Overview

Cosmic reionization is the last major phase transition of the universe — the epoch when light from the first stars and galaxies ionised the intergalactic hydrogen and reshaped subsequent galaxy formation. My goal is to understand this transition self-consistently, modelling galaxies, the radiation they emit, and the intergalactic medium they live in within a single simulation framework.

As PI of the Lumina project I lead a new generation of cosmological radiation-hydrodynamic simulations targeting reionization, built on the moving-mesh code AREPO with the GPU-accelerated radiation-transport solver I developed during my time at MIT. Lumina evolves a 500 cMpc box from the cosmic dark ages down to z = 3 with nearly half a trillion resolution elements — large enough to capture both the galaxy populations that drive hydrogen reionization and the rare quasars responsible for He II reionization, while still resolving the sources themselves.

Four linked goals shape the project: tracking how galaxies form, grow, and respond to radiation feedback from cosmic dawn to z = 3; measuring when and where hydrogen reionization occurs and which galaxies drive it; modelling AGN-driven He II reionization and its thermal imprint on the intergalactic medium; and producing predictions for high-redshift galaxy surveys, the Lyman-α forest, 21-cm cosmology, line-intensity mapping, and quasar absorption spectroscopy. Lumina also adopts improved initial conditions with separate baryon and dark-matter transfer functions and their relative streaming velocity — physics that single-fluid initial conditions miss at high redshift.

The science I care about is rate-limited by the codes available to do it. Reionization, SIDM, and small-scale MHD all demand simulations that are simultaneously high-resolution, large-volume, and physically rich — a combination that is only achievable on exascale hardware, with carefully designed numerical methods underneath.

For reionization specifically, the radiation-transport step is the bottleneck. I rewrote AREPO-RT to run on GPUs and developed a communication strategy that avoids per-step host–device synchronization, demonstrating efficient strong scaling on pre-exascale systems and opening the door to the next generation of radiation-hydrodynamic simulations (Zier et al. 2024).

Earlier I traced a long-standing source of spurious noise in moving-mesh shear flows to the first-order flux integration along cell interfaces and replaced it with higher-order Gauss–Legendre quadrature, eliminating the artefact; optimised vector kernels limit the overhead to ≈30 % for ideal MHD (Zier & Springel 2022). I am also one of the co-developers of GADGET-4 — the open-source successor to GADGET-2 — where I contributed the SPH implementation (density- and pressure-based formulations with time-dependent artificial viscosity) and vectorised gravity and SPH kernels (Springel, Pakmor, Zier & Reinecke 2021).

Accretion disks are the workhorses through which gas accumulates onto protostars and black holes; the angular-momentum transport that regulates this flow comes from disk-scale instabilities. Using the shearing-box approximation in AREPO I performed convergence studies of the magnetorotational instability across magnetic-field configurations, comparing the moving-mesh results to established static-grid codes and characterising the numerical Prandtl-number behaviour (Zier & Springel 2022).

For self-gravitating disks I added a TreePM solver for the shearing box and ran an extensive study of fragmentation and gravito-turbulence across cooling strengths, box sizes, and resolutions, mapping out where cool disks settle into a quasi-steady turbulent state versus fragmenting into bound clumps (Zier & Springel 2023).