Polydisperse hard spheres: crystallization kinetics in small systems and role of local structure

Abstract

We study numerically the crystallization of a hard-sphere mixture with 8% polydispersity. Although often used as a model glass former, for small system sizes we observe crystallization in molecular dynamics simulations. This opens the possibility to study the competition between crystallization and structural relaxation of the melt, which typically is out of reach due to the disparate timescales. We quantify the dependence of relaxation and crystallization times on density and system size. For one density and system size we perform a detailed committor analysis to investigate the suitability of local structures as order parameters to describe the crystallization process. We find that local structures are strongly correlated with generic bond order and add little information to the reaction coordinate.

Confinement-induced demixing and crystallization

“…Additionally, for the study of structural relaxation in supercooled liquids and glasses a size dispersity must be introduced to prevent crystallization even at very small packing fractions [6,12,13]. Despite their popularity as model glass formers, recent observations in bulk have revealed that Gaussiandistributed hard spheres [37][38][39][40] as well as the often used Kob-Anderson model [41,42] form crystals already in the supercooled regime. This crystallization is induced by a process called fractionation which describes the separation of a homogeneous fluid into different liquid or crystalline fractions with very different particle-size distributions.…”

Kinetic crystallisation instability in liquids with short-ranged attractions

“…To determine the binodal we use a polydisperse system to prevent crystallization. In our five-component system, small systems (N 500) can exhibit fluctuations to crystalline states [63], but for the system sizes we consider here we have never observed crystallization when the polydispersity δ ≥ 0.08 [23].…”

Employing Artificial Neural Networks to Identify Reaction Coordinates and Pathways for Self-Assembly

“…Even for simple systems, it is challenging to uncover reaction coordinates for such highly cooperative transitions as occurring in nucleation and self-assembly. While optimal linear combinations of a set of candidate order parameters have been determined through maximum-likelihood methods, − autoencoder architectures have the potential to uncover strongly nonlinear optimal mappings to reaction coordinates. Our results constitute a step toward a comprehensive computational understanding for multistep and hierarchical self-assembly through a combination of unsupervised dimensionality reduction of a set of structural descriptors with Markov state modeling.…”