Challenges and opportunities in atomistic simulations of glasses: a review

Abstract

Atomistic modeling and simulations have been pivotal in our understanding of the glassy state. Indeed, atomistic modeling offers direct access to the structure and dynamics of atoms in glasses—which is typically hidden from conventional experiments. Simulations also offer a more economical, faster alternative to systematic experiments to decode composition-property relationships and accelerate the discovery of new glasses with desirable properties and functionalities. However, the atomistic modeling of glasses remains plagued by a series of challenges, e.g., high computational cost, limited accessible timescale, lack of accurate interatomic forcefields, etc. These challenges often result in the existence of discrepancies between simulation and experimental data, thereby limiting the predictive power of atomistic modeling. Here, we review recent accomplishments and remaining challenges facing the atomistic modeling of glasses. We discuss future opportunities offered by the seamless integration of simulations, knowledge, experiments, and machine learning in advancing glass modeling to a new era.

Keywords

• Molecular dynamics
• Interatomic forcefield
• Reserve Monte Carlo
• Machine learning
• Accelerated simulation
• Differentiable programming
• Inverse design