Large scale MD simulations of nucleation

We present preliminary results from large scale molecular dynamics (MD) simulations of homogenous vapor to liquid nucleation. The simulations contain between one and eight billion Lennard-Jones atoms and were run for up to 56 million time-steps. The large particle numbers (over 10⁴ times larger than previous simulations, see e.g. [2]) have several advantages: i) Resolving and quantifying nucleation at low supersaturations becomes possible within an accessible number of simulation time-steps, in spite of the very slow nucleation. ii) Even after forming many stable droplets the depletion of the vapor phase is negligible, i.e. the supersaturation remains constant during the simulations. iii) Excellent statistics on liquid droplet abundances and microscopic properties over a wide range in droplet sizes. iv) Simulations can be run efficiently on a large number of cpus. First, direct comparisons to laboratory experiments[6] are now possible: we find excellent agreement in the nucleation rates at kT = 0.3ε and somewhat lower rates in the simulations at kT = 0.4ε. At low temperatures, modified classical nucleation theory significantly underestimates the nucleation rates (by up to 10⁹) and at kT = 1.0ε it overestimates the nucleation rates by up to 10⁵. The semi-phenomenological model[3] matches the nucleation rates and the cluster size distributions found in previous MD simulations at higher supersaturations quite well[2]. But at the lower supersaturations probed here, its predictions differ from the MD results by large factors (up to 10^3.5). We will also present MD results on cluster size distributions, free energy functions, sticking probabilities and condensation and evaporation rates. The microscopic properties (shapes, density profiles, binding energies, etc.) of the large numbers of droplets formed are presented in a separate contribution to this conference (Angélil et. al).