Machine Learning Predictions of Simulated Self-Diffusion Coefficients for Bulk and Confined Pure Liquids.

Area of Science:

• Physical Chemistry
• Computational Science

Background:

• Predicting fluid diffusion is crucial for applications in separations, catalysis, batteries, and subsurface energy.
• Existing empirical and machine learning (ML) models primarily focus on bulk fluids.

Purpose of the Study:

• To apply artificial neural network (ANN) models for predicting self-diffusion coefficients of real liquids in bulk and confined (pore) environments.
• To assess the accuracy of ANN models in reproducing diffusion data from molecular dynamics (MD) simulations.

Main Methods:

• Generated training data using MD simulations of Lennard-Jones particles for 14 diverse molecules.
• Created pore models (planar, cylindrical, hexagonal) with carbon atom walls.
• Trained ANN models using simple descriptors to predict diffusion coefficients.

Main Results:

• Simulated bulk liquid diffusion coefficients showed excellent agreement with experimental data.
• ANN models accurately reproduced MD diffusion data for both bulk and confined liquids.
• Observed increased mobility in large pores compared to bulk liquids was accurately predicted.

Conclusions:

• ANN models offer a reliable method for predicting fluid diffusion in both bulk and porous media.
• Accurate diffusion prediction in confined environments can accelerate development in materials science and chemical engineering.
• This approach facilitates the design and optimization of technologies relying on fluid transport in porous materials.