An explainable ML model for binary LJ fluids.

Area of Science:

• Computational physics
• Statistical mechanics
• Materials science

Background:

• Lennard-Jones (LJ) fluids are fundamental models for molecular interactions.
• Binary LJ fluids offer insights into complex fluid mixtures and phase behavior.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting radial distribution functions (RDFs) in binary LJ fluids.
• To assess the ML model's accuracy and extrapolation capabilities across various conditions.

Main Methods:

• Molecular dynamics (MD) simulations were used to generate RDF data for binary LJ mixtures.
• A machine learning model was constructed, discretizing RDFs to reduce dimensionality and improve efficiency.
• The ML model was trained and validated using simulation data across different compositions and temperatures.

Main Results:

• The ML model accurately predicts RDFs for previously unseen binary LJ fluid mixtures.
• The model demonstrates extrapolation capabilities within the compositional-temperature phase space.
• Analysis indicates particle size ratio significantly influences the mixture's microstructure.

Conclusions:

• The developed ML model is effective for predicting RDFs in binary LJ fluids.
• The study highlights the importance of particle size ratio in determining fluid microstructure.
• Limitations exist when the ML model encounters physical regimes outside its training data.