Living on the edge. A quantitative warning on boundary artifacts in the IllustrisTNG

Ana Mitrašinović · 2025

Read Living on the edge. A quantitative warning on boundary artifacts in the IllustrisTNG by Ana Mitrašinović (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-2511-03562.txt.

Introduction

Ana Mitrasinovic warns that the edges of a famous galaxy simulation quietly corrupt some of its galaxies. Cosmological boxes wrap around at their faces, a practical trick that can fake or strip matter when galaxies cross the seam. Galaxies oddly starved of dark matter serve as tracers for the damage.

Their numbers grow as box size shrinks, and they cluster near the edges, with the smallest box hit hardest. Hand checks confirm abrupt, unphysical mass losses right at boundary crossings. In a mid-size box, telling seam damage from true tidal stripping in crossing clusters stays genuinely hard. The author closes with concrete safeguards for anyone mining that irreplaceable simulation for galaxy structure claims.

Catalog record

CREATORAna Mitrašinović
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-03562

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Living on the edge. A quantitative warning on boundary artifacts in the IllustrisTNG Ana Mitrašinović Abstract Periodic boundary conditions (PBCs) are a practical necessity in cosmological simulations, but they can also introduce numerical artifacts. We quantified the prevalence of PBC-related artifacts in the IllustrisTNG using dark-matter-deprived galaxies (DMDGs) as tracers. We found that their occurrence scales inversely with simulation volume. We demonstrated that this excess population is spatially correlated with the box edges; the smallest TNG50 box is most affected by these problems. This is a unique and irreplaceable resource for studying galaxy structure. Manual inspection confirmed abrupt, unphysical mass-loss events coincident with boundary crossings. We also highlight the challenge of disentangling numerical artifacts from genuine tidally stripped galaxies in boundary-crossing clusters in TNG100. We conclude by recommending a set of mandatory sanity checks that include positional verification, mass history analysis, and even exclusion of the buffer zone near the edges. These strategies ensure the robustness of the scientific results derived from these invaluable simulations. 1 Introduction Cosmological simulations of galaxy formation are a valuable contemporary tool for bridging the gap between theoretical knowledge and vast observational data ( 47 ; 5 , for reviews of the ecosystem of cosmological simulations, see) . The real Universe is effectively infinite, whereas cosmological simulations model the large-scale structure of the Universe within a finite cubic volume. As a solution, simulations use periodic boundary conditions ( 11 ; 42 , PBCs; e.g.,) . When a particle, galaxy (typically subhalo), or halo crosses one boundary of the simulation box, it reenters from the opposite side, as if the simulation volume tiles infinitely, ensuring that structures can evolve without artificial edges. This solution works very well in practice, but it is not perfect and has its pitfalls. The identification of halos and subhalos in cosmological simulations relies on specialized structure-finding algorithms, each with advantages and limitations. Highly efficient, in a computational sense, is the SUBFIND algorithm ( 41 ) , which is a configuration-space algorithm that operates by detecting gravitationally bound overdensities. This algorithm is relevant to our work because it is used to produce subhalo catalogs in IllustrisTNG. Studies comparing various structure-finding algorithms ( 14 ; 24 ; 7 , e.g.,) have shown, however, that SUBFIND becomes less reliable for subhalos in very dense environments relative to algorithms that are based on phase space ( 2 ; 6 , e.g.,) or history space ( 9 ; 18 , e.g.,) . These findings suggest that analogous errors can be expected for subhalos near the edges of the periodic simulation box, where parts of structures lie across the boundary wrap, and local density or binding energy calculations may be perturbed, resulting in multiple artificial subhalos of nonphysical origin (i.e., numerical artifacts). Although alternative algorithms have the potential to resolve some of the issues, they can be more computationally demanding and still require careful treatment of periodic wrapping. Thus, while the choice of (sub)halo finder can affect the prevalence of numerical artifacts, no method is entirely immune to issues arising at the edges of the simulation box. The discovery of dark-matter-deprived galaxies (DMDGs) by 45 ; 44 has raised questions about the validity of the standard cosmological model. It soon became clear that these galaxies naturally occur as a result of tidal stripping ( 31 ; 22 ; 13 ; 17 ; 21 ; 23 ; 30 ; 4 , e.g.,) . Although this formation mechanism is probably dominant, DMDGs can also form from tidally stripped material in the aftermath of a high-velocity head-on collision between two gas-rich galaxies ( 39 ; 38 ; 16 ; 46 ; 12 ; 15 ) . Most studies that explored DMDGs in simulations have focused on the satellite galaxy population, reinforcing the idea that DMDGs are indeed satellites. Some observed ( 8 ; 3 , e.g.,) and even simulated ( 20 , e.g.,) DMDGs can be found in isolation, however, far from more massive galaxies. In an effort to explore these isolated DMDGs, 37 serendipitously found an issue with the implementation of the PBCs that causes regular galaxies crossing the edge of the simulation box to become torn apart, unrelated to any (sub)halo-finding algorithm. Ruediger Pakmor and Volker Springel traced and identified the bug as a force-calculation error in GADGET-2 ( 42 ) and its derivatives (namely, GADGET-3 , Arepo , and GIZMO ), and they promptly fixed the issue. The resolution of this issue will improve the robustness and precision of the upcoming simulations, but numerical artifacts related to this issue ( 37 ) or the particular subhalo-finding algorithm still exist in the publicly available IllustrisTNG data. It is important to note that while observational searches for DMDGs ( 45 ; 44 , e.g.,) often focus on low-mass satellite galaxies, the artifacts identified by 37 were primarily massive galaxies, suggesting that numerical issues at the boundary can affect a wide range of galaxy masses. This motivates a broad search for these effects. The widespread use of IllustrisTNG simulations, particularly the high-resolution TNG50 volume ( 27 ; 35 ) , for studying galaxy evolution necessitates a clear understanding of their numerical limitations. The TNG50 simulation offers an unparalleled combination of resolution and cosmological context, making it a cornerstone for studying galaxy formation and internal structure. Given current trends in numerical astrophysics and the fact that newer simulations (e.g., MillenniumTNG or TNG-Cluster) focus on enlarging the simulation box and not on improving particle resolution ( 1 ; 10 ; 33 ; 26 , e.g.,) , it is highly unlikely that the unique TNG50 will become irrelevant anytime soon. Although previous work has identified the existence of artifacts caused by PBCs ( 37 ) , the prevalence and scaling of these artifacts with simulation volume have not been quantified. The small volume of TNG50, while enabling its high resolution (which offers an invaluable resource for in-depth studies of galaxy structure in a cosmological context), may also increase the rate of numerical artifacts. This is because the small volume inherently increases the frequency of periodic boundary crossings. The motivation for this work therefore is to provide an urgent, quantitative assessment of this issue. We provide the first quantitative estimate of the prevalence of PBC-related artifacts, using DMDGs as a proxy, in the three main IllustrisTNG volumes (TNG50, TNG100, and TNG300). Our goal is to demonstrate how the prevalence of these artifacts scales with box size and to highlight the specific challenges and interpretational pitfalls that are primarily present in TNG50 but, to a certain extent, also in TNG100. By doing so, we aim to provide not only a necessary warning, but also a constructive guide with practical mitigation strategies to safeguard the robustness of current and future scientific analyses that rely on these invaluable simulations. This paper is organized as follows. In Sect. 2 we briefly introduce the simulations and describe our sample and proxies. In Sect. 3 we present the results of our calculations, present an illustrative example of a numerical artifact, and analyze the optimal value of the buffer size by examining the fractions of DMDGs in these boundary regions. Finally, in Sect. 4 , we discuss our results and their broader implications and offer comments on mitigation strategies. 2 Simulations

Read the complete text