PI: Alexandre Refregier (ETHZ)

January 1, 2025 – December 31, 2027

Approximately 380’000 years after the Big Bang, the temperature in our Universe dropped sufficiently to allow the formation of neutral hydrogen and helium atoms. Over time, gravitational instability led to the formation of compact structures that hosted the first luminous objects, such as primordial stars, galaxies, and outflow radiation from supermassive black holes. Only 200 million years after the Big Bang, the Universe transitioned once again from an initial neutral and cold state to a highly ionized and warm plasma, marking the end of the period devoid of any light, the cosmological dark ages, and, instead, the beginning of the era of galaxies. This transition is often divided into the Cosmic Dawn (CD) and the Epoch of Reionization (EoR).
The former indicates the time when the first luminous objects appear. In contrast, the latter encompasses the period in which the radiation of these sources propagates into the vast intergalactic medium (IGM), altering its thermal and ionization state to form latter-day structures. The Square Kilometre Array Observatory (SKAO) project is an international effort to build the world’s largest radio telescope. The low-frequency component of this telescope, SKA-Low2, promises to provide direct evidence of the Universe’s transformation throughout the Cosmic Dawn (CD) and the Epoch of Reionization (EoR). Therefore, SKAO will revolutionize our understanding of the Universe and answer the pressing questions: What are the primary sources of reionization? When does reionization begin and end? What is the topology of the ionized region? What can be learned about early galaxies from the reionization and vice-versa? How does reionization later affect galaxy formation?
A new generation of precursor telescopes is also being constructed, and the first results are starting to be reported. Numerical simulations provide a powerful method to study the CD and the EoR. They, however, require careful physical treatment, which can only be achieved with Radiative Transfer (RT) modeling. In
this approach, the propagation of photons from the primordial sources into the intergalactic medium (IGM) is computed via raytracing algorithms. Meanwhile, the chemistry evolution of the primordial gas is tracked by solving high-order differential equations (ODE).
CD and EoR are inherently multi-scale problems, requiring codes that simulate large-scale dynamics while resolving physical processes that occur on galactic scales. Several authors have tackled this problem. Ongoing works such as the CoDa3 and the Thesan RD4 projects, require a few thousand nodes with graphical processing units (GPUs) for a computational cost of millions of GPU and CPU-core hours. At the moment, they can only simulate a small portion of the observed Universe, and the large-scale structures can be simulated only at the expense of lower resolution. Therefore, current RT codes will struggle to concurrently account for the large volumes and the physical process that occurs on the small scales detectable with SKA-Low observations.
With this project, we propose to develop the first code that is not only able to simulate the largescale required by future SKA-Low observations but also includes a complete treatment of all the relevant physical processes occurring on the galactic scale. The most time-consuming part of cosmological simulation for reionization is the raytracing algorithm. This involves the computation of the hydrogen column density along the path of the ionizing photons. The
C2Ray is a massively parallel raytracing and chemistry code that has been extensively employed in reionization simulations, and it often requires millions of CPU-core hours simulated on several thousand computing nodes run on high-performance computers (HPC) to simulate large-scale volume. In recent years, an updated version of this code, the pyC2Ray, presented the Accelerated Short-characteristics Octahedral RAytracing (ASORA) method, a novel raytracing algorithm specifically designed to run on GPUs. The algorithm is written in C++/CUDA and wrapped in a Python interface that allows easy and customized code use without compromising computational efficiency. In our recent work, we demonstrated that a simple refactoring from CPU to GPU of the raytracing algorithm employed in reionization simulations led to a speedup factor of 600, indicating that these algorithms require dedicated development to achieve computational efficiency and physical veracity. Moreover, pyC2Ray takes large time steps to solve the ODE, over which astrophysical quantities are considered constant. This assumption allows us
to find an analytical solution, but because of the radiative feedback, the assumption on the time step is inaccurate, and it requires updating the density field before repeating the raytracing step. This process must be repeated several times before finding the exact solution, which becomes time-consuming.
In January 2022, Switzerland joined the SKAO with nine national institutions that constituted the SKA Swiss Consortium (SKACH) with the goal of managing Switzerland’s scientific and technological contribution to the SKAO. The proposed project will leverage the unique expertise of SKACH in the domains of cosmology, astrophysics, and computer science (Machine Learning and HPC). Over the course of the project, the team will work on (1) enabling multi-frequency radiation that includes a variety of radiative mechanisms relevant to cosmic reionization. (2) develop an adaptable particle-to-mesh framework for raytracing to smooth particle-based hydrodynamical simulation on multi-node domain decomposition, and (3) extend the chemistry evolution to include helium and, most importantly, the implication of AI in the chemistry solver for an accurate and faster the convergence of the numerical solution.