Machine learning of an implicit solvent for dynamic Monte Carlo simulations.

Area of Science:

• Computational physics and chemistry
• Polymer science
• Soft matter physics

Background:

• The Bond Fluctuation Model (BFM) is a coarse-grained simulation method for polymers, membranes, and soft matter.
• Explicit solvent models are computationally expensive for localized systems, while implicit models face ergodicity issues.
• Simulating complex polymer systems like coagulated chains requires efficient solvent modeling.

Purpose of the Study:

• To introduce a novel artificial neural network (NN)-based implicit solvent model for BFM simulations.
• To overcome the limitations of traditional implicit solvent models in polymer simulations.
• To develop a computationally efficient method for studying universal polymer properties.

Main Methods:

• Training an artificial neural network (NN) using BFM simulation data of single homopolymers in explicit solvent.
• Implementing the trained NN as an implicit solvent model within the BFM framework.
• Validating the NN-based model by simulating polymer properties and comparing them to known behaviors.

Main Results:

• The NN-based implicit solvent model successfully reproduces universal macroscopic polymer properties, including the coil-globule transition.
• Simulations accurately capture static and dynamic bond autocorrelation and mean square displacement.
• Learned NN parameters demonstrate transferability from single-chain to multi-homopolymer systems.

Conclusions:

• The developed NN-based implicit solvent model offers an efficient and accurate alternative for BFM simulations.
• This method effectively addresses computational costs associated with explicit solvent models in complex polymer systems.
• The model's ability to reproduce key polymer phenomena and parameter transferability highlight its versatility and potential for broader applications in soft matter research.