High-Precision Prediction of Thermal Conductivity of Metals by Molecular Dynamics Simulation in Combination with Machine Learning Approach

Abstract

Molecular dynamics (MD) simulation is a powerful tool to estimate materials properties from atomistic viewpoint. However, the scope of application of MD simulations is limited to problems where the Newton's equation of motion for atoms is dominant. Therefore, it is inherently insufficient to estimate thermal conductivity of metallic materials, which consists of phononic and electronic components. In this study, machine learning (ML)-based regression model is employed to predict thermal conductivity of metals with high accuracy using deficient results from MD simulations. A regression analysis with the least absolute shrinkage and selection operator (Lasso) including electrical conductivity as predictor variables successfully predict the thermal conductivity of metals with negative temperature dependence, which indicates a significant contribution of electrons to thermal conduction in metals. It should be stressed that our prediction is better than the conventional estimation from the Wiedemann–Franz law. This study shows us new possibilities of new ML approach for improving the accuracy of physical properties obtained from MD simulations.