Prof. Bjørk Hammer

Aarhus University, Denmark
Bjørk Hammer is a professor at Department of Physics and Astronomy at Aarhus University since 2000. He obtained his PhD degree in 1993 from the Technical University of Denmark, and has previously worked at The Fritz-Haber-Institute of the Max-Planck-Gesellschaft, Berlin, Germany, at the Technical University of Denmark, and at Aalborg University, Denmark.

Talk title: Surface structure from machine learning
In recent years, machine learning interatomic potentials (MLIPs) have been developed to a stage, where they are replacing Density Functional Theory (DFT) in e.g. molecular dynamics calculations [1]. For structure determination it remains, however, a challenge to formulate active learning protocols, where new DFT training data are provided to improve the MLIPs, while the searches progress and discover new atomic motifs [2]. In this work, strategies for how to collect the training data and for how to construct the MLIPs will be presented. The training data may e.g. be constructed using the same technology as in image generation (diffusion models) [3] while the construction of the MLIPs may involve neural networks and Gaussian Process Regression methods [4]. Examples of surface structures solved will include complex ultra-thin surface oxides on Ag(111) [5] and Pt₃Sn [6]. Our computer codes are available as an open-source python package, AGOX [7, 8].