Machine learning based insights of seeded congruent crystal growth of LiNbO3 in glass

Abstract

The seeded crystal growth of LiNbO${}_3$ in glass under the isothermal conditions has been studied using a machine-learned clustering algorithm trained on a combination of static and dynamic structural features. Our findings contradict the sharp crystal-glass interface assumption of classical nucleation theory (CNT). The growth of the seed occurs via the attachment of a group of atoms rather than single atoms. The predictions from the machine-learned simulations helped us compare the growth rate of seeds across various initial seed-sizes and temperature. Simulations with multiple seeds show that the growth rate of a seed is enhanced by the presence of another seed in its vicinity.