Evaluation of Representations and Response Models for Polarizable Force Fields.

Area of Science:

• Computational chemistry
• Molecular modeling
• Physical chemistry

Background:

• Classical simulations of condensed-phase systems need accurate electronic polarization models.
• Polarization models consist of representation and response components.
• Common representations include charge redistribution and point dipoles.

Purpose of the Study:

• To analyze the accuracy of different polarization models by decoupling representation and response.
• To compare the maximal achievable accuracy of charge redistribution and point-dipole representations.
• To evaluate the performance of common response functions for point-dipole models.

Main Methods:

• Decoupling polarization representations from their response models.
• Comparing QM electrostatic potentials (ESPs) with model predictions for small molecules.
• Optimizing polarizabilities to fit QM ESPs for point-dipole models.
• Testing self-consistent and direct inducible point polarizability response functions.

Main Results:

• Atom-centered dipole models achieve higher accuracy than charge redistribution models in replicating QM ESPs.
• The widely used inducible dipole response model significantly degrades accuracy for point-dipole representations.
• Optimal charge redistribution can outperform inducible dipole models, even for out-of-plane polarization.
• The accuracy of inducible dipole models is limited by the response function, not solely the dipole representation.

Conclusions:

• Improvement of polarization response models is crucial for enhancing the accuracy of molecular simulations.
• Current inducible dipole response functions do not fully exploit the potential accuracy of the point-dipole representation.
• Further research into advanced response functions is needed for more reliable classical simulations.