pith. sign in

arxiv: 2605.22698 · v1 · pith:C4ENN63Pnew · submitted 2026-05-21 · ⚛️ physics.chem-ph

Machine Learning Interatomic Potentials: Advancing Open-Source Software for Efficient and Scalable Molecular Simulation

Christoph Brunken , Titouan Cormier , Lucien Walewski , Marco Carobene , Yessine Khanfir , Zachary Weller-Davies , Miguel Bragan\c{c}a , Armand Picard
show 10 more authors
Adrien Pichard Leon Wehrhan Heloise Chomet Eszter Varga-Umbrich Marie Bluntzer Massimo Bortone Valentin Heyraud Silvia Acosta-Guti\'errez Jules Tilly Olivier Peltre
This is my paper

Pith reviewed 2026-05-22 03:29 UTC · model grok-4.3

classification ⚛️ physics.chem-ph
keywords machine learning interatomic potentialsmolecular simulationequivariant operationsmixture of expertselectrostatic modelingopen source libraryscalable trainingNPT ensembles
0
0 comments X

The pith

mlip v2 introduces a modular API redesign, e3j backend, and eSEN architecture to make machine learning interatomic potentials faster and more scalable for molecular simulations.

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

The paper presents mlip v2 as an updated open-source library that unifies training, data handling, and simulation tools for machine learning interatomic potentials. It achieves this through an API redesign for greater flexibility, a new high-performance backend for equivariant operations, and added features such as the eSEN model for large datasets plus better electrostatic and ensemble methods. A sympathetic reader would care because these changes could lower the computational barriers that currently limit widespread use of accurate atomistic modeling in chemistry and materials research.

Core claim

The authors introduce mlip v2 as a new generation of the mlip library that advances efficient and scalable molecular simulation through a unified and extensible framework. The release includes a targeted API redesign with improved modularity and control for flexible customization of workflows, integration of the e3j high-performance backend for equivariant operations to accelerate inference and simulations, the eSEN architecture with a Mixture-of-Experts formulation for scalable training on large datasets, improved electrostatics via more physically grounded charge modeling and long-range interactions, and advanced simulation features including NPT ensembles and nudged elastic band methods.

What carries the argument

mlip v2, a unified extensible framework that combines API modularity, the e3j backend for equivariant operations, and the eSEN Mixture-of-Experts architecture to support flexible customization and high-performance molecular simulations.

Load-bearing premise

The new integrations of e3j, eSEN, and electrostatic improvements deliver the stated performance gains and physical accuracy without introducing errors or requiring extensive user tuning.

What would settle it

Direct comparison of wall-clock inference time and energy accuracy on a standard test set such as liquid water or small organic molecules before and after switching to the e3j backend and eSEN model.

Figures

Figures reproduced from arXiv: 2605.22698 by Adrien Pichard, Armand Picard, Christoph Brunken, Eszter Varga-Umbrich, Heloise Chomet, Jules Tilly, Leon Wehrhan, Lucien Walewski, Marco Carobene, Marie Bluntzer, Massimo Bortone, Miguel Bragan\c{c}a, Olivier Peltre, Silvia Acosta-Guti\'errez, Titouan Cormier, Valentin Heyraud, Yessine Khanfir, Zachary Weller-Davies.

Figure 1
Figure 1. Figure 1: Results of end-to-end runtime benchmarks on MACE and NequIP, comparing the [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Validation set mean absolute errors (MAE) for energy per atom (meV/atom) and atomic [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of vibrational frequencies computed with DFT, a MACE model trained on [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Evaluation of energy MAE distribution per global charge. A comparison of energy [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: NPT molecular dynamics validation for VisNet-v2 water (501 molecules, 300 K, 1 atm). [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Validation set root-mean-squared errors (RMSE) for energy per atom (meV/atom) and [PITH_FULL_IMAGE:figures/full_fig_p028_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Validation set mean absolute errors (MAE) and root-mean-squared errors (RMSE) for atomic [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
read the original abstract

Machine learning interatomic potentials (MLIPs) enable atomistic simulations with near ab initio accuracy at significantly reduced computational cost, but their broader adoption is often limited by fragmented tooling, limited scalability, and inflexible software design. We present mlip v2, a new generation of the mlip library that advances efficient and scalable molecular simulation through a unified and extensible framework. The new release features a targeted API redesign with improved modularity and control, enabling flexible customization of training, data processing, and simulation workflows. It further integrates a new high-performance backend for equivariant operations, e3j, significantly accelerating model inference and simulations. In addition, the framework introduces a range of entirely new capabilities, including the eSEN architecture with a Mixture-of-Experts formulation for scalable training on large and diverse datasets, improved handling of electrostatics through more physically grounded charge modeling and long-range interaction treatment, and advanced simulation features such as NPT ensembles and nudged elastic band methods. Together, these extensions significantly broaden the scope of MLIP applications, enabling efficient modeling of complex, reactive, and out-of-equilibrium systems, and bridging the gap between ML research and practical molecular simulation applications. The library is available on GitHub and on PyPI under the Apache license 2.0.

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 mlip v2, an updated open-source library for machine learning interatomic potentials (MLIPs). It describes a targeted API redesign for improved modularity in training, data processing, and simulation workflows; integration of a new high-performance e3j backend for equivariant operations to accelerate inference and simulations; the eSEN architecture incorporating a Mixture-of-Experts formulation for scalable training on large datasets; enhanced electrostatics via more physically grounded charge modeling and long-range interactions; and new simulation capabilities including NPT ensembles and nudged elastic band methods. The library is released under the Apache 2.0 license on GitHub and PyPI, with the goal of broadening MLIP applications to complex, reactive, and out-of-equilibrium systems.

Significance. If the described features deliver the claimed accelerations and accuracy improvements, the work could meaningfully advance practical adoption of MLIPs by providing a more extensible and performant open-source framework. The emphasis on modularity, equivariant performance, scalable architectures, and physically motivated electrostatics addresses real barriers in the field. However, the manuscript supplies no quantitative benchmarks, timing comparisons, error metrics against reference data, or scaling studies, which substantially weakens the ability to evaluate the significance of these contributions.

major comments (2)
  1. Abstract and description of new capabilities: the assertions that the e3j backend 'significantly accelerat[es] model inference and simulations' and that eSEN enables 'scalable training on large and diverse datasets' are central to the paper's claim of advancement, yet no timing data, scaling curves, ablation studies, or comparisons to mlip v1 or competing libraries (e.g., MACE, NequIP) are provided to substantiate them.
  2. Description of improved handling of electrostatics: the claim of 'more physically grounded charge modeling and long-range interaction treatment' is load-bearing for the assertion of improved physical accuracy, but the manuscript provides neither implementation details (e.g., charge equilibration scheme or Ewald summation parameters) nor validation against ab initio electrostatic energies or forces.
minor comments (2)
  1. The manuscript would benefit from a dedicated section or table summarizing the new features with version numbers, dependencies, and installation instructions to aid reproducibility for users.
  2. Clarify the relationship between the eSEN Mixture-of-Experts formulation and existing equivariant architectures; a brief comparison or reference to prior MoE work in MLIPs would improve context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript describing mlip v2. We address each of the major comments in detail below and outline the revisions we plan to make to strengthen the paper.

read point-by-point responses

  1. Referee: Abstract and description of new capabilities: the assertions that the e3j backend 'significantly accelerat[es] model inference and simulations' and that eSEN enables 'scalable training on large and diverse datasets' are central to the paper's claim of advancement, yet no timing data, scaling curves, ablation studies, or comparisons to mlip v1 or competing libraries (e.g., MACE, NequIP) are provided to substantiate them.

    Authors: We agree with the referee that quantitative benchmarks are essential to substantiate the performance claims. The current manuscript focuses on describing the new software features and their design rationale. In the revised manuscript, we will add timing comparisons for the e3j backend versus the previous implementation, scaling studies for the eSEN model on large datasets, and direct comparisons with other MLIP libraries such as MACE and NequIP. These additions will be included in a new section on performance evaluation. revision: yes

  2. Referee: Description of improved handling of electrostatics: the claim of 'more physically grounded charge modeling and long-range interaction treatment' is load-bearing for the assertion of improved physical accuracy, but the manuscript provides neither implementation details (e.g., charge equilibration scheme or Ewald summation parameters) nor validation against ab initio electrostatic energies or forces.

    Authors: We acknowledge that the manuscript would benefit from more detailed descriptions and validation for the electrostatics improvements. In the revised version, we will expand the relevant section to include specifics on the charge modeling approach, including the charge equilibration scheme and parameters for long-range interactions such as Ewald summation. Additionally, we will provide validation results comparing the modeled electrostatic energies and forces to ab initio reference data. revision: yes

Circularity Check

0 steps flagged

No significant circularity: software release note with no derivations, equations, or load-bearing predictions

full rationale

The manuscript describes new features in the mlip v2 library (API redesign, e3j backend, eSEN MoE architecture, electrostatic improvements, NPT and NEB simulation capabilities) but contains no equations, no claimed first-principles derivations, and no predictions that could reduce to fitted inputs or self-citations by construction. All statements are declarative descriptions of implemented software components rather than a derivation chain. No self-definitional loops, fitted-input-as-prediction patterns, or uniqueness theorems imported from prior author work appear. The paper is self-contained as a release announcement; external validation of performance claims is a separate correctness issue, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

The central claims rest on the existence and correct functioning of newly introduced software components whose performance benefits are asserted without supporting measurements in the abstract; no free parameters or formal axioms are invoked because this is a tooling announcement rather than a theoretical derivation.

invented entities (2)
  • e3j backend no independent evidence
    purpose: High-performance backend for equivariant operations to accelerate inference and simulations
    New component introduced to improve speed of model operations; no independent evidence provided in abstract.
  • eSEN architecture no independent evidence
    purpose: Mixture-of-Experts formulation for scalable training on large and diverse datasets
    New model architecture presented for handling bigger data; no independent evidence or benchmarks in abstract.

pith-pipeline@v0.9.0 · 5841 in / 1391 out tokens · 40422 ms · 2026-05-22T03:29:11.227186+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.

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

87 extracted references · 87 canonical work pages · 5 internal anchors

  1. [1]

    Machine learning interatomic poten- tials: library for efficient training, model development and simulation of molecular systems,

    Christoph Brunken, Olivier Peltre, Heloise Chomet, Lucien Walewski, Manus McAuliffe, Valentin Heyraud, Solal Attias, Martin Maarand, Yessine Khanfir, Edan Toledo, Fabio Falcioni, Marie Bluntzer, Silvia Acosta-Gutiérrez, and Jules Tilly. Machine learning interatomic poten- tials: library for efficient training, model development and simulation of molecular...

    work page

  2. [2]

    URLhttps://arxiv.org/abs/2505.22397

    work page arXiv

  3. [3]

    Schoenholz and Ekin D

    Samuel S. Schoenholz and Ekin D. Cubuk. Jax, m.d.: A framework for differentiable physics,

    work page

  4. [4]

    URLhttps://arxiv.org/abs/1912.04232

    work page arXiv 1912

  5. [5]

    Diversity-driven training of machine- learned force fields

    Maitreyee Sharma Priyadarshini and Connor Ganley. Diversity-driven training of machine- learned force fields. InAI for Accelerated Materials Design - NeurIPS 2025, 2025. URL https://openreview.net/forum?id=WNuRO48JSn

    work page 2025

  6. [6]

    Enhancing non-local interaction modeling for ab initio biomolecular calculations and simulations with visnet-pima.bioRxiv, 2026

    Taoyong Cui, Zihan Wang, and Tong Wang. Enhancing non-local interaction modeling for ab initio biomolecular calculations and simulations with visnet-pima.bioRxiv, 2026. doi: 10.64898/2026.03.18.712561. URL https://www.biorxiv.org/content/early/2026/ 03/20/2026.03.18.712561

    work page doi:10.64898/2026.03.18.712561 2026

  7. [7]

    Skillpuzzler: A self-evolving agentic framework for materials and chemistry research with minimal reliance on predefined tools

    Xu Huang, Junwu Chen, Philippe Schwaller, and Gerbrand Ceder. Skillpuzzler: A self-evolving agentic framework for materials and chemistry research with minimal reliance on predefined tools. InNeurIPS 2025 AI for Science Workshop, 2025. URL https://openreview.net/ forum?id=Mc1JTnKQIH

    work page 2025

  8. [8]

    Mlipaudit: A benchmarking tool for machine learned interatomic potentials, 2025

    Leon Wehrhan, Lucien Walewski, Marie Bluntzer, Heloise Chomet, Jules Tilly, Christoph Brunken, and Silvia Acosta-Gutiérrez. Mlipaudit: A benchmarking tool for machine learned interatomic potentials, 2025. URLhttps://arxiv.org/abs/2511.20487

    work page arXiv 2025

  9. [9]

    Lawrence Zitnick

    Saro Passaro and C. Lawrence Zitnick. Reducing so(3) convolutions to so(2) for efficient equivariant gnns. InProceedings of the 40th International Conference on Machine Learning, ICML’23. JMLR.org, 2023

    work page 2023

  10. [10]

    Wood, Luis Barroso-Luque, Daniel S

    Xiang Fu, Brandon M. Wood, Luis Barroso-Luque, Daniel S. Levine, Meng Gao, Misko Dzamba, and C. Lawrence Zitnick. Learning smooth and expressive interatomic potentials for physical property prediction, 2025. URLhttps://arxiv.org/abs/2502.12147

    work page arXiv 2025

  11. [11]

    Wood, Misko Dzamba, Xiang Fu, Meng Gao, Muhammed Shuaibi, Luis Barroso- Luque, Kareem Abdelmaqsoud, Vahe Gharakhanyan, John R

    Brandon M. Wood, Misko Dzamba, Xiang Fu, Meng Gao, Muhammed Shuaibi, Luis Barroso- Luque, Kareem Abdelmaqsoud, Vahe Gharakhanyan, John R. Kitchin, Daniel S. Levine, Kyle Michel, Anuroop Sriram, Taco Cohen, Abhishek Das, Ammar Rizvi, Sushree Jagriti Sahoo, Zachary W. Ulissi, and C. Lawrence Zitnick. Uma: A family of universal models for atoms,

    work page

  12. [12]

    URLhttps://arxiv.org/abs/2506.23971

    work page arXiv

  13. [13]

    Mailoa, Mordechai Kornbluth, Nicola Molinari, Tess E

    Simon Batzner, Albert Musaelian, Lixin Sun, Mario Geiger, Jonathan P. Mailoa, Mordechai Kornbluth, Nicola Molinari, Tess E. Smidt, and Boris Kozinsky. E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials.Nature Communications, 13(1), May 2022. ISSN 2041-1723. doi: 10.1038/s41467-022-29939-5. URL http://dx.doi.org/ ...

    work page doi:10.1038/s41467-022-29939-5 2022

  14. [14]

    Brehmer, S

    Johann Brehmer, Sönke Behrends, Pim de Haan, and Taco Cohen. Does equivariance matter at scale?, 2024. URLhttps://arxiv.org/abs/2410.23179

    work page arXiv 2024

  15. [15]

    K. T. Schütt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, and K.-R. Müller. Schnet – a deep learning architecture for molecules and materials.The Journal of Chemical Physics, 148 (24), March 2018. ISSN 1089-7690. doi: 10.1063/1.5019779. URL http://dx.doi.org/10. 1063/1.5019779. 11

    work page doi:10.1063/1.5019779 2018

  16. [16]

    Schütt, Pan Kessel, Michael Gastegger, Kim A

    Kristof T. Schütt, Pan Kessel, Michael Gastegger, Kim A. Nicoli, Alexandre Tkatchenko, and Klaus-Robert Müller. SchNetPack: A Deep Learning Toolbox For Atomistic Systems.Journal of Chemical Theory and Computation, 15(1):448–455, 2019. doi: 10.1021/acs.jctc.8b00908. URLhttps://doi.org/10.1021/acs.jctc.8b00908

    work page doi:10.1021/acs.jctc.8b00908 2019

  17. [17]

    Identifying anyonic topological order in fractional quantum anomalous Hall systems

    Kristof T. Schütt, Stefaan S. P. Hessmann, Niklas W. A. Gebauer, Jonas Lederer, and Michael Gastegger. SchNetPack 2.0: A neural network toolbox for atomistic machine learning.The Journal of Chemical Physics, 158(14):144801, 04 2023. ISSN 0021-9606. doi: 10.1063/5. 0138367. URLhttps://doi.org/10.1063/5.0138367

    work page doi:10.1063/5 2023

  18. [18]

    Anstine, Roman Zubatyuk, and Olexandr Isayev

    Dylan M. Anstine, Roman Zubatyuk, and Olexandr Isayev. Aimnet2: a neural network potential to meet your neutral, charged, organic, and elemental-organic needs.Chemical Science,

    work page

  19. [20]

    Grossman

    Tian Xie and Jeffrey C. Grossman. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties.Physical Review Letters, 120(14), April

    work page

  20. [21]

    doi: 10.1103/physrevlett.120.145301

    ISSN 1079-7114. doi: 10.1103/physrevlett.120.145301. URL http://dx.doi.org/ 10.1103/PhysRevLett.120.145301

    work page doi:10.1103/physrevlett.120.145301

  21. [22]

    E(n) equivariant graph neural networks, 2022

    Victor Garcia Satorras, Emiel Hoogeboom, and Max Welling. E(n) equivariant graph neural networks, 2022. URLhttps://arxiv.org/abs/2102.09844

    work page arXiv 2022

  22. [23]

    Directional message passing for molecular graphs

    Johannes Gasteiger, Janek Groß, and Stephan Günnemann. Directional message passing for molecular graphs. InInternational Conference on Learning Representations (ICLR), 2020

    work page 2020

  23. [24]

    Margraf, and Stephan Günnemann

    Johannes Gasteiger, Shankari Giri, Johannes T. Margraf, and Stephan Günnemann. Fast and uncertainty-aware directional message passing for non-equilibrium molecules. InMachine Learning for Molecules Workshop, NeurIPS, 2020

    work page 2020

  24. [25]

    Gemnet: Univer- sal directional graph neural networks for molecules

    Johannes Gasteiger, Florian Becker, and Stephan Günnemann. Gemnet: Univer- sal directional graph neural networks for molecules. In M. Ranzato, A. Beygelz- imer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advances in Neu- ral Information Processing Systems, volume 34, pages 6790–6802. Curran Associates, Inc., 2021. URL https://proceedings.neu...

    work page 2021

  25. [26]

    Gemnet: Universal directional graph neural networks for molecules, 2024

    Johannes Gasteiger, Florian Becker, and Stephan Günnemann. Gemnet: Universal directional graph neural networks for molecules, 2024. URLhttps://arxiv.org/abs/2106.08903

    work page arXiv 2024

  26. [27]

    Enhancing geometric representations for molecules with equivariant vector-scalar interactive message passing.Nature Communications, 15(1), January 2024

    Yusong Wang, Tong Wang, Shaoning Li, Xinheng He, Mingyu Li, Zun Wang, Nanning Zheng, Bin Shao, and Tie-Yan Liu. Enhancing geometric representations for molecules with equivariant vector-scalar interactive message passing.Nature Communications, 15(1), January 2024. ISSN 2041-1723. doi: 10.1038/s41467-023-43720-2. URL http://dx.doi.org/10.1038/ s41467-023-43720-2

    work page doi:10.1038/s41467-023-43720-2 2024

  27. [28]

    Ab initio characterization of protein molecular dynamics with ai2bmd.Nature, 635(8040):1019–1027, November 2024

    Tong Wang, Xinheng He, Mingyu Li, Yatao Li, Ran Bi, Yusong Wang, Chaoran Cheng, Xiangzhen Shen, Jiawei Meng, He Zhang, Haiguang Liu, Zun Wang, Shaoning Li, Bin Shao, and Tie-Yan Liu. Ab initio characterization of protein molecular dynamics with ai2bmd.Nature, 635(8040):1019–1027, November 2024. ISSN 1476-4687. doi: 10.1038/s41586-024-08127-z. URLhttp://dx...

    work page doi:10.1038/s41586-024-08127-z 2024

  28. [29]

    3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data

    Maurice Weiler, Mario Geiger, Max Welling, Wouter Boomsma, and Taco Cohen. 3d steerable cnns: Learning rotationally equivariant features in volumetric data, 2018. URL https:// arxiv.org/abs/1807.02547

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  29. [30]

    Clebsch-Gordan Nets: a Fully Fourier Space Spherical Convolutional Neural Network

    Risi Kondor, Zhen Lin, and Shubhendu Trivedi. Clebsch-gordan nets: a fully fourier space spherical convolutional neural network, 2018. URL https://arxiv.org/abs/1806.09231. 12

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  30. [31]

    Ilyes Batatia, Simon Batzner, Dávid Péter Kovács, Albert Musaelian, Gregor N. C. Simm, Ralf Drautz, Christoph Ortner, Boris Kozinsky, and Gábor Csányi. The design space of e(3)- equivariant atom-centered interatomic potentials, 2022. URL https://arxiv.org/abs/ 2205.06643

    work page arXiv 2022

  31. [32]

    Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds

    Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, and Patrick Riley. Tensor field networks: Rotation- and translation-equivariant neural networks for 3d point clouds, 2018. URLhttps://arxiv.org/abs/1802.08219

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  32. [33]

    Ilyes Batatia, Dávid Péter Kovács, Gregor N. C. Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields, 2023. URLhttps://arxiv.org/abs/2206.07697

    work page arXiv 2023

  33. [34]

    Harry Moore, Nicholas J

    Dávid Péter Kovács, J. Harry Moore, Nicholas J. Browning, Ilyes Batatia, Joshua T. Horton, Yixuan Pu, Venkat Kapil, William C. Witt, Ioan-Bogdan Magd˘au, Daniel J. Cole, and Gábor Csányi. Mace-off: Short-range transferable machine learning force fields for organic molecules. Journal of the American Chemical Society, May 2025. ISSN 1520-5126. doi: 10.1021/...

    work page doi:10.1021/jacs 2025

  34. [35]

    Elena, Dávid P

    Ilyes Batatia, Philipp Benner, Yuan Chiang, Alin M. Elena, Dávid P. Kovács, Janosh Riebesell, Xavier R. Advincula, Mark Asta, Matthew Avaylon, William J. Baldwin, Fabian Berger, Noam Bernstein, Arghya Bhowmik, Samuel M. Blau, Vlad C ˘arare, James P. Darby, Sandip De, Flaviano Della Pia, V olker L. Deringer, Rokas Elijošius, Zakariya El-Machachi, Fabio Fal...

    work page

  35. [36]

    URLhttps://arxiv.org/abs/2401.00096

    work page internal anchor Pith review Pith/arXiv arXiv

  36. [37]

    Evaluation of the mace force field architecture: From medicinal chemistry to materials science.The Journal of Chemical Physics, 159(4), July 2023

    Dávid Péter Kovács, Ilyes Batatia, Eszter Sára Arany, and Gábor Csányi. Evaluation of the mace force field architecture: From medicinal chemistry to materials science.The Journal of Chemical Physics, 159(4), July 2023. ISSN 1089-7690. doi: 10.1063/5.0155322. URL http://dx.doi.org/10.1063/5.0155322

    work page doi:10.1063/5.0155322 2023

  37. [38]

    Baldwin, Domantas Kuryla, Joseph Hart, Elliott Kasoar, Alin M

    Ilyes Batatia, William J. Baldwin, Domantas Kuryla, Joseph Hart, Elliott Kasoar, Alin M. Elena, Harry Moore, Mikołaj J. Gawkowski, Benjamin X. Shi, Venkat Kapil, Panagiotis Kourtis, Ioan-Bogdan Magd˘au, and Gábor Csányi. Mace-polar-1: A polarisable electrostatic foundation model for molecular chemistry, 2026. URLhttps://arxiv.org/abs/2602.19411

    work page arXiv 2026

  38. [39]

    Krishnapriyan

    Shengjie Luo, Tianlang Chen, and Aditi S. Krishnapriyan. Enabling efficient equivariant operations in the fourier basis via gaunt tensor products, 2024. URL https://arxiv.org/ abs/2401.10216

    work page arXiv 2024

  39. [40]

    The price of freedom: Exploring expressivity and runtime tradeoffs in equivariant tensor products

    YuQing Xie, Ameya Daigavane, Mit Kotak, and Tess Smidt. The price of freedom: Exploring expressivity and runtime tradeoffs in equivariant tensor products. InForty-second Interna- tional Conference on Machine Learning, 2025. URL https://openreview.net/forum? id=EvIwwGYTLc

    work page 2025

  40. [41]

    Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics, 2026

    YuQing Xie, Ameya Daigavane, Mit Kotak, and Tess Smidt. Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics, 2026. URLhttps://arxiv.org/abs/2602. 21466. 13

    work page 2026

  41. [42]

    Integral formulas for vector spherical tensor products, 2026

    Valentin Heyraud, Zachary Weller-Davies, and Jules Tilly. Integral formulas for vector spherical tensor products, 2026. URLhttps://arxiv.org/abs/2603.08630

    work page arXiv 2026

  42. [43]

    Equiformer: Equivariant graph attention transformer for 3d atomistic graphs

    Yi-Lun Liao and Tess Smidt. Equiformer: Equivariant graph attention transformer for 3d atomistic graphs. InThe Eleventh International Conference on Learning Representations, 2023. URLhttps://openreview.net/forum?id=KwmPfARgOTD

    work page 2023

  43. [44]

    Equiformerv2: Improved equivariant transformer for scaling to higher-degree representations

    Yi-Lun Liao, Brandon M Wood, Abhishek Das, and Tess Smidt. Equiformerv2: Improved equivariant transformer for scaling to higher-degree representations. InThe Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum? id=mCOBKZmrzD

    work page 2024

  44. [45]

    E2former: An efficient and equivariant transformer with linear-scaling tensor products

    Yunyang Li, Lin Huang, Zhihao Ding, Xinran Wei, Chu Wang, Han Yang, Zun Wang, Chang Liu, Yu Shi, Peiran Jin, Tao Qin, Mark Gerstein, and Jia Zhang. E2former: An efficient and equivariant transformer with linear-scaling tensor products. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026. URL https://openreview. net/forum?id...

    work page 2026

  45. [46]

    E2former-v2: On-the-fly equivariant attention with linear activation memory, 2026

    Lin Huang, Chengxiang Huang, Ziang Wang, Yiyue Du, Chu Wang, Haocheng Lu, Yunyang Li, Xiaoli Liu, Arthur Jiang, and Jia Zhang. E2former-v2: On-the-fly equivariant attention with linear activation memory, 2026. URLhttps://arxiv.org/abs/2601.16622

    work page arXiv 2026

  46. [47]

    Wood, Aditi S

    Eric Qu, Brandon M. Wood, Aditi S. Krishnapriyan, and Zachary W. Ulissi. A recipe for scalable attention-based mlips: unlocking long-range accuracy with all-to-all node attention,

    work page

  47. [48]

    URLhttps://arxiv.org/abs/2603.06567

    work page arXiv

  48. [49]

    Elhag, Arun Raja, Alex Morehead, Samuel M

    Ahmed A. Elhag, Arun Raja, Alex Morehead, Samuel M. Blau, Hongtao Zhao, Christian Tyrchan, Eva Nittinger, Garrett M. Morris, and Michael M. Bronstein. Learning inter-atomic potentials without explicit equivariance, 2026. URL https://arxiv.org/abs/2510.00027

    work page arXiv 2026

  49. [50]

    Jraph: A library for graph neural networks in jax., 2020

    Jonathan Godwin, Thomas Keck, Peter Battaglia, Victor Bapst, Thomas Kipf, Yujia Li, Kimberly Stachenfeld, Petar Veliˇckovi´c, and Alvaro Sanchez-Gonzalez. Jraph: A library for graph neural networks in jax., 2020. URLhttp://github.com/deepmind/jraph

    work page 2020

  50. [51]

    Unke and Markus Meuwly

    Oliver T. Unke and Markus Meuwly. Physnet: A neural network for predicting energies, forces, dipole moments and partial charges.Journal of Chemical Theory and Computation, 15(6): 3678–3693, 2019. doi: 10.1021/acs.jctc.9b00181

    work page doi:10.1021/acs.jctc.9b00181 2019

  51. [52]

    Machine learning of force fields towards molecular dynamics simulations of proteins at DFT accuracy

    Christoph Brunken, Sebastien Boyer, Mustafa Omar, Bakary N’tji Diallo, Karim Beguir, Nico- las Lopez Carranza, and Oliver Bent. Machine learning of force fields towards molecular dynamics simulations of proteins at DFT accuracy. InICLR 2024 Workshop on Geometry- grounded Graph Machine Learning (GEM), 2024. URL https://openreview.net/forum? id=3YjclODCyq

    work page 2024

  52. [53]

    Aimnet2: a neural network potential to meet your neutral, charged, organic, and elemental-organic needs.Chemical Science, 16:10228,

    Dylan Anstine, Roman Zubatyuk, and Olexandr Isayev. Aimnet2: a neural network potential to meet your neutral, charged, organic, and elemental-organic needs.Chemical Science, 16:10228,

    work page

  53. [54]

    doi: 10.1039/D4SC08572H

    work page doi:10.1039/d4sc08572h

  54. [55]

    A new approach to variable metric algorithms.The Computer Journal, 13(3): 317–322, 1970

    Roger Fletcher. A new approach to variable metric algorithms.The Computer Journal, 13(3): 317–322, 1970

    work page 1970

  55. [56]

    A family of variable metric updates derived by variational means.Mathematics of Computation, 24(109):23–26, 1970

    Donald Goldfarb. A family of variable metric updates derived by variational means.Mathematics of Computation, 24(109):23–26, 1970

    work page 1970

  56. [57]

    Conditioning of quasi-newton methods for function minimization.Mathemat- ics of Computation, 24(111):647–656, 1970

    David F Shanno. Conditioning of quasi-newton methods for function minimization.Mathemat- ics of Computation, 24(111):647–656, 1970

    work page 1970

  57. [58]

    A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives.The Journal of Chemical Physics, 111(15):7010–7022, 1999

    Graeme Henkelman and Hannes Jónsson. A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives.The Journal of Chemical Physics, 111(15):7010–7022, 1999. doi: 10.1063/1.480097. 14

    work page doi:10.1063/1.480097 1999

  58. [59]

    Does hessian data improve the performance of machine learning potentials?Journal of Chemical Theory and Computation, 21 (14):6698–6710, 2025

    Austin Rodriguez, Justin S Smith, and Jose L Mendoza-Cortes. Does hessian data improve the performance of machine learning potentials?Journal of Chemical Theory and Computation, 21 (14):6698–6710, 2025

    work page 2025

  59. [60]

    Projected hessian learning: Fast curvature supervision for accurate machine- learning interatomic potentials.arXiv preprint arXiv:2603.04523, 2026

    Austin Rodriguez, Justin S Smith, Sakib Matin, Nicholas Lubbers, Kipton Barros, and Jose L Mendoza-Cortes. Projected hessian learning: Fast curvature supervision for accurate machine- learning interatomic potentials.arXiv preprint arXiv:2603.04523, 2026

    work page arXiv 2026

  60. [61]

    Shoot from the hip: Hessian interatomic potentials without derivatives, 2025

    Andreas Burger, Luca Thiede, Nikolaj Rønne, Varinia Bernales, Nandita Vijaykumar, Tejs Vegge, Arghya Bhowmik, and Alán Aspuru-Guzik. Shoot from the hip: Hessian interatomic potentials without derivatives, 2025. URLhttps://arxiv.org/abs/2509.21624

    work page arXiv 2025

  61. [62]

    Towards fast, specialized machine learning force fields: Distilling foundation models via energy hessians

    Ishan Amin, Sanjeev Raja, and Aditi Krishnapriyan. Towards fast, specialized machine learning force fields: Distilling foundation models via energy hessians. arxiv 2025, doi: 10.48550.arXiv preprint arXiv.2501.09009, 2025

    work page arXiv 2025

  62. [63]

    Elena, Sam Walton Norwood, Thomas Wolf, and Gábor Csányi

    Ilyes Batatia, Chen Lin, Joseph Hart, Elliott Kasoar, Alin M. Elena, Sam Walton Norwood, Thomas Wolf, and Gábor Csányi. Cross learning between electronic structure theories for unifying molecular, surface, and inorganic crystal foundation force fields, 10 2025. URL https://arxiv.org/pdf/2510.25380

    work page arXiv 2025

  63. [64]

    Improved tangent estimate in the nudged elastic band method for finding minimum energy paths and saddle points.The Journal of Chemical Physics, 113(22):9978–9985, 12 2000

    Graeme Henkelman and Hannes Jónsson. Improved tangent estimate in the nudged elastic band method for finding minimum energy paths and saddle points.The Journal of Chemical Physics, 113(22):9978–9985, 12 2000. ISSN 0021-9606. doi: 10.1063/1.1323224. URL https://doi.org/10.1063/1.1323224

    work page doi:10.1063/1.1323224 2000

  64. [65]

    Uberuaga, and Hannes Jónsson

    Graeme Henkelman, Blas P. Uberuaga, and Hannes Jónsson. A climbing image nudged elastic band method for finding saddle points and minimum energy paths.The Journal of Chemical Physics, 113(22):9901–9904, 12 2000. ISSN 0021-9606. doi: 10.1063/1.1329672. URL https://doi.org/10.1063/1.1329672

    work page doi:10.1063/1.1329672 2000

  65. [66]

    Improved initial guess for minimum energy path calculations.The Journal of Chemical Physics, 140(21):214106, 06 2014

    Søren Smidstrup, Andreas Pedersen, Kurt Stokbro, and Hannes Jónsson. Improved initial guess for minimum energy path calculations.The Journal of Chemical Physics, 140(21):214106, 06 2014. ISSN 0021-9606. doi: 10.1063/1.4878664. URL https://doi.org/10.1063/1. 4878664

    work page doi:10.1063/1.4878664 2014

  66. [67]

    Ferguson

    Kim-Hung Chow and David M. Ferguson. Isothermal-isobaric molecular dynamics simulations with monte carlo volume sampling.Computer Physics Communications, 91(1):283–289,

    work page

  67. [68]

    doi: https://doi.org/10.1016/0010-4655(95)00059-O

    ISSN 0010-4655. doi: https://doi.org/10.1016/0010-4655(95)00059-O. URL https: //www.sciencedirect.com/science/article/pii/001046559500059O

    work page doi:10.1016/0010-4655(95)00059-o

  68. [69]

    Brandsdal

    Johan Åqvist, Petra Wennerström, Martin Nervall, Sinisa Bjelic, and Bjørn O. Brandsdal. Molecular dynamics simulations of water and biomolecules with a monte carlo constant pressure algorithm.Chemical Physics Letters, 384(4):288–294, 2004. ISSN 0009-2614. doi: https: //doi.org/10.1016/j.cplett.2003.12.039. URL https://www.sciencedirect.com/science/ articl...

    work page doi:10.1016/j.cplett.2003.12.039 2004

  69. [70]

    GabrieleCorso,HannesStärk,BowenJing,ReginaBarzilay,andTommiJaakkola

    Peter Eastman, Jason Swails, John D. Chodera, Robert T. McGibbon, Yutong Zhao, Kyle A. Beauchamp, Lee-Ping Wang, Andrew C. Simmonett, Matthew P. Harrigan, Chaya D. Stern, Rafal P. Wiewiora, Bernard R. Brooks, and Vijay S. Pande. Openmm 7: Rapid development of high performance algorithms for molecular dynamics.PLOS Computational Biology, 13 (7):1–17, 07 20...

    work page doi:10.1371/journal.pcbi.1005659 2017

  70. [71]

    H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and J. R. Haak. Molecular dynamics with coupling to an external bath.The Journal of Chemical Physics, 81(8):3684–3690,

    work page

  71. [72]

    doi: 10.1063/1.448118. 15

    work page doi:10.1063/1.448118

  72. [73]

    Parrinello and A

    M. Parrinello and A. Rahman. Polymorphic transitions in single crystals: A new molecular dynamics method.Journal of Applied Physics, 52(12):7182–7190, 1981. doi: 10.1063/1.328693

    work page doi:10.1063/1.328693 1981

  73. [74]

    Characterizing dependence of samples along the langevin dynamics and algorithms via contraction of ϕ-mutual information

    Jiaming Liang, Siddharth Mitra, and Andre Wibisono. Characterizing dependence of samples along the langevin dynamics and algorithms via contraction of ϕ-mutual information. In Proceedings of Thirty Eighth Conference on Learning Theory, volume 291 ofProceedings of Machine Learning Research, pages 3730–3731. PMLR, 30 Jun–04 Jul 2025. URL https: //proceeding...

    work page 2025

  74. [75]

    Levine, Muhammed Shuaibi, Evan Walter Clark Spotte-Smith, Michael G

    Daniel S. Levine, Muhammed Shuaibi, Evan Walter Clark Spotte-Smith, Michael G. Taylor, Muhammad R. Hasyim, Kyle Michel, Ilyes Batatia, Gábor Csányi, Misko Dzamba, Peter Eastman, Nathan C. Frey, Xiang Fu, Vahe Gharakhanyan, Aditi S. Krishnapriyan, Joshua A. Rackers, Sanjeev Raja, Ammar Rizvi, Andrew S. Rosen, Zachary Ulissi, Santiago Vargas, C. Lawrence Zi...

    work page 2025

  75. [76]

    EquiformerV3: Scaling Efficient, Expressive, and General SE(3)-Equivariant Graph Attention Transformers

    Yi-Lun Liao, Alexander J. Hoffman, Sabrina C. Shen, Alexandre Duval, Sam Walton Norwood, and Tess Smidt. Equiformerv3: Scaling efficient, expressive, and general se(3)-equivariant graph attention transformers, 2026. URLhttps://arxiv.org/abs/2604.09130

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  76. [77]

    Dotson, Raimondas Galvelis, John E

    Peter Eastman, Pavan Kumar Behara, David L. Dotson, Raimondas Galvelis, John E. Herr, Josh T. Horton, Yuezhi Mao, John D. Chodera, Benjamin P. Pritchard, Yuanqing Wang, Gianni De Fabritiis, and Thomas E. Markland. Spice, a dataset of drug-like molecules and peptides for training machine learning potentials.Scientific Data, 10, January 2023. doi: 10.1038/ ...

    work page doi:10.1038/s41597-022-01882-6 2023

  77. [78]

    Transition1x - a dataset for building generalizable reactive machine learning potentials.Scientific Data, 9(1), December 2022

    Mathias Schreiner, Arghya Bhowmik, Tejs Vegge, Jonas Busk, and Ole Winther. Transition1x - a dataset for building generalizable reactive machine learning potentials.Scientific Data, 9(1), December 2022. ISSN 2052-4463. doi: 10.1038/s41597-022-01870-w. URL http: //dx.doi.org/10.1038/s41597-022-01870-w

    work page doi:10.1038/s41597-022-01870-w 2022

  78. [79]

    Ogunfowora, Sanjay S

    Qiyuan Zhao, Sai Mahit Vaddadi, Michael Woulfe, Lawal A. Ogunfowora, Sanjay S. Garimella, Olexandr Isayev, and Brett M. Savoie. Comprehensive exploration of graphically defined reaction spaces.Scientific Data, 10(1), March 2023. ISSN 2052-4463. doi: 10.1038/ s41597-023-02043-z. URLhttp://dx.doi.org/10.1038/s41597-023-02043-z

    work page doi:10.1038/s41597-023-02043-z 2023

  79. [80]

    Pyscf: the python- based simulations of chemistry framework.Wiley Interdisciplinary Reviews: Computational Molecular Science, 8(1):e1340, 2018

    Qiming Sun, Timothy C Berkelbach, Nick S Blunt, George H Booth, Sheng Guo, Zhendong Li, Junzi Liu, James D McClain, Elvira R Sayfutyarova, Sandeep Sharma, et al. Pyscf: the python- based simulations of chemistry framework.Wiley Interdisciplinary Reviews: Computational Molecular Science, 8(1):e1340, 2018

    work page 2018

  80. [81]

    Torchani: a free and open source pytorch-based deep learning implementation of the ani neural network potentials.Journal of chemical information and modeling, 60(7):3408–3415, 2020

    Xiang Gao, Farhad Ramezanghorbani, Olexandr Isayev, Justin S Smith, and Adrian E Roitberg. Torchani: a free and open source pytorch-based deep learning implementation of the ani neural network potentials.Journal of chemical information and modeling, 60(7):3408–3415, 2020

    work page 2020

Showing first 80 references.