How to quantify and avoid finite size effects in computational studies of crystal nucleation: The case of homogeneous crystal nucleation

Abstract: Finite size artifacts arise in molecular simulations of nucleation when critical nuclei are too close to their periodic images. A rigorous determination of what constitutes too close is, however, a major challenge. Recently, we devised rigorous heuristics for detecting such artifacts based on our investigation of how system size impacts the rate of heterogeneous ice nucleation (Hussain, Haji-Akbari, \emph{J. Chem. Phys.} \textbf{154}, 014108, \textbf{2021}). We identified the prevalence of critical nuclei spanning across the periodic boundary, and the thermodynamic and structural properties of the liquid occupying the inter-image region as indicators of finite size artifacts. Here, we further probe the performance of such heuristics by examining the dependence of homogeneous crystal nucleation rates in the Lennard-Jones liquid on system size. The rates depend non-monotonically on system size and vary by almost six orders of magnitude for the range of system sizes considered here. We confirm that the prevalence of spanning critical nuclei is the primary indicator of finite size artifacts and almost fully explains the observed variations in rate. Proximity, or structuring of the inter-image liquid, however, is not as strong of an indicator due to the fragmented nature of crystalline nuclei. As a result, the dependence of rate on system size is subtle for the systems with a minuscule fraction of spanning critical nuclei. These observations indicate that our heuristics are universally applicable to different modes of nucleation (homogeneous and heterogeneous) in different systems even if they might be overly stringent for homogeneous nucleation,~e.g.,~in the LJ system.