Accelerating Lattice Thermal Conductivity Calculations in MXenes: A Machine Learning Force Field Approach.

Abstract

Traditionally, lattice thermal conductivity is evaluated using the phonon Boltzmann transport equation (PBTE) in combination with density functional theory (DFT) calculations. However, this approach is computationally intensive. In this study, we predicted lattice thermal conductivity of Ti₂C and Ti₃C₂ MXenes, along with their functionalized variants featuring surface terminations (O, F, OH), by using active DFT-based on-the-fly machine learning force fields (MLFF). The predicted thermal conductivities of Ti₂C and Ti₃C₂ are 73.10 and 101.15 W m^-1 K^-1, respectively, close to previously calculated DFT values. The introduction of surface functional groups significantly reduces the lattice thermal conductivity. Furthermore, the MLFF-based predictions of lattice thermal conductivity are tens to thousands of times faster than conventional DFT calculations, dramatically accelerating the study of thermal transport in MXenes. This efficiency highlights the potential of MLFF as a powerful tool for exploring and optimizing the thermal properties of two-dimensional materials.