pith. sign in

arxiv: 2605.19589 · v1 · pith:6YVWAN4Rnew · submitted 2026-05-19 · 💻 cs.LG · physics.flu-dyn

Physics-Informed Graph Neural Network Surrogates for Turbulent Nanoparticle Dispersion in Dental Clinical Environments

Takshak Shende , Viktor Popov This is my paper

Pith reviewed 2026-05-20 06:36 UTC · model grok-4.3

classification 💻 cs.LG physics.flu-dyn
keywords graph neural networksphysics-informed learningaerosol dispersiondental aerosolssurrogate modelingparticle trackingturbulent flowOpenFOAM
0
0 comments X

The pith

ELGIN graph surrogate tracks dental aerosol clouds with lower error and 37 times the speed of full CFD on tested case.

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

The paper presents ELGIN, a physics-informed graph neural network that jointly predicts air flow on a clinic mesh and the motion of small spray particles. Traditional Reynolds-Averaged Navier-Stokes simulations with particle tracking give accurate results but take too long to support real-time decisions in three-dimensional clinic rooms. ELGIN replaces the slow solver with a multi-head Graph Transformer for the flow field, a Lagrangian Interaction Network for parcels, and a four-stage curriculum that keeps long autoregressive rollouts stable. On the single reported case it reduces mean parcel displacement error from 19.56 percent to 16.20 percent of room width and radius-of-gyration error from 9.85 percent to 6.58 percent, while finishing a 26-second rollout in roughly 64 seconds on a 4 GB GPU. The authors note that full training across the twenty-case sweep is still in progress and will replace the current single-case numbers.

Core claim

ELGIN couples a multi-head Graph Transformer equipped with Jacobi-preconditioned learnable pressure projection and turbulence closure to a sigmoid-gated Lagrangian Interaction Network through differentiable inverse-distance mesh-parcel coupling, then advances parcels with a symplectic Stormer-Verlet integrator; a four-stage physics-informed curriculum enables stable 260-step rollouts that match foam-extend reactingParcelFoam results more closely than a Lagrangian-only baseline while delivering a 37-fold wall-time reduction.

What carries the argument

Eulerian-Lagrangian Graph Interaction Network (ELGIN) that links graph-based Eulerian flow prediction on OpenFOAM polyhedral meshes to Lagrangian parcel motion via differentiable inverse-distance coupling and symplectic integration.

If this is right

  • Once the multi-case checkpoint exists, per-appointment infection-risk screening in full 3D clinic geometries becomes computationally feasible.
  • Different combinations of ventilation rate and handpiece spray speed can be evaluated in seconds rather than hours, supporting rapid protocol optimization.
  • Ensemble rollouts become practical, allowing uncertainty quantification in airborne pathogen transmission estimates.

Where Pith is reading between the lines

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

  • The same graph-coupling pattern could accelerate dispersion modeling inside other enclosed high-traffic spaces such as operating theaters or aircraft cabins.
  • Adding a lightweight inactivation module for pathogen viability would convert the present transport surrogate directly into a risk-prediction engine.
  • Online assimilation of room-sensor measurements could correct the surrogate in real time and reduce drift during long clinical sessions.

Load-bearing premise

That accuracy and speedup measured on the single Sweep_Case_03 will persist after full retraining on the twenty-case sweep and that the four-stage curriculum will continue to prevent gradient explosion at clinically relevant ventilation and spray speeds.

What would settle it

Complete 16/2/2 retraining of ELGIN on the full twenty-case sweep followed by evaluation on held-out cases; if mean parcel displacement error fails to stay below 17 percent of room width or wall-time speedup falls below 30 times the foam-extend baseline, the central claim does not hold.

Figures

Figures reproduced from arXiv: 2605.19589 by Takshak Shende, Viktor Popov.

Figure 1
Figure 1. Figure 1: FIG. 1. (a) Dental treatment room: 4 m [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Overview of the [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Rollout diagnostics for M0 and [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Particle cloud snapshots at [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Eulerian [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The box is centred on the dentist’s head height directly above the obstacle ridge of the [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Breathing Zone Exposure (BZE) fraction vs. time for [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Peak BZE vs. Air Changes per Hour (ACH) for all 20 CFD cases. Points coloured by spray velocity [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Non-dimensional mean-squared displacement vs. non-dimensional lag time [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
read the original abstract

Dental aerosol procedures produce sub-50 micrometre nuclei that can remain airborne for long periods in enclosed clinics, creating pathways for airborne pathogen transmission. Reynolds-Averaged Navier-Stokes (RANS) simulations with Euler-Lagrange particle tracking capture this transport accurately but require very long run times per scenario, which precludes real-time clinical decision support in 3D. We present the Eulerian-Lagrangian Graph Interaction Network (ELGIN), a physics-informed graph surrogate that jointly predicts carrier-flow dynamics on the OpenFOAM polyhedral mesh and the per-parcel motion of the polydisperse spray cloud. ELGIN couples a multi-head Graph Transformer with Jacobi-preconditioned learnable pressure projection and a turbulence-closure head to a sigmoid-gated Lagrangian Interaction Network through differentiable inverse-distance mesh-parcel coupling, and advances parcels with a symplectic Stormer-Verlet integrator. A four-stage physics-informed curriculum stabilises 260-step autoregressive rollouts without gradient explosion. A parameter sweep with foam-extend 4.1 OpenFOAM reactingParcelFoam across clinically relevant ventilation rates and handpiece spray speeds provides CFD ground truth. This article reports a single-case demonstration in which both ELGIN and a Lagrangian-only baseline (M0) are trained and evaluated on Sweep_Case_03 of a twenty-case sweep; full 16/2/2 retraining is in progress and will replace all reported metrics. On this case, ELGIN tracks the foam-extend particle cloud much more closely than M0: mean parcel displacement error falls from 19.56% to 16.20% of room width and cloud radius-of-gyration error from 9.85% to 6.58%. A 26-second rollout completes in ~64 s on a 4 GB GPU, approximately 37x faster than the foam-extend reference pipeline, toward per-appointment infection-risk screening once the multi-case checkpoint is in place.

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 the Eulerian-Lagrangian Graph Interaction Network (ELGIN), a physics-informed graph neural network surrogate for turbulent nanoparticle dispersion in dental clinics. It couples a multi-head Graph Transformer for carrier-flow prediction on OpenFOAM polyhedral meshes (with Jacobi-preconditioned pressure projection and turbulence closure) to a sigmoid-gated Lagrangian Interaction Network via differentiable inverse-distance coupling, advancing parcels with a symplectic Stormer-Verlet integrator. A four-stage physics-informed curriculum is used to stabilize 260-step autoregressive rollouts. Ground truth comes from a twenty-case foam-extend reactingParcelFoam parameter sweep over ventilation rates and handpiece speeds; the paper reports accuracy and speedup results only for the single case Sweep_Case_03 (with full 16/2/2 retraining noted as in progress) and shows ELGIN outperforming a Lagrangian-only baseline M0.

Significance. If the reported error reductions and 37x speedup generalize after full multi-case retraining, ELGIN could enable real-time per-appointment infection-risk screening by replacing computationally expensive RANS Euler-Lagrange simulations, addressing a key bottleneck for clinical decision support in enclosed dental environments.

major comments (2)
  1. [Abstract] Abstract: The headline metrics (mean parcel displacement error reduced from 19.56% to 16.20% of room width; cloud radius-of-gyration error from 9.85% to 6.58%) and the ~37x speedup (26 s rollout in ~64 s on 4 GB GPU) are obtained exclusively after training and rollout on Sweep_Case_03. Because the manuscript states that full 16/2/2 retraining on the twenty-case sweep “is in progress” and “will replace all reported metrics,” the central claim that ELGIN supplies a clinically usable surrogate rests on unverified generalization across ventilation and spray-speed variations.
  2. [Methods / Curriculum description] The four-stage physics-informed curriculum is asserted to prevent gradient explosion in longer autoregressive rollouts, yet no ablation, stability metrics, or rollout-error curves are supplied for regimes other than the single reported Sweep_Case_03; this leaves the stability claim for clinically relevant parameter ranges unsupported.
minor comments (2)
  1. [Abstract] Abstract: The exact numerical ranges of ventilation rates and handpiece spray speeds in the twenty-case sweep are not stated; supplying these bounds would strengthen the clinical-relevance argument.
  2. [Model architecture] Notation: The differentiable inverse-distance mesh-parcel coupling operator and the precise form of the sigmoid gate between the Eulerian and Lagrangian heads would benefit from an explicit equation or pseudocode block.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback and for highlighting the importance of demonstrating generalization in our work. We address each major comment below and describe the revisions we will implement.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline metrics (mean parcel displacement error reduced from 19.56% to 16.20% of room width; cloud radius-of-gyration error from 9.85% to 6.58%) and the ~37x speedup (26 s rollout in ~64 s on 4 GB GPU) are obtained exclusively after training and rollout on Sweep_Case_03. Because the manuscript states that full 16/2/2 retraining on the twenty-case sweep “is in progress” and “will replace all reported metrics,” the central claim that ELGIN supplies a clinically usable surrogate rests on unverified generalization across ventilation and spray-speed variations.

    Authors: We agree that the reported headline metrics and speedup are from the single Sweep_Case_03 demonstration and that the manuscript already states the full 16/2/2 retraining is in progress. To address this concern directly, we will revise the abstract and the opening of the results section to explicitly frame the current numbers as a preliminary single-case validation of the ELGIN architecture. The discussion will be updated to moderate claims about clinical usability until multi-case generalization is confirmed, while retaining the description of the ongoing retraining as the path to broader validation. revision: yes

  2. Referee: [Methods / Curriculum description] The four-stage physics-informed curriculum is asserted to prevent gradient explosion in longer autoregressive rollouts, yet no ablation, stability metrics, or rollout-error curves are supplied for regimes other than the single reported Sweep_Case_03; this leaves the stability claim for clinically relevant parameter ranges unsupported.

    Authors: The four-stage curriculum was selected after internal experiments to stabilize the 260-step autoregressive rollouts on the reported case. We acknowledge the absence of ablations and cross-regime stability curves. In the revised manuscript we will add an ablation table comparing curriculum stages and rollout-error curves specifically for Sweep_Case_03. Because the full multi-case retraining remains in progress, we cannot yet supply equivalent stability metrics across the entire twenty-case parameter sweep; this limitation will be stated explicitly in the revised text. revision: partial

standing simulated objections not resolved
  • Quantitative stability metrics, ablations, and generalization results across the full twenty-case ventilation/handpiece-speed sweep, as the 16/2/2 retraining is still underway and the corresponding checkpoints are not yet available.

Circularity Check

0 steps flagged

No significant circularity; performance metrics are empirical comparisons to independent OpenFOAM CFD ground truth.

full rationale

The paper trains both ELGIN and the M0 baseline on external foam-extend reactingParcelFoam data for Sweep_Case_03 and directly measures parcel displacement error reduction (19.56% to 16.20% of room width), gyration error reduction (9.85% to 6.58%), and 37x wall-clock speedup against the same reference pipeline. These quantities are obtained by forward simulation of the trained model and comparison to held-out CFD output; they do not reduce by construction to any fitted parameter, self-citation, or ansatz imported from prior work by the same authors. The four-stage curriculum and differentiable coupling are architectural choices whose effect is quantified externally rather than defined into the reported numbers. The single-case demonstration therefore remains self-contained against an independent benchmark, with full multi-case results noted as forthcoming but not required for the present claims.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on CFD simulations serving as ground truth and on the stability of the physics-informed training curriculum for autoregressive rollouts.

free parameters (1)
  • GNN hyperparameters and curriculum stage weights
    Learning rates, layer dimensions, and loss weighting factors chosen during training to stabilize 260-step rollouts.
axioms (1)
  • domain assumption RANS with Euler-Lagrange particle tracking on OpenFOAM polyhedral meshes provides accurate ground truth for carrier flow and polydisperse spray transport.
    Invoked as the source of training data and evaluation benchmark in the parameter sweep.

pith-pipeline@v0.9.0 · 5895 in / 1528 out tokens · 52799 ms · 2026-05-20T06:36:21.257097+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith.Foundation.RealityFromDistinction reality_from_one_distinction unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    ELGIN couples a multi-head Graph Transformer with Jacobi-preconditioned learnable pressure projection and a turbulence-closure head to a sigmoid-gated Lagrangian Interaction Network through differentiable inverse-distance mesh-parcel coupling, and advances parcels with a symplectic Störmer-Verlet integrator. A four-stage physics-informed curriculum stabilises 260-step autoregressive rollouts.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

124 extracted references · 124 canonical work pages · 7 internal anchors

  1. [1]

    author author S. K. \ Harrel \ and\ author J. Molinari ,\ 10.14219/jada.archive.2004.0207 journal journal Journal of the American Dental Association \ volume 135 ,\ pages 429 ( year 2004 ) NoStop

    work page doi:10.14219/jada.archive.2004.0207 2004

  2. [2]

    author author R. E. \ Micik , author R. L. \ Miller , author M. A. \ Mazzarella , \ and\ author G. Ryge ,\ 10.1177/00220345690480010501 journal journal Journal of Dental Research \ volume 48 ,\ pages 49 ( year 1969 ) NoStop

    work page doi:10.1177/00220345690480010501 1969

  3. [3]

    Li , author H

    author author Y. Li , author H. Qian , author J. Hang , author X. Chen , author P. Cheng , author H. Ling , author S. Wang , author P. Liang , author J. Li , author S. Xiao , author J. Wei , author L. Liu , author B. J. \ Cowling , \ and\ author M. Kang ,\ 10.1016/j.buildenv.2021.107788 journal journal Building and Environment \ volume 196 ,\ pages 107788...

    work page doi:10.1016/j.buildenv.2021.107788 2021

  4. [4]

    Zemouri , author H

    author author C. Zemouri , author H. de Soet , author W. Crielaard , \ and\ author A. Laheij ,\ 10.1371/journal.pone.0178007 journal journal PLoS ONE \ volume 12 ,\ pages e0178007 ( year 2017 ) NoStop

    work page doi:10.1371/journal.pone.0178007 2017

  5. [5]

    Morawska \ and\ author D

    author author L. Morawska \ and\ author D. K. \ Milton ,\ 10.1093/cid/ciaa939 journal journal Clinical Infectious Diseases \ volume 71 ,\ pages 2311 ( year 2020 ) NoStop

    work page doi:10.1093/cid/ciaa939 2020

  6. [6]

    \ Wang , author K

    author author C.-C. \ Wang , author K. A. \ Prather , author J. Sznitman , author J. L. \ Jimenez , author S. S. \ Lakdawala , author Z. Tufekci , \ and\ author L. C. \ Marr ,\ 10.1126/science.abd9149 journal journal Science \ volume 373 ,\ pages eabd9149 ( year 2021 ) NoStop

    work page doi:10.1126/science.abd9149 2021

  7. [7]

    author author J. H. \ Ferziger \ and\ author M. Peri\' c ,\ 10.1007/978-3-642-56026-2 title Computational Methods for Fluid Dynamics ,\ edition 3rd \ ed.\ ( publisher Springer ,\ address Berlin ,\ year 2002 ) NoStop

    work page doi:10.1007/978-3-642-56026-2 2002

  8. [8]

    author author F. R. \ Menter ,\ 10.2514/3.12149 journal journal AIAA Journal \ volume 32 ,\ pages 1598 ( year 1994 ) NoStop

    work page doi:10.2514/3.12149 1994

  9. [9]

    Cunningham ,\ 10.1098/rspa.1910.0024 journal journal Proceedings of the Royal Society of London A \ volume 83 ,\ pages 357 ( year 1910 ) NoStop

    author author E. Cunningham ,\ 10.1098/rspa.1910.0024 journal journal Proceedings of the Royal Society of London A \ volume 83 ,\ pages 357 ( year 1910 ) NoStop

    work page doi:10.1098/rspa.1910.0024 1910

  10. [10]

    author author S. B. \ Pope ,\ 10.1017/CBO9780511840531 title Turbulent Flows \ ( publisher Cambridge University Press ,\ address Cambridge ,\ year 2000 ) NoStop

    work page doi:10.1017/cbo9780511840531 2000

  11. [11]

    Einstein ,\ 10.1002/andp.19053220806 journal journal Annalen der Physik \ volume 322 ,\ pages 549 ( year 1905 ) NoStop

    author author A. Einstein ,\ 10.1002/andp.19053220806 journal journal Annalen der Physik \ volume 322 ,\ pages 549 ( year 1905 ) NoStop

    work page doi:10.1002/andp.19053220806 1905

  12. [12]

    author author J. K. \ Gupta , author C.-H. \ Lin , \ and\ author Q. Chen ,\ 10.1111/j.1600-0668.2009.00623.x journal journal Indoor Air \ volume 20 ,\ pages 31 ( year 2010 ) NoStop

    work page doi:10.1111/j.1600-0668.2009.00623.x 2009

  13. [13]

    Guo , author W

    author author X. Guo , author W. Li , \ and\ author F. Iorio ,\ in\ 10.1145/2939672.2939738 booktitle Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining \ ( year 2016 )\ pp.\ pages 481--490 NoStop

    work page doi:10.1145/2939672.2939738 2016

  14. [14]

    Smith, Ayya Alieva, Qing Wang, Michael P

    author author D. Kochkov , author J. A. \ Smith , author A. Alieva , author Q. Wang , author M. P. \ Brenner , \ and\ author S. Hoyer ,\ 10.1073/pnas.2101784118 journal journal Proceedings of the National Academy of Sciences \ volume 118 ,\ pages e2101784118 ( year 2021 ) NoStop

    work page doi:10.1073/pnas.2101784118 2021

  15. [16]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators

    author author L. Lu , author P. Jin , author G. Pang , author Z. Zhang , \ and\ author G. E. \ Karniadakis ,\ 10.1038/s42256-021-00302-5 journal journal Nature Machine Intelligence \ volume 3 ,\ pages 218 ( year 2021 ) NoStop

    work page doi:10.1038/s42256-021-00302-5 2021

  16. [17]

    Raissi, P

    author author M. Raissi , author P. Perdikaris , \ and\ author G. E. \ Karniadakis ,\ 10.1016/j.jcp.2018.10.045 journal journal Journal of Computational Physics \ volume 378 ,\ pages 686 ( year 2019 ) NoStop

    work page doi:10.1016/j.jcp.2018.10.045 2018

  17. [18]

    Gilmer , author K

    author author J. Gilmer , author K. T. \ Schütt , author O. A. \ von Lilienfeld , author A. Tkatchenko , \ and\ author K.-R. \ Müller ,\ in\ https://proceedings.mlr.press/v70/gilmer17a.html booktitle Proceedings of the 34th International Conference on Machine Learning (ICML) ,\ Vol. volume 70 \ ( year 2017 )\ pp.\ pages 1263--1272 NoStop

    work page 2017

  18. [20]

    Sanchez-Gonzalez , author J

    author author A. Sanchez-Gonzalez , author J. Godwin , author T. Pfaff , author R. Ying , author J. Leskovec , \ and\ author P. Battaglia ,\ in\ https://proceedings.mlr.press/v119/sanchez-gonzalez20a.html booktitle Proceedings of the 37th International Conference on Machine Learning (ICML) ,\ Vol.\ volume 119 \ ( publisher PMLR ,\ year 2020 )\ pp.\ pages ...

    work page 2020

  19. [21]

    Pfaff , author M

    author author T. Pfaff , author M. Fortunato , author A. Sanchez-Gonzalez , \ and\ author P. W. \ Battaglia ,\ in\ https://openreview.net/forum?id=roNqG-hehdt booktitle International Conference on Learning Representations (ICLR) \ ( year 2021 ) NoStop

    work page 2021

  20. [22]

    Hanke , author T

    author author N. Hanke , author T. Alber , \ and\ author D. M. \ Kochmann ,\ 10.1016/j.apor.2025.104424 journal journal Applied Ocean Research \ volume 154 ,\ pages 104424 ( year 2025 ) NoStop

    work page doi:10.1016/j.apor.2025.104424 2025

  21. [23]

    Aminimajd , author J

    author author A. Aminimajd , author J. Maia , \ and\ author A. Singh ,\ 10.1039/D4SM01391C journal journal Soft Matter \ volume 21 ,\ pages 2826 ( year 2025 ) NoStop

    work page doi:10.1039/d4sm01391c 2025

  22. [24]

    Mlinarevi\' c , author G

    author author M. Mlinarevi\' c , author G. K. \ Holt , \ and\ author A. Agnello ,\ https://arxiv.org/abs/2503.00274 journal journal arXiv preprint arXiv:2503.00274 \ ( year 2025 ) NoStop

    work page arXiv 2025

  23. [26]

    Lino , author T

    author author M. Lino , author T. Pfaff , \ and\ author N. Thuerey ,\ in\ https://arxiv.org/abs/2504.02843 booktitle International Conference on Learning Representations (ICLR) \ ( year 2025 ) NoStop

    work page arXiv 2025

  24. [27]

    Veli c kovi\' c , author G

    author author P. Veli c kovi\' c , author G. Cucurull , author A. Casanova , author A. Romero , author P. Li\` o , \ and\ author Y. Bengio ,\ in\ https://openreview.net/forum?id=rJXMpikCZ booktitle International Conference on Learning Representations (ICLR) \ ( year 2018 ) NoStop

    work page 2018

  25. [28]

    Brody , author U

    author author S. Brody , author U. Alon , \ and\ author E. Yahav ,\ in\ https://openreview.net/forum?id=F72ximsx7C1 booktitle International Conference on Learning Representations (ICLR) \ ( year 2022 ) NoStop

    work page 2022

  26. [29]

    Hamiltonian neural networks, 2019

    author author S. Greydanus , author M. Dzamba , \ and\ author J. Yosinski ,\ https://arxiv.org/abs/1906.01563 journal journal Advances in Neural Information Processing Systems \ volume 32 ( year 2019 ) NoStop

    work page arXiv 1906

  27. [31]

    Hairer, C

    author author E. Hairer , author C. Lubich , \ and\ author G. Wanner ,\ 10.1007/3-540-30666-8 title Geometric Numerical Integration: Structure-Preserving Algorithms for Ordinary Differential Equations ,\ edition 2nd \ ed.\ ( publisher Springer ,\ address Berlin ,\ year 2006 ) NoStop

    work page doi:10.1007/3-540-30666-8 2006

  28. [32]

    Donn \'e , author U

    author author A. Donn \'e , author U. Siddiqui , author P. Bhatt , \ and\ author M. Kaandorp ,\ @noop title CORGI : GNNs with convolutional residual global interactions for Lagrangian simulation , \ ( year 2025 ),\ note arXiv:2511.22938; submitted to ICLR 2026 ,\ http://arxiv.org/abs/2511.22938 arXiv:2511.22938 [cs.LG] NoStop

    work page arXiv 2025

  29. [33]

    Wei , author Y

    author author T. Wei , author Y. Yao , author Y. Shao , author J. Bian , \ and\ author T. Wang ,\ in\ @noop booktitle Advances in Neural Information Processing Systems (NeurIPS) ,\ Vol. volume 37 \ ( year 2024 ) NoStop

    work page 2024

  30. [34]

    Chen , author F

    author author Z. Chen , author F. Xu , author J. Wang , author R. Hu , author Q. Chen , \ and\ author S. Gao ,\ 10.1016/j.buildenv.2023.110056 journal journal Building and Environment \ volume 233 ,\ pages 110056 ( year 2023 ) NoStop

    work page doi:10.1016/j.buildenv.2023.110056 2023

  31. [35]

    Ferracina , author P

    author author F. Ferracina , author P. Beeler , author M. Halappanavar , author B. Krishnamoorthy , author M. Minutoli , \ and\ author L. Fierce ,\ 10.1021/acsestair.4c00261 journal journal ACS ES&T Air \ volume 2 ,\ pages 1426 ( year 2025 ) NoStop

    work page doi:10.1021/acsestair.4c00261 2025

  32. [37]

    author author M. R. \ Maxey \ and\ author J. J. \ Riley ,\ 10.1063/1.864230 journal journal Physics of Fluids \ volume 26 ,\ pages 883 ( year 1983 ) NoStop

    work page doi:10.1063/1.864230 1983

  33. [38]

    author author W. F. \ Wells ,\ 10.1093/oxfordjournals.aje.a118097 journal journal American Journal of Epidemiology \ volume 20 ,\ pages 611 ( year 1934 ) NoStop

    work page doi:10.1093/oxfordjournals.aje.a118097 1934

  34. [39]

    Long short -term memory,

    author author S. Hochreiter \ and\ author J. Schmidhuber ,\ 10.1162/neco.1997.9.8.1735 journal journal Neural Computation \ volume 9 ,\ pages 1735 ( year 1997 ) NoStop

    work page doi:10.1162/neco.1997.9.8.1735 1997

  35. [40]

    Masked label prediction: Unified message passing model for semi-supervised classification

    author author Y. Shi , author Z. Huang , author S. Feng , author H. Zhong , author W. Wang , \ and\ author Y. Sun ,\ in\ https://arxiv.org/abs/2009.03509 booktitle Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI) \ ( year 2021 ) NoStop

    work page arXiv 2009

  36. [42]

    Attention Is All You Need

    author author A. Vaswani , author N. Shazeer , author N. Parmar , author J. Uszkoreit , author L. Jones , author A. N. \ Gomez , author L. Kaiser , \ and\ author I. Polosukhin ,\ in\ https://arxiv.org/abs/1706.03762 booktitle Advances in Neural Information Processing Systems ,\ Vol. volume 30 \ ( year 2017 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  37. [43]

    author author D. P. \ Kingma \ and\ author M. Welling ,\ in\ https://arxiv.org/abs/1312.6114 booktitle International Conference on Learning Representations (ICLR) \ ( year 2014 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  38. [44]

    author author C. W. \ Oseen ,\ @noop journal journal Arkiv f\"or Matematik, Astronomi och Fysik \ volume 6 ( year 1910 ) NoStop

    work page 1910

  39. [45]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    author author A. Paszke , author S. Gross , author F. Massa , author A. Lerer , author J. Bradbury , author G. Chanan , et al. ,\ in\ https://arxiv.org/abs/1912.01703 booktitle Advances in Neural Information Processing Systems ,\ Vol. volume 32 \ ( year 2019 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  40. [46]

    Fast Graph Representation Learning with PyTorch Geometric

    author author M. Fey \ and\ author J. E. \ Lenssen ,\ in\ https://arxiv.org/abs/1903.02428 booktitle ICLR Workshop on Representation Learning on Graphs and Manifolds \ ( year 2019 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  41. [47]

    author author P. J. \ Werbos ,\ 10.1109/5.58337 journal journal Proceedings of the IEEE \ volume 78 ,\ pages 1550 ( year 1990 ) NoStop

    work page doi:10.1109/5.58337 1990

  42. [48]

    and Eaton, John K

    author author S. Balachandar \ and\ author J. K. \ Eaton ,\ 10.1146/annurev.fluid.010908.165243 journal journal Annual Review of Fluid Mechanics \ volume 42 ,\ pages 111 ( year 2010 ) NoStop

    work page doi:10.1146/annurev.fluid.010908.165243 2010

  43. [49]

    Li , author G

    author author Y. Li , author G. M. \ Leung , author J. W. \ Tang , author X. Yang , author C. Y. H. \ Chao , author J. Z. \ Lin , author J. W. \ Lu , author P. V. \ Nielsen , author J. Niu , author H. Qian , author A. C. \ Sleigh , author H. J. J. \ Su , author J. Sundell , author T. W. \ Wong , \ and\ author P. L. \ Leung ,\ 10.1111/j.1600-0668.2006.0044...

    work page doi:10.1111/j.1600-0668.2006.00445.x 2006

  44. [50]

    author author ASHRAE ,\ @noop title ANSI/ASHRAE standard 170-2021: Ventilation of health care facilities , \ type Tech. Rep. \ ( institution American Society of Heating, Refrigerating and Air-Conditioning Engineers ,\ address Atlanta, GA ,\ year 2021 )\ note prescribes minimum 12 air changes per hour for airborne infection isolation rooms NoStop

    work page 2021

  45. [51]

    author author G. I. \ Taylor ,\ 10.1112/plms/s2-20.1.196 journal journal Proceedings of the London Mathematical Society \ volume 20 ,\ pages 196 ( year 1921 ) NoStop

    work page doi:10.1112/plms/s2-20.1.196 1921

  46. [52]

    author author G. K. \ Batchelor ,\ 10.1017/S0305004100027687 journal journal Proceedings of the Cambridge Philosophical Society \ volume 48 ,\ pages 345 ( year 1952 ) NoStop

    work page doi:10.1017/s0305004100027687 1952

  47. [53]

    author author M. D. \ Allen \ and\ author O. G. \ Raabe ,\ 10.1080/02786828508959055 journal journal Aerosol Science and Technology \ volume 4 ,\ pages 269 ( year 1985 ) NoStop

    work page doi:10.1080/02786828508959055 1985

  48. [54]

    Schiller \ and\ author A

    author author L. Schiller \ and\ author A. Naumann ,\ @noop journal journal Zeitschrift des Vereines Deutscher Ingenieure \ volume 77 ,\ pages 318 ( year 1933 ) NoStop

    work page 1933

  49. [55]

    Clift , author J

    author author R. Clift , author J. R. \ Grace , \ and\ author M. E. \ Weber ,\ @noop journal journal Academic Press, New York \ ( year 1978 ) ,\ note standard reference for particle settling velocities and non-dimensional drag correlations NoStop

    work page 1978

  50. [56]

    author author P. G. \ Saffman ,\ 10.1017/S0022112065000824 journal journal Journal of Fluid Mechanics \ volume 22 ,\ pages 385 ( year 1965 ) NoStop

    work page doi:10.1017/s0022112065000824 1965

  51. [57]

    author author J. B. \ McLaughlin ,\ 10.1017/S0022112091001751 journal journal Journal of Fluid Mechanics \ volume 224 ,\ pages 261 ( year 1991 ) NoStop

    work page doi:10.1017/s0022112091001751 1991

  52. [58]

    Jasak , author A

    author author H. Jasak , author A. Jemcov , \ and\ author Z. Tukovic ,\ in\ @noop booktitle International Workshop on Coupled Methods in Numerical Dynamics \ ( address Dubrovnik, Croatia ,\ year 2007 )\ pp.\ pages 1--20 NoStop

    work page 2007

  53. [59]

    author author H. J. \ Bae \ and\ author A. Lozano-Dur\' a n ,\ 10.1038/s41467-022-28957-7 journal journal Nature Communications \ volume 13 ,\ pages 1443 ( year 2022 ) NoStop

    work page doi:10.1038/s41467-022-28957-7 2022

  54. [60]

    Quattromini , author P

    author author M. Quattromini , author P. Morra , author K. Srinivasan , author A. Cavalieri , author C. Cossu , author D. S. \ Henningson , \ and\ author A. Lozano-Dur\' a n ,\ 10.1103/PhysRevFluids.10.024603 journal journal Physical Review Fluids \ volume 10 ,\ pages 024603 ( year 2025 ) NoStop

    work page doi:10.1103/physrevfluids.10.024603 2025

  55. [61]

    and Peri\'

    Ferziger, Joel H. and Peri\'. Computational Methods for Fluid Dynamics , edition =. 2002 , doi =

    work page 2002

  56. [62]

    2022 , type =

    Xing, Victor , title =. 2022 , type =

    work page 2022

  57. [63]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =

    Guo, Xiaoxiao and Li, Wei and Iorio, Francesco , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2016 , doi =

    work page 2016

  58. [64]

    and Alieva, Ayya and Wang, Qing and Brenner, Michael P

    Kochkov, Dmitrii and Smith, Jamie A. and Alieva, Ayya and Wang, Qing and Brenner, Michael P. and Hoyer, Stephan , title =. Proceedings of the National Academy of Sciences , volume =. 2021 , doi =

    work page 2021

  59. [65]

    Fourier Neural Operator for Parametric Partial Differential Equations

    Li, Zongyi and Kovachki, Nikola and Azizzadenesheli, Kamyar and Liu, Burigede and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , title =. arXiv preprint arXiv:2010.08895 , year =

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  60. [66]

    Nature Machine Intelligence , volume =

    Lu, Lu and Jin, Pengzhan and Pang, Guofei and Zhang, Zhongqiang and Karniadakis, George Em , title =. Nature Machine Intelligence , volume =. 2021 , doi =

    work page 2021

  61. [67]

    Journal of Computational Physics , volume =

    Raissi, Maziar and Perdikaris, Paris and Karniadakis, George Em , title =. Journal of Computational Physics , volume =. 2019 , doi =

    work page 2019

  62. [68]

    Applied Mathematics and Mechanics , volume =

    Li, Shengze and others , title =. Applied Mathematics and Mechanics , volume =. 2024 , doi =

    work page 2024

  63. [69]

    and von Lilienfeld, O

    Gilmer, Justin and Schütt, Kristof T. and von Lilienfeld, O. Anatole and Tkatchenko, Alexandre and Müller, Klaus-Robert , title =. Proceedings of the 34th International Conference on Machine Learning (ICML) , pages =. 2017 , url =

    work page 2017

  64. [70]

    Relational inductive biases, deep learning, and graph networks

    Battaglia, Peter W. and Hamrick, Jessica B. and Bapst, Victor and others , title =. arXiv preprint arXiv:1806.01261 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  65. [71]

    Jane and Lozano-Dur\'

    Bae, H. Jane and Lozano-Dur\'. Scientific multi-agent reinforcement learning for wall-models of turbulent flows , journal =. 2022 , doi =

    work page 2022

  66. [72]

    Causality of energy-containing eddies in wall turbulence , journal =

    Lozano-Dur\'. Causality of energy-containing eddies in wall turbulence , journal =. 2019 , doi =

    work page 2019

  67. [73]

    Mean-flow data assimilation using physics-constrained graph neural networks , journal =

    Quattromini, Michele and Morra, Pierluigi and Srinivasan, Krishnan and Cavalieri, Andr\'. Mean-flow data assimilation using physics-constrained graph neural networks , journal =. 2025 , doi =

    work page 2025

  68. [74]

    Proceedings of the 37th International Conference on Machine Learning (ICML) , pages =

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

    work page 2020

  69. [75]

    arXiv preprint arXiv:2211.10228 , year =

    Kumar, Krishna and Vantassel, Joseph , title =. arXiv preprint arXiv:2211.10228 , year =

    work page arXiv

  70. [76]

    , title =

    Pfaff, Tobias and Fortunato, Meire and Sanchez-Gonzalez, Alvaro and Battaglia, Peter W. , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  71. [77]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Yang, Can and Wang, Zhenzhong and Liu, Junyuan and Gong, Yunpeng and Jiang, Min , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2026 , doi =

    work page 2026

  72. [78]

    arXiv preprint arXiv:2501.07373 , year =

    Sharma, Vinay and Fink, Olga , title =. arXiv preprint arXiv:2501.07373 , year =

    work page arXiv

  73. [79]

    Advances in Neural Information Processing Systems , volume =

    Greydanus, Samuel and Dzamba, Misko and Yosinski, Jason , title =. Advances in Neural Information Processing Systems , volume =. 2019 , url =

    work page 2019

  74. [80]

    et al.Lagrangian Neural Networks

    Cranmer, Miles and Greydanus, Sam and Hoyer, Stephan and Battaglia, Peter and Spergel, David and Ho, Shirley , title =. arXiv preprint arXiv:2003.04630 , year =

    work page arXiv 2003

  75. [81]

    and Kaiser, Lukasz and Polosukhin, Illia , title =

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N. and Kaiser, Lukasz and Polosukhin, Illia , title =. Advances in Neural Information Processing Systems , volume =. 2017 , url =

    work page 2017

  76. [82]

    International Conference on Learning Representations (ICLR) , year =

    Brody, Shaked and Alon, Uri and Yahav, Eran , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  77. [83]

    Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI) , year =

    Shi, Yunsheng and Huang, Zhengjie and Feng, Shikun and Zhong, Hui and Wang, Wenjing and Sun, Yu , title =. Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI) , year =

    work page

  78. [84]

    and Welling, Max , title =

    Kingma, Diederik P. and Welling, Max , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  79. [85]

    International Conference on Learning Representations (ICLR) , year =

    Lino, Mario and Pfaff, Tobias and Thuerey, Nils , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  80. [86]

    Particle-based plasma simulation using a graph neural network , journal =

    Mlinarevi\'. Particle-based plasma simulation using a graph neural network , journal =. 2025 , url =

    work page 2025

Showing first 80 references.