Grand-Canonical Equivariant Neural Potentials for Electrochemical Interfaces

Area of Science:

• Computational chemistry
• Materials science
• Electrochemistry

Background:

• Electrochemical reactions under constant potential are crucial for energy storage, catalysis, and corrosion.
• Conventional machine learning potentials struggle to model voltage insensitivity, limiting simulations of potential-dependent interfacial phenomena.
• A unified framework incorporating grand-canonical constraints is needed for accurate, scalable modeling.

Purpose of the Study:

• To present a novel constant-potential, E(3)-equivariant message-passing neural network (CPMPNN).
• To integrate grand-canonical electronic structure principles with a dynamic excess-charge parameter for improved accuracy.
• To enable scalable simulations of potential-dependent interfacial phenomena.

Main Methods:

• Developed a CPMPNN integrating grand-canonical principles and a global excess-charge parameter.
• Utilized a multihead attention mechanism for dynamic charge redistribution.
• Employed an E(3)-equivariant graph neural network to encode atomic geometry, preserving Euclidean symmetry.

Main Results:

• CPMPNN demonstrated first-principles accuracy comparable to grand-canonical DFT.
• Achieved a three-orders-of-magnitude computational speedup compared to DFT.
• Successfully modeled potential-modulated thermodynamics, charge distribution, and transition states in electrocatalysis.

Conclusions:

• CPMPNN provides a transferable framework for operando modeling of electrified interfaces.
• Enables mechanistic insights into potential-controlled electrocatalysis.
• Bridges first-principles accuracy with molecular dynamics scalability for advanced simulations.