Atomistic simulations of electrolyte solutions at metallic interfaces

C. J. Stein

The interface between metallic electrodes and aqueous salt solutions poses a severe challenge for atomistic simulations. This is due to several competing and subtle effects, such as the hydrogen bond network of water, charge-transfer to the electrode, ion solvation, electrostatic screening, and hydrophobic effects, that must be accurately incorporated to arrive at a realistic description. The different time scales of the relevant processes, such as ion diffusion and elementary reaction steps at the electrode surface, require long simulation times and enhanced sampling methods. Two methods have emerged recently that might replace density functional theory-based ab initio molecular dynamics simulations because of their significantly lower computational cost. These methods are Grimme’s extended tight-binding methods[1] and machine-learned interatomic potentials (MLIPs) such as MACE[2]. I will discuss how fine-tuning of the parameters of tight-binding methods improves the prediction of dynamic properties in salt solutions. [3] Concerning MLIPs, I will discuss what accuracy can be achieved at what computational cost and how suitable active-learning training methods lead to improved yet lightweight MLIPs for the specific application of atomistic simulations of electrolyte solutions at metallic interfaces. [4] In the last part of my talk, I will outline how we aim to include the effect of bias potentials on the electrode surface in an efficient manner.

1. C. Bannwarth; S. Ehlert; S. Grimme, J. Chem. Theory Comput. 2019 15, 1652–1671.
2. I. Batatia; D. P. Kovács; G. N. C. Simm; C. Ortner; G. Csányi, NeurIPS 2022 35, 11423–11436.
3. e.g S. Moradi; P. Dabbaghi; C. J. Stein, ChemPhysChem 2025 e202500463.
4. e.g a) A. Beiersdorfer; L. Hetzel; C. Staake; F. Deißenbeck; C. J. Stein, chemRxiv 2025 DOI:10.26434/chemrxiv-2025-z8dhb, b) L. Hetzel; C. J. Stein, chemRxiv 2025.