A universal graph deep learning interatomic potential for the periodic table (2022-02-05T00:00:00.000000Z)

TL;DR

A machine learning-accelerated pathway to the discovery of synthesizable materials with exceptional properties is demonstrated, designed and trained on diverse crystal data comprising 89 elements, enabling materials discovery across a vast chemical space without retraining.

Abstract

Interatomic potentials (IAPs), which describe the potential energy surface of atoms, are a fundamental input for atomistic simulations. However, existing IAPs are either fitted to narrow chemistries or too inaccurate for general applications. Here we report a universal IAP for materials based on graph neural networks with three-body interactions (M3GNet). The M3GNet IAP was trained on the massive database of structural relaxations performed by the Materials Project over the past ten years and has broad applications in structural relaxation, dynamic simulations and property prediction of materials across diverse chemical spaces. About 1.8 million materials from a screening of 31 million hypothetical crystal structures were identified to be potentially stable against existing Materials Project crystals based on M3GNet energies. Of the top 2,000 materials with the lowest energies above the convex hull, 1,578 were verified to be stable using density functional theory calculations. These results demonstrate a machine learning-accelerated pathway to the discovery of synthesizable materials with exceptional properties. A machine learning interatomic potential model is designed and trained on diverse crystal data comprising 89 elements, enabling materials discovery across a vast chemical space without retraining.

Authors

Chi Chen

4 papers

S. Ong

6 papers

TL;DR

Abstract

Authors

References82 items

Crystal Diffusion Variational Autoencoder for Periodic Material Generation

NewtonNet: a Newtonian message passing network for deep learning of interatomic potentials and forces

Atomistic Line Graph Neural Network for improved materials property predictions

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

SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials

Bridging the gap between simulated and experimental ionic conductivities in lithium superionic conductors

Equivariant message passing for the prediction of tensorial properties and molecular spectra

AtomSets as a hierarchical transfer learning framework for small and large materials datasets

Robust model benchmarking and bias-imbalance in data-driven materials science: a case study on MODNet

Learning properties of ordered and disordered materials from multi-fidelity data

Crystal Structure Search with Random Relaxations Using Graph Networks

Benchmarking materials property prediction methods: the Matbench test set and Automatminer reference algorithm

Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet

structural relaxation

Directional Message Passing for Molecular Graphs

Machine learning enabled autonomous microstructural characterization in 3D samples

Embedded Atom Neural Network Potentials: Efficient and Accurate Machine Learning with a Physically Inspired Representation.

A Performance and Cost Assessment of Machine Learning Interatomic Potentials.

Lithium Chlorides and Bromides as Promising Solid-State Chemistries for Fast Ion Conductors with Good Electrochemical Stability.

The NOMAD laboratory: from data sharing to artificial intelligence

Machine Learning With Python

A novel approach to describe chemical environments in high-dimensional neural network potentials.

PhysNet: A Neural Network for Predicting Energies, Forces, Dipole Moments, and Partial Charges.

Library-Based LAMMPS Implementation of High-Dimensional Neural Network Potentials.

Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals

High-throughput density-functional perturbation theory phonons for inorganic materials

Towards exact molecular dynamics simulations with machine-learned force fields

Atomate: A high-level interface to generate, execute, and analyze computational materials science workflows

Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties.

Deep Potential Molecular Dynamics: a scalable model with the accuracy of quantum mechanics

The atomic simulation environment-a Python library for working with atoms.

SchNet: A continuous-filter convolutional neural network for modeling quantum interactions

Li3Y(PS4)2 and Li5PS4Cl2: New Lithium Superionic Conductors Predicted from Silver Thiophosphates using Efficiently Tiered Ab Initio Molecular Dynamics Simulations

Machine learning of accurate energy-conserving molecular force fields

Machine learning based interatomic potential for amorphous carbon

The thermodynamic scale of inorganic crystalline metastability

Quantum-chemical insights from deep tensor neural networks

Reproducibility in density functional theory calculations of solids

Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Potentials

The Open Quantum Materials Database (OQMD): assessing the accuracy of DFT formation energies

FireWorks: a dynamic workflow system designed for high‐throughput applications

First principles phonon calculations in materials science

Rational Composition Optimization of the Lithium-Rich Li3OCl1–xBrx Anti-Perovskite Superionic Conductors

AiiDA: Automated Interactive Infrastructure and Database for Computational Science

Charting the complete elastic properties of inorganic crystalline compounds

Adam: A Method for Stochastic Optimization

Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials

Commentary: The Materials Project: A materials genome approach to accelerating materials innovation

Ab initio study of the stabilities of and mechanism of superionic transport in lithium-rich antiperovskites

Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis

AFLOWLIB.ORG: A distributed materials properties repository from high-throughput ab initio calculations

CALYPSO: A method for crystal structure prediction

Scikit-learn: Machine Learning in Python

Gaussian approximation potentials: the accuracy of quantum mechanics, without the electrons.

Thermal transport properties of MgO and Nd2Zr2O7 pyrochlore by molecular dynamics simulation

Restoring the density-gradient expansion for exchange in solids and surfaces.

Matplotlib: A 2D Graphics Environment

Generalized neural-network representation of high-dimensional potential-energy surfaces.

USPEX - Evolutionary crystal structure prediction

Structural relaxation made simple.

Oxidation energies of transition metal oxides within the GGA+U framework

The Amber biomolecular simulation programs

The Inorganic Crystal Structure Database (ICSD)—Present and Future

Generalized Gradient Approximation Made Simple.

Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set

Ab initio molecular dynamics for liquid metals.

Fast parallel algorithms for short-range molecular dynamics

UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations

Band theory and Mott insulators: Hubbard U instead of Stoner I.

New empirical approach for the structure and energy of covalent systems.

AMBER: Assisted model building with energy refinement. A general program for modeling molecules and their interactions

Seaborn: Statistical Data Visualization

Pandas Development

Searching for Activation Functions

Statsmodels: Econometric and Statistical Modeling with Python

assisted model building with energy refinement