PhyNEO: A Neural-Network-Enhanced Physics-Driven Force Field Development Workflow for Bulk Organic Molecule and...

Area of Science:

• Computational Chemistry
• Materials Science
• Polymer Science

Background:

• Accurate molecular dynamics (MD) simulations require generalizable force fields for organic polymers and biomolecules.
• Conventional empirical force fields lack accuracy due to neglecting key physics (e.g., polarization, many-body dispersion) and rely on limited experimental data for parameterization.
• This limits their transferability to new systems.

Purpose of the Study:

• To introduce PhyNEO, a novel, general, and fully ab initio force field construction strategy.
• To develop a method capable of generating accurate bulk potentials using only quantum chemistry data from small clusters.
• To overcome limitations of conventional force fields in accuracy and transferability.

Main Methods:

• PhyNEO employs a hybrid physics-driven and data-driven approach.
• It carefully separates long-/short-range and nonbonding/bonding interactions.
• Utilizes quantum chemistry data from small molecular clusters for parameterization.

Main Results:

• PhyNEO demonstrates superior accuracy in both microscopic and bulk properties for poly(ethylene oxide) and polyethylene glycol systems compared to conventional force fields.
• The method shows increased data efficiency and scalability for machine learning models.
• Successfully mitigates limitations of pure data-driven methods in handling long-range interactions.

Conclusions:

• PhyNEO offers a promising framework for developing advanced, accurate, and transferable force fields for diverse organic molecular systems.
• The strategy enhances the reliability of molecular dynamics simulations.
• Validates the effectiveness of combining physics-based principles with data-driven approaches.