报告时间：2026年6月5日（星期五）上午10:30
报告地点：新能源大楼附楼102会议室
报告摘要：
Machine learning approaches in materials science and chemistry continue to evolve with sophistication. We have endeavored to participate in this exciting direction by applying machine-learning tools to catalysts and reaction dynamics. The development of DFT methodologies continues to mature as a useful tool for computational materials science and machine learning motifs. As “a rising tide lifts all boats”, then so too do these advances in DFT contribute to advances in molecular dynamics capabilities with DFT used for direct evaluation of energies and forces. For machine learning algorithms, particularly time-series convolution methods, rapid, evolving acquisition of molecular-dynamics simulation data is imperative for training and validation. We will discuss the latest developments of FIREBALL, an efficient, DFT molecular dynamics code using pseudo-atomic orbitals (numerical basis set) that does not require fitting of parameters. FIREBALL2020 has greatly improved speed and numerical accuracy with implementations for ASE and with Multiwfn and the package is a very useful tool for generating data used in machine learning models that we have developed. The ability to calculate exactly the non-adiabatic molecular dynamics couplings in a local orbital basis has also enabled us to generate pertinent excited state potential energy surfaces and explore these with correct Hellmann-Feynman force representation. We will discuss two recent directions where we are continually making an impact withing machine learning applications. For example, in reaction dynamics, we will present recent results of how we can use pattern recognition tools to define causal links in Diels-Alder reactions from molecular dynamics simulations.
报告人简介：
James P. Lewis教授，博士生导师，国家高层次人才计划入选者，华南师范大学“杰出人才”引进。从头算密度泛函理论 (DFT) 软件包 FIREBALL 的主要开发者、活跃程序员和发行者。研究方向着重于高效的密度泛函理论的发展和程序开发、应用于环境理论化学以及复杂体系的多尺度模拟，继续设计研发理论化学计算及分子模拟软件包。研发经历丰富，主持各项研究课题经费总额超1000 万美元。发表论文110多篇，总引用次数超过7000多次，H-index42。
报告联系人：苏韧