Molecular dynamics simulations of thermal conductivity of carbon nanotubes: Resolving the effects of computational parameters

RN Salaway and LV Zhigilei, INTERNATIONAL JOURNAL OF HEAT AND MASS TRANSFER, 70, 954-964 (2014).

DOI: 10.1016/j.ijheatmasstransfer.2013.11.065

Predicting thermal conductivity, k, of carbon nanotubes (CNTs) has been the focus of many molecular dynamics (MD) simulation studies reported in the literature. The values of k obtained in these studies exhibit a large, up to an order of magnitude, variability that is commonly attributed to the variations in the computational setups adopted in different studies. The sensitivity of the computational results to the choice of individual parameters of the simulation setups, however, has not been systematically investigated and is often overlooked when the predicted values of k are compared across the literature. Here we present the results of several series of simulations specifically designed to evaluate the effects of common computational parameters of non-equilibrium MD (NEMD), such as the type of boundary conditions, size and location of heat bath regions, definition of the CNT length, and the choice of interatomic potential, on the computational predictions. The length dependence of thermal conductivity is found to exhibit a gradual transition from a strong increase of k with CNT length for nanotubes that are shorter than similar to 200 nm to a much weaker dependence for longer CNTs, reflecting the transition from ballistic to diffusive- ballistic heat transport regimes. The effect of increasing length of thermal bath regions is found to be nearly indistinguishable from the effect of increasing length of the unperturbed region between the bath regions, suggesting that the value of k is defined by the total length of the CNT (including the length of the heat bath regions) in NEMD simulations employing uni-directional heat flux. The choice of interatomic potential is shown to be responsible for an up to fourfold variability in predictions of k for otherwise identical simulation conditions. Overall, the results of this study help elucidate the cause of quantitative discrepancies across published data and provide recommendations on the choice of simulation setups that may improve the consistency of the computational predictions. (C) 2013 Elsevier Ltd. All rights reserved.

Return to Publications page