Learning the bulk and interfacial physics of liquid-liquid phase separation with neural density functionals.

Area of Science:

• Physical Chemistry
• Computational Physics
• Materials Science

Background:

• Understanding phase coexistence in binary mixtures is crucial for predicting material properties.
• Interfacial phenomena significantly influence bulk behavior and material applications.
• Classical density functional theory provides a framework for studying fluid phase behavior.

Purpose of the Study:

• To investigate bulk and interfacial phenomena in binary mixtures using simulation-based supervised machine learning and classical density functional theory.
• To accurately predict liquid-liquid and liquid-vapor binodals for a symmetrical Lennard-Jones mixture.
• To determine interfacial tensions and contact angles across the fluid phase diagram.

Main Methods:

• Simulation-based supervised machine learning
• Classical density functional theory
• Development of a neural density functional

Main Results:

• The trained neural density functional accurately predicts liquid-liquid and liquid-vapor binodals.
• Accurate predictions of interfacial tensions across the entire fluid phase diagram were achieved.
• Contact angles at fluid-fluid interfaces were determined along the triple-phase coexistence line.

Conclusions:

• The study confirms no wetting transition in the investigated symmetrical mixture.
• Machine learning integrated with density functional theory offers a powerful approach for studying complex fluid systems.
• Accurate prediction of interfacial properties is essential for understanding and designing materials.