Datensatz

Zusammenfassung

The efficiency and reliability of artificial intelligence (AI)-driven physics, chemistry, biophysics, materials science and engineering depends on the acquisition of sufficient, high-quality data. Due to its all-electron, full potential treatment, and its scalability to larger systems without precision limitations, FHI-aims provides accurate ab initio data from a wide range of computer simulations, such as electronic-structure calculations and molecular dynamics. To leverage the capabilities of AI models, workflows that seamlessly integrate AI tools with FHI-aims are essential. These workflows automate the acquisition of data and their use by AI. Thus, they facilitate the iterative data exchange between AI models and simulations, allowing FHI-aims to be used as a powerful AI-guided calculation engine. Also, interpretable AI models aid in analyzing the generated data. Furthermore, AI complements ab initio studies as it enables to perform simulations at larger time and length scales. In turn, also the AI models must incorporate the physics required for an accurate representation of the ab initio data. This contribution highlights workflows developed to integrate FHI-aims with AI and future challenges.