pith. sign in

arxiv: 2605.15305 · v4 · pith:X4GIYMAWnew · submitted 2026-05-14 · 💻 cs.GR · cs.LG

WorldParticle: Unified World Simulation of Lagrangian Particle Dynamics via Transformer

Caoliwen Wang , Minghao Guo , Siyuan Chen , Heng Zhang , Mengdi Wang , Xingyu Ni , Hanson Sun , Kunyi Wang
show 8 more authors
Zherong Pan Kui Wu Lingjie Liu Yin Yang Chenfanfu Jiang Taku Komura Wojciech Matusik Peter Yichen Chen
This is my paper

Pith reviewed 2026-05-21 08:11 UTC · model grok-4.3

classification 💻 cs.GR cs.LG
keywords unified particle simulationtransformer architectureLagrangian dynamicsprediction-correctiontoken mergingphysical modelinggeneralization
0
0 comments X

The pith

A single transformer architecture unifies simulation of cloth, elastic solids, fluids, granular materials, and molecular dynamics on Lagrangian particles.

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

The paper sets out to build one learning-based simulator that covers six distinct physical dynamics categories instead of requiring separate solvers for each. It advances particles first with an explicit predictor that applies known external forces, then applies a learned transformer corrector to compute the missing interaction effects. The corrector works in three stages: a tokenizer that captures local particle and boundary interactions, a super-token encoder that repeatedly merges tokens to cut the number in half at each layer, and a decoder that uses cross-attention from the compact super-token set to output per-particle corrections. Because the same trained model handles all categories and still works on new materials, boundaries, and forces, the approach removes the need to redesign or retrain a solver when the physics changes. A sympathetic reader would care because this reduces repeated engineering work whenever a new phenomenon or scene must be simulated.

Core claim

The paper claims that a prediction-correction scheme on a shared Lagrangian particle representation, driven by a single transformer corrector, suffices to model cloth, elastic solids, Newtonian and non-Newtonian fluids, granular materials, and molecular dynamics. The corrector consists of a particle tokenizer encoding local particle-particle, particle-boundary, and topology interactions; a super-token encoder that alternates self-attention with progressive merging to halve the token count at each level; and a super-token decoder that lifts the compact representation back to full particle resolution via cross-attention to predict residual position and velocity updates. The resulting model, on

What carries the argument

The learned corrector that tokenizes particles, hierarchically merges them into super tokens through alternating self-attention and token merging, then decodes corrections via cross-attention from the reduced set.

If this is right

  • The same architecture generalizes across the six dynamics categories to unseen materials, boundary configurations, initial conditions, and external forces.
  • The model supports downstream tasks such as interactive control, inverse design, and learning from real-world manipulation data.
  • Progressive token merging reduces attention cost by halving the token count at each successive encoder layer.
  • The approach reduces the need for per-phenomenon solver engineering.

Where Pith is reading between the lines

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

  • If the corrector generalizes as claimed, the same trained weights could be applied to mixed scenes containing several of the six phenomena at once without additional engineering.
  • The hierarchical merging pattern could be reused in other large-scale particle systems where full pairwise attention is too expensive.
  • Training on combined data from multiple dynamics categories might improve robustness to entirely new physics not present in the original training set.
  • The design opens a route to real-time graphics applications that currently switch between separate simulators.

Load-bearing premise

The learned corrector can accurately capture and generalize inter-particle interactions for all six dynamics categories without per-category retraining or architectural changes.

What would settle it

Apply the trained model to a previously unseen material property or boundary configuration drawn from one of the six categories and check whether the predicted particle trajectories deviate significantly from reference solver outputs.

Figures

Figures reproduced from arXiv: 2605.15305 by Caoliwen Wang, Chenfanfu Jiang, Hanson Sun, Heng Zhang, Kui Wu, Kunyi Wang, Lingjie Liu, Mengdi Wang, Minghao Guo, Peter Yichen Chen, Siyuan Chen, Taku Komura, Wojciech Matusik, Xingyu Ni, Yin Yang, Zherong Pan.

Figure 1
Figure 1. Figure 1: We propose a unified transformer-based neural simulator for Lagrangian particle dynamics. Left: Our model handles diverse particle systems, including proteins, elastic solids, fluids, and cloth, within a single architecture. Right: The learned simulator supports downstream tasks including interactive control, inverse design, and learning from real-world observations. ∗Equal contribution †Corresponding auth… view at source ↗
Figure 2
Figure 2. Figure 2: Prediction-correction particle transformer. Given the intermediate state from the prediction step (Eq. 1), the correction step proceeds in three stages within one timestep: the particle tokenizer encodes local interactions into per-particle tokens, the super-token encoder compresses them into dynamically generated super tokens via self-attention and token merging, and the super-token decoder refines partic… view at source ↗
Figure 3
Figure 3. Figure 3: Token merging visualization. Particle tokens (blue spheres, left) are merged into super tokens (yellow cubes, right). Cube size and color intensity reflect how many original particles each super token represents. methods and reduced-order models [Chang et al. 2025; Chen et al. 2023; Fulton et al. 2019], but the weights here are feature-dependent and recomputed for every timestep rather than fixed by geomet… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results across five of six dynamics categories: Newtonian fluids, cloth, granular sand, non-Newtonian fluids, and elastic solids. Example sequences show Newtonian fluids flowing with varying viscosity and obstacle placements in a container, cloth colliding with a sphere, granular sand collapsing with varied initial shapes and friction, non-Newtonian fluids with varying rheological properties on… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation on local and global communication. Three variants on the fluid task: w/o the super-token encoder-decoder (local only), w/o the tokenizer’s neighborhood branches (global only), and the full model. Removing either component degrades rollout quality. 4.3 Ablation Studies The ablations below isolate two core design choices: the interplay between local tokenization and global super-token communication,… view at source ↗
Figure 8
Figure 8. Figure 8: Generalization to unseen initial and boundary conditions. First row: the training sequence whose initial configuration is closest to the test case in the second row. Orange boxes highlight differences in fluid volume and container geometry. Despite substantial differences, the model produces a stable 800-frame rollout that closely tracks the ground truth. Frame=150 Frame=300 w/o Obstacle Feature w/ Obstacl… view at source ↗
Figure 6
Figure 6. Figure 6: Side-chain rotation prediction on proteins 1a62_A and 16pk_A. Starting from different initial conformations (orange), the model predicts side-chain rotational dynamics (blue) at 50 fs timesteps, 100× larger than the 0.5 fs steps used in traditional simulations. Resolution Twist Speed Frame=200 Frame=250 Resolution Twist Speed [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 10
Figure 10. Figure 10: Interactive force control. Left: user-specified force parameters (application region, direction, and duration). Right: three examples of the resulting model rollouts under forces not seen during training. Lift Cloth Stretch Zebra Fold Rope [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 7
Figure 7. Figure 7: Long rollouts and unseen actuation. Cloth-twisting sequences under actuation strengths not seen during training. The model remains stable up to 250 frames, beyond the 200-frame training horizon. Initial Condition Frame=50 GT Ours Frame=200 GT Ours Frame=800 GT Ours [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 12
Figure 12. Figure 12: Generalization to particle sampling density. The fluid model, trained on approximately 7k particles, produces stable rollouts under altered initial particle arrangements (7k) and increased particle counts (8k, 9k). Optimized Initial Guess Iteration 0.232 0.236 0.240 0.244 0.248 Friction 0 100 200 300 400 500 600 700 800 900 Iteration 0.000 0.003 0.005 0.008 0.010 0.013 0.015 0.018 Energy 0 100 200 300 400… view at source ↗
Figure 15
Figure 15. Figure 15: Baseline comparison on fluid, cloth, and sand. Rollout se￾quences for DeepLagrangian [Ummenhofer et al. 2020], GNS [Sanchez￾Gonzalez et al. 2020], Neural Operator [Li et al. 2021; Viswanath et al. 2024], and our model, compared against ground truth (GT). Reported errors are auto-regressive rollout MSE for position and velocity, averaged over parti￾cles, frames, and test sequences. 0.000 0.003 0.005 0.009 … view at source ↗
Figure 16
Figure 16. Figure 16: Decoder attention patterns. (a) Influence maps of selected super tokens on particles at two frames across 8 decoder layers; warmer colors indicate stronger influence. Each super token affects spatially distributed particles, reflecting global coupling. (b) Cross-attention heat maps from decoder layers 1, 3, and 8 (rows: super tokens, columns: particles). Horizontal high-response bands indicate that super … view at source ↗
read the original abstract

A unified simulator that can model diverse physical phenomena without solver-specific redesign is a long-standing goal across simulation science. We present a learning-based particle simulator built on a single transformer architecture to model cloth, elastic solds, Newtonian and non-Newtonian fluids, granular materials, and molecular dynamics. Our model follows a prediction-correction design on a shared Lagrangian particle representation. An explicit predictor first advances particles under the known external forces, producing an intermediate state that captures externally driven motion but not inter-particle interactions. A learned corrector then predicts the residual position and velocity updates through three stages: a particle tokenizer that encodes local particle-particle, particle-boundary, and topology-guided interactions; a super-token encoder that hierarchically merges particle tokens into a compact set of super tokens via alternating self-attention and token merging; and a super-token decoder that lifts these super tokens back to particle resolution through cross-attention to predict per-particle position and velocity corrections. Progressive token merging reduces the attention cost at successive encoder layers by halving the token count at each level, and the decoder communicates through the compact super-token set rather than full particle-to-particle attention. Across the six dynamics categories, the same architecture generalizes to unseen materials, boundary configurations, initial conditions, and external forces. We further demonstrate downstream interactive control, inverse design, and learning from real-world manipulation data, reducing the need for per-phenomenon solver engineering.

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 / 2 minor

Summary. The manuscript presents WorldParticle, a single transformer-based architecture for unified Lagrangian particle simulation of six dynamics categories (cloth, elastic solids, Newtonian/non-Newtonian fluids, granular materials, and molecular dynamics). It uses a prediction-correction scheme: an explicit predictor advances particles under external forces, while a learned corrector employs a particle tokenizer for local and topology-guided interactions, a super-token encoder with alternating self-attention and progressive halving merges, and a cross-attention decoder to predict per-particle residuals. The central claim is that this fixed architecture generalizes to unseen materials, boundary configurations, initial conditions, and external forces, enabling downstream tasks such as interactive control, inverse design, and learning from real-world data without per-phenomenon solver redesign.

Significance. If the generalization claims hold with strong empirical support, the work would represent a meaningful advance toward solver-agnostic simulation in graphics and physics, potentially reducing engineering overhead across disparate phenomena. The hierarchical token-merging strategy for computational efficiency and the shared architecture for diverse physics would be notable if validated; downstream applications in control and inverse problems add practical value. However, the absence of quantitative benchmarks in the core description limits immediate assessment of impact.

major comments (2)
  1. [Abstract] Abstract and architecture description: the generalization claim across all six dynamics to unseen materials, boundaries, initial conditions, and forces is load-bearing for the paper's contribution, yet no error metrics, baselines, ablation studies, or quantitative validation results are referenced to support it. Without these, the soundness of the central claim cannot be evaluated.
  2. [super-token encoder] Super-token encoder description: progressive token merging that halves the token count at each level implicitly assumes salient interaction information survives aggressive downsampling. This assumption is critical for the learned corrector's ability to capture short-range potentials (molecular dynamics) or contact networks (granular flow) and to generalize without per-category changes; loss of local structure would directly undermine the uniform-architecture claim.
minor comments (2)
  1. [Abstract] Abstract contains a typo: 'elastic solds' should read 'elastic solids'.
  2. Notation for the three-stage corrector (tokenizer, encoder, decoder) is introduced without an accompanying diagram or pseudocode, which would clarify the data flow from particle tokens to super-tokens and back.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our generalization claims and architectural assumptions. We respond to each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract and architecture description: the generalization claim across all six dynamics to unseen materials, boundaries, initial conditions, and forces is load-bearing for the paper's contribution, yet no error metrics, baselines, ablation studies, or quantitative validation results are referenced to support it. Without these, the soundness of the central claim cannot be evaluated.

    Authors: We agree that the abstract would benefit from explicit references to supporting quantitative evidence. The manuscript body (Sections 4–6) reports error metrics (position and velocity L2 errors), baseline comparisons against per-phenomenon particle simulators, and ablation studies isolating the tokenizer, hierarchical encoder, and decoder across all six dynamics categories, including generalization tests on unseen materials, boundaries, initial conditions, and forces. We will revise the abstract to cite these key quantitative results and performance highlights. revision: yes

  2. Referee: [super-token encoder] Super-token encoder description: progressive token merging that halves the token count at each level implicitly assumes salient interaction information survives aggressive downsampling. This assumption is critical for the learned corrector's ability to capture short-range potentials (molecular dynamics) or contact networks (granular flow) and to generalize without per-category changes; loss of local structure would directly undermine the uniform-architecture claim.

    Authors: The referee correctly identifies that preservation of local structure is essential. Our design addresses this by performing local encoding in the particle tokenizer at full resolution prior to any merging; each super-token encoder layer then applies self-attention before halving, enabling aggregation of salient features. Experiments on molecular dynamics and granular materials confirm that short-range potentials and contact networks are captured accurately under the shared architecture. We will add a dedicated paragraph and supporting visualizations in the method section to explicitly discuss how local information is retained through the hierarchy. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected; model is a trained architecture with empirical generalization claims

full rationale

The paper proposes a single transformer architecture for unified particle simulation across six dynamics categories using an explicit predictor for external forces followed by a learned corrector. The corrector consists of a particle tokenizer, hierarchical super-token encoder with progressive merging, and cross-attention decoder. These components are trained on external data to predict residuals, and claims of generalization to unseen materials, boundaries, and forces rest on empirical results rather than any derivation that reduces by construction to fitted inputs or self-citations. No equations or steps in the provided description equate predictions to inputs tautologically, and the token-merging design is an efficiency choice whose validity is testable via performance on held-out cases. The derivation chain is self-contained as a data-driven proposal without load-bearing self-referential reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit information on free parameters, axioms, or invented entities; all such elements remain unidentified.

pith-pipeline@v0.9.0 · 5837 in / 1190 out tokens · 61364 ms · 2026-05-21T08:11:47.158406+00:00 · methodology

Review history (2 revisions) →

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

96 extracted references · 96 canonical work pages · 4 internal anchors

  1. [1]

    2019 , booktitle=

    Decoupled Weight Decay Regularization , author=. 2019 , booktitle=

    work page 2019

  2. [2]

    2024 , booktitle=

    FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning , author=. 2024 , booktitle=

    work page 2024

  3. [3]

    and Lin, Huancheng and Komura, Taku , title =

    Huang, Kemeng and Chitalu, Floyd M. and Lin, Huancheng and Komura, Taku , title =. 2024 , issue_date =. doi:10.1145/3643028 , journal =

    work page doi:10.1145/3643028 2024

  4. [4]

    2025 , publisher =

    Huang, Kemeng and Lu, Xinyu and Lin, Huancheng and Komura, Taku and Li, Minchen , title =. 2025 , issue_date =. doi:10.1145/3735126 , journal =

    work page doi:10.1145/3735126 2025

  5. [5]

    2017 , booktitle=

    Attention Is All You Need , author=. 2017 , booktitle=

    work page 2017

  6. [6]

    2016 , booktitle=

    Deep Residual Learning for Image Recognition , author=. 2016 , booktitle=

    work page 2016

  7. [7]

    2020 , eprint=

    GLU Variants Improve Transformer , author=. 2020 , eprint=

    work page 2020

  8. [8]

    2020 , booktitle=

    Query-Key Normalization for Transformers , author=. 2020 , booktitle=

    work page 2020

  9. [9]

    2024 , issue_date =

    Su, Jianlin and Ahmed, Murtadha and Lu, Yu and Pan, Shengfeng and Bo, Wen and Liu, Yunfeng , title =. 2024 , issue_date =. doi:10.1016/j.neucom.2023.127063 , journal =

    work page doi:10.1016/j.neucom.2023.127063 2024

  10. [10]

    2019 , booktitle=

    Root Mean Square Layer Normalization , author=. 2019 , booktitle=

    work page 2019

  11. [11]

    Computer Graphics Forum , volume=

    Deep Fluids: A Generative Network for Parameterized Fluid Simulations , author=. Computer Graphics Forum , volume=. 2019 , doi=

    work page 2019

  12. [12]

    Computer Graphics Forum , volume=

    Latent Space Physics: Towards Learning the Temporal Evolution of Fluid Flow , author=. Computer Graphics Forum , volume=. 2019 , doi=

    work page 2019

  13. [13]

    2023 , booktitle=

    CROM: Continuous Reduced-Order Modeling of PDEs Using Implicit Neural Representations , author=. 2023 , booktitle=

    work page 2023

  14. [14]

    and Carlberg, Kevin and Grinspun, Eitan , title =

    Chang, Yue and Chen, Peter Yichen and Wang, Zhecheng and Chiaramonte, Maurizio M. and Carlberg, Kevin and Grinspun, Eitan , title =. 2023 , publisher =. doi:10.1145/3610548.3618158 , booktitle =

    work page doi:10.1145/3610548.3618158 2023

  15. [15]

    and Chen, Peter Yichen and Grinspun, Eitan , title =

    Chang, Yue and Benchekroun, Otman and Chiaramonte, Maurizio M. and Chen, Peter Yichen and Grinspun, Eitan , title =. 2025 , issue_date =. doi:10.1145/3731148 , journal =

    work page doi:10.1145/3731148 2025

  16. [16]

    2025 , publisher =

    Liu, Mengfei and Chang, Yue and Wang, Zhecheng and Chen, Peter Yichen and Grinspun, Eitan , title =. 2025 , publisher =. doi:10.1145/3757377.3763810 , booktitle =

    work page doi:10.1145/3757377.3763810 2025

  17. [17]

    2025 , issue_date =

    Chen, Siyuan and Chen, Yixin and Panuelos, Jonathan and Benchekroun, Otman and Chang, Yue and Grinspun, Eitan and Wang, Zhecheng , title =. 2025 , issue_date =. doi:10.1145/3730826 , journal =

    work page doi:10.1145/3730826 2025

  18. [18]

    2026 , eprint=

    Low-Rank Koopman Deformables with Log-Linear Time Integration , author=. 2026 , eprint=

    work page 2026

  19. [19]

    2026 , eprint=

    Factorized Neural Implicit DMD for Parametric Dynamics , author=. 2026 , eprint=

    work page 2026

  20. [20]

    2026 , eprint=

    Odd-DC: Generalizable Neural Model Reduction via Odd Difference-of-Convex Structure , author=. 2026 , eprint=

    work page 2026

  21. [21]

    2016 , booktitle=

    Interaction Networks for Learning about Objects, Relations and Physics , author=. 2016 , booktitle=

    work page 2016

  22. [22]

    2019 , booktitle=

    Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids , author=. 2019 , booktitle=

    work page 2019

  23. [23]

    2020 , booktitle=

    Lagrangian Fluid Simulation with Continuous Convolutions , author=. 2020 , booktitle=

    work page 2020

  24. [24]

    , title =

    Sanchez-Gonzalez, Alvaro and Godwin, Jonathan and Pfaff, Tobias and Ying, Rex and Leskovec, Jure and Battaglia, Peter W. , title =. Proceedings of the 37th International Conference on Machine Learning , articleno =. 2020 , publisher =

    work page 2020

  25. [25]

    2021 , booktitle=

    Learning Mesh-Based Simulation with Graph Networks , author=. 2021 , booktitle=

    work page 2021

  26. [26]

    2023 , booktitle=

    Learning Neural Constitutive Laws from Motion Observations for Generalizable PDE Dynamics , author=. 2023 , booktitle=

    work page 2023

  27. [27]

    and Matusik, Wojciech and Grinspun, Eitan and Carlberg, Kevin and Jiang, Chenfanfu and Chen, Peter Yichen , title =

    Zong, Zeshun and Li, Xuan and Li, Minchen and Chiaramonte, Maurizio M. and Matusik, Wojciech and Grinspun, Eitan and Carlberg, Kevin and Jiang, Chenfanfu and Chen, Peter Yichen , title =. 2023 , publisher =. doi:10.1145/3610548.3618207 , booktitle =

    work page doi:10.1145/3610548.3618207 2023

  28. [28]

    Modi, Vismay and Sharp, Nicholas and Perel, Or and Sueda, Shinjiro and Levin, David I. W. , title =. 2024 , issue_date =. doi:10.1145/3658184 , journal =

    work page doi:10.1145/3658184 2024

  29. [29]

    2021 , booktitle=

    Fourier Neural Operator for Parametric Partial Differential Equations , author=. 2021 , booktitle=

    work page 2021

  30. [30]

    2023 , booktitle=

    GNOT: A General Neural Operator Transformer for Operator Learning , author=. 2023 , booktitle=

    work page 2023

  31. [31]

    2024 , booktitle=

    Transolver: A Fast Transformer Solver for PDEs on General Geometries , author=. 2024 , booktitle=

    work page 2024

  32. [32]

    2024 , booktitle=

    Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators , author=. 2024 , booktitle=

    work page 2024

  33. [33]

    2021 , booktitle=

    Point Transformer , author=. 2021 , booktitle=

    work page 2021

  34. [34]

    2022 , booktitle=

    Point Transformer V2: Grouped Vector Attention and Partition-based Pooling , author=. 2022 , booktitle=

    work page 2022

  35. [35]

    2021 , booktitle=

    PoinTr: Diverse Point Cloud Completion with Geometry-Aware Transformers , author=. 2021 , booktitle=

    work page 2021

  36. [36]

    2022 , booktitle=

    Masked Autoencoders for Point Cloud Self-supervised Learning , author=. 2022 , booktitle=

    work page 2022

  37. [37]

    2023 , issue_date =

    Wang, Peng-Shuai , title =. 2023 , issue_date =. doi:10.1145/3592131 , journal =

    work page doi:10.1145/3592131 2023

  38. [38]

    2023 , eprint=

    Swin3D: A Pretrained Transformer Backbone for 3D Indoor Scene Understanding , author=. 2023 , eprint=

    work page 2023

  39. [39]

    2022 , booktitle=

    ViewFormer: NeRF-free Neural Rendering from Few Images Using Transformers , author=. 2022 , booktitle=

    work page 2022

  40. [40]

    2021 , booktitle=

    IBRNet: Learning Multi-View Image-Based Rendering , author=. 2021 , booktitle=

    work page 2021

  41. [41]

    2025 , booktitle=

    LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias , author=. 2025 , booktitle=

    work page 2025

  42. [42]

    2025 , publisher =

    Zeng, Chong and Dong, Yue and Peers, Pieter and Wu, Hongzhi and Tong, Xin , title =. 2025 , publisher =. doi:10.1145/3721238.3730595 , booktitle =

    work page doi:10.1145/3721238.3730595 2025

  43. [43]

    2026 , booktitle=

    P3D: Scalable Neural Surrogates for High-Resolution 3D Physics Simulations with Global Context , author=. 2026 , booktitle=

    work page 2026

  44. [44]

    2025 , booktitle=

    PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation , author=. 2025 , booktitle=

    work page 2025

  45. [45]

    2017 , booktitle=

    A Compositional Object-Based Approach to Learning Physical Dynamics , author=. 2017 , booktitle=

    work page 2017

  46. [46]

    2018 , booktitle=

    SPNets: Differentiable Fluid Dynamics for Deep Neural Networks , author=. 2018 , booktitle=

    work page 2018

  47. [47]

    2017 , booktitle=

    Accelerating Eulerian Fluid Simulation With Convolutional Networks , author=. 2017 , booktitle=

    work page 2017

  48. [48]

    2017 , issue_date =

    Chu, Mengyu and Thuerey, Nils , title =. 2017 , issue_date =. doi:10.1145/3072959.3073643 , journal =

    work page doi:10.1145/3072959.3073643 2017

  49. [49]

    , author Grinspun, E

    Xie, You and Franz, Aleksandra and Chu, Mengyu and Thuerey, Nils , title =. 2018 , issue_date =. doi:10.1145/3197517.3201304 , journal =

    work page doi:10.1145/3197517.3201304 2018

  50. [50]

    and Jameson, Antony and Kochenderfer, Mykel J

    Morton, Jeremy and Witherden, Freddie D. and Jameson, Antony and Kochenderfer, Mykel J. , title =. Proceedings of the 32nd International Conference on Neural Information Processing Systems , pages =. 2018 , publisher =

    work page 2018

  51. [51]

    2021 , issue_date =

    Bertiche, Hugo and Madadi, Meysam and Escalera, Sergio , title =. 2021 , issue_date =. doi:10.1145/3478513.3480479 , journal =

    work page doi:10.1145/3478513.3480479 2021

  52. [52]

    2022 , issue_date =

    Bertiche, Hugo and Madadi, Meysam and Escalera, Sergio , title =. 2022 , issue_date =. doi:10.1145/3550454.3555491 , journal =

    work page doi:10.1145/3550454.3555491 2022

  53. [53]

    2024 , booktitle=

    SENC: Handling Self-collision in Neural Cloth Simulation , author=. 2024 , booktitle=

    work page 2024

  54. [54]

    Journal of Computational Physics , volume=

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , author=. Journal of Computational Physics , volume=. 2019 , doi=

    work page 2019

  55. [55]

    Nature Machine Intelligence , volume=

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , author=. Nature Machine Intelligence , volume=. 2021 , doi=

    work page 2021

  56. [56]

    2020 , booktitle=

    Multipole Graph Neural Operator for Parametric Partial Differential Equations , author=. 2020 , booktitle=

    work page 2020

  57. [57]

    ACM/IMS Journal of Data Science , volume=

    Physics-Informed Neural Operator for Learning Partial Differential Equations , author=. ACM/IMS Journal of Data Science , volume=. 2024 , doi=

    work page 2024

  58. [58]

    2023 , booktitle=

    Geometry-Informed Neural Operator for Large-Scale 3D PDEs , author=. 2023 , booktitle=

    work page 2023

  59. [59]

    Transactions on Machine Learning Research , year=

    Multi-Grid Tensorized Fourier Neural Operator for High-Resolution PDEs , author=. Transactions on Machine Learning Research , year=

    work page

  60. [60]

    Transactions on Machine Learning Research , year=

    U-NO: U-shaped Neural Operators , author=. Transactions on Machine Learning Research , year=

    work page

  61. [61]

    2023 , booktitle=

    Factorized Fourier Neural Operators , author=. 2023 , booktitle=

    work page 2023

  62. [62]

    2023 , booktitle=

    Convolutional Neural Operators for Robust and Accurate Learning of PDEs , author=. 2023 , booktitle=

    work page 2023

  63. [63]

    2024 , booktitle=

    BENO: Boundary-Embedded Neural Operators for Elliptic PDEs , author=. 2024 , booktitle=

    work page 2024

  64. [64]

    2021 , booktitle=

    Choose a Transformer: Fourier or Galerkin , author=. 2021 , booktitle=

    work page 2021

  65. [65]

    Transactions on Machine Learning Research , year=

    Transformer for Partial Differential Equations' Operator Learning , author=. Transactions on Machine Learning Research , year=

    work page

  66. [66]

    2024 , booktitle=

    DPOT: Auto-Regressive Denoising Operator Transformer for Large-Scale PDE Pre-Training , author=. 2024 , booktitle=

    work page 2024

  67. [67]

    2024 , booktitle=

    Poseidon: Efficient Foundation Models for PDEs , author=. 2024 , booktitle=

    work page 2024

  68. [68]

    2024 , booktitle=

    Pretraining Codomain Attention Neural Operators for Solving Multiphysics PDEs , author=. 2024 , booktitle=

    work page 2024

  69. [69]

    2025 , booktitle=

    Geometry Aware Operator Transformer as an Efficient and Accurate Neural Surrogate for PDEs on Arbitrary Domains , author=. 2025 , booktitle=

    work page 2025

  70. [70]

    2022 , booktitle=

    Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling , author=. 2022 , booktitle=

    work page 2022

  71. [71]

    2024 , booktitle=

    Point Transformer V3: Simpler, Faster, Stronger , author=. 2024 , booktitle=

    work page 2024

  72. [72]

    The Newton Contributors , title =

    work page

  73. [73]

    2024 , issue_date =

    Chen, Anka He and Liu, Ziheng and Yang, Yin and Yuksel, Cem , title =. 2024 , issue_date =. doi:10.1145/3658179 , journal =

    work page doi:10.1145/3658179 2024

  74. [74]

    Bender, Jan and others , license =

    work page

  75. [75]

    Polyscope , author =

    work page

  76. [76]

    Nucleic Acids Research , volume =

    Vander Meersche, Yann and Cretin, Gabriel and Gheeraert, Aria and Gelly, Jean-Christophe and Galochkina, Tatiana , title =. Nucleic Acids Research , volume =. 2024 , month =

    work page 2024

  77. [77]

    2023 , eprint=

    OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials , author=. 2023 , eprint=

    work page 2023

  78. [78]

    Journal of Visual Communication and Image Representation , volume=

    Position based dynamics , author=. Journal of Visual Communication and Image Representation , volume=. 2007 , publisher=

    work page 2007

  79. [79]

    ACM Transactions on Graphics (TOG) , volume=

    A material point method for snow simulation , author=. ACM Transactions on Graphics (TOG) , volume=. 2013 , publisher=

    work page 2013

  80. [80]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Phystwin: Physics-informed reconstruction and simulation of deformable objects from videos , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

Showing first 80 references.