(115j) Divergence between Structure and Nucleation Outcome in Heterogeneous Nucleation of Close-Packed Crystals

Crystallization is a process in which molecules and/or building blocks within a melt or a solution assemble into a crystalline phase, and can occur in a wide variety of atomic, molecular, metallic, polymeric and colloidal systems over a wide range of time and length scales. Crystallization almost always proceeds through a process known as nucleation and growth wherein a sufficiently large crystalline seed emerges during nucleation, which then grows by absorbing more building blocks until the system reaches thermodynamic equilibrium. Under most circumstances, nucleation occurs heterogeneously in which an external surface facilitates crystallization by decreasing nucleation barriers. Despite being the dominant mechanism of crystallization and self-assembly, there are major gaps in our understanding of how surface chemistry and topography impacts the kinetics and mechanism of heterogeneous nucleation.

In recent decades, molecular simulations have emerged as attractive tools for filling this gap and have been successfully used for probing the physics of nucleation in wide variety of systems. One of the necessary ingredients of such explorations is the construction of good collective variables (CVs) that accurately describe the progress of nucleation. Historically, such order parameters have been developed by comparing the static structures of bulk target crystals and the corresponding fluids [1].

In this talk, I will discuss our recent work [2] in which we use molecular dynamics (MD) simulations, jumpy forward flux sampling (jFFS) [3] and machine learning to probe heterogeneous nucleation of close-packed crystals in the Lennard-Jones and hard sphere systems. We demonstrate that traditional CVs based on structural properties of bulk crystals fail at describing the progress of heterogeneous nucleation despite being superb descriptors of homogeneous nucleation of the same crystals. There is therefore a divergence between traditional notions of crystalline structure and nucleation outcome, which is accompanied by an intriguing dynamical anomaly wherein crystalline particles with few crystalline neighbors tend to move faster than their liquid-like counterparts. Through high throughput screening of hundreds of thousands of CVs and machine learning, we systematically assess different strategies to refine such CVs. In particular, we find readjusting Steinhardt bond order parameter cutoffs and pruning crystalline bridges to be effective strategies in improving CV efficacy. Employing such optimal CVs is pivotal for proper characterization of heterogeneous nucleation in metallic and colloidal systems that typically form such close-packed crystals.

[1] Steinhardt PJ, Nelson DR, Ronchetti M, Phys. Rev. B, 28: 784 (1983).

[2] Domingues TS, Hussain S, Haji-Akbari A, J. Phys. Chem. Lett., 15: 1279 (2024).

[3] Haji-Akbari A, J. Chem. Phys., 149: 072303 (2018).