Machine-Learning Molecular Dynamics Study on the Structure and Glass Transition of Calcium Aluminosilicate Glasses

Area of Science:

• Materials Science
• Computational Chemistry
• Glass Science

Background:

• Calcium aluminosilicate (CAS) glasses possess valuable thermal and mechanical properties for industrial use.
• Understanding structure-property relationships in CAS glasses, especially in peraluminous compositions, remains a challenge.
• Classical molecular dynamics struggle to accurately represent the aluminum coordination environment in CAS systems.

Purpose of the Study:

• To develop a machine-learning potential (MLP) for accurate simulation of CAS glass structures.
• To investigate structure-property relationships, including aluminum coordination and thermal behavior, in CAS glasses.
• To model the influence of composition on five-coordinated aluminum (Al⁵) and oxygen triclusters (TBOs).

Main Methods:

• Developed a machine-learning potential trained on density functional theory (DFT) data for CAS systems.
• Employed melt-quench simulations using machine learning-based molecular dynamics (MLMD) to generate glass structures.
• Conducted heating simulations to determine glass transition temperature (Tg) and analyze thermal evolution.

Main Results:

• MLMD accurately reproduced experimental CAS glass densities and the fraction of five-coordinated aluminum (Al⁵).
• Observed an increase in Al⁵ and oxygen triclusters (TBOs) in the peraluminous region, consistent with experiments.
• Quantitatively analyzed the compositional dependence of Al⁵ and TBO near the glass transition temperature (Tg).

Conclusions:

• Machine learning-based molecular dynamics (MLMD) provides an accurate approach for modeling CAS glass structures.
• The study offers valuable insights into the role of aluminum in structural rearrangements and thermal behavior.
• This MLMD framework enables robust investigation of structure-property relationships in complex glass systems.