ASTROPHYSICS

CLONES: Digital Twins of the Local Universe

Principal Investigator:
Jenny Sorce

Affiliation:
Leibniz-Institut für Astrophysik Potsdam, Potsdam, Germany

Local Project ID:
pn57ne, pr74je

HPC Platform used:
SuperMUC-NG PH1 at LRZ

Date published:

Introduction

Two billion light-years wide, the local Universe, is a formidable observational playground for astrophysicists. This tiny bit of the Universe hosts billions of billions of stars, planets gathered in galaxies, including our very own: the Milky Way. We now have a huge amount of data at different wavelengths of tons of galaxies and galaxy clusters in our neighborhood to understand better the Universe, its formation and evolution as well as that of its constituents. However, analyzing and interpreting properly all the observations of our Cosmic Home requires sophisticated cosmological simulations that need to be run on powerful supercomputers.

Nevertheless, typical cosmological simulations reproduce only statistically the large-scale structure of the Universe called the cosmic web. This web defines the distribution of matter, galaxies included, in the Universe into knots, sheets, filaments and voids. In order to refine our conclusions, new simulations designed to resemble the local Universe not only statistically, but also down to the details (galaxies and galaxy clusters), become essential. Thus, CLONES (Constrained LOcal & Nesting Environment Simulations) were born [1]. These digital twins of the local Universe have since been used for several projects. The CLONES initial conditions (a set of dark matter particles with initial positions and velocities) have almost all been run forward in time at SuperMUC-NG to produce various digital twins. Among them, there are those of our own galaxy and its neighbor, Andromeda (HESTIA project), of our closest cluster neighbor, Virgo (Light on the Virgo cluster project) e.g. [2], of another cluster, Coma (Coma Connectivity project) [3], of the inner part of the Universe in its infancy (CoDa project) and of the full local Universe (SLOW and LOCALIZATION projects) e.g. [4].

Still, the first two projects used only zooms of small parts of larger simulation boxes. The third one followed only dark matter particles and the fourth one stopped at high redshifts, namely way before reaching our cosmic time. The fifth project reached our cosmic time in the full box and followed both dark matter and baryonic particles. However, given the scientific ambition of our new LOCALIZATION project, a CLONE with an even higher resolution is required. The new CLONE of the local Universe will thus have a higher resolution than our previous full-box CLONE. In addition, it will include additional physics, in particular non-isotropic AGN jet feedback and black hole spin-dependent feedback. These two additional components permit a more accurate modelling of the physics related to the energetic black holes happening in galaxy clusters. These phenomena are indeed fully involved in galaxy formation and evolution via their effect on the gas (baryonic matter).

All the CLONES, but for our last one, have already been presented in previous editions. Consequently, this paper focuses only on our latest higher resolution full-boxCLONE, that includes additional physics. It is currently run for the LOCALIZATION project at LRZ on SuperMUC-NG with the RAMSES code [5].

Results and Methods

Following the methodology, we developed and presented in [1] and references therein, the replica of our Cosmic Home is obtained via initial conditions that are run forward in time down to our cosmic time (redshift z=0). In order to properly resolve galaxies of interest for our project, the dark matter particle mass is ~109 times the solar mass. There are more than 8.5 billion of them in the simulation box. In addition, the gas distribution and physics properties are represented by cells of less than 10,000 light-years aside in the densest regions. There are an additional more than 2 billion stars and black holes particles.

Already more than sixty million cpu hours were used to simulate the full formation and evolution of our cosmic neighborhood. From 5,280 up to 31,680 cores performed the calculation either from the thin or fat nodes available on SuperMUC-NG. About 200 snapshots were written, half were kept for reasonable storage reasons. One snapshot is indeed 8.3 TB of data distributed into 26,411 files. A hundred snapshots means almost a petabyte of data distributed into almost three millions files. The number do not include the pre-computing work required to prepare the initial conditions nor the post-computing time that has been and will be necessary to further analyze the outputs.

In parallel, the RAMSES code set up has been profiled for performance on SuperMUC-NG. A few hotspots were identified and optimized such as a reduction of compilation time and of memory footprint.

Figure 1 shows different steps of the formation of our Cosmic Home. The cosmic web becomes more and more pronounced as structures form and evolve. Filaments become more pronounced. Knots and filaments are filled with matter, galaxies included, while voids, and to a lesser extent sheets, are slowly emptied from their galaxies. Filaments constitute highways for galaxies falling into galaxy clusters that already gather hundreds of them. While the blueish color stands for the distribution of dark matter in the local Universe replica, the reddish color shows the gas.

Ongoing Research / Outlook

The next step is to analyze the simulation as an observer would and to compare it with observations in order to better understand them. To mimic observations, an observer must then be placed at the center of the box. Synthetic observations made by this observer are then derived. With the CLONE, our ultimate goal is to correct observationally derived (cosmological) parameters from any source of observational biases.

References and Links

[1] J.G. Sorce, MNRAS 478 (2018), 5199.

[2] T. Lebeau, J.G. Sorce, N. Aghanim, E. Hernandez-Martinez, K. Dolag, ArXiv231002326L, A&A in press

[3] N. Malavasi, J.G. Sorce, K. Dolag, N. Aghanim, A&A 675 (2023), A76.

[4] K. Dolag, J.G. Sorce, S. Pilipenko, E. Hernandez-Martinez, M.

Valentini, S. Gottlöber, N. Aghanim, I. Khabibullin, A&A 677 (2023), A169.

[5] Teyssier, R. A&A 385 (2002).