Efficient Approximations of Complete Interatomic Potentials for Crystal Property Prediction

Yuchao Lin,Keqiang Yan,Youzhi Luo,Yi Liu,Xiaoning Qian,Shuiwang Ji

2023-11-07

Abstract:We study property prediction for crystal materials. A crystal structure consists of a minimal unit cell that is repeated infinitely in 3D space. How to accurately represent such repetitive structures in machine learning models remains unresolved. Current methods construct graphs by establishing edges only between nearby nodes, thereby failing to faithfully capture infinite repeating patterns and distant interatomic interactions. In this work, we propose several innovations to overcome these limitations. First, we propose to model physics-principled interatomic potentials directly instead of only using distances as in many existing methods. These potentials include the Coulomb potential, London dispersion potential, and Pauli repulsion potential. Second, we model the complete set of potentials among all atoms, instead of only between nearby atoms as in existing methods. This is enabled by our approximations of infinite potential summations, where we extend the Ewald summation for several potential series approximations with provable error bounds. Finally, we propose to incorporate our computations of complete interatomic potentials into message passing neural networks for representation learning. We perform experiments on the JARVIS and Materials Project benchmarks for evaluation. Results show that the use of interatomic potentials and complete interatomic potentials leads to consistent performance improvements with reasonable computational costs. Our code is publicly available as part of the AIRS library (<a class="link-external link-https" href="https://github.com/divelab/AIRS/tree/main/OpenMat/PotNet" rel="external noopener nofollow">this https URL</a>).

Chemical Physics,Machine Learning

What problem does this paper attempt to address?

The paper aims to address the key challenges in predicting the properties of crystalline materials, especially in accurately capturing the interatomic interactions within an infinite range. Existing Graph Neural Network (GNN) methods, when dealing with crystal structures, only construct graphs by establishing edges between locally adjacent nodes, which leads to an inability to precisely capture the long-range interatomic interactions caused by the infinite repetition of the unit cell in three-dimensional space. The paper introduces PotNet, a novel graph-based deep learning approach with the following innovative aspects: 1. It directly uses interatomic potential energies guided by physical principles as edge features, instead of merely using distance information. These potential energies include Coulombic, London dispersion, and Pauli repulsion potentials. 2. PotNet considers the complete set of potential energies between all atoms, not just the potential energies between neighboring atoms. This is achieved through an approximation of infinite potential energy summation, which includes an algorithm with provable error bounds. 3. It integrates the computed interatomic potential energies and their complete set into a message-passing neural network for representation learning. The paper validates the effectiveness of PotNet through experiments on the JARVIS and Materials Project benchmark datasets, showing that the use of interatomic potential energies and the complete set of such energies significantly improves performance at a reasonable computational cost. In summary, the problem that the paper attempts to solve is the accurate representation and utilization of interatomic interactions within an infinite range for the prediction of crystalline material properties, with the aim of enhancing prediction accuracy.

Efficient Approximations of Complete Interatomic Potentials for Crystal Property Prediction

Interpretable Ensemble Learning for Materials Property Prediction with Classical Interatomic Potentials: Carbon as an Example

Improving Material Property Prediction by Leveraging the Large-Scale Computational Database and Deep Learning

Leveraging Large-scale Computational Database and Deep Learning for Accurate Prediction of Material Properties

Accelerating material property prediction using generically complete isometry invariants

Material Property Prediction Using Graphs Based on Generically Complete Isometry Invariants

Lightweight equivariant interaction graph neural network for accurate and efficient interatomic potential and force predictions

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

A Crystal-Specific Pre-Training Framework for Crystal Material Property Prediction

Predicting the Volumes of Crystals

Recent advances in machine learning interatomic potentials for cross-scale computational simulation of materials

Utilizing Wyckoff Sites to Construct Machine-Learning-Driven Interatomic Potentials for Crystalline Materials: A Case Study on α-Alumina

Machine-learning interatomic potentials for materials science

Enhancing crystal structure prediction by combining computational and experimental data via graph networks

Machine-learning correction to density-functional crystal structure optimization

Topological representations of crystalline compounds for the machine-learning prediction of materials properties

Accurate Property Prediction with Interpretable Machine Learning Model for Small Datasets Via Transformed Atom Vector

How to represent crystal structures for machine learning: Towards fast prediction of electronic properties

LLM-Prop: Predicting Physical And Electronic Properties Of Crystalline Solids From Their Text Descriptions

Fundamental Microscopic Properties as Predictors of Large-Scale Quantities of Interest: Validation through Grain Boundary Energy Trends