Point + Gaussian charge model for electrostatic interactions derived by machine learning.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Quantum Chemistry

Background:

• Point charge models are insufficient for short-range electrostatic interactions due to electron cloud overlap and charge shielding.
• Existing methods like Thole-screening and Gaussian-distributed charges address these limitations in molecular simulations.

Purpose of the Study:

• To demonstrate the practical equivalence of Thole-screening and Gaussian-distributed charge models.
• To quantitatively compare electrostatic interactions in alkali-halides using advanced computational methods and simplified models.
• To develop accurate machine learning models for alkali-halide simulations.

Main Methods:

• Symmetry-Adapted Perturbation Theory (SAPT) for high-level electrostatic interaction calculations.
• Comparison of Thole-screening and Gaussian-distributed charge models.
• Machine learning model training using the Alexandria Chemistry Toolkit.

Main Results:

• Thole-screening and Gaussian-distributed charge models exhibit numerically similar screening functions and related parameters.
• Electrostatic interactions in alkali-halides are not always weaker than point charge predictions, suggesting limitations of simple models.
• Machine learning models based on Gaussian-distributed charges accurately reproduce SAPT energies for alkali-halides.

Conclusions:

• Thole-screening and Gaussian-distributed charge models offer equivalent approaches to short-range electrostatic interactions.
• Accurate modeling of electrostatic interactions in ion pairs may require more sophisticated atomic models beyond simple point charges.
• Machine learning, combined with appropriate atomic models, provides a powerful tool for accurate molecular simulations.