A numerical simulation of cosmic structure formation reproduces both large- and smaller-scale features of a representative volume of the Universe from early in its history to the present day. See Article p.177
Perhaps the greatest triumph of modern cosmology is that a model with only six parameters can explain the vast majority of observational data from the first minutes of the Universe to the present day1. This standard model posits that 95% of the Universe today is composed of enigmatic 'dark matter' and 'dark energy'. Paradoxically, modelling the dynamics of the remaining 5% — normal, 'baryonic' matter — has proved to be the more challenging task. On page 177 of this issue, Vogelsberger et al.2 describe a numerical simulation of the formation of cosmic structure that captures both the large-scale distribution of baryonic material and its properties in individual galactic systems through cosmic time.
Tracking the evolution of baryonic matter is a daunting undertaking because of the huge range of physical scales involved in the processes that shape galaxies and larger structures (Fig. 1). To cover a representative portion of the Universe, cosmologists must study cosmic volumes that are at least 100 million parsecs (326 million light years) across. By contrast, the natural scale of star formation is approximately 1 parsec, and accretion of gas by supermassive black holes occurs on even smaller scales. Numerical simulations have long been the tool of choice for tackling these problems. But even with the most powerful supercomputers, it has been impossible to run a simulation large enough to model the large-scale distribution of gas, stars and dark matter while keeping sufficient detail to accurately capture individual galaxies. It turns out that simulating the Universe can be a difficult endeavour.
Vogelsberger and colleagues attack this problem from all sides. Their simulation — named Illustris — incorporates more than 10 billion individual cells to represent the gas in the simulation volume; this is nearly a tenfold increase on its predecessors. The numerical code3 used to perform their simulation employs a new approach, based on an unstructured and adaptive computational mesh that follows fluid flows, to solve the equations that describe the time evolution of baryonic matter within cosmic structures. And finally, the physical phenomena included in the simulation are rich and complex: the authors consider gas cooling, stellar evolution, energy input from supernova explosions, production of chemical elements, and gas accretion by supermassive black holes (along with accompanying radiative feedback), to list just a few.
If this all sounds somewhat complicated, do not be fooled: it is extremely complicated. Many of these processes are not understood from first principles, and they interact in complex, nonlinear ways. Additionally, the relevant physical scales are often (much) smaller than can be directly resolved, even with Illustris. This requires computationally efficient models that accurately encapsulate the underlying physics. Running the simulation was therefore no mean feat: it took approximately 16 million CPU (central processing unit) hours. The end result, however, is a simulated Universe that looks an awful lot like the real one.
A mock observation of Illustris set to mimic the Hubble Ultra Deep Field4, the deepest picture of the cosmos ever taken, can easily pass for the real thing when the two are viewed side by side (see Fig. 1b, c of the paper2). Images of galaxies from the simulation are also impressively realistic (see Fig. 1a of the paper), an accomplishment that has previously been possible only for simulations of individual galaxies. This is not just window dressing: such procedures allow direct and meaningful confrontation of theory with data. A wide array of quantitative measures agrees with observations of the real Universe as well. For example, previous generations of simulations had great difficulty capturing the observed distribution of elements heavier than hydrogen and helium contained in stars. Illustris reproduces these observations, not just for the Universe as a whole, but also as a function of galactic stellar mass. In addition, the simulation matches the abundance of these heavier elements in dense gas clouds.
Of course, Illustris does not mark the end of cosmological simulations of galaxy formation. Although its computational volume is immense, it is not large enough to model the formation of rare cosmological objects (for example, powerful black holes observed in the early Universe). And the level of detail is still not fine enough to study the faintest galaxies surrounding the Milky Way. Star formation in low-mass galaxies occurs earlier and faster in Illustris than in the real Universe, a difficulty shared by almost all models of galaxy formation5. However, such issues point the way for future advances.
One clear goal for observers and theorists alike is to understand in detail the ways in which energy and momentum from evolving and exploding stars affect the properties of gas in and around galaxies6,7. A promising computational approach is to combine large-volume simulations such as Illustris, which are necessarily limited in the level of detail they can resolve, with simulations that focus all of their power on individual galaxies, which sacrifice the ability to study galaxy formation in a statistical manner. If the knowledge gained from these focused simulations can be incorporated into large-scale computational efforts, ever more accurate numerical realizations of the Universe should be possible, guiding our investigations into the underlying physical processes.
Computationally, the ability to reach the scales necessary to model star formation directly, while encompassing thousands of galaxies similar to the Milky Way, is still a distant dream. But thanks to the Illustris simulation, creating a realistic virtual Universe of gas, stars, black holes and dark matter is already a reality.
Planck Collaboration. Astron. Astrophys. (in the press); preprint at http://arxiv.org/abs/1303.5076 (2014).
Vogelsberger, M. et al. Nature 509, 177–182 (2014).
Springel, V. Mon. Not. R. Astron. Soc. 401, 791–851 (2010).
Illingworth, G. D. et al. Astrophys. J. Suppl. Ser. 209, 6 (2013).
Weinmann, S. M. et al. Mon. Not. R. Astron. Soc. 426, 2797–2812 (2012).
Agertz, O., Kravtsov, A. V., Leitner, S. N. & Gnedin, N. Y. Astrophys. J. 770, 25 (2013).
Hopkins, P. F. et al. Preprint at http://arxiv.org/abs/1311.2073 (2013).
About this article
Monthly Notices of the Royal Astronomical Society (2016)