Machine Learning (ML)-based force fields offer the potential to achieve quantum-level accuracy across spatiotemporal scales. This paper presents a method that fuses Density Functional Theory (DFT) calculations and experimental data to train an ML potential for titanium. By combining both data sources, the model simultaneously satisfies various target objectives, leading to higher accuracy compared to single-source models. This approach corrects DFT inaccuracies while minimally affecting other properties. The methodology is widely applicable for generating highly accurate ML potentials for various materials.

Sebastien Röcken, Julija Zavadlav