Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

BoostMD: Accelerating molecular sampling by leveraging ML force field features from previous time-steps

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read BoostMD accelerates machine-learned force-field molecular dynamics by more than eightfold using previous node features.

desk verdict A promising surrogate idea with a real 8x speedup, but the Boltzmann-sampling claim rests on a qualitative figure and an untested architectural simplification. read the letter →

arxiv 2412.18633 v1 pith:ZSAJ7XWA submitted 2024-12-21 physics.chem-ph cond-mat.stat-mechcs.LGphysics.comp-phstat.ML

Lars L. Schaaf , Ilyes Batatia , Christoph Brunken , Thomas D. Barrett , Jules Tilly This is my paper · ORCID
classification physics.chem-phcond-mat.stat-mechcs.LGphysics.comp-phstat.ML
keywords machinelearningforcefieldsmoleculardynamicsaccelerationequivariantmessagepassingsurrogatemodelBoltzmannsamplingfeaturereusemetadynamics
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 proposes BoostMD, a surrogate model that makes machine-learning force-field molecular dynamics faster by not recomputing the expensive internal features at every time step. The full reference force field is evaluated every N steps, and a small, fast BoostMD model predicts the energy change for the intervening steps from the previous node features and the atoms' positional changes. The authors claim this yields more than an eightfold speedup over the reference model, transfers to dipeptides not seen in training, and samples the correct equilibrium (Boltzmann) distribution, including under metadynamics. If true, long-timescale simulations of peptides and other organic molecules would become practical at near-quantum accuracy.

What carries the argument

The central object is the BoostMD equivariant message-passing layer. For each atom it takes the current positions, the reference positions from the last full evaluation of the reference force field, and the reference node features, forming change vectors that are aligned by the optimal rotation between the reference and current neighborhoods (computed with the Kabsch algorithm). Spherical harmonics of the current displacement and of the change vector are combined with the rotated reference features through learnable tensor products to build a message basis, then a MACE-style product basis creates many-body features, and a small readout predicts the per-atom energy change. A multiplicative factor on the displacement ensures zero energy change when nothing has moved. In the reported experiments a single-layer BoostMD model with a five-angstrom receptive field is used, with the reference model evaluated every ten steps, and the rotational reference framing is omitted during training and inference.

What would settle it

Run BoostMD with a larger reference interval N or with a molecular system that rotates substantially between frames, then compare the sampled dihedral free-energy surface to the reference force field; if the free-energy surfaces diverge or the force error grows with the angle of rotation between reference and current frames, the small-rotation assumption fails.

Watch

Extended reading notes

Core claim

The discovery is that the internal per-atom features of an equivariant machine-learning force field change slowly during a molecular dynamics trajectory, so a lightweight surrogate conditioned on the features from a recent reference evaluation can accurately predict energy changes and carry most time steps. Running the reference only every N steps and BoostMD in between yields more than eightfold speedup on a large system with over three thousand atoms, with energy and force errors comparable to the reference model's own error against density functional theory. The surrogate reproduces the reference free-energy surface for an unseen dipeptide and remains stable over ten nanoseconds, demonstrating that the accelerated trajectory still samples the correct equilibrium ensemble.

Load-bearing premise

The reported experiments skip the rotational alignment step that the architecture's equivariance argument requires, so the results depend on molecular rotations between reference updates being small enough that omitting the alignment barely changes the forces and the sampled distribution.

Editorial extensions

If this is right

  • Molecular dynamics with machine-learning force fields can run about eight times faster, bringing microsecond-timescale simulations of organic molecules within practical reach.
  • The acceleration compounds with enhanced sampling: BoostMD carries metadynamics and reproduces the reference free-energy surface for a molecule outside its training set.
  • The boost model is small and local, so it should parallelize across GPUs more easily than the full reference model, easing scale-up to large systems.
  • The surrogate's error is close to the reference model's own error against density functional theory, so the practical accuracy bottleneck becomes the reference force field rather than the acceleration wrapper.
  • Transferability to unseen dipeptides suggests the scheme may generalize beyond its training molecules, though the paper demonstrates this only for dipeptides.

Reading between the lines

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

  • If the omission of rotational reference framing is stressed with a larger reference interval or highly rotatable molecules, the reported sampling accuracy may degrade; the paper does not test this.
  • The method's energy is not exactly conserved at the reference-update steps, and the authors rely on thermostats to maintain correct sampling; combining BoostMD with multiple-time-step integrators or adaptive reference intervals could control drift.
  • Because the surrogate is conditioned on any machine-learning force field's node features, the same BoostMD recipe could likely be applied to other equivariant force fields, not just the MACE-based reference.
  • The eightfold speedup may be a conservative figure: the parallel producer/consumer evaluation described but not implemented would remove the serial bottleneck and could deliver the full model speedup.
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 BoostMD, a surrogate architecture that reuses node features of a reference machine-learned force field (MACE-OFF23-M) computed at a previous MD time step to predict energy and force changes over a short interval, enabling the reference model to be evaluated only every N steps. The authors report speedups up to about 8.6x on a dipeptide dataset, transferability to an unseen dipeptide, and a qualitative Ramachandran free-energy comparison from metadynamics that they interpret as accurate Boltzmann sampling. The manuscript also discusses architectural design choices, including a Kabsch-based rotational reference frame, readout strategies, and conservation properties.

Significance. If the central sampling claim is established, BoostMD would be a practically valuable method: it could accelerate high-accuracy MLFF molecular dynamics by roughly an order of magnitude with only small bias in the sampled equilibrium ensemble. The paper identifies a promising direction—reusing expressive reference features across time steps—and includes concrete speedup/error measurements, a comparison against a similarly fast MACE baseline, and a transferability test. However, the current evidence does not yet support the full claim: the experiments were run with an architectural variant that omits the rotational reference framing that the paper itself argues is necessary for equivariance and conservation, and the only sampling evidence is a visual free-energy comparison without quantitative error bars or convergence diagnostics. These are load-bearing gaps that can likely be addressed within the manuscript's scope.

major comments (4)
  1. [§5.1 and Appendix C.1] Section 5.1 and Appendix A.1 introduce a per-atom Kabsch optimal rotation Rbar_i as necessary for rotational equivariance and for conservation of momentum/angular momentum between BoostMD steps, but Appendix C.1 states: "All experimental results do not use the rotational reference framing during training or inference, due to the associated computational cost." Consequently, every reported result—Table 1 RMSE/speedups, the 10 ns stability statement, and the Figure 3 free-energy comparison—was produced by a model that is not rotationally equivariant between the stored reference features and evolving coordinates. The §5.2 claim that the method is "energy, momentum and angular momentum conserving" between boost steps is therefore not supported for the tested model. Please rerun the core experiments with the symmetry-preserving framing, or provide quantitative diagnostics (e.g., angular-momentum drift, energy drift, net torque) demonstrating that the omission is benign over the N=10 interval used in the sampling experiment.
  2. [§6.2 / Figure 3] The sampling evidence for the central Boltzmann-distribution claim consists solely of a qualitative comparison of Ramachandran free-energy surfaces from a single 5 ns metadynamics run per method (Figure 3). There is no quantitative distribution discrepancy metric, no error bars, no replicate runs, and no convergence analysis. Please add a quantitative comparison (e.g., Kullback-Leibler or Jensen-Shannon divergence between the two free-energy histograms, with bootstrap or block uncertainty estimates), report the metadynamics bias parameters for both runs, and state whether the runs are statistically converged. Without these, the statement that BoostMD "accurately samples the ground-truth Boltzmann distribution" is not demonstrated.
  3. [§5.2] The conservation/sampling argument relies on the cited result that thermostats can ensure correct sampling despite energy-conservation violation (Ref. [38]), but that result is not directly established for this specific state-dependent, piecewise potential whose energy surface is refreshed every N steps and whose forces are not gradients of a fixed global potential. The paper should either provide a theoretical argument applicable to this setting or add empirical diagnostics, such as comparing sampled distributions for different boost intervals N, measuring energy/angular-momentum jumps at refresh events, and checking stationarity and ergodicity of the trajectory. This is load-bearing for the claim that the sampled distribution is Boltzmann.
  4. [Table 1 and §1] The headline "more than eight times faster" claim in Section 1 and the Abstract is only supported by the Lref=0 variants in Table 1 (speedups 8.4 and 8.6); the Lref=1 variant, which has better force RMSE, achieves only 2.3x. Since the scalar-feature variant is the one that delivers the advertised speedup, the paper should explicitly qualify the speedup claim with this accuracy/speed trade-off and, ideally, show a broader trade-off curve rather than presenting the 8x figure as the general BoostMD performance.
minor comments (5)
  1. [Appendix C.1] The statement that all experimental results omit the rotational reference framing is currently buried in an appendix; this is a major limitation of the experiments as presented and should be stated prominently in the main text, alongside a discussion of its implications.
  2. [§6.2 / Figure 3] The paper calls the MACE-OFF model the "ground truth" for the Boltzmann distribution, but it is itself an approximate MLFF, not the true quantum-mechanical distribution; this should be stated explicitly to avoid overclaiming.
  3. [§5.2] The effective speedup formula appears as "1 + (N − 1)/s n" and is garbled; please define all symbols (n or N) and present the formula cleanly.
  4. [Throughout] There are several typos: "transitionally" should be "translationally" (§5.1), "comapred" should be "compared" (Appendix C.1), "prallelisation" should be "parallelisation" (Appendix C.1), and "BostMD" in the Figure 1 caption should be "BoostMD".
  5. [§6.1] The sentence "MACE-OFF23-M achieves an error of 0.85 meV/atom compared to DFT, with BoostMD models showing similar or lower accuracy relative to MACE-OFF23-M" is confusing: "lower accuracy" is ambiguous; please clarify whether BoostMD errors are lower or higher than the reference model's DFT error.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: BoostMD is a supervised surrogate whose sampling benchmark tests fidelity to the reference MLFF, not a construction-level reduction.

full rationale

The paper's chain of claims is empirical rather than derivational: BoostMD is trained as a surrogate to reproduce the energies and forces of MACE-OFF23-M, and its reported speedups, error metrics, and sampling results are measurements of that surrogate's fidelity and efficiency. Nothing in the architecture makes the surrogate's outputs equal to the reference model's outputs by construction; the surrogate has fewer parameters, a different receptive field, and is evaluated at configurations and on molecules not seen in training, so the observed agreement in Figure 3 is a nontrivial generalization result. The 'ground-truth Boltzmann distribution' is the reference MACE-OFF23-M model's own equilibrium distribution, so the validation demonstrates surrogate fidelity rather than independent physical accuracy; this is a benchmark limitation, not a circular derivation. Self-citations to MACE and MACE-OFF23 are design choices and reference-model choices, not load-bearing proofs of BoostMD's acceleration or sampling claims. The rotational-framing inconsistency between Section 5.1 and Appendix C.1 is a correctness and validation gap (the tested model is not rotationally equivariant as claimed), but it does not reduce any claimed result to its inputs. Therefore no circularity step meets the required standard of exhibiting a specific reduction by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on hand-chosen hyperparameters N=10 and a 5 Å cutoff, on the untested sufficiency of previous-step features, on thermostat-mediated sampling under non-conservative forces, and on using the reference model itself as the ground truth for validation. The most fragile item is the unquantified omission of the rotational reference framing: the architecture section motivates it as necessary for equivariance, but all experiments skip it.

free parameters (3)
  • Boost interval N = 10
    The reference model is evaluated every 10 steps in the sampling experiment (Figure 3 caption); this hyperparameter trades speed against accuracy but is not swept, and the speedup claims depend on it.
  • Cutoff radius = 5 Å
    Single-layer BoostMD uses a 5 Å receptive field, restricting locality; this is chosen by design for speed and parallelization, and it determines which atoms are included in the local energy prediction.
  • Model size (channels, max spherical harmonic degree, correlation order) = 125 channels, l_max=2, correlation order=2 for XS; 256 channels for XS+ MACE baseline
    Architecture sizes are hand-chosen to make BoostMD fast; the XS+ MACE comparison gives the baseline additional channels to compensate for its speed, so the comparison depends on these size choices.
assumptions (5)
  • domain assumption Node features from a previous reference step remain informative enough to predict energy changes for the next N steps.
    Core enabling assumption of BoostMD, presented in Section 2 and Figure 1; no formal bound or degradation analysis is provided.
  • domain assumption A thermostat ensures correct canonical sampling even when the surrogate energy is not exactly conservative.
    Invoked in Section 5.2 via Jones and Leimkuhler; the empirical claim that energy non-conservation is minimal relative to thermostat noise is stated but not quantified.
  • domain assumption MACE-OFF23-M is an adequate ground truth for the Boltzmann distribution of the test dipeptide.
    The "ground-truth" free energy surface in Figure 3 is generated by the same reference model that produced the BoostMD training labels, so physical correctness of the sampled distribution is assumed rather than tested against experiment or DFT.
  • ad hoc to paper Omitting the Kabsch rotational reference framing has negligible effect over N-step intervals.
    Appendix C.1 states experiments do not use rotational reference framing; the paper does not analyze the resulting equivariance error. This is specific to this paper's experimental setup.
  • standard math Equivariant tensor product algebra from MACE and e3nn is correct and implemented as described.
    The architecture relies on Wigner-D matrices, Clebsch-Gordan coefficients, and the MACE product basis from prior work; these are treated as background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BoostMD: Accelerating molecular sampling by leveraging ML force field features from previous time-steps." pith.science (2026). https://pith.science/paper/ZSAJ7XWA

@misc{pith2026241218633,
  author       = {Pith},
  title        = {Pith review of: BoostMD: Accelerating molecular sampling by leveraging ML force field features from previous time-steps},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZSAJ7XWA}},
  note         = {Machine review of arXiv:2412.18633}
}
abstract

Simulating atomic-scale processes, such as protein dynamics and catalytic reactions, is crucial for advancements in biology, chemistry, and materials science. Machine learning force fields (MLFFs) have emerged as powerful tools that achieve near quantum mechanical accuracy, with promising generalization capabilities. However, their practical use is often limited by long inference times compared to classical force fields, especially when running extensive molecular dynamics (MD) simulations required for many biological applications. In this study, we introduce BoostMD, a surrogate model architecture designed to accelerate MD simulations. BoostMD leverages node features computed at previous time steps to predict energies and forces based on positional changes. This approach reduces the complexity of the learning task, allowing BoostMD to be both smaller and significantly faster than conventional MLFFs. During simulations, the computationally intensive reference MLFF is evaluated only every $N$ steps, while the lightweight BoostMD model handles the intermediate steps at a fraction of the computational cost. Our experiments demonstrate that BoostMD achieves an eight-fold speedup compared to the reference model and generalizes to unseen dipeptides. Furthermore, we find that BoostMD accurately samples the ground-truth Boltzmann distribution when running molecular dynamics. By combining efficient feature reuse with a streamlined architecture, BoostMD offers a robust solution for conducting large-scale, long-timescale molecular simulations, making high-accuracy ML-driven modeling more accessible and practical.

Figures

Figures reproduced from arXiv: 2412.18633 by the authors.

Figure 1
Figure 1. Properties of hidden features and the BostMD architecture. (a) Showing the node features of a foundation model for organic molecules (MACE-OFF23 [1]) as a function of MD steps. After 150 fs the molecule is artificially split in two. The hidden features oscillate during MD and only significantly change under this sever configurational change. (b) Proposed method of using node features from previous time steps to pred… view at source ↗
Figure 2
Figure 2. BoostMD architecture and reference framing Showing the steps of BoostMD models, highlighting the reference framing step. The reference node features and reference positions are transformed to make BoostMD translationally and rotationally equivariant between steps as detailed in equation 2. The figures show the receptive field of an atom i with neighbors j, showing both the current (blue) and reference (red) position… view at source ↗
Figure 3
Figure 3. Free energy surface of unseen dipeptide. Comparison of the samples obtained by running met-dynamics using the ground truth MACE-OFF model and BoostMD. The free en￾ergy of the Ramachandran plot, is directly related to the marginalized Boltzmann distribution exp [−F(ϕ, ψ)/kBT], where ϕ, ψ are the dihedral angles marked in rendered molecule (left) . Dur￾ing BoostMD, the reference model is evaluated every 10 steps. Both… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Node feature properties Showing the fluctuations of the reference node features as a function of simulation time (a) and the change in node features as a function of change in position inside the atomic environment. total receptive field. A single layer reduces the rec…

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. Machine Learning Interatomic Potentials: library for efficient training, model development and simulation of molecular systems

    physics.chem-ph 2025-05 conditional novelty 6.0 of 10

    InstaDeep's mlip library ports MACE, NequIP, and ViSNet to JAX with a JAX-MD backend, ships SPICE2-trained organics models, reports faster MD steps than its own Torch routes, and proposes a faster gated MACE variant i...

Reference graph

Works this paper leans on

45 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [23]

    & Kozinsky, B

    Fu, X., Musaelian, A., Johansson, A., Jaakkola, T. & Kozinsky, B. Learning Interatomic Potentials at Multiple Scales. http://arxiv.org/abs/2310.13756 (Oct. 2023)

    arXiv 2023

  2. [38]

    & Leimkuhler, B

    Jones, A. & Leimkuhler, B. Adaptive stochastic methods for sampling driven molecular systems. The Journal of Chemical Physics 135, 084125. ISSN : 0021-9606, 1089-7690. https://pubs. aip.org/jcp/article/135/8/084125/72033/Adaptive-stochastic-methods-for- sampling-driven (2024) (Aug. 28, 2011)

    work page 2024

  3. [1]

    Kovács, D. P.et al. MACE-OFF23: Transferable Machine Learning Force Fields for Organic Molecules Dec. 29, 2023. arXiv: 2312.15211[physics]. http://arxiv.org/abs/2312. 15211 (2024)

    arXiv 2024

  4. [2]

    & Parrinello, M

    Behler, J. & Parrinello, M. Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces. Phys. Rev. Lett. 98, 146401. ISSN : 0031-9007, 1079-7114. https: //link.aps.org/doi/10.1103/PhysRevLett.98.146401 (2024) (Apr. 2, 2007)

    doi:10.1103/physrevlett.98.146401 2024

  5. [3]

    Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. nature 596, 583–589 (2021)

    work page 2021

  6. [4]

    W., Xu, L

    Chen, B. W., Xu, L. & Mavrikakis, M. Computational methods in heterogeneous catalysis. Chemical Reviews 121, 1007–1048 (2020)

    work page 2020

  7. [5]

    P., Payne, M

    Bartók, A. P., Payne, M. C., Kondor, R. & Csányi, G. Gaussian Approximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons. Phys. Rev. Lett. 104, 136403. ISSN : 0031-9007, 1079-7114. https://link.aps.org/doi/10.1103/PhysRevLett.104. 136403 (2024) (Apr. 1, 2010)

    doi:10.1103/physrevlett.104 2024

  8. [6]

    Schütt, K. et al. SchNet: A continuous-filter convolutional neural network for modeling quantum interactions in Advances in Neural Information Processing Systems 30 (Curran Associates, Inc., 2017). https : / / proceedings . neurips . cc / paper / 2017 / hash / 303ed4c69846ab36c2904d3ba8573050-Abstract.html (2024)

    work page 2024

Show all 45 references
  1. [7]

    Shapeev, A. V . Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Potentials. Multiscale Model. Simul. 14, 1153–1173. ISSN : 1540-3459, 1540-3467. http : //epubs.siam.org/doi/10.1137/15M1054183 (2024) (Jan. 2016)

    2024 doi

  2. [8]

    Bochkarev, A. et al. Efficient parametrization of the atomic cluster expansion. Phys. Rev. Materials 6, 013804. ISSN : 2475-9953. https : / / link . aps . org / doi / 10 . 1103 / PhysRevMaterials.6.013804 (2024) (Jan. 24, 2022)

    2024

  3. [9]

    & Ceriotti, M

    Nigam, J., Pozdnyakov, S. & Ceriotti, M. Recursive evaluation and iterative contraction of N-body equivariant features. The Journal of chemical physics 153. Publisher: AIP Publishing. https://pubs.aip.org/aip/jcp/article/153/12/121101/1023428 (2024) (2020)

    2024

  4. [10]

    Batzner, S. et al. E(3)-equivariant graph neural networks for data-efficient and accurate inter- atomic potentials. Nat Commun 13, 2453. ISSN : 2041-1723. https://www.nature.com/ articles/s41467-022-29939-5 (2024) (May 4, 2022)

    2024

  5. [11]

    P., Simm, G

    Batatia, I., Kovács, D. P., Simm, G. N. C., Ortner, C. & Csányi, G. MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields Jan. 26,

  6. [12]

    Hannagan, R. T. et al. First-principles design of a single-atom–alloy propane dehydrogenation catalyst. Science 372, 1444–1447 (2021)

    2021

  7. [13]

    L., Fako, E., De, S., Schäfer, A

    Schaaf, L. L., Fako, E., De, S., Schäfer, A. & Csányi, G. Accurate energy barriers for cat- alytic reaction pathways: an automatic training protocol for machine learning force fields. npj Computational Materials 9, 180 (2023)

    2023

  8. [14]

    Rhodes, B. et al. 17O NMR spectroscopy reveals CO2 speciation and dynamics in hydroxide- based carbon capture materials. ChemPhysChem, e202400941 (2024)

    2024

  9. [15]

    & Deringer, V

    Zhou, Y ., Zhang, W., Ma, E. & Deringer, V . L. Device-scale atomistic modelling of phase-change memory materials. Nature Electronics 6, 746–754 (2023). 8

    2023

  10. [16]

    Batatia, I. et al. A foundation model for atomistic materials chemistry Mar. 1, 2024. arXiv: 2401.00096[cond- mat, physics:physics] . http://arxiv.org/abs/2401.00096 (2024)

    2024 arXiv

  11. [17]

    Elijošius, R. et al. Zero Shot Molecular Generation via Similarity Kernels. http://arxiv. org/abs/2402.08708 (Feb. 2024)

    2024 arXiv

  12. [18]

    & Harvey, J

    Ivkovi´c, Ž., Jover, J. & Harvey, J. Transfer learning based on atomic feature extraction for the prediction of experimental 13 C chemical shifts. Digital Discovery 3, 2242–2251 (2024)

    2024

  13. [19]

    & Laio, A

    Bussi, G. & Laio, A. Using metadynamics to explore complex free-energy landscapes. Nat Rev Phys 2, 200–212. ISSN : 2522-5820. https://www.nature.com/articles/s42254-020- 0153-0 (2024) (Mar. 6, 2020)

    2024

  14. [20]

    Umbrella sampling

    Kästner, J. Umbrella sampling. WIREs Computational Molecular Science 1. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/wcms.66, 932–942. ISSN : 1759-0884. https: //onlinelibrary.wiley.com/doi/abs/10.1002/wcms.66 (2024) (2011)

    2024 doi

  15. [21]

    Berne, B. J. Molecular Dynamics in Systems with Multiple Time Scales: Reference System Propagator Algorithms in Computational Molecular Dynamics: Challenges, Methods, Ideas (eds Deuflhard, P. et al.) (Springer, Berlin, Heidelberg, 1999), 297–317. ISBN : 978-3-642- 58360-5

    1999

  16. [22]

    Tuckerman, M., Berne, B. J. & Martyna, G. J. Reversible multiple time scale molecular dynamics. The Journal of Chemical Physics 97, 1990–2001. ISSN : 0021-9606. https://doi. org/10.1063/1.463137 (2024) (Aug. 1, 1992)

    2024 doi

  17. [24]

    Tabak, E. G. & Turner, C. V . A Family of Nonparametric Density Estima- tion Algorithms. Communications on Pure and Applied Mathematics 66. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/cpa.21423, 145–164. ISSN : 1097-0312. https: //onlinelibrary.wiley.com/doi/abs/1...

    2024 doi

  18. [25]

    Noé, F., Olsson, S., Köhler, J. & Wu, H. Boltzmann Generators – Sampling Equilibrium States of Many-Body Systems with Deep Learning July 12, 2019. arXiv: 1812.01729[cond- mat,physics:physics,stat]. http://arxiv.org/abs/1812.01729 (2024)

    2024 arXiv

  19. [26]

    & Noé, F

    Klein, L. & Noé, F. Transferable Boltzmann Generators June 20, 2024. arXiv: 2406 . 14426[physics,stat]. http://arxiv.org/abs/2406.14426 (2024)

    2024 arXiv

  20. [27]

    I., Stimper, V ., Simm, G

    Midgley, L. I., Stimper, V ., Simm, G. N. C., Schölkopf, B. & Hernández-Lobato, J. M.Flow Annealed Importance Sampling Bootstrap Mar. 7, 2023. arXiv: 2208.01893[cs, q- bio, stat]. http://arxiv.org/abs/2208.01893 (2024)

    2024 arXiv

  21. [28]

    Midgley, L. I. et al. SE(3) Equivariant Augmented Coupling Flows Mar. 5, 2024. arXiv: 2308.10364[physics]. http://arxiv.org/abs/2308.10364 (2024)

    2024 arXiv

  22. [29]

    & von Lilienfeld, O

    Coretti, A., Falkner, S., Weinreich, J., Dellago, C. & von Lilienfeld, O. A.Boltzmann Generators and the New Frontier of Computational Sampling in Many-Body Systems2024. arXiv: 2404. 16566[physics]. http://arxiv.org/abs/2404.16566 (2024)

    2024 arXiv

  23. [30]

    Rotskoff, G. M. Sampling thermodynamic ensembles of molecular systems with generative neural networks: Will integrating physics-based models close the generalization gap? Cur- rent Opinion in Solid State and Materials Science 30, 101158. ISSN : 1359-0286. https:// www.scienced...

    2024

  24. [31]

    Klein, L. et al. Timewarp: Transferable Acceleration of Molecular Dynamics by Learning Time- Coarsened Dynamics Dec. 1, 2023. arXiv: 2302.01170[cond- mat, physics:physics, stat]. http://arxiv.org/abs/2302.01170 (2024)

    2024 arXiv

  25. [32]

    & Smidt, T

    Geiger, M. & Smidt, T. e3nn: Euclidean Neural Networks July 18, 2022. arXiv: 2207 . 09453[cs]. http://arxiv.org/abs/2207.09453 (2024)

    2024 arXiv

  26. [33]

    Lie groups and Lie algebras 247-267 (Elements of the History of Mathematics, 1994)

    Bourbaki, N. Lie groups and Lie algebras 247-267 (Elements of the History of Mathematics, 1994)

    1994

  27. [34]

    A solution for the best rotation to relate two sets of vectors

    Kabsch, W. A solution for the best rotation to relate two sets of vectors. Acta Cryst A 32. Publisher: International Union of Crystallography, 922–923. ISSN : 0567-7394. https : / / journals.iucr.org/a/issues/1976/05/00/a12999/ (2024) (Sept. 1, 1976). 9

    2024

  28. [35]

    & Krishnapriyan, A

    Qu, E. & Krishnapriyan, A. S. The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains. arXiv preprint arXiv:2410.24169 (2024)

    2024 arXiv

  29. [36]

    & Smidt, T

    Liao, Y .-L., Wood, B., Das, A. & Smidt, T. Equiformerv2: Improved equivariant transformer for scaling to higher-degree representations. arXiv preprint arXiv:2306.12059 (2023)

    2023 arXiv

  30. [37]

    & Csányi, G

    Várnai, C., Bernstein, N., Mones, L. & Csányi, G. Tests of an Adaptive QM/MM Calculation on Free Energy Profiles of Chemical Reactions in Solution. J. Phys. Chem. B 117, 12202–12211. ISSN : 1520-6106, 1520-5207. https://pubs.acs.org/doi/10.1021/jp405974b (2024) (Oct. 10, 2013)

    2024 doi

  31. [39]

    Eastman, P. et al. SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials. Sci Data 10. Publisher: Nature Publishing Group, 11. ISSN : 2052-4463. https://www.nature.com/articles/s41597-022-01882-6 (2024) (Jan. 4, 2023)

    2024

  32. [40]

    & Rao, F

    Gfeller, D., De Los Rios, P., Caflisch, A. & Rao, F. Complex network analysis of free-energy landscapes. Proceedings of the National Academy of Sciences 104, 1817–1822 (2007)

    2007

  33. [41]

    Darby, J. P.et al. Tensor-Reduced Atomic Density Representations.Phys. Rev. Lett.131, 028001. ISSN : 0031-9007, 1079-7114. https://link.aps.org/doi/10.1103/PhysRevLett.131. 028001 (2024) (July 13, 2023)

    2024 doi

  34. [42]

    Bonomi, M. et al. PLUMED: A portable plugin for free-energy calculations with molecular dynamics. Computer Physics Communications 180, 1961–1972 (2009). 10 Appendices A Detailed equations and BoostMD architecture In this section we provide the details of the Boost MD architect...

    2009

  35. [44]

    Direct energy readout The reference model has an energy associated with each of its layers t. We perform a readout on each of the BoostMD layers t′ for each reference layer energy t, E(t) i = Eref,(t) i + X t′ MLP(t′)(h(t′) i ) X j |xij|, (11) where the multiplicative factor |...

  36. [45]

    h(t) i = R¯θi (Hi,k) + X t′ MLP(t′)(h(t′) i ) X j |xij| (12) The original readout of the reference model Rt can then be used to compute the energy as E(t) i = Rt h(t) i

    Reference node feature readout Here we predict the changes to the reference’s models node features. h(t) i = R¯θi (Hi,k) + X t′ MLP(t′)(h(t′) i ) X j |xij| (12) The original readout of the reference model Rt can then be used to compute the energy as E(t) i = Rt h(t) i . (13) T...

  37. [2023]

    http://arxiv.org/ abs/2206.07697 (2024)

    arXiv: 2206.07697[cond- mat, physics:physics, stat] . http://arxiv.org/ abs/2206.07697 (2024)

    2024 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.