Pith. sign in

REVIEW 4 major objections 5 minor 49 references

DEQuify your force field: More efficient simulations using deep equilibrium models

T0 review · 4 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 By recasting a state-of-the-art equivariant force-field network as a deep equilibrium model and warm-starting its fixed-point solver from the previous simulation step, the paper reports 10-20% gains in both accuracy and speed on the MD17…

desk verdict First real DEQ-for-force-fields paper with a genuinely useful fixed-point reuse trick; the 10-20% claim is too broad and the OC20 numbers don't back it up, but it deserves a serious referee. read the letter →

arxiv 2509.08734 v1 pith:BS5F5G54 submitted 2025-09-10 cs.LG cs.AI

Andreas Burger , Luca Thiede , Alán Aspuru-Guzik , Nandita Vijaykumar This is my paper · ORCID
classification cs.LGcs.AI
keywords deepequilibriummodelsmachinelearningforcefieldsmoleculardynamicsequivariantgraphneuralnetworksfixed-pointreuseimplicittemporalcontinuityEquiformerV2
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

Machine-learning force fields currently treat every molecular dynamics step as an independent prediction, ignoring that a simulation is continuous and consecutive states are nearly identical. The paper argues that this temporal continuity is usable prior information and shows how to exploit it: recast the state-of-the-art EquiformerV2 architecture as a deep equilibrium model (DEQ), in which a few weight-tied layers are iterated to a fixed point instead of being stacked into a deep explicit network. At inference, the fixed-point solver is warm-started from the previous time step's fixed point, recycling intermediate features rather than recomputing them. The authors report improving both accuracy and speed by 10-20% compared to the non-DEQ base model on MD17, MD22, and OC20 200k, with memory cost no longer growing with effective depth. If correct, this makes temporal reuse a broadly applicable axis of force-field design, complementary to the symmetry priors that drove earlier progress.

What carries the argument

The central object is the DEQ fixed-point layer: the $L$ Equiformer transformer blocks are replaced by $L_{\text{DEQ}} \ll L$ weight-tied blocks $g_\theta$, and the node features are defined as the equilibrium $h^* = g_\theta\!\left((h^* + \tilde{x})\,\|\tilde{x}\|/\|h^* + \tilde{x}\|\right)$, where $\tilde{x}$ is the encoder embedding injected at every solver step and the norm rescaling prevents feature growth. Four supporting mechanisms carry the argument: Anderson acceleration (chosen over Broyden's method for stability) as the root solver; warm-starting from the previous time step's fixed point during inference; implicit-function-theorem gradients instead of backpropagating through solver iterations, making training memory independent of depth; and the sparse fixed-point-correction loss plus recurrent path dropout for training stability, with the inference tolerance relaxed from $10^{-4}$ to $10^{-1}$ to cut solver steps.

What would settle it

Run DEQuiformer at the training tolerance ($\epsilon = 10^{-4}$) on all MD17/MD22 molecules and log Anderson solver steps per sample across training; if any system or checkpoint fails to converge within the 40-step maximum, the model has no well-defined forward pass and the IFT gradient does not exist. A complementary test: measure the force deviation between warm-started and zero-initialized predictions on trajectories binned by inter-step displacement; if the deviation grows substantially with step size instead of staying below the reported 1%, the temporal-continuity premise that carries the speedup fails precisely where it is needed.

Watch

Extended reading notes

Core claim

Stated on the paper's own terms: the explicit stack of EquiformerV2 transformer blocks can be replaced by an implicit-depth layer whose output is the fixed point $h^* = f_\theta(h^*, x)$ of one or two equivariant blocks, solved with Anderson acceleration. Because MD trajectories are integrated with time steps small relative to the fastest atomic motion, the fixed point at step $t$ is close to the one at step $t+1$, so initializing the solver with the previous fixed point cuts the average number of solver steps from 5-6 to about 3 on Aspirin and from roughly 29 to 11 in OC20 relaxations. This fixed-point reuse, combined with a relaxed test-time solver tolerance of $10^{-1}$, is what turns the DEQ from slow into 10-20% faster than EquiformerV2; without both techniques, no speedup appears. Training uses implicit-function-theorem gradients instead of backpropagation through the solver trajectory, so memory use is independent of effective depth, while a sparse fixed-point-correction loss and recurrent dropout keep training stable. The paper also reports that warm-starting shifts predicted forces by less than 1% relative to zero initialization, so the reuse approximately preserves the Markov property of the dynamics.

Load-bearing premise

The Anderson-accelerated fixed-point solver reliably converges to a well-defined equilibrium for every molecule and at every point in training, so the forward pass exists and the implicit-function-theorem gradient is valid; convergence is verified on Aspirin only, not on the other MD17/MD22 systems or OC20.

Editorial extensions

If this is right

  • On MD17, DEQuiformer is on average 19% faster and 15% more accurate than EquiformerV2, and wins on 5 of 8 molecules; on MD22 it outperforms EquiformerV2 on average and fits the double-walled nanotube system that runs the 8-layer baseline out of memory.
  • On OC20 200k, a two-layer DEQuiformer improves force error to 0.035 eV/Å from EquiformerV2's 0.038 eV/Å at a fraction of the parameter count, and the accuracy gap persists when EquiformerV2 is scaled to 14 layers, consistent with the paper's suggestion that the fixed-point formulation itself carries an inductive bias.
  • In OC20 relaxation simulations, the speedup appears only when fixed-point reuse and the relaxed tolerance ($10^{-1}$) are combined, reducing the number of layer evaluations from roughly 29 to about 11 per step.
  • Warm-starting approximately preserves the Markov property: predicted forces deviate by less than 1% (0.4% on Aspirin, 0.8% on OC20 relaxations) from zero-initialized predictions, so the reuse does not measurably bias the dynamics.
  • Training memory no longer grows with effective depth, which is what allows more expressive models to be trained on larger systems that exceed the explicit-stack baseline's memory budget.

Reading between the lines

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

  • The scheme's structure mirrors SCF warm-starting in ab initio MD, which suggests a testable gradient the paper does not measure: the reuse speedup should scale with trajectory smoothness, helping most at very small integration steps and degrading as the time step grows.
  • The post-training tolerance trade-off could plausibly be made adaptive per time step, spending more solver steps exactly when the configuration has moved far from the previous fixed point, such as during bond breaking or barrier crossing.
  • Because the solver's step count adapts to input difficulty, the DEQ form may generalize better out of distribution, with effective depth growing for unfamiliar geometries; this could be tested directly on elevated-temperature or far-from-equilibrium trajectories.
  • AIMD practice goes beyond simple warm-starting to Pulay mixing and predictor-corrector extrapolation of the density, a toolbox the authors explicitly flag as future work; applying the same ideas to fixed-point initialization could plausibly cut solver steps below the observed 3 per step.
  • The reported energy error on OC20 (0.498 vs 0.392 eV) is worse than the base model even though the force error improves, so the paper's headline 'better accuracy' claim is cleanest read as a force-accuracy claim; a fair reader should check whether the energy gap persists on larger splits.
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

4 major / 5 minor

Summary. The paper proposes DEQuiformer, a deep equilibrium variant of EquiformerV2 for machine-learned force fields. The deep stack of equivariant transformer layers is replaced by a weight-tied block whose output is defined as a fixed point of an Anderson-accelerated solver, trained with implicit-function-theorem gradients. The main idea is to exploit the temporal continuity of molecular dynamics by warm-starting the fixed-point solver from the previous timestep and relaxing the solver tolerance at inference. The authors report improved force accuracy and 10-20% speedups on MD17, MD22, and OC20 200k relative to EquiformerV2, with up to 5x fewer parameters and reduced training memory.

Significance. If the claims hold, the paper identifies a useful and underused inductive bias for ML force fields—temporal continuity—and demonstrates a practical way to obtain deep-model accuracy at low memory cost. The strengths are the clean connection to SCF restart practices, the memory-efficient IFT training, and the competitive MD17/MD22 force errors at lower parameter counts. However, the significance is currently undercut by overclaimed OC20 numbers and by the absence of fixed-point convergence diagnostics outside a single molecule, so the central empirical claims are not yet fully supported.

major comments (4)
  1. [Abstract, §4.1, Table 1b] The abstract's claim that the method 'improve[s] both accuracy and speed by 10%-20% on ... OC20 200k' is not supported by the reported numbers. In Table 1b, DEQuiformer's OC20 energy error is 0.498 eV versus 0.392 eV for EquiformerV2 (27% worse), while the force error improves from 0.038 to 0.035 eV/Å (about 8%). The caption of Table 1b states that DEQuiformer is 'more accurate than EquiformerV2 on forces and energy', which is directly contradicted by the same table. The OC20 relaxation speedup in Table 1a is 12.38 s versus 12.92 s, approximately 4%, not 10-20%. The authors should report force and energy errors separately and restrict the 10-20% claim to the datasets and metrics where it actually holds.
  2. [§4.1, 'Speedup in relaxations', Table 1a] The OC20 speed comparison and accuracy comparison use different baselines. Table 1a compares the one-layer DEQuiformer with the 14-layer EquiformerV2, whereas Table 1b reports accuracy only for the 8-layer EquiformerV2 (force 0.038 eV/Å, energy 0.392 eV); the accuracy of the 14-layer model is not reported. A speed advantage over a 14-layer model does not by itself establish a speed advantage at matched accuracy, because the 14-layer model may not be the same accuracy point as the DEQuiformer. To support the 'faster than EquiformerV2' claim, the authors should report relaxation times and accuracies for the same baseline (e.g., 8-layer) and for the 14-layer model on the same dataset.
  3. [§3.1, Eq. (9), Fig. 4a, Table 3, Appendix A.3] The DEQ output is well-defined only if the Anderson-accelerated solver converges to h* = f_theta(h*, x), and the IFT gradient in Eq. (9) is valid only at such a fixed point. The paper states no contraction or stability condition for f_theta and demonstrates convergence only on Aspirin (Fig. 4a); the tolerance-sensitivity study (Fig. 2a) is also on Aspirin only. During inference the solver stops at the relaxed tolerance epsilon_FP_reuse_test = 1e-1 with a cap of 40 forward steps (Table 3), so for other molecules or training stages the returned iterate may not be the equilibrium. The Markov-property test in Appendix A.3 reports a mean relative force deviation of 0.4% on Aspirin and 0.8% on OC20, but it compares warm-started relaxed predictions with zero-initialized relaxed predictions, not with tightly converged fixed points, so it cannot rule out a systematic bias from early stopping. Please provide per-system fixed-point residuals, the fraction of samples hitting the 40-step cap, and a comparison of warm-started relaxed predictions with tightly converged predictions across all MD17/MD22 molecules and OC20.
  4. [Tables 1, 2, 4; NeurIPS Checklist item 7] The central quantitative claims rest on single training runs without error bars. The checklist acknowledges this, but the headline 10-20% improvements and the '5/8 best' statements in Section 4.1 have no variance estimates, and the OC20 relaxation time difference in Table 1a (12.38 ± 0.33 s vs 12.92 ± 0.26 s) is small relative to the reported deviations. At minimum, the authors should report multiple seeds for the main comparisons, or explicitly state which reported differences are within run-to-run noise.
minor comments (5)
  1. [Throughout] Typos such as 'fa,r' (Abstract), 'Immproved accuracy' (Section 4.1 heading), 'themself' (Introduction), 'variies' (Section 4), 'furhter' (Conclusion), and 'are are common practice' (Section 4.1) should be corrected.
  2. [§3.1 and Table 3] The text says the first inference timestep uses epsilon_train = 1e-4 and later timesteps use epsilon_FP_reuse_test = 1e-1, but Table 3 lists a single inference tolerance of 1e-1; please label the table entry as applying to reuse steps and state the first-step tolerance.
  3. [Figure 2b and Appendix A.3] The solver-step distribution without fixed-point reuse does not sum to 100% in the main figure; the appendix explains this is due to a long tail, but the main text should note the log-scale tail to avoid confusion.
  4. [§4 and Tables 1a, 4] 'Time is measured as the forward pass' should be clarified for Table 1a and Table 4 to state whether reported times include the whole relaxation trajectory, the DEQ solver iterations, and any integration overhead.
  5. [Appendix A.2] Since the method is a modification of a specific EquiformerV2 commit and the TorchDEQ library, please specify the exact commits and versions of all dependencies, and describe any modifications to the solver, to aid reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's central claims are measured benchmark comparisons, not constructed from fitted inputs or self-citations.

full rationale

The paper is an empirical architecture-and-systems contribution: it replaces EquiformerV2's stacked layers with a DEQ fixed-point layer, warm-starts inference from the previous time step, and measures accuracy and speed on public benchmarks (MD17, MD22, OC20 200k). No equation in the paper defines the reported accuracy or speed in terms of a fitted parameter or a self-citation. The DEQ fixed-point equation h* = f_theta(h*,x), the IFT gradient (Eq. 9), and the inference tolerance epsilon_FP_reuse_test = 1e-1 are standard DEQ machinery explicitly borrowed from prior work (Bai et al. and related DEQ references), none of which is authored by the present paper's authors and none of which is fit to the benchmark targets and then renamed as a prediction. The relaxed tolerance is chosen from an Aspirin validation curve (Fig. 2a) and then applied uniformly, but that is ordinary hyperparameter selection and does not statistically force the subsequently measured test errors or wall-clock timings on other molecules. The fixed-point convergence concern raised in review is a correctness and robustness risk, since convergence is only spot-checked on Aspirin, but it is not a circularity: the claims stand or fall on the experiments, not on a derivation that assumes its own conclusion. Self-citation is not load-bearing. Therefore no circular step meets the evidentiary bar.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities, mediators, or conservation laws are introduced. DEQuiformer is a neural architecture, not a postulated phenomenon. The free parameters listed are hyperparameters or tolerances chosen by hand that directly influence the claimed speed-accuracy tradeoff.

free parameters (4)
  • Inference fixed-point tolerance epsilon_FP_reuse_test = 0.1
    Chosen by hand as the sweet spot on the Aspirin validation curve (figure 2a). This threshold directly controls the speed-accuracy tradeoff during inference and is essential for the claimed speedup.
  • Training fixed-point tolerance epsilon_train = 1e-4
    Taken from prior DEQ work [3]. It controls the accuracy of the forward fixed-point solve during training and underpins the validity of IFT gradients.
  • Fixed-point correction loss terms = 3
    Number of intermediate solver states used for the sparse fixed-point correction loss, taken from [3]; affects training stability and final accuracy.
  • DEQ layer count = 1 or 2
    Design choice for the implicit layer count; the authors compare 1- and 2-layer DEQuiformer against 4-14 layer EquiformerV2. This determines parameter count and accuracy.
assumptions (4)
  • domain assumption The DEQ layer has a unique fixed point and Anderson acceleration converges to it.
    The paper relies on the DEQ framework from [5] and does not prove existence or uniqueness for the specific equivariant transformer block; convergence is only verified empirically on Aspirin (figure 4a).
  • standard math The Implicit Function Theorem gradient is applicable, so Eq. (9) gives correct gradients.
    Used in section 3.1, memory-efficient training relies on this identity; the paper assumes differentiability of f_theta and invertibility of the Jacobian at the fixed point.
  • domain assumption Consecutive MD frames are similar enough that warm-starting reduces solver iterations without changing predictions.
    This is the core inductive bias, tested only indirectly via solver-step counts (figure 2b) and the Markov property check (appendix A.3), not guaranteed by the data.
  • ad hoc to paper The weight-tied EquiformerV2 block, with input injection and normalization, defines a valid f_theta for the DEQ.
    Section 3.1 and algorithm 2 introduce the normalization and input injection; the paper assumes this modified layer behaves like a standard DEQ layer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DEQuify your force field: More efficient simulations using deep equilibrium models." pith.science (2026). https://pith.science/paper/BS5F5G54

@misc{pith2026250908734,
  author       = {Pith},
  title        = {Pith review of: DEQuify your force field: More efficient simulations using deep equilibrium models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BS5F5G54}},
  note         = {Machine review of arXiv:2509.08734}
}
abstract

Machine learning force fields show great promise in enabling more accurate molecular dynamics simulations compared to manually derived ones. Much of the progress in recent years was driven by exploiting prior knowledge about physical systems, in particular symmetries under rotation, translation, and reflections. In this paper, we argue that there is another important piece of prior information that, thus fa,r hasn't been explored: Simulating a molecular system is necessarily continuous, and successive states are therefore extremely similar. Our contribution is to show that we can exploit this information by recasting a state-of-the-art equivariant base model as a deep equilibrium model. This allows us to recycle intermediate neural network features from previous time steps, enabling us to improve both accuracy and speed by $10\%-20\%$ on the MD17, MD22, and OC20 200k datasets, compared to the non-DEQ base model. The training is also much more memory efficient, allowing us to train more expressive models on larger systems.

Figures

Figures reproduced from arXiv: 2509.08734 by the authors.

Figure 1
Figure 1. Comparison of the EquiformerV2 and DEQuiformer architectures. While the Equiformer [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Reusing fixed-points and relaxing the solver threshold lead to better inference speed. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Results on MD17/22 and OC20 200k: DEQuiformer is faster and more accurate than [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: DEQuiformer enjoys stable training dynamics, reaching lower train and test error. [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: DEQuiformer enjoys stable training dynamics, reaching lower train and test error than [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Markov property. Initializing from the previous fixed-point, compared to initializing from zero, leads to very small deviation in forces ∆Frel below one percent. This means, initialization from the past fixed point has almost no effect on the accuracy of the prediction…
Figure 7
Figure 7. Figure 7: Error scaling with more epochs and model size. [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Examining DEQuiformers fixed-point behaviour. [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 42 canonical work pages

  1. [1]

    Cormorant: Covariant molecular neural networks.Advances in neural information processing systems, 32, 2019

    Brandon Anderson, Truong Son Hy, and Risi Kondor. Cormorant: Covariant molecular neural networks.Advances in neural information processing systems, 32, 2019. 1

    work page 2019

  2. [2]

    Anderson

    Donald G. Anderson. Iterative procedures for nonlinear integral equations.J. ACM, 12(4):547–560, October 1965. 3.1, A.3

    work page 1965

  3. [3]

    Zico Kolter

    Shaojie Bai, Zhengyang Geng, Yash Savani, and J. Zico Kolter. Deep Equilibrium Optical Flow Estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 620–630, 2022. 3.1, 3.1, 3.1, 3.1, A.1.2, A.3

    work page 2022

  4. [4]

    Trellis networks for sequence modeling.arXiv preprint arXiv:1810.06682, 2018

    Shaojie Bai, J Zico Kolter, and Vladlen Koltun. Trellis networks for sequence modeling.arXiv preprint arXiv:1810.06682, 2018. 3.1

    arXiv 2018

  5. [5]

    Zico Kolter, and Vladlen Koltun

    Shaojie Bai, J. Zico Kolter, and Vladlen Koltun. Deep Equilibrium Models. InAdvances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. 1, 3.1, 3.1, 3.1, 3.1, 3.1, A.3

    work page 2019

  6. [6]

    Stabilizing Equilibrium Models by Jacobian Regularization

    Shaojie Bai, Vladlen Koltun, and Zico Kolter. Stabilizing Equilibrium Models by Jacobian Regularization. InProceedings of the 38th International Conference on Machine Learning, pages 554–565. PMLR, July 2021. 3.1

    work page 2021

  7. [7]

    Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons.Physical review letters, 104(13):136403, 2010

    Albert P Bartók, Mike C Payne, Risi Kondor, and Gábor Csányi. Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons.Physical review letters, 104(13):136403, 2010. 1

    work page 2010

  8. [8]

    Kovacs, Gregor Simm, Christoph Ortner, and Gabor Csanyi

    Ilyes Batatia, David P. Kovacs, Gregor Simm, Christoph Ortner, and Gabor Csanyi. MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields. Advances in Neural Information Processing Systems, 35:11423–11436, December 2022. 1, 2

    work page 2022

Show all 49 references
  1. [9]

    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):245...

    2022

  2. [10]

    Generalized neural-network representation of high- dimensional potential-energy surfaces.Physical review letters, 98(14):146401, 2007

    Jörg Behler and Michele Parrinello. Generalized neural-network representation of high- dimensional potential-energy surfaces.Physical review letters, 98(14):146401, 2007. 1

    2007

  3. [11]

    How attentive are graph attention networks?arXiv preprint arXiv:2105.14491, 2021

    Shaked Brody, Uri Alon, and Eran Yahav. How attentive are graph attention networks?arXiv preprint arXiv:2105.14491, 2021. A.1.1

    2021 arXiv

  4. [12]

    C. G. Broyden. A Class of Methods for Solving Nonlinear Simultaneous Equations.Mathematics of Computation, 19(92):577–593, 1965. 3.1

    1965

  5. [13]

    Deep equilibrium diffusion restoration with parallel sampling

    Jiezhang Cao, Yue Shi, Kai Zhang, Yulun Zhang, Radu Timofte, and Luc Van Gool. Deep equilibrium diffusion restoration with parallel sampling. InCVPR, 2024. 3.1, A.1.2

    2024

  6. [14]

    Unified approach for molecular dynamics and density- functional theory.Physical review letters, 55(22):2471, 1985

    Richard Car and Mark Parrinello. Unified approach for molecular dynamics and density- functional theory.Physical review letters, 55(22):2471, 1985. 1

    1985

  7. [15]

    Unke, Adil Kabylda, Huziel E

    Stefan Chmiela, Valentin Vassilev-Galindo, Oliver T. Unke, Adil Kabylda, Huziel E. Sauceda, Alexandre Tkatchenko, and Klaus-Robert Müller. Accurate global machine learning force fields for molecules with hundreds of atoms.Science Advances, 9(2):eadf0873, 2023. 4

    2023

  8. [16]

    Bartel, and Gerbrand Ceder

    Bowen Deng, Peichen Zhong, KyuJung Jun, Janosh Riebesell, Kevin Han, Christopher J. Bartel, and Gerbrand Ceder. Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling.Nature Machine Intelligence, 5(9):1031–1041, Sep 2023. 1

    2023

  9. [17]

    Molecular dynamics simulations and drug discovery.BMC biology, 9:1–9, 2011

    Jacob D Durrant and J Andrew McCammon. Molecular dynamics simulations and drug discovery.BMC biology, 9:1–9, 2011. 1 11

    2011

  10. [18]

    Jfb: Jacobian-free backpropagation for implicit networks

    Samy Wu Fung, Howard Heaton, Qiuwei Li, Daniel McKenzie, Stanley Osher, and Wotao Yin. Jfb: Jacobian-free backpropagation for implicit networks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 6648–6656, 2022. 3.1, A.1.2

    2022

  11. [19]

    A theoretically grounded application of dropout in recurrent neural networks

    Yarin Gal and Zoubin Ghahramani. A theoretically grounded application of dropout in recurrent neural networks. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016. 3.1

    2016

  12. [20]

    GemNet: Universal Directional Graph Neural Networks for Molecules

    Johannes Gasteiger, Florian Becker, and Stephan Günnemann. GemNet: Universal Directional Graph Neural Networks for Molecules. InAdvances in Neural Information Processing Systems, volume 34, pages 6790–6802. Curran Associates, Inc., 2021. 1

    2021

  13. [21]

    Directional message passing for molecular graphs.arXiv preprint arXiv:2003.03123, 2020

    Johannes Gasteiger, Janek Groß, and Stephan Günnemann. Directional message passing for molecular graphs.arXiv preprint arXiv:2003.03123, 2020. 1, 2

    2003 arXiv

  14. [22]

    Zico Kolter

    Zhengyang Geng and J. Zico Kolter. TorchDEQ: A Library for Deep Equilibrium Models, October 2023. 3.1, A.2, A.3

    2023

  15. [23]

    One-step diffusion distillation via deep equilibrium models

    Zhengyang Geng, Ashwini Pokle, and J Zico Kolter. One-step diffusion distillation via deep equilibrium models. InThirty-seventh Conference on Neural Information Processing Systems,

  16. [24]

    On training implicit models.Advances in Neural Information Processing Systems, 34:24247–24260, 2021

    Zhengyang Geng, Xin-Yu Zhang, Shaojie Bai, Yisen Wang, and Zhouchen Lin. On training implicit models.Advances in Neural Information Processing Systems, 34:24247–24260, 2021. A.1.2

    2021

  17. [25]

    Molecular dynamics simulation for all.Neuron, 99(6):1129–1143, 2018

    Scott A Hollingsworth and Ron O Dror. Molecular dynamics simulation for all.Neuron, 99(6):1129–1143, 2018. 1

    2018

  18. [26]

    Finkler, Stefan Goedecker, and Jörg Behler

    Tsz Wai Ko, Jonas A. Finkler, Stefan Goedecker, and Jörg Behler. A fourth-generation high- dimensional neural network potential with accurate electrostatics including non-local charge transfer.Nature Communications, 12(1):398, Jan 2021. 1

    2021

  19. [27]

    Time-reversible always stable predictor–corrector method for molecular dynamics of polarizable molecules.Journal of computational chemistry, 25(3):335–342, 2004

    Jiˇrí Kolafa. Time-reversible always stable predictor–corrector method for molecular dynamics of polarizable molecules.Journal of computational chemistry, 25(3):335–342, 2004. 1

    2004

  20. [28]

    Long-short-range message-passing: A physics-informed framework to capture non-local interaction for scalable molecular dynamics simulation.arXiv preprint arXiv:2304.13542, 2023

    Yunyang Li, Yusong Wang, Lin Huang, Han Yang, Xinran Wei, Jia Zhang, Tong Wang, Zun Wang, Bin Shao, and Tie-Yan Liu. Long-short-range message-passing: A physics-informed framework to capture non-local interaction for scalable molecular dynamics simulation.arXiv preprint arXiv:...

    2023 arXiv

  21. [29]

    Long-short-range message-passing: A physics-informed framework to capture non-local interaction for scalable molecular dynamics simulation, 2024

    Yunyang Li, Yusong Wang, Lin Huang, Han Yang, Xinran Wei, Jia Zhang, Tong Wang, Zun Wang, Bin Shao, and Tie-Yan Liu. Long-short-range message-passing: A physics-informed framework to capture non-local interaction for scalable molecular dynamics simulation, 2024. 2

    2024

  22. [30]

    Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs, February 2023

    Yi-Lun Liao and Tess Smidt. Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs, February 2023. 1, A.1.1, A.1.1, A.2, A.2

    2023

  23. [31]

    EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representations, March 2024

    Yi-Lun Liao, Brandon Wood, Abhishek Das, and Tess Smidt. EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representations, March 2024. 1, 2, 2, 2.1, A.1.1, A.2, A.3

    2024

  24. [32]

    Force fields for small molecules.Biomolecular simulations: Methods and protocols, pages 21–54, 2019

    Fang-Yu Lin and Alexander D MacKerell. Force fields for small molecules.Biomolecular simulations: Methods and protocols, pages 21–54, 2019. 1

    2019

  25. [33]

    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. 1, 2

    2022

  26. [34]

    Owen, Mordechai Kornbluth, and Boris Kozinsky

    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, February 2023. 1, 2 12

    2023

  27. [35]

    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, volume 202 ofICML’23, pages 27420–27438. JMLR.org, July 2023. 1, 2, 2

    2023

  28. [36]

    David A Pearlman, David A Case, James W Caldwell, Wilson S Ross, Thomas E Cheatham III, Steve DeBolt, David Ferguson, George Seibel, and Peter Kollman. Amber, a package of com- puter programs for applying molecular mechanics, normal mode analysis, molecular dynamics and free e...

    1995

  29. [37]

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

    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,

  30. [38]

    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. 1, 2

    2021

  31. [39]

    Moment tensor potentials: A class of systematically improvable inter- atomic potentials.Multiscale Modeling & Simulation, 14(3):1153–1173, 2016

    Alexander V Shapeev. Moment tensor potentials: A class of systematically improvable inter- atomic potentials.Multiscale Modeling & Simulation, 14(3):1153–1173, 2016. 1

    2016

  32. [40]

    From molecules to materials: Pre-training large generalizable models for atomic property prediction.arXiv preprint arXiv:2310.16802, 2023

    Nima Shoghi, Adeesh Kolluru, John R Kitchin, Zachary W Ulissi, C Lawrence Zitnick, and Brandon M Wood. From molecules to materials: Pre-training large generalizable models for atomic property prediction.arXiv preprint arXiv:2310.16802, 2023. 2

    arXiv

  33. [41]

    Applications of molecular dynamics simulation in protein study.Membranes, 12(9):844, 2022

    Siddharth Sinha, Benjamin Tam, and San Ming Wang. Applications of molecular dynamics simulation in protein study.Membranes, 12(9):844, 2022. 1

    2022

  34. [42]

    Tensor field networks: Rotation-and translation-equivariant neural networks for 3d point clouds.arXiv preprint arXiv:1802.08219, 2018

    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. 2, 2

    2018 arXiv

  35. [43]

    Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials.Journal of Computational Physics, 285:316–330, 2015

    Aidan P Thompson, Laura P Swiler, Christian R Trott, Stephen M Foiles, and Garritt J Tucker. Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials.Journal of Computational Physics, 285:316–330, 2015. 1

    2015

  36. [44]

    Visnet: an equivariant geometry-enhanced graph neural network with vector-scalar interactive message passing for molecules.arXiv preprint arXiv:2210.16518,

    Yusong Wang, Shaoning Li, Xinheng He, Mingyu Li, Zun Wang, Nanning Zheng, Bin Shao, Tie-Yan Liu, and Tong Wang. Visnet: an equivariant geometry-enhanced graph neural network with vector-scalar interactive message passing for molecules.arXiv preprint arXiv:2210.16518,

    arXiv

  37. [45]

    Efficiently incorporating quintuple interactions into geometric deep learning force fields.Advances in Neural Information Processing Systems, 36, 2024

    Zun Wang, Guoqing Liu, Yichi Zhou, Tong Wang, and Bin Shao. Efficiently incorporating quintuple interactions into geometric deep learning force fields.Advances in Neural Information Processing Systems, 36, 2024. 1

    2024

  38. [46]

    Amber: Assisted model building with energy refinement

    Paul K Weiner and Peter A Kollman. Amber: Assisted model building with energy refinement. a general program for modeling molecules and their interactions.Journal of Computational Chemistry, 2(3):287–303, 1981. 1

    1981

  39. [47]

    Lightweight equivariant model for efficient machine learning interatomic potentials, 2024

    Ziduo Yang, Xian Wang, Yifan Li, Qiujie Lv, Calvin Yu-Chian Chen, and Lei Shen. Lightweight equivariant model for efficient machine learning interatomic potentials, 2024. 1b 13 A Appendix A.1 Additional Background A.1.1 EquiformerV2 architecture We provide some further detail ...

    2024

  40. [48]

    Limitations

    from commit b7e7a0d. The DEQ solver is adapted from the TorchDEQ library [22]. For MD17/MD22 each model is trained on a single AMD MI100 GPU with 32GB GPU RAM for 1000 epochs, which takes 12 to 72 hours. For OC20 200k training takes about 40 to 120 hours for six epochs. A.3 Ad...

  41. [49]

    • Depending on the country in which research is conducted, IRB approval (or equivalent) may be required for any human subjects research

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

    2025

Pith tools

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