Dec 4, 2024
9:30am - 9:45am
Hynes, Level 2, Room 209
Domagoj Fijan1,Brandon Butler1,Maria Ward Rashidi1,Sharon Glotzer1
University of Michigan1
Domagoj Fijan1,Brandon Butler1,Maria Ward Rashidi1,Sharon Glotzer1
University of Michigan1
Nucleation and crystallization processes are critical in diverse materials, such as colloidal, soft, molecular, metallic, and ionic crystals. While there's considerable understanding of nucleation in simple crystals, research on complex crystals' nucleation is limited due to the lack of suitable computational tools. A significant challenge is the lack of a tool that can capture the key property of crystalline materials—symmetry—in a continuous and efficient manner. Another major issue is the absence of system-independent tools for the automatic analysis of molecular simulation trajectories to identify those exhibiting interesting behavior. We address these problems using machine learning-assisted tools that automate the detection of rare events (both post-analysis and online) and introduce an order parameter that describes local structures with continuous point group symmetry.<br/>To enhance the automation of event detection and manage the data-intensive demands of these simulations, we developed dupin, a Python-based tool for detecting rare events in molecular simulation trajectories. While primarily designed for molecular simulations, dupin is also versatile enough to analyze experimental data, such as trajectories from microscopy experiments. This tool automates the detection of rare events, opening new research avenues in the study of rare events, nucleation, and crystallization. It enables the automated analysis of vast amounts of trajectories, facilitating the curation and preparation of datasets for machine learning applications focused on nucleation pathways. Additionally, dupin can operate on-the-fly during simulations or experiments, allowing for real-time adjustments based on detected events. An example of its utility is enabling very high-frequency sampling of molecular dynamics particle positions upon detecting a rare event minimizing data storage needs and enabling detailed studies of nucleation events at very high temporal resolution.<br/>Given our focus on nucleation and crystallization, we have developed an order parameter specifically designed to quantify the structural evolution of local environments in complex crystals to work in tandem with dupin. Complex crystals feature different local environments; understanding nucleation at the most fundamental level necessitates tracking the progression of symmetry elements across these environments. Traditional order parameters, such as Steinhardt's Parameters (SOP) and its variants, often combine multiple symmetries into a single value, without useful and meaningful geometric or structural significance. In contrast, our Point Group Order Parameter (PGOP) provides a per-particle value that enables the continuous quantification of point group symmetries exhibited by particle arrangements. This method effectively quantifies the alignment of local environments with specific symmetry operations, enabling the distinct differentiation of local environments within complex crystals.<br/>We illustrate dupin’s capability to automatically detect diverse events in soft matter simulations, such as complex nucleation processes and polymer collapses. Employing both tools in tandem allows for a deeper understanding of nucleation in complex crystallization pathways, analyzed through molecular dynamics simulations interacting with various potentials. We highlight dupin’s ability to perform real-time, on-line detection, which enables the sampling of exceptionally detailed nucleation process trajectories. We showcase PGOP’s utility in effectively quantifying different crystalline environments in various crystals during the nucleation process. Collectively, these tools not only advance the study of nucleation and crystallization but also open new avenues for research in large-scale crystallization projects and the application of emerging fields, such as machine learning, to explore nucleation and crystallization pathways.