Machine Learning Accelerated Material Discovery at High Pressure

Title: Machine Learning Accelerated Material Discovery at High Pressure

Time: 9:00 - 10:00 AM, Thursday, July 18th, 2024

Place: Onsite only: Conference room A417, HPSTAR (Beijing)

Host:  Dr. Huiyang Gou

Abstract:

Predicting new phases and simulating corresponding phase transitions in solid materials requires an accurate description of the potential energy surface (PES). While density-functional theory (DFT)-based calculations can provide the required accuracy, they are computationally prohibitive for large systems and/or extended simulation times. In this talk, I will present a popular approach for predicting new structures and simulating reconstructive phase transitions, which integrates random structure search, metadynamics simulation and machine learning representation of high-dimensional PES. This machine learning accelerated method can reach an accuracy comparable to DFT-based calculations but with computational costs reduced by several orders of magnitude less and a nearly linear scaling with system size. I will demonstrate the dynamics simulation of pressure-induced phase transitions in gallium nitride and silicon. Using a half-million atom simulation box, our large-scale simulation reveals the phase transition in remarkable detail, showcasing a sequential change in the phase transition mechanism from collective modes to nucleation and growth. Additionally, I will illustrate the static simulation for structure prediction of sodium under high pressure, where we successfully identified a crystal structure corresponding to a long-elusive phase of sodium discovered experimentally in 2008.