Visual Analysis of Structure Formation in Cosmic Evolution Karsten Schatz * Christoph M ¨ uller * Patrick Gralka * Moritz Heinemann * Alexander Straub * Christoph Schulz * Matthias Braun * Tobias Rau * Michael Becher * Patrick Diehl † Dominic Marcello † Juhan Frank † Thomas M ¨ uller ‡ Steffen Frey * Guido Reina * Daniel Weiskopf * Thomas Ertl * ABSTRACT The IEEE SciVis 2019 Contest targets the visual analysis of structure formation in the cosmic evolution of the universe from when the universe was five million years old up to now. In our submission, we analyze high-dimensional data to get an overview, then inves- tigate the impact of Active Galactic Nuclei (AGNs) using various visualization techniques, for instance, an adapted filament filtering method for detailed analysis and particle flow in the vicinity of fil- aments. Based on feedback from domain scientists on these initial visualizations, we also analyzed X-ray emissions and star formation areas. The conversion of star-forming gas to stars and the resulting increasing molecular weight of the particles could be observed. Index Terms: Human-centered computing—Visualization; 1 I NTRODUCTION The SciVis Contest 2019 is dedicated to the visual analysis of a cosmological simulation that models the evolution of the universe from the age of five million years (z = 200) up to now (z = 0), where z is the redshift. The simulation, run using the Hardware/Hybrid Accelerated Cosmology Code (HACC) [7, 11], comprises 64 3 dark matter particles and 64 3 baryon particles contained in a cubic domain with a side length 64 Mpc / h (galactic distance with h reflecting the uncertainty of the Hubble constant). The task of the contest was to summarize the provided data set. No further details were provided. Furthermore, the organizers of the contest recommended including data derived from the ones provided by the simulation, namely the temperature of the baryons and thermodynamic entropy. Moreover, studying the impact of Active Galactic Nuclei (AGNs) on their surroundings is in line with the general goal of the HACC simulation. Our approach to the analysis adopts a variety of different tech- niques from all areas of visualization: We use parallel coordi- nates plots (PCPs) and scatter plot matrices (SPLOMs) to provide an overview of the quantities and to discover patterns in multi- dimensional data. Additionally, these techniques facilitate selection and filtering using brushing and linking with 3D views of the parti- cles or the vector field, thus enabling the user to locate specific points and ranges in space. As additional means to provide an overview and detailed analysis, we employ direct displays of the particles as spheres with different quantities being mapped to size and color, arrow glyphs to detect the velocity of the particles as well as direct volume rendering and iso-surface rendering of reconstructed parti- cle densities. As the majority of matter in the universe resides in galactic filaments and many interesting processes like star formation happen there [1], we developed a filament detection algorithm that can separate these structures from the surrounding medium. For * Visualization Research Center (VISUS), University of Stuttgart, Ger- many. e-mail: firstname.lastname@visus.uni-stuttgart.de † Louisiana State University, Baton Rouge, United States. e-mail: {frank,dmarcello}@phys.lsu.edu, pdiehl@cct.lsu.edu ‡ Max Planck Institute for Astronomy, Heidelberg, Germany. e-mail: tmueller@mpia.de the velocity vectors, we use surface line integral convolution (LIC), which allows us to show the direction of the particles on the surface of dense areas, for instance, on filament structures. Furthermore, we want to obtain a visual representation of the data that can be related to quantities observed in the real universe. To this end, we compute a scalar field showing the X-ray emission in space and time from the given data. All these techniques are smoothly combined in a visual analytics tool based on MegaMol [8], which facilitates high-performance rendering for interacting with large simulation data. Our approach allows researchers to explore, summarize, and quantify the structure formation in cosmic evolution. The presented system is the result of a collaboration between domain scientists and visualization experts. 2 OVERVIEW OF THE DATA We first want to report on the different steps of the analysis process for the simulation data using our system. In this section, we start with an overview of the data—following the visual information seeking mantra by Shneiderman [16]. The subsequent sections address various details of the analysis, as well as adapted or novel techniques. Initially, we tried to inspect the first and last simulation step by depicting particles as solid glyphs. However, this approach was not viable since particles are evenly distributed initially and throughout the simulation high particle densities lead to significant occlusion. To account for the multivariate nature of the data set and to quickly filter basic quantities present in the simulation data, we decided to incorporate a SPLOM and a PCP. The first time step is shown in Fig. 1(a). Please note the evenly distributed position components, indicating that the particles are ini- tially positioned on a uniform grid. While this circumstance would also have become visible with a simple spatial volume rendering, one might have missed the non-evenly distributed high velocities and gravitational potential. A little less obvious is the equal dis- tribution of the two classes of particles (all baryon particles and dark matter particles have equal mass each). Presumably, these are the initial values of the simulation. The last time step is shown in Fig. 1(b). Contrary to the beginning of the simulation, one can see the fine filament structures emergent in the position as well as the normally distributed velocity. As expected, we also see a strong correlation between temperature and internal energy. Moreover, one can recognize that large smoothing length correlates with slow parti- cles, which is also backed up by correlation with the gravitational potential. Furthermore, entropy is not evenly distributed. The PCP at the top allows for easy detection of correlations between the given magnitudes, as the column order is interchangeable. It also provides brushing and linking based filtering options on the whole data as well as single particles as shown in red in Fig. 1(b). When inspecting all time steps as an animation (cf. supplemental video), we noticed some leaping values such as molecular weight and density around time step 35 and an overall trend that everything is slowing down, which was to be expected. 3 FILAMENT DETECTION Especially in later simulation steps of the universe, we observed matter forming filament structures. Based on previous research [1],