Hierarchical Deep Potential with Structure Constraints for Efficient Coarse-Grained Modeling.

Area of Science:

• Computational materials science
• Polymer physics
• Machine learning applications

Background:

• Coarse-grained molecular dynamics (CGMD) simplifies simulations of large systems but requires accurate force fields.
• Developing precise CGMD force fields for complex polymers remains a significant challenge.
• Existing methods often struggle with capturing essential structural and energetic properties.

Purpose of the Study:

• To introduce a novel framework, hierarchical deep potential with structure constraints (HDP-SC), for constructing accurate coarse-grained force fields for polymers.
• To enhance the accuracy and efficiency of coarse-grained models for polymer materials.
• To leverage machine learning for advanced force field development.

Main Methods:

• Integration of a prior energy term from direct Boltzmann inversion with a deep neural network potential.
• Training the deep neural network using hierarchical bead environment descriptors.
• Incorporation of structure constraints to guide the force field construction.

Main Results:

• The HDP-SC framework successfully reproduces structural distributions and potentials of mean force.
• Validation using polystyrene systems demonstrates accurate reproduction of structural properties.
• The model shows applicability at larger simulation scales, indicating scalability.

Conclusions:

• The HDP-SC framework offers a promising approach for developing accurate and efficient coarse-grained force fields for polymers.
• Machine learning techniques, particularly deep neural networks, are crucial for advancing CGMD simulations of complex materials.
• This methodology enhances the predictive power of coarse-grained models in polymer science.