Machine Learning-Assisted Multi-Target Coarse-Graining Strategy for Polystyrene.

Area of Science:

• Computational chemistry
• Materials science
• Polymer physics

Background:

• Coarse-grained (CG) molecular dynamics is crucial for bridging atomistic simulations and macroscopic experiments.
• Developing CG models with simultaneous structural, thermodynamic, and dynamical consistency is a significant challenge.

Purpose of the Study:

• To present a machine learning-assisted, multi-objective parameterization strategy for atactic polystyrene (PS).
• To construct a dynamically consistent CG model by including diffusion coefficients in the optimization.

Main Methods:

• Utilized a 2:1 mapping scheme for polystyrene.
• Integrated Support Vector Regression (SVR) and Particle Swarm Optimization (PSO) to optimize Lennard-Jones parameters.
• Optimized parameters to reproduce atomistic-level radial distribution functions, density, cohesive energy density, and self-diffusion coefficients.

Main Results:

• Achieved remarkable agreement between the CG force field and all-atom (AA) simulations across multiple observables.
• Successfully reproduced structural, thermodynamic, and dynamical properties of polystyrene.
• Demonstrated the effectiveness of including diffusion coefficients for dynamic consistency.

Conclusions:

• The developed strategy establishes a robust framework for predictive polymer modeling.
• This methodology can be extended to materials discovery and rational polymer design.
• The machine learning-assisted approach enhances the accuracy and reliability of coarse-grained models.