Structural Coarse-Graining via Multiobjective Optimization with Differentiable Simulation.

Area of Science:

• Molecular Simulations
• Soft Matter Physics
• Computational Chemistry

Background:

• Structure-based coarse-graining (CG) is vital for simulating large soft matter systems.
• Current methods optimize CG interactions by matching static correlation functions from all-atom (AA) models.
• Developing transferable CG models across temperatures remains a challenge.

Purpose of the Study:

• Introduce a versatile differentiable coarse-graining (DiffCG) method.
• Construct robust and transferable CG models for polymer melts.
• Demonstrate the ability to reproduce structural and thermodynamic properties across temperatures.

Main Methods:

• Combine multiobjective optimization with differentiable simulation.
• Iteratively optimize effective potentials to match multiple target properties.
• Develop a CG model for polystyrene (PS) melts by optimizing bonded and nonbonded potentials.

Main Results:

• The DiffCG approach successfully created a CG-PS model that reproduces AA structural characteristics and thermodynamic pressure.
• The optimized CG model demonstrates transferability across temperatures (400-600 K).
• The model accurately predicts radial distribution functions and density at various temperatures, including unseen ones.

Conclusions:

• DiffCG offers a promising route for developing accurate and transferable CG models for complex soft matter.
• Multiobjective optimization combined with differentiable simulation enhances CG model development.
• The developed temperature-transferable CG-PS model significantly advances polymer simulation capabilities.