Enhancing Glassy Dynamics Prediction by Incorporating Displacement from the Initial to Equilibrium State.

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Chemistry

Background:

• The glass transition is a complex phenomenon.
• Machine learning (ML) models enhance predictions of glassy dynamics by considering particle displacement.
• The directional aspect of particle vibrations within cages has been overlooked.

Purpose of the Study:

• To incorporate vectorial displacement into ML models for predicting glassy dynamics.
• To introduce the Equivariance-Constrained Invariant Graph Neural Network (EIGNN) for improved structural encoding.
• To demonstrate the significance of particle displacement orientation in glassy dynamics.

Main Methods:

• Utilizing vectorial displacement from initial to equilibrium states as a structural input for ML models.
• Developing and applying the Equivariance-Constrained Invariant Graph Neural Network (EIGNN).
• Validating EIGNN on a 3D Kob-Andersen system from the GlassBench dataset.

Main Results:

• EIGNN significantly enhances the understanding of structure-dynamics correlations in glassy systems.
• The model demonstrates robust temperature transferability.
• A simplified model (EIGNN++) shows displacement parameters represent local bond orientation order.

Conclusions:

• Vectorial displacement is a critical factor in predicting glassy dynamics.
• The orientation of cage dynamics plays a crucial role in improving predictive models.
• EIGNN offers a powerful framework for studying the glass transition.