Macro-Dipole-Constrained Learning of Atomic Charges for Accurate Electrostatic Potentials at Electrochemical...

Area of Science:

• Computational Chemistry
• Electrochemistry
• Materials Science

Background:

• Extracting interfacial charge distributions from ab initio molecular dynamics is difficult due to thermal fluctuations obscuring weak electric fields.
• Standard methods struggle to reliably determine atomic charges in complex electrochemical systems.

Purpose of the Study:

• To introduce a novel macro-dipole-constrained charge partitioning scheme, SMILE (Scalar Macro-dipole Integrated LEarning).
• To enable accurate inference of atomic charges using readily available simulation data.
• To improve the analysis of interfacial charge distributions in electrochemical reactions.

Main Methods:

• Developed SMILE, a charge partition scheme using instantaneous atomic coordinates and total dipole moment.
• SMILE constrains the macro-dipole, preserving global electrostatics and local potentials without explicit partitioning.
• Validated the method on nanoconfined water, Mg^{2+} dissolution, and a Mg vicinal surface under anodic bias.

Main Results:

• SMILE successfully eliminates qualitative errors in charge decomposition compared to unconstrained methods.
• The method accurately infers atomic charges, providing reliable interfacial charge distributions.
• Demonstrated effectiveness across diverse electrochemical interfaces.

Conclusions:

• SMILE is a computationally inexpensive and data-efficient method for charge partitioning in electrochemical simulations.
• Enables the development of charge-aware machine learning potentials for advanced simulations.
• Facilitates bias-controlled, nanosecond-scale simulations of realistic electrochemical systems.