DeePCG: Constructing coarse-grained models via deep neural networks

Area of Science:

• Computational chemistry
• Materials science
• Statistical mechanics

Background:

• Coarse-grained models simplify complex molecular systems.
• Traditional methods often rely on approximations like two- and three-body potentials.
• Accurate many-body interactions are crucial for reliable simulations.

Purpose of the Study:

• Introduce a general framework for constructing many-body coarse-grained potentials.
• Develop a novel approach without ad hoc approximations.
• Enhance the efficiency of molecular simulations.

Main Methods:

• Developed the Deep Coarse-Grained Potential (DeePCG) scheme.
• Utilized a neural network trained on full atomistic data.
• Preserved system symmetries during neural network training.
• Applied DeePCG to liquid water using oxygen coordinates.

Main Results:

• Achieved highly accurate many-body coarse-grained potentials.
• Demonstrated significantly faster sampling of coarse-grained configurations.
• Obtained excellent agreement in correlation functions (two-body, three-body, and higher-order) between DeePCG and atomistic models for liquid water.

Conclusions:

• DeePCG offers a powerful and generalizable framework for coarse-grained modeling.
• The method effectively captures complex many-body interactions.
• DeePCG significantly improves simulation efficiency while maintaining high accuracy.