Erasing dielectric breakdown artifacts to machine-learn charged Pt-water interfaces

Area of Science:

• Computational Chemistry
• Materials Science
• Electrochemistry

Background:

• The response analysis in z-orientation (RAZOR) model uses perturbation theory for atomistic simulations of electrified interfaces.
• Previous work demonstrated RAZOR's accuracy for adsorbates on metallic surfaces in implicit solvents.
• Real-world electrochemical applications necessitate explicit inclusion of solvent layers, such as water (H2O).

Purpose of the Study:

• To benchmark the performance of the RAZOR model for the Pt(111)-H2O interface, including explicit water layers.
• To assess RAZOR's ability to reproduce ab initio molecular dynamics findings for bias-induced changes in interfacial properties.

Main Methods:

• Machine learning the energy and force response to applied bias charges using the RAZOR model.
• Benchmarking against ab initio molecular dynamics simulations of the Pt(111)-H2O interface.
• Implementing a specialized training procedure to prevent learning artifacts from dielectric breakdown and charge transfer.

Main Results:

• RAZOR reliably reproduces ab initio molecular dynamics results for bias-induced changes in H2O density and orientation profiles.
• The model accurately captures bias-induced changes in the interfacial potential.
• A specific training strategy successfully avoids artifacts related to water dielectric breakdown and charge transfer.

Conclusions:

• The RAZOR model, with appropriate training, can accurately describe electrified interfaces with explicit solvent layers.
• RAZOR provides quantitative predictions within a ±20 μC cm⁻² window around the neutral-charged cell, suitable for many electrochemical studies.
• This advancement extends the applicability of RAZOR to more realistic electrochemical systems.