Toward a Mobility-Preserving Coarse-Grained Model: A Data-Driven Approach.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Coarse-grained molecular dynamics (MD) simulations offer faster calculations for material design.
• Existing coarse-graining methods accurately predict structure but artificially increase molecular mobility.
• This artificial acceleration hinders the accurate dynamic properties prediction.

Purpose of the Study:

• To develop a data-driven approach for generating coarse-grained models that preserve all-atom molecular mobility.
• To create a machine learning model that predicts simulation-ready, mobility-preserving coarse-grained potentials.
• To address the drawback of artificial mobility acceleration in coarse-grained simulations.

Main Methods:

• Designed an artificial neural network (ANN) model.
• The ANN takes all-atom force field (FF) parameters as input.
• The ANN directly outputs simulation-ready coarse-grained potentials that preserve molecular mobility.
• Used 2,3,4-trimethylpentane as a model system for proof of principle.

Main Results:

• Quantified artificial mobility acceleration using an acceleration factor (ratio of coarse-grained to all-atom diffusion coefficients).
• The best ANN model reduced the acceleration factor to approximately 2.
• Typical coarse-grained models can exhibit acceleration factors as high as 7.
• Achieved a significant reduction in the artificial acceleration of molecular dynamics.

Conclusions:

• The developed data-driven method successfully generates coarse-grained potentials with preserved molecular mobility.
• This approach offers a route to accurate dynamic properties in coarse-grained simulations.
• The method is expected to be valuable for the computational materials science community.