Public Research Seminar by Advanced Materials Thrust - Machine Learning Promises a New Paradigm for Computational Materials Science

In modern computational materials science, great efforts have been made to develop simulation methods, e.g., density functional theory (DFT) and molecular dynamics. These simulation methods can help researchers understand mechanisms, predict properties and design new materials. Despite these successes, there remain multiple experimental phenomena that can hardly be described by conventional atomistic/molecular simulation methods, which severely impede us from further understanding and designing advanced functional materials. Recently, computational materials science is undergoing a second revolution empowered by machine learning (ML). ML methods do not exclusively rely on the theoretical understanding of materials but take a data-driven approach to solve the problems. In this talk, I will first present my earlier works on using the conventional DFT method to inversely design functional materials (e.g. electrides) and explore their phase diagram and novel properties under extreme conditions. In the second part of my talk, I will mainly focus on my recent works on applying ML to predict the notorious properties of materials, i.e. lifetime of Li-ion batteries and high-temperature superconductivity, which are challenging for conventional simulation methods.