Large Scale Molecular Dynamics Simulations of Homogeneous Nucleation

TL;DR

This study uses large-scale molecular dynamics simulations with up to eight billion atoms to investigate homogeneous vapor-to-liquid nucleation, providing detailed data that aligns well with experiments and tests classical and semi-phenomenological nucleation models.

Contribution

First large-scale MD simulations of vapor nucleation covering extensive temperature and supersaturation ranges, enabling direct comparison with experiments and evaluation of nucleation theories.

Findings

Good agreement between MD simulations and experimental nucleation rates.

Classical nucleation theory significantly underestimates rates at low temperatures.

Semi-phenomenological models are closer but still have discrepancies.

Abstract

We present results from large-scale molecular dynamics (MD) simulations of homogeneous vapor-to-liquid nucleation. The simulations contain between one and eight billion Lennard-Jones (LJ) atoms, covering up to 1.2 {\mu}s (56 million time-steps). They cover a wide range of supersaturation ratios, S=1.55 to 10^4, and temperatures from kT = 0.3 to 1.0 {\epsilon} (where {\epsilon} is the depth of the LJ potential, and k the Boltzmann constant). We have resolved nucleation rates as low as 10^{17} cm^{-3} s^{-1} (in the argon system), and critical cluster sizes as large as 100 atoms. Recent argon nucleation experiments probe nucleation rates in an overlapping range, making the first direct comparison between laboratory experiments and molecular dynamics simulations possible: We find very good agreement within the uncertainties, which are mainly due to the extrapolations of argon and LJ saturation curves to very low temperatures. The self-consistent, modified classical nucleation model of Girshick and Chiu [J. Chem. Phys. 93, 1273 (1990)] underestimates the nucleation rates by up to 9 orders of magnitudes at low temperatures, and at kT = 1.0 {\epsilon} it overestimates them by up to 10^5. The predictions from a semi-phenomenological model by Laaksonen et al. [Phys. Rev. E 49, 5517 (1994)] are much closer to our MD results, but still differ by factors of up to 104 in some cases. At low temperatures, the classical theory predicts critical clusters sizes, which match the simulation results (using the first nucleation theorem) quite well, while the semi-phenomenological model slightly underestimates them. At kT = 1.0 {\epsilon} the critical sizes from both models are clearly too small. (abridged)