报告题目 (Title)：Orbital-free density functional theory for molecular systems using deep learning（结合深度学习的无轨道密度泛函理论分子体系研究）
报告人 (Speaker)：刘畅（微软研究院科学智能中心）
报告时间 (Time)：2024年9月20日(周五)16:00
报告地点 (Place):腾讯会议室：417-230-544
邀请人 (Inviter)：李永乐 副教授
主办部门：金沙威尼斯欢乐娱人城物理系
摘要 (Abstract)：
Calculating molecular properties is the cornerstone for many vital industry problems, including drug discovery and material design. These properties are commonly solved by Kohn-Sham density functional theory (KSDFT), which still has a large cost scaling, while more scalable alternatives like orbital-free DFT (OFDFT) suffer from accuracy issues, especially for non-periodic molecular systems. In this work, we propose M-OFDFT, an OFDFT implementation that achieves the same level of accuracy as KSDFT on molecules, while maintaining the lower cost scaling. M-OFDFT uses a deep learning model to better approximate the kinetic energy density functional, which is the key component for OFDFT. We build the essential non-locality into the model, which is made affordable by the concise density representation as expansion coefficients under an atomic basis. With techniques to address unconventional learning challenges, M-OFDFT achieves a comparable accuracy to KSDFT on a wide range of molecules untouched by OFDFT before. More attractively, M-OFDFT extrapolates well to molecules much larger than those seen in training, which unleashes the appealing scaling of OFDFT for studying large molecules, representing an advancement of the accuracy-efficiency trade-off frontier in electronic structure methods.
