Thermal Transport in Amorphous Materials with Ab Initio Machine Learning Potentials

Glassy materials are found ubiquitously in nature and are essential components in technological applications. Their thermal conductivity controls heat flows from planetary-level macroscales down to the microscale of electronic components. While lattice thermal conduction in crystals has been extensively studied, much less is known about the relation between structural properties and heat transport in glasses.<br/>Here we exploit the recently developed quasi-harmonic Green-Kubo approach based on lattice dynamics to investigate heat conduction in two systems of broad interest at both the fundamental and the applicational level: diamond-like carbon and amorphous ice. Both systems exhibit structural transitions as a function of the vitrification conditions. In the case of amorphous ice, there are at least two forms separated by a first-order phase transition. Using machine learning potentials trained on ab initio density functional theory datasets enables accuracy and transferability over the broad range of local structures encountered in these materials and allows us to probe vitrification at various thermodynamic conditions.<br/>Through our lattice dynamics approach, we sort out the structural features that lead to vibrational modes diffusion and localization, which modulate the nature and the magnitude of heat conduction in the different forms of glassy carbon and amorphous ice.