long-time integration and extrapolation beyond the training trajectories.

What would settle it

Integrate the trained model along a new temperature-density trajectory outside the training set and compare its predicted time series of charge-state populations and mean charge directly against an independent full NLTE calculation performed on the same trajectory; systematic deviations larger than a few percent would falsify the claim.

Figures

Figures reproduced from arXiv: 2604.16664 by Byoung Ick Cho, Haewon Jeong, Hai P. Le, Jeongwoo Nam, Mark E. Foord, Min Sang Cho, William Anderson, Youngsoo Choi.

Figure 1
Figure 1. Figure 1: Overview of the physics-informed LaSDI (pLaSDI) framework for NLTE atomic kinetics. An autoencoder compresses the high-dimensional population vector n ∈ ℝ𝑁 into a latent state z ∈ ℝ𝑑 (𝑑 ≪ 𝑁), where dynamics are governed by an explicit ODE identified via DMDc with plasma conditions u(𝑡) = (𝑇 (𝑡), 𝜌(𝑡)) as control variables. Physics-informed constraints enforce Hurwitz stability and steady-state consistency … view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the data generation workflow. (a) Radiation–hydrodynamic simula￾tion of a laser-heated tin plasma using FLASH, showing the spatial distribution of density. (b) Temperature–density trajectories of Lagrangian particles extracted from the FLASH simulation, with the color map indicating the sample count across the parameter space. (c) TD–NLTE population distributions generated by SCFLY using the te… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of microscopic population predictions for a representative validation case. (a) Fractional population from the SCFLY reference, (b) fractional population from the pLaSDI prediction, and (c) their pointwise absolute error. (d) shows the corresponding w￾scaled populations from SCFLY and pLaSDI, highlighting agreement in the low-population regime. All panels share the same y-axis range. 18 [PITH_F… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of macroscopic observables for a representative validation case. (a– c) Charge-state distribution: (a) SCFLY reference, (b) pLaSDI prediction, (c) pointwise absolute error. (d) Population conservation (∑𝑖 𝑛𝑖 − 1); the deviation remains at the 10−13 level. (e) Mean charge state .̄𝑞 6.3 Role of Physics-informed Constraints and Steady-state Gen￾eralization This subsection examines the role of the H… view at source ↗
Figure 5
Figure 5. Figure 5: Effect of the Hurwitz stability constraint on long-time population predictions. (a) Reference SCFLY population distribution up to 4 ns; the inset shows the prescribed tem￾perature and density histories, which are held fixed beyond 4 ns. (b) Without the Hurwitz constraint; the learned operator 𝐴 has large eigenvalues with positive real parts (severely vio￾lating the Hurwitz condition), and the predicted pop… view at source ↗
Figure 6
Figure 6. Figure 6: Effect of the steady state constraint on the predicted equilibrium population and charge state distribution at 𝑇 = 45 eV and 𝜌 = 10−5 g/cc. (a, b) Without the constraint, the predicted population and CSD are visibly shifted from the reference. (c, d) With the constraint, the predicted distributions closely overlap the reference. The population panels are zoomed in to super-configuration indices 200–1000 (o… view at source ↗
Figure 7
Figure 7. Figure 7: Steady-state population predictions from the analytically computed latent equilib￾rium z ∗ = −𝐴−1(a + 𝐵u ∗ ), compared with SCFLY reference solutions across four densities (10−5 , 10−4 , 10−3 , 10−2 g/cc). Within each density block, temperature increases from 1 to 50 eV. (a–d) SCFLY reference, (e–h) pLaSDI prediction, (i–l) pointwise absolute error. Columns correspond to increasing density from left to rig… view at source ↗
read the original abstract

Non-local thermodynamic equilibrium (NLTE) calculations remain a major computational bottleneck in radiation--hydrodynamics, while most existing machine-learning surrogates treat NLTE as a static input--output mapping rather than a kinetic evolution problem. Here, we present a physics-informed Latent Space Dynamics Identification (pLaSDI) framework specifically designed for NLTE atomic kinetics, which captures the time-dependent atomic kinetics of non-equilibrium plasmas through an explicit reduced governing equation. To ensure the physical reliability of the reduced model, we impose physics-informed loss terms that enforce macroscopic consistency, dynamical stability, and convergence to the correct steady state during long-time integration. Applied to tin NLTE population data generated along hydrodynamically modeled temperature--density trajectories relevant to extreme ultraviolet (EUV) lithography plasmas, the model accurately reproduces charge-state evolution and mean charge state with errors below 2\%, achieves speedups of approximately $5\times10^{4}$--$10^{5}$, and remains stable outside the training trajectories by converging toward physically admissible states and the correct steady-state solution under fixed plasma conditions. These results show that careful physics-informed design of the latent dynamics, rather than data fitting alone, is essential for constructing fast, stable, and physically reliable extrapolative surrogates for time-dependent NLTE kinetics.

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

3 major / 1 minor

Summary. The manuscript introduces a physics-informed Latent Space Dynamics Identification (pLaSDI) framework for time-dependent NLTE atomic kinetics. It constructs a reduced-order model in a learned latent space governed by an explicit ODE, with additional physics-informed loss terms enforcing macroscopic consistency, dynamical stability, and convergence to the correct steady state. Applied to tin NLTE population data generated along hydrodynamically modeled temperature-density trajectories relevant to EUV lithography, the model is reported to reproduce charge-state evolution and mean charge with errors below 2%, deliver speedups of 5×10^4–10^5, and remain stable outside the training set by converging to physically admissible states and the correct steady-state solution under fixed plasma conditions.

Significance. If the extrapolation and long-time fidelity claims hold, the work would represent a meaningful advance for radiation-hydrodynamics simulations by replacing expensive NLTE rate-equation solves with a fast, physics-constrained latent ODE surrogate. The explicit emphasis on designing the latent dynamics (rather than pure data fitting) to enforce stability and steady-state behavior addresses a key limitation of existing static ML surrogates for NLTE. The reported speedups and sub-2% errors on relevant tin data indicate practical potential for EUV and fusion applications, provided the physics-informed losses prove sufficient for out-of-distribution trajectories.

major comments (3)
  1. [Abstract] Abstract: the claim that the model 'accurately reproduces charge-state evolution ... with errors below 2%' and 'remains stable outside the training trajectories' is presented without any description of validation protocols, train/test splits, error bars, or sensitivity to the physics-loss weights; these omissions make the quantitative support for the central accuracy and reliability claims difficult to evaluate.
  2. [Abstract] Abstract: the assertion of stability and physical admissibility outside training trajectories is supported only by convergence to the correct steady state under fixed T/n conditions; no quantitative results are given on error growth during integrations longer than the training horizon or on performance when T(t) and n_e(t) follow previously unseen time-varying paths.
  3. [Methods (physics-informed loss terms)] Physics-informed loss terms (described in the methods): the sufficiency of the soft penalties for macroscopic consistency, dynamical stability, and steady-state convergence to keep the latent dynamics faithful to the full NLTE rate equations during long-time integration and extrapolation is asserted but not demonstrated; no ablation studies, loss-weight sensitivity tests, or comparisons with and without individual loss terms are reported.
minor comments (1)
  1. [Abstract] The speedup range 5×10^4–10^5 should be clarified with respect to whether it includes data generation, latent-space projection overhead, or only the core ODE integration.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We are grateful to the referee for their thorough review and valuable suggestions, which have helped us improve the clarity and completeness of our manuscript. We address each of the major comments below in a point-by-point manner. We have made revisions to the manuscript, including additions to the abstract, methods, and results sections, to provide the requested details on validation protocols, quantitative extrapolation results, and ablation studies.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the model 'accurately reproduces charge-state evolution ... with errors below 2%' and 'remains stable outside the training trajectories' is presented without any description of validation protocols, train/test splits, error bars, or sensitivity to the physics-loss weights; these omissions make the quantitative support for the central accuracy and reliability claims difficult to evaluate.

    Authors: We agree that the abstract, due to its brevity, does not detail the validation approach. The full manuscript describes the train/test split on different hydrodynamic trajectories and reports errors as averages over multiple test cases. To improve accessibility, we have revised the abstract to include a short phrase on the validation: 'validated on held-out trajectories with mean errors below 2%'. Error bars from sensitivity to random seeds are now mentioned in the results section and referenced. For physics-loss weight sensitivity, we have added a dedicated paragraph in the methods explaining the selection process and robustness tests. revision: yes

  2. Referee: [Abstract] Abstract: the assertion of stability and physical admissibility outside training trajectories is supported only by convergence to the correct steady state under fixed T/n conditions; no quantitative results are given on error growth during integrations longer than the training horizon or on performance when T(t) and n_e(t) follow previously unseen time-varying paths.

    Authors: The referee correctly notes that the abstract highlights the fixed-condition steady-state test. However, the manuscript body includes demonstrations of stability for time-dependent extrapolations, where the model is integrated on unseen T(t), n_e(t) paths and shown to remain physically admissible without divergence. To strengthen this, we have added quantitative plots of error accumulation over extended time horizons (up to 3 times training length) in the revised results, confirming bounded error growth, and included performance metrics for additional unseen varying paths. These additions provide the requested quantitative support. revision: partial

  3. Referee: [Methods (physics-informed loss terms)] Physics-informed loss terms (described in the methods): the sufficiency of the soft penalties for macroscopic consistency, dynamical stability, and steady-state convergence to keep the latent dynamics faithful to the full NLTE rate equations during long-time integration and extrapolation is asserted but not demonstrated; no ablation studies, loss-weight sensitivity tests, or comparisons with and without individual loss terms are reported.

    Authors: We acknowledge that while the overall performance with the physics-informed losses is shown, explicit ablations were not included in the original submission. We have now conducted ablation experiments removing each loss term individually and compared the resulting models' accuracy, stability, and extrapolation capability. These results, along with sensitivity analysis varying the loss weights by factors of 0.1 to 10, are added to the methods section and a new supplementary figure. The ablations confirm that each term contributes to the observed fidelity, particularly for long-time behavior, thus demonstrating the sufficiency of the soft penalties. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected; derivation remains self-contained

full rationale

The pLaSDI framework learns latent dynamics from NLTE trajectory data while augmenting training with independent physics-informed loss terms that enforce macroscopic consistency, dynamical stability, and steady-state convergence. These losses constitute external constraints drawn from the underlying rate equations rather than redefinitions of the fitted quantities themselves. No step in the presented chain reduces a claimed prediction or uniqueness result to a fitted parameter or self-citation by construction; the reported reproduction errors, speedups, and out-of-sample stability are outcomes of the trained model rather than tautological restatements of the training procedure. The central claim therefore retains independent content beyond its inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the premise that a low-dimensional latent ODE plus three physics loss terms can faithfully represent high-dimensional NLTE rate equations. Free parameters include the latent dimension and the relative weights of the physics losses, both of which are tuned to data. The key domain assumption is that the chosen macroscopic constraints are sufficient to enforce physical admissibility outside the training distribution.

free parameters (2)
  • latent dimension
    Dimensionality of the reduced space; chosen or tuned during model development.
  • physics loss weights
    Relative strengths of the consistency, stability, and steady-state loss terms; adjusted to achieve reported performance.
axioms (1)
  • domain assumption A low-dimensional latent dynamics model can approximate the essential time evolution of NLTE atomic populations.
    Invoked by the choice of LaSDI architecture for the kinetic problem.

pith-pipeline@v0.9.0 · 5554 in / 1583 out tokens · 63217 ms · 2026-05-10T06:50:20.566688+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Cretin—a radiative transfer capability for laboratory plasmas

    Howard A Scott. Cretin—a radiative transfer capability for laboratory plasmas. Journal of Quantitative Spectroscopy and Radiative Transfer , 71(2-6):689–701, 2001

    work page 2001

  2. [2]

    Yechiel Frank and Howard A. Scott. The bands method for tabulating nlte material properties. High Energy Density Physics , 43:100998, 2022

    work page 2022

  3. [3]

    Using tabulated nlte data for hohlraum simulations

    HA Scott, JA Harte, ME Foord, and DT Woods. Using tabulated nlte data for hohlraum simulations. Physics of Plasmas , 29(8), 2022

    work page 2022

  4. [4]

    Scott, Steve Langer, and Yechiel Frank

    Howard A. Scott, Steve Langer, and Yechiel Frank. Radiation-hydrodynamics modeling of lpp euv sources. In Laser-Produced Plasmas. IntechOpen, 2022. A vailable via OSTI as Lawrence Livermore National Laboratory public release

    work page 2022

  5. [5]

    Deep learning for nlte spectral opacities

    G Kluth, KD Humbird, BK Spears, JL Peterson, HA Scott, MV Patel, J Koning, M Marinak, L Divol, and CV Young. Deep learning for nlte spectral opacities. Physics of Plasmas , 27(5), 2020

    work page 2020

  6. [6]

    Vander Wal, Ryan G

    Michael D. Vander Wal, Ryan G. McClarren, and Kelli D. Humbird. Neural network surrogate models for absorptivity and emissivity spectra of multiple elements. Machine Learning: Science and Technology , 3(4):045036, 2022

    work page 2022

  7. [7]

    Transfer learning as a method to reproduce high-fidelity non-local thermodynamic equilibrium opacities in simulations

    Michael D Vander Wal, Ryan G McClarren, and Kelli D Humbird. Transfer learning as a method to reproduce high-fidelity non-local thermodynamic equilibrium opacities in simulations. Journal of Plasma Physics , 89(1):895890103, 2023

    work page 2023

  8. [8]

    Schaeuble, W

    M.-A. Schaeuble, W. E. Lewis, S. Hansen, and T. Nagayama. Using deep learning to develop a fast, high-resolution, optically thin NLTE spectral model. Physics of Plasmas , 32(8):083302, 2025. 25

    work page 2025

  9. [9]

    Physics-informed transformation toward improving the machine-learned nlte models of icf simulations

    Min Sang Cho, Paul E Grabowski, Kowshik Thopalli, Thathachar S Jayram, Michael J Barrow, Jayaraman J Thiagarajan, Rushil Anirudh, Hai P Le, Howard A Scott, Joshua B Kallman, et al. Physics-informed transformation toward improving the machine-learned nlte models of icf simulations. Physical Review Research, 7(2):023150, 2025

    work page 2025

  10. [10]

    Ionization disequilibrium in k-and l-shell ions

    David T Bishel, EV Marley, MB Schneider, DA Liedahl, RF Heeter, ME Foord, GE Kemp, Y Frank, JA Emig, G Pérez-Callejo, et al. Ionization disequilibrium in k-and l-shell ions. Physics of Plasmas , 30(7), 2023

    work page 2023

  11. [11]

    M. S. Cho, A. L. Milder, W. Rozmus, H. P. Le, H. A. Scott, D. T. Bishel, D. Turnbull, S. B. Libby, and M. E. Foord. Reduced model of ionization lag in intense laser-produced plasmas. Phys. Rev. Lett. , 134:185101, May 2025

    work page 2025

  12. [12]

    Ionization dynamics of intense laser-produced argon plasmas revealed by nlte modeling

    Min Sang Cho, AL Milder, W Rozmus, Hai P Le, HA Scott, David T Bishel, D Turnbull, Steve B Libby, and Mark E Foord. Ionization dynamics of intense laser-produced argon plasmas revealed by nlte modeling. Physics of Plasmas , 32(10), 2025

    work page 2025

  13. [13]

    Latent space dynamics learning for stiff collisional-radiative models

    Xuping Xie, Qi Tang, and Xianzhu Tang. Latent space dynamics learning for stiff collisional-radiative models. Machine Learning: Science and Technology , 5(4):045070, 2024

    work page 2024

  14. [14]

    Deep learning surrogate models to solve time-dependent nlte absorption and emission spectra

    Jingsong Zhang, Wengu Chen, Xiaoying Han, Peng Song, and Han Wang. Deep learning surrogate models to solve time-dependent nlte absorption and emission spectra. High Energy Density Physics , 56:101199, 2025

    work page 2025

  15. [15]

    Direct measurements of the ionization potential depression in a dense plasma

    O Ciricosta, SM Vinko, H-K Chung, B-I Cho, CRD Brown, T Burian, J Chalupskỳ, K Engelhorn, R W Falcone, C Graves, et al. Direct measurements of the ionization potential depression in a dense plasma. Physical review letters , 109(6):065002, 2012

    work page 2012

  16. [16]

    Le, Mark Sherlock, and Howard A

    Hai P. Le, Mark Sherlock, and Howard A. Scott. Influence of atomic kinetics on inverse bremsstrahlung heating and nonlocal thermal transport. Phys. Rev. E , 100:013202, Jul 2019

    work page 2019

  17. [17]

    M. S. Cho, H.-K. Chung, M. E. Foord, S. B. Libby, and B. I. Cho. Numerical in- vestigation of nonequilibrium electron effects on the collisional ionization rate in the collisional-radiative model. Phys. Rev. E , 109:045207, Apr 2024

    work page 2024

  18. [18]

    Fries, Xiaolong He, and Youngsoo Choi

    William D. Fries, Xiaolong He, and Youngsoo Choi. Lasdi: Parametric latent space dynamics identification. Computer Methods in Applied Mechanics and Engineering , 399:115436, 2022

    work page 2022

  19. [19]

    glasdi: Parametric physics-informed greedy latent space dynamics identification

    Xiaolong He, Youngsoo Choi, William D Fries, Jonathan L Belof, and Jiun-Shyan Chen. glasdi: Parametric physics-informed greedy latent space dynamics identification. Jour- nal of Computational Physics , 489:112267, 2023. 26

    work page 2023

  20. [20]

    Christophe Bonneville, Youngsoo Choi, Debojyoti Ghosh, and Jonathan L. Belof. Gplasdi: Gaussian process-based interpretable latent space dynamics identification through deep autoencoder. Computer Methods in Applied Mechanics and Engineer- ing, 418:116535, 2024

    work page 2024

  21. [21]

    A comprehensive review of latent space dynamics identification algorithms for intrusive and non-intrusive reduced-order-modeling

    Christophe Bonneville, Xiaolong He, April Tran, Jun Sur Park, William Fries, Daniel A Messenger, Siu Wun Cheung, Yeonjong Shin, David M Bortz, Debojyoti Ghosh, et al. A comprehensive review of latent space dynamics identification algorithms for intrusive and non-intrusive reduced-order-modeling. arXiv preprint arXiv:2403.10748 , 2024

    work page arXiv 2024

  22. [22]

    Mardt, L

    A. Mardt, L. Pasquali, H. Wu, and Frank Noé. Vampnets for deep learning of molecular kinetics. Nature Communications, 9(5), 2018

    work page 2018

  23. [23]

    Lusch, J

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

    work page 2018

  24. [24]

    Takeishi, Y

    N. Takeishi, Y. Kawahara, and T. Yairi. Learning koopman invariant subspaces for dynamic mode decomposition. Advances in neural information processing systems , 30, 2017

    work page 2017

  25. [25]

    Andrievsky, P

    S. Andrievsky, P. Bonifacio, E. Caffau, S. Korotin, M. Spite, F. Spite, L. Sbordone, and A. V. Zhukova. Galactic evolution of copper in the light of nlte computations. Monthly Notices of the Royal Astronomical Society , 473(3):3377–3384, 09 2017

    work page 2017

  26. [26]

    Modern Methods in Collisional-Radiative Modeling of Plasmas

    Yuri Ralchenko. Modern Methods in Collisional-Radiative Modeling of Plasmas . Springer, 2016

    work page 2016

  27. [27]

    Opacity calculation for aluminum, iron, and gold plasmas using flychk code

    Min Sang Cho, K Matsuo, Shuhei Fujioka, Sang June Hahn, and Byoung Ick Cho. Opacity calculation for aluminum, iron, and gold plasmas using flychk code. Journal of Quantitative Spectroscopy and Radiative Transfer , 257:107369, 2020

    work page 2020

  28. [28]

    Vander Wal, Ryan G

    Michael D. Vander Wal, Ryan G. McClarren, and Kelli D. Humbird. Neural network surrogate models for absorptivity and emissivity spectra of multiple elements. Machine Learning with Applications , 8:100308, 2022

    work page 2022

  29. [29]

    J. L. Proctor, S. L. Brunton, and J. N. Kutz. Dynamic mode decomposition with control. SIAM Journal on Applied Dynamical Systems , 15(1):142–161, 2016

    work page 2016

  30. [30]

    Brunton, Joshua L

    Steven L. Brunton, Joshua L. Proctor, and J. Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the National Academy of Sciences , 113(15):3932–3937, 2016

    work page 2016

  31. [31]

    J Schmid

    P. J Schmid. Dynamic mode decomposition of numerical and experimental data. Journal of fluid mechanics , 656:5–28, 2010. 27

    work page 2010

  32. [32]

    Zhang, Y

    Z. Zhang, Y. Shin, and G. Em Karniadakis. Gfinns: Generic formalism informed neu- ral networks for deterministic and stochastic dynamical systems. Philosophical Trans- actions of the Royal Society A: Mathematical, Physical and Engineering Sciences , 380(2229):20210207, 06 2022

    work page 2022

  33. [33]

    TQ Chen, Y

    R. TQ Chen, Y. Rubanova, Jesse B., and D. K Duvenaud. Neural ordinary differential equations. Advances in neural information processing systems , 31, 2018

    work page 2018

  34. [34]

    Longhi, D

    A. Longhi, D. Lathouwers, and Zoltán Perkó. Latent space modeling of parametric and time-dependent pdes using neural odes. Computer Methods in Applied Mechanics and Engineering, 448:118394, 2026

    work page 2026

  35. [35]

    Nathan Kutz, Bingni W

    Urban Fasel, Eurika Kaiser, J. Nathan Kutz, Bingni W. Brunton, and Steven L. Brun- ton. SINDy with control: A tutorial. In 2021 60th IEEE Conference on Decision and Control (CDC) , pages 16–21, 2021

    work page 2021

  36. [36]

    Messenger and David M

    Daniel A. Messenger and David M. Bortz. Weak sindy for partial differential equations. Journal of Computational Physics , 443:110525, 2021

    work page 2021

  37. [37]

    Flychk: Generalized population kinetics and spectral model for rapid spectroscopic analysis for all elements

    H-K Chung, MH Chen, WL Morgan, Yuri Ralchenko, and R W Lee. Flychk: Generalized population kinetics and spectral model for rapid spectroscopic analysis for all elements. High energy density physics , 1(1):3–12, 2005

    work page 2005

  38. [38]

    H.-K. Chung. The How To for FLYCHK . National Institute of Standards and Technol- ogy, 2008. FLYCHK Manual, November 2008

    work page 2008

  39. [39]

    Enhanced extreme ultraviolet conversion efficiency of a 2 𝜇m laser-driven preformed tin-droplet target using short picosecond pre-pulses

    ZY Shi, Y Yuan, WP Wang, YY Ma, XY Sun, N Lin, and YX Leng. Enhanced extreme ultraviolet conversion efficiency of a 2 𝜇m laser-driven preformed tin-droplet target using short picosecond pre-pulses. Physics of Plasmas , 30(4), 2023

    work page 2023

  40. [40]

    Characterization of charge state distributions in near-infrared laser-driven tin plasmas for efficient euv generation

    Jang Hyeob Sohn, Hyun-Kyung Chung, Hyungyu Yu, Inki Jeong, Sang June Hahn, Young-Gui Yoon, Hyyong Suk, and Byoung Ick Cho. Characterization of charge state distributions in near-infrared laser-driven tin plasmas for efficient euv generation. Re- sults in Physics , 75:108340, 2025

    work page 2025

  41. [41]

    Fryxell, K

    B. Fryxell, K. Olson, P. Ricker, F. X. Timmes, M. Zingale, D. Q. Lamb, P. MacNeice, R. Rosner, J. W. Truran, and H. Tufo. Flash: An adaptive mesh hydrodynamics code for modeling astrophysical thermonuclear flashes. The Astrophysical Journal Supplement Series, 131(1):273–334, 2000

    work page 2000

  42. [42]

    Atomic processes modeling of x-ray free electron laser produced plasmas using scfly code

    H-K Chung, BI Cho, O Ciricosta, SM Vinko, JS Wark, and R W Lee. Atomic processes modeling of x-ray free electron laser produced plasmas using scfly code. In Aip conference proceedings, volume 1811. AIP Publishing, 2017

    work page 2017

  43. [43]

    M. S. Cho, H.-K. Chung, M. E. Foord, S. B. Libby, and B. I. Cho. Numerical in- vestigation of nonequilibrium electron effects on the collisional ionization rate in the collisional-radiative model. Physical Review E , 109(4):045207, 2024. 28

    work page 2024