Pith. sign in

REVIEW 3 major objections 6 minor 34 references

A Hyperbolic Neural Closure for M1 Radiation Transfer

T0 review · 3 major / 6 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read A neural M1 radiation closure can keep real characteristic speeds by learning a symmetrizable flux Jacobian, then recovering the pressure tensor by path integration.

desk verdict Solid Jacobian-level construction for hyperbolic M1 closures with real empirical gains over Levermore; the frozen-feature theory is clean, but the gradient-dependent PDE used in practice is not covered by the proof. read the letter →

arxiv 2607.10364 v1 pith:HBMK4FJC submitted 2026-07-11 cs.LG cs.AIcs.CEcs.NAmath.NA

Bongseok Kim , Jiahao Zhang , Johannes Krotz , Dinshaw Balsara , Ryan McClarren , Guang Lin This is my paper · ORCID
classification cs.LGcs.AIcs.CEcs.NAmath.NA
keywords radiationtransferM1methodmomentclosuredeeplearningdiscontinuousGalerkinhyperbolicityentropysymmetrization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Radiation transport is expensive to simulate in full angular detail, so the M1 method replaces it with equations for energy and flux and needs a closure for the unknown pressure tensor. Classical analytic closures are limited in strongly anisotropic regimes, while unconstrained machine-learned closures can produce complex wave speeds and break numerical solvers. This paper builds the closure by learning the flux Jacobian itself: one network produces symmetric matrix factors and another produces a strictly convex entropy whose Hessian is a positive definite symmetrizer, so the directional Jacobian is similar to a symmetric matrix and has real eigenvalues for each frozen feature set. The pressure is recovered by integrating that Jacobian field along a fixed path, optionally using local gradient features the classical models lack. Experiments on lattice, beam-crossing, and crooked-pipe problems show better pressure accuracy than Levermore and more accurate, stable discontinuous Galerkin solutions.

What carries the argument

Hyperbolic neural closure: flux Jacobians J = S H, with S assembled from a symmetric-matrix network and H = Hess(η_θ) ≻ 0 from an input-convex entropy network; pressure components are reconstructed by midpoint line integration of the resulting Jacobian rows along a prescribed path from the origin.

What would settle it

On a standard M1 benchmark (lattice, beam crossing, or crooked pipe), sample directional Jacobians of the trained closure on the DG solution states and find a nonzero fraction of complex eigenvalues, or observe solver breakdown that wave-speed clipping cannot prevent while a Levermore run remains stable.

Watch

Extended reading notes

Core claim

For the M1 radiative transfer system, parameterizing the closure-induced flux Jacobian as the product of a learned symmetric matrix and the Hessian of a learned strictly convex entropy yields directional Jacobians similar to symmetric matrices, hence real eigenvalues, for each frozen set of auxiliary features; recovering the radiation pressure by numerical path integration of that Jacobian then gives a data-driven closure that is more accurate than classical analytic closures and remains stable when coupled to discontinuous Galerkin solvers.

Load-bearing premise

Real eigenvalues are guaranteed only when the gradient-based features are treated as fixed parameters; once those features depend on the solution itself, the paper does not prove the closed system stays well posed.

Editorial extensions

If this is right

  • DG and other high-order hyperbolic solvers can use learned M1 pressure closures without the usual risk of non-real characteristic speeds from unconstrained regression.
  • Gradient-based auxiliary features can be fed into M1 closures to improve off-diagonal pressure accuracy beyond pure local analytic Eddington factors.
  • The same symmetrizable Jacobian-plus-path-integration pattern can be tried for other unclosed hyperbolic moment systems while keeping the known physical flux structure intact.
  • Training can include an explicit wave-speed penalty so learned closures respect a prescribed maximum propagation speed in practice.

Reading between the lines

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

  • Because hyperbolicity is only frozen-parameter, long-time or strongly discontinuous runs may still need ad-hoc wave-speed clipping or limiters until a full variable-coefficient theory exists.
  • Path dependence of the reconstructed pressure is small in the reported tests, so integrability may be soft-constrained by data rather than hard algebraic curl-free conditions.
  • The same construction could be stress-tested on three-dimensional M1 systems or multi-group radiation, where deriving hand-crafted hyperbolicity conditions becomes harder.
  • If the method generalizes, it offers a template for hybrid physics-ML closures that keep the known conservation structure and only learn the unclosed higher moments.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes a hyperbolic neural (HN) closure for the two-dimensional M1 radiation-transfer moment system. Rather than regressing the radiation pressure tensor directly, the authors parameterize the closure-induced flux Jacobians as J = S H, where S is assembled from a neural network that outputs free coefficients of symmetric matrices and H is the Hessian of a strictly convex input-convex entropy network (with a positive quadratic term). For each frozen feature vector γ the directional Jacobian is similar to a symmetric matrix and therefore has real eigenvalues (Theorem 3.1). The pressure tensor is recovered by midpoint path integration of the learned Jacobian field along a straight-line path in state space, optionally conditioned on local differential features G(u). A soft wave-speed penalty is added during training. On three Monte-Carlo-referenced benchmarks (steady lattice, time-dependent beam crossing, crooked pipe) the HN closure reports substantially lower pressure-tensor MSE/R² than the Levermore closure and yields lower relative L2 errors when coupled to a modal discontinuous Galerkin solver, with sampled directional spectra remaining real.

Significance. Hyperbolicity failure is a genuine obstacle for unconstrained ML moment closures in radiative transfer and kinetic theory; solvers that rely on characteristic information (DG, approximate Riemann solvers, WENO) can break when eigenvalues become complex. The paper’s structural route—entropy symmetrization plus a free symmetric factor, followed by path reconstruction—avoids deriving problem-specific algebraic coefficient conditions and is therefore more systematic than several recent gradient-based hyperbolic ML closures. The architecture constraints (ICNN with α>0, forced first Jacobian rows that preserve the exact energy flux, symmetric S) correctly implement the classical symmetrization theory for the frozen-parameter system. The three-benchmark numerical campaign, including eigenvalue histograms with Im(λ)=0 on large samples and consistent gains over Levermore in both closure and DG solution accuracy, is a concrete empirical contribution. If the frozen-to-gradient gap can be clarified and the reconstruction–Jacobian consistency better controlled, the framework is a useful template for hyperbolic ML closures beyond M1.

major comments (3)
  1. Theorem 3.1 establishes symmetrizable hyperbolicity only for the parametric system with frozen feature vector γ. Immediately after the theorem (Sec. 3.2) the authors state that when γ is instantiated as the actual local differential map G(u), “a rigorous well-posedness analysis of the corresponding full closed PDE system is beyond the scope of the present work.” The Abstract and contribution list nevertheless claim a construction that “guarantees real eigenvalues of the Jacobian associated with ML closures” and a closure that “remains stable in discontinuous Galerkin simulations.” The DG experiments therefore operate outside the proven regime: the closed system is genuinely gradient-dependent, residual wave-speed violations still occur (~1–2 % of states in the lattice tests, requiring clipping of the Lax–Friedrichs face speed, Sec. 4.1), and stability is empirical. The central claim is d
  2. The pressure tensor used by the DG solver is obtained by numerical path integration of a learned Jacobian field that is not constrained to be integrable (Sec. 3.2: “exact integrability conditions are not explicitly enforced”). Consequently the Jacobian of the reconstructed map (E,F)↦P may differ from the parameterized J = S H whose real eigenvalues were guaranteed. Appendix E shows that line versus quadratic paths produce similar pressure accuracy, but does not verify that the discrete flux Jacobian of the reconstructed P remains close to the learned J or retains real spectra. Because the solver evaluates P_θ (and D_θ = Ē^{-1} P_θ) rather than the learned J directly (Eqs. 51–54), this consistency gap is load-bearing for the claim that the hyperbolicity construction protects the discretized system. A quantitative check—e.g., spectra of the Jacobian of the integrated P on the same sample s
  3. Comparisons are restricted to the classical Levermore closure. The introduction cites the Huang et al. series of gradient-based and symmetrizable hyperbolic ML closures for radiative transfer (Refs. [28–31]), yet none of those methods appear as baselines in Sec. 4. Without at least one such comparison (or a clear statement why they are inapplicable to the present M1 setting), it is difficult to assess whether the accuracy and stability gains are due to the symmetrizable Jacobian construction itself or simply to the use of gradient features and a flexible neural map. Adding one competitive ML hyperbolic baseline on at least the lattice or crooked-pipe problem would substantially strengthen the empirical claim.
minor comments (6)
  1. Figure 2 caption and surrounding text contain residual funding-acknowledgment fragments (“273 2053746, DMS-2134209…”) that appear to be copy-paste artifacts; they should be removed.
  2. Notation for the symmetric network is inconsistent: N_S in Table 1 / Algorithm 1 versus N_f in Table 7. Unify the symbol.
  3. In Sec. 3.2 the off-diagonal reconstruction averages two independent path integrals P^{(x)}_{xy} and P^{(y)}_{xy}; a short remark on why this average is preferred over a single consistent definition would help readers.
  4. Fig. 9 caption still says “steady lattice problem” while the section is the beam-crossing problem; correct the caption.
  5. The wave-speed bound c and penalty weight ρ_ws are free training parameters; their chosen values (or a sensitivity check) should be stated explicitly for each experiment.
  6. Appendix B’s determinant/trace comparison is informative; a brief pointer from the main text (e.g., after the lattice closure-accuracy discussion) would make it easier to find.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: hyperbolicity is an explicit structural construction; accuracy is measured against independent Monte Carlo and Levermore baselines.

full rationale

The paper’s load-bearing chain is (i) classical entropy symmetrization (Thm. 2.2 / Thm. 3.1), (ii) a neural parameterization J = S H with S symmetric by assembly and H = Hess(η_θ) ≻ 0 from an ICNN, and (iii) recovery of P by path integration of that Jacobian, trained against Monte Carlo pressure data and evaluated on held-out MC and DG solutions versus Levermore. Real eigenvalues for each frozen feature vector γ follow by similarity to a symmetric matrix by construction; the authors state this as a design guarantee, not as an empirical discovery derived from data. Closure and solution accuracy claims rest on external MC references and classical analytic closures, with standard train/test splits—not on re-labeling fitted constants as predictions. Related citations (SymCLaw, Huang et al., Dafermos/Tadmor) supply background; none is a self-authored uniqueness theorem that forces the present result. The known gap that hyperbolicity is proven only for frozen γ (not the full gradient-dependent PDE) is a scope/correctness limitation, not a circular reduction of outputs to inputs. No self-definitional loop, fitted-as-prediction, or load-bearing self-citation chain is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central claim rests on classical entropy symmetrization, the standard M1 truncation of radiative transfer, and a collection of training/architecture choices that define the learned map. No new physical particles or forces are postulated; the invented object is the algorithmic closure itself. Free parameters are the usual neural and regularization knobs plus the soft wave-speed bound used in training and clipping.

free parameters (5)
  • Neural network weights Θ (η_θ, N_S, N_A)
    All closure accuracy is obtained by supervised fitting of ~24,202 parameters to Monte Carlo pressure data; the reported MSE/R² are properties of this fit.
  • Quadratic entropy coefficient α > 0
    Hand-chosen strict-convexity term in the ICNN entropy that guarantees H ≻ 0; not derived from radiation physics.
  • Wave-speed bound c and penalty weight ρ_ws
    Training regularizer and DG clipping threshold; residual 1–2% violations show the bound is an operational choice, not a theorem.
  • Quadrature count N_q and integration path (line vs quadratic β)
    Reconstruction of P from ∇P depends on discrete path integration choices (N_q=4 in experiments); path independence is not enforced.
  • Network architecture hyperparameters (widths [128], [128,128], [64,64], activations)
    Capacity and inductive bias are selected by the authors; results are conditional on this architecture family.
assumptions (4)
  • standard math Existence of a strictly convex entropy implies hyperbolicity of a first-order conservation law via symmetrization (classical Godunov–Mock / Dafermos theory).
    Invoked as Thm. 2.2 and extended to frozen parameters in Thm. 3.1; standard continuum PDE theory.
  • domain assumption The M1 two-moment truncation with local pressure map P(E,F) is an adequate reduced model of the RTE for the regimes studied.
    Sec. 3.1 derives the moment system and adopts the standard M1 constitutive form; validity of low-order moments under strong anisotropy is a modeling assumption.
  • ad hoc to paper Freezing gradient features γ = G(u) is sufficient to transfer hyperbolicity conclusions to the practical gradient-dependent closure used in DG.
    Authors state full well-posedness of the closed PDE is beyond scope (Sec. 3.2); this is the load-bearing modeling leap from Thm. 3.1 to the implemented solver.
  • domain assumption Monte Carlo angular solutions provide ground-truth pressure tensors for supervised training and error metrics.
    All quantitative claims in Sec. 4 compare against MC references with stated particle counts; statistical MC error is not propagated into reported R².
invented entities (1)
  • Hyperbolic neural (HN) closure: Jacobian field J=SH plus path-integrated pressure with optional G(u)
    purpose: Replace analytic Eddington factors with a learned, structure-preserving map that keeps real characteristic speeds.
    The entity is an algorithmic construction, not a new physical field; independent evidence is limited to the paper's own MC/DG experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Hyperbolic Neural Closure for M1 Radiation Transfer." pith.science (2026). https://pith.science/paper/HBMK4FJC

@misc{pith2026260710364,
  author       = {Pith},
  title        = {Pith review of: A Hyperbolic Neural Closure for M1 Radiation Transfer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HBMK4FJC}},
  note         = {Machine review of arXiv:2607.10364}
}
read the original abstract

In radiation transfer simulations, an M1 method achieves substantial computational savings by replacing the full angular transport equation with a low-order moment system. Because this reduced system is not closed, a closure model is required to represent the unknown higher-order moments using lower-order moments. While machine learning (ML)-based closures can improve accuracy beyond classical analytic closures, unconstrained learned closures may produce non-real characteristic speeds and consequently cause numerical solver breakdown. To guarantee real eigenvalues of the Jacobian associated with ML closures, we propose a hyperbolic neural closure for the M1 radiative transfer system. Rather than directly predicting closure terms, we parameterize the Jacobian through two neural networks: (i) a symmetric matrix network and (ii) a strictly convex entropy network whose Hessian defines a positive definite symmetrizer. These components are combined to yield a Jacobian that is similar to a symmetric matrix, thereby ensuring real eigenvalues. The closure is then reconstructed by numerical integration of the learned Jacobian field along a prescribed integration path. Numerical experiments show that the proposed closure not only achieves higher closure accuracy than classical analytic closures, but also improves solution accuracy and remains stable in discontinuous Galerkin simulations for radiative transfer problems.

Figures

Figures reproduced from arXiv: 2607.10364 by the authors.

Figure 1
Figure 1. Components of the proposed hyperbolic neural closure for the two-dimensional M1 system [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Lattice benchmark configuration on a 7 × 7 square domain with (x, y) ∈ [0, 7] × [0, 7]. The blue filled cells are the obstacle (Obs) regions, with (σa, σs, Q) = (10, 0, 0). The white cells are the background (Bg) regions, with (σa, σs, Q) = (0, 1, 0). The red striped central cell is the source region, with (σa, σs, Q) = (0, 1, 1). Each highlighted block has size 1 × 1 [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Training loss histories of the HN closure, showing the component losses for [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Comparison between reference and predicted closure outputs [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]
Figure 5
Figure 5. Figure 5: Eigenvalue histograms of the directional Jacobian [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: Comparison across four material settings of the checkerboard problem, where Obs denotes [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Beam-crossing configuration on a 7 × 7 square domain Ω = [0, 7] × [0, 7], where x ∈ [0, 7] and y ∈ [0, 7]. Material coefficients: σs = 0 everywhere, σa = 10 in the boundary frame (x < 0.5 or x > 6.5, or y < 0.5 or y > 6.5), and σa = 0.02 in the interior. The problem is…
Figure 8
Figure 8. Figure 8: Training loss histories of the HN closure for the beam crossing problem, showing the [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: Scatter plots comparing reference and predicted closure outputs [PITH_FULL_IMAGE:figures/full_fig_p026_9.png]
Figure 10
Figure 10. Figure 10: Eigenvalue histograms of the directional Jacobian [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]
Figure 11
Figure 11. Figure 11: Beam crossing comparison at four time instances [PITH_FULL_IMAGE:figures/full_fig_p027_11.png]
Figure 12
Figure 12. Figure 12: Crooked pipe configuration on a 7 × 7 square domain Ω = [0, 7] × [0, 7], where x ∈ [0, 7] and y ∈ [0, 7]. The blue filled region is absorbing, with σa = 10 and σs = 0. The gray pipe region is scattering, with σa = 0 and σs = 1. A unit square source patch, shown in red…
Figure 13
Figure 13. Figure 13: Training loss histories of the HN closure, showing the component losses for [PITH_FULL_IMAGE:figures/full_fig_p030_13.png]
Figure 14
Figure 14. Figure 14: Actual vs predicted scatter comparison of closure outputs [PITH_FULL_IMAGE:figures/full_fig_p031_14.png]
Figure 15
Figure 15. Figure 15: Eigenvalue histograms of the directional Jacobian [PITH_FULL_IMAGE:figures/full_fig_p031_15.png]
Figure 16
Figure 16. Figure 16: Solution comparison for the crooked pipe problem on a [PITH_FULL_IMAGE:figures/full_fig_p032_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 1 linked inside Pith

  1. [1]

    González, E

    M. González, E. Audit, and P. Huynh. HERACLES: a three-dimensional radi- ation hydrodynamics code.Astronomy & Astrophysics, 464(2):429–435, 2007

    2007

  2. [2]

    J. A. Fleck and J. D. Cummings. An implicit Monte Carlo scheme for calcu- lating time and frequency dependent nonlinear radiation transport.Journal of Computational Physics, 8(3):313–342, 1971

    1971

  3. [3]

    Cahalan, Lazaros Oreopoulos, Alexander Marshak, K

    Robert F. Cahalan, Lazaros Oreopoulos, Alexander Marshak, K. Franklin Evans, Anthony B. Davis, Robert Pincus, Ken H. Yetzer, Bernhard Mayer, Roger Davies, Thomas P. Ackerman, Howard W. Barker, Eugene E. Clothiaux, Robert G. Ellingson, Michael J. Garay, Evgueni Kassianov, Stefan Kinne, An- dreas Macke, William O’Hirok, Philip T. Partain, Sergei M. Prigarin...

    2005

  4. [4]

    A. D. Klose, U. Netz, J. Beuthan, and A. H. Hielscher. Optical tomography us- ing the time-independent equation of radiative transfer. part 1: Forward model. Journal of Quantitative Spectroscopy and Radiative Transfer, 72(5):691–713, 2002

    2002

  5. [5]

    Chai, HaeOk S

    John C. Chai, HaeOk S. Lee, and Suhas V. Patankar. Finite volume method for radiation heat transfer.Journal of Thermophysics and Heat Transfer, 8(3):419– 425, 1994

    1994

  6. [6]

    David Levermore

    C. David Levermore. Moment closure hierarchies for kinetic theories.Journal of Statistical Physics, 83(5):1021–1065, 1996

    1996

  7. [7]

    E. M. Murchikova, E. Abdikamalov, and T. Urbatsch. Analytic closures for M1 neutrino transport.Monthly Notices of the Royal Astronomical Society, 469(2):1725–1737, 2017. 34

    2017

  8. [8]

    C. D. Levermore. Relating eddington factors to flux limiters.Journal of Quan- titative Spectroscopy and Radiative Transfer, 31:149–160, 1984

    1984

Show all 34 references
  1. [9]

    G. N. Minerbo. Maximum entropy eddington factors.Journal of Quantitative Spectroscopy and Radiative Transfer, 20:541–545, 1978

    1978

  2. [10]

    Approximate models for radiative transfer.Bulletin of the Insti- tute of Mathematics Academia Sinica (New Series), 2(2):409–432, 2007

    Martin Frank. Approximate models for radiative transfer.Bulletin of the Insti- tute of Mathematics Academia Sinica (New Series), 2(2):409–432, 2007

    2007

  3. [11]

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

    2011

  4. [12]

    Dafermos.Hyperbolic Conservation Laws in Continuum Physics

    Constantine M. Dafermos.Hyperbolic Conservation Laws in Continuum Physics. Springer, 2005

    2005

  5. [13]

    Godlewski and P.-A

    E. Godlewski and P.-A. Raviart.Numerical Approximation of Hyperbolic Sys- tems of Conservation Laws. Springer, 2013

    2013

  6. [14]

    W. H. Reed and T. R. Hill. Triangular mesh methods for the neutron transport equation. Technical report, Los Alamos Scientific Laboratory, 1973

    1973

  7. [15]

    Cockburn and C.-W

    B. Cockburn and C.-W. Shu. TVB runge–kutta local projection discontinuous galerkin finite element method for conservation laws II: General framework. Mathematics of Computation, 52:411–435, 1989

    1989

  8. [16]

    Cockburn, S

    B. Cockburn, S. Hou, and C.-W. Shu. TVB runge–kutta local projection dis- continuous galerkin finite element method for conservation laws IV: The multi- dimensional case.Journal of Computational Physics, 54:545–581, 1990

    1990

  9. [17]

    Cockburn and C.-W

    B. Cockburn and C.-W. Shu. The runge–kutta discontinuous galerkin method for conservation laws V: Multidimensional systems.Journal of Computational Physics, 141:199–224, 1998

    1998

  10. [18]

    D. S. Balsara, S. Garain, V. Florinski, and W. Boscheri. An efficient class of WENO schemes with adaptive order for unstructured meshes.Journal of Computational Physics, 404, 2020

    2020

  11. [19]

    D. S. Balsara, J. Li, and G. Montecinos. An efficient, second order accurate, universal generalized riemann problem solver based on the HLLI riemann solver. Journal of Computational Physics, 375:1238–1269, 2018. 35

    2018

  12. [20]

    D. S. Balsara and D. Bhoriya. Multidimensional HLLI generalized riemann problem solver for conservation laws: The two-dimensional case for structured meshes.Computers and Fluids, 301:106791, 2025

    2025

  13. [21]

    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, Zhiwen 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

    2019

  14. [22]

    Scoggins, Jiequn Han, and Marc Massot

    James B. Scoggins, Jiequn Han, and Marc Massot. Machine learning moment closures for accurate and efficient simulation of polydisperse evaporating sprays. InAIAA Scitech 2021 Forum, page 1786, 2021

    2021

  15. [23]

    A neural network closure for the euler–poisson system based on kinetic simulations.Kinetic and Related Models, 2022

    Loic Bois et al. A neural network closure for the euler–poisson system based on kinetic simulations.Kinetic and Related Models, 2022

    2022

  16. [24]

    Learning invariance preserving moment closure model for boltzmann–bgk equation.Communications in Mathematics and Statistics, 2023

    Zhengyi Li, Bin Dong, and Yanli Wang. Learning invariance preserving moment closure model for boltzmann–bgk equation.Communications in Mathematics and Statistics, 2023

    2023

  17. [25]

    Porteous, Ming Tse P

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

    2023

  18. [26]

    Steffen Schotthofer, Tianbai Xiao, Martin Frank, and Cory D. 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

    2021

  19. [27]

    SteffenSchotthofer, M.Paul Laiu, MartinFrank, andCory D.Hauck. Structure- preserving neural networks for the regularized entropy-based closure of a lin- ear, kinetic, radiative transport equation.Journal of Computational Physics, 533:113967, 2025

    2025

  20. [28]

    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

    2022

  21. [29]

    Huang, Y

    J. Huang, Y. Cheng, A. J. Christlieb, and L. F. Roberts. Machine learning moment closure models for the radiative transfer equation II: Enforcing global 36 hyperbolicity in gradient based closures.Multiscale Modeling & Simulation, 21(2):489–512, 2023

    2023

  22. [30]

    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

    2023

  23. [31]

    Machine learning moment closure models for the radiative trans- fer equation iv: Enforcing symmetrizable hyperbolicity in two dimensions, 2026

    Juntao Huang. Machine learning moment closure models for the radiative trans- fer equation iv: Enforcing symmetrizable hyperbolicity in two dimensions, 2026. arXiv:2604.20143

    2026 arXiv

  24. [32]

    Entropy stable schemes

    Eitan Tadmor. Entropy stable schemes. InHandbook of Numerical Analysis, volume 17, pages 467–493. Elsevier, 2016

    2016

  25. [33]

    Parametric hyperbolic conservation laws: A unified framework for conservation, entropy stability, and hyperbolicity.arXiv preprint arXiv:2601.21080, 2026

    Lizuo Liu, Lu Zhang, and Anne Gelb. Parametric hyperbolic conservation laws: A unified framework for conservation, entropy stability, and hyperbolicity.arXiv preprint arXiv:2601.21080, 2026

    2026

  26. [34]

    A hybrid monte carlo, discontinuous galerkin method for linear kinetic transport equations.Journal of Computational Physics, 514:113253, 2024

    Johannes Krotz, Cory D Hauck, and Ryan G McClarren. A hybrid monte carlo, discontinuous galerkin method for linear kinetic transport equations.Journal of Computational Physics, 514:113253, 2024. Appendix A. Sensitivity analysis Toexaminetherelativeimportanceoftheinputvariables...

    2024

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.