Machine Learning the Energetics of Electrified Solid-Liquid Interfaces.

Area of Science:

• Computational Chemistry
• Materials Science
• Surface Science

Background:

• Understanding the energetics of electrified metal surfaces is crucial for catalysis and electrochemistry.
• Existing methods struggle to efficiently incorporate the effects of applied electrical bias on surface properties.

Purpose of the Study:

• To develop a novel machine-learning (ML) approach for accurately calculating the energetics of electrified metal surfaces.
• To extend ML interatomic potentials to include finite bias effects up to second order.

Main Methods:

• A response-augmented ML approach using local descriptors to learn work functions.
• Incorporation of Born effective charges to stabilize the ML model.
• Extension of ML interatomic potential architectures to second-order bias effects.

Main Results:

• The ML approach efficiently captures the energy changes due to bias charges.
• Demonstrated accurate prediction of energetics for electrified metal surfaces.
• Rationalized the pH dependence of adsorption site preference for OH on Cu(100).

Conclusions:

• The developed ML method provides an efficient and accurate way to study electrified metal surfaces.
• The findings offer insights into charge-induced site switching phenomena affecting molecular adsorption.
• This work enables better understanding and design of electrochemical systems.