By Greg Snyder, Postoctoral Fellow at STScI
For decades, we have sought to use cosmological simulations to better understand our universe as seen by galaxy surveys. Owing to the continually increasing availability of computing power, the latest simulations can directly follow dark matter, stars, and gas in entire galaxy populations. Rapid progress was made toward this goal by adding significant feedback to regulate star formation rates, and by overcoming challenges with numerical methods. Now, hydrodynamical simulations yield increasingly realistic galaxies in large regions of the universe, setting the stage for jointly analyzing the statistics of galaxy morphology in surveys and in theory. Such exercises are needed to measure the observability timescales of galaxy formation events and therefore assign significance to a given observed galaxy state.
Figure 1: Real versus synthetic Hubble Space Telescope (HST) Ultra Deep Field (UDF) from . Left: data from the HST eXtreme Deep Field release . Right: in the same units, intensity, and color scale, a sightline through the Illustris Simulation of galaxy formation, with similar depth and resolution. Current simulations produce galaxies with colors and shapes broadly similar to observed ones.
A number of simulation projects can be analyzed in this manner. For example, the Illustris Simulation  calculated galaxy formation over cosmic time in (106 Mpc)3. Illustris applied galaxy physics models consisting of primordial and metal line cooling, star formation, gas recycling, metal enrichment, supermassive black hole (SMBH) growth, and gas heating and ejection by feedback from supernovae and SMBHs. Models were chosen to match the z=0 stellar mass and halo occupation functions, plus the cosmic history of star formation rate (SFR) density, achieving a reasonable mix of spirals and ellipticals.
To make statistical tests regarding galaxy morphology, we process the simulation data into ideal synthetic images, convolve these with telescope point-spread functions, re-bin to CCD pixel scales, and add sky noise. Recent such simulations of thousands of galaxies achieve a resolution element size ~1 kiloparsec, and therefore they are well-matched to imaging by the Sloan Digital Sky Survey at z > 0.05 and Hubble Space Telescope (HST) at z > 1. Fig. 2 shows a sample in mass and redshift as if the galaxies were observed in deep HST images. We are creating synthetic data for up to ~10,000 simulated galaxy histories, at each epoch measuring quantitative automated morphologies from several directions in many filters. With these measurements, we have confirmed that  produces a roughly realistic distribution of quantitative galaxy structures at z~0, shown in Fig. 3. Where it is not realistic in the nearby or distant universe, we expect to learn more about the limitations of current large-scale galaxy formation models. Regardless, we are able to extract information about the observability of mergers and other events, which cannot be measured without simulations of this type.
Figure 2: Synthetic HST data of individual galaxies , in ACS/F606W, WFC3/F125W and F160W filters with sky noise comparable to the Ultra Deep Field. From left to right in each panel, galaxy models at z=5, 4, 3, 2, 1, and 0. From top to bottom, galaxy models with stellar mass M = 109 to 1011.5 solar masses.
Figure 3: Low-redshift comparison between galaxy morphology from the Illustris Project [1, right] and a volume-limited sample of galaxies [4, left]. This shows current simulations can span the observed space of galaxy structure with roughly the correct dependence on SFR as measured by optical colors.
I have also been studying very high-resolution zoom-in hydrodynamical simulations . These allow accurate modeling of the rest-frame optical and ultraviolet flux via dust radiative transfer  to create mock HST images , study the effects of dust, and trace galaxy growth and interactions with high precision in both space in time. From many individual galaxy simulations, we are quantifying the distribution of galaxy evolutionary paths and calibrating estimators of the galaxy merger rate at z > 1. We are using multiple simulation sets with different assumptions to verify our estimates of observability timescales.
Galaxy formation simulated at any scale continues to have significant uncertainties, including the treatment of star formation, chemical enrichment, SMBH accretion, and the creation and evolution of outflows. Assumptions made to treat these processes today may be unphysical, leading to low-mass galaxies that are too old at z=0, and SMBH feedback that is too efficient at removing gas from the centers of halos, among other issues. Even so, recent progress in the field allows us to construct statistically relevant experiments with which to improve our understanding of galaxy physics and robustly measure important events during their formation.
- Vogelsberger et al. 2014 (Nature, 509, 177)
- Illingworth et al. 2013 (ApJS, 209, 6)
- Genel et al. 2014
- Lotz et al. 2008 (ApJ, 672, 177)
- Ceverino et al. 2013
- Jonsson 2006 (MNRAS, 372, 2)
- Moody et al. 2014
- Torrey et al, in preparation
- Snyder et al, in preparation