CoChat

Constructing Collective Variables Using Invariant Learned Representations

Open access
Šípka, Martin, Erlebach, Andreas, Grajciar, Lukáš · 2023 · Journal of Chemical Theory and Computation

The DOI matches its Crossref record; 50 of the 50 references it cites were found in a registry or an index.

Abstract

On the time scales accessible to atomistic numerical modeling, chemical reactions are considered rare events. Therefore, the atomistic simulations are commonly biased along a low-dimensional representation of a chemical reaction in an atomic structure space, i.e., along the collective variables. However, suitable collective variables are often complicated to guess a priori . We propose a novel method of collective variable discovery based on dimensionality reduction of the atomic representation vectors. These linear-scaling and invariant representations can be either fixed (untrained) or learned by supervised training of the end-to-end machine learning potential. The learned representations are expected to reflect not only the structural but also the energetic features of the system that are transferable to all of the reactive transformation covered by the machine learning potential. We demonstrate our approach on four high-barrier reactions ranging from a simple gas-phase hydrogen jump reaction to complex reactions in periodic models of industrially relevant heterogeneous catalysts. High data efficiency, automatized feature extraction, favorable scaling, and retention of inherent invariances are all properties that are expected to enable fast and largely automatic construction of suitable collective variables even in highly complex reactive scenarios such as reactive/catalytic transformations at solid–liquid interfaces.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1021/acs.jctc.2c00729

Cites

50 of 50 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Training of our collective variables.
Figure 1 Training of our collective variables.CC BY-NC-ND · PubMed Central
Figure 2 Single neural network for energy and collective variable prediction.
Figure 2 Single neural network for energy and collective variable prediction.CC BY-NC-ND · PubMed Central
Figure 3 Projections of our collective variables to important distances. Red points represent reactants, blue points represent products, and green ones are slow-growth benchmark dynamics obtained using CVd given in eq 1. In the bottom-right panel, we concatenate all of the datasets, putting reactant data first, slow-growth data in the middle, and product data last, while keeping the order of simulation steps. We then plot the evolution of a collective variable on these simulated data points. Only blue and red points were used for training.
Figure 3 Projections of our collective variables to important distances. Red points represent reactants, blue points represent products, and green ones are slow-growth benchmark dynamics obtained using CVd given in eq 1. In the bottom-right panel, we concatenate all of the datasets, putting reactant data first, slow-growth data in the middle, and product data last, while keeping the order of simulation steps. We then plot the evolution of a collective variable on these simulated data points. Only blue and red points were used for training.CC BY-NC-ND · PubMed Central
Figure 4 Keto–enol transition: evolution of the trained collective variable driven using the steered dynamics (moving restraint technique). Positive CV values represent reactants, while negative values represent products. The value of the reference expert-chosen CV is also plotted (see the vertical axis on the right).
Figure 4 Keto–enol transition: evolution of the trained collective variable driven using the steered dynamics (moving restraint technique). Positive CV values represent reactants, while negative values represent products. The value of the reference expert-chosen CV is also plotted (see the vertical axis on the right).CC BY-NC-ND · PubMed Central
Figure 5 Illustration of the hydrogen jump in a chabazite unit cell (with a single aluminum in the unit cell).
Figure 5 Illustration of the hydrogen jump in a chabazite unit cell (with a single aluminum in the unit cell).CC BY-NC-ND · PubMed Central
Figure 6 CV evolution and approximate free energy surface for the hydrogen jump in chabazite evaluated via metadynamics simulations. (a) Evolution of the expert-chosen CV during the simulation. (b) Evolution of a pretrained NNP-based collective variable. Diffusive behavior takes place, suggesting that the metadynamics converged. (c) Evolution of the ACSF-based collective variable. Diffusive behavior is not achieved, indicating that the collective variable can be expected to be of lower quality than in the case of pretrained NNP-based CVs. (d) Predicted FES profiles from metadynamics with all three CVs.
Figure 6 CV evolution and approximate free energy surface for the hydrogen jump in chabazite evaluated via metadynamics simulations. (a) Evolution of the expert-chosen CV during the simulation. (b) Evolution of a pretrained NNP-based collective variable. Diffusive behavior takes place, suggesting that the metadynamics converged. (c) Evolution of the ACSF-based collective variable. Diffusive behavior is not achieved, indicating that the collective variable can be expected to be of lower quality than in the case of pretrained NNP-based CVs. (d) Predicted FES profiles from metadynamics with all three CVs.CC BY-NC-ND · PubMed Central
Figure 7 Two-dimensional CV example for the hydrogen jump in chabazite using NNP-based CVs. (a) Free energy surface in two dimensions. (b) Evolution of trained collective variables for the training data. The reactant structures are shown in red, the product structures in blue, and the steered dynamics between the two in green.
Figure 7 Two-dimensional CV example for the hydrogen jump in chabazite using NNP-based CVs. (a) Free energy surface in two dimensions. (b) Evolution of trained collective variables for the training data. The reactant structures are shown in red, the product structures in blue, and the steered dynamics between the two in green.CC BY-NC-ND · PubMed Central
Figure 8 Chabazite hydrolysis: evolution of the trained collective variable using steered dynamics. Larger CV values represent products, while smaller values correspond to reactants. Representative structures of reactant, product, and candidate TS structures are depicted in the insets.
Figure 8 Chabazite hydrolysis: evolution of the trained collective variable using steered dynamics. Larger CV values represent products, while smaller values correspond to reactants. Representative structures of reactant, product, and candidate TS structures are depicted in the insets.CC BY-NC-ND · PubMed Central
Figure 9 CV evolution and approximate free energy surface for the geminal complex formation. Notice the our machine-learned CV (ML-CV) is inverted with respect to the reference one. (a) Evolution of the ML-CV based on the pretrained model. (b) Evolution of the reference expert-chosen CV. (c) Predicted FES profiles from metadynamics with ML-CV and the reference CV.
Figure 9 CV evolution and approximate free energy surface for the geminal complex formation. Notice the our machine-learned CV (ML-CV) is inverted with respect to the reference one. (a) Evolution of the ML-CV based on the pretrained model. (b) Evolution of the reference expert-chosen CV. (c) Predicted FES profiles from metadynamics with ML-CV and the reference CV.CC BY-NC-ND · PubMed Central
Figure 10 Projections of our collective variables onto important interatomic distances and the expert-chosen CV: (a) collective variable based on a pretrained model; (b) collective variable based on the ACSF representations. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics using CVd from eq 3. Training was done with only blue and red points. The bottom-right panel in each plot shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.
Figure 10 Projections of our collective variables onto important interatomic distances and the expert-chosen CV: (a) collective variable based on a pretrained model; (b) collective variable based on the ACSF representations. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics using CVd from eq 3. Training was done with only blue and red points. The bottom-right panel in each plot shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.CC BY-NC-ND · PubMed Central
Figure 11 Projections of our collective variables onto important interatomic distances and the expert-chosen CV. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics using CVd from eq 5. Training was done with only blue and red points. The bottom-right panel shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.
Figure 11 Projections of our collective variables onto important interatomic distances and the expert-chosen CV. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics using CVd from eq 5. Training was done with only blue and red points. The bottom-right panel shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.CC BY-NC-ND · PubMed Central
Figure 12 Projections of our collective variable based on the pretrained model onto important interatomic distances and the reference expert-chosen collective variable (see eq 4) describing the chabazite hydrolysis test reaction. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics. Training was done with only blue and red points. The bottom-right panel shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.
Figure 12 Projections of our collective variable based on the pretrained model onto important interatomic distances and the reference expert-chosen collective variable (see eq 4) describing the chabazite hydrolysis test reaction. Red points represent reactants, blue points represent products, and green ones are data points from the slow-growth dynamics. Training was done with only blue and red points. The bottom-right panel shows the evolution of a collective variable on a concatenated dataset of equilibrium and slow-growth simulations.CC BY-NC-ND · PubMed Central
Figure 13 Reactant in the chabazite hydrolysis reaction. The atomic indices are drawn to better understand the collective variable used.
Figure 13 Reactant in the chabazite hydrolysis reaction. The atomic indices are drawn to better understand the collective variable used.CC BY-NC-ND · PubMed Central
Figure 14 Reactant in our geminal complex formation reaction. The atomic indices are drawn to better understand the collective variable used.
Figure 14 Reactant in our geminal complex formation reaction. The atomic indices are drawn to better understand the collective variable used.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 50 references

Research works printed but not linked: 5

  1. Duvenaud D., 2015, 28th Conference on Neural Information Processing Systems (NIPS 2015), 2224
  2. Schütt K., 2017, 31st Conference on Neural Information Processing Systems (NIPS 2017)
  3. Schütt K. T., 2021, Proc. Mach. Learn. Res., 139, 9377
  4. Huang H., 2018, 32nd Conference on Neural Information Processing Systems (NeurIPS 2018)
  5. Gebauer N. W. A., 2019, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), 7566

Cited by

See all 21

Metadata

Authors
Šípka, M., Erlebach, A., & Grajciar, L.
Affiliations
Charles University (CZ)
Corresponding author
  • Šípka, Martin
  • Grajciar, Lukáš
Venue
Journal of Chemical Theory and Computation
Published
2023-01-25
Volume
19
Issue
3
Pages
887-901
ISSN
  • 1549-9618 (ISSN-L)
  • 1549-9626
OpenAlex
W4318021309
Cited by
21
References
49
Citations by year
2026: 3 · 2025: 3 · 2024: 13 · 2023: 2
FWCI
1.55
Type
journal article
Language
en
Open access
Locations
  • Journal of Chemical Theory and Computation (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
Funders
  • Ministerstvo ?kolství, Mláde?e a Telovýchovy (90140)
  • Grantová Agentura ?eské Republiky (20-26767Y)
  • Univerzita Karlova v Praze (CZ.02.1.01/0.0/0.0/15 003/0000417, PRIMUS/20/SCI/004)
Topics
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Advanced Chemical Physics Studies
  • Physical Sciences › Chemistry › Physical and Theoretical Chemistry › Advanced Physical and Chemical Molecular Interactions
Keywords
collective variables, invariant representations, learned representations, dimensionality reduction, rare events, machine learning interatomic potentials, atomistic simulation, heterogeneous catalysis, periodic systems, solid-liquid interface, data efficiency

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Constructing Collective Variables Using Invariant Learned Representations. https://app.cochat.ai/papers/10.1021/acs.jctc.2c00729
The paper
Šípka, M., Erlebach, A., & Grajciar, L. (2023). Constructing Collective Variables Using Invariant Learned Representations. Journal of Chemical Theory and Computation. https://doi.org/10.1021/acs.jctc.2c00729

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.