Accurate evaluation of thermal conductivity using molecular simulation can be challenging depending on the system’s dynamical behavior. Two computational strategies are commonly employed in this field: equilibrium molecular dynamics (EMD) and nonequilibrium molecular dynamics (NEMD). The EMD approach, which relies on the Green–Kubo formalism, requires extensive sampling of the heat flux time autocorrelation function. The NEMD approach, which relies on the direct implementation of Fourier’s law, requires obeying the linear regime and checking for any finite size effects. Here, using a united atom model for methane and a flexible zeolite framework, we discuss the fundamental aspects of such methods by analyzing the convergence in time and/or in size. We show that both approaches can be rationalized by invoking characteristic time and length scales, tc and Lc. In an equivalent manner as the convergence in EMD approach is achieved when the simulation time t ≫ tc, the convergence in NEMD methods requires system sizes L ≫ Lc (in the latter case, smaller simulation boxes lead to finite size effects due to phonon scattering at the temperature gradient boundaries). Notably, EMD and NEMD data collapse on the same curve when rescaled through these parameters.
Souza et al. (Tue,) studied this question.