Machine learning enabled molecular dynamics-Monte Carlo framework for nanoconfined fluid adsorption.

Area of Science:

• Computational physics and chemistry
• Artificial intelligence in scientific simulation
• Nanofluidics and materials science

Background:

• Nanoconfined fluids are critical for energy production, carbon sequestration, and separations.
• Classical molecular dynamics (MD) is computationally expensive, limiting scale.
• Existing hybrid methods lack generalizability.

Purpose of the Study:

• To develop an AI-assisted hybrid MD-Monte Carlo (MDMC) framework.
• To overcome the limitations of computational cost and generalizability in MD simulations.
• To enable efficient and accurate simulation of nanoconfined fluids.

Main Methods:

• Developed an AI-assisted MDMC framework learning transition statistics from MD trajectories.
• Encoded molecular motion into neural network-predicted displacement actions.
• Reduced dimensionality while preserving MD-level accuracy.

Main Results:

• Achieved efficient sampling with robust generality.
• Demonstrated framework accuracy and transferability across diverse conditions (temperature, pressure).
• Validated performance across various spatial scales and complex nano-scale geometries.

Conclusions:

• The AI-assisted MDMC framework provides a versatile path for simulating nanoconfined fluids.
• This approach significantly enhances computational efficiency for engineering applications.
• The method offers accurate and generalizable simulations for complex systems.