Thermal conductivity of nanoporous silicon: Molecular dynamics simulations and machine learning prediction

The thermal conductivity of nanoporous silicon (p-Si) was investigated using equilibrium molecular dynamics simulations and machine learning techniques. Several interatomic potentials were evaluated, with the Tersoff potential selected for its stability and accuracy. Thermal conductivity values were computed over a wide range of temperatures and porosities. A Symbolic Regression algorithm was used to derive an analytical expression describing its dependence on temperature and porosity. Additionally, Random Forest and Gradient Boosting models were trained to reconstruct heat current autocorrelation functions and predict thermal conductivity, outperforming Support Vector Regression. The combined approach enables accurate, data-efficient prediction of thermal transport in porous nanostructures, supporting the design of materials for thermal management and energy conversion applications.