Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

Park, Cheol Woo; Kornbluth, Mordechai; Vandermause, Jonathan; Wolverton, Chris; Kozinsky, Boris; Mailoa, Jonathan P.

doi:10.1038/s41524-021-00543-3

Cited by 106 publications

(87 citation statements)

References 49 publications

(73 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One such method involves using ML, which has emerged as a powerful tool in the materials science community and enabled researchers to devise highly accurate models that describe increasingly complex trends [25,26,27,28]. Models have already been developed for a diverse range of applications, from predicting material properties [29,30,31,32], to DFT energetics [33,32], and even force fields for molecular dynamics [34]. ML models have also been used to predict materials synthesizability.…”

Section: Correlatedmentioning

confidence: 99%

Machine Learned Synthesizability Predictions Aided by Density Functional Theory

Lee¹,

Sarker²,

Saal³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

A grand challenge of materials science is the computational prediction of synthesis pathways for novel compounds. Data-driven and machine learning (ML) approaches have made significant progress in addressing a portion of this problem, namely, predicting whether a compound is synthesizable or not. However, some recent attempts to do so have not incorporated energetic or phase stability information. Here, we combine thermodynamic stability calculated using density functional theory (DFT) with composition-based features to train a ML model that predicts a material's synthesizability. We focus on compositions with ABC stoichiometry and predict synthesizability in the half-heusler (HH) structure. Our model is trained on experimentally reported ABC compositions, and achieves a cross-validated precision of 0.83 and recall of 0.85. Our model shows improvement in identifying compositions that do not form HH when compared to a similar HH synthesizability model from a previous study. Our model identifies 163 synthesizable candidates out of 4141 unreported ABC compositions. More notably, 38 stable compositions are predicted unsynthesizable while 103 unstable compositions are predicted synthesizable; these findings otherwise cannot be made using DFT stability alone. This study presents a new approach for accurately predicting synthesizability, and identifies newly synthesizable candidate HH compositions for further experimental exploration.

show abstract

Section: Correlatedmentioning

confidence: 99%

Machine Learned Synthesizability Predictions Aided by Density Functional Theory

Lee¹,

Sarker²,

Saal³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…While the above works are mainly based on NNs, there have also been development of graph neural network force field (GNNFF) framework [117,118] that bypasses both computational bottlenecks. GNNFF can predict atomic forces directly using automatically extracted structural features that are not only translationally-invariant, but rotationally-covariant to the coordinate space of the atomic positions.…”

Section: Force Field Developmentmentioning

confidence: 99%

Recent Advances and Applications of Deep Learning Methods in Materials Science

Choudhary,

DeCost,

Chen

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…GemNets (Klicpera et al, 2021) are related to DimeNets, and incorporate angle information on quadruplets of points within a graph message passing scheme using an architecture optimized for molecular datasets. GNNFF (Park et al, 2021) generates rotation-covariant results by computing a weighted sum of modulated input vectors based on a graph message passing scheme.…”

Section: Related Workmentioning

confidence: 99%

Geometric Algebra Attention Networks for Small Point Clouds

Spellings¹

2021

Preprint

View full text Add to dashboard Cite

Much of the success of deep learning is drawn from building architectures that properly respect underlying symmetry and structure in the data on which they operate-a set of considerations that have been united under the banner of geometric deep learning. Often problems in the physical sciences deal with relatively small sets of points in two-or three-dimensional space wherein translation, rotation, and permutation equivariance are important or even vital for models to be useful in practice. In this work, we present rotation-and permutation-equivariant architectures for deep learning on these small point clouds, composed of a set of products of terms from the geometric algebra and reductions over those products using an attention mechanism. The geometric algebra provides valuable mathematical structure by which to combine vector, scalar, and other types of geometric inputs in a systematic way to account for rotation invariance or covariance, while attention yields a powerful way to impose permutation equivariance. We demonstrate the usefulness of these architectures by training models to solve sample problems relevant to physics, chemistry, and biology.

show abstract

Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

Cited by 106 publications

References 49 publications

Machine Learned Synthesizability Predictions Aided by Density Functional Theory

Machine Learned Synthesizability Predictions Aided by Density Functional Theory

Recent Advances and Applications of Deep Learning Methods in Materials Science

Geometric Algebra Attention Networks for Small Point Clouds

Contact Info

Product

Resources

About