Enhancing Empirical Energy Functions Using Physics- and Machine Learning-Based Extensions: Structure, Dynamics and...

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Physical Chemistry

Background:

• Empirical energy functions are crucial for molecular simulations.
• Accurate representation of electrostatic interactions, like σ-holes, is challenging with traditional atom-centered point charges (PCs).
• Halogenated benzenes and chlorinated phenols present specific modeling challenges due to their electronic structures.

Purpose of the Study:

• To evaluate the impact of replacing components of empirical energy functions.
• To assess the performance of minimal distributed charge models (MDCM) versus point charges (PCs).
• To investigate the utility of neural network-based energy functions for molecular simulations.

Main Methods:

• Assessed effects of replacing individual contributions in empirical energy functions.
• Employed minimal distributed charge models (MDCM) and neural network-based energy functions.
• Compared simulation results with experimental data for hydration free energies and infrared spectroscopy.

Main Results:

• MDCM models overestimated hydration free energies without reparametrization of van der Waals parameters.
• Scaling van der Waals ranges by 10-20% brought most halogenated benzene and chlorinated phenol results within experimental error.
• Neural network energy functions, after van der Waals parameter adaptation, also achieved experimental agreement; they slightly improved infrared spectroscopy predictions.

Conclusions:

• Refining empirical energy functions for specific applications enhances simulation quantitative accuracy and physical basis.
• Advanced electrostatic models like MDCM require careful parameterization, particularly for van der Waals interactions.
• Empirical energy functions have reached a high level of maturity for the studied molecular systems.