Reionization in the Illustris Universe and Novel Numerical Methods

Abstract

Numerical simulation methods provide powerful tools to study astrophysical processes in cosmic structure formation. Further advancing their utility requires to improve their accuracy and to account for more of the relevant physics. In this thesis, we pursue this goal by developing novel numerical approaches for studying the epoch of cosmic reionization and for simulating hydrodynamical flows with accurate higher-order methods. We introduce a novel GPU-based radiative transfer code designed to study cosmic reionization. Our implementation of radiative transfer uses either a cone-based or a moment-based advection method and is able to accurately follow the epoch of cosmic reionization in postprocessing. To validate our methods, we consider a number of standard reionization test problems. We then apply our implementation to the state-of-the-art Illustris simulation of galaxy formation. We find that the stellar populations of the galaxies forming in Illustris are able to reionize the universe at an epoch consistent with observations. In particular, our results reproduce Lyman-alpha constraints for the reionization history and yield an optical depth towards the surface of last scattering of tau = 0.065, which is in reassuring agreement with recent Planck observations. In our simulations, reionization proceeds ‘inside-out’ and predicts an evolving size distribution of ionized bubbles that is characterized by ever larger maximum sizes of the bubbles with time, whereas the abundance of small bubbles stays relatively constant over an extended period until reionization is completed. The results obtained with both of our radiative transfer schemes are rather similar, suggesting that the details of these methods are not a major source of uncertainty. We also present the implementation of a novel hydrodynamics solver based on a discontinuous Galerkin method. To this end we design and add an adaptive mesh refinement module to the hydrodynamical moving-mesh code AREPO. As a first application of this new tool, we discuss simulations of driven subsonic turbulence. There, we find an enlarged inertial range for our discontinuous Galerkin simulations compared with finite volume methods for an equal number of degrees of freedom. Furthermore, the overall compute time to solution at a prescribed accuracy is shorter as well for the new discontinuous Galerkin code, demonstrating the potential of this technique for future astrophysical applications.