pith. sign in

arxiv: 2604.20143 · v1 · submitted 2026-04-22 · 🧮 math.NA · cs.LG· cs.NA· physics.comp-ph

Machine learning moment closure models for the radiative transfer equation IV: enforcing symmetrizable hyperbolicity in two dimensions

Juntao Huang This is my paper

Pith reviewed 2026-05-10 00:19 UTC · model grok-4.3

classification 🧮 math.NA cs.LGcs.NAphysics.comp-ph
keywords machine learningmoment closureradiative transfer equationsymmetrizable hyperbolicityP_N modelblock-tridiagonal structurenumerical methodstwo dimensions
0
0 comments X

The pith

Machine-learned moment closures for two-dimensional radiative transfer can be made symmetrizably hyperbolic by construction.

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

This paper develops a machine learning approach for closing the moment equations of the radiative transfer equation in two spatial and two angular dimensions. It keeps the lower-order blocks of the classical P_N model fixed and replaces only the highest-order block row with a learned closure. By exploiting the symmetry and block-tridiagonal structure of the P_N coefficient matrices, the authors derive conditions that let the closure be parametrized using a symmetric positive definite matrix and symmetric blocks. This parametrization ensures the full system remains symmetrizably hyperbolic without extra constraints during training. The result is a data-driven model that improves accuracy over the standard P_N approximation while preserving the structural properties needed for well-posedness.

Core claim

The central claim is that the structural properties of the P_N model—symmetric coefficient matrices with block-tridiagonal form—extend to the ML moment model when only the highest block row is modified. This allows a block-diagonal symmetrizer and explicit algebraic conditions on the closure that guarantee symmetrizable hyperbolicity. These conditions naturally parametrize the closure as a symmetric positive definite matrix together with symmetric closure blocks, which can be learned from data.

What carries the argument

The block-diagonal symmetrizer together with algebraic conditions on the highest-order closure blocks, derived from the symmetry and block-tridiagonal form of the P_N coefficient matrices.

If this is right

  • The ML moment system is symmetrizably hyperbolic by construction when the closure satisfies the derived conditions.
  • Training proceeds without additional hyperbolicity penalties or constraints.
  • Numerical tests show improved accuracy over the classical P_N model while hyperbolicity is maintained.
  • Only the highest-order block needs to be learned, leaving the rest of the P_N structure unchanged.

Where Pith is reading between the lines

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

  • The same parametrization strategy could extend to three-dimensional settings or other kinetic equations if analogous structural properties exist.
  • The method might combine with positivity or entropy constraints to produce more robust closures for radiative transfer problems.
  • Similar symmetry-based conditions could apply to moment models in other fields where hyperbolicity must be guaranteed after data-driven modification.

Load-bearing premise

That replacing only the highest-order block row with a learned closure preserves enough of the P_N structure for the same symmetrizer and algebraic conditions to enforce hyperbolicity in the target applications.

What would settle it

A numerical test or Jacobian eigenvalue analysis in which a closure satisfying the symmetry and positive-definiteness conditions produces a system that loses symmetrizable hyperbolicity, or a closure violating the conditions that nonetheless remains hyperbolic.

Figures

Figures reproduced from arXiv: 2604.20143 by Juntao Huang.

Figure 1
Figure 1. Figure 1: Architecture of the hyperbolicity-preserving neural network closure model. The input [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Task 1: Comparison of PN models for the single-mode sine wave initial condition in (4.1). The figure shows the 1D cuts at y = 0 and t = 1 for the zeroth-order moment u0 from P2, P3, P4, P5, P6, and P10. We observe that P2 exhibits a large error, and P3 already shows a sig￾nificant improvement. The other higher-order models P4, P5, and P6 are all nearly indistinguishable from the P10 model. KN = HMy. We the… view at source ↗
Figure 3
Figure 3. Figure 3: Task 1: Comparison for the ML closures and the linear [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Task 2: Visualization of the initial conditions for the generated multi-sine [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Task 2: Hyperparameter sweep for the multi-sine training problem with fixed material. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Task 2: Comparison for the ML closure with [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Task 2: Comparison for the ML closure with [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Task 2: Comparison for the ML closure with [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Task 3: Left: All 100 randomly sampled seeds are shown in the ( [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Task 3: Comparison for the ML closure with [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
read the original abstract

This is our fourth work in the series on machine learning (ML) moment closure models for the radiative transfer equation (RTE). In the first three papers of this series, we considered the RTE in slab geometry in 1D1V (i.e. one dimension in physical space and one dimension in angular space), and introduced a gradient-based ML moment closure [1], then enforced the hyperbolicity through a symmetrizer [2], or together with physical characteristic speeds by learning the eigenvalues of the Jacobian matrix [3]. Here, we extend our framework to the RTE in 2D2V (i.e. two dimensions in physical space and two dimensions in angular space). The main idea is to preserve the leading part of the classical $P_N$ model and modify only the highest-order block row. By analyzing the structural properties of the $P_N$ model, we show that its coefficient matrices are symmetric and admit a block-tridiagonal structure. Then we use this property to introduce a block-diagonal symmetrizer for the ML moment model and derive explicit algebraic conditions on the closure blocks which guarantee the symmetrizable hyperbolicity of the resulting ML system. These conditions lead to a natural parametrization of the closure in terms of a symmetric positive definite matrix together with symmetric closure blocks, which can be learned from data while automatically enforcing symmetrizable hyperbolicity by construction. The numerical results show that the proposed framework improves upon the classical $P_N$ model while maintaining hyperbolicity.

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

0 major / 3 minor

Summary. The manuscript extends the authors' prior work on machine learning moment closures for the radiative transfer equation to the 2D2V setting. It retains the leading blocks of the classical P_N model and replaces only the highest-order closure row with a learned model. Structural analysis of the P_N coefficient matrices establishes their symmetry and block-tridiagonal form, which is used to construct a block-diagonal symmetrizer and derive algebraic conditions on the closure blocks. These conditions yield a parametrization in terms of a symmetric positive definite matrix together with symmetric blocks; the parameters can be learned from data while enforcing symmetrizable hyperbolicity by algebraic construction. Numerical experiments indicate improved accuracy relative to P_N while preserving hyperbolicity.

Significance. If the derivation is correct, the work supplies a systematic, structure-preserving route to machine-learned closures for multi-dimensional hyperbolic moment systems. The explicit algebraic parametrization that guarantees symmetrizable hyperbolicity by construction is a clear strength, removing the need for post-training hyperbolicity checks and building directly on the authors' earlier papers in the series. This approach is likely to be useful for stable numerical schemes in radiative transfer applications where classical P_N closures are insufficient.

minor comments (3)
  1. [Numerical results] Numerical results section: The manuscript should report quantitative error norms (e.g., L2 or L-infinity errors against a reference solution) for the learned closures versus P_N on the test problems to make the claimed improvement concrete and reproducible.
  2. [Abstract and Introduction] The abstract and introduction refer to 'symmetrizable hyperbolicity' without a brief reminder of the precise definition used (e.g., existence of a symmetrizer that renders the system symmetric hyperbolic). Adding one sentence would improve accessibility.
  3. [References] References: The citations to the three preceding papers in the series should include full titles and arXiv identifiers for completeness.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. We appreciate the recognition that the algebraic parametrization enforces symmetrizable hyperbolicity by construction and builds on our prior work in the series.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation begins from the established structural properties (symmetry and block-tridiagonal form) of the classical P_N coefficient matrices in 2D2V, which are analyzed directly in the paper rather than assumed from prior self-work. Explicit algebraic conditions on the highest-order closure blocks are then derived to admit a block-diagonal symmetrizer. These conditions motivate a parametrization (symmetric positive definite matrix plus symmetric blocks) that enforces symmetrizable hyperbolicity by algebraic construction for any learned values inside the set. This is an intentional constraint on the ML model, not a reduction of the claimed result to its own inputs. The leading P_N blocks are taken from standard literature, and numerical experiments serve only as confirmation. No step equates a prediction or first-principles claim to a fitted quantity or self-citation chain by the paper's own equations.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on the symmetry and block-tridiagonal structure of P_N coefficient matrices in 2D, which are used to construct the symmetrizer. The ML parameters are learned but constrained to the symmetric positive definite form.

free parameters (1)
  • entries of symmetric positive definite matrix and symmetric closure blocks
    These are the trainable parameters of the ML model; their values are fitted to data but constrained to maintain the algebraic conditions for hyperbolicity.
axioms (1)
  • domain assumption Coefficient matrices of the classical P_N model in 2D2V are symmetric and admit a block-tridiagonal structure.
    Invoked to derive the block-diagonal symmetrizer and algebraic conditions on the closure.

pith-pipeline@v0.9.0 · 5584 in / 1384 out tokens · 28540 ms · 2026-05-10T00:19:12.807283+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

49 extracted references · 49 canonical work pages

  1. [1]

    Christlieb, and Luke F

    Juntao Huang, Yingda Cheng, Andrew J. Christlieb, and Luke F. Roberts. Machine learning moment closure models for the radiative transfer equation I: Directly learning a gradient based closure.Journal of Computational Physics, 453:110941, 2022

    work page 2022

  2. [2]

    Christlieb, Luke F

    Juntao Huang, Yingda Cheng, Andrew J. Christlieb, Luke F. Roberts, and Wen-An Yong. Machine learning moment closure models for the radiative transfer equation II: Enforcing global hyperbolicity in gradient based closures, 2021

    work page 2021

  3. [3]

    Christlieb, and Luke F

    Juntao Huang, Yingda Cheng, Andrew J. Christlieb, and Luke F. Roberts. Machine learning moment closure models for the radiative transfer equation III: Enforcing hyperbolicity and physical characteristic speeds.Journal of Scientific Computing, 94(1):7, 2023

    work page 2023

  4. [4]

    Pergamon Press, Oxford, UK, 1973

    Gerald C Pomraning.The Equations of Radiation Hydrodynamics. Pergamon Press, Oxford, UK, 1973

    work page 1973

  5. [5]

    Evaluation of quadrature schemes for the discrete ordinates method.Journal of Quantitative Spectroscopy and Radiative Transfer, 84(4):423–435, 2004

    Rainer Koch and Ralf Becker. Evaluation of quadrature schemes for the discrete ordinates method.Journal of Quantitative Spectroscopy and Radiative Transfer, 84(4):423–435, 2004

    work page 2004

  6. [6]

    Optical tomography using the time-independent equation of radiative transfer—part 1: forward model.Journal of Quantitative Spectroscopy and Radiative Transfer, 72(5):691–713, 2002

    Alexander D Klose, Uwe Netz, J¨ urgen Beuthan, and Andreas H Hielscher. Optical tomography using the time-independent equation of radiative transfer—part 1: forward model.Journal of Quantitative Spectroscopy and Radiative Transfer, 72(5):691–713, 2002

    work page 2002

  7. [7]

    Adaptive sparse grid discontinuous galerkin method: review and software implementation.Communications on Applied Mathematics and Computation, 6(1):501–532, 2024

    Juntao Huang, Wei Guo, and Yingda Cheng. Adaptive sparse grid discontinuous galerkin method: review and software implementation.Communications on Applied Mathematics and Computation, 6(1):501–532, 2024

    work page 2024

  8. [8]

    A reduced basis method for radiative transfer equation.Journal of Scientific Computing, 91(1):5, 2022

    Zhichao Peng, Yanlai Chen, Yingda Cheng, and Fengyan Li. A reduced basis method for radiative transfer equation.Journal of Scientific Computing, 91(1):5, 2022

    work page 2022

  9. [9]

    A review of low-rank methods for time-dependent kinetic simulations.Journal of Computa- tional Physics, 538:114191, 2025

    Lukas Einkemmer, Katharina Kormann, Jonas Kusch, Ryan G McClarren, and Jing-Mei Qiu. A review of low-rank methods for time-dependent kinetic simulations.Journal of Computa- tional Physics, 538:114191, 2025

    work page 2025

  10. [10]

    On the kinetic theory of rarefied gases.Communications on pure and applied mathematics, 2(4):331–407, 1949

    Harold Grad. On the kinetic theory of rarefied gases.Communications on pure and applied mathematics, 2(4):331–407, 1949

    work page 1949

  11. [11]

    On the radiative equilibrium of a stellar atmosphere.The Astrophysical Journal, 99:180, 1944

    S Chandrasekhar. On the radiative equilibrium of a stellar atmosphere.The Astrophysical Journal, 99:180, 1944

    work page 1944

  12. [12]

    Relating eddington factors to flux limiters.Journal of Quantitative Spec- troscopy and Radiative Transfer, 31(2):149–160, 1984

    CD Levermore. Relating eddington factors to flux limiters.Journal of Quantitative Spec- troscopy and Radiative Transfer, 31(2):149–160, 1984

    work page 1984

  13. [13]

    Analytic closures for M1 neutrino transport.Monthly Notices of the Royal Astronomical Society, 469(2):1725–1737, 2017

    EM Murchikova, Ernazar Abdikamalov, and Todd Urbatsch. Analytic closures for M1 neutrino transport.Monthly Notices of the Royal Astronomical Society, 469(2):1725–1737, 2017

    work page 2017

  14. [14]

    High-order entropy-based closures for linear transport in slab geometry.Com- munications in Mathematical Sciences, 9(1):187–205, 2011

    Cory D Hauck. High-order entropy-based closures for linear transport in slab geometry.Com- munications in Mathematical Sciences, 9(1):187–205, 2011

    work page 2011

  15. [15]

    High-order entropy-based closures for linear transport in slab geometry ii: A computational study of the optimization problem

    Graham W Alldredge, Cory D Hauck, and Andre L Tits. High-order entropy-based closures for linear transport in slab geometry ii: A computational study of the optimization problem. SIAM Journal on Scientific Computing, 34(4):B361–B391, 2012. 19

    work page 2012

  16. [16]

    Adaptive change of basis in entropy-based moment closures for linear kinetic equations.Journal of Computational Physics, 258:489–508, 2014

    Graham W Alldredge, Cory D Hauck, Dianne P OLeary, and Andr´ e L Tits. Adaptive change of basis in entropy-based moment closures for linear kinetic equations.Journal of Computational Physics, 258:489–508, 2014

    work page 2014

  17. [17]

    PositiveP N closures.SIAM Journal on Scientific Comput- ing, 32(5):2603–2626, 2010

    Cory Hauck and Ryan McClarren. PositiveP N closures.SIAM Journal on Scientific Comput- ing, 32(5):2603–2626, 2010

    work page 2010

  18. [18]

    Robust and accurate filtered spherical harmonics expansions for radiative transfer.Journal of Computational Physics, 229(16):5597–5614, 2010

    Ryan G McClarren and Cory D Hauck. Robust and accurate filtered spherical harmonics expansions for radiative transfer.Journal of Computational Physics, 229(16):5597–5614, 2010

    work page 2010

  19. [19]

    Implicit filteredP N for high- energy density thermal radiation transport using discontinuous galerkin finite elements.Jour- nal of Computational Physics, 321:624–643, 2016

    Vincent M Laboure, Ryan G McClarren, and Cory D Hauck. Implicit filteredP N for high- energy density thermal radiation transport using discontinuous galerkin finite elements.Jour- nal of Computational Physics, 321:624–643, 2016

    work page 2016

  20. [20]

    A new spherical harmonics scheme for multi-dimensional radiation transport I

    David Radice, Ernazar Abdikamalov, Luciano Rezzolla, and Christian D Ott. A new spherical harmonics scheme for multi-dimensional radiation transport I. static matter configurations. Journal of Computational Physics, 242:648–669, 2013

    work page 2013

  21. [21]

    Approximating theM 2 method by the ex- tended quadrature method of moments for radiative transfer in slab geometry.Kinetic & Related Models, 9(2):237, 2016

    Graham W Alldredge, Ruo Li, and Weiming Li. Approximating theM 2 method by the ex- tended quadrature method of moments for radiative transfer in slab geometry.Kinetic & Related Models, 9(2):237, 2016

    work page 2016

  22. [22]

    A nonlinear hyperbolic model for radiative transfer equation in slab geometry.SIAM Journal on Applied Mathematics, 80(6):2388–2419, 2020

    Yuwei Fan, Ruo Li, and Lingchao Zheng. A nonlinear hyperbolic model for radiative transfer equation in slab geometry.SIAM Journal on Applied Mathematics, 80(6):2388–2419, 2020

    work page 2020

  23. [23]

    A nonlinear moment model for radiative transfer equation in slab geometry.Journal of Computational Physics, 404:109128, 2020

    Yuwei Fan, Ruo Li, and Lingchao Zheng. A nonlinear moment model for radiative transfer equation in slab geometry.Journal of Computational Physics, 404:109128, 2020

    work page 2020

  24. [24]

    Deep learning.nature, 521(7553):436–444, 2015

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

    work page 2015

  25. [25]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    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

  26. [26]

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

    work page 2019

  27. [27]

    Solving high-dimensional partial differential equations using deep learning.Proceedings of the National Academy of Sciences, 115(34):8505– 8510, 2018

    Jiequn Han, Arnulf Jentzen, and Weinan E. Solving high-dimensional partial differential equations using deep learning.Proceedings of the National Academy of Sciences, 115(34):8505– 8510, 2018

    work page 2018

  28. [28]

    Uniformly accurate machine learning-based hydrodynamic models for kinetic equations.Proceedings of the National Academy of Sciences, 116(44):21983–21991, 2019

    Jiequn Han, Chao Ma, Zheng Ma, and Weinan E. Uniformly accurate machine learning-based hydrodynamic models for kinetic equations.Proceedings of the National Academy of Sciences, 116(44):21983–21991, 2019

    work page 2019

  29. [29]

    Learning thermodynamically stable and Galilean invariant partial differential equations for non-equilibrium flows.Journal of Non-Equilibrium Thermodynamics, 2021

    Juntao Huang, Zhiting Ma, Yizhou Zhou, and Wen-An Yong. Learning thermodynamically stable and Galilean invariant partial differential equations for non-equilibrium flows.Journal of Non-Equilibrium Thermodynamics, 2021

    work page 2021

  30. [30]

    Conservation-dissipation formalism of irreversible thermodynamics.Journal of Non-Equilibrium Thermodynamics, 40(2):67–74, 2015

    Yi Zhu, Liu Hong, Zaibao Yang, and Wen-An Yong. Conservation-dissipation formalism of irreversible thermodynamics.Journal of Non-Equilibrium Thermodynamics, 40(2):67–74, 2015. 20

    work page 2015

  31. [31]

    A neural network closure for the euler-poisson system based on kinetic simulations.Kinetic and Related Models, 15(1):49– 89, 2021

    Leo Bois, Emmanuel Franck, Laurent Navoret, and Vincent Vigon. A neural network closure for the euler-poisson system based on kinetic simulations.Kinetic and Related Models, 15(1):49– 89, 2021

    work page 2021

  32. [32]

    Machine learning surrogate models for Landau fluid closure.Physics of Plasmas, 27(4):042502, 2020

    Chenhao Ma, Ben Zhu, Xue-Qiao Xu, and Weixing Wang. Machine learning surrogate models for Landau fluid closure.Physics of Plasmas, 27(4):042502, 2020

    work page 2020

  33. [33]

    Deep learning surrogate model for kinetic Landau-fluid closure with collision.AIP Advances, 10(7):075108, 2020

    Libo Wang, XQ Xu, Ben Zhu, Chenhao Ma, and Yi-An Lei. Deep learning surrogate model for kinetic Landau-fluid closure with collision.AIP Advances, 10(7):075108, 2020

    work page 2020

  34. [34]

    Neural network representability of fully ionized plasma fluid model closures.Physics of Plasmas, 27(7):072106, 2020

    Romit Maulik, Nathan A Garland, Joshua W Burby, Xian-Zhu Tang, and Prasanna Bal- aprakash. Neural network representability of fully ionized plasma fluid model closures.Physics of Plasmas, 27(7):072106, 2020

    work page 2020

  35. [35]

    Messenger, B

    Daniel Messenger, Ben Southworth, Hans Hammer, and Luis Chacon. Learning interpretable closures for thermal radiation transport in optically-thin media using wsindy.arXiv preprint arXiv:2510.11840, 2025

    work page arXiv 2025

  36. [36]

    Physics-informed neural networks for solving forward and inverse flow problems via the boltzmann-bgk formulation.Journal of Computational Physics, 447:110676, 2021

    Qin Lou, Xuhui Meng, and George Em Karniadakis. Physics-informed neural networks for solving forward and inverse flow problems via the boltzmann-bgk formulation.Journal of Computational Physics, 447:110676, 2021

    work page 2021

  37. [37]

    Physics-informed neural networks for solving multiscale mode-resolved phonon boltzmann transport equation.Materials Today Physics, 19:100429, 2021

    Ruiyang Li, Eungkyu Lee, and Tengfei Luo. Physics-informed neural networks for solving multiscale mode-resolved phonon boltzmann transport equation.Materials Today Physics, 19:100429, 2021

    work page 2021

  38. [38]

    Physics informed neural networks for simulating radiative transfer.Journal of Quantitative Spectroscopy and Radiative Transfer, 270:107705, 2021

    Siddhartha Mishra and Roberto Molinaro. Physics informed neural networks for simulating radiative transfer.Journal of Quantitative Spectroscopy and Radiative Transfer, 270:107705, 2021

    work page 2021

  39. [39]

    Using neural networks to accelerate the solution of the boltzmann equation.Journal of Computational Physics, 443:110521, 2021

    Tianbai Xiao and Martin Frank. Using neural networks to accelerate the solution of the boltzmann equation.Journal of Computational Physics, 443:110521, 2021

    work page 2021

  40. [40]

    Neural-network based collision operators for the boltzmann equation.Journal of Computational Physics, 470:111541, 2022

    Sean T Miller, Nathan V Roberts, Stephen D Bond, and Eric C Cyr. Neural-network based collision operators for the boltzmann equation.Journal of Computational Physics, 470:111541, 2022

    work page 2022

  41. [41]

    Data-driven construction of a generalized kinetic collision operator from molecular dynamics.Physical Review Letters, 135(18):185101, 2025

    Yue Zhao, Joshua Burby, Andrew Christlieb, and Huan Lei. Data-driven construction of a generalized kinetic collision operator from molecular dynamics.Physical Review Letters, 135(18):185101, 2025

    work page 2025

  42. [42]

    From molecular dynamics to kinetic models: data- driven generalized collision operators in 1d3v plasmas.arXiv preprint arXiv:2603.27828, 2026

    Yue Zhao, Guosheng Fu, and Huan Lei. From molecular dynamics to kinetic models: data- driven generalized collision operators in 1d3v plasmas.arXiv preprint arXiv:2603.27828, 2026

    work page arXiv 2026

  43. [43]

    Data-driven, structure-preserving approximations to entropy-based moment closures for kinetic equations.Communications in Mathematical Sciences, 21(4):885–913, 2023

    William A Porteous, Ming Tse P Laiu, and Cory D Hauck. Data-driven, structure-preserving approximations to entropy-based moment closures for kinetic equations.Communications in Mathematical Sciences, 21(4):885–913, 2023

    work page 2023

  44. [44]

    A structure-preserving surrogate model for the closure of the moment system of the boltzmann equation using convex deep neural networks

    Steffen Schotth¨ ofer, Tianbai Xiao, Martin Frank, and Cory Hauck. A structure-preserving surrogate model for the closure of the moment system of the boltzmann equation using convex deep neural networks. InAIAA Aviation 2021 Forum, page 2895, 2021. 21

    work page 2021

  45. [45]

    Input convex neural networks

    Brandon Amos, Lei Xu, and J Zico Kolter. Input convex neural networks. InInternational Conference on Machine Learning, pages 146–155. PMLR, 2017

    work page 2017

  46. [46]

    Steffen Schotth¨ ofer, M Paul Laiu, Martin Frank, and Cory D Hauck. Structure-preserving neu- ral networks for the regularized entropy-based closure of a linear, kinetic, radiative transport equation.Journal of Computational Physics, 533:113967, 2025

    work page 2025

  47. [47]

    Hyperbolic machine learning moment closures for the bgk equations.Multiscale Modeling & Simulation, 23(1):187–217, 2025

    Andrew J Christlieb, Mingchang Ding, Juntao Huang, and Nicholas A Krupansky. Hyperbolic machine learning moment closures for the bgk equations.Multiscale Modeling & Simulation, 23(1):187–217, 2025

    work page 2025

  48. [48]

    Machine learning-based moment closure model for the linear boltzmann equation with uncertainties.Computer Methods in Applied Mechanics and Engineering, 450:118569, 2026

    Juntao Huang, Liu Liu, Kunlun Qi, and Jiayu Wan. Machine learning-based moment closure model for the linear boltzmann equation with uncertainties.Computer Methods in Applied Mechanics and Engineering, 450:118569, 2026

    work page 2026

  49. [49]

    Starmap—a second order staggered grid method for spherical harmonics moment equations of radiative transfer.ACM Transactions on Mathe- matical Software (TOMS), 41(1):1–28, 2014

    Benjamin Seibold and Martin Frank. Starmap—a second order staggered grid method for spherical harmonics moment equations of radiative transfer.ACM Transactions on Mathe- matical Software (TOMS), 41(1):1–28, 2014. 22

    work page 2014