pith. sign in

arxiv: 2606.04878 · v1 · pith:ACVUTDITnew · submitted 2026-06-03 · ⚛️ physics.comp-ph · physics.plasm-ph

Sparse and low-rank kinetic distribution estimation

Georgii Oblapenko , Lambert Theisen , Rostislav-Paul Wilhelm , Michael Herty , Manuel Torrilhon This is my paper

Pith reviewed 2026-06-28 03:26 UTC · model grok-4.3

classification ⚛️ physics.comp-ph physics.plasm-ph
keywords kinetic distributionssparsitylow-rank decompositionentropic quadratureVlasov-Maxwellmoment preservationmemory-efficient storagehigh-dimensional data
0
0 comments X

The pith

An extension to entropic quadrature enforces sparsity while a new low-rank decomposition preserves moments for memory-efficient kinetic distribution storage.

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

The paper develops techniques for storing high-dimensional kinetic distributions with reduced memory use while retaining key properties in a kinetic theory setting. It extends the entropic quadrature method to enforce sparsity and introduces a low-rank decomposition that keeps moment information intact. These are demonstrated on model kinetic distributions and on data from high-resolution Vlasov-Maxwell simulations. A sympathetic reader would care because full representations of such distributions consume prohibitive memory in large-scale computations. If the claims hold, the approaches make repeated storage and manipulation of these distributions feasible without discarding essential physical content.

Core claim

The paper establishes that extending entropic quadrature to enforce sparsity, combined with a new low-rank decomposition that preserves moment information, permits memory-efficient storage of high-dimensional kinetic distributions while retaining essential features, as shown through application to model cases and Vlasov-Maxwell simulation outputs.

What carries the argument

The sparsity-enforcing extension to entropic quadrature together with the moment-preserving low-rank decomposition, which together reduce storage requirements for kinetic distributions.

If this is right

  • High-dimensional distributions arising in Vlasov-Maxwell simulations can be stored with substantially lower memory.
  • Moment information remains available for use in the reduced representations.
  • The same reductions apply both to constructed model distributions and to data extracted from actual kinetic simulations.
  • Key physical features of the original distributions are retained after the sparsity and low-rank steps.

Where Pith is reading between the lines

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

  • The same storage reductions could apply to high-dimensional data in other computational physics domains that rely on distribution functions.
  • Embedding the methods inside existing kinetic codes would allow simulations at higher resolution or in more dimensions than current memory limits permit.
  • Systematic checks on conservation properties beyond moments would clarify the range of physical regimes where the approximations remain reliable.

Load-bearing premise

Enforcing sparsity and applying the low-rank decomposition will retain all necessary physical features of the distributions without unacceptable accuracy loss.

What would settle it

A side-by-side run of a Vlasov-Maxwell simulation in which the sparse or low-rank stored distribution produces visibly different moment evolution or instability compared with the unreduced reference solution.

Figures

Figures reproduced from arXiv: 2606.04878 by Georgii Oblapenko, Lambert Theisen, Manuel Torrilhon, Michael Herty, Rostislav-Paul Wilhelm.

Figure 1
Figure 1. Figure 1: Relative L2 error in the VDF (left) and absolute error in next predicted mo￾ments (right) as a function of the fraction of used degrees of freedom, Maxwell-Boltzmann distribution. in the SPEQ approach, values of g that are smaller than ε = 10−7 were set to 0; a parameter study of the impact of the choice of ε on the errors and solution sparsity is performed for the Mott–Smith distribution. 4.1.1. Maxwell–B… view at source ↗
Figure 2
Figure 2. Figure 2: Slices along the vz axis (k = 20) of the hidden truth distribution and distribu￾tions reconstructed using the SPEQ method for various values of λ, Maxwell-Boltzmann distribution. moments, which are odd, remain zero almost to machine precision. Since the Maxwell–Boltzmann distribution can be decomposed exactly into a product of rank-1 tensors, the LRMR method exhibits virtually no error for any rank r ≥ 1. … view at source ↗
Figure 3
Figure 3. Figure 3: Relative L2 error in the VDF (left) and absolute error in next predicted moments (right) as a function of the fraction of used degrees of freedom, Druyvesteyn distribution. for the reduction in the values of the higher-order moments caused by the induced sparsity in the distribution tails. For very sparse representations, this leads to the “hollowing-out” seen above. 4.1.2. Druyvesteyn distribution Next, w… view at source ↗
Figure 4
Figure 4. Figure 4: Slices along the vz axis (k = 20) of the hidden truth distribution and distributions reconstructed using the SPEQ method for various values of λ, Druyvesteyn distribution. are zero, and since the symmetry is well-retained regardless of the values of λ, this is reflected in the very low absolute errors in the predicted values of the unknown moments, as seen on the right-hand side of 3. The same holds for th… view at source ↗
Figure 5
Figure 5. Figure 5: Slices along the vz axis (k = 20) of the hidden truth distribution and distributions reconstructed using the LRMR method for various ranks r, Druyvesteyn distribution. 4.1.3. Mott–Smith distribution Finally, we consider a 3-dimensional version of the Mott–Smith distribu￾tion [28] describing the velocity distribution before, inside, and after a shock￾wave. Originally develop to produce a 1-dimensional solut… view at source ↗
Figure 6
Figure 6. Figure 6: Relative L2 error in the VDF (left) and relative error in next predicted moments (right) as a function of the fraction of used degrees of freedom, Mott-Smith distribution. where α = 1 1 + exp(x) , (51) n1 = 1, n2 = M2 γ + 1 2 + M2 (γ − 1) (52) v1 = (√ γM, 0, 0), v2 =  1 n2 √ γM, 0, 0  , (53) T1 = 1, T2 = 1 − γ + 2γM2 1 + γ 1 n2 . (54) Here x is the position along the streamline and defines the mixing of … view at source ↗
Figure 7
Figure 7. Figure 7: Relative L2 error in the VDF (left) and relative error in next predicted moments (right) of total order 7 as a function of the fraction of used degrees of freedom for different values of cut-off threshold parameter ε, Mott-Smith distribution [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Slices along the vz axis (k = 25) of the hidden truth distribution and recon￾structed distributions for various values of λ, Mott-Smith distribution. to a reduction in both VDF reconstruction error and error in the next pre￾dicted moments (as seen on the right subplot), consistent with the behaviour of the non-sparse entropy minimization-based approach [40]. It should be noted that even for high degrees of… view at source ↗
Figure 9
Figure 9. Figure 9: Slices along the vz axis (k = 25) of the hidden truth distribution and recon￾structed distributions for various values of λ, Mott-Smith distribution. predicted moments remain quite small, on the order of 1%, and only a fac￾tor of 2 larger than the errors in the non-sparse solution. For the LRMR method, the role of the rank is not as clear, with the choice of r not having a significant impact on the solutio… view at source ↗
Figure 10
Figure 10. Figure 10: Results of the beam-driven instability simulation produced by NuFI. Different [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Slices along the vz axis (k = 40) of the hidden truth distribution and dis￾tributions reconstructed using SPEQ with entropy minimization for various values of λ, beam-driven instability at t = 300; reference NuFI solution in the upper left. tions triggers the formation of magnetic islands. The configuration space is set to [0, 2π]×[0, 2π] and the numerical velocity space is chosen as [−30, 30]× [−30, 30] … view at source ↗
Figure 12
Figure 12. Figure 12: Slices along the vz axis (k = 40) of the hidden truth distribution and distri￾butions reconstructed using SPEQ with minimization of Kullback-Leibler divergence w.r.t the original data for various values of λ, beam-driven instability at t = 300; reference NuFI solution in the upper left. retaining the very general shape of the velocity distribution function and losing physically significant details. Theref… view at source ↗
Figure 13
Figure 13. Figure 13: Slices along the vz axis (k = 32) of the hidden truth distribution and distri￾butions reconstructed using SPEQ minimization of Kullback-Leibler divergence w.r.t the original data for various values of λ, beam-driven instability at t = 300; reference NuFI solution in the upper left. tails of the distribution. For the low-rank distributions recovered with the LRMR method, we plot slightly different slices, … view at source ↗
Figure 14
Figure 14. Figure 14: Slices along the vz axis (k = 64) of the hidden truth distribution and distribu￾tions reconstructed using the LRMR method for various ranks r, beam-driven instability at t = 300; reference NuFI solution in the upper left. theory, given that we know the underlying distribution, we can construct moment measurement matrices of arbitrary order, although this leads to significant usage of machine memory and hi… view at source ↗
Figure 15
Figure 15. Figure 15: Slices along the vz axis (k = 40) of the hidden truth distribution and distribu￾tions reconstructed using the LRMR method for various ranks r, beam-driven instability at t = 300; reference NuFI solution in the upper left. 10−1 100 101 102 % of D.o.F. 10−14 10−11 10−8 10−5 10−2 101 || fpred − ftrue|| 2 || ftrue|| 2 SPEQ, M.E., mmax = 4 SPEQ, M.E., mmax = 6 SPEQ, K.L., mmax = 4 SPEQ, K.L., mmax = 6 LRMR, mm… view at source ↗
Figure 16
Figure 16. Figure 16: Relative L2 error in the VDF (left) and absolute error in next predicted mo￾ments (right) as a function of the fraction of used degrees of freedom, NuFI simulation results at t = 100. constraints; thus, additional data- or physics-informed priors are needed to achieve better solution accuracy. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Relative L2 error in the VDF (left) and absolute error in next predicted mo￾ments (right) as a function of the fraction of used degrees of freedom, NuFI simulation results at t = 200. 10−1 100 101 102 % of D.o.F. 10−14 10−11 10−8 10−5 10−2 || fpred − ftrue|| 2 || ftrue|| 2 SPEQ, M.E., mmax = 4 SPEQ, M.E., mmax = 6 SPEQ, K.L., mmax = 4 SPEQ, K.L., mmax = 6 LRMR, mmax = 4 LRMR, mmax = 6 LRMR, mmax = 10 10−1… view at source ↗
Figure 18
Figure 18. Figure 18: Relative L2 error in the VDF (left) and absolute error in next predicted mo￾ments (right) as a function of the fraction of used degrees of freedom, NuFI simulation results at t = 300. For the SPEQ method, conservation of a larger number of moments also has little impact on the solution quality. For the entropy minimization-based approach, the errors are in general large, and it is likely that incorporatio… view at source ↗
read the original abstract

In this paper, we consider methods that allow for memory-efficient storage of high-dimensional distributions and retain certain key features thereof, specifically in a kinetic theory context. We propose an extension to the entropic quadrature method that allows for enforcing sparsity, and propose a new low-rank decomposition approach that ensures preservation of moment information. The methods are applied to several model kinetic distributions, as well as to distributions obtained from high-resolution kinetic simulations of the Vlasov--Maxwell system.

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

1 major / 0 minor

Summary. The paper proposes an extension to the entropic quadrature method that enforces sparsity and introduces a new low-rank decomposition that preserves moment information. These are intended to enable memory-efficient storage of high-dimensional kinetic distributions while retaining key features, with applications to model distributions and outputs from Vlasov-Maxwell simulations.

Significance. If the sparsity enforcement and low-rank approach demonstrably retain physical features such as moments with acceptable accuracy loss, the work would offer practical tools for memory reduction in high-dimensional kinetic simulations common in plasma physics.

major comments (1)
  1. [Abstract] Abstract: no derivation details, error analysis, or quantitative results are provided, so the central claims about preservation of features and memory efficiency cannot be evaluated.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their comments on our manuscript. We respond to the major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: no derivation details, error analysis, or quantitative results are provided, so the central claims about preservation of features and memory efficiency cannot be evaluated.

    Authors: Abstracts are intentionally concise overviews and do not contain detailed derivations, error analyses, or quantitative results; these elements are provided in the body of the manuscript. The extension of entropic quadrature for sparsity, the moment-preserving low-rank decomposition, associated error analysis, and quantitative results on feature preservation and memory efficiency are presented in Sections 3–5, with applications to model distributions and Vlasov–Maxwell data. revision: no

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes an extension to entropic quadrature for sparsity enforcement and a new low-rank decomposition to preserve moments in kinetic distributions from Vlasov-Maxwell simulations. These are presented as methodological contributions applied to model distributions and simulation outputs. No self-definitional steps, fitted inputs renamed as predictions, load-bearing self-citations, or ansatzes smuggled via prior work are identifiable from the provided description or abstract. The derivation chain relies on external techniques (entropic quadrature) without reducing the central claims to tautological fits or self-referential definitions. The result is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no information on free parameters, axioms, or invented entities; ledger left empty pending full text.

pith-pipeline@v0.9.1-grok · 5605 in / 959 out tokens · 32963 ms · 2026-06-28T03:26:37.681451+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

40 extracted references · 9 canonical work pages · 2 internal anchors

  1. [1]

    Moment closure approxi- mations of the Boltzmann equation based onφ-divergences

    Abdelmalik, M., Van Brummelen, E., 2016. Moment closure approxi- mations of the Boltzmann equation based onφ-divergences. Journal of Statistical Physics 164, 77–104

    2016

  2. [2]

    SomeQuestionsintheTheoryofMoments

    Aheizer, N., Krein, M., 1962. SomeQuestionsintheTheoryofMoments. volume 2 ofTranslations of Mathematical Monographs. American Math- ematical Society

    1962

  3. [3]

    A parallel low-rank solver for the six-dimensional Vlasov–Maxwell equations

    Allmann-Rahn, F., Grauer, R., Kormann, K., 2022. A parallel low-rank solver for the six-dimensional Vlasov–Maxwell equations. Journal of Computational Physics 469, 111562

    2022

  4. [4]

    Low-rank tensor methods for partial differential equations

    Bachmayr, M., 2023. Low-rank tensor methods for partial differential equations. Acta Numerica 32, 1–121

    2023

  5. [5]

    A gallery of maximum- entropy distributions: 14 and 21 moments

    Boccelli, S., Giroux, F., McDonald, J.G., 2024. A gallery of maximum- entropy distributions: 14 and 21 moments. Journal of Statistical Physics 191, 39

    2024

  6. [6]

    Entropic quadrature for moment ap- proximations of the Boltzmann-BGK equation

    Böhmer, N., Torrilhon, M., 2020. Entropic quadrature for moment ap- proximations of the Boltzmann-BGK equation. Journal of Computa- tional Physics 401, 108992

    2020

  7. [7]

    Modeling and simulating the spatial spread of an epidemic through multiscale kinetic transport equations

    Boscheri, W., Dimarco, G., Pareschi, L., 2021. Modeling and simulating the spatial spread of an epidemic through multiscale kinetic transport equations. Mathematical Models and Methods in Applied Sciences 31, 1059–1097

    2021

  8. [8]

    Enhancing sparsity by reweightedl 1 minimization

    Candes, E.J., Wakin, M.B., Boyd, S.P., 2008. Enhancing sparsity by reweightedl 1 minimization. Journal of Fourier analysis and applications 14, 877–905

    2008

  9. [9]

    Rarefied gas dynamics: from basic concepts to actual calculations

    Cercignani, C., 2000. Rarefied gas dynamics: from basic concepts to actual calculations. volume 21. Cambridge university press

    2000

  10. [10]

    Introduction to plasma physics and controlled fusion

    Chen, F., 2015. Introduction to plasma physics and controlled fusion. Springer. 35

    2015

  11. [11]

    Numerical solution of the Boltzmann equation with S-model collision integral using tensor decompositions

    Chikitkin, A.V., Kornev, E.K., Titarev, V.A., 2021. Numerical solution of the Boltzmann equation with S-model collision integral using tensor decompositions. Computer Physics Communications 264, 107954

    2021

  12. [12]

    Robust and conservative dy- namical low-rank methods for the Vlasov equation via a novel macro- micro decomposition

    Coughlin, J., Hu, J., Shumlak, U., 2024. Robust and conservative dy- namical low-rank methods for the Vlasov equation via a novel macro- micro decomposition. Journal of Computational Physics 509, 113055

    2024

  13. [13]

    Interpolatory dynamical low-rank approximation for the 3+ 3d Boltzmann–BGK equation

    Dektor, A., Einkemmer, L., 2025. Interpolatory dynamical low-rank approximation for the 3+ 3d Boltzmann–BGK equation. Journal of Computational Physics , 114515

    2025

  14. [14]

    Numerical methods for kinetic equa- tions

    Dimarco, G., Pareschi, L., 2014. Numerical methods for kinetic equa- tions. Acta Numerica 23, 369–520

    2014

  15. [15]

    Asymptotic-preserving dy- namical low-rank method for the stiff nonlinear Boltzmann equation

    Einkemmer, L., Hu, J., Zhang, S., 2025a. Asymptotic-preserving dy- namical low-rank method for the stiff nonlinear Boltzmann equation. Journal of Computational Physics , 114112

  16. [16]

    A review of low-rank methods for time-dependent kinetic simu- lations

    Einkemmer, L., Kormann, K., Kusch, J., McClarren, R.G., Qiu, J.M., 2025b. A review of low-rank methods for time-dependent kinetic simu- lations. Journal of Computational Physics , 114191

  17. [17]

    Krylov-based adaptive-rank implicit time integrators for stiff problems with applica- tion to nonlinear Fokker-Planck kinetic models

    El Kahza, H., Taitano, W., Qiu, J.M., Chacón, L., 2024. Krylov-based adaptive-rank implicit time integrators for stiff problems with applica- tion to nonlinear Fokker-Planck kinetic models. Journal of Computa- tional Physics 518, 113332

    2024

  18. [18]

    Deformed decomposi- tion for non-negative tensors, in: The 29th International Conference on Artificial Intelligence and Statistics

    Ghalamkari, K., Taborsky, P., Mørup, M., 2026. Deformed decomposi- tion for non-negative tensors, in: The 29th International Conference on Artificial Intelligence and Statistics

    2026

  19. [19]

    A conservative low rank tensor method for the Vlasov dynamics

    Guo, W., Qiu, J.M., 2024. A conservative low rank tensor method for the Vlasov dynamics. SIAM Journal on Scientific Computing 46, A232– A263

    2024

  20. [20]

    Hermitian transformations of deficiency-index (1, 1), Jacobi matrices and undetermined moment problems

    Hamburger, H.L., 1944. Hermitian transformations of deficiency-index (1, 1), Jacobi matrices and undetermined moment problems. Am. J. Math. 66, 489–522. 36

    1944

  21. [21]

    The BGK approximation of kinetic models for traffic

    Herty, M., Puppo, G., Roncoroni, S., Visconti, G., 2020. The BGK approximation of kinetic models for traffic. Kinet. Relat. Models 13, 279–307

    2020

  22. [22]

    An adaptive dynamical low rank method for the nonlinear Boltzmann equation

    Hu, J., Wang, Y., 2022. An adaptive dynamical low rank method for the nonlinear Boltzmann equation. Journal of Scientific Computing 92, 75

    2022

  23. [23]

    The numerical flow iteration for the Vlasov-Poisson equation

    Kirchhart, M., Wilhelm, R.P., 2024. The numerical flow iteration for the Vlasov-Poisson equation. SIAM Journal on Scientific Computing 46, A1972–A1997. doi:10.1137/23M154710X

    work page doi:10.1137/23m154710x 2024

  24. [24]

    Dynamical low-rank approximation

    Koch, O., Lubich, C., 2007. Dynamical low-rank approximation. SIAM Journal on Matrix Analysis and Applications 29, 434–454

    2007

  25. [25]

    G., & Bader, B

    Kolda, T.G., Bader, B.W., 2009. Tensor Decompositions and Applica- tions. SIAM Rev. 51, 455–500. doi:10.1137/07070111X

    work page doi:10.1137/07070111x 2009

  26. [26]

    On information and sufficiency

    Kullback, S., Leibler, R.A., 1951. On information and sufficiency. The annals of mathematical statistics 22, 79–86

    1951

  27. [27]

    Radiative heat transfer

    Modest, M.F., Mazumder, S., 2021. Radiative heat transfer. Academic press

    2021

  28. [28]

    The solution of the Boltzmann equation for a shock wave

    Mott-Smith, H.M., 1951. The solution of the Boltzmann equation for a shock wave. Physical Review 82, 885

    1951

  29. [29]

    UseofTensor-Traindecompositionswithadiscrete velocity Boltzmann solver

    Oblapenko, G., 2023. UseofTensor-Traindecompositionswithadiscrete velocity Boltzmann solver. arXiv preprint arXiv:2303.15142

    work page arXiv 2023

  30. [30]

    Specqk.jl.https://github.com/ACoM-RWTH/specqk

    Oblapenko, G., 2026. Specqk.jl.https://github.com/ACoM-RWTH/specqk. doi:10.5281/zenodo.20026362

    work page doi:10.5281/zenodo.20026362 2026

  31. [31]

    sparse and low-rank kinetic distri- bution estimation

    Oblapenko, G., Theisen, L., Wilhelm, R.P., Torrilhon, M., Herty, M., . Simulation data for "sparse and low-rank kinetic distri- bution estimation".https://doi.org/10.5281/zenodo.19855012. doi:10.5281/zenodo.19855012

    work page doi:10.5281/zenodo.19855012

  32. [32]

    Sparse reconstruction of multi-dimensional kinetic distributions

    Oblapenko, G., Torrilhon, M., Herty, M., 2026. Sparse reconstruction of multi-dimensional kinetic distributions. Kinetic and Related Models 20, 80–104. 37

    2026

  33. [33]

    The Moment Problem

    Schmüdgen, K., 2017. The Moment Problem. Graduate Texts in Math- ematics, Springer International Publishing

    2017

  34. [34]

    The problem of moments

    Shohat, J., Tamarkin, J., 1945. The problem of moments. AMS, Provi- dence

    1945

  35. [35]

    SparseTensorMoments.jl (version v0.1)

    Theisen, L., 2026. SparseTensorMoments.jl (version v0.1). doi:10.5281/zenodo.20305509

    work page doi:10.5281/zenodo.20305509 2026

  36. [36]

    The Mott- Smith solution to the regular shock reflection problem

    Timokhin, M.Y., Kudryavtsev, A., Bondar, Y.A., 2022. The Mott- Smith solution to the regular shock reflection problem. Journal of Fluid Mechanics 950, A14

    2022

  37. [37]

    High fidelity simulations of the multi- species Vlasov-Maxwell system with the numerical flow iteration

    Wilhelm, R.P., Bacchini, F., 2025. High fidelity simulations of the multi- species Vlasov-Maxwell system with the numerical flow iteration. Ac- cepted to proceedings in 16th International Conference on Numerical Modeling of Space Plasma Flows (ASTRONUM 2025)

    2025

  38. [38]

    Nufi beam instability simulation data.https://doi.org/10.5281/zenodo.19816196

    Wilhelm, R.P., Bacchini, F., 2026. Nufi beam instability simulation data.https://doi.org/10.5281/zenodo.19816196. doi:10.5281/zenodo.19816196

    work page doi:10.5281/zenodo.19816196 2026

  39. [39]

    Extendingthenumericalflowiterationtothemulti- species Vlasov-Maxwell system through Hamiltonian splitting

    Wilhelm, R.P., Bacchini, F., Schöps, S., Torrilhon, M., Merkel, M., Kirchhart, M., 2025. Extendingthenumericalflowiterationtothemulti- species Vlasov-Maxwell system through Hamiltonian splitting. URL: https://arxiv.org/abs/2511.11322,arXiv:2511.11322

    work page internal anchor Pith review arXiv 2025

  40. [40]

    On Nonlinear Closures for Moment Equations Based on Orthogonal Polynomials

    Yilmaz, E., Oblapenko, G., Torrilhon, M., 2024. On nonlinear closures for moment equations based on orthogonal polynomials. arXiv preprint arXiv:2407.05894 . 38

    work page internal anchor Pith review Pith/arXiv arXiv 2024