pith. machine review for the scientific record. sign in

arxiv: 2604.22087 · v1 · submitted 2026-04-23 · 💻 cs.MS

Recognition: unknown

JetSCI: A Hybrid JAX-PETSc Framework for Scalable Differentiable Simulation

Alberto Cattaneo , M Keith Ballard , Robert M. Kirby , Varun Shankar
Authors on Pith no claims yet

Pith reviewed 2026-05-08 12:41 UTC · model grok-4.3

classification 💻 cs.MS
keywords hybrid JAX-PETSc frameworkdifferentiable simulationfinite element methodmicromechanicsscientific machine learningautomatic differentiationscalable solvers
0
0 comments X

The pith

JetSCI merges JAX and PETSc to deliver scalable differentiable finite element simulations that outperform pure JAX code on heterogeneous micromechanics problems.

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

The paper introduces JetSCI, a hybrid framework that pairs JAX's automatic differentiation and GPU vectorization with PETSc's MPI-based solvers and preconditioners. This combination supports large-scale differentiable modeling for scientific machine learning while retaining the robustness needed for distributed-memory architectures. For finite element discretizations of heterogeneous micromechanics, the hybrid approach achieves better efficiency and accuracy than JAX-only implementations by exposing multilevel parallelism within and across nodes.

Core claim

JetSCI uses JAX for GPU-parallel differentiable discretizations of the governing equations and PETSc for the robust, scalable solution of the resulting linear and nonlinear systems on distributed-memory hardware, thereby unifying automatic differentiability with mature HPC capabilities for micromechanics problems.

What carries the argument

The JAX-PETSc interface that routes differentiable finite-element operators from JAX into PETSc's solvers while preserving end-to-end automatic differentiation and adding multilevel (GPU + MPI) parallelism.

If this is right

  • Differentiable surrogate models and data-driven constitutive laws can now be trained inside large-scale, distributed finite-element workflows.
  • Multilevel parallelism becomes available without rewriting existing JAX discretization code.
  • Preconditioning and nonlinear solver strategies from PETSc become directly usable inside gradient-based optimization loops.

Where Pith is reading between the lines

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

  • The same hybrid pattern could extend to other physics domains that combine local GPU kernels with global distributed solves, such as incompressible flow or structural dynamics.
  • Users could embed learned constitutive relations directly into existing PETSc-based production codes with minimal additional coding.
  • Performance gains may grow with problem size, suggesting JetSCI becomes more advantageous precisely when JAX-only codes hit memory or solver limits.

Load-bearing premise

The interface between JAX and PETSc preserves full automatic differentiability, scalability, and compatibility without introducing significant overhead for the targeted finite-element micromechanics discretizations.

What would settle it

Run the same heterogeneous micromechanics finite-element problem at increasing mesh sizes on a multi-node cluster; if JetSCI does not show lower wall-clock time or higher solution accuracy than an equivalent JAX-only implementation once the problem exceeds single-node memory, the performance claim does not hold.

Figures

Figures reproduced from arXiv: 2604.22087 by Alberto Cattaneo, M Keith Ballard, Robert M. Kirby, Varun Shankar.

Figure 1
Figure 1. Figure 1: JetSCI workflow. Element information and global problem structure both feed into view at source ↗
Figure 2
Figure 2. Figure 2: JetSCI software stack. JAX / fea-in-jax handles local differenti￾ated finite-element kernels and Newton iteration, CuPy and DLPack provide device-array interoperability, ctypes together with petsc4py and mpi4py bridges to distributed PETSc objects, and PETSc performs the sparse linear solve. this copy is negligible compared to the alternative of failing to fit the problem in memory at all. IV. IMPLEMENTATI… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of time to converge for JAX-only matrix-explicit and view at source ↗
Figure 6
Figure 6. Figure 6: Convergence time for direct solvers in CuPy and JetSCI. view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of time to convergence between PETSc and JAX solvers view at source ↗
Figure 7
Figure 7. Figure 7: Convergence time for JetSCI solvers with Jacobi and ILU precondi view at source ↗
Figure 8
Figure 8. Figure 8: Convergence time for JetSCI on single GPUs vs multiple GPUs on view at source ↗
read the original abstract

The rapid rise of scientific machine learning (SciML) has expanded the role of differentiable modeling, surrogate modeling, and data-driven constitutive laws in large-scale simulation. The JAX framework provides an attractive environment for these workflows through automatically differentiable programs, vectorization, GPU acceleration, and while enabling seamless learning of surrogate models. However, large-scale simulation still relies on mature HPC infrastructure. Libraries, such as PETSc, provide scalable MPI-based parallelism, robust linear and nonlinear solvers, and advanced preconditioning capabilities that remain difficult to reproduce in JAX-only workflows. We present JetSCI, a hybrid JAX-PETSc framework that unifies these complementary strengths. JetSCI uses JAX for GPU-parallel differentiable discretizations and PETSc for robust, scalable solution of the resulting systems on distributed-memory architectures, exposing multilevel parallelism through GPU acceleration within nodes and MPI parallelism across nodes. For finite element discretizations of heterogeneous micromechanics problems, JetSCI outperforms JAX-only implementations in efficiency and accuracy.

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 / 0 minor

Summary. The paper introduces JetSCI, a hybrid JAX-PETSc framework that assigns GPU-parallel differentiable finite-element discretizations to JAX and scalable distributed linear/nonlinear solves to PETSc, claiming that the resulting multilevel parallelism yields better efficiency and accuracy than pure-JAX implementations for heterogeneous micromechanics problems.

Significance. If the hybrid coupling can be shown to preserve end-to-end automatic differentiability while delivering measurable gains in wall-clock time, memory, and solution accuracy on distributed architectures, the work would provide a practical bridge between the differentiable-programming ecosystem and mature HPC libraries, enabling larger-scale SciML workflows in micromechanics and related fields.

major comments (3)
  1. [Abstract] Abstract: the central claim that JetSCI 'outperforms JAX-only implementations in efficiency and accuracy' for finite-element discretizations of heterogeneous micromechanics problems is stated without any accompanying benchmarks, timing tables, error metrics, strong-scaling curves, or memory profiles; this absence leaves the empirical superiority unsupported.
  2. [Abstract / Introduction] The description of automatic differentiation through the PETSc solve step is absent; no custom vjp rules, adjoint formulations, checkpointing strategy, or data-movement overhead analysis between JAX arrays and PETSc distributed vectors is supplied, which is load-bearing for the claim that full differentiability is retained with negligible cost.
  3. [Abstract] No concrete implementation details or performance data are given for the multilevel parallelism (GPU within nodes + MPI across nodes) on the targeted heterogeneous micromechanics discretizations, preventing assessment of whether the hybrid interface actually reduces overhead relative to a pure JAX or pure PETSc baseline.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment point-by-point below. Where the presentation in the abstract or introduction can be strengthened by referencing or summarizing material already present in the body of the manuscript, we have revised accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that JetSCI 'outperforms JAX-only implementations in efficiency and accuracy' for finite-element discretizations of heterogeneous micromechanics problems is stated without any accompanying benchmarks, timing tables, error metrics, strong-scaling curves, or memory profiles; this absence leaves the empirical superiority unsupported.

    Authors: We agree that the abstract would be strengthened by a brief reference to the supporting results. The full manuscript contains the requested benchmarks, timing tables, error metrics, strong-scaling curves, and memory profiles in Section 4. We have revised the abstract to include a concise summary of the key empirical gains (e.g., wall-clock time reductions and accuracy improvements on the heterogeneous micromechanics test cases) while retaining the abstract's brevity. This change directly supports the central claim without altering the manuscript's technical content. revision: yes

  2. Referee: [Abstract / Introduction] The description of automatic differentiation through the PETSc solve step is absent; no custom vjp rules, adjoint formulations, checkpointing strategy, or data-movement overhead analysis between JAX arrays and PETSc distributed vectors is supplied, which is load-bearing for the claim that full differentiability is retained with negligible cost.

    Authors: The mechanism for automatic differentiation through the PETSc solve is described in Section 3.2, including the custom VJP rules, the adjoint formulation, checkpointing strategy, and analysis of JAX-PETSc data movement overhead. We acknowledge that a high-level statement of this capability is missing from the abstract and introduction. We have added one sentence to the abstract and a short paragraph to the introduction that summarizes how end-to-end differentiability is preserved at negligible cost. These revisions make the differentiability claim explicit at the front of the paper while pointing readers to the detailed implementation. revision: yes

  3. Referee: [Abstract] No concrete implementation details or performance data are given for the multilevel parallelism (GPU within nodes + MPI across nodes) on the targeted heterogeneous micromechanics discretizations, preventing assessment of whether the hybrid interface actually reduces overhead relative to a pure JAX or pure PETSc baseline.

    Authors: Section 3.1 details the multilevel parallelism strategy (JAX for intra-node GPU discretizations and PETSc for inter-node MPI solves), and Section 4 provides the corresponding performance data and overhead comparisons against pure-JAX and pure-PETSc baselines on the heterogeneous micromechanics problems. We have revised the abstract to include a brief mention of the hybrid parallelization approach and the observed overhead reductions, with explicit references to Sections 3 and 4 for the concrete implementation and quantitative results. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework description with benchmarking claims

full rationale

The paper introduces JetSCI as a hybrid JAX-PETSc implementation for differentiable FE simulations and supports its central claim (outperformance vs. JAX-only on heterogeneous micromechanics problems) via direct empirical comparison of efficiency and accuracy. No derivation chain, fitted parameters renamed as predictions, self-definitional equations, or load-bearing self-citations appear. The hybrid interface and differentiability assertions are engineering statements whose validity rests on implementation details and benchmarks rather than reducing to the inputs by construction. This is the expected non-finding for a software-framework paper.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on the assumption that established properties of JAX and PETSc can be combined without loss of differentiability or scalability, plus standard finite-element theory for micromechanics.

axioms (3)
  • standard math JAX provides reliable automatic differentiation and GPU vectorization for discretizations.
    Invoked as the basis for the differentiable part of the workflow.
  • standard math PETSc supplies robust, scalable MPI solvers and preconditioners for distributed-memory systems.
    Invoked as the basis for the scalable solver component.
  • domain assumption Finite-element discretizations are appropriate for heterogeneous micromechanics problems.
    The performance comparison is stated specifically for these discretizations.

pith-pipeline@v0.9.0 · 5478 in / 1322 out tokens · 64465 ms · 2026-05-08T12:41:37.131900+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

51 extracted references · 34 canonical work pages

  1. [1]

    Boehmet al., Nature Reviews Physics 10.1038/s42254-021- 00417-4 (2022)

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,”Nature Reviews Physics, vol. 3, no. 6, pp. 422–440, 2021, https://doi.org/10.1038/s42254-021- 00314-5

    work page doi:10.1038/s42254-021- 2021

  2. [2]

    Operator learning: A statistical perspective,

    U. Subedi and A. Tewari, “Operator learning: A statistical perspective,” Annual Review of Statistics and Its Application, vol. 13, pp. 123–148, 2026, https://doi.org/10.1146/annurev-statistics-042424-070908

    work page doi:10.1146/annurev-statistics-042424-070908 2026

  3. [3]

    Deep learning in computational mechanics: a review,

    L. Herrmann and S. Kollmannsberger, “Deep learning in computational mechanics: a review,”Computational Mechanics, vol. 74, pp. 281–331, 2024, https://doi.org/10.1007/s00466-023-02434-4

    work page doi:10.1007/s00466-023-02434-4 2024

  4. [4]

    N.et al.A Review on Data-Driven Constitutive Laws for Solids.Archives of Computational Methods in Engineering32, 1841–1883 (2025)

    J. N. Fuhg, G. Anantha Padmanabha, N. Bouklas, B. Bahmani, W. Sun, N. N. Vlassis, M. Flaschel, P. Carrara, and L. De Lorenzis, “A review on data-driven constitutive laws for solids,”Archives of Computa- tional Methods in Engineering, vol. 32, no. 3, pp. 1841–1883, 2025, https://doi.org/10.1007/s11831-024-10196-2

    work page doi:10.1007/s11831-024-10196-2 2025

  5. [5]

    Physics- informed data-driven discovery of constitutive models with applica- tion to strain-rate-sensitive soft materials,

    K. Upadhyay, J. N. Fuhg, N. Bouklas, and K. T. Ramesh, “Physics- informed data-driven discovery of constitutive models with applica- tion to strain-rate-sensitive soft materials,”Computational Mechanics, vol. 74, pp. 1–30, 2024, https://doi.org/10.1007/s00466-024-02497-x

    work page doi:10.1007/s00466-024-02497-x 2024

  6. [6]

    R., Millman, K

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. del R ´ıo, M. Wiebe, P. Peterson, P. G ´erard-Marchant, K. Sheppard, T. Reddy, W. Weckesser, H. Abbasi, C. Gohlke, and T. E. Oliphant, “Array progr...

    work page doi:10.1038/s41586-020-2649-2 2020

  7. [7]

    E., et al

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

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

  8. [8]

    JAX: composable transformations of Python+NumPy programs,

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, and S. Wanderman-Milne, “JAX: composable transformations of Python+NumPy programs,” 2018. [Online]. Available: https://github.com/jax-ml/jax

    2018

  9. [9]

    JAX-FEM: A differentiable gpu-accelerated 3d finite ele- ment solver for automatic inverse design and mechanistic data sci- ence,

    T. Xue, S. Liao, Z. Gan, C. Park, X. Xie, W. K. Liu, and J. Cao, “JAX-FEM: A differentiable gpu-accelerated 3d finite ele- ment solver for automatic inverse design and mechanistic data sci- ence,”Computer Physics Communications, vol. 291, p. 108802, 2023, https://doi.org/10.1016/j.cpc.2023.108802

    work page doi:10.1016/j.cpc.2023.108802 2023

  10. [10]

    Efficient management of parallelism in object oriented numerical software li- braries,

    S. Balay, W. D. Gropp, L. C. McInnes, and B. F. Smith, “Efficient management of parallelism in object oriented numerical software li- braries,” inModern Software Tools in Scientific Computing, E. Arge, A. M. Bruaset, and H. P. Langtangen, Eds. Boston, MA: Birkh ¨auser, 1997, pp. 163–202

    1997

  11. [11]

    PETSc/TAO users manual,

    S. Balay, S. Abhyankar, M. F. Adams, S. Benson, J. Brown, P. Brune, K. Buschelman, E. M. Constantinescu, L. Dalcin, A. Dener, V . Ei- jkhout, J. Faibussowitsch, W. D. Gropp, V . Hapla, T. Isaac, P. Jo- livet, D. Karpeev, D. Kaushik, M. G. Knepley, F. Kong, S. Kruger, D. A. May, L. C. McInnes, R. T. Mills, L. Mitchell, T. Munson, J. E. Roman, K. Rupp, P. S...

    work page doi:10.2172/3025790 2026

  12. [12]

    PETSc Web page,

    S. Balay, S. Abhyankar, M. F. Adams, J. Brown, P. Brune, K. Buschelman, L. Dalcin, V . Eijkhout, W. D. Gropp, D. Kaushik, M. G. Knepley, L. C. McInnes, K. Rupp, B. F. Smith, S. Zampini, and H. Zhang, “PETSc Web page,” http://www.mcs.anl.gov/petsc, 2015. [Online]. Available: http://www.mcs.anl.gov/petsc

    2015

  13. [13]

    ChatGPT (GPT-5.3),

    OpenAI, “ChatGPT (GPT-5.3),” https://chat.openai.com/, 2026, used for grammar correction assistance

    2026

  14. [14]

    Jax-sso: Differentiable finite element analysis solver for structural optimization and seamless integration with neural networks.arXiv preprint arXiv:2407.20026, 2024

    G. Wu, “Jax-sso: Differentiable finite element analysis solver for structural optimization and seamless integration with neural networks,”arXiv preprint arXiv:2407.20026, 2024, https://doi.org/10.48550/arXiv.2407.20026

    work page doi:10.48550/arxiv.2407.20026 2024

  15. [15]

    Adcme: Learning spatially-varying physical fields using deep neural networks,

    K. Xu and E. Darve, “Adcme: Learning spatially-varying physical fields using deep neural networks,”arXiv preprint arXiv:2011.11955, 2020, https://doi.org/10.48550/arXiv.2011.11955

    work page doi:10.48550/arxiv.2011.11955 2011

  16. [16]

    D. A. Ham, P. H. J. Kelly, L. Mitchell, C. J. Cotter, R. C. Kirby, K. Sagiyama, N. Bouziani, S. V orderwuelbecke, T. J. Gregory, J. Bet- teridge, D. R. Shapero, R. W. Nixon-Hill, C. J. Ward, P. E. Farrell, P. D. Brubeck, I. Marsden, T. H. Gibson, M. Homolya, T. Sun, A. T. T. McRae, F. Luporini, A. Gregory, M. Lange, S. W. Funke, F. Rathgeber, G.-T. Bercea...

    work page doi:10.25561/104839 2023

  17. [17]

    DOLFINx: the next generation FEniCS problem solving environment,

    I. A. Baratta, J. P. Dean, J. S. Dokken, M. Habera, J. S. Hale, C. N. Richardson, M. E. Rognes, M. W. Scroggs, N. Sime, and G. N. Wells, “DOLFINx: the next generation FEniCS problem solving environment,” preprint, 2023

    2023

  18. [18]

    dolfin-adjoint 2018.1: au- tomated adjoints for fenics and firedrake,

    S. K. Mitusch, S. W. Funke, and J. S. Dokken, “dolfin-adjoint 2018.1: au- tomated adjoints for fenics and firedrake,”Journal of Open Source Soft- ware, vol. 4, no. 38, p. 1292, 2019, https://doi.org/10.21105/joss.01292

    work page doi:10.21105/joss.01292 2018

  19. [19]

    Physics-driven machine learning models coupling pytorch and firedrake,

    N. Bouziani and D. A. Ham, “Physics-driven machine learning models coupling pytorch and firedrake,”arXiv preprint arXiv:2303.06871, 2023, https://doi.org/10.48550/arXiv.2303.06871

    work page doi:10.48550/arxiv.2303.06871 2023

  20. [20]

    Differentiable program- ming across the pde and machine learning barrier,

    N. Bouziani, D. A. Ham, and A. Farsi, “Differentiable program- ming across the pde and machine learning barrier,”arXiv preprint arXiv:2409.06085, 2024, https://doi.org/10.48550/arXiv.2409.06085

    work page doi:10.48550/arxiv.2409.06085 2024

  21. [21]

    Formopt: A fenicsx toolbox for level set-based shape optimization supporting parallel computing,

    J. D. D ´ıaz-Avalos and A. Laurain, “Formopt: A fenicsx toolbox for level set-based shape optimization supporting parallel computing,”arXiv preprint arXiv:2601.05709, 2026, https://doi.org/10.48550/arXiv.2601.05709

    work page doi:10.48550/arxiv.2601.05709 2026

  22. [22]

    Instead of rewriting foreign code for machine learning, automatically synthesize fast gradients,

    W. S. Moses and V . Churavy, “Instead of rewriting foreign code for machine learning, automatically synthesize fast gradients,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 12 472–12 485. [Online]. Available: https://dl.acm.org/doi/abs/10.5555/ 3495724.3496770

    work page arXiv 2020

  23. [23]

    Reverse-mode automatic differentiation and optimization of gpu kernels via enzyme,

    W. S. Moses, V . Churavy, L. Paehler, J. H ¨uckelheim, S. H. K. Narayanan, M. Schanen, and J. Doerfert, “Reverse-mode automatic differentiation and optimization of gpu kernels via enzyme,” inSC ’21: Proceedings of the International Conference for High Perfor- mance Computing, Networking, Storage and Analysis, 2021, pp. 1–18, https://doi.org/10.1145/345881...

    work page doi:10.1145/3458817.3476165 2021

  24. [24]

    Neutrino Production via $e^-e^+$ Collision at $Z$-boson Peak

    W. S. Moses, S. H. K. Narayanan, L. Paehler, V . Churavy, M. Schanen, J. H ¨uckelheim, J. Doerfert, and P. Hovland, “Scalable automatic differ- entiation of multiple parallel paradigms through compiler augmentation,” inSC ’22: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, 2022, pp. 1–18, http...

    work page Pith review doi:10.1109/sc41404.2022.00065 2022

  25. [25]

    Scalable analysis and design using automatic differentiation,

    J. Andrej, T. Kolev, and B. S. Lazarov, “Scalable analysis and design using automatic differentiation,”arXiv preprint arXiv:2506.00746, 2025, https://doi.org/10.48550/arXiv.2506.00746

    work page doi:10.48550/arxiv.2506.00746 2025

  26. [26]

    The deal.ii finite element library: Design, features, and insights,

    D. Arndt, W. Bangerth, D. Davydov, T. Heister, L. Heltai, M. Kro- nbichler, M. Maier, J.-P. Pelteret, B. Turcksin, and D. Wells, “The deal.ii finite element library: Design, features, and insights,”Comput- ers & Mathematics with Applications, vol. 81, pp. 407–422, 2021, https://doi.org/10.1016/j.camwa.2020.02.022

    work page doi:10.1016/j.camwa.2020.02.022 2021

  27. [27]

    Algorithm 755: Adol-c: a package for the automatic differentiation of algorithms written in c/c++,

    A. Griewank, D. Juedes, and J. Utke, “Algorithm 755: Adol-c: a package for the automatic differentiation of algorithms written in c/c++,”ACM Transactions on Mathematical Software, vol. 22, no. 2, pp. 131–167, 1996, https://doi.org/10.1145/229473.229474

    work page doi:10.1145/229473.229474 1996

  28. [28]

    Automatic differentiation of c++ codes on emerging many-core architectures with sacado,

    E. T. Phipps, R. P. Pawlowski, A. G. Salinger, R. A. Bartlett, D. M. Gay, and . Kokkos, “Automatic differentiation of c++ codes on emerging many-core architectures with sacado,”ACM Transactions on Mathemat- ical Software, 2022, https://doi.org/10.1145/3560262

    work page doi:10.1145/3560262 2022

  29. [29]

    4.0 MOOSE: Enabling massively parallel multiphysics simulation,

    L. Harbour, G. Giudicelli, A. D. Lindsay, P. German, J. Hansel, C. Icenhour, M. Li, J. M. Miller, R. H. Stogner, P. Behne, D. Yankura, Z. M. Prince, C. DeChant, D. Schwen, B. W. Spencer, M. Tano, N. Choi, Y . Wang, M. Nezdyur, Y . Miao, T. Hu, S. Kumar, C. Matthews, B. Langley, N. Nobre, A. Blair, C. MacMackin, H. B. Rocha, E. Palmer, J. Carter, J. Meier,...

    work page doi:10.1016/j.softx.2025.102264 2025

  30. [30]

    Automatic differentiation in metaphysicl and its applications in moose,

    A. Lindsay, R. Stogner, D. Gaston, D. Schwen, C. Matthews, W. Jiang, L. K. Aagesen, R. Carlsen, F. Kong, A. Slaugh- teret al., “Automatic differentiation in metaphysicl and its applications in moose,”Nuclear Technology, pp. 1–18, 2021, https://doi.org/10.1080/00295450.2020.1838877

    work page doi:10.1080/00295450.2020.1838877 2021

  31. [31]

    lifex: A flexible, high performance library for the numer- ical solution of complex finite element problems,

    P. C. Africa, “lifex: A flexible, high performance library for the numer- ical solution of complex finite element problems,”SoftwareX, vol. 20, p. 101252, 2022, https://doi.org/10.1016/j.softx.2022.101252

    work page doi:10.1016/j.softx.2022.101252 2022

  32. [32]

    High-performance derivative computations using codipack,

    M. Sagebaum, T. Albring, and N. R. Gauger, “High-performance derivative computations using codipack,”ACM Transactions on Mathematical Software, vol. 45, no. 4, pp. 1–26, 2019, https://doi.org/10.1145/3356900

    work page doi:10.1145/3356900 2019

  33. [33]

    Jaxbind: Bind any function to jax,

    J. Rothet al., “Jaxbind: Bind any function to jax,”arXiv preprint arXiv:2403.08847, 2024

    work page arXiv 2024

  34. [34]

    Parallel distributed computing using python,

    L. D. Dalcin, R. R. Paz, P. A. Kler, and A. Cosimo, “Parallel distributed computing using python,”Advances in Water Resources, vol. 34, no. 9, pp. 1124–1139, 2011, new Computational Methods and Software Tools. [Online]. Available: https://www.sciencedirect. com/science/article/pii/S0309170811000777

    2011

  35. [35]

    fea-in-jax,

    K. Ballard, “fea-in-jax,” 2024, finite element method implementation in JAX; accessed 2026-04-07. [Online]. Available: https://github.com/ KeithBallard/fea-in-jax

    2024

  36. [36]

    Jax documentation,

    JAX authors, “Jax documentation,” 2026, accessed 2026-04-07. [Online]. Available: https://docs.jax.dev/en/latest/quickstart.html

    2026

  37. [37]

    The influence of microstructure randomness on prediction of fiber properties in compos- ites,

    M. K. Ballard, W. R. McLendon, and J. D. Whitcomb, “The influence of microstructure randomness on prediction of fiber properties in compos- ites,”Journal of Composite Materials, vol. 48, no. 29, pp. 3605–3620, 2014, https://doi.org/10.1177/0021998313511654

    work page doi:10.1177/0021998313511654 2014

  38. [38]

    Prediction of tow ar- chitecture and stress distributions for a 3d woven composite,

    M. K. Ballard and J. D. Whitcomb, “Prediction of tow ar- chitecture and stress distributions for a 3d woven composite,” inProceedings of the 32nd ASC Technical Conference, 2017, https://doi.org/10.12783/asc2017/15402

    work page doi:10.12783/asc2017/15402 2017

  39. [39]

    Outgassing Environment of Spacecraft: An Overview

    Ballard, M. Keith and Whitcomb, John D., “Stress analysis of 3d textile composites using high performance computing: new insights and challenges,”IOP Conference Series: Materials Science and Engi- neering, vol. 406, no. 1, p. 012004, 2018, https://doi.org/10.1088/1757- 899X/406/1/012004

    work page doi:10.1088/1757- 2018

  40. [40]

    PETSc overview,

    “PETSc overview,” https://petsc.org/release/overview/, 2026, accessed 2026-04-07

    2026

  41. [41]

    petsc4py: Petsc for python,

    L. Dalcin, “petsc4py: Petsc for python,” 2026, accessed 2026-04-07. [Online]. Available: https://petsc.org/release/petsc4py/

    2026

  42. [42]

    petsc4py documentation,

    Dalcin, Lisandro, “petsc4py documentation,” 2026, accessed 2026- 04-07. [Online]. Available: https://petsc.org/release/petsc4py/reference/ petsc4py.html

    2026

  43. [43]

    Advances in Water Resources , volume =

    L. D. Dalcin, R. R. Paz, and M. Storti, “Parallel distributed computing using python,”Advances in Water Resources, vol. 34, no. 9, pp. 1124– 1139, 2011, https://doi.org/10.1016/j.advwatres.2011.04.013

    work page doi:10.1016/j.advwatres.2011.04.013 2011

  44. [44]

    mpi4py: Status update after 12 years of development,

    L. Dalcin and Y .-L. L. Fang, “mpi4py: Status update after 12 years of development,”Computing in Science & Engineering, vol. 23, no. 4, pp. 47–54, 2021, https://doi.org/10.1109/MCSE.2021.3083216

    work page doi:10.1109/mcse.2021.3083216 2021

  45. [45]

    Mpi for python documentation,

    L. Dalcin, “Mpi for python documentation,” 2025, accessed 2026-04-07. [Online]. Available: https://mpi4py.readthedocs.io/

    2025

  46. [46]

    CuPy: A NumPy-compatible library for NVIDIA GPU calculations,

    R. Okuta, Y . Unno, D. Nishino, S. Hido, and C. Loomis, “CuPy: A NumPy-compatible library for NVIDIA GPU calculations,” in Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-First Annual Conference on Neural Information Processing Systems (NIPS), 2017. [Online]. Available: http://learningsys.org/nips17/ assets/papers/paper 16.pdf

    2017

  47. [47]

    Cupy documentation,

    CuPy developers, “Cupy documentation,” 2026, accessed 2026-04-07. [Online]. Available: https://docs.cupy.dev/en/stable/overview.html

    2026

  48. [48]

    Python specification for dlpack,

    Data APIs Consortium and DLPack contributors, “Python specification for dlpack,” 2026, accessed 2026-04-07. [Online]. Available: https: //dmlc.github.io/dlpack/latest/python spec.html

    2026

  49. [49]

    Dlpack: Open in-memory tensor structure,

    DLPack contributors, “Dlpack: Open in-memory tensor structure,” 2026, accessed 2026-04-07. [Online]. Available: https://github.com/ dmlc/dlpack

    2026

  50. [50]

    [Online]

    Python Software Foundation,ctypes — A foreign function library for Python, Python Software Foundation, 2026, accessed 2026-04-07. [Online]. Available: https://docs.python.org/3/library/ctypes.html

    2026

  51. [51]

    A supernodal approach to sparse partial pivoting,

    J. W. Demmel, S. C. Eisenstat, J. R. Gilbert, X. S. Li, and J. W. H. Liu, “A supernodal approach to sparse partial pivoting,”SIAM Journal on Matrix Analysis and Applications, vol. 20, no. 3, pp. 720–755, 1999, https://doi.org/10.1137/S0895479895291765

    work page doi:10.1137/s0895479895291765 1999