JOURNAL OF CHEMICAL PHYSICS, cilt.140, sa.24, 2014 (SCI-Expanded)
Determining the lattice thermal conductivity (kappa) of nanostructures is especially challenging in that, aside from the phonon-phonon scattering present in large systems, the scattering of phonons from the system boundary greatly influences heat transport, particularly when system length (L) is less than the average phonon mean free path (MFP). One possible route to modeling kappa in these systems is through molecular dynamics (MD) simulations, inherently including both phonon-phonon and phonon-boundary scattering effects in the classical limit. Here, we compare current MD methods for computing kappa in nanostructures with both L <= MFP and L >> MFP, referred to as mean free path constrained (cMFP) and unconstrained (uMFP), respectively. Using a (10,0) CNT (carbon nanotube) as a benchmark case, we find that while the uMFP limit of. is well-defined through the use of equilibrium MD and the time-correlation formalism, the standard equilibrium procedure for kappa is not appropriate for the treatment of the cMFP limit because of the large influence of boundary scattering. To address this issue, we define an appropriate equilibrium procedure for cMFP systems that, through comparison to high-fidelity non-equilibrium methods, is shown to be the low thermal gradient limit to non-equilibrium results. Further, as a means of predicting kappa in systems having L >> MFP from cMFP results, we employ an extrapolation procedure based on the phenomenological, boundary scattering inclusive expression of Callaway [Phys. Rev. 113, 1046 (1959)]. Using. from systems with L <= 3 mu m in the extrapolation, we find that the equilibrium uMFP kappa of a (10,0) CNT can be predicted within 5%. The equilibrium procedure is then applied to a variety of carbon-based nanostructures, such as graphene flakes (GF), graphene nanoribbons (GNRs), CNTs, and icosahedral fullerenes, to determine the influence of size and environment (suspended versus supported) on kappa. Concerning the GF and GNR systems, we find that the supported samples yield consistently lower values of kappa and that the phonon-boundary scattering remains dominant at large lengths, with L = 0.4 mu m structures exhibiting a third of the periodic result. We finally characterize the effect of shape in CNTs and fullerenes on kappa, showing the angular components of conductivity in CNTs and icosahedral fullerenes are similar for a given circumference. (C) 2014 AIP Publishing LLC.