Dresden 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

O: Fachverband Oberflächenphysik

O 18: New methods: Theory – Poster

O 18.6: Poster

Monday, March 9, 2026, 18:00–20:00, P2

Calculating surface energy using machine learning interatomic potentials — •Friedrich Neumann¹, Tom Barnowsky^2,3, Rico Friedrich^2,3, and Jens Kortus¹ — ¹TU Bergakademie Freiberg - Institut für Theoretische Physik, Leipziger Str. 23, 09599 Freiberg — ²Technische Universität Dresden - Theoretische Chemie, 01062 Dresden — ³Helmholtz-Zentrum Dresden-Rossendorf e.V., Bautzner Landstraße 400, 01328 Dresden

The surface energy of a material is a fundamental property that governs the behaviour of solid interfaces in a wide range of phenomena. While density functional theory (DFT) provides highly accurate values, its computational scaling limits its applicability for large systems and systematic surface-energy screening. In this work, a workflow was developed to estimate surface energies using machine-learning interatomic potentials (ML-IAPs). Atomic Cluster Expansion (ACE) models were fitted to DFT-based molecular-dynamics data for Al, Pt, and Au, and effective two-body contributions were extracted from the resulting potentials. These were combined with a bond-counting approach to compute surface energies and Wulff constructions. Although in our present ACE model the absolute magnitudes of the pair interactions are influenced by many-body interaction, preventing quantitative reproduction of surface energies, the approach still captures general trends and identifies the dominant facets in the Wulff constructions. In this specific formulation, the results illustrate how many-body interaction can affect the apparent pair contribution.

Keywords: Machine learning; interatomic potentials; surface energy; Wulff construction