Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Clifford Group Equivariant Diffusion Models for 3D Molecular Generation

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Diffusing over all Clifford multivector grades, not just vectors, improves 3D molecular generation.

desk verdict Legitimate but modest extension of equivariant diffusion to Clifford multivector latents; the all-grade advantage over one-vector is within noise and the attribution claim is untested. read the letter →

arxiv 2504.15773 v2 pith:CZZVJH6M submitted 2025-04-22 cs.LG cs.AI

Cong Liu , Sharvaree Vadgama , David Ruhe , Erik Bekkers , Patrick Forré This is my paper · ORCID
classification cs.LGcs.AI
keywords CliffordalgebramultivectorsequivariantdiffusionmoleculargenerationgeometricproductlatentE(n)equivarianceQM9
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

The paper sets out to show that diffusion models for 3D molecules need not live on plain Euclidean vectors: by embedding molecular coordinates in a Clifford algebra and diffusing over every grade of the resulting multivector, a generative model can use scalars, vectors, bivectors, and trivectors at once. The paper constructs two variants, a one-vector diffusion that is essentially diffusion on coordinates and an all-grade diffusion that first encodes the molecule into a complete multivector latent space, then adds noise separately to each grade and denoises the whole multivector. On unconditional generation from the QM9 benchmark, the all-grade variant reaches 99.0 percent atom stability, 89.7 percent molecular stability, 96.4 percent validity, and 96.3 percent valid-and-unique molecules, matching or exceeding vector-only and higher-grade baselines on most metrics. If the claim holds, geometric structure encoded in higher Clifford grades is a usable resource for molecular generation rather than a mathematical decoration.

What carries the argument

The load-bearing object is the Clifford algebra Cl(R3), whose elements are multivectors written as a direct sum of grade-0 scalars, grade-1 vectors, grade-2 bivectors (oriented plane segments), and grade-3 trivectors (oriented volume elements), combined through the geometric product. The all-grade diffusion model uses this algebra in two places: a Clifford-EGNN encoder lifts an input molecular graph into a latent multivector with nontrivial higher-grade components, and the forward and backward diffusion processes are defined independently on each grade subspace, with the joint distribution factored as a product of Gaussian per grade. The denoising network is another Clifford-EGNN that outputs noise in the same algebra, so equivariance to the Euclidean group is preserved throughout. This lets the generative process reason jointly about positions and the higher-order geometric features associated with planes and volumes, which is what the paper claims improves generation.

What would settle it

Take the all-grade CDM and replace the encoded bivector and trivector features with random noise of the same dimension while keeping the encoder capacity fixed; if atom stability, molecular stability, validity, and valid-and-unique rates stay statistically unchanged, the claimed benefit of higher-grade geometric structure is not supported.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a diffusion process carried out over the full Clifford algebra Cl(R3), not just its grade-one subspace, generates higher-quality molecules. The coordinates X are identified with a one-vector in Clifford space, and a Clifford-EGNN encoder fills the scalar, bivector, and trivector components with learned geometric features; the grade-one component is reset to the true coordinates to preserve geometry. Gaussian noise is added independently to each grade subspace, and the denoising network predicts multivector-valued noise, so sampling reconstructs all grades jointly. Readout projects the denoised one-vector back to Euclidean space. Empirically the all-grade model matches or exceeds the one-vector model and most listed baselines on validity, uniqueness, and atom stability, which the authors read as evidence that higher-grade subspaces carry useful geometric signal for diffusion.

Load-bearing premise

The load-bearing premise is that the bivector and trivector latent components produced by the encoder carry real geometric signal that aids denoising; if those channels are only extra network capacity, the paper's central comparison does not establish its conclusion.

Editorial extensions

If this is right

  • All-grade Clifford diffusion produces more valid and unique molecules than one-vector diffusion, so extending the diffusion state space beyond coordinates is a viable path for molecular generation.
  • The one-vector CDM already matches the best non-diffusion baselines, so even the minimal Clifford embedding is competitive for unconditional generation.
  • Because noise is added and predicted per grade in a factored Gaussian, the standard denoising objective carries over unchanged, suggesting the same recipe can be applied to other E(n)-equivariant generation tasks.
  • CDM uses equal or fewer parameters than the comparison methods, so the quality gains are not bought by a larger network on this benchmark.

Reading between the lines

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

  • The paper does not ablate the encoder, so the all-grade gain might come from extra capacity; a controlled comparison that fills the extra grade channels with random features would tell whether the Clifford product itself matters.
  • A natural extension is to couple the independent per-grade noise into a joint noise model, since a molecule's geometry imposes correlations between positions, orientations, and volumes that independent additive noise ignores.
  • The same multivector-latent recipe could be tested on larger molecular or protein generation tasks, where planar and volumetric features may carry more information than in small benchmark molecules.
  • If higher-grade features genuinely help, readout could go beyond grade-one projection, for example using bivector components to predict local orientation or bond geometry instead of discarding them at the end.
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

3 major / 4 minor

Summary. The paper introduces Clifford Diffusion Models (CDMs), a family of E(n)-equivariant diffusion models for unconditional 3D molecular generation on QM9. Two variants are proposed: a one-vector CDM, which embeds atomic coordinates as grade-1 multivectors and denoises them with a Clifford-EGNN, and an all-grade CDM, which uses a Clifford-EGNN encoder to produce latent multivectors with nonzero scalar, bivector, and trivector components and then runs diffusion independently on each grade subspace. The reported experiments compare these models against several equivariant diffusion and flow-matching baselines, with CDM (all-grade) achieving the best numbers among CDM variants and competitive results overall. The paper argues that the results validate the use of Clifford grade-k subspaces for generative modeling.

Significance. If the central claim were fully established, the work would be a meaningful step: it extends latent diffusion from the vector-valued latent spaces of GeoLDM to the full multivector structure of a Clifford algebra, and it demonstrates that Clifford-EGNN backbones can be used as diffusion denoisers. The one-vector CDM is already competitive with strong baselines, which is a useful empirical data point. However, the paper's headline contribution, that all-grade diffusion improves quality and that this validates the geometric information in higher-grade Clifford subspaces, is not supported by the evidence as presented. The reported margins over the one-vector variant are within one another's error bars, the comparison is not controlled, and the conclusion section explicitly hedges the effectiveness claim. The paper would need a proper ablation or substantially softened claims before the central contribution can be accepted. No code or training details are provided, which further limits reproducibility and the ability to assess the source of the observed behavior.

major comments (3)
  1. [Section 4, Table 1] The claim that "CDMs with all-grade diffusion in general generate molecules with higher quality" is not supported by the reported statistics. For every metric, the one-vector and all-grade CDM results overlap under the reported standard deviations: atom stability 99.0±0.2 vs 98.9±0.0, molecular stability 89.7±1.4 vs 89.6±0.2, validity 96.4±1.0 vs 96.0±0.3, and valid & unique 96.3±1.0 vs 95.8±0.3. Since the all-grade variant also uses a substantially different architecture (an added encoder), the observed differences cannot be attributed to the diffusion process on higher-grade subspaces. This sentence is load-bearing for the abstract and conclusion, and it should either be backed by a proper statistical comparison or removed.
  2. [Section 3.2 and Section 4] The all-grade model differs from the one-vector model by more than the diffusion space: it adds a learned Clifford-EGNN encoder that produces latent bivector, trivector, and scalar features. There is no ablation that removes these higher-grade channels, replaces them with unstructured features of matched dimension and capacity, or otherwise isolates the effect of the Clifford structure. Without such an ablation, the improvement (or its absence) could be due to additional model capacity, a different training signal, or seed noise, rather than the geometric meaning of the Clifford subspaces. The paper's own Section 5 states that "its effectiveness compared to Clifford one-vector diffusion remains under evaluation," which directly undercuts the stronger wording in Section 4 and the abstract.
  3. [Section 3.2] The training and sampling procedure for the all-grade model is underspecified, which affects the interpretability of the central comparison. It is not stated whether the encoder E is trained jointly with the denoising network or pretrained and frozen, what the latent dimension is, how the grade-1 component is kept equal to X during training and sampling, or how the scalar, bivector, and trivector targets are normalized. These details are necessary both for reproducibility and for determining whether the all-grade variant is a latent diffusion model over coordinates alone or a generative model over a learned latent space.
minor comments (4)
  1. [Section 2.2] The text contains a typo: "For simplicitly" should read "For simplicity."
  2. [Section 4] The table caption and text would benefit from clarifying the exact definition and denominator of "Valid & Unique" and from reporting standard deviations for all methods, since several rows (GDM-AUG, PΘNITA, the Data row, and the one-vector CDM's atom stability) list no uncertainty.
  3. [Section 4, footnote 2] The statement that CDMs have "equal or less parameter counts compared to baseline models" is not accompanied by a table of parameter counts or architecture sizes, making the fairness claim unverifiable.
  4. [Section 3.2, footnote 1] The footnote defining "Clifford space" as "any product of Clifford algebras: Cl(R3)^k" is ambiguous; it should state explicitly whether the latent space is a direct sum of k copies of the algebra and how this interacts with the grade-wise diffusion described in the main text.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the paper's claims are empirical and benchmarked against external baselines, with the all-grade advantage itself explicitly hedged.

full rationale

I walked the derivation chain and found no step where a prediction is equivalent to an input by construction. The core claims are empirical: Table 1 compares CDM against external baselines on QM9, and the strongest statement, 'CDMs with all-grade diffusion in general generate molecules with higher quality,' is a data-dependent comparison, not a defined identity. The all-grade model's higher-grade latents are produced by a learned Clifford-EGNN encoder, initialized from X, and diffused; no equation in Section 3 forces the reported metric values, and the paper itself qualifies the result in Section 5: 'While its effectiveness compared to Clifford one-vector diffusion remains under evaluation.' Self-citations to prior Clifford network works (Ruhe et al. 2023a; Liu et al. 2024b) are used as architectural building blocks, not as evidence for the present empirical outcome, so they are not load-bearing in a circular sense. The absence of an ablation that removes or replaces the higher-grade channels is a legitimate validity concern about attribution, but it is not circularity: the experiment could in principle have gone either way. No uniqueness theorem is invoked, no fitted parameter is relabeled as a prediction, and no known result is renamed. The paper is therefore self-contained with respect to its empirical evaluation, and the only mild issue is normal self-citation of prior work by the same group.

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

The central claim is empirical, so the main assumptions are the geometric meaningfulness of learned higher-grade features and the standard evaluation protocol.

assumptions (4)
  • standard math Clifford algebra Cl(R^3) decomposes into grade-0,1,2,3 subspaces with a positive definite quadratic form.
    Background from Section 2.2; standard linear algebra.
  • domain assumption E(3) invariance of the generative distribution is required for physical validity.
    Section 1 asserts this, citing Xu et al. (2022). Standard in equivariant molecular generation.
  • ad hoc to paper The Clifford-EGNN encoder produces higher-grade latent components that are geometrically meaningful and useful for denoising.
    This is a design premise of Section 3.2; no ablation in the paper tests it.
  • domain assumption QM9 metrics (atom stability, mol stability, validity, valid & unique) are the appropriate evaluation of generative quality.
    Section 4 uses standard QM9 metrics from prior work; no justification beyond citation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Clifford Group Equivariant Diffusion Models for 3D Molecular Generation." pith.science (2026). https://pith.science/paper/CZZVJH6M

@misc{pith2026250415773,
  author       = {Pith},
  title        = {Pith review of: Clifford Group Equivariant Diffusion Models for 3D Molecular Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CZZVJH6M}},
  note         = {Machine review of arXiv:2504.15773}
}
abstract

This paper explores leveraging the Clifford algebra's expressive power for $\E(n)$-equivariant diffusion models. We utilize the geometric products between Clifford multivectors and the rich geometric information encoded in Clifford subspaces in \emph{Clifford Diffusion Models} (CDMs). We extend the diffusion process beyond just Clifford one-vectors to incorporate all higher-grade multivector subspaces. The data is embedded in grade-$k$ subspaces, allowing us to apply latent diffusion across complete multivectors. This enables CDMs to capture the joint distribution across different subspaces of the algebra, incorporating richer geometric information through higher-order features. We provide empirical results for unconditional molecular generation on the QM9 dataset, showing that CDMs provide a promising avenue for generative modeling.

Figures

Figures reproduced from arXiv: 2504.15773 by the authors.

Figure 1
Figure 1. During sampling of Clifford all-grade diffusion models, subspace features are initialized [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

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. Conditional Clifford-Steerable CNNs for PDE Modeling

    cs.LG 2025-10 conditional novelty 5.0 of 10

    Conditional Clifford-Steerable CNNs, which condition the equivariant kernel on pooled input features, improve PDE forecasting accuracy but do not prove the claimed complete kernel basis.

Reference graph

Works this paper leans on

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

  1. [1]

    Ballard, Joshua Bambrick, Sebastian W

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J. Ballard, Joshua Bambrick, Sebastian W. Bodenstein, David A. Evans, Chia-Chun Hung, Michael O’Neill, David Reiman, Kathryn Tunyasuvunakool, Zachary Wu, Akvilė Žemgulytė, Eirini Arvaniti, Charles Beattie, Ottavia Bertolli, Alex...

    2024

  2. [2]

    Erik J Bekkers, Sharvaree Vadgama, Rob Hesselink, Putri A Van der Linden, and David W. Romero. Fast, expressive se(n) equivariant networks through weight-sharing in position-orientation space. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  3. [3]

    Se (3)-stochastic flow matching for protein backbone generation

    Joey Bose, Tara Akhound-Sadegh, Guillaume Huguet, Kilian FATRAS, Jarrid Rector-Brooks, Cheng-Hao Liu, Andrei Cristian Nica, Maksym Korablyov, Michael M Bronstein, and Alexander Tong. Se (3)-stochastic flow matching for protein backbone generation. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  4. [4]

    Geometric and Physical Quantities Improve E (3) Equivariant Message Passing

    Johannes Brandstetter, Rob Hesselink, Elise van der Pol, Erik J Bekkers, and Max Welling. Geometric and Physical Quantities Improve E (3) Equivariant Message Passing . In International Conference on Learning Representations, 2022

    work page 2022

  5. [5]

    Geometric Algebra Transformer

    Johann Brehmer, Pim de Haan, S \"o nke Behrends, and Taco Cohen. Geometric Algebra Transformer . In Neural Information Processing Systems, 2023

    work page 2023

  6. [6]

    Equivariant neural diffusion for molecule generation

    Fran c ois Cornet, Grigory Bartosh, Mikkel N Schmidt, and Christian A Naesseth. Equivariant neural diffusion for molecule generation. In 38th Conference on Neural Information Processing Systems, 2024

    work page 2024

  7. [7]

    Mixed continuous and categorical flow matching for 3d de novo molecule generation

    Ian Dunn and David Ryan Koes. Mixed continuous and categorical flow matching for 3d de novo molecule generation. ArXiv, 2024

    work page 2024

  8. [8]

    E (n) equivariant normalizing flows

    Victor Garcia Satorras, Emiel Hoogeboom, Fabian Fuchs, Ingmar Posner, and Max Welling. E (n) equivariant normalizing flows. Advances in Neural Information Processing Systems, 2021

    work page 2021

Show all 35 references
  1. [9]

    Symmetry-adapted generation of 3d point sets for the targeted discovery of molecules

    Niklas Gebauer, Michael Gastegger, and Kristof Sch \"u tt. Symmetry-adapted generation of 3d point sets for the targeted discovery of molecules. Advances in neural information processing systems, 2019

    2019

  2. [10]

    Denoising diffusion probabilistic models

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

    2020

  3. [11]

    Equivariant diffusion for molecule generation in 3d

    Emiel Hoogeboom, V ctor Garcia Satorras, Cl \'e ment Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3d. In International conference on machine learning, 2022

    2022

  4. [12]

    Mudiff: Unified diffusion for complete molecule generation

    Chenqing Hua, Sitao Luan, Minkai Xu, Zhitao Ying, Jie Fu, Stefano Ermon, and Doina Precup. Mudiff: Unified diffusion for complete molecule generation. In Learning on Graphs Conference, pp.\ 33--1. PMLR, 2024

    2024

  5. [13]

    Learning joint 2d & 3d diffusion models for complete molecule generation, 2023

    Han Huang, Leilei Sun, Bowen Du, and Weifeng Lv. Learning joint 2d & 3d diffusion models for complete molecule generation, 2023

    2023

  6. [14]

    Elucidating the design space of diffusion-based generative models, 2022

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models, 2022

    2022

  7. [15]

    Clifford Group Equivariant Simplicial Message Passing Networks

    Cong Liu, David Ruhe, Floor Eijkelboom, and Patrick Forr \'e . Clifford Group Equivariant Simplicial Message Passing Networks . In The Twelfth International Conference on Learning Representations, 2024 a

    2024

  8. [16]

    Multivector neurons: Better and faster o(n)-equivariant clifford GNN s

    Cong Liu, David Ruhe, and Patrick Forr \'e . Multivector neurons: Better and faster o(n)-equivariant clifford GNN s. In ICML 2024 Workshop on Geometry-grounded Representation Learning and Generative Modeling, 2024 b

    2024

  9. [17]

    Quantum chemistry structures and properties of 134 kilo molecules

    Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 0 (1), 2014

    2014

  10. [18]

    Clifford Group Equivariant Neural Networks

    David Ruhe, Johannes Brandstetter, and Patrick Forr\'e. Clifford Group Equivariant Neural Networks . In Neural Information Processing Systems, 2023 a

    2023

  11. [19]

    Gupta, Steven De Keninck, Max Welling, and Johannes Brandstetter

    David Ruhe, Jayesh K. Gupta, Steven De Keninck, Max Welling, and Johannes Brandstetter. Geometric Clifford Algebra Networks . In International Conference on Machine Learning, 2023 b

    2023

  12. [20]

    E (n) equivariant graph neural networks

    V ctor Garcia Satorras, Emiel Hoogeboom, and Max Welling. E (n) equivariant graph neural networks. In International conference on machine learning, 2021

    2021

  13. [21]

    Gregor N. C. Simm, Robert Pinsler, G \'a bor Cs \'a nyi, and Jos \'e Miguel Hern \'a ndez-Lobato. Symmetry-aware actor-critic for 3d molecular design. In International Conference on Learning Representations, 2021

    2021

  14. [22]

    Graphvae: Towards generation of small graphs using variational autoencoders

    Martin Simonovsky and Nikos Komodakis. Graphvae: Towards generation of small graphs using variational autoencoders. In International Conference on Artificial Neural Networks, 2018

    2018

  15. [23]

    Equivariant flow matching with hybrid probability transport for 3d molecule generation

    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, 36, 2024

    2024

  16. [24]

    Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics

    Jonas Spinner, Victor Bres \'o , Pim de Haan, Tilman Plehn, Jesse Thaler, and Johann Brehmer. Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics . 2024

    2024

  17. [25]

    On the utility of equivariance and symmetry breaking in deep learning architectures on point clouds, 2025

    Sharvaree Vadgama, Mohammad Mohaiminul Islam, Domas Buracus, Christian Shewmake, and Erik Bekkers. On the utility of equivariance and symmetry breaking in deep learning architectures on point clouds, 2025

    2025

  18. [26]

    Midi: Mixed graph and 3d denoising diffusion for molecule generation

    Clement Vignac, Nagham Osman, Laura Toni, and Pascal Frossard. Midi: Mixed graph and 3d denoising diffusion for molecule generation. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, 2023

    2023

  19. [27]

    Watson, David Juergens, Nathaniel Bennett, Brian Trippe, Jaekyung Yim, Helen Eisenach, William Ahern, Andrew Borst, Robert Ragotte, Lukas Milles, Basile I

    James L. Watson, David Juergens, Nathaniel Bennett, Brian Trippe, Jaekyung Yim, Helen Eisenach, William Ahern, Andrew Borst, Robert Ragotte, Lukas Milles, Basile I. M. Wicky, Nikita Hanikel, Samuel J. Pellock, Alexis Courbet, William Sheffler, Jianyi Wang, Keenan Bett, Asim Be...

    2023

  20. [28]

    Geodiff: A geometric diffusion model for molecular conformation generation

    Minkai Xu, Lantao Yu, Yang Song, Chence Shi, Stefano Ermon, and Jian Tang. Geodiff: A geometric diffusion model for molecular conformation generation. In International Conference on Learning Representations, 2022

    2022

  21. [29]

    Geometric latent diffusion models for 3d molecule generation

    Minkai Xu, Alexander S Powers, Ron O Dror, Stefano Ermon, and Jure Leskovec. Geometric latent diffusion models for 3d molecule generation. In International Conference on Machine Learning, 2023

    2023

  22. [30]

    Se (3) diffusion model with application to protein backbone generation

    Jason Yim, Brian L Trippe, Valentin De Bortoli, Emile Mathieu, Arnaud Doucet, Regina Barzilay, and Tommi Jaakkola. Se (3) diffusion model with application to protein backbone generation. In International Conference on Machine Learning, 2023

    2023

  23. [31]

    Clifford-Steerable Convolutional Neural Networks

    Maksim Zhdanov, David Ruhe, Maurice Weiler, Ana Lucic, Johannes Brandstetter, and Patrick Forr\'e. Clifford-Steerable Convolutional Neural Networks . In International Conference on Machine Learning, 2024

    2024

  24. [32]

    write newline

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

  25. [33]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  26. [34]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  27. [35]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.