Spin-Weighted Spherical Harmonics Enable Complete and Scalable $\mathrm{E}(3)$-Equivariant Networks

Andrii Kryvenko; Chenxing Liang; Chuan Li; Jianwen Xie; Shuiwang Ji; Wendi Yu; Xiaofeng Qian; Yuchao Lin

arxiv: 2607.01408 · v1 · pith:6R7GDMIAnew · submitted 2026-07-01 · 💻 cs.LG · cs.AI

Spin-Weighted Spherical Harmonics Enable Complete and Scalable E(3)-Equivariant Networks

Chenxing Liang , Yuchao Lin , Andrii Kryvenko , Wendi Yu , Chuan Li , Jianwen Xie , Xiaofeng Qian , Shuiwang Ji This is my paper

Pith reviewed 2026-07-03 21:19 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords E(3)-equivariant networksspin-weighted spherical harmonicsGaunt tensor productantisymmetric interactions3D atomistic modelingparity-odd componentsequivariant basisClebsch-Gordan tensor product

0 comments

The pith

Spin-weighted spherical harmonics recover the antisymmetric interactions missing from efficient Gaunt tensor products in E(3)-equivariant networks.

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

The paper seeks to make E(3)-equivariant networks both complete and scalable for 3D atomistic modeling by addressing the incompleteness of the Gaunt Tensor Product. GTP lowers complexity from the O(L^6) of Clebsch-Gordan products but omits antisymmetric paths; SpinGTP generalizes the construction to Spin-Weighted Spherical Harmonics to restore those paths. A reader would care because the resulting networks can then handle chiral materials and non-centrosymmetric geometries without reverting to the slower full tensor product. The method is shown to match CGTP accuracy on standard benchmarks while outperforming GTP on parity-sensitive tasks.

Core claim

SpinGTP overcomes GTP incompleteness by generalizing from scalar functions to Spin-Weighted Spherical Harmonics; the algebraic properties of SWSH recover the missing antisymmetric interactions, preserve GTP's asymptotic efficiency, and supply a more expressive equivariant basis that naturally includes parity-odd components of tensor products.

What carries the argument

SpinGTP, the generalization of the Gaunt Tensor Product that operates on Spin-Weighted Spherical Harmonics instead of scalar spherical harmonics to capture antisymmetric paths.

If this is right

SpinGTP reaches accuracies comparable to full CGTP on Tetris, 3BPA, SPICE-MACE-OFF, and OC20 benchmarks.
SpinGTP shows superior performance on tasks involving chiral materials and non-centrosymmetric geometries.
The approach maintains the asymptotic efficiency of the original GTP.
The resulting basis naturally accounts for parity-odd components without extra machinery.

Where Pith is reading between the lines

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

Large-scale atomistic simulations could adopt SpinGTP to gain completeness on asymmetric systems while retaining GTP-level scaling.
The same SWSH generalization might apply to other incomplete tensor-product constructions in equivariant models beyond E(3).
Practitioners modeling systems with handedness or inversion asymmetry now have a concrete replacement for incomplete GTP layers.

Load-bearing premise

The algebraic properties of Spin-Weighted Spherical Harmonics are sufficient to recover every missing antisymmetric path from the GTP construction without approximation error or increase in asymptotic complexity.

What would settle it

Run SpinGTP and GTP side-by-side on a chiral-molecule benchmark where full CGTP is known to succeed; if SpinGTP reaches CGTP-level accuracy while GTP remains measurably worse, the claim holds.

read the original abstract

$\mathrm{E}(3)$-equivariant networks are promising for 3D atomistic system modeling, yet their scalability is limited by the $O(L^6)$ complexity of the Clebsch-Gordan Tensor Product (CGTP). The recently proposed Gaunt Tensor Product (GTP) reduces the complexity but is unable to capture the antisymmetric paths, resulting in incomplete expressivity. In this work, we present SpinGTP, an approach to overcome the GTP incompleteness by generalizing from scalar functions to Spin-Weighted Spherical Harmonics (SWSH). By relying on the algebraic properties of SWSH, SpinGTP recovers the missing antisymmetric interactions while maintaining the asymptotic efficiency of GTP. It also allows for a more expressive equivariant basis that naturally accounts for the parity-odd components of tensor products. We evaluate SpinGTP across diverse benchmarks, including Tetris, 3BPA, SPICE-MACE-OFF, and OC20. Our results show that SpinGTP achieves accuracies comparable to full CGTP. Notably, by explicitly capturing antisymmetric paths, SpinGTP exhibits superior performance in tasks involving chiral materials and non-centrosymmetric geometries. This work provides a complete, scalable, and mathematically rigorous path toward high-order equivariance in large-scale 3D atomistic system simulations.

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.

Desk Editor's Note private letter to a colleague

SpinGTP uses SWSH to restore antisymmetric paths to GTP while keeping its scaling, with solid benchmark gains on chiral tasks but thin math details in the abstract.

read the letter

The main takeaway is that this paper gives a concrete algebraic fix for the incompleteness in Gaunt tensor products by moving to spin-weighted spherical harmonics. That step recovers the missing antisymmetric interactions and parity-odd components without changing the leading complexity.

The construction itself is the new piece. It generalizes the scalar case directly and the authors show it produces a more expressive basis. On the experiments, SpinGTP matches full CGTP accuracy on Tetris, 3BPA, SPICE-MACE-OFF, and OC20 while pulling ahead on chiral and non-centrosymmetric cases. That lines up with the claim that the added paths matter in practice.

The soft spot is the level of verification. The abstract states the algebraic properties suffice for exact recovery and unchanged scaling, but the provided text does not include the explicit derivation or complexity proof. The stress-test concern about whether every antisymmetric channel is spanned and whether any hidden L-dependent factors appear is reasonable to check in the full math section. If those details are there and clean, the central claim holds; if they rely on unstated assumptions, the completeness part needs tightening.

Experiments are run on standard benchmarks and the gains on chiral materials are reported plainly, which is useful. No obvious circularity or invented quantities.

This paper is for people working on scalable high-order E(3)-equivariant models for molecules and materials. Readers who care about the expressivity-scalability tradeoff will get something concrete from the construction. It is grounded enough in the existing GTP and CGTP literature to deserve a serious referee, even if revisions will focus on the derivation details and ablations.

I would send it to peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes SpinGTP, an E(3)-equivariant architecture that generalizes the Gaunt Tensor Product (GTP) by replacing scalar spherical harmonics with Spin-Weighted Spherical Harmonics (SWSH). The central claim is that the algebraic properties of the SWSH basis recover all antisymmetric tensor-product paths omitted by GTP, preserve GTP's asymptotic efficiency, and naturally incorporate parity-odd components, yielding accuracies comparable to full Clebsch-Gordan tensor products (CGTP) on Tetris, 3BPA, SPICE-MACE-OFF, and OC20 while improving results on chiral and non-centrosymmetric tasks.

Significance. If the claimed exact algebraic recovery holds without approximation or hidden complexity costs, the result would supply a complete yet scalable route to high-order E(3) equivariance for atomistic modeling, directly addressing the expressivity-efficiency trade-off that currently limits GTP-based networks on chiral systems.

major comments (2)

[§3 (SpinGTP construction)] The central claim (abstract and §3 construction) that SWSH algebraic properties exactly recover every GTP-missing antisymmetric path with no approximation and no change to leading-order complexity is load-bearing yet unsupported by an explicit derivation showing that the weighted basis spans the full antisymmetric subspace of the CGTP tensor product and that the resulting contraction retains GTP scaling; without this, the completeness statement remains an unverified assertion.
[Table 2 and §5.2] Table 2 and the associated benchmark discussion report comparable or superior accuracy on chiral tasks but provide no ablation isolating the contribution of the recovered antisymmetric channels versus other modeling choices; this leaves open whether the observed gains are attributable to the claimed completeness or to incidental differences in basis expressivity.

minor comments (2)

[§3.1] The notation for spin weights and the precise definition of the SWSH tensor contraction should be introduced with an explicit equation before the complexity argument is made.
[Figure 1] Figure 1 caption should state the precise L values and spin weights used in the illustrated basis change.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We address the two major comments point-by-point below. Where the comments correctly identify gaps in the current presentation, we will revise the manuscript accordingly.

read point-by-point responses

Referee: [§3 (SpinGTP construction)] The central claim (abstract and §3 construction) that SWSH algebraic properties exactly recover every GTP-missing antisymmetric path with no approximation and no change to leading-order complexity is load-bearing yet unsupported by an explicit derivation showing that the weighted basis spans the full antisymmetric subspace of the CGTP tensor product and that the resulting contraction retains GTP scaling; without this, the completeness statement remains an unverified assertion.

Authors: We agree that the manuscript would benefit from a more explicit derivation of the completeness property. The current §3 presents the construction via the algebraic properties of SWSH (their rotation transformation laws and weight selection rules) that recover the antisymmetric tensor-product paths omitted by scalar spherical harmonics, but does not include a self-contained proof that the resulting basis exactly spans the full antisymmetric subspace of the CGTP product while preserving the leading O(L^4) scaling of GTP. In the revised manuscript we will add this derivation, either as an expanded subsection or a dedicated appendix, showing the relevant Clebsch-Gordan coefficients and contraction identities. revision: yes
Referee: [Table 2 and §5.2] Table 2 and the associated benchmark discussion report comparable or superior accuracy on chiral tasks but provide no ablation isolating the contribution of the recovered antisymmetric channels versus other modeling choices; this leaves open whether the observed gains are attributable to the claimed completeness or to incidental differences in basis expressivity.

Authors: The referee is correct that no dedicated ablation isolating only the antisymmetric channels appears in the current experiments. The reported comparisons hold all other architectural choices fixed and contrast SpinGTP against both GTP (which lacks the channels) and full CGTP (which includes them), so the performance differences on chiral and non-centrosymmetric tasks are consistent with the completeness claim. Nevertheless, an explicit ablation would strengthen the attribution. We will add a clarifying paragraph in §5.2 that uses the GTP baseline to isolate the effect and, if space permits, include a limited additional experiment that toggles only the antisymmetric paths. revision: partial

Circularity Check

0 steps flagged

No circularity; central claim is direct algebraic generalization from SWSH properties.

full rationale

The paper's derivation chain rests on the stated algebraic properties of Spin-Weighted Spherical Harmonics to recover antisymmetric tensor-product paths from the GTP construction while preserving asymptotic scaling. This is presented as a mathematical generalization from scalar functions, not as a fitted parameter, self-defined quantity, or result justified solely by overlapping self-citations. No equations or steps in the provided abstract reduce the completeness or efficiency claim to the inputs by construction. Empirical benchmarks serve as external validation rather than load-bearing support for the algebraic step. The construction is therefore self-contained against the listed circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unproven-in-abstract algebraic closure property of SWSH for antisymmetric paths; no free parameters or new entities are mentioned.

axioms (1)

domain assumption Algebraic properties of Spin-Weighted Spherical Harmonics recover all missing antisymmetric interactions from GTP without altering asymptotic complexity
Invoked as the basis for SpinGTP recovering completeness while keeping GTP efficiency.

pith-pipeline@v0.9.1-grok · 5797 in / 1259 out tokens · 27469 ms · 2026-07-03T21:19:39.983867+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

36 extracted references · 8 canonical work pages · 2 internal anchors

[1]

Cormorant: Covariant molecular neural networks

Brandon Anderson, Truong Son Hy, and Risi Kondor. Cormorant: Covariant molecular neural networks. Advances in neural information processing systems, 32, 2019

2019
[2]

Mace: Higher order equivariant message passing neural networks for fast and accurate force fields.Advances in neural information processing systems, 35:11423–11436, 2022

Ilyes Batatia, David P Kovacs, Gregor Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields.Advances in neural information processing systems, 35:11423–11436, 2022

2022
[3]

The design space of e (3)-equivariant atom-centred interatomic potentials.Nature Machine Intelligence, 7(1):56, 2025

Ilyes Batatia, Simon Batzner, Dávid Péter Kovács, Albert Musaelian, Gregor NC Simm, Ralf Drautz, Christoph Ortner, Boris Kozinsky, and Gábor Csányi. The design space of e (3)-equivariant atom-centred interatomic potentials.Nature Machine Intelligence, 7(1):56, 2025

2025
[4]

E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials.Nature communications, 13(1):2453, 2022

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):2453, 2022

2022
[5]

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veličković. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges.arXiv preprint arXiv:2104.13478, 2021

work page internal anchor Pith review Pith/arXiv arXiv 2021
[6]

Open catalyst 2020 (oc20) dataset and community challenges.Acs Catalysis, 11(10):6059–6072, 2021

LowikChanussot, AbhishekDas, SiddharthGoyal, ThibautLavril, MuhammedShuaibi, MorganeRiviere, Kevin Tran, Javier Heras-Domingo, Caleb Ho, Weihua Hu, et al. Open catalyst 2020 (oc20) dataset and community challenges.Acs Catalysis, 11(10):6059–6072, 2021

2020
[7]

Rotation invariance and equivariance in 3d deep learning: a survey

Jiajun Fei and Zhidong Deng. Rotation invariance and equivariance in 3d deep learning: a survey. Artificial Intelligence Review, 57(7):168, 2024

2024
[8]

Augmenting molecular graphs with geometries via machine learning interatomic potentials.arXiv preprint arXiv:2507.00407, 2025

Cong Fu, Yuchao Lin, Zachary Krueger, Haiyang Yu, Maho Nakata, Jianwen Xie, Emine Kucukbenli, Xiaofeng Qian, and Shuiwang Ji. Augmenting molecular graphs with geometries via machine learning interatomic potentials.arXiv preprint arXiv:2507.00407, 2025

work page arXiv 2025
[9]

Margraf and Stephan G

Johannes Gasteiger, Shankari Giri, Johannes T Margraf, and Stephan Günnemann. Fast and uncertainty- aware directional message passing for non-equilibrium molecules.arXiv preprint arXiv:2011.14115, 2020

work page arXiv 2011
[10]

Gemnet: Universal directional graph neural networks for molecules.Advances in neural information processing systems, 34:6790–6802, 2021

Johannes Gasteiger, Florian Becker, and Stephan Günnemann. Gemnet: Universal directional graph neural networks for molecules.Advances in neural information processing systems, 34:6790–6802, 2021

2021
[11]

Euclidean neural networks: e3nn, April 2022

Mario Geiger, Tess Smidt, Alby M., Benjamin Kurt Miller, Wouter Boomsma, Bradley Dice, Kostiantyn Lapchevskyi, Maurice Weiler, Michał Tyszkiewicz, Simon Batzner, Dylan Madisetti, Martin Uhrin, Jes Frellsen, Nuri Jung, Sophia Sanborn, Mingjian Wen, Josh Rackers, Marcel Rød, and Michael Bailey. Euclidean neural networks: e3nn, April 2022. URLhttps://doi.org...

work page doi:10.5281/zenodo 2022
[12]

Spin-s spherical harmonics and ð.Journal of Mathematical Physics, 8(11):2155–2161, 1967

Joshua N Goldberg, Alan J MacFarlane, Ezra T Newman, Fritz Rohrlich, and EC George Sudarshan. Spin-s spherical harmonics and ð.Journal of Mathematical Physics, 8(11):2155–2161, 1967

1967
[13]

Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces.Digital Discovery, 1(3):333–343, 2022

Mojtaba Haghighatlari, Jie Li, Xingyi Guan, Oufan Zhang, Akshaya Das, Christopher J Stein, Farnaz Heidar-Zadeh, Meili Liu, Martin Head-Gordon, Luke Bertels, et al. Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces.Digital Discovery, 1(3):333–343, 2022

2022
[14]

Ffts for the 2-sphere- improvements and variations.Journal of Fourier analysis and applications, 9(4):341–385, 2003

Dennis M Healy Jr, Daniel N Rockmore, Peter J Kostelec, and Sean Moore. Ffts for the 2-sphere- improvements and variations.Journal of Fourier analysis and applications, 9(4):341–385, 2003

2003
[15]

Quantum theory of angular momemtum

Valery Kelmanovich Khersonskii, Anatoly Nikolaevich Moskalev, and Dmitry Alexandrovich Varshalovich. Quantum theory of angular momemtum. World Scientific, 1988

1988
[16]

The principles behind equivariant neural networks for physics and chemistry.Proceedings of the National Academy of Sciences, 122(41):e2415656122, 2025

Risi Kondor. The principles behind equivariant neural networks for physics and chemistry.Proceedings of the National Academy of Sciences, 122(41):e2415656122, 2025

2025
[17]

Mace-off: Short-range transferable machine learning force fields for organic molecules.Journal of the American Chemical Society, 147(21):17598–17611, 2025

Dávid Péter Kovács, J Harry Moore, Nicholas J Browning, Ilyes Batatia, Joshua T Horton, Yixuan Pu, Venkat Kapil, William C Witt, Ioan-Bogdan Magdau, Daniel J Cole, et al. Mace-off: Short-range transferable machine learning force fields for organic molecules.Journal of the American Chemical Society, 147(21):17598–17611, 2025

2025
[18]

Dávid Péter Kovács, Cas van der Oord, Jiri Kucera, Alice E. A. Allen, Daniel J. Cole, Christoph Ortner, and Gábor Csányi. Linear atomic cluster expansion force fields for organic molecules: Beyond rmse. Journal of Chemical Theory and Computation, 17(12):7696–7711, 2021. doi: 10.1021/acs.jctc.1c00647. URLhttps://doi.org/10.1021/acs.jctc.1c00647. PMID: 34735161

work page doi:10.1021/acs.jctc.1c00647 2021
[19]

Scalfani, Daniel Probst, Kazuya Ujihara, Jeremy Monat, and Juuso Lehtivarjo

Greg Landrum, Paolo Tosco, Brian Kelley, Ricardo Rodriguez, David Cosgrove, Riccardo Vianello, sriniker, Peter Gedeck, Gareth Jones, Eisuke Kawashima, NadineSchneider, Dan Nealschneider, tadhurst cdd, Andrew Dalke, Matt Swain, Brian Cole, Samo Turk, Aleksandr Savelev, Niels Maeder, Yakov Pechersky, Alain Vaucher, Maciej Wójcikowski, Rachel Walker, Hussein...

work page doi:10.5281/zenodo.19922430 2026
[20]

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. URLhttps://openreview.net/forum?id=l...

2026
[21]

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. URL https: //openreview.net/forum?id=KwmPfARgOTD

2023
[22]

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. URLhttps://openreview.net/forum?id=mCOBKZmrzD

2024
[23]

Tensor decomposition networks for accelerating machine learning force field computations

Yuchao Lin, Cong Fu, Zachary Krueger, Haiyang Yu, Maho Nakata, Jianwen Xie, Emine Kucukbenli, Xiaofeng Qian, and Shuiwang Ji. Tensor decomposition networks for accelerating machine learning force field computations. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URLhttps://openreview.net/forum?id=9vKJyCUfMH. 15 Spin-W...

2025
[24]

Spherical message passing for 3D molecular graphs

Yi Liu, Limei Wang, Meng Liu, Yuchao Lin, Xuan Zhang, Bora Oztekin, and Shuiwang Ji. Spherical message passing for 3D molecular graphs. InInternational Conference on Learning Representations, 2022

2022
[25]

Enabling efficient equivariant operations in the fourier basis via gaunt tensor products

Shengjie Luo, Tianlang Chen, and Aditi S Krishnapriyan. Enabling efficient equivariant operations in the fourier basis via gaunt tensor products. InThe Twelfth International Conference on Learning Representations, 2024

2024
[26]

Learning local equivariant representations for large-scale atomistic dynamics

Albert Musaelian, Simon Batzner, Anders Johansson, Lixin Sun, Cameron J Owen, Mordechai Kornbluth, and Boris Kozinsky. Learning local equivariant representations for large-scale atomistic dynamics. Nature Communications, 14(1):579, 2023

2023
[27]

Reducing so (3) convolutions to so (2) for efficient equivariant gnns

Saro Passaro and C Lawrence Zitnick. Reducing so (3) convolutions to so (2) for efficient equivariant gnns. InInternational conference on machine learning, pages 27420–27438. PMLR, 2023

2023
[28]

General relativistic null-cone evolutions with a high-order scheme.General Relativity and Gravitation, 45(5):1069–1094, 2013

Christian Reisswig, Nigel Bishop, and Denis Pollney. General relativistic null-cone evolutions with a high-order scheme.General Relativity and Gravitation, 45(5):1069–1094, 2013

2013
[29]

Schnet: A continuous-filter convolutional neural network for modeling quantum interactions.Advances in neural information processing systems, 30, 2017

Kristof Schütt, Pieter-Jan Kindermans, Huziel Enoc Sauceda Felix, Stefan Chmiela, Alexandre Tkatchenko, and Klaus-Robert Müller. Schnet: A continuous-filter convolutional neural network for modeling quantum interactions.Advances in neural information processing systems, 30, 2017

2017
[30]

Equivariant message passing for the prediction of tensorial properties and molecular spectra

Kristof Schütt, Oliver Unke, and Michael Gastegger. Equivariant message passing for the prediction of tensorial properties and molecular spectra. InInternational conference on machine learning, pages 9377–9388. PMLR, 2021

2021
[31]

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.arXiv preprint arXiv:1802.08219, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[32]

Scalars are universal: Equivariantmachinelearning, structured likeclassicalphysics.Advancesinneuralinformation processing systems, 34:28848–28863, 2021

Soledad Villar, David W Hogg, Kate Storey-Fisher, Weichi Yao, and Ben Blum-Smith. Scalars are universal: Equivariantmachinelearning, structured likeclassicalphysics.Advancesinneuralinformation processing systems, 34:28848–28863, 2021

2021
[33]

Comenet: Towards complete and efficient message passing for 3d molecular graphs.Advances in Neural Information Processing Systems, 35:650–664, 2022

Limei Wang, Yi Liu, Yuchao Lin, Haoran Liu, and Shuiwang Ji. Comenet: Towards complete and efficient message passing for 3d molecular graphs.Advances in Neural Information Processing Systems, 35:650–664, 2022

2022
[34]

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. InInternational Conference on Machine Learning, pages 68599–68625. PMLR, 2025

2025
[35]

Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics.arXiv preprint arXiv:2602.21466, 2026

YuQing Xie, Ameya Daigavane, Mit Kotak, and Tess Smidt. Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics.arXiv preprint arXiv:2602.21466, 2026

work page arXiv 2026
[36]

Artificial intelligence for science in quantum, atomistic, and continuum systems.Foundations and Trends®in Machine Learning, 18(4):385–849, 2025

Xuan Zhang, Limei Wang, Jacob Helwig, Youzhi Luo, Cong Fu, Yaochen Xie, Meng Liu, Yuchao Lin, Zhao Xu, Keqiang Yan, et al. Artificial intelligence for science in quantum, atomistic, and continuum systems.Foundations and Trends®in Machine Learning, 18(4):385–849, 2025. 16 Spin-Weighted Spherical Harmonics Enable Complete and ScalableE(3)-Equivariant Networ...

2025

[1] [1]

Cormorant: Covariant molecular neural networks

Brandon Anderson, Truong Son Hy, and Risi Kondor. Cormorant: Covariant molecular neural networks. Advances in neural information processing systems, 32, 2019

2019

[2] [2]

Mace: Higher order equivariant message passing neural networks for fast and accurate force fields.Advances in neural information processing systems, 35:11423–11436, 2022

Ilyes Batatia, David P Kovacs, Gregor Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields.Advances in neural information processing systems, 35:11423–11436, 2022

2022

[3] [3]

The design space of e (3)-equivariant atom-centred interatomic potentials.Nature Machine Intelligence, 7(1):56, 2025

Ilyes Batatia, Simon Batzner, Dávid Péter Kovács, Albert Musaelian, Gregor NC Simm, Ralf Drautz, Christoph Ortner, Boris Kozinsky, and Gábor Csányi. The design space of e (3)-equivariant atom-centred interatomic potentials.Nature Machine Intelligence, 7(1):56, 2025

2025

[4] [4]

E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials.Nature communications, 13(1):2453, 2022

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):2453, 2022

2022

[5] [5]

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges

Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veličković. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges.arXiv preprint arXiv:2104.13478, 2021

work page internal anchor Pith review Pith/arXiv arXiv 2021

[6] [6]

Open catalyst 2020 (oc20) dataset and community challenges.Acs Catalysis, 11(10):6059–6072, 2021

LowikChanussot, AbhishekDas, SiddharthGoyal, ThibautLavril, MuhammedShuaibi, MorganeRiviere, Kevin Tran, Javier Heras-Domingo, Caleb Ho, Weihua Hu, et al. Open catalyst 2020 (oc20) dataset and community challenges.Acs Catalysis, 11(10):6059–6072, 2021

2020

[7] [7]

Rotation invariance and equivariance in 3d deep learning: a survey

Jiajun Fei and Zhidong Deng. Rotation invariance and equivariance in 3d deep learning: a survey. Artificial Intelligence Review, 57(7):168, 2024

2024

[8] [8]

Augmenting molecular graphs with geometries via machine learning interatomic potentials.arXiv preprint arXiv:2507.00407, 2025

Cong Fu, Yuchao Lin, Zachary Krueger, Haiyang Yu, Maho Nakata, Jianwen Xie, Emine Kucukbenli, Xiaofeng Qian, and Shuiwang Ji. Augmenting molecular graphs with geometries via machine learning interatomic potentials.arXiv preprint arXiv:2507.00407, 2025

work page arXiv 2025

[9] [9]

Margraf and Stephan G

Johannes Gasteiger, Shankari Giri, Johannes T Margraf, and Stephan Günnemann. Fast and uncertainty- aware directional message passing for non-equilibrium molecules.arXiv preprint arXiv:2011.14115, 2020

work page arXiv 2011

[10] [10]

Gemnet: Universal directional graph neural networks for molecules.Advances in neural information processing systems, 34:6790–6802, 2021

Johannes Gasteiger, Florian Becker, and Stephan Günnemann. Gemnet: Universal directional graph neural networks for molecules.Advances in neural information processing systems, 34:6790–6802, 2021

2021

[11] [11]

Euclidean neural networks: e3nn, April 2022

Mario Geiger, Tess Smidt, Alby M., Benjamin Kurt Miller, Wouter Boomsma, Bradley Dice, Kostiantyn Lapchevskyi, Maurice Weiler, Michał Tyszkiewicz, Simon Batzner, Dylan Madisetti, Martin Uhrin, Jes Frellsen, Nuri Jung, Sophia Sanborn, Mingjian Wen, Josh Rackers, Marcel Rød, and Michael Bailey. Euclidean neural networks: e3nn, April 2022. URLhttps://doi.org...

work page doi:10.5281/zenodo 2022

[12] [12]

Spin-s spherical harmonics and ð.Journal of Mathematical Physics, 8(11):2155–2161, 1967

Joshua N Goldberg, Alan J MacFarlane, Ezra T Newman, Fritz Rohrlich, and EC George Sudarshan. Spin-s spherical harmonics and ð.Journal of Mathematical Physics, 8(11):2155–2161, 1967

1967

[13] [13]

Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces.Digital Discovery, 1(3):333–343, 2022

Mojtaba Haghighatlari, Jie Li, Xingyi Guan, Oufan Zhang, Akshaya Das, Christopher J Stein, Farnaz Heidar-Zadeh, Meili Liu, Martin Head-Gordon, Luke Bertels, et al. Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces.Digital Discovery, 1(3):333–343, 2022

2022

[14] [14]

Ffts for the 2-sphere- improvements and variations.Journal of Fourier analysis and applications, 9(4):341–385, 2003

Dennis M Healy Jr, Daniel N Rockmore, Peter J Kostelec, and Sean Moore. Ffts for the 2-sphere- improvements and variations.Journal of Fourier analysis and applications, 9(4):341–385, 2003

2003

[15] [15]

Quantum theory of angular momemtum

Valery Kelmanovich Khersonskii, Anatoly Nikolaevich Moskalev, and Dmitry Alexandrovich Varshalovich. Quantum theory of angular momemtum. World Scientific, 1988

1988

[16] [16]

The principles behind equivariant neural networks for physics and chemistry.Proceedings of the National Academy of Sciences, 122(41):e2415656122, 2025

Risi Kondor. The principles behind equivariant neural networks for physics and chemistry.Proceedings of the National Academy of Sciences, 122(41):e2415656122, 2025

2025

[17] [17]

Mace-off: Short-range transferable machine learning force fields for organic molecules.Journal of the American Chemical Society, 147(21):17598–17611, 2025

Dávid Péter Kovács, J Harry Moore, Nicholas J Browning, Ilyes Batatia, Joshua T Horton, Yixuan Pu, Venkat Kapil, William C Witt, Ioan-Bogdan Magdau, Daniel J Cole, et al. Mace-off: Short-range transferable machine learning force fields for organic molecules.Journal of the American Chemical Society, 147(21):17598–17611, 2025

2025

[18] [18]

Dávid Péter Kovács, Cas van der Oord, Jiri Kucera, Alice E. A. Allen, Daniel J. Cole, Christoph Ortner, and Gábor Csányi. Linear atomic cluster expansion force fields for organic molecules: Beyond rmse. Journal of Chemical Theory and Computation, 17(12):7696–7711, 2021. doi: 10.1021/acs.jctc.1c00647. URLhttps://doi.org/10.1021/acs.jctc.1c00647. PMID: 34735161

work page doi:10.1021/acs.jctc.1c00647 2021

[19] [19]

Scalfani, Daniel Probst, Kazuya Ujihara, Jeremy Monat, and Juuso Lehtivarjo

Greg Landrum, Paolo Tosco, Brian Kelley, Ricardo Rodriguez, David Cosgrove, Riccardo Vianello, sriniker, Peter Gedeck, Gareth Jones, Eisuke Kawashima, NadineSchneider, Dan Nealschneider, tadhurst cdd, Andrew Dalke, Matt Swain, Brian Cole, Samo Turk, Aleksandr Savelev, Niels Maeder, Yakov Pechersky, Alain Vaucher, Maciej Wójcikowski, Rachel Walker, Hussein...

work page doi:10.5281/zenodo.19922430 2026

[20] [20]

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. URLhttps://openreview.net/forum?id=l...

2026

[21] [21]

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. URL https: //openreview.net/forum?id=KwmPfARgOTD

2023

[22] [22]

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. URLhttps://openreview.net/forum?id=mCOBKZmrzD

2024

[23] [23]

Tensor decomposition networks for accelerating machine learning force field computations

Yuchao Lin, Cong Fu, Zachary Krueger, Haiyang Yu, Maho Nakata, Jianwen Xie, Emine Kucukbenli, Xiaofeng Qian, and Shuiwang Ji. Tensor decomposition networks for accelerating machine learning force field computations. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URLhttps://openreview.net/forum?id=9vKJyCUfMH. 15 Spin-W...

2025

[24] [24]

Spherical message passing for 3D molecular graphs

Yi Liu, Limei Wang, Meng Liu, Yuchao Lin, Xuan Zhang, Bora Oztekin, and Shuiwang Ji. Spherical message passing for 3D molecular graphs. InInternational Conference on Learning Representations, 2022

2022

[25] [25]

Enabling efficient equivariant operations in the fourier basis via gaunt tensor products

Shengjie Luo, Tianlang Chen, and Aditi S Krishnapriyan. Enabling efficient equivariant operations in the fourier basis via gaunt tensor products. InThe Twelfth International Conference on Learning Representations, 2024

2024

[26] [26]

Learning local equivariant representations for large-scale atomistic dynamics

Albert Musaelian, Simon Batzner, Anders Johansson, Lixin Sun, Cameron J Owen, Mordechai Kornbluth, and Boris Kozinsky. Learning local equivariant representations for large-scale atomistic dynamics. Nature Communications, 14(1):579, 2023

2023

[27] [27]

Reducing so (3) convolutions to so (2) for efficient equivariant gnns

Saro Passaro and C Lawrence Zitnick. Reducing so (3) convolutions to so (2) for efficient equivariant gnns. InInternational conference on machine learning, pages 27420–27438. PMLR, 2023

2023

[28] [28]

General relativistic null-cone evolutions with a high-order scheme.General Relativity and Gravitation, 45(5):1069–1094, 2013

Christian Reisswig, Nigel Bishop, and Denis Pollney. General relativistic null-cone evolutions with a high-order scheme.General Relativity and Gravitation, 45(5):1069–1094, 2013

2013

[29] [29]

Schnet: A continuous-filter convolutional neural network for modeling quantum interactions.Advances in neural information processing systems, 30, 2017

Kristof Schütt, Pieter-Jan Kindermans, Huziel Enoc Sauceda Felix, Stefan Chmiela, Alexandre Tkatchenko, and Klaus-Robert Müller. Schnet: A continuous-filter convolutional neural network for modeling quantum interactions.Advances in neural information processing systems, 30, 2017

2017

[30] [30]

Equivariant message passing for the prediction of tensorial properties and molecular spectra

Kristof Schütt, Oliver Unke, and Michael Gastegger. Equivariant message passing for the prediction of tensorial properties and molecular spectra. InInternational conference on machine learning, pages 9377–9388. PMLR, 2021

2021

[31] [31]

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.arXiv preprint arXiv:1802.08219, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[32] [32]

Scalars are universal: Equivariantmachinelearning, structured likeclassicalphysics.Advancesinneuralinformation processing systems, 34:28848–28863, 2021

Soledad Villar, David W Hogg, Kate Storey-Fisher, Weichi Yao, and Ben Blum-Smith. Scalars are universal: Equivariantmachinelearning, structured likeclassicalphysics.Advancesinneuralinformation processing systems, 34:28848–28863, 2021

2021

[33] [33]

Comenet: Towards complete and efficient message passing for 3d molecular graphs.Advances in Neural Information Processing Systems, 35:650–664, 2022

Limei Wang, Yi Liu, Yuchao Lin, Haoran Liu, and Shuiwang Ji. Comenet: Towards complete and efficient message passing for 3d molecular graphs.Advances in Neural Information Processing Systems, 35:650–664, 2022

2022

[34] [34]

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. InInternational Conference on Machine Learning, pages 68599–68625. PMLR, 2025

2025

[35] [35]

Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics.arXiv preprint arXiv:2602.21466, 2026

YuQing Xie, Ameya Daigavane, Mit Kotak, and Tess Smidt. Asymptotically fast clebsch-gordan tensor products with vector spherical harmonics.arXiv preprint arXiv:2602.21466, 2026

work page arXiv 2026

[36] [36]

Artificial intelligence for science in quantum, atomistic, and continuum systems.Foundations and Trends®in Machine Learning, 18(4):385–849, 2025

Xuan Zhang, Limei Wang, Jacob Helwig, Youzhi Luo, Cong Fu, Yaochen Xie, Meng Liu, Yuchao Lin, Zhao Xu, Keqiang Yan, et al. Artificial intelligence for science in quantum, atomistic, and continuum systems.Foundations and Trends®in Machine Learning, 18(4):385–849, 2025. 16 Spin-Weighted Spherical Harmonics Enable Complete and ScalableE(3)-Equivariant Networ...

2025