You reap what you sow: On the impact of nuclei morphology on seeded molecular dynamics simulations

Seeded molecular dynamics represents an increasingly popular approach to investigate crystal nucleation via computer simulations. This method involves the insertion of crystalline seeds into the supercooled liquid phase (often over a range of temperatures or sizes) in order to measure their evolution in time. When dealing with the prototypical scenario of crystal nucleation from Lennard-Jones melts, these seeds are artificially constructed to be approximately spherical fcc nuclei. In addition, the order parameter used to monitor the time evolution of a seed is typically chosen as the number of crystal-like atoms within it-consistent with the tenets of classical nucleation theory. However, evidence suggests that these artificially constructed seeds might be rather different from the crystalline nuclei formed during unbiased molecular dynamics simulations. In particular, previous studies of Lennard-Jones crystallization indicate that non-spherical, as well as bcc, nuclei might be involved with the nucleation process. In this work, we assess the impact of the choice of the initial crystalline nuclei in the context of seeded molecular dynamics by directly comparing two different classes of seeds. Specifically, we consider either crystalline nuclei extracted from "brute force" nucleation trajectories ("unbiased seed") or artificially constructed fcc spherical nuclei ("constructed seeds"). We show that the properties of these two classes of seeds, most notably their committor probability distributions, are markedly different. We also discuss the importance of choosing an appropriate order parameter for seeded molecular dynamics simulations and the implications of our results in the context of estimating crystal nucleation rates via computer simulations.