The Design Space of E(3)-Equivariant Atom-Centered Interatomic Potentials

Abstract

The rapid progress of machine learning interatomic potentials over the past couple of years produced a number of new architectures. Particularly notable among these are the Atomic Cluster Expansion (ACE), which unified many of the earlier ideas around atom density-based descriptors, and Neural Equivariant Interatomic Potentials (NequIP), a message passing neural network with equivariant features that showed state of the art accuracy. In this work, we construct a mathematical framework that unifies these models: ACE is generalised so that it can be recast as one layer of a multi-layer architecture. From another point of view, the linearised version of NequIP is understood as a particular sparsification of a much larger polynomial model. Our framework also provides a practical tool for systematically probing different choices in the unified design space. We demonstrate this by an ablation study of NequIP via a set of experiments looking at in- and out-of-domain accuracy and smooth extrapolation very far from the training data, and shed some light on which design choices are critical for achieving high accuracy. Finally, we present BOTNet (Body-Ordered-Tensor-Network), a much-simplified version of NequIP, which has an interpretable architecture and maintains accuracy on benchmark datasets.

Cited by in corpus (22)

• Evaluation of the MACE Force Field Architecture: from Medicinal Chemistry to Materials Science
• MLatom 3: Platform for machine learning-enhanced computational chemistry simulations and workflows
• Accelerating CALPHAD-based Phase Diagram Predictions in Complex Alloys Using Universal Machine Learning Potentials: Opportunities and Challenges
• Deep Coarse-grained Potentials via Relative Entropy Minimization
• Force-Field-Enhanced Neural Network Interactions: from Local Equivariant Embedding to Atom-in-Molecule properties and long-range effects
• E(n)-Equivariant Cartesian Tensor Passing Potential
• On the design space between molecular mechanics and machine learning force fields
• Artificial Intelligence in Materials Science and Engineering: Current Landscape, Key Challenges, and Future Trajectorie
• Completeness of Atomic Structure Representations
• Neural Network Potential with Multi-Resolution Approach Enables Accurate Prediction of Reaction Free Energies in Solution
• Data efficiency and extrapolation trends in neural network interatomic potentials
• Hybrid Classical/Machine-Learning Force Fields for the Accurate Description of Molecular Condensed-Phase Systems
• Stress and heat flux via automatic differentiation
• Probing the effects of broken symmetries in machine learning
• Direct imaging of carbohydrate stereochemistry
• BIP: Boost Invariant Polynomials for Efficient Jet Tagging
• Prediction rigidities for data-driven chemistry
• Automated High-throughput Organic Crystal Structure Prediction via Population-based Sampling
• Machine learning based modeling of disordered elemental semiconductors: understanding the atomic structure of a-Si and a-C
• Scaling Graph Neural Networks to Large Proteins
• Unifying O(3) Equivariant Neural Networks Design with Tensor-Network Formalism
• Decoding lithium's subtle phase stability with a machine learning force field