Bayesian Optimization for High-Dimensional Coarse-Grained Model Parameterization: A Case Study on Pebax Polymer.

Area of Science:

• Computational materials science
• Polymer physics
• Statistical mechanics

Background:

• Coarse-grained (CG) force fields offer computational efficiency for materials simulations.
• Traditional CG model parametrization uses sequential top-down and bottom-up methods, limiting joint parameter optimization.
• Bayesian optimization (BO) has been limited to low-dimensional problems, hindering its application to complex CG models.

Purpose of the Study:

• To extend Bayesian optimization (BO) for high-dimensional coarse-grained (CG) model parametrization.
• To challenge the assumption that BO is unsuitable for complex CG models with numerous parameters.
• To develop an efficient CG model for Pebax-1657 by optimizing multiple physical properties simultaneously.

Main Methods:

• Utilized the tree-structured Parzen estimator (TPE) model, a variant of Bayesian optimization (BO).
• Applied BO-TPE to a 41-parameter CG model of Pebax-1657.
• Simultaneously optimized structural (density, radius of gyration) and thermodynamic (glass transition temperature) properties.

Main Results:

• Successfully parametrized a high-dimensional (41 parameters) CG model of Pebax-1657 using BO-TPE.
• The optimized CG model accurately reproduced key physical properties of the atomistic representation.
• BO-TPE demonstrated faster convergence and consistent improvements over traditional parametrization methods.

Conclusions:

• Bayesian optimization (BO), specifically BO-TPE, is a viable and effective strategy for optimizing high-dimensional CG force fields.
• This extended BO approach enables accurate and efficient CG model development for complex polymers like Pebax-1657.
• The framework offers a significant advancement over conventional methods for CG model parametrization in materials simulations.