Event Date
Molecular Modeling in the Age of AI: From Energy Materials to Device Simulations
Atomistic modeling, rooted in density functional theory and molecular dynamics, has been a cornerstone of materials science research for nearly four decades. This approach offers a magnifying lens into the atomistic texture of complex systems, enabling the interpretation of experimental data, elucidating structure-function relationships of materials, and exploring extreme conditions. However, on one hand, the efficacy of molecular simulations is hindered by the inherently high computational costs of electronic structure calculations, which restrict the size and complexity of systems accessible to accurate electronic structure calculations to a mere few hundred atoms. On the other hand, the use of empirical potentials grants access to billion-atom models, albeit at the expense of diminished accuracy and transferability.
In this talk, I will illustrate how the integration of advanced statistical learning methods is catalyzing a paradigm shift in materials modeling, combining accuracy, transferability, and computational efficiency. I will discuss the use of machine learning models to predict the stability of new intermetallic compounds for energy-related applications, and to probe the limit of heat transport in inorganic crystals. I will also illustrate the development and the application of ab initio quality machine-learning potentials to simulate materials crystallization at extreme conditions and thermal dissipation in electronic devices.