.png){kind=icon}
.png){kind=icon}
Return
Deep learning-based density functionals empower AI for materials
.png){kind=icon}
OA
AI
DOI：10.1016/j.matt.2022.05.044.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Density functional theory (DFT) is a widely adopted methodology that gives quantum-level understanding of matter and guides materials discovery. In parallel, artificial intelligence (AI) for materials is an emerging interdisciplinary research direction whose major purpose is to accelerate the process of materials discovery. However, the shortage of informative data-which are crucial for model training-makes accurate prediction, design, synthesis, and characterization a challenge for materials. Experimental data are regarded as expensive, while the relatively less expensive data obtained from DFT calculations suffer from systematic errors rooted in approximated density functionals. Recently, Kirkpatrick et al. constructed the density functional DeepMind 21 (DM21) via deep learning, which solved fractional electron problems and outperformed hand-designed functionals. The foundation of DM21 reveals that data calculated from accurate AI-designed functionals can be supplementary to experimental data in prediction models training and improve models' performance, thus accelerating material discovery processes.
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.