Data-driven coarse-grained modeling of polymers in solution with structural and dynamic properties conserved.

Area of Science:

• Computational chemistry and materials science
• Polymer physics and simulations

Background:

• Atomistic simulations provide detailed polymer behavior but are computationally expensive.
• Coarse-grained (CG) modeling simplifies systems but often struggles to conserve dynamic and structural properties.
• The generalized Langevin equation (GLE) framework offers a path for accurate CG modeling.

Purpose of the Study:

• To develop a data-driven coarse-grained (CG) modeling approach for polymers in solution.
• To ensure the CG model conserves both dynamic and structural properties of the atomistic system.
• To enhance computational efficiency in polymer simulations.

Main Methods:

• Utilized the generalized Langevin equation (GLE) framework.
• Employed a two-stage Gaussian process-based Bayesian optimization to infer the non-Markovian memory kernel from velocity autocorrelation function (VACF) data.
• Developed an active learning method to identify hydrodynamic scaling in VACF and memory kernel, accelerating kernel inference.
• Compared deep learning and iterative Boltzmann inversion for constructing the CG potential.
• Mapped the GLE onto an extended Markovian process for computational efficiency.

Main Results:

• The proposed CG modeling robustly and accurately reproduced dynamic and structural properties of polymers in solution.
• Demonstrated conservation of key properties compared to reference atomistic simulations.
• Assessed accuracy and computational efficiency using a model star-polymer solution system across various concentrations.

Conclusions:

• The developed data-driven CG modeling approach effectively captures polymer dynamics and structures in solution.
• The method offers a computationally efficient alternative to atomistic simulations while maintaining high accuracy.
• This approach provides a powerful tool for studying complex polymer systems.