Current browse context:
cond-mat.mtrl-sci
Change to browse by:
References & Citations
Condensed Matter > Materials Science
Title: Scaling Crystal Structure Relaxation with a Universal Trustworthy Deep Generative Model
(Submitted on 1 Apr 2024)
Abstract: The evolution of AI and high-throughput technologies has boosted a rapid increase in the number of new materials, challenging our computational ability to comprehensively analyze their properties. Relaxed crystal structures often serve as the foundational basis for further property calculations. However, determining equilibrium structures traditionally involves computationally expensive iterative calculations. Here, we develop DeepRelax, an efficient deep generative model designed for rapid structural relaxation without any iterative process. DeepRelax learns the equilibrium structural distribution, enabling it to predict relaxed structures directly from their unrelaxed counterparts. The ability to perform structural relaxation in just a few hundred milliseconds per structure, combined with the scalability of parallel processing, makes DeepRelax particularly useful for large-scale virtual screening. To demonstrate the universality of DeepRelax, we benchmark it against three different databases of X-Mn-O oxides, Materials Project, and Computational 2D Materials Database with various types of materials. In these tests, DeepRelax exhibits both high accuracy and efficiency in structural relaxation, as further validated by DFT calculations. Finally, we integrate DeepRelax with an implementation of uncertainty quantification, enhancing its reliability and trustworthiness in material discovery. This work provides an efficient and trustworthy method to significantly accelerate large-scale computations, offering substantial advancements in the field of computational materials science.
Link back to: arXiv, form interface, contact.