pith. sign in

arxiv: 2605.28368 · v2 · pith:BYFN5K64new · submitted 2026-05-27 · 💻 cs.LG · cond-mat.mtrl-sci· physics.app-ph

LEIA: Learned Environment for Interactive Architected Materials

Haiqian Yang , Yuan Cao , Markus J. Buehler This is my paper

Pith reviewed 2026-06-29 13:35 UTC · model grok-4.3

classification 💻 cs.LG cond-mat.mtrl-sciphysics.app-ph
keywords world modelarchitected materialsautoregressive predictionsurrogate modelingde novo designfinite element validationMicroPlate benchmark3D unstructured meshes
0
0 comments X

The pith

A learned world model generates step-by-step stress and deformation fields on three-dimensional architected material meshes.

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

The paper introduces LEIA as a world model that accepts sequential boundary conditions from a user and produces corresponding deformation and stress fields through autoregressive prediction. It addresses the challenges of nonlinear material behavior, history dependence, and large unstructured meshes that traditional simulation methods handle slowly. The approach is evaluated on the MicroPlate benchmark, which includes both explicitly resolved lattice microstructures and implicitly modeled homogeneous plates, and is compared against four baseline methods. LEIA is shown to support surrogate-guided searches that generate and rank new material design candidates, with the rankings validated for stress accuracy against finite element ground truth. If correct, this would allow engineers to explore and refine architected material designs interactively at speeds not feasible with direct simulation alone.

Core claim

LEIA is a learned environment that applies boundary conditions step by step to large three-dimensional unstructured meshes of architected plates and generates autoregressive responses for deformation and stress fields. It covers two regimes of microstructure modeling in the MicroPlate benchmark and enables efficient candidate generation and ranking for de novo designs, where the stress-accurate rankings are validated by finite element ground truth.

What carries the argument

The autoregressive learned model that produces deformation and stress field responses to sequential user-specified loading on unstructured 3D meshes while capturing nonlinear and history-dependent behavior.

If this is right

  • Candidate generation and ranking for de novo architected material designs becomes efficient through surrogate guidance.
  • Stress-accurate candidate ranking is achievable and matches finite element ground truth.
  • Real-time observation of deformation and stress fields is possible under step-by-step boundary condition application.
  • The model operates across both explicit three-dimensional lattice microstructures and implicit homogeneous plate representations.
  • The MicroPlate benchmark provides a consistent testbed for comparing world models on architected material tasks.

Where Pith is reading between the lines

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

  • Interactive loading sequences could support testing of materials under complex or time-varying conditions not covered in initial training data.
  • The approach might integrate into larger design loops that combine multiple physical objectives beyond stress.
  • Extension to hierarchical structures spanning more length scales could follow if the autoregressive structure scales without added error.
  • The benchmark could standardize evaluation of learned simulators for other nonlinear engineering domains with history dependence.

Load-bearing premise

The autoregressive model generalizes to unseen loading sequences and microstructures without accumulating errors that would invalidate the stress predictions used for ranking.

What would settle it

Apply the surrogate-guided search to produce a ranked list of new microstructures, then run full finite element analysis on the top candidates and check whether their actual stress values preserve the same ranking order as LEIA predicted.

Figures

Figures reproduced from arXiv: 2605.28368 by Haiqian Yang, Markus J. Buehler, Yuan Cao.

Figure 1
Figure 1. Figure 1: (a) Inference throughput of LEIA and FEM vs. mesh size. LEIA achieves high per-step inference speedup. Dashed line: interactive threshold (30 FPS). (b) A MicroPlate lattice plate (301,565 nodes). Hierarchical materials, structures whose property emerges from geom￾etry organized across multiple length scales, are among the most effective designs in nature [1, 2]. Bone, nacre, and wood achieve combinations o… view at source ↗
Figure 2
Figure 2. Figure 2: LEIA architecture. The tokenizer compresses the physical fields St on an arbitrary mesh X into K latent tokens Zt via Perceiver cross-attention. The dynamics transformer, conditioned on the boundary condition action At, predicts the next latent state autoregressively. The decoder reconstructs the physical states at mesh nodes. 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Autoregressive rollout on a MicroPlate lattice (301,565 nodes, 1,382,904 tetrahedra). Columns show selected timesteps. Top: ground-truth von Mises stress. Middle: LEIA prediction. Bottom: absolute von Mises stress error [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Surrogate-guided beam search over MicroPlate. (a) Topology evolution: best design at selected iterations, showing progressive material removal as the search optimizes for high stretch resistance, low shear resistance, and low volume fraction. (b) Convergence of the design metric. Solid: LEIA. Dashed: FEM validation at each iteration, confirming that the surrogate ranking is reliable. (c) Search landscape: … view at source ↗
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Unit cell of a MicroPlate lattice. The compact graph (4 beams, shown as colored rods) is replicated under all 48 operations of the cubic symmetry group Oh, producing 48 × 4 = 192 beams whose smooth-min-blended signed distance field yields the final mesh (gray). Each color represents one seed beam and all its symmetric copies. Dataset statistics. Each topology has multiple boundary condition trajectories (8… view at source ↗
Figure 7
Figure 7. Figure 7: Library of MicroPlate unit cells. Unit cells (RVEs) corresponding to the 63 architected lattices, shown after Oh symmetry expansion. Top: 55 training topologies. Bottom: 8 held-out topologies for evaluating generalization to unseen geometries. The number above each cell is the plate mesh size (in thousands of nodes) after 5 × 5 × 1 tiling and TetGen volumetric meshing. 0.4 0.2 0.0 0.2 0.4 0.6 0.8 PC1 (55% … view at source ↗
Figure 8
Figure 8. Figure 8: PCA projection of architected-lattice features. PCA projection of the 15-dimensional graph feature vectors (node/beam counts, spatial spread, beam lengths, radii, connectivity) for all 63 architected lattices. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Per-frame normalized work error on MicroPlate. Per-frame normalized work error |W − Wˆ |/Wmax across 55 architected lattices during 30-step autoregressive rollout. Solid line: median. Shaded band: interquartile range. 0 10 20 30 Frame 0 2 4 6 8 10 12 W o r k, W Plate ID 026 031 036 140 224 239 288 305 317 338 [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Work vs. time on ten MicroPlate shapes. Solid: FEM ground truth. Dashed: prediction. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
read the original abstract

World models have enabled interactive exploration of game environments and robotic manipulation, but physical engineering remains beyond their reach: real materials exhibit nonlinear constitutive laws, carry history-dependent internal state, undergo inertial dynamics, and may possess hierarchical structures spanning multiple length scales. We present LEIA (Learned Environment for Interactive Architected materials), a world model that lets engineers apply boundary conditions step by step and observe the resulting deformation and stress fields in real time. LEIA handles large three-dimensional unstructured meshes and generates autoregressive responses to user-specified loading. We introduce MicroPlate, a benchmark of architected plates spanning two regimes of microstructure modeling: architected lattices that resolve microstructure explicitly through three-dimensional geometry, and a homogeneous plate where microstructural change is modeled implicitly through internal degrees of freedom. MicroPlate is used to assess LEIA alongside four baseline methods across both regimes. Finally, we demonstrate that LEIA enables efficient candidate generation and ranking for fast surrogate-guided search for de novo designs of architected materials, with stress-accurate candidate ranking validated by finite element ground truth.

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

2 major / 1 minor

Summary. The paper introduces LEIA, a learned autoregressive world model for real-time interactive simulation of 3D architected materials under user-specified boundary conditions. It handles nonlinear constitutive behavior, history-dependent states, and large unstructured meshes in both explicit-lattice and implicit-microstructure regimes. The work presents the MicroPlate benchmark to evaluate LEIA against four baselines and demonstrates its application to surrogate-guided de novo design search, claiming stress-accurate candidate ranking validated by finite-element ground truth.

Significance. If the validation claims hold with quantified error bounds, LEIA would represent a meaningful advance in applying world models to physical engineering domains that require history-dependent mechanics and multi-scale structures. The introduction of the MicroPlate benchmark itself provides a reusable testbed for future surrogate models in architected materials.

major comments (2)
  1. [Abstract] Abstract: the central claim that LEIA supplies 'stress-accurate candidate ranking validated by finite element ground truth' for de novo search is load-bearing, yet the provided text reports neither per-step nor cumulative L2 stress errors, nor any bound on autoregressive drift under distribution shift to unseen loading sequences or microstructures. Without these quantities the ranking accuracy premise cannot be assessed.
  2. [MicroPlate benchmark and evaluation] MicroPlate evaluation section (implied by the benchmark description): the comparison to four baseline methods is asserted but no tables or figures quantify stress-field accuracy, ranking correlation with FE, or error growth versus sequence length, which directly determines whether the surrogate-guided search claim is supported.
minor comments (1)
  1. [Abstract] The abstract would benefit from a single sentence stating the key quantitative result (e.g., mean stress error or ranking correlation) that supports the 'stress-accurate' claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. The comments correctly identify that the current manuscript does not supply the specific quantitative error metrics needed to fully substantiate the abstract claim and the surrogate-guided design results. We will revise the manuscript to include these quantities.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that LEIA supplies 'stress-accurate candidate ranking validated by finite element ground truth' for de novo search is load-bearing, yet the provided text reports neither per-step nor cumulative L2 stress errors, nor any bound on autoregressive drift under distribution shift to unseen loading sequences or microstructures. Without these quantities the ranking accuracy premise cannot be assessed.

    Authors: We agree that the abstract claim is load-bearing and that the manuscript currently lacks the requested per-step and cumulative L2 stress errors as well as explicit bounds on autoregressive drift. In the revised version we will add these metrics (computed on held-out loading sequences and microstructures) to the results section, include them in the abstract, and provide the corresponding tables and figures. revision: yes

  2. Referee: [MicroPlate benchmark and evaluation] MicroPlate evaluation section (implied by the benchmark description): the comparison to four baseline methods is asserted but no tables or figures quantify stress-field accuracy, ranking correlation with FE, or error growth versus sequence length, which directly determines whether the surrogate-guided search claim is supported.

    Authors: The referee is correct that the present manuscript does not contain the requested tables or figures for stress-field L2 accuracy, ranking correlation with FE ground truth, or error growth versus sequence length. We will expand the MicroPlate evaluation section with these quantitative results (including direct comparison to the four baselines) to support the design-search claims. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical ML model with external validation

full rationale

The paper presents an autoregressive neural world model trained on simulation data for architected materials, with all claims about stress-accurate ranking supported by direct comparison to independent finite-element ground truth on held-out cases. No closed-form derivations, self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text; performance assertions reduce to standard supervised learning plus external benchmarking rather than any reduction to the model's own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be extracted beyond the implicit assumption that a neural world model can capture history-dependent nonlinear mechanics on unstructured meshes.

pith-pipeline@v0.9.1-grok · 5716 in / 1062 out tokens · 22425 ms · 2026-06-29T13:35:17.398855+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

73 extracted references · 65 canonical work pages · 7 internal anchors

  1. [1]

    M. A. Meyers, Po-Yu Chen, A. Lin, and Yasuaki Seki. Biological materials: Struc- ture and mechanical properties.Progress in Materials Science, 53:1–206, 2008. doi: 10.1016/j.pmatsci.2007.05.002

    work page doi:10.1016/j.pmatsci.2007.05.002 2008

  2. [2]

    Wegst, Hao Bai, E

    U. Wegst, Hao Bai, E. Saiz, A. Tomsia, and Robert O. Ritchie. Bioinspired structural materials. Nature Materials, 14:23–36, 2015. doi: 10.1038/nmat4089

    work page doi:10.1038/nmat4089 2015

  3. [3]

    J. U. Surjadi and C. M. Portela. Enabling three-dimensional architected materials across length scales and timescales.Nature Materials, 24:493–505, 2025. doi: 10.1038/s41563-025-02119-8

    work page doi:10.1038/s41563-025-02119-8 2025

  4. [4]

    X. Xia, C. Spadaccini, and J. Greer. Responsive materials architected in space and time.Nature Reviews Materials, 7:683–701, 2022. doi: 10.1038/s41578-022-00450-z

    work page doi:10.1038/s41578-022-00450-z 2022

  5. [5]

    Xiaojian Wang, Shanqing Xu, Shiwei Zhou, W. Xu, M. Leary, Peter Choong, M. Qian, M. Brandt, and Y . Xie. Topological design and additive manufacturing of porous metals for bone scaffolds and orthopaedic implants: A review.Biomaterials, 83:127–141, 2016. doi: 10.1016/j.biomaterials.2016.01.012

    work page doi:10.1016/j.biomaterials.2016.01.012 2016

  6. [6]

    Tancogne-Dejean, A

    T. Tancogne-Dejean, A. Spierings, and D. Mohr. Additively-manufactured metallic micro-lattice materials for high specific energy absorption under static and dynamic loading.Acta Materialia, 116:14–28, 2016. doi: 10.1016/j.actamat.2016.05.054

    work page doi:10.1016/j.actamat.2016.05.054 2016

  7. [7]

    Raney, X

    Pengcheng Jiao, Jochen Mueller, J. Raney, X. Zheng, and Amir H. Alavi. Mechanical metamate- rials and beyond.Nature Communications, 14:6004, 2023. doi: 10.1038/s41467-023-41679-8

    work page doi:10.1038/s41467-023-41679-8 2023

  8. [8]

    Shaikeea, Huachen Cui, M

    A. Shaikeea, Huachen Cui, M. O’Masta, X. Zheng, and V . Deshpande. The toughness of mechanical metamaterials.Nature Materials, 21:297–304, 2022. doi: 10.1038/s41563-021- 01182-1

    work page doi:10.1038/s41563-021- 2022

  9. [9]

    Watanabe

    Xiaoyang Zheng, Xubo Zhang, Ta-Te Chen, and I. Watanabe. Deep learning in mechanical metamaterials: From prediction and generation to inverse design.Advanced Materials, 35: e2302530, 2023. doi: 10.1002/adma.202302530

    work page doi:10.1002/adma.202302530 2023

  10. [10]

    Kochmann

    Jan-Hendrik Bastek, WaiChing Sun, and Dennis M. Kochmann. Physics-informed diffusion models. InInternational Conference on Learning Representations, 2025. URL https:// arxiv.org/abs/2403.14404

    work page arXiv 2025

  11. [11]

    Kochmann

    Li Zheng, Siddhant Kumar, and Dennis M. Kochmann. Algebraic language models for in- verse design of metamaterials via diffusion transformers.Nature Machine Intelligence, 8 (4):628–640, 2026. doi: 10.1038/s42256-026-01218-8. URL https://doi.org/10.1038/ s42256-026-01218-8

    work page doi:10.1038/s42256-026-01218-8 2026

  12. [12]

    Hirschler, R

    T. Hirschler, R. Bouclier, P. Antolin, and A. Buffa. Reduced order modeling based inexact FETI- DP solver for lattice structures.International Journal for Numerical Methods in Engineering, 125, 2024. doi: 10.1002/nme.7419

    work page doi:10.1002/nme.7419 2024

  13. [13]

    Ef- ficient fine-scale simulation of nonlinear hyperelastic lattice structures.arXiv preprint arXiv:2603.10741, 2026

    Clément Guillet, Thibaut Hirschler, Pierre Jolivet, Pablo Antolin, and Robin Bouclier. Ef- ficient fine-scale simulation of nonlinear hyperelastic lattice structures.arXiv preprint arXiv:2603.10741, 2026. URLhttps://arxiv.org/abs/2603.10741

    work page arXiv 2026

  14. [14]

    Gongora, Caleb Friedman, D

    Aldair E. Gongora, Caleb Friedman, D. Newton, T. Yee, Zachary Doorenbos, Brian Giera, E. Duoss, T. Yong-Jin Han, Kyle T. Sullivan, and Jennifer N. Rodriguez. Accelerating the design of lattice structures using machine learning.Scientific Reports, 14, 2024. doi: 10.1038/s41598- 024-63204-7

    work page doi:10.1038/s41598- 2024

  15. [15]

    LatticeGraphNet: a two-scale graph neural operator for simulating lattice structures.Engineering with Computers, 41(2):829–844, 2024

    Ayush Jain, Ehsan Haghighat, and Sai Nelaturi. LatticeGraphNet: a two-scale graph neural operator for simulating lattice structures.Engineering with Computers, 41(2):829–844, 2024. doi: 10.1007/s00366-024-02034-7

    work page doi:10.1007/s00366-024-02034-7 2024

  16. [16]

    Glaesener, and D

    Jan-Hendrik Bastek, Siddhant Kumar, Bastian Telgen, R. Glaesener, and D. Kochmann. Invert- ing the structure–property map of truss metamaterials by deep learning.Proceedings of the National Academy of Sciences, 119, 2022. doi: 10.1073/pnas.2111505119. 10

    work page doi:10.1073/pnas.2111505119 2022

  17. [17]

    Kochmann

    Jan-Hendrik Bastek and D. Kochmann. Inverse design of nonlinear mechanical metamaterials via video denoising diffusion models.Nature Machine Intelligence, 5:1466–1475, 2023. doi: 10.1038/s42256-023-00762-x

    work page doi:10.1038/s42256-023-00762-x 2023

  18. [18]

    Karapiperis, Siddhant Kumar, and D

    Li Zheng, K. Karapiperis, Siddhant Kumar, and D. Kochmann. Unifying the design space and optimizing linear and nonlinear truss metamaterials by generative modeling.Nature Communications, 14, 2023. doi: 10.1038/s41467-023-42068-x

    work page doi:10.1038/s41467-023-42068-x 2023

  19. [19]

    White, William J

    Daniel A. White, William J. Arrighi, Jun Kudo, and Seth E. Watts. Multiscale topology optimization using neural network surrogate models.Computer Methods in Applied Mechanics and Engineering, 346:1118–1135, 2019. doi: 10.1016/j.cma.2018.09.007. URL https://doi. org/10.1016/j.cma.2018.09.007

    work page doi:10.1016/j.cma.2018.09.007 2019

  20. [20]

    Weaver, Chris H

    Bolei Deng, Ahmad Zareei, Xiaoxiao Ding, James C. Weaver, Chris H. Rycroft, and Katia Bertoldi. Inverse design of mechanical metamaterials with target nonlinear re- sponse via a neural accelerated evolution strategy.Advanced Materials, 34(41), 2022. doi: 10.1002/adma.202206238. URLhttps://doi.org/10.1002/adma.202206238

    work page doi:10.1002/adma.202206238 2022

  21. [21]

    Xue, Shuheng Liao, Zhengtao Gan, Chanwook Park, Xiaoyu Xie, Wing Kam Liu, and Jian Cao

    T. Xue, Shuheng Liao, Zhengtao Gan, Chanwook Park, Xiaoyu Xie, Wing Kam Liu, and Jian Cao. JAX-FEM: A differentiable GPU-accelerated 3D finite element solver for automatic inverse design and mechanistic data science.Computer Physics Communications, 291:108802,

  22. [22]

    doi: 10.1016/j.cpc.2023.108802

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

  23. [23]

    Recurrent World Models Facilitate Policy Evolution

    David Ha and Jürgen Schmidhuber. Recurrent world models facilitate policy evolution. In Advances in Neural Information Processing Systems 31, pages 2451–2463, 2018. URL https: //arxiv.org/abs/1809.01999

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    Mastering Diverse Domains through World Models

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy P. Lillicrap. Mastering diverse domains through world models.CoRR, abs/2301.04104, 2023. URL https://arxiv.org/ abs/2301.04104

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    Learning Interactive Real-World Simulators

    Sherry Yang, Yilun Du, Seyed Kamyar Seyed Ghasemipour, Jonathan Tompson, Leslie Pack Kaelbling, Dale Schuurmans, and Pieter Abbeel. Learning interactive real-world simulators. In The Twelfth International Conference on Learning Representations, ICLR 2024, 2024. URL https://arxiv.org/abs/2310.06114

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    Perceiver IO: A General Architecture for Structured Inputs & Outputs

    Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier J. Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, and João Carreira. Perceiver IO: A general architecture for structured inputs & outputs. InThe Tenth International Confere...

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  27. [27]

    Fieldperceiver: Domain agnostic transformer model to predict multiscale physical fields and nonlinear material properties through neural ologs.Materials Today, 57: 9–25, 2022

    Markus J Buehler. Fieldperceiver: Domain agnostic transformer model to predict multiscale physical fields and nonlinear material properties through neural ologs.Materials Today, 57: 9–25, 2022. doi: 10.1016/j.mattod.2022.05.020

    work page doi:10.1016/j.mattod.2022.05.020 2022

  28. [28]

    Solving high- dimensional pdes with latent spectral models

    Haixu Wu, Tengge Hu, Huakun Luo, Jianmin Wang, and Mingsheng Long. Solving high- dimensional pdes with latent spectral models. InInternational Conference on Machine Learning, ICML 2023, Proceedings of Machine Learning Research, pages 37417–37438. PMLR, 2023. URLhttps://arxiv.org/abs/2301.12664

    work page arXiv 2023

  29. [29]

    Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators

    Benedikt Alkin, Andreas Fürst, Simon Schmid, Lukas Gruber, Markus Holzleitner, and Johannes Brandstetter. Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators. InAdvances in Neural Information Processing Systems 37, 2024. URL https: //arxiv.org/abs/2402.12365

    work page arXiv 2024

  30. [30]

    Wang, Etienne Le Naour, Jean-Noël Vittaut, and Patrick Gallinari

    Louis Serrano, Thomas X. Wang, Etienne Le Naour, Jean-Noël Vittaut, and Patrick Gallinari. AROMA: Preserving spatial structure for latent PDE modeling with local neural fields. In Advances in Neural Information Processing Systems 37, NeurIPS 2024, 2024. URL https: //arxiv.org/abs/2406.02176

    work page arXiv 2024

  31. [31]

    Latent neural operator for solving forward and inverse PDE problems

    Tian Wang and Chuang Wang. Latent neural operator for solving forward and inverse PDE problems. InAdvances in Neural Information Processing Systems 37, NeurIPS 2024, 2024. URLhttps://arxiv.org/abs/2406.03923. 11

    work page arXiv 2024

  32. [32]

    Fourier neural operator with learned deformations for pdes on general geometries.Journal of Machine Learning Research, 24:388:1–388:26, 2023

    Zongyi Li, Daniel Zhengyu Huang, Burigede Liu, and Anima Anandkumar. Fourier neural operator with learned deformations for pdes on general geometries.Journal of Machine Learning Research, 24:388:1–388:26, 2023. URLhttps://jmlr.org/papers/v24/23-0064.html

    2023

  33. [33]

    Battaglia

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W. Battaglia. Learning mesh-based simulation with graph networks. In9th International Conference on Learning Representations, ICLR 2021, 2021. URLhttps://arxiv.org/abs/2010.03409

    work page arXiv 2021

  34. [34]

    Levin, and Maria Shugrina

    Rishit Dagli, Donglai Xiang, Vismay Modi, Charles Loop, Clement Fuji Tsang, Anka He Chen, Anita Hu, Gavriel State, David I.W. Levin, and Maria Shugrina. V oMP: Predicting volumetric mechanical property fields.arXiv preprint arXiv:2510.22975, 2025. URL https: //arxiv.org/abs/2510.22975

    work page arXiv 2025

  35. [35]

    Genie: Generative interactive environments

    Jake Bruce, Michael Dennis, Ashley Edwards, Jack Parker-Holder, Yuge Shi, Edward Hughes, Matthew Lai, Aditi Mavalankar, Richie Steigerwald, Chris Apps, Yusuf Aytar, Sarah Bechtle, Feryal Behbahani, Stephanie Chan, Nicolas Heess, Lucy Gonzalez, Simon Osindero, Sherjil Ozair, Scott Reed, Jingwei Zhang, Konrad Zolna, Jeff Clune, Nando de Freitas, Satinder Si...

    2024

  36. [36]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In IEEE/CVF International Conference on Computer Vision, ICCV 2023, pages 4195–4205,

    2023

  37. [37]

    doi: 10.1109/ICCV51070.2023.00387

    work page doi:10.1109/iccv51070.2023.00387 2023

  38. [38]

    FiLM: Visual Reasoning with a General Conditioning Layer

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron C. Courville. FiLM: Visual reasoning with a general conditioning layer. InProceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, pages 3942–3951, 2018. URL https://arxiv.org/ abs/1709.07871

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  39. [39]

    Transolver++: An accurate neural solver for PDEs on million-scale geome- tries

    Huakun Luo, Haixu Wu, Hang Zhou, Lanxiang Xing, Yichen Di, Jianmin Wang, and Ming- sheng Long. Transolver++: An accurate neural solver for PDEs on million-scale geome- tries. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 41432–41449. PMLR, 2025. URL https://arxiv.org/ab...

    work page arXiv 2025

  40. [40]

    Differentiability in Unrolled Training of Neural Physics Simulators on Transient Dynamics.Computer Methods in Applied Mechanics and Engineering, 433:117441, 2025

    Bjoern List, Li-Wei Chen, Kartik Bali, and Nils Thuerey. Differentiability in Unrolled Training of Neural Physics Simulators on Transient Dynamics.Computer Methods in Applied Mechanics and Engineering, 433:117441, 2025. doi: 10.1016/j.cma.2024.117441

    work page doi:10.1016/j.cma.2024.117441 2025

  41. [41]

    Improving long-range interactions in graph neural simulators via Hamiltonian dynamics

    Tai Hoang, Alessandro Trenta, Alessio Gravina, Niklas Freymuth, Philipp Becker, Davide Bacciu, and Gerhard Neumann. Improving long-range interactions in graph neural simulators via Hamiltonian dynamics. InInternational Conference on Learning Representations, 2026. URLhttps://arxiv.org/abs/2511.08185

    work page arXiv 2026

  42. [42]

    SineNet: Learning temporal dynamics in time-dependent partial differential equations

    Xuan Zhang, Jacob Helwig, Yuchao Lin, Yaochen Xie, Cong Fu, Stephan Wojtowytsch, and Shuiwang Ji. SineNet: Learning temporal dynamics in time-dependent partial differential equations. InInternational Conference on Learning Representations, 2024. URL https: //arxiv.org/abs/2403.19507

    work page arXiv 2024

  43. [43]

    Sobolev Training for Neural Networks

    Wojciech M. Czarnecki, Simon Osindero, Max Jaderberg, Grzegorz Swirszcz, and Razvan Pascanu. Sobolev training for neural networks. InAdvances in Neural Information Processing Systems 30, pages 4278–4287, 2017. URLhttps://arxiv.org/abs/1706.04859

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  44. [44]

    Sobolev Training for Operator Learning

    Namkyeong Cho, Junseung Ryu, and Hyung Ju Hwang. Sobolev Training for Operator Learning. Journal of Computational Physics, 542:114319, 2025. doi: 10.1016/j.jcp.2025.114408

    work page doi:10.1016/j.jcp.2025.114408 2025

  45. [45]

    Sobolev acceleration for neural networks

    Jong Kwon Oh, Hanbaek Lyu, and Hwijae Son. Sobolev acceleration for neural networks. 2025. URLhttps://arxiv.org/abs/2509.19773. 12

    work page arXiv 2025

  46. [46]

    Derivative-Informed Fourier Neural Operator: Universal Approximation and Applications to PDE-Constrained Optimization

    Boyuan Yao, Dingcheng Luo, Lianghao Cao, Nikola Kovachki, Thomas O’Leary-Roseberry, and Omar Ghattas. Derivative-Informed Fourier Neural Operator: Universal Approximation and Applications to PDE-Constrained Optimization. 2025. URLhttps://arxiv.org/abs/ 2512.14086

    work page arXiv 2025

  47. [47]

    Learning Memory and Material Dependent Constitutive Laws

    Kaushik Bhattacharya, Lianghao Cao, George Stepaniants, Andrew Stuart, and Margaret Trautner. Learning Memory and Material Dependent Constitutive Laws. 2025. URL https: //arxiv.org/abs/2502.05463

    work page arXiv 2025

  48. [48]

    Derivative-enhanced deep operator network

    Yuan Qiu, Nolan Bridges, and Peng Chen. Derivative-enhanced deep operator network. In Advances in Neural Information Processing Systems 37, NeurIPS 2024, 2024. URL https: //arxiv.org/abs/2402.19242

    work page arXiv 2024

  49. [49]

    Gradient-enhanced physics- informed neural networks for forward and inverse PDE problems.Computer Methods in Applied Mechanics and Engineering, 393:114823, 2022

    Jeremy Yu, Lu Lu, Xuhui Meng, and George Em Karniadakis. Gradient-enhanced physics- informed neural networks for forward and inverse PDE problems.Computer Methods in Applied Mechanics and Engineering, 393:114823, 2022. doi: 10.1016/j.cma.2022.114823

    work page doi:10.1016/j.cma.2022.114823 2022

  50. [50]

    Gemnet: Universal directional graph neural networks for molecules

    Johannes Gasteiger, Florian Becker, and Stephan Günnemann. Gemnet: Universal directional graph neural networks for molecules. InAdvances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, pages 6790–6802, 2021. URL https://proceedings.neurips.cc/ paper/2...

    2021

  51. [51]

    Price, I., Sanchez-Gonzalez, A., Alet, F., Ewalds, T., El- Kadi, A., Stott, J., Mohamed, S., Battaglia, P

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd E. Zickler, Jonathan T. Barron, and Pratul P. Srinivasan. Ref-nerf: Structured view-dependent appearance for neural radi- ance fields. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 5481–5490. IEEE, 2022. doi: 10.1109/CVPR52688.2022.00541

    work page doi:10.1109/cvpr52688.2022.00541 2022

  52. [52]

    Stress field prediction in cantilevered structures using convolutional neural networks.Journal of Computing and Information Science in Engineering, 20, 2020

    Zhenguo Nie, Haoliang Jiang, and Levent Burak Kara. Stress field prediction in cantilevered structures using convolutional neural networks.Journal of Computing and Information Science in Engineering, 20, 2020. doi: 10.1115/1.4044097

    work page doi:10.1115/1.4044097 2020

  53. [53]

    Siboni, and Dierk Raabe

    Jaber Rezaei Mianroodi, Nima H. Siboni, and Dierk Raabe. Teaching solid mechanics to artificial intelligence—a fast solver for heterogeneous materials.npj Computational Materials, 7, 2021. doi: 10.1038/s41524-021-00571-z

    work page doi:10.1038/s41524-021-00571-z 2021

  54. [54]

    Anoop Krishnan

    Meer Mehran Rashid, Tanu Pittie, Souvik Chakraborty, and N.M. Anoop Krishnan. Learning the stress-strain fields in digital composites using Fourier neural operator.iScience, 25:105452,

  55. [55]

    doi: 10.1016/j.isci.2022.105452

    work page doi:10.1016/j.isci.2022.105452 2022

  56. [56]

    Khorrami, Jaber R

    Mohammad S. Khorrami, Jaber R. Mianroodi, Nima H. Siboni, Pawan Goyal, Bob Svend- sen, Peter Benner, and Dierk Raabe. An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials.npj Computational Materials, 9, 2023. doi: 10.1038/s41524-023-00991-z

    work page doi:10.1038/s41524-023-00991-z 2023

  57. [57]

    Abueidda, Seid Koric, Nahil A

    Diab W. Abueidda, Seid Koric, Nahil A. Sobh, and Huseyin Sehitoglu. Deep learning for plasticity and thermo-viscoplasticity.International Journal of Plasticity, 136:102852, 2021. doi: 10.1016/j.ijplas.2020.102852

    work page doi:10.1016/j.ijplas.2020.102852 2021

  58. [58]

    Junyan He, Seid Koric, Shashank Kushwaha, Jaewan Park, Diab Abueidda, and Iwona Jasiuk. Novel DeepONet architecture to predict stresses in elastoplastic structures with variable complex geometries and loads.Computer Methods in Applied Mechanics and Engineering, 415:116277,

  59. [59]

    doi: 10.1016/j.cma.2023.116277

    work page doi:10.1016/j.cma.2023.116277 2023

  60. [60]

    Junyan He, Seid Koric, Diab Abueidda, Ali Najafi, and Iwona Jasiuk. Geom-DeepONet: A point-cloud-based deep operator network for field predictions on 3D parameterized ge- ometries.Computer Methods in Applied Mechanics and Engineering, 429:117130, 2024. doi: 10.1016/j.cma.2024.117130

    work page doi:10.1016/j.cma.2024.117130 2024

  61. [61]

    Geometry-Informed Neural Operator Transformer

    Qibang Liu, Weiheng Zhong, Hadi Meidani, Diab Abueidda, Seid Koric, and Philippe Geubelle. Geometry-Informed Neural Operator Transformer. 2025. URL https://arxiv.org/abs/ 2504.19452. 13

    work page arXiv 2025

  62. [62]

    A comprehensive comparison of neural operators for 3D industry-scale engineering designs, 2025

    Weiheng Zhong, Qibang Liu, Diab Abueidda, Seid Koric, and Hadi Meidani. A comprehensive comparison of neural operators for 3D industry-scale engineering designs, 2025. URL https: //arxiv.org/abs/2510.05995

    work page arXiv 2025

  63. [63]

    Khorrami, Pawan Goyal, Jaber R

    Mohammad S. Khorrami, Pawan Goyal, Jaber R. Mianroodi, Bob Svendsen, Peter Benner, and Dierk Raabe. A physics-encoded Fourier neural operator approach for surrogate modeling of divergence-free stress fields in solids. 2024. URLhttps://arxiv.org/abs/2408.15408

    work page arXiv 2024

  64. [64]

    Iparraguirre, Iciar Alfaro, David Gonzalez, and Elias Cueto

    Mikel M. Iparraguirre, Iciar Alfaro, David Gonzalez, and Elias Cueto. MeshGraphNet- Transformer: Scalable mesh-based learned simulation for solid mechanics. 2026. URL https://arxiv.org/abs/2601.23177

    work page arXiv 2026

  65. [65]

    A new family of Constitutive Artificial Neural Networks towards automated model discovery.Computer Methods in Applied Mechanics and Engineering, 403: 115731, 2023

    Kevin Linka and Ellen Kuhl. A new family of Constitutive Artificial Neural Networks towards automated model discovery.Computer Methods in Applied Mechanics and Engineering, 403: 115731, 2023. doi: 10.1016/j.cma.2022.115731

    work page doi:10.1016/j.cma.2022.115731 2023

  66. [66]

    Automating modeling in mechanics: LLMs as designers of physics- constrained neural networks for constitutive modeling of materials

    Marius Tacke, Matthias Busch, Kian Abdolazizi, Jonas Eichinger, Kevin Linka, Christian Cyron, and Roland Aydin. Automating modeling in mechanics: LLMs as designers of physics- constrained neural networks for constitutive modeling of materials. 2025. URL https: //arxiv.org/abs/2512.01735

    work page arXiv 2025

  67. [67]

    GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training

    Haixu Wu, Minghao Guo, Zongyi Li, Zhiyang Dou, Mingsheng Long, Kaiming He, and Wojciech Matusik. GeoPT: Scaling physics simulation via lifted geometric pre-training. 2026. URLhttps://arxiv.org/abs/2602.20399

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  68. [68]

    Automotive Crash Dynamics Modeling Accelerated with Machine Learning

    Mohammad Amin Nabian, Sudeep Chavare, Deepak Akhare, Rishikesh Ranade, Ram Cherukuri, and Srinivas Tadepalli. Automotive Crash Dynamics Modeling Accelerated with Machine Learning. 2025. URLhttps://arxiv.org/abs/2510.15201

    work page arXiv 2025

  69. [69]

    P3D: Highly scalable 3D neural surrogates for physics simulations with global context

    Benjamin Holzschuh, Georg Kohl, Florian Redinger, and Nils Thuerey. P3D: Highly scalable 3D neural surrogates for physics simulations with global context. InInternational Conference on Learning Representations, 2026. URLhttps://arxiv.org/abs/2509.10186

    work page arXiv 2026

  70. [70]

    Asano, and Max Welling

    Winfried van den Dool, Maksim Zhdanov, Yuki M. Asano, and Max Welling. Adaptive mesh- quantization for neural PDE solvers, 2025. URLhttps://arxiv.org/abs/2511.18474

    work page arXiv 2025

  71. [71]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. InAdvances in Neural Informa- tion Processing Systems, volume 30. Curran Associates, Inc., 2017. URL https://papers. nips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html

    2017

  72. [72]

    Stewart and Lallit Anand

    Eric M. Stewart and Lallit Anand. A large deformation viscoelasticity theory for elas- tomeric materials and its numerical implementation in the open-source finite element program FEniCSx.International Journal of Solids and Structures, 303:113023, 2024. doi: 10.1016/j.ijsolstr.2024.113023

    work page doi:10.1016/j.ijsolstr.2024.113023 2024

  73. [73]

    nodes": [[0.24, 0.61, 0.46], [0.15, 0.18, 0.19], [0.44, 0.57, 0.85], [0.64, 0.12, 0.51]],

    Johannes Brandstetter, Daniel E. Worrall, and Max Welling. Message passing neural PDE solvers. InThe Tenth International Conference on Learning Representations, ICLR 2022, 2022. URLhttps://arxiv.org/abs/2202.03376. 14 A Constitutive Models and FEM Solver We consider finite deformation solid mechanics on a body in reference configurationΩ0. The dis- placem...

    work page arXiv 2022