Comparison of atomic-level simulation methods for computing thermal conductivity

TLDR

The study compares equilibrium and nonequilibrium methods for computing thermal conductivity. Using a Sillinger‑Weber silicon model, the authors analyze nonlinear response, thermal equilibration, statistical averaging, and finite‑size effects, showing that extrapolation to infinite size yields reliable nonequilibrium results. Equilibrium Green‑Kubo results are insensitive to the local‑energy definition, and the equilibrium and nonequilibrium methods agree and match experimental silicon conductivity.

Abstract

We compare the results of equilibrium and nonequilibrium methods to compute thermal conductivity. Using Sillinger-Weber silicon as a model system, we address issues related to nonlinear response, thermal equilibration, and statistical averaging. In addition, we present an analysis of finite-size effects and demonstrate how reliable results can be obtained when using nonequilibrium methods by extrapolation to an infinite system size. For the equilibrium Green-Kubo method, we show that results for the thermal conductivity are insensitive to the choice of the definition of local energy from the many-body part of the potential. Finally, we show that the results obtained by the equilibrium and nonequilibrium methods are consistent with each other and for the case of Si are in reasonable agreement with experimental results.