Systematic coarse-graining in nucleation theory

In this work, we show that the standard method to obtain nucleation rate-predictions with the aid of atomistic Monte Carlo simulations leads to nucleation rate predictions that deviate 3 - 5 orders of magnitude from the recent brute-force molecular dynamics simulations [Diemand et al., J. Chem. Phys. 139, 074309 (2013)] conducted in the experimental accessible supersaturation regime for Lennard-Jones argon. We argue that this is due to the truncated state space the literature mostly relies on, where the number of atoms in a nucleus is considered the only relevant order parameter. We here formulate the nonequilibrium statistical mechanics of nucleation in an extended state space, where the internal energy and momentum of the nuclei are additionally incorporated. We show that the extended model explains the lack in agreement between the molecular dynamics simulations by Diemand et al. and the truncated state space. We demonstrate additional benefits of using the extended state space; in particular, the definition of a nucleus temperature arises very naturally and can be shown without further approximation to obey the fluctuation law of McGraw and LaViolette. In addition, we illustrate that our theory conveniently allows to extend existing theories to richer sets of order parameters.