My research combines computational physics and astrophysics to study the universe at scales from protoplanetary disks to the large-scale structure during cosmic reionization. Below you'll find an overview of the major themes and individual projects.
Simulating the epoch when the first stars and galaxies ionized the intergalactic medium, using the THESAN-ZOOM project to study Population III star formation and the long-term impact of reionization on galaxy evolution.
Using high-resolution zoom-in simulations within the THESAN framework, I study how Population III stars — the very first generation of stars — continue forming throughout cosmic reionization. These metal-free stars play a crucial role in the chemical enrichment and evolution of the early universe.
Reionization is not a uniform process — different regions of the universe are ionized at different times. I investigate how the external reionization environment imprints long-term signatures on galaxy evolution, shaping properties like star formation rates, gas content, and morphology.
During my master's thesis I simulated Milky Way–like disk galaxies including self-gravity, cooling, star formation, and stellar feedback. The model was used to analyze structure and turbulence across different phases of the ISM.
Developing and optimizing radiation transport methods for the AREPO moving-mesh code, including GPU acceleration and efficient parallel communication for exascale simulations.
Modern astrophysical simulations demand exascale computing resources. I adapted the radiation transport solver in AREPO-RT for GPU architectures and developed an efficient communication strategy to minimize overhead. This enables large-scale radiation-hydrodynamics simulations on next-generation supercomputers.
Implementing non-ideal MHD effects — Ohmic diffusion, ambipolar diffusion, and the Hall effect — enabling accurate simulations of magnetized, weakly-ionized astrophysical environments.
In dense, neutral gas the ideal MHD approximation breaks down. I implemented Ohmic and ambipolar diffusion for the first time on AREPO's moving mesh, using a least-squares gradient reconstruction on cell interfaces. Simulations of magnetized protostellar collapse demonstrate the stability and accuracy of the approach.
The Hall effect can destabilize numerical MHD schemes, and most codes address this by globally increasing numerical diffusion. I implemented a more targeted approach — locally increasing Ohmic diffusion only where necessary — achieving stable and accurate results even in complex magnetized collapse simulations.
Studying gravitational, magnetorotational, and Rossby wave instabilities in protoplanetary disks using the shearing box approximation implemented in AREPO.
Cool, self-gravitating disks can fragment or develop a turbulent quasi-steady state depending on the cooling efficiency. I implemented a self-gravity solver for the shearing box in AREPO using the TreePM method and performed an extensive study of fragmentation and turbulence across different cooling strengths, box sizes, and resolutions.
The MRI is a leading candidate for driving angular momentum transport in ionized accretion disks. Using the shearing box in AREPO, I performed simulations with different magnetic field configurations and compared the results with static grid codes, finding good agreement and interesting differences in the numerical Prandtl number behavior.
The Rossby wave instability arises in regions of strong density or temperature gradients in accretion disks, leading to the formation of vortices. I implemented a Fourier filter for runtime mode analysis and showed that AREPO on an arbitrary mesh achieves accuracy comparable to polar-grid codes optimized for this problem.
The shearing box approximation models a small patch of a rotating disk in a co-rotating frame. I implemented this framework for ideal MHD in AREPO, including Coriolis and tidal forces, shearing-periodic boundary conditions, and demonstrated second-order convergence in a range of test problems.
Improving the accuracy of moving-mesh hydrodynamics through higher-order flux integration, contributing to major simulation codes including AREPO and GADGET-4, and leveraging vectorized and GPU-accelerated kernels.
Moving-mesh codes can suffer from grid noise in shear flows. I identified that this noise originates from the first-order flux integration and implemented higher-order Gauss–Legendre rules to eliminate it. To manage the computational overhead, I developed optimized vector-instruction-based kernels, limiting the performance cost to roughly 30% for ideal MHD simulations.
GADGET-4 is the open-source successor to the widely used GADGET-2. I contributed the SPH implementation — supporting both density-based and pressure-based formulations with time-dependent artificial viscosity — along with optimized computational kernels using explicit vector instructions.
A Bonnor–Ebert sphere is a self-gravitating, isothermal gas cloud in pressure equilibrium. I derived a new analytical stability criterion for the case of an additional one-sided pressure from a stellar wind, confirmed it with SPH simulations, and implemented a HEALPix-based stellar wind feedback model.
In my bachelor's thesis I analyzed the stability of an inclined layer of liquid mercury to convection. Using Galerkin methods and pseudo-spectral simulations I characterized linear instability thresholds and nonlinear pattern formation as a function of inclination angle for the low-Prandtl-number regime.