Mardochée Réveil, PhD
Back to Publications

Scalable Crystal Structure Relaxation Using an Iteration-Free Deep Generative Model with Uncertainty Quantification

Ziduo Yang, Yi-Ming Zhao, Xian Wang, Xiaoqing Liu, Xiuying Zhang, Yifan Li, Qiujie Lv, Calvin Yu-Chian Chen, Lei Shen
4/1/2024

Abstract

In computational molecular and materials science, determining equilibrium structures is the crucial first step for accurate subsequent property calculations. However, the recent discovery of millions of new crystals and complex twisted structures has challenged traditional computational methods, both ab initio and machine-learning-based, due to their computationally intensive iterative processes. To address these scalability issues, here we introduce DeepRelax, a deep generative model capable of performing geometric crystal structure relaxation rapidly and without iterations. DeepRelax learns the equilibrium structural distribution, enabling it to predict relaxed structures directly from their unrelaxed ones. The ability to perform structural relaxation at the millisecond level per structure, combined with the scalability of parallel processing, makes DeepRelax particularly useful for large-scale virtual screening. We demonstrate DeepRelax's reliability and robustness by applying it to five diverse databases, including oxides, Materials Project, two-dimensional materials, van der Waals crystals, and crystals with point defects. DeepRelax consistently shows high accuracy and efficiency, validated by density functional theory calculations. Finally, we enhance its trustworthiness by integrating uncertainty quantification. This work significantly accelerates computational workflows, offering a robust and trustworthy machine-learning method for material discovery and advancing the application of AI for science. Code for DeepRelax is available at https://github.com/Shen-Group/DeepRelax.

AI-Generated Overview

  • Research Focus: The study introduces DeepRelax, a deep generative model designed to perform scalable and iteration-free geometric crystal structure relaxation, overcoming limitations of traditional computational methods in materials science.

  • Methodology: DeepRelax employs a periodicity-aware equivariant graph neural network (PaEGNN) to predict relaxation quantities from unrelaxed structures, followed by a numerical Euclidean distance geometry solver for final structure determination, while also incorporating uncertainty quantification methods.

  • Results: DeepRelax demonstrates high accuracy and efficiency in predicting relaxed structures across various databases (including oxides and 2D materials), significantly outperforming state-of-the-art models like Cryslator in terms of speed (approximately 100 times faster) and predictive accuracy.

  • Key Contributions: The model combines a novel architecture for capturing crystal periodicity and uncertainty quantification, making it a powerful tool for directly predicting crystal relaxation without iterative processes traditionally used in density functional theory (DFT).

  • Significance: This research addresses critical computational scalability issues in materials discovery and design, providing a robust and efficient framework that can accelerate the identification of energetically favorable structures from large material databases.

  • Broader Applications: DeepRelax can be applied in various fields, such as chemical reactions, drug design, and materials science, enabling high-throughput virtual screening of new materials and fostering advancements in AI-driven material discovery approaches.

Relevant Links

Stay Updated

Subscribe to my Substack for periodic updates on AI and Materials Science