Thermal and phase transition properties of van der Waals heterostructures by equivariant graph neural networks

Simulations using molecular dynamics (MD) are particularly suitable to study the physical properties of complex materials at the atomic scale but require an accurate description of their potential energy surface (PES). In the past couple of years, much progress has been made in the development of machine learning interatomic potentials (ML-IP) trained on ab initio simulations to describe such PES. Recently, ML-IP relying on equivariant graph neural networks (GNN) have shown very high accuracy and data efficiency compared to other ML-IP methods.

The goal of this PhD is to build equivariant graph neural networks potentials to perform MD simulations to study the thermal and phase transitions properties of van der Waals heterostructures such as chalcogenides superlattices and stacked 2D materials.

Facebook

Twitter