Study of Complex Thermal Transport Enabled by Machine Learning Potential

Ruiqiang Guo*, Cui Zheng, Lin Cheng

Shandong Institute of Advanced Technology, Jinan, 250103, China

EXTENDED ABSTRACT: As the electro-mechanical devices and energy conversion systems approach micro/nanoscale, micro/nanoscale heat transfer becomes a research frontier and hot topic. Understanding heat transfer at micro/nano scale is the key to solving problems in thermal management and energy conversion efficiency of relevant systems. Atomistic simulation has been widely used for studying micro/nanoscale heat transfer, however, it has many limitations in modeling complex systems. On one hand, the atomistic simulation based on empirical potentials can model relatively large systems but usually has low prediction accuracy; on the other hand, ab initio calculation is very accurate but usually only suitable for small systems. Recently, machine learning emerges as a powerful approach for developing interatomic potentials, offering opportunities to solving complex thermal transport issues. We have developed a method for developing machine learning interatomic potentials of materials containing point defects and grain boundaries, which has been demonstrated to be both accurate and efficient in modeling their thermal properties. The developed method provides an effective tool for studying complex thermal transport.