报告标题标题题目 (Title)：Accurate and Transferable Molecular-orbital-based machine learning for molecular modelling（用于份子建模的、切确且可迁徙的、基于份子轨道的机械进修方式）
报告人 (Speaker)：Sherry Cheng 程立雪（美国加州理工学院CalTech）
报告时候 (Time)：2021年9月3日 (周五) 10:30
集会号：828 7316 8207
约请人 (Inviter)：李永乐 副传授
Quantum simulation is a is a powerful tool for chemists to understand the chemical processes and discover their nature accurately by expensive wavefunction theory (WFT) or approximately by cheap density function theory (DFT). However, the cost-accuracy trade-offs in electronic structure methods limit the application of quantum simulation to large chemical and biological systems. An accurate, transferable, and physical-driven molecular modelling framework, i.e., molecular orbital based machine learning (MOB-ML), is introduced to provide accurate wavefunction-quality molecular descriptions with at most DFT level computational cost. Preserving all the physical constraints, molecular orbital based (MOB) features represent the chemical space faithfully in both supervised learning for molecular property by scalable exact Gaussian processes and unsupervised learning for chemical space explorations. MOB-ML is not only the most accurate method in the low data regime, but also scalable to big data modelling to provide a universal density matrix functional. As an exciting and general new tool to tackle various problems in chemistry, MOB-ML offers great accuracies of predicting total energies of organic and transition-metal containing molecules, non-covalent interactions in the protein backbone-backbone, and transition-state energies. The availability of analytical gradient of MOB-ML opens an avenue of applying MOB-ML to provide accurate potential energy surfaces (PESs) for molecular dynamics simulations, and we further support this by applying PESs obtained from MOB-ML to simulate diffusion Monte Carlo accurately and efficiently for computational spectroscopy.