Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 97: Solid-liquid interfaces: Reactions and electrochemistry III
O 97.10: Talk
Friday, March 13, 2026, 12:00–12:15, TRE/PHYS
Machine Learning the Energetics of Electrified Solid-Liquid Interfaces — •Nicolas Bergmann1, Nicéphore Bonnet2, Nicola Marzari2, Karsten Reuter1, and Nicolas G. Hörmann1 — 1Fritz-Haber-Institut der MPG, Berlin — 2Theory and Simulation of Materials, EPFL, Lausanne
Machine learning interatomic potentials (MLIPs) accelerate many aspects of computational chemistry. However, MLIPs fail to describe the biases introduced at applied potential conditions for typical electrocatalytic systems, where effects of the inner double layer play a critical role [1]. Here, we present the "Response Analysis in z-ORientation" (RAZOR) model [2], which machine-learns the work function, the first-order energy change to bias charges q. RAZOR is stabilized by additionally learning the atomic force derivative to q, equivalent to Born effective charges. The approach extends MLIPs to the variable charge case, by adding bias-induced energy and force changes to traditional zero-bias MLIPs. This enables large-scale molecular dynamics simulations at finite bias. We demonstrate RAZOR’s capabilities by investigating OH adsorption on Cu(100) and explicit Pt(111)-H2O interfaces, faithfully recreating ab initio and experimental results.
[1] N. Bergmann et al., J. Chem. Theory Comput. 19, 8815 (2023).
[2] N. Bergmann et al., Phys. Rev. Lett. 135, 146201 (2025).
Keywords: Machine Learning; Electrochemistry; DFT; Computational Chemistry; Modeling