Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

The Price of Freedom: Exploring Expressivity and Runtime Tradeoffs in Equivariant Tensor Products

T0 review · 2 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that the reported speedups of Gaunt and matrix tensor products over the Clebsch-Gordan tensor product are purchased by discarding expressive capacity, and that once runtime is normalized by expressivity, only a fast…

desk verdict GTP/MTP speedups indeed come from lost expressivity, but the expressivity measure is a parameter count rather than task capacity, so the 'only true speedup' claim needs that caveat. read the letter →

arxiv 2506.13523 v2 pith:6VRR4MV6 submitted 2025-06-16 cs.LG cs.AI

YuQing Xie , Ameya Daigavane , Mit Kotak , Tess Smidt This is my paper · ORCID
classification cs.LGcs.AI
keywords E(3)-equivariantneuralnetworkstensorproductoperationsClebsch-GordanGauntmatrixexpressivityinteractabilitysphericalharmonics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that the reported speedups of the Gaunt tensor product and the matrix tensor product over the standard Clebsch-Gordan tensor product are not free: the faster operations are not the same operation, and their asymptotic gains come with reduced capacity to construct distinct bilinear interactions. To make that precise, the authors define the expressivity of a tensor product operation as the dimension of the space of equivariant bilinear maps obtainable by composing the operation with equivariant linear layers, and interactability as the set of irrep triples it can couple. Under this measure GTP and MTP have expressivity $O(L)$ versus $O(L^3)$ for CGTP, and once runtime is divided by expressivity every alternative has the same asymptotics as sparse CGTP, leaving the fast spherical-harmonic transform variant as the only true speedup. The paper also contributes a simpler spherical-grid implementation of GTP that is a drop-in replacement and about 30% faster in training the MACE interatomic potential, plus microbenchmarks showing that asymptotic order, FLOP counts, GPU utilization, and wall-clock time disagree enough to make application-specific benchmarking necessary. The practical stakes are that choosing a tensor product changes what a model can learn, not merely how fast it runs.

What carries the argument

The load-bearing object is the tensor product operation (TPO), an equivariant bilinear map $T: X' \times Y' \to Z'$ used in place of the mathematical tensor product, together with a definition of expressivity as the dimension of the space of equivariant bilinear maps constructible by composing $T$ with equivariant linear layers on the inputs and output. The engine of the argument is the selection rule, a condition on irrep labels that a TPO must satisfy for an interaction to be nonzero, which determines interactability. For GTP, treating irreps up to degree $L$ as spherical-harmonic coefficients turns the operation into pointwise multiplication of two functions on the sphere $S^2$ followed by spherical-harmonic decomposition, and the resulting parity condition forces $\ell_1+\ell_2+\ell_3$ to be even, excluding antisymmetric paths. For MTP, embedding irreps into tensor-product representations and multiplying the resulting matrices inherits the CGTP triangle condition but still emits only one channel per output irrep. The runtime analysis leans on the sparsity of the Clebsch-Gordan coefficients, where only paths with $m_1+m_2=m_3$ survive, cutting the naive CGTP complexity from $O(L^6)$ to $O(L^5)$.

What would settle it

The chiral Tetris classification is the sharpest test: under the paper's selection rules, no GTP network of any width or depth can construct the antisymmetric $1_o \otimes 1_o \to 1_e$ path needed to tell mirror-image pieces apart, so a GTP model that reaches above-chance accuracy on the chiral pair would refute the claim that the missing interactions are structurally unreachable. A complementary check measures wall-clock time of the S2FFT variant against sparse CGTP at increasing $L$; if CGTP times fail to track $O(L^5)$ while S2FFT tracks $O(L^2 \log^2 L)$, the expressivity-normalized asymptotics of Table 1 would need to be revisited.

Watch

Extended reading notes

Core claim

The paper's central claim is that most operations called tensor products in $E(3)$-equivariant networks are equivariant bilinear maps the authors rename tensor product operations (TPOs), and that substituting a TPO for the true tensor product trades away computable functions, not just speed. Measured by the dimension of constructible bilinearities, GTP and MTP keep only $O(L)$ of the $O(L^3)$ bilinear degrees of freedom that CGTP retains, and they merge the multiple paths yielding the same output irrep into a single channel. Consequently, when runtimes are normalized by expressivity, GTP and MTP fall back to the same $O(L^2)$ scaling as sparse CGTP, and the only genuinely faster option is GTP computed through a fast spherical-harmonic transform, at $O(L^2 \log^2 L)$ runtime and $O(L \log^2 L)$ per unit of expressivity. The selection-rule analysis shows GTP is inherently symmetric and cannot produce antisymmetric interactions such as the cross product, and the paper demonstrates the consequence with a chiral-classification experiment in which CGTP networks reach perfect accuracy while GTP networks cannot separate mirror-image pieces. It further shows the often-quoted $O(L^6)$ runtime for CGTP is not tight, since exploiting the sparsity of the Clebsch-Gordan coefficients brings it down to $O(L^5)$.

Load-bearing premise

The whole comparison stands or falls on the premise that the dimension of the space of bilinear maps a tensor product operation can construct is a faithful proxy for what the operation contributes to a learned model, a modeling choice the paper states explicitly and qualifies by conceding that tasks which do not need the missing interactions lose nothing in practice.

Editorial extensions

If this is right

  • A practitioner comparing GTP or MTP with CGTP by wall-clock time is comparing different operations; the speed advantage is real only for tasks whose needed interaction paths survive the TPO's selection rules.
  • The spherical-grid implementation of GTP is a drop-in replacement for the Fourier-basis implementation with identical asymptotics and about 30% faster training of the MACE interatomic potential.
  • Networks that must represent chiral or antisymmetric information cannot be built on GTP alone, because paths such as the cross product are structurally absent rather than merely expensive.
  • Asymptotic order does not predict wall-clock behavior: sparse CGTP has the fewest FLOPs but the slowest wall time, while GTP's higher FLOP count is offset by higher GPU utilization.
  • The commonly cited $O(L^6)$ cost of CGTP overstates the gap by an order of $L$; exploiting Clebsch-Gordan sparsity gives $O(L^5)$, narrowing the asymptotic gap the $O(L^3)$ alternatives advertise.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension of the paper's own logic: on tasks dominated by symmetric couplings GTP and MTP should match CGTP accuracy at lower cost, while on tasks with strong antisymmetric coupling they should lag regardless of width; the paper itself concedes the loss can be irrelevant when the missing interactions are not needed.
  • If the expressivity proxy is accepted, the practical Pareto frontier is nearly degenerate after normalization, which suggests the next real speedup for equivariant layers will come from faster spherical-harmonic transforms or GPU-native sparse CGTP kernels rather than from new bilinear interaction designs.
  • Interactability offers a quick theoretical screen: any proposed tensor product operation can be classified by which irrep triples it permits and how many independent bilinearities it constructs, letting designers check for expressivity loss before writing GPU code.
  • The paper's framing implies that benchmarking claims about tensor products should always be paired with an expressivity measure; otherwise a reported speedup may simply be a report of an operation that computes less.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. This paper analyzes the runtime and expressivity tradeoffs of several equivariant tensor product operations (TPOs) used in E(3)-equivariant neural networks: the Clebsch-Gordan tensor product (CGTP), the Gaunt tensor product (GTP), and the matrix tensor product (MTP). It introduces two measures, expressivity (the dimension of the space of constructible bilinear maps, Definition 2.2) and interactability (which irrep triples can interact, Definition 2.3), and uses them to argue that reported asymptotic speedups of GTP and MTP over sparse CGTP come from reduced expressivity. The paper proposes a simplified grid-based implementation of GTP, which it claims is a drop-in replacement for the original Fourier-based GTP and is about 30% faster in practice, and it presents extensive microbenchmarks on GPU and CPU. It also demonstrates experimentally that GTP cannot classify chiral 3D Tetris pieces because it lacks antisymmetric interactions, while CGTP can. The main theoretical findings are summarized in Table 1: after normalizing runtime by the expressivity measure, most TPOs have O(L^2) asymptotic runtime per unit expressivity, with only a variant using fast spherical transforms (S2FFT) achieving O(L log^2 L).

Significance. If the claims hold, this is a useful and timely contribution to the equivariant neural network literature. The paper provides the first systematic comparison of TPOs along both asymptotic and empirical axes, and its emphasis on the distinction between a true tensor product and a TPO clarifies an important conceptual point. The asymptotic analyses in Appendix E are careful, the expressivity and interactability definitions are explicit and mathematically well posed, and the Tetris experiment gives a clean behavioral demonstration of the interactability claims. The grid-based GTP implementation is simple and seems genuinely faster in practice, and the public code release supports reproducibility. The main significance hinges on whether the expressivity proxy is accepted as the right normalization for the headline 'no true speedup' conclusion; this is the central point that needs strengthening.

major comments (2)
  1. [Definition 2.2; §4 / Table 1] The headline conclusion that 'when normalized for expressivity, most TPOs have the same asymptotics as sparse CGTP' and that GTP/MTP 'do not truly speed up tensor products but rather remove degrees of freedom' relies entirely on Definition 2.2, which measures expressivity as the dimension of the parameterized family of bilinear maps. The paper itself labels this a proxy in Section 2.3, and Section 5.4 concedes that missing interactions may be irrelevant to a task. Moreover, Appendix H and the Tetris experiment (Section 6.2) demonstrate that two TPOs with the same expressivity dimension can have opposite task-level capabilities: for example, a one-path CGTP can realize the pseudoscalar needed for the chiral Tetris task while GTP cannot, even though both have comparable expressivity dimension. As written, the normalized-runtime comparison establishes a statement about a specific parameter-count proxy, not a general statement about practical speed per unit of usable expressivity. Please either scope the conclusion explicitly to this proxy, or provide additional evidence (for example, interactability-weighted counts or a task-based calibration) that supports the stronger practical claim.
  2. [§6.1 / Appendix E.2.2] The paper calls the grid GTP a 'drop-in replacement' for the Fourier GTP, but I could not find any test of numerical output equivalence. The grid construction replaces exact spherical-harmonic transforms by quadrature (Appendix E.2.2), so the two implementations are not bitwise identical and may differ for finite L and for the chosen quadrature rule. Please report the maximum pointwise or normwise deviation between grid GTP and Fourier GTP on random inputs across a range of L, or provide an error bound derived from the quadrature rule. Without this, the 30% speedup claim is not tied to a demonstrated equivalence of the operation that is being sped up.
minor comments (5)
  1. [Section 7 / Abstract / Section 6.1] The abstract and Section 6.1 report a 30% speedup for grid GTP, but the conclusion (Section 7) says it 'perfroms 20% faster in practice'; please reconcile the two numbers.
  2. [Section 2.5] There is a typo in Section 2.5: 'this construction is cheap aaand so we focus' should read 'and'.
  3. [Section 5.2] Section 5.2 contains the placeholder phrase '(some details about the hardware here)', which should be replaced with an actual hardware description or removed.
  4. [Appendix L.2] In Appendix L.2, 'we could not colleft hardware-counters for them' should read 'collect'.
  5. [Table 4 / Appendix E.4] In the caption of Table 4, the phrase 'for output settings' is slightly ambiguous; specifying that the three settings are SISO, SIMO, and MIMO would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: expressivity is a stated modeling definition, runtime analyses are independent, benchmarks are external, and the Tetris experiment provides a behavioral check.

full rationale

The central claims of the paper are built on an explicitly stated proxy, Definition 2.2, which defines expressivity as the dimension of the space of constructible bilinear maps. This is a transparent modeling choice, not a hidden fitted parameter, and the paper does not present it as an empirical prediction. The runtime asymptotics in Table 1 are derived independently in Appendix E from the algebraic structure of each tensor product operation, and the expressivity counts in Appendix F are direct consequences of the same definitions and of known Clebsch-Gordan/Gaunt selection rules. Dividing runtime by expressivity is then arithmetic on these independently obtained quantities. The wall-clock and FLOP benchmarks in Section 5 are external measurements, and the Tetris experiment in Section 6.2 is an independent behavioral test of the interactability claim, showing that GTP's missing antisymmetric path has observable consequences. The paper's self-citations, such as Geiger & Smidt (2022) or Batzner et al. (2022), are contextual references to standard equivariant-network tools and are not load-bearing for the paper's main theoretical or empirical conclusions. The limitation noted in Section 5.4, that reduced expressivity may not matter for tasks that do not need the missing interactions, is an honest qualification rather than a circular rescue. The skeptical concern that dimension is not a perfect measure of task-level capacity is a validity criticism of the proxy, not evidence that the derivation reduces to its own inputs.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central analysis rests on three premises: the expressivity proxy (Section 2.3), the exactness of the spherical grid quadrature (Appendix E.2.2), and standard representation-theory facts. No free parameters are fitted to data, and no new physical or mathematical entities are postulated.

assumptions (3)
  • domain assumption The dimension of the space of constructible bilinear maps is a meaningful proxy for expressivity of a tensor product operation.
    Section 2.3 uses this proxy to rank TPOs; the headline conclusion that faster TPOs trade speed for expressivity depends on this modeling choice.
  • domain assumption The O(L^2)-point spherical grid exactly computes the pointwise product projection for band-limited signals, making grid GTP equivalent to Fourier GTP.
    Appendix E.2.2 assumes the quadrature rule exactly integrates spherical harmonics up to the product bandwidth; if the specific grid used is not exact, grid GTP is an approximation and the 'drop-in replacement' claim needs qualification.
  • standard math Schur's lemma and the Clebsch-Gordan and Gaunt coefficient selection rules as stated in the cited literature.
    Used throughout: equivariant linear layers between irreps are scalar multiples; GTP selection rules are derived from Gaunt coefficients in Appendix K.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Price of Freedom: Exploring Expressivity and Runtime Tradeoffs in Equivariant Tensor Products." pith.science (2026). https://pith.science/paper/6VRR4MV6

@misc{pith2026250613523,
  author       = {Pith},
  title        = {Pith review of: The Price of Freedom: Exploring Expressivity and Runtime Tradeoffs in Equivariant Tensor Products},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6VRR4MV6}},
  note         = {Machine review of arXiv:2506.13523}
}
abstract

$E(3)$-equivariant neural networks have demonstrated success across a wide range of 3D modelling tasks. A fundamental operation in these networks is the tensor product, which interacts two geometric features in an equivariant manner to create new features. Due to the high computational complexity of the tensor product, significant effort has been invested to optimize the runtime of this operation. For example, Luo et al. (2024) recently proposed the Gaunt tensor product (GTP) which promises a significant speedup. In this work, we provide a careful, systematic analysis of a number of tensor product operations. In particular, we emphasize that different tensor products are not performing the same operation. The reported speedups typically come at the cost of expressivity. We introduce measures of expressivity and interactability to characterize these differences. In addition, we realized the original implementation of GTP can be greatly simplified by directly using a spherical grid at no cost in asymptotic runtime. This spherical grid approach is faster on our benchmarks and in actual training of the MACE interatomic potential by 30%. Finally, we provide the first systematic microbenchmarks of the various tensor product operations. We find that the theoretical runtime guarantees can differ wildly from empirical performance, demonstrating the need for careful application-specific benchmarking. Code is available at https://github.com/atomicarchitects/PriceofFreedom.

Figures

Figures reproduced from arXiv: 2506.13523 by the authors.

Figure 1
Figure 1. Overall schematic of an equivariant bilinearity, where the two inputs x and y are passed through linear layers and combined using the tensor product operation T to form z. The output irreps z are passed through a final linear layer. In the case of the CGTP, this would consist of elementwise multiplication of x and y followed by contraction with the Clebsch-Gordan coefficients to form output irreps z . Each irrep of … view at source ↗
Figure 2
Figure 2. Schematic of GTP. We interpret input irreps as scalar SH coefficients to create spherical signals. We then take pointwise products of the two signals to create a new signal which we de￾compose back into scalar SH coefficients. on S 2 gives us a new function fx · fy, also on S 2 . Then, converting back to irreps gives us the Gaunt tensor product: x ⊗GTP y = FromSphere(fx · fy) (3) Selection rules for GTP can be deriv… view at source ↗
Figure 3
Figure 3. Schematic of the process in taking a matrix tensor prod￾uct. We embed input irreps into a tensor product rep. We then interact using matrix multiplication before decomposing the result￾ing tensor product rep back into a direct sum of irreps. We also analyze another interaction introduced in the new e3x framework in the FusedTensor class (Unke & Maennel, 2024; Maennel et al., 2024). The key idea is that a tensor prod… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Top: Analysis of tensor products compute scaling on a RTX A5500 GPU: Total GFLOPs (Left), Total walltime (Middle), and Average throughput in GFLOPs/s (Right). Bottom: Analysis of tensor products compute scaling per path on RTX A5500 GPU: (Left) Total Walltime / Express…
Figure 5
Figure 5. Figure 5: Training the MACE interatomic force field with the orig￾inal GTP (Luo et al., 2024) and our faster implementation GTP (Grid) for the same number of GPU hours on the 3BPA and revised MD17 datasets. 6.2. Classifying 3D Tetris Pieces We consider a simple task of classifyi…
Figure 7
Figure 7. Figure 7: Analysis of SISO, SIMO and MIMO ( [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Analysis of SISO, SIMO and MIMO ( [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: Analysis of SISO, SIMO and MIMO ( [PITH_FULL_IMAGE:figures/full_fig_p026_9.png]
Figure 10
Figure 10. Figure 10: Analysis of SISO, SIMO and MIMO ( [PITH_FULL_IMAGE:figures/full_fig_p027_10.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Reconstructing local environments from concise atomistic representations

    physics.comp-ph 2026-07 conditional novelty 6.0 of 10

    Atomic environments can be recovered from what amounts to dozens of rotation-invariant numbers, and the same inversion reveals new pairs of distinct geometries that the descriptors cannot tell apart.

Reference graph

Works this paper leans on

50 extracted references · 28 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    and Xia, T

    Aykent, S. and Xia, T. Gotennet: Rethinking efficient 3d equivariant graph neural networks. In The Thirteenth International Conference on Learning Representations

    work page

  3. [3]

    and Isard, M

    Barham, P. and Isard, M. Machine learning systems are stuck in a rut. In Proceedings of the Workshop on Hot Topics in Operating Systems, HotOS '19, pp.\ 177–183, New York, NY, USA, 2019. Association for Computing Machinery. ISBN 9781450367271. doi:10.1145/3317550.3321441. URL https://doi.org/10.1145/3317550.3321441

    arXiv 2019

  4. [4]

    P., Simm, G

    Batatia, I., Kovacs, D. P., Simm, G. N. C., Ortner, C., and Csanyi, G. MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields . In Oh, A. H., Agarwal, A., Belgrave, D., and Cho, K. (eds.), Advances in Neural Information Processing Systems, 2022. URL https://openreview.net/forum?id=YPpSngE-ZU

    work page 2022

  5. [5]

    P., Kornbluth, M., Molinari, N., Smidt, T

    Batzner, S., Musaelian, A., Sun, L., Geiger, M., Mailoa, J. P., Kornbluth, M., Molinari, N., Smidt, T. E., and Kozinsky, B. E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials . 13, May 2022. URL https://doi.org/10.1038/s41467-022-29939-5

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

  6. [6]

    Beentjes, C. H. Quadrature on a spherical surface. Working note available on the website http://people. maths. ox. ac. uk/beentjes/Essays, 2015

    work page 2015

  7. [7]

    An efficient sparse kernel generator for o(3)-equivariant deep networks, 2025

    Bharadwaj, V., Glover, A., Buluc, A., and Demmel, J. An efficient sparse kernel generator for o(3)-equivariant deep networks, 2025. URL https://arxiv.org/abs/2501.13986

    arXiv 2025

  8. [8]

    J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., Vander P las, J., Wanderman- M ilne, S., and Zhang, Q

    Bradbury, J., Frostig, R., Hawkins, P., Johnson, M. J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., Vander P las, J., Wanderman- M ilne, S., and Zhang, Q. JAX : composable transformations of P ython+ N um P y programs, 2018. URL http://github.com/jax-ml/jax

    2018

Show all 50 references
  1. [9]

    Does equivariance matter at scale?, 2024

    Brehmer, J., Behrends, S., de Haan, P., and Cohen, T. Does equivariance matter at scale?, 2024. URL https://arxiv.org/abs/2410.23179

    2024 arXiv

  2. [10]

    Are high-degree representations really unnecessary in equivariant graph neural networks? In The Thirty-eighth Annual Conference on Neural Information Processing Systems

    Cen, J., Li, A., Lin, N., Ren, Y., Wang, Z., and Huang, W. Are high-degree representations really unnecessary in equivariant graph neural networks? In The Thirty-eighth Annual Conference on Neural Information Processing Systems

  3. [11]

    Christensen, A. S. and von Lilienfeld, O. A. On the role of gradients for machine learning of molecular energies and forces, 2020. URL https://arxiv.org/abs/2007.09593

    2020 arXiv

  4. [12]

    G., Mavor-Parker, A

    Cobb, O., Wallis, C. G., Mavor-Parker, A. N., Marignier, A., Price, M. A., d'Avezac, M., and McEwen, J. Efficient generalized spherical cnns. In International Conference on Learning Representations

  5. [13]

    S., Dresselhaus, G., and Jorio, A

    Dresselhaus, M. S., Dresselhaus, G., and Jorio, A. Group theory: application to the physics of condensed matter. Springer Science & Business Media, 2007

    2007

  6. [14]

    A., Chou, H.-Y., Stadler, M., Hoehnerbach, M., Wang, T., Lin, D., Kucukbenli, E., Sprueill, H

    Firoz, J., Pellegrini, F., Geiger, M., Hsu, D., Bilbrey, J. A., Chou, H.-Y., Stadler, M., Hoehnerbach, M., Wang, T., Lin, D., Kucukbenli, E., Sprueill, H. W., Batatia, I., Xantheas, S. S., Lee, M., Mundy, C., Csanyi, G., Smith, J. S., Sadayappan, P., and Choudhury, S. Optimizi...

    2025 arXiv

  7. [15]

    T., and Muller, K

    Frank, T., Unke, O. T., and Muller, K. R. So3krates: Equivariant attention for interactions on arbitrary length-scales in molecular systems. In Advances in Neural Information Processing Systems, 2022

    2022

  8. [16]

    C., Soklaski, R., Axelrod, S., Samsi, S., Gómez-Bombarelli, R., Coley, C

    Frey, N. C., Soklaski, R., Axelrod, S., Samsi, S., Gómez-Bombarelli, R., Coley, C. W., and Gadepally, V. Neural scaling of deep chemical models. Nature Machine Intelligence, 5 0 (11): 0 1297--1305, October 2023

    2023

  9. [17]

    A recipe for charge density prediction, 2024

    Fu, X., Rosen, A., Bystrom, K., Wang, R., Musaelian, A., Kozinsky, B., Smidt, T., and Jaakkola, T. A recipe for charge density prediction, 2024

    2024

  10. [18]

    Gaunt, J. A. Iv. the triplets of helium. Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character, 228 0 (659-669): 0 151--196, 1929

    1929

  11. [19]

    and Smidt, T

    Geiger, M. and Smidt, T. e3nn: Euclidean neural networks. arXiv preprint arXiv:2207.09453, 2022

    2022 arXiv

  12. [20]

    M., Rockmore, D

    Healy, D. M., Rockmore, D. N., Kostelec, P. J., and Moore, S. FFTs for the 2-sphere-improvements and variations . Journal of Fourier analysis and applications, 9: 0 341--385, 2003

    2003

  13. [21]

    G., Vignac, C., and Welling, M

    Hoogeboom, E., Satorras, V. G., Vignac, C., and Welling, M. Equivariant diffusion for molecule generation in 3d, 2022

    2022

  14. [22]

    Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Z \' dek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., Back, T., Peter...

    2021

  15. [23]

    N-body networks: a covariant hierarchical neural network architecture for learning atomic potentials, 2018

    Kondor, R. N-body networks: a covariant hierarchical neural network architecture for learning atomic potentials, 2018. URL https://arxiv.org/abs/1803.01588

    2018 arXiv

  16. [24]

    Clebsch-gordan nets: a fully fourier space spherical convolutional neural network

    Kondor, R., Lin, Z., and Trivedi, S. Clebsch-gordan nets: a fully fourier space spherical convolutional neural network. Advances in Neural Information Processing Systems, 31, 2018

    2018

  17. [25]

    P., Oord, C

    Kov \'a cs, D. P., Oord, C. v. d., Kucera, J., Allen, A. E. A., Cole, D. J., Ortner, C., and Cs \'a nyi, G. Linear atomic cluster expansion force fields for organic molecules: Beyond rmse. Journal of Chemical Theory and Computation, 17 0 (12): 0 7696--7711, Dec 2021. ISSN 1549...

    2021 doi

  18. [26]

    Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive fourier neural operators

    Kurth, T., Subramanian, S., Harrington, P., Pathak, J., Mardani, M., Hall, D., Miele, A., Kashinath, K., and Anandkumar, A. Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive fourier neural operators. In Proceedings of the Platform for Advanced...

    2023

  19. [27]

    Lebedev, V. I. Quadratures on a sphere. USSR Computational Mathematics and Mathematical Physics, 16 0 (2): 0 10--24, 1976

    1976

  20. [28]

    H., Yadollahpour, P., Watkins, A., Frey, N

    Lee, J. H., Yadollahpour, P., Watkins, A., Frey, N. C., Leaver-Fay, A., Ra, S., Cho, K., Gligorijevic, V., Regev, A., and Bonneau, R. Equifold: Protein structure prediction with a novel coarse-grained structure representation. bioRxiv, 2022. doi:10.1101/2022.10.07.511322. URL ...

    2022 doi

  21. [29]

    and Smidt, T

    Liao, Y.-L. and Smidt, T. Equiformer: Equivariant graph attention transformer for 3d atomistic graphs. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=KwmPfARgOTD

    2023

  22. [30]

    Luo, S., Chen, T., and Krishnapriyan, A. S. Enabling efficient equivariant operations in the fourier basis via gaunt tensor products. arXiv preprint arXiv:2401.10216, 2024

    2024 arXiv

  23. [31]

    T., and M \"u ller, K.-R

    Maennel, H., Unke, O. T., and M \"u ller, K.-R. Complete and efficient covariants for 3d point configurations with application to learning molecular quantum properties. arXiv preprint arXiv:2409.02730, 2024

    2024 arXiv

  24. [32]

    McLaren, A. D. Optimal numerical integration on a sphere. Mathematics of Computation, 17 0 (84): 0 361--383, 1963

    1963

  25. [33]

    J., Kornbluth, M., and Kozinsky, B

    Musaelian, A., Batzner, S., Johansson, A., Sun, L., Owen, C. J., Kornbluth, M., and Kozinsky, B. Learning local equivariant representations for large-scale atomistic dynamics. Nature Communications, 14 0 (1): 0 579, 2023

    2023

  26. [34]

    cuequivariance, 2024

    NVIDIA. cuequivariance, 2024. URL https://docs.nvidia.com/cuda/cuequivariance/index.html

    2024

  27. [35]

    J., Torrisi, S

    Owen, C. J., Torrisi, S. B., Xie, Y., Batzner, S., Bystrom, K., Coulter, J., Musaelian, A., Sun, L., and Kozinsky, B. Complexity of many-body interactions in transition metals via machine-learned force fields from the TM23 data set. Npj Comput. Mater., 10 0 (1), May 2024

    2024

  28. [36]

    and Zitnick, C

    Passaro, S. and Zitnick, C. L. Reducing so (3) convolutions to so (2) for efficient equivariant gnns. In International Conference on Machine Learning, pp.\ 27420--27438. PMLR, 2023

    2023

  29. [37]

    A., Tecot, L., Geiger, M., and Smidt, T

    Rackers, J. A., Tecot, L., Geiger, M., and Smidt, T. E. A recipe for cracking the quantum scaling limit with machine learned electron densities. Mach. Learn.: Sci. Technol., 4 0 (1): 0 015027, February 2023

    2023

  30. [38]

    Xla : Compiling machine learning for peak performance, 2020

    Sabne, A. Xla : Compiling machine learning for peak performance, 2020

    2020

  31. [39]

    High-rank irreducible cartesian tensor decomposition and bases of equivariant spaces

    Shao, S., Li, Y., Lin, Z., and Cui, Q. High-rank irreducible cartesian tensor decomposition and bases of equivariant spaces. arXiv preprint arXiv:2412.18263, 2024

    2024 arXiv

  32. [40]

    and Liang, R

    Snider, D. and Liang, R. Operator fusion in xla: Analysis and evaluation, 2023. URL https://arxiv.org/abs/2301.13062

    2023 arXiv

  33. [41]

    W., Descoteaux, M

    Tan, C. W., Descoteaux, M. L., Kotak, M., de Miranda Nascimento, G., Kavanagh, S. R., Zichi, L., Wang, M., Saluja, A., Hu, Y. R., Smidt, T., Johansson, A., Witt, W. C., Kozinsky, B., and Musaelian, A. High-performance training and inference for deep equivariant interatomic pot...

    2025 arXiv

  34. [42]

    Why the compute throughputs value is different from the actual performance (peak performance)?, 2023

    TensorCoreWeight. Why the compute throughputs value is different from the actual performance (peak performance)?, 2023. URL https://forums.developer.nvidia.com/t/why-the-compute-throughputs-value-is-different-from-the-actual-performance-peak-performance/227563/5

    2023

  35. [43]

    Tensor field networks: Rotation- and translation-equivariant neural networks for 3d point clouds, 2018

    Thomas, N., Smidt, T., Kearnes, S., Yang, L., Li, L., Kohlhoff, K., and Riley, P. Tensor field networks: Rotation- and translation-equivariant neural networks for 3d point clouds, 2018. URL https://arxiv.org/abs/1802.08219

    2018 arXiv

  36. [44]

    Unke, O. T. and Maennel, H. E3x: E(3) -equivariant deep learning made easy. arXiv preprint arXiv:2401.07595, 2024

    2024 arXiv

  37. [45]

    A., Moskalev, A

    Varshalovich, D. A., Moskalev, A. N., and Khersonskii, V. K. Quantum theory of angular momentum. World Scientific, 1988

    1988

  38. [46]

    3d steerable cnns: Learning rotationally equivariant features in volumetric data, 2018

    Weiler, M., Geiger, M., Welling, M., Boomsma, W., and Cohen, T. 3d steerable cnns: Learning rotationally equivariant features in volumetric data, 2018. URL https://arxiv.org/abs/1807.02547

    2018 arXiv

  39. [47]

    Fast and accurate spherical harmonics products

    Xin, H., Zhou, Z., An, D., Yan, L.-Q., Xu, K., Hu, S.-M., and Yau, S.-T. Fast and accurate spherical harmonics products. ACM Trans. Graph., 40 0 (6): 0 280--1, 2021

    2021

  40. [48]

    Hierarchical roofline analysis: How to collect data using performance tools on intel cpus and nvidia gpus, 2020

    Yang, C. Hierarchical roofline analysis: How to collect data using performance tools on intel cpus and nvidia gpus, 2020. URL https://arxiv.org/abs/2009.02449

    2020 arXiv

  41. [49]

    Hierarchical roofline performance analysis for deep learning applications, 2020

    Yang, C., Wang, Y., Farrell, S., Kurth, T., and Williams, S. Hierarchical roofline performance analysis for deep learning applications, 2020. URL https://arxiv.org/abs/2009.05257

    2020 arXiv

  42. [50]

    Higher-rank irreducible cartesian tensors for equivariant message passing

    Zaverkin, V., Alesiani, F., Maruyama, T., Errica, F., Christiansen, H., Takamoto, M., Weber, N., and Niepert, M. Higher-rank irreducible cartesian tensors for equivariant message passing. In The Thirty-eighth Annual Conference on Neural Information Processing Systems

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.