COSMOS: Clustering, Orbital- and Simulation-based Modeling for Observational Surveys

• Applicant: Dr. Tobias Buck, since 1/2026
• Subject Area: Astrophysics and Astronomy ; Methods in Artificial Intelligence and Machine Learning
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 573580088

The COSMOS project aims to reconstruct the formation and accretion history of the Milky Way using novel data-driven methods in galactic archaeology. At the center of the project is the development of robust, unsupervised clustering algorithms (AstroLink, FuzzyCat) and simulation-based inference (SBI) techniques to reliably identify and physically interpret stellar substructures in large astrometric and spectroscopic surveys such as Gaia, APOGEE, SDSS-V, and 4MOST. Scientifically, the project seeks to infer the mass distribution, orbital properties, and infall times of disrupted satellite galaxies that have shaped the structure of the Milky Way’s stellar halo. We also investigate how chemical and kinematic information can be used to trace these past events, and to what extent chemical abundance patterns alone are sufficient to draw conclusions about the Galaxy’s formation history and the underlying cosmological models. This latter point in particular will help to inform extragalactic surveys where kinematic data will not be available and will put the Milky Way in context with the larger galaxy population. Methodologically, COSMOS will develop and employ differentiable, GPU-accelerated N-body simulations to generate realistic, controllable models of accreted systems within a dynamically evolving Milky Way potential. All resulting tools and models will be made available as sustainable, open-source software through a close collaboration with the Scientific Software Center at Heidelberg University. COSMOS thus integrates cutting-edge astrophysics, numerical simulation, machine learning, and scientific software development in a high-impact framework with strong relevance for the international research community.

• DFG Programme: Research Grants
• Ehemaliger Antragsteller: Dr. William Oliver, Ph.D., until 1/2026