Do We Know How to Model Reionization?
Nickolay Y. Gnedin · 2025
Read Do We Know How to Model Reionization? by Nickolay Y. Gnedin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-10571.txt.
Introduction
Nickolay Gnedin compares two large computer models of how the early universe became transparent, in a solo paper from 2025. That change, called reionization, happened when light from the first stars and galaxies split the gas around them. Modeling it well matters for reading many other signals from that era.
The two models, named CROC and Thesan, start from like sources of light. Yet the patterns of their radiation fields differ at the same cosmic time. Gnedin shows the patterns can be matched at large scales if the snapshots are allowed to sit at unlike times. At small scales, gaps remain that are not yet explained. The paper warns that timing in such models is less firm than modelers often assume.
Catalog record
| CREATOR | Nickolay Y. Gnedin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-10571
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Do We Know How to Model Reionization? Nickolay Y. Gnedin Abstract I compare the power spectra of the radiation fields from two recent sets of fully-coupled simulations that model cosmic reionization: ”Cosmic Reionization On Computers” (CROC) and ”Thesan”. While both simulations have similar power spectra of the radiation sources, the power spectra of the photoionization rate are significantly different at the same values of cosmic time or the same values of the mean neutral hydrogen fraction. However, the power spectra of the photoionization rate can be matched at large scales for the two simulations when the matching snapshots are allowed to vary independently. I.e., on large scales, the radiation field in two simulations proceeds through the same evolutionary stages, but the timing of these stages is different in different simulations and is not parameterized by an easily interpretable physical quantity like the mean neutral fraction or the mean free path. On small scales, large differences are present and remain partially unexplained. Both CROC and Thesan use the Variable Eddington Tensor approximation for modeling radiative transfer, but adopt different closure relations (optically thin OTVET versus M1). The role of this key difference is tested by using smaller simulations with a new cosmological simulation code that implements both closure relations in a controlled environment (the same hydro, cooling, and gravity solvers and the star formation recipe). In these controlled tests, both the M1 closure and the OTVET ansatz follow the expected behavior from a simple analytical approximation, demonstrating that the differences in the 2-point function of the radiation field induced by the choice of the Eddington tensor are not dominant. keywords cosmology, galaxy formation, intergalactic medium, reionization, radiative transfer, numerical methods 1 Introduction One of the primary challenges in modeling the real universe is that it is often very hard to validate numerical codes - different codes may all pass simple tests perfectly, and yet diverge in realistic simulations. Hence, ”comparison projects” have been used extensively to validate codes that use very different numerical approaches by modeling the same physical system, beginning with the pioneering ”Santa Barbara Cluster Comparison” project ( Frenk et al. 1999 ) and first N-body code comparisons ( Knebe et al. 2000 ) . For modeling cosmic reionization, an accurate radiative transfer solver is especially important. ”Iliev tests” ( Iliev et al. 2009 ; Iliev et al. 2006 ) serve as the de facto standard for cosmological radiative transfer codes, but they test radiative transfer solvers and their coupling to hydrodynamics in restricted, idealized settings. So far, no inter-comparison of different simulations of reionization, similar to existing comparisons of other cosmological codes like AGORA ( Kim et al. 2014 ) , has been done. Figure 1: Volume weighted neutral (solid lines) and ionized (dashed lines) fraction of hydrogen as functions of redshift for CROC (blue) and Thesan (red) runs used in this work. The data points from Fan et al. 2006 and Bosman et al. 2022 were used as the calibration data for CROC and Thesan simulations, respectively. I also show in magenta Thesan-WC-2, one of Thesan’s calibration runs. Figure 2: Slices through CROC ”C40A” (top) and Thesan-1 (bottom) simulations at times when each simulation is approximately at mid-point of reionization ( \langle X_{\rm HI}\rangle_{V}\approx 0.5 ). Three panels show the neutral fraction, the photoionization rate, and the radiation source (in arbitrary units). The inserts are centered on the strongest peak in the photoionization rate within these randomly chosen slices. The Thesan box is cropped to match the size of a smaller CROC box. Two simulation sets that serve as representative examples of the current ”state-of-the-art” are publicly available: ”Cosmic Reionization On Computers” ( Gnedin 2014 ; Gnedin & Kaurov 2014 ; Gnedin 2022 , CROC,) and ”Thesan” ( Kannan et al. 2022 ; Garaldi et al. 2022 ; Garaldi et al. 2024 ) . The two sets of simulations are sufficiently similar so that a reasonable agreement can be expected between them. They model similar \sim 100 cMpc volumes with similar spatial resolutions of 100-300 proper parsecs, have almost identical mass resolution ( 2000^{3} ), and model a similar range of relevant physical effects: hydrodynamics, non-equilibrium cooling and ionization, radiative transfer, and star formation and stellar feedback. In modeling the latter, Thesan does a noticeably better job than CROC, as CROC only includes star formation in molecular gas and has been shown to underpredict star formation rates and UV luminosities of most massive galaxies ( Zhu et al. 2020 ) , while Thesan uses BPASS for stellar spectra, and includes binary stars, which emit ionizing photons up to 3 times longer than single stars. Despite their similarities, the two simulation sets have sufficiently different reionization histories, shown in Figure 1 . Since both simulation projects match the observed evolution of the galaxy luminosity function at z<10 reasonably well, the difference between them is not due to a difference in how sources of ionizing radiation are modeled (this is also illustrated quantitatively below in Fig. 3 ), but rather in how ionizing radiation propagates through the simulation volume and how (and where) it is absorbed. Both simulation projects use the Variable Eddington Tensor approximation for modeling radiative transfer on a grid with a moments method, adopt a ”reduced speed of light” approximation ( Gnedin & Abel 2001 ) for evolving the radiation field forward in time, and feature the ”effective escape fraction” to account for photon absorptions below the spatial resolution of the simulations. The main difference between the two projects is in different ansatzes for the Variable Eddington tensor - this may serve as one reason for the difference between them. Another major difference is in the actual equations they are solving. Both these differences and several other, less obvious numerical effects are explored in this paper. 2 Simulations As has already been mentioned, the two simulation sets (”CROC” and ”Thesan” hereafter) have sufficiently different reionization histories. Hence, the comparison between them is not trivial; the corresponding snapshots between the two simulations must be ”matched” somehow. One possible way to match them is to compare snapshots at the same value of the volume-weighted mean neutral fraction, \langle X_{\rm HI}\rangle_{V} (the mass-weighted fraction during and after the overlap of ionized bubbles is dominated by DLAs and is not representative of the ionization state of the IGM). Figure 2 shows a visualization of the two simulations at around the ”mid-point” of reionization when \langle X_{\rm HI}\rangle_{V}\approx 0.5 . Thus matched, the two simulations are not drastically different: the distribution of sources is similar, ionized bubble morphology is also similar (although hydrogen is more ionized inside the bubbles in Thesan than in CROC, and CROC has more small bubbles, again reflecting its deficiency in modeling the largest galaxies), and the largest difference seems to be in the distribution of the photoionization rate inside the bubbles. Exploring this difference is the key goal of this paper. Publicly available Thesan data include uniform grids with a number of relevant physical fields, with a small twist: the ionizing luminosity is binned in redshift, not real space. Since the clustering in redshift space is suppressed on small scales compared to real space, that particular choice of the Thesan team complicates the comparison between the