arxiv: 2605.06140 · v1 · submitted 2026-05-07 · 💻 cs.LG · cs.AI

Recognition: unknown

SymDrift: One-Shot Generative Modeling under Symmetries

Samir Darouich , Vinh Tong , Llu\'is Pastor-P\'erez , Tanja Bien , Loay Mualem , Mathias Niepert

Authors on Pith no claims yet

Pith reviewed 2026-05-08 13:46 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords generative modelingsymmetriesdrifting modelsmolecular conformersone-shot generationequivariant modelstransition states

0 comments

The pith

SymDrift lets drifting models generate symmetric structures like molecular conformers in one step by symmetrizing the drift field rather than the training data.

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

The paper seeks to make one-shot drifting models respect global symmetries such as rotations without paying the cost of fully symmetrizing the empirical distribution. Standard equivariant drifting models produce a different field than the symmetrized target, so the authors replace that mismatch with two explicit fixes: a drift computed after optimal coordinate alignment and an embedding that is invariant under the symmetry group by design. If these fixes work, single-step sampling becomes accurate for physical systems while cutting compute by up to forty times relative to multi-step baselines. Readers who generate molecules or transition states at scale would then gain a practical route to high-throughput virtual screening and reaction exploration.

Core claim

SymDrift resolves the symmetry mismatch in drifting models by defining a symmetrized drift via optimal alignment in coordinate space and a G-invariant embedding that removes symmetry ambiguity, so that the generated drifting field matches the one induced by the symmetrized target distribution and enables accurate one-shot sampling.

What carries the argument

Symmetrized drift obtained by optimal alignment together with a G-invariant embedding that removes symmetry ambiguity by construction.

If this is right

One-step sampling becomes practical for rotation-invariant molecular distributions.
Generation cost drops by up to forty times while staying competitive with multi-step equivariant models.
Performance on standard conformer and transition-state benchmarks exceeds other one-shot methods.
High-throughput tasks such as virtual screening become feasible on modest hardware.

Where Pith is reading between the lines

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

The same alignment-plus-invariant-embedding pattern could apply to other finite or continuous symmetry groups.
Real-time molecular design loops could incorporate the method directly when latency matters more than marginal accuracy gains.
Integration with existing equivariant architectures might further reduce any remaining per-sample alignment overhead.

Load-bearing premise

That applying optimal alignment to the drift or using an invariant embedding produces a field equivalent to the fully symmetrized target without introducing bias or hidden per-sample costs that cancel the claimed speed gain.

What would settle it

Generate a large set of samples with SymDrift and with a baseline that explicitly symmetrizes every training example, then test whether their marginal distributions over rotation-invariant features differ beyond sampling noise.

Figures

Figures reproduced from arXiv: 2605.06140 by Llu\'is Pastor-P\'erez, Loay Mualem, Mathias Niepert, Samir Darouich, Tanja Bien, Vinh Tong.

**Figure 1.** Figure 1: Conceptual illustration of SymDrift. a The base drift assigns highest kernel weights ˜k(x, y) to y + 3 , despite y + 2 being the closest conformer under symmetry, leading to a biased drifting field Vˆ + p (x) ̸= V+ pG (x). b SymDrift incorporates symmetries via 1) explicit alignment or 2) Ginvariant embedding space. c This leads to SymDrift correctly identifying y + 2 in the symmetrized space, producing t… view at source ↗

**Figure 2.** Figure 2: presents an ablation study on the GEOM-QM9 dataset, assessing how different formulations of the drifting field, ranging from base coordinate and iterative alignment to embedding space representations, affect generative performance. Validating Theorem 1, we observe a complete training collapse when applying the base coordinate drift to an equivariant architecture. The model achieves 0% coverage at the defin… view at source ↗

read the original abstract

Generative modeling of physical systems, such as molecules, requires learning distributions that are invariant under global symmetries, such as rotations in three-dimensional space. Equivariant diffusion and flow matching models can incorporate such invariances effectively, even when trained on a non-invariant empirical distribution, but they typically rely on costly multi-step sampling. Recently, drifting models have emerged as an efficient alternative, enabling single-step generation and achieving state-of-the-art performance in generative modeling tasks. However, we show that drifting models face a symmetry-specific challenge, since an equivariant generator does not generally produce the same drifting field as the one obtained from the symmetrized target distribution. Addressing this issue would require expensive symmetrization of the empirical distribution. To avoid this cost, we propose SymDrift, a framework that makes the drifting field itself symmetry-aware. We introduce two complementary strategies: (i) a symmetrized drift in coordinate space based on optimal alignment, and (ii) a $G$-invariant embedding that removes symmetry ambiguity by construction. Empirically, SymDrift outperforms existing one-shot methods on standard benchmarks for conformer and transition state generation, while remaining competitive with significantly more expensive multi-step approaches. By enabling one-shot inference, SymDrift reduces computational overhead by up to 40$\times$ compared to existing baselines, making it promising for high-throughput applications such as virtual drug screening and large-scale reaction network exploration.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

SymDrift gives a workable fix for symmetry in one-shot drifting models for molecules, but the optimal-alignment step may not fully match full symmetrization.

read the letter

The paper's main move is to notice that drifting models, which are fast and one-step, do not automatically produce the right drift field when the target distribution is symmetrized under rotations or permutations. They offer two practical patches: symmetrized drift via optimal alignment in coordinate space, and a G-invariant embedding that bakes the symmetry in from the start. Both avoid the cost of explicitly averaging the empirical measure over the group at training time. On the benchmarks they report, this lets SymDrift beat other one-shot baselines for conformer and transition-state generation while staying competitive with much slower multi-step equivariant methods, with claimed speed-ups up to 40x. That speed advantage is the clearest practical payoff for high-throughput uses like virtual screening. The construction itself is straightforward and the empirical numbers are presented cleanly. The soft spot is the one flagged in the stress-test note. Optimal alignment picks a single representative rather than integrating over the orbit, and for groups that include both continuous rotations and discrete permutations of identical atoms this choice is not guaranteed to recover the exact vector field that full symmetrization would give. If the learned drift ends up slightly off, the generated distribution could carry residual bias even when standard metrics look fine. The abstract does not include the derivation or the ablation that would show the two fields are close enough in practice, so a referee would need to see that evidence. The rest of the paper appears to build on existing drifting-model and equivariant-diffusion literature without obvious citation gaps. This is the kind of targeted engineering paper that people working on fast generative models for physical systems would want to read. It is worth sending to peer review so the math and the controls can be checked properly.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes SymDrift, a framework for one-shot generative modeling of symmetry-invariant distributions in physical systems like molecules. It identifies that standard drifting models do not produce the correct drifting field when the generator is equivariant but the target is symmetrized, and introduces two strategies to address this: symmetrized drift via optimal alignment in coordinate space and G-invariant embeddings. The approach is evaluated on conformer and transition state generation tasks, where it outperforms other one-shot methods and competes with multi-step approaches while offering substantial computational savings.

Significance. If the proposed constructions correctly yield unbiased symmetry-aware drifting fields, this work has the potential to significantly impact high-throughput generative tasks in chemistry and physics by providing an efficient one-shot alternative to multi-step equivariant diffusion models. The claimed 40× reduction in overhead is a notable practical advantage for applications such as virtual drug screening.

major comments (1)

[The symmetrized drift construction (Section 3)] The central claim that the symmetrized drift based on optimal alignment produces a field equivalent to that from the fully symmetrized target distribution is not sufficiently supported. Optimal alignment selects a single best representative rather than integrating over the group orbit, and for groups involving continuous rotations and discrete permutations of atoms, this may not recover the same vector field. This equivalence is load-bearing for the assertion that SymDrift avoids bias without the explicit averaging cost.

minor comments (2)

The paper would benefit from an explicit statement of the group G in the introduction, including whether it includes reflections or only proper rotations.
[Experimental section] Include ablation studies separating the contributions of the two proposed strategies (optimal alignment vs. G-invariant embedding) to the performance gains.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and insightful feedback on our manuscript. We address the major comment point-by-point below and have revised the paper to strengthen the theoretical justification for the symmetrized drift construction.

read point-by-point responses

Referee: [The symmetrized drift construction (Section 3)] The central claim that the symmetrized drift based on optimal alignment produces a field equivalent to that from the fully symmetrized target distribution is not sufficiently supported. Optimal alignment selects a single best representative rather than integrating over the group orbit, and for groups involving continuous rotations and discrete permutations of atoms, this may not recover the same vector field. This equivalence is load-bearing for the assertion that SymDrift avoids bias without the explicit averaging cost.

Authors: We appreciate this precise observation on the distinction between selecting a single optimal representative and explicit integration over the group orbit. In Section 3, the symmetrized drift is defined by solving for the group element g* that minimizes the squared Euclidean distance between the generated point and the target after applying g, which is the standard optimal alignment procedure used in molecular conformation tasks. While this is a pointwise selection rather than an average, we note that the resulting vector field is the gradient of the aligned distance and, under the assumption that the optimal alignment is unique (which holds almost surely for generic configurations under continuous rotations), it coincides with the field induced by the symmetrized distribution. To make this rigorous, the revised manuscript now includes a formal proposition establishing the equivalence of the expected drift fields, along with a brief error analysis for the discrete permutation component when multiple alignments are possible. We have also added a small-scale numerical verification comparing the aligned drift to a Monte-Carlo averaged symmetrized field on toy examples. These additions directly address the load-bearing claim without changing the method or empirical results. revision: yes

Circularity Check

0 steps flagged

No significant circularity; algorithmic constructions are independent of target result

full rationale

The abstract and context present SymDrift as a new framework introducing symmetrized drift via optimal alignment and G-invariant embedding to address a symmetry challenge in drifting models. No equations, derivations, or parameter-fitting steps are supplied that reduce by construction to the symmetrized target distribution or to self-citations. Claims rest on empirical outperformance on benchmarks rather than any self-definitional, fitted-input, or uniqueness-imported reduction. The derivation chain is therefore self-contained and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the domain assumption that physical systems possess global symmetries (rotations, etc.) and on the technical assumption that optimal alignment or invariant embeddings can be computed efficiently enough to preserve the one-shot advantage. No free parameters or new physical entities are mentioned.

axioms (1)

domain assumption Physical systems such as molecules are subject to global symmetries (e.g., rotations in 3D space) that the generative distribution must respect.
Stated in the first sentence of the abstract as a requirement for generative modeling of physical systems.

invented entities (2)

Symmetrized drift in coordinate space no independent evidence
purpose: To produce a symmetry-aware drifting field via optimal alignment without explicit symmetrization of the empirical distribution.
Introduced as strategy (i) to address the mismatch between equivariant generators and symmetrized targets.
G-invariant embedding no independent evidence
purpose: To remove symmetry ambiguity by construction in the embedding space.
Introduced as strategy (ii) complementary to the coordinate-space approach.

pith-pipeline@v0.9.0 · 5575 in / 1401 out tokens · 55436 ms · 2026-05-08T13:46:24.701547+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

63 extracted references · 7 canonical work pages · 2 internal anchors

[1]

Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

2020
[2]

Score-based generative modeling through stochastic differential equations

Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

2021
[3]

Generative Modeling via Drifting

Mingyang Deng, He Li, Tianhong Li, Yilun Du, and Kaiming He. Generative modeling via drifting.arXiv preprint arXiv:2602.04770, 2026

work page internal anchor Pith review arXiv 2026
[4]

Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InThe Eleventh International Conference on Learning Representations, 2023

2023
[5]

Flow straight and fast: Learning to generate and transfer data with rectified flow

Xingchao Liu, Chengyue Gong, and qiang liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations, 2023

2023
[6]

Stochastic interpolants: A unifying framework for flows and diffusions.Journal of Machine Learning Research, 26(209): 1–80, 2025

Michael Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions.Journal of Machine Learning Research, 26(209): 1–80, 2025

2025
[7]

Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35: 26565–26577, 2022

Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35: 26565–26577, 2022

2022
[8]

Mean flows for one-step generative modeling

Zhengyang Geng, Mingyang Deng, Xingjian Bai, J Zico Kolter, and Kaiming He. Mean flows for one-step generative modeling. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

2025
[9]

Diffusion transformers with representation autoencoders

Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with representation autoencoders. InThe Fourteenth International Conference on Learning Repre- sentations, 2026

2026
[10]

Propmolflow: property-guided molecule generation with geometry-complete flow matching.Nature Computa- tional Science, pages 1–10, 2026

Cheng Zeng, Jirui Jin, Connor Ambrose, George Karypis, Mark Transtrum, Ellad B Tadmor, Richard G Hennig, Adrian Roitberg, Stefano Martiniani, and Mingjie Liu. Propmolflow: property-guided molecule generation with geometry-complete flow matching.Nature Computa- tional Science, pages 1–10, 2026

2026
[11]

Equivariant flow matching with hybrid probability transport for 3d molecule generation.Advances in Neural Information Processing Systems, 2023

Yuxuan Song, Jingjing Gong, Minkai Xu, Ziyao Cao, Yanyan Lan, Stefano Ermon, Hao Zhou, and Wei-Ying Ma. Equivariant flow matching with hybrid probability transport for 3d molecule generation.Advances in Neural Information Processing Systems, 2023

2023
[12]

Equivariant flow matching.Advances in Neural Information Processing Systems, 2023

Leon Klein, Andreas Krämer, and Frank Noé. Equivariant flow matching.Advances in Neural Information Processing Systems, 2023

2023
[13]

Efficient molecular con- former generation with SO(3)-averaged flow matching and reflow

Zhonglin Cao, Mario Geiger, Allan Dos Santos Costa, Danny Reidenbach, Karsten Kreis, Tomas Geffner, Franco Pellegrini, Guoqing Zhou, and Emine Kucukbenli. Efficient molecular con- former generation with SO(3)-averaged flow matching and reflow. InForty-second International Conference on Machine Learning, 2025

2025
[14]

Improving and generalizing flow-based genera- tive models with minibatch optimal transport.Transactions on Machine Learning Research,

Alexander Tong, Kilian FATRAS, Nikolay Malkin, Guillaume Huguet, Yanlei Zhang, Jarrid Rector-Brooks, Guy Wolf, and Yoshua Bengio. Improving and generalizing flow-based genera- tive models with minibatch optimal transport.Transactions on Machine Learning Research,
[15]

Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

2024
[16]

Proteina: Scaling flow-based protein structure generative models.arXiv preprint arXiv:2503.00710, 2025

Tomas Geffner, Kieran Didi, Zuobai Zhang, Danny Reidenbach, Zhonglin Cao, Jason Yim, Mario Geiger, Christian Dallago, Emine Kucukbenli, Arash Vahdat, et al. Proteina: Scaling flow-based protein structure generative models.arXiv preprint arXiv:2503.00710, 2025. 11

work page arXiv 2025
[17]

Equivariant flows: exact likelihood generative learning for symmetric densities

Jonas Köhler, Leon Klein, and Frank Noé. Equivariant flows: exact likelihood generative learning for symmetric densities. InInternational conference on machine learning, pages 5361–5370. PMLR, 2020

2020
[18]

Equivariant diffusion for molecule generation in 3D

Emiel Hoogeboom, Víctor Garcia Satorras, Clément Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3D. InProceedings of the 39th International Conference on Machine Learning, 2022

2022
[19]

Rao- blackwell gradient estimators for equivariant denoising diffusion

Vinh Tong, Trung-Dung Hoang, Anji Liu, Guy Van den Broeck, and Mathias Niepert. Rao- blackwell gradient estimators for equivariant denoising diffusion. InProceedings of the 39th Conference on Neural Information Processing Systems, 2025

2025
[20]

Et-flow: Equivariant flow-matching for molecular conformer generation.Advances in Neural Information Processing Systems, 37:128798–128824, 2024

Majdi Hassan, Nikhil Shenoy, Jungyoon Lee, Hannes Stärk, Stephan Thaler, and Dominique Beaini. Et-flow: Equivariant flow-matching for molecular conformer generation.Advances in Neural Information Processing Systems, 37:128798–128824, 2024

2024
[21]

Consistency models

Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. In Proceedings of the 40th International Conference on Machine Learning, 2023

2023
[22]

Learning to discretize denoising diffusion ODEs

Vinh Tong, Dung Trung Hoang, Anji Liu, Guy Van den Broeck, and Mathias Niepert. Learning to discretize denoising diffusion ODEs. InThe Thirteenth International Conference on Learning Representations, 2025

2025
[23]

Instaflow: One step is enough for high-quality diffusion-based text-to-image generation

Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion-based text-to-image generation. InThe Twelfth International Conference on Learning Representations, 2023

2023
[24]

Simple reflow: Improved techniques for fast flow models

Beomsu Kim, Yu-Guan Hsieh, Michal Klein, marco cuturi, Jong Chul Ye, Bahjat Kawar, and James Thornton. Simple reflow: Improved techniques for fast flow models. InThe Thirteenth International Conference on Learning Representations, 2025

2025
[25]

Slimflow: Training smaller one-step diffusion models with rectified flow

Yuanzhi Zhu, Xingchao Liu, and Qiang Liu. Slimflow: Training smaller one-step diffusion models with rectified flow. InEuropean Conference on Computer Vision, pages 342–359. Springer, 2024

2024
[26]

Improving the training of rectified flows

Sangyun Lee, Zinan Lin, and Giulia Fanti. Improving the training of rectified flows. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

2024
[27]

Better informed distance geometry: using what we know to improve conformation generation.Journal of chemical information and modeling, 55 (12):2562–2574, 2015

Sereina Riniker and Gregory A Landrum. Better informed distance geometry: using what we know to improve conformation generation.Journal of chemical information and modeling, 55 (12):2562–2574, 2015

2015
[28]

Predicting molecular conformation via dynamic graph score matching.Advances in neural information processing systems, 34: 19784–19795, 2021

Shitong Luo, Chence Shi, Minkai Xu, and Jian Tang. Predicting molecular conformation via dynamic graph score matching.Advances in neural information processing systems, 34: 19784–19795, 2021

2021
[29]

Learning gradient fields for molecular conformation generation

Chence Shi, Shitong Luo, Minkai Xu, and Jian Tang. Learning gradient fields for molecular conformation generation. InInternational conference on machine learning, pages 9558–9568. PMLR, 2021

2021
[30]

Geodiff: A geo- metric diffusion model for molecular conformation generation.arXiv preprint arXiv:2203.02923, 2022

Minkai Xu, Lantao Yu, Yang Song, Chence Shi, Stefano Ermon, and Jian Tang. Geodiff: A geo- metric diffusion model for molecular conformation generation.arXiv preprint arXiv:2203.02923, 2022

work page arXiv 2022
[31]

Torsional diffusion for molecular conformer generation.Advances in neural information processing systems, 35:24240–24253, 2022

Bowen Jing, Gabriele Corso, Jeffrey Chang, Regina Barzilay, and Tommi Jaakkola. Torsional diffusion for molecular conformer generation.Advances in neural information processing systems, 35:24240–24253, 2022

2022
[32]

NExt-mol: 3d diffusion meets 1d language modeling for 3d molecule generation

Zhiyuan Liu, Yanchen Luo, Han Huang, Enzhi Zhang, Sihang Li, Junfeng Fang, Yaorui Shi, Xiang Wang, Kenji Kawaguchi, and Tat-Seng Chua. NExt-mol: 3d diffusion meets 1d language modeling for 3d molecule generation. InThe Thirteenth International Conference on Learning Representations, 2025. 12

2025
[33]

Energy-guided flow matching enables few-step conformer generation and ground-state identification.arXiv preprint arXiv:2512.22597, 2025

Guikun Xu, Xiaohan Yi, Peilin Zhao, and Yatao Bian. Energy-guided flow matching enables few-step conformer generation and ground-state identification.arXiv preprint arXiv:2512.22597, 2025

work page arXiv 2025
[34]

Nudged elastic band method for finding minimum energy paths of transitions

Hannes Jónsson, Greg Mills, and Karsten W Jacobsen. Nudged elastic band method for finding minimum energy paths of transitions. InClassical and quantum dynamics in condensed phase simulations, pages 385–404. World Scientific, 1998

1998
[35]

Uberuaga, and Hannes Jónsson

Graeme Henkelman, Blas P. Uberuaga, and Hannes Jónsson. A climbing image nudged elastic band method for finding saddle points and minimum energy paths.J. Chem. Phys., 113(22): 9901–9904, December 2000. ISSN 0021-9606

2000
[36]

Search for stationary points on surfaces.J

Ajit Banerjee, Noah Adams, Jack Simons, and Ron Shepard. Search for stationary points on surfaces.J. Phys. Chem., 89(1):52–57, January 1985. ISSN 0022-3654, 1541-5740

1985
[37]

A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives.J

Graeme Henkelman and Hannes Jónsson. A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives.J. Chem. Phys., 111(15):7010–7022, October 1999. ISSN 0021-9606

1999
[38]

Generating transition states of isomerization reactions with deep learning.Physical Chemistry Chemical Physics, 22(41):23618–23626, 2020

Lagnajit Pattanaik, John B Ingraham, Colin A Grambow, and William H Green. Generating transition states of isomerization reactions with deep learning.Physical Chemistry Chemical Physics, 22(41):23618–23626, 2020

2020
[39]

Accurate transition state generation with an object-aware equivariant elementary reaction diffusion model.Nature computational science, 3(12):1045–1055, 2023

Chenru Duan, Yuanqi Du, Haojun Jia, and Heather J Kulik. Accurate transition state generation with an object-aware equivariant elementary reaction diffusion model.Nature computational science, 3(12):1045–1055, 2023

2023
[40]

Yuan, Anup Kumar, Xingyi Guan, Eric D

Eric C.-Y . Yuan, Anup Kumar, Xingyi Guan, Eric D. Hermes, Andrew S. Rosen, Judit Zádor, Teresa Head-Gordon, and Samuel M. Blau. Analytical ab initio hessian from a deep learning potential for transition state optimization.Nat. Commun., 15(1):8865, October 2024. ISSN 2041-1723

2024
[41]

Kitchin, Zachary W

Brook Wander, Muhammed Shuaibi, John R. Kitchin, Zachary W. Ulissi, and C. Lawrence Zitnick. Cattsunami: Accelerating transition state energy calculations with pretrained graph neural networks.ACS Catal., 15(7):5283–5294, 2025

2025
[42]

Diffusion-based generative ai for exploring transition states from 2d molecular graphs.Nature Communications, 15(1):341, 2024

Seonghwan Kim, Jeheon Woo, and Woo Youn Kim. Diffusion-based generative ai for exploring transition states from 2d molecular graphs.Nature Communications, 15(1):341, 2024

2024
[43]

Goflow: efficient transition state geometry prediction with flow matching and e (3)-equivariant neural networks.Digital Discovery, 4(12):3492–3501, 2025

Leonard Galustian, Konstantin Mark, Johannes Karwounopoulos, Maximilian P-P Kovar, and Esther Heid. Goflow: efficient transition state geometry prediction with flow matching and e (3)-equivariant neural networks.Digital Discovery, 4(12):3492–3501, 2025

2025
[44]

Adaptive transition-state refinement with learned equilibrium flows.Journal of Chemical Information and Modeling, 66(4):2154–2165, 2026

Samir Darouich, Vinh Tong, Tanja Bien, Johannes Kästner, and Mathias Niepert. Adaptive transition-state refinement with learned equilibrium flows.Journal of Chemical Information and Modeling, 66(4):2154–2165, 2026

2026
[45]

motsart: Accelerating automated transition state search with generative models in a low-data regime

Leonard Galustian, Johannes Karwounopoulos, Tori Demuth, Jasper De Landsheere, Konstantin Mark, Maximilian P-P Kovar, Anton Zamyatin, Dennis Svatunek, and Esther Heid. motsart: Accelerating automated transition state search with generative models in a low-data regime. ChemRxiv, 2026(0417), 2026

2026
[46]

Equivariant score-based gen- erative models provably learn distributions with symmetries efficiently.arXiv preprint arXiv:2410.01244, 2024

Ziyu Chen, Markos A Katsoulakis, and Benjamin J Zhang. Equivariant score-based gen- erative models provably learn distributions with symmetries efficiently.arXiv preprint arXiv:2410.01244, 2024

work page internal anchor Pith review arXiv 2024
[47]

A solution for the best rotation to relate two sets of vectors.Foundations of Crystallography, 32(5):922–923, 1976

Wolfgang Kabsch. A solution for the best rotation to relate two sets of vectors.Foundations of Crystallography, 32(5):922–923, 1976

1976
[48]

The hungarian method for the assignment problem.Naval research logistics quarterly, 2(1-2):83–97, 1955

Harold W Kuhn. The hungarian method for the assignment problem.Naval research logistics quarterly, 2(1-2):83–97, 1955

1955
[49]

Geom, energy-annotated molecular conforma- tions for property prediction and molecular generation.Scientific data, 9(1):185, 2022

Simon Axelrod and Rafael Gomez-Bombarelli. Geom, energy-annotated molecular conforma- tions for property prediction and molecular generation.Scientific data, 9(1):185, 2022. 13

2022
[50]

Crest—a program for the exploration of low-energy molecular chemical space.The Journal of Chemical Physics, 160(11), 2024

Philipp Pracht, Stefan Grimme, Christoph Bannwarth, Fabian Bohle, Sebastian Ehlert, Gereon Feldmann, Johannes Gorges, Marcel Müller, Tim Neudecker, Christoph Plett, et al. Crest—a program for the exploration of low-energy molecular chemical space.The Journal of Chemical Physics, 160(11), 2024

2024
[51]

Coley, Regina Barzilay, Klavs Jensen, William Green, and Tommi S

Octavian-Eugen Ganea, Lagnajit Pattanaik, Connor W. Coley, Regina Barzilay, Klavs Jensen, William Green, and Tommi S. Jaakkola. Geomol: Torsional geometric generation of molecular 3d conformer ensembles. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021

2021
[52]

Sampling 3d molecular conformers with diffusion transformers.arXiv preprint arXiv:2506.15378, 2025

J Thorben Frank, Winfried Ripken, Gregor Lied, Klaus-Robert MÃžller, Oliver T Unke, and Stefan Chmiela. Sampling 3d molecular conformers with diffusion transformers.arXiv preprint arXiv:2506.15378, 2025

work page arXiv 2025
[53]

Generative Sliced MMD Flows with Riesz Kernels

Ping He, Om Khangaonkar, Hamed Pirsiavash, Yikun Bai, and Soheil Kolouri. Sinkhorn-drifting generative models.arXiv preprint arXiv:2603.12366, 2026

work page arXiv 2026
[54]

Swallowing the bitter pill: Simplified scalable conformer generation

Yuyang Wang, Ahmed AA Elhag, Navdeep Jaitly, Joshua M Susskind, and Miguel Ángel Bautista. Swallowing the bitter pill: Simplified scalable conformer generation. InInternational Conference on Machine Learning, pages 50400–50418. PMLR, 2024

2024
[55]

High accuracy barrier heights, enthalpies, and rate coefficients for chemical reactions.Scientific Data, 9(1):417, 2022

Kevin Spiekermann, Lagnajit Pattanaik, and William H Green. High accuracy barrier heights, enthalpies, and rate coefficients for chemical reactions.Scientific Data, 9(1):417, 2022

2022
[56]

Joshi, Cristian Bodnar, Simon V

Chaitanya K. Joshi, Cristian Bodnar, Simon V . Mathis, Taco Cohen, and Pietro Liò. On the expressive power of geometric graph neural networks. InInternational Conference on Machine Learning, 2023

2023
[57]

Weisfeiler leman for euclidean equivariant machine learning

Snir Hordan, Tal Amir, and Nadav Dym. Weisfeiler leman for euclidean equivariant machine learning. InInternational Conference on Machine Learning, 2024

2024
[58]

On the expressive power of sparse geometric mpnns

Yonatan Sverdlov and Nadav Dym. On the expressive power of sparse geometric mpnns. In International Conference on Learning Representations, 2025

2025
[59]

Machine learning of reaction properties via learned repre- sentations of the condensed graph of reaction.Journal of Chemical Information and Modeling, 62(9):2101–2110, 2021

Esther Heid and William H Green. Machine learning of reaction properties via learned repre- sentations of the condensed graph of reaction.Journal of Chemical Information and Modeling, 62(9):2101–2110, 2021. 14 SYMDRIFT: ONE-SHOTGENERATIVEMODELING UNDERSYMMETRIES ADDITIONALMATERIAL A Extented Background 16 B Theoretical Proofs 16 B.1 Gradient definition . ...

2021
[60]

Expression forV + pG(x)
[61]

Expression for ˆV+ p (x)
[62]

-R” denotes Recall, and “-P

Comparison. Step 1: Symmetrized distribution drift.By definition, V+ pG(x) = Ey+∼pG k(x,y +)(y+ −x) Ey+∼pG k(x,y +) . Usingp G(y) =E g∼G[p(g−1y)], we rewrite the numerator: ApG(x) =E y+∼pEg∼G k(x, gy+)(gy+ −x) =E g∼GEy+∼p k(x, gy+)(gy+ −x) Using orthogonality of G, we have k(x, gy) =k(g −1x,y), and we rewrite gy−x=g(y−g −1x). Hence, ApG(x) =E g∼G h gE y∼p...
[63]

Guidelines: • The answer [N/A] means that the paper does not involve crowdsourcing nor research with human subjects

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 (or an equivalent approval/review based on the requirements of your country or ...