Fast and Sample-Efficient Interatomic Neural Network Potentials for Molecules and Materials Based on Gaussian Moments

2021/09/20 by Viktor Zaverkin, David Holzmüller, Ingo Steinwart +1 · 1 voice · 1 citation
Chemistry · Computer Science · Materials Science · Mathematics · Physics and Astronomy · #Computational Drug Discovery Methods #Machine Learning in Materials Science #Various Chemistry Research Topics #physics.comp-ph #stat.ML

paper · pdf · doi:10.1021/acs.jctc.1c00527

Manuscript accepted for publication in J. Chem. Theory Comput.; Code published at https://gitlab.com/zaverkin_v/gmnn

arxiv created 2021/09/20 · openalex publication_date 2021/09/29 · arxiv updated 2021/10/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01

Abstract

Artificial neural networks (NNs) are one of the most frequently used machine learning approaches to construct interatomic potentials and enable efficient large-scale atomistic simulations with almost ab initio accuracy. However, the simultaneous training of NNs on energies and forces, which are a prerequisite for, e.g., molecular dynamics simulations, can be demanding. In this work, we present an improved NN architecture based on the previous GM-NN model [V. Zaverkin and J. Kästner, J. Chem. Theory Comput. 16, 5410-5421 (2020)], which shows an improved prediction accuracy and considerably reduced training times. Moreover, we extend the applicability of Gaussian moment-based interatomic potentials to periodic systems and demonstrate the overall excellent transferability and robustness of the respective models. The fast training by the improved methodology is a pre-requisite for training-heavy workflows such as active learning or learning-on-the-fly.

Citations

• Generalized Gradient Approximation Made Simple
• CHARMM general force field: A force field for drug‐like molecules compatible with the CHARMM all‐atom additive biological force fields
• Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning
• Merck molecular force field. I. Basis, form, scope, parameterization, and performance of MMFF94
• Quantum-chemical insights from deep tensor neural networks
• The atomic simulation environment—a Python library for working with atoms
• Less is more: Sampling chemical space with active learning
• Gaussian Error Linear Units (GELUs)
• Wide Neural Networks of Any Depth Evolve as Linear Models Under Gradient Descent
• Sigmoid-weighted linear units for neural network function approximation in reinforcement learning
• Equivariant message passing for the prediction of tensorial properties and molecular spectra

Cited by

• Breaking scaling relations with inverse catalysts: a machine learning exploration of trends in CO₂ hydrogenation energy barriers

Discussions

• We extend our previous framework for batch active learning from fully-connected NNs to atomistic NNs and test it on different chemical data sets against conventional methods. We use GM-NN (https://doi [bsky, 0 points, 1 comments]

Related

• SG-NNP: Species-separated Gaussian Neural Network Potential with Linear Elemental Scaling and Optimized Dimensions for Multi-component Materials
• Automated training of neural-network interatomic potentials
• INN-FF: A Scalable and Efficient Machine Learning Potential for Molecular Dynamics
• Interatomic Potential in a Simple Dense Neural Network Representation
• Machine-learning interatomic potentials from a users perspective: A comparison of accuracy, speed and data efficiency