Next generation Monte Carlo radiative transfer: Reinforcement Learning

• Applicant: Professor Dr. Sebastian Wolf
• Subject Area: Astrophysics and Astronomy
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 561178007

Monte Carlo radiative transfer (MCRT) plays an essential role in simulating the physical properties and observables of astrophysical objects. However, the increasing capabilities of modern telescopes, especially in terms of resolution and sensitivity, and the increasing availability of complementary observational data require simulation approaches that allow for greater complexity, higher resolution and the incorporation of additional physical processes. These extensions often lead to immense computational requirements that exceed the capabilities of traditional MCRT methods. The aim of this project is to integrate reinforcement learning (RL) methods into MCRT simulations in order to extend their applicability to more complex astrophysical systems and observational data. Specifically, the goals of this project are (a) to develop a first generation of RL-enhanced MCRT algorithms that can effectively guide photon packets by optimizing both computational performance and accuracy, and (b) to apply these algorithms to exemplary, challenging problems in planet formation and exoplanet characterization, which currently require extremely high computational effort. Achieving these goals will significantly improve the efficiency and precision of MCRT simulations and enable previously infeasible radiative transfer simulations to be tackled.

• DFG Programme: Research Grants