Applicant
Dr. Christian Fromm
Computational Astrophysics
Universität Würzburg
Project Overview
Relativistic jets launched from supermassive black holes are among the most powerful phenomena in the Universe. Despite significant progress in observing and modelling relativistic jets, the detailed physical processes responsible for launching and powering these jets are still not fully understood. In recent years, major advances have been achieved through general relativistic magnetohydrodynamic (GRMHD) simulations. However, to compare simulations with observations, the plasma properties from the simulations must be converted into observable radiation using general relativistic radiative transfer (GRRT) calculations.
Current GRRT calculations usually employ the so-called fast-light approximation, where the plasma is assumed to remain unchanged while photons travel through the simulation domain. Although computationally efficient, this approximation neglects the finite speed of light and can become inaccurate when modelling time-dependent emission. A more realistic treatment requires slow-light radiative transfer, where the evolution of the plasma is taken into account during photon propagation. However, such calculations require access to many simulation snapshots and lead to large memory and data I/O demands.
The aim of this project is to extend the GRRT code BHOSS to enable efficient slow-light calculations for large GRMHD simulations. The project is divided into two main work packages. The first focuses on reducing memory usage and improving data access. Photon geodesics will be precomputed and stored so that they can be reused in repeated radiative transfer calculations. In addition, a native adaptive mesh refinement (AMR) reader will be implemented, allowing direct access to the hierarchical grid structure used by modern GRMHD simulations. The central development will be a lazy-loading approach based on HDF5 data format. Instead of loading complete simulation snapshots into memory, only the small spatial regions and time intervals required along a photon path will be read from disk when needed and released immediately afterwards. This keeps the memory requirements low and largely independent of the total simulation size.
The second work package focuses on improving computational performance. Since individual photon rays can be treated independently, radiative transfer calculations are naturally parallel. We will therefore extend the current OpenMP implementation of BHOSS to a hybrid MPI-OpenMP version. MPI will distribute photon rays across compute nodes, while OpenMP will provide parallelisation within each node.
Overall, this KONWIHR project will deliver the computational framework necessary for fully time-dependent, slow-light radiative transfer simulations, while bridging the gap between large-scale astrophysical plasma simulations and next-generation observations of accreting black holes and time-domain phenomena.