Seeing the Difference between Cosmological Simulations
IEEE Computer Graphics and Applications
V isualizing the time-variant results of a simulation can help scientists see patterns that would be difficult to find using only statistics. Furthermore, visualizing the differences between multiple simulations can let scientists directly analyze the simulations' consistency. In this article, we describe our interactive application for viewing the differences between multiple time-variant cosmological simulations. Using programmable shaders and multiple visualization techniques, the
... illustrates how the properties of millions of particles vary across simulations. In doing so, we aim to help scientists find patterns in variance across dimensions, simulation parameters, and time. The driving force behind this study is a series of cosmological particle data sets with multiple dimensions of data as well as temporal changes and variability across all of the dimensions. We specifically examine past scenarios wherein a particle visualization requires a defined spatial element and multiple additional dimensions. 1,2 Expanding on our previous work, 3 we also show examples of the multiple structural components of the data sets, which can be difficult to qualitatively investigate without visualization. Cosmological simulation comparison In simulating the universe's evolution, cosmologists employ various different simulation algorithms. Understanding the effect of algorithmic differences on the outcome is critical in ensuring that judgments based on the results are accurate. We examine the results of a cosmological particle simulation generated for a study of cosmological simulation robustness. 4 Our goal is to visualize inconsistencies between different simulations that begin with the same initial conditions. 1 The simulation begins as 256 3 particles arranged evenly on a cubical grid. The particles are then moved by small amounts to establish the correct cosmological initial conditions (see Figure 1 , next page), which are constrained by observations of the cosmic microwave background and the distribution of galaxies on large scales. The particles then move under the influence of gravity in an expanding universe for a number of time steps until the current epoch is reached. As is typical for cosmological simulations, periodic boundary conditions are imposed. When a particle moves past one edge of an axis, it appears on the other end. This property is intriguing from a visualization perspective because the axes wrap. As we'll discuss later, we address this important hurdle in our visualization implementation. The numerous simulator algorithms used to compute interparticle forces and the approximated values that drive them present results with quantifiable differences. 4,5 Some simulators use hierarchical sampling of the system phase space distribution function while others simplify by distance. 5 Each simulation was run separately, and as a result of their implementation differences, every simulation produced slight deviations in the final particle distributions and velocities. Past visualization approaches have made significant progress in displaying similar types of particle data. These methods have merit in their effective portrayal of the particles' position and velocity. 1 However, we now have the benefit of hindsight and newer technology. Earlier approaches to vi-As cosmology simulations help us understand the universe, we must understand how the results of different simulations vary. Visualization and modern graphics hardware can now provide the ability to visually explore these differences interactively.