Melt crystallization mechanism analyzed with dimensional reduction of high-dimensional data representing distribution...

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Melt crystallization is crucial for industries like semiconductors, food, and ice manufacturing.
• Understanding the molecular mechanisms of melt crystallization is challenging due to experimental limitations in observing melt structures.
• Computer simulations, particularly molecular dynamics (MD), are vital for studying melt structures, but require proper analysis of temporal structural evolution.

Purpose of the Study:

• To investigate the time evolution of structural order during melt crystallization using an unsupervised machine learning technique.
• To elucidate the detailed mechanism of melt crystallization, including nucleation and structural rearrangement.
• To develop a method applicable to complex, real-world materials.

Main Methods:

• Employed dimensional reduction (DR), an unsupervised machine learning technique, to analyze high-dimensional data from molecular dynamics (MD) simulations.
• Utilized atom-atom pair distribution functions and nearest-neighbor angular distribution functions as input for DR.
• Simulated a supercooled Lennard-Jones melt undergoing crystallization.

Main Results:

• The study identified crystallization occurring through sequential activation processes: initial nucleation of a distorted crystal structure followed by its reconstruction into a more stable form.
• Dimensional reduction effectively captured the time evolution of local structures during the crystallization process.
• The findings provide insights into the dynamic changes in atomic arrangements during phase transitions.

Conclusions:

• The developed DR method offers a powerful approach to study crystallization mechanisms from disordered systems.
• This technique is applicable to complex, multi-component materials, advancing the understanding of real-world crystallization phenomena.
• The findings contribute to a deeper comprehension of fundamental crystallization processes relevant to various industrial applications.