Machine learning prediction of self-diffusion in Lennard-Jones fluids.

Area of Science:

• Computational physics
• Materials science
• Chemical engineering

Background:

• Accurate prediction of self-diffusion is crucial for understanding fluid behavior.
• Existing empirical models for Lennard-Jones fluids have limitations in predictive accuracy.
• Machine learning offers potential for developing more robust predictive models.

Purpose of the Study:

• To explore and compare different machine learning methods for predicting self-diffusion in Lennard-Jones fluids.
• To evaluate the performance of Random Forest (RF) and Artificial Neural Network (ANN) regression models.
• To assess the impact of feature engineering on model performance.

Main Methods:

• Development and characterization of multiple RF and ANN regression models.
• Utilized a database of diffusion constants from molecular dynamics simulations.
• Feature engineering was applied and its effect on RF model performance was analyzed.
• Model predictions were compared against an established empirical relationship for LJ fluid diffusion.

Main Results:

• ANN regression models demonstrated superior predictive accuracy for self-diffusion compared to existing empirical relationships.
• RF models, particularly with enhanced feature engineering, also showed competitive performance.
• The study quantified prediction errors for all developed ML models.

Conclusions:

• ANN regression models represent a significant advancement for predicting self-diffusion in Lennard-Jones fluids.
• Machine learning approaches, especially ANNs, offer a powerful alternative to traditional empirical methods in fluid dynamics.
• Feature engineering plays a key role in optimizing the performance of machine learning models for physical property prediction.