Vibrational spectrum and entropy in simulation of melting

Abstract

We present a detailed analysis of entropy reconstruction from a velocity autocorrelation function in molecular dynamics simulation for solid and liquid states. The reconstruction is based on the vibrational density of states (VDOS) and for the liquid phase is known as a two-phase thermodynamic (2PT) model. We show that adequate accuracy of VDOS is required for successful application of this technique in the solid phase. We study the convergence of VDOS and entropy on the number of particles, time step and simulation time using aluminum as an example. We also examine the influence of temperature upon VDOS. Our analysis demonstrates that systems containing less than 500 atoms of aluminum do not reproduce the phonon density of state (PhDOS) of the crystal. Nevertheless the error of entropy calculation decreases quickly with the increase of the number of particles in simulation. We note strong influence of high temperatures on VDOS and the difference between VDOS and PhDOS near melting. We show that a time step of 0.5fs or less and trajectories of more than 10,000 time steps are required to obtain good accuracy of entropy in the solid phase. We use quantum molecular dynamics and the 2PT model with a memory function representation for the gas-like component to obtain a new point on the melting curve of aluminum at a pressure of 171GPa, that agree well with previous experimental works and calculations.