Thermal Transport in Solid Materials with Defects and Dislocations Based on a Machine Learning Method

Thermal transport properties are essential for sustainable energy and thermal management. It is also important to incorporate the effect of defects, which widely exist in materials in real world, when calculating thermal transport properties. The prevailing methods for the calculation of thermal transport properties would be either not accurate, using the molecular dynamic (MD) simulation, or time consuming, when applying first-principles calculations. Recently, people applied a reliable method to calculate the transport properties with the combination of the advantages of MD and first-principles through machine learning. In this work, we used a machine learning method based on gaussian regression, called Gaussian Approximation Potential¹ (GAP) method, to achieve a fast and accurate prediction of thermal conductivity of unknown solid materials. We built interatomic potentials using data from density functional theory (DFT) calculation through the training of GAP and applied the potential to calculate the thermal conductivity of cubic boron nitride (cBN), which is a material with ultra-high thermal conductivity². Our simulation results were compared with first-principles calculation and experimental results and showed great agreement. This method was then extended to explore the change of thermal conductivity with the effect of defects and dislocations. Instead of compensating relaxation time with dislocation models³, our method incorporates the effect of dislocations to interatomic potentials. And our results showed good agreement with the reference.<br/><br/>1. Bartók, A. P., <i>et al.</i>, Physical Review Letters (2010) 104 (13), 136403<br/>2. Chen, K., <i>et al.</i>, Science (2020) 367 (6477), 555<br/>3. Klemens, P. G., Proceedings of the Physical Society. Section A (1955) 68 (12), 1113