Jadwal Sholat

Memuat jadwal sholatโ€ฆ

Fisika & Matematika editorial

Open AccessOA2024

Machine Learning Assisted Crystal Structure Prediction Made Simple

This review explores how machine learning, including graph neural networks, force fields, and generative models, accelerates crystal structure prediction and addresses bottlenecks in conventional methods.
Chuan-Nan Li; Han-Pu Liang; Bai-Qing Zhao; Suhuai Wei; Xie Zhangยท Journal of Materials Informaticsยท 2024ยท DOI 10.20517/jmi.2024.18

The core problem

Crystal structure prediction (CSP) is fundamental in condensed matter physics and materials science, impacting theoretical research, new material discovery, and technological advancement. Traditional trial-and-error experimental synthesis is time-consuming, labor-intensive, and inadequate for the growing demand for new materials. Machine learning (ML) methods have recently emerged as powerful tools to significantly boost CSP. This review provides a comprehensive overview of ML models applied in CSP, covering general steps, bottlenecks, crystal structure representation, and specific ML techniques such as graph neural networks (GNNs), ML force fields, and generative models (VAEs, GANs, diffusion models). The authors also discuss remaining challenges in ML-assisted CSP.

Innovation

The review systematically examines the integration of ML into CSP. It first outlines the general steps for CSP and identifies bottlenecks in conventional methods, such as the vast search space and high computational cost. The representation of crystal structures is discussed, emphasizing how ML models can effectively encode structural information. The authors then illustrate the workflow of ML-assisted CSP, where ML models are trained on existing data to predict stable structures or guide search algorithms. Key ML approaches are reviewed: GNNs for learning from graph representations of crystals, ML force fields for accurate and efficient energy calculations, and generative models for sampling novel structures. The review synthesizes findings from various studies to demonstrate the effectiveness of these methods.
Introduction
Crystal structure prediction (CSP) is fundamental in condensed matter physics and materials science, impacting theoretical research, new material discovery, and technological advancement. Traditional trial-and-error experimental synthesis is time-consuming, labor-intensive, and inadequate for the growing demand for new materials. Machine learning (ML) methods have recently emerged as powerful tools to significantly boost CSP. This review provides a comprehensive overview of ML models applied in CSP, covering general steps, bottlenecks, crystal structure representation, and specific ML techniques such as graph neural networks (GNNs), ML force fields, and generative models (VAEs, GANs, diffusion models). The authors also discuss remaining challenges in ML-assisted CSP.

Why it matters

The review discusses the remaining challenges in ML-assisted CSP, such as data scarcity, model interpretability, and the need for robust representations. It emphasizes that while ML models have advanced CSP, integrating them with physical constraints and ensuring generalizability across different material classes remain open problems. The authors suggest future directions, including the development of more sophisticated generative models and the combination of ML with high-throughput experiments. Overall, the review concludes that ML-assisted CSP holds great potential to revolutionize materials discovery, but further research is needed to overcome current limitations and fully realize its benefits.

Who should read this

Pembaca Research Digest dan praktisi rumpun ini.

Opening member contentโ€ฆ