AG Fromm

To study the dynamics of the magnetised plasma around black holes and the jet launching and propagation we perform 3D General Relativistic Magneto-Hydrodynamic (GRMHD) simulations.  During the simulations we include electron thermodynaimcs i.e. the heating of electrons via magnetic reconnection as well as radiative cooling. In more recent simulations we include the effect of a physical resisitivity. For the jet propagation simulations we include realistic ambient medium conditions. Our simulations are performed with the BHAC and KHARMA codes which we modify and continously expand to our needs.  Below we provide a 3D rendering of our simulations including jet launching from a rotating black hole and the propagation of the jet into the ambient medium. The magnetic field lines are shown in gray and in red. Notice the growth of different instabilities including kinks and Kelvin Helmholtz instabilities. 

Understanding how black holes launch and power relativistic jets requires large-scale magnetohydrodynamic simulations that are often computationally expensive. My recent research explores the use of physics-informed neural operators as AI surrogates for the Black Hole Accretion Code (BHAC), enabling key plasma processes to be modeled with significantly greater efficiency. We demonstrated that incorporating the governing physical equations into the training process allows neural operators to recover complex phenomena such as magnetic reconnection and plasmoid formation, even when only sparse simulation data are available. In parallel, we developed transformer-based surrogate models capable of capturing the evolution of relativistic jets directly on adaptive mesh refinement grids. These results highlight the potential of AI-driven surrogates to accelerate numerical astrophysics and open new opportunities for exploring the physics of black hole accretion and jet formation across large parameter spaces.

In order to compare our 3D simulations of accreting black holes and relativistic jets with observations across the electro-magnetic spectrum we perform radiative transfer (RT) calculations. During the RT calculations we include different electron distribution functions and evolve the radiating particles including sub-grid models inspired from kinetic particle-in-cell (PIC) simulations.  Our main emission process is synchrotron radiation generated by relativistic electrons gyrating the magnetic field. However, we also include bremsstrahlung from the accretion disk and inverse compton scattering to compute the high energy emission i.e. X-rays and Gamma-rays. Below we show as example averaged images of MAD models for SgrA* and their broad band spectral energy distribution (SED).

For the direct comparison with observations especially with high-resolution VLBI observations it is necessary to generate synthetic data including the technical capabilities of the observing array, i.e. the limited coverage of the Fourier plane, the temperature of the receivers and the properties of the atmosphere i.e. the water vapor. In addition to the data generation, we also carry out image reconstruction using advance entropy based reconstruction algorithms. Our methods and tools can also be used to perform array optimisation studies (searching for the best suited locations of additional telescopes) or optimising the orbital parameters of space-based telescopes. Below we show the reconstruction of an M87-like object using the next generation VLA (ngVLA) in combination with telescopes in Germany at an observing frequency of 96 GHz.  

We use of archival Very Long Baseline Interferometry (VLBI) observations black holes from the MOJAVE survey  and the Boston Blazar Program. We re-analyse these observations by applying modern imaging algorithms extracting their spectro-polarimetric properties. The observations are accompanied by our black hole monitoring program TELAMON and  future EHT observations of nearby SMBH systems as well as by  newly proposed observations of candidate neutrino-associated accreting black hole systems. These evolving data sets form the observational foundation of our research and are used during the comparison with the developed numerical models. Given the complexity of the numerical simulations and synthetic data generation, a direct feed-back loop between the observational data and the numerical models is challenging. However, my developed \texttt{GENA} code based on non-gradient evolutionary algorithms (EAs) showed that it is possible to incorporate obserational data in the numerical modelling. Furthermore,  we explore the capabilities of EAs and combine them with artificial neural networks and GANs i.e. generating black holes images via neural networks. Below we show a comparison between classical image reconstruction methods CLEAN) and maximum entropy based methods (MEM) for  Mrk501 at 43 Ghz  based on archival data from the Boston Blazar program.