Finding defects in glasses through machine learning.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Structural defects significantly influence the properties of glasses, including their kinetic, thermodynamic, and mechanical behaviors.
• Quantum tunneling two-level systems (TLS) are rare defects that dominate the physics of glasses at very low temperatures.
• Identifying TLS in computer simulations is challenging due to their low density.

Purpose of the Study:

• To develop an efficient machine learning approach for exploring glass potential energy landscapes.
• To identify and characterize quantum tunneling two-level systems (TLS) within glass models.
• To accelerate the discovery and analysis of TLS, improving our understanding of their microscopic nature.

Main Methods:

• Introduction of a machine learning approach to efficiently explore potential energy landscapes of glass models.
• Design of an algorithm to rapidly predict quantum splitting between amorphous configurations.
• Utilizing classical simulations to generate amorphous configurations for analysis.

Main Results:

• The machine learning approach enables efficient exploration and identification of desired defect classes, specifically TLS.
• The developed algorithm significantly speeds up the prediction of quantum splitting for TLS.
• Computational effort is redirected towards identifying more TLS, rather than abundant non-tunneling defects.

Conclusions:

• Machine learning provides an effective strategy for identifying rare structural defects like TLS in glasses.
• The developed algorithm enhances the efficiency of studying TLS, crucial for low-temperature glass physics.
• Interpretation of the ML model offers direct physical insights into the microscopic nature of TLS.