Bias Conditioning and Long-Range Electrostatics for Consistent Machine-Learning Potentials of Electrified Interfaces
Machine-learning interatomic potentials (MLIPs) of electrified interfaces need the bias as input, but it is unclear whether a scalar electrode charge suffices and whether short-range models capture the non-local response. We add charge and field conditioning to short- and long-range MLIPs. State-of-the-art field-conditioned short-range models overpolarize thick water films nearly twofold and miss work-function shifts at stepped Pt. Our long-range model captures both, needs only the electrode charge at metals as input, and reproduces ab initio simulations of the Pt(111)--water interface.
Authors
- Philip Loche (ORCID: https://orcid.org/0000-0002-9112-0010)
- Christopher Johannes Stein (ORCID: https://orcid.org/0000-0003-2050-4866)
- Patrick Rinke (ORCID: https://orcid.org/0000-0003-1898-723X)
- Lucas B. T. de Kam (ORCID: https://orcid.org/0009-0000-3943-6619)
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.48550/arxiv.2611.21026
- Primary Topic
- Machine Learning in Materials Science
- Type
- preprint