We investigate the physics driving the cosmic star formation (SF) history using the more than fifty large, cosmological, hydrodynamical simulations that together comprise the OverWhelmingly Large Simulations (OWLS) project. We systematically vary the parameters of the model to determine which physical processes are dominant and which aspects of the model are robust. Generically, we find that SF is limited by the build-up of dark matter haloes at high redshift, reaches a broad maximum at intermediate redshift, then decreases as it is quenched by lower cooling rates in hotter and lower density gas, gas exhaustion, and self-regulated feedback from stars and black holes. The higher redshift SF is therefore mostly determined by the cosmological parameters and to a lesser extent by photo-heating from reionization. The location and height of the peak in the SF history, and the steepness of the decline towards the present, depend on the physics and implementation of stellar and black hole feedback. Mass loss from intermediate-mass stars and metal-line cooling both boost the SF rate at late times. Galaxies form stars in a self-regulated fashion at a rate controlled by the balance between, on the one hand, feedback from massive stars and black holes and, on the other hand, gas cooling and accretion. Paradoxically, the SF rate is highly insensitive to the assumed SF law. This can be understood in terms of self-regulation: if the SF efficiency is changed, then galaxies adjust their gas fractions so as to achieve the same rate of production of massive stars. Self-regulated feedback from accreting black holes is required to match the steep decline in the observed SF rate below redshift two, although more extreme feedback from SF, for example in the form of a top-heavy IMF at high gas pressures, can help.Comment: Accepted for publication in MNRAS, 27 pages and 18 figures. Revised version: minor change

We study the rate at which gas accretes on to galaxies and haloes and investigate whether the accreted gas was shocked to high temperatures before reaching a galaxy. For this purpose, we use a suite of large cosmological, hydrodynamical simulations from the OverWhelmingly Large Simulations project, which uses a modified version of the smoothed particle hydrodynamics code gadget‐3. We improve on previous work by considering a wider range of halo masses and redshifts, by distinguishing between accretion on to haloes and accretion on to galaxies, by including important feedback processes and by comparing simulations with different physics. Gas accretion is mostly smooth, with mergers only becoming important for groups and clusters. The specific rate of the gas accretion on to haloes is, like that for dark matter, only weakly dependent on the halo mass. For halo masses Mhalo≫ 1011 M⊙, it is relatively insensitive to feedback processes. In contrast, accretion rates on to galaxies are determined by radiative cooling and by outflows driven by supernovae and active galactic nuclei. Galactic winds increase the halo mass at which the central galaxies grow the fastest by about two orders of magnitude to Mhalo∼ 1012 M⊙. Gas accretion is bimodal, with maximum past temperatures either of the order of the virial temperature or ≲105 K. The fraction of the gas accreted on to haloes in the hot mode is insensitive to feedback and metal‐line cooling. It increases with decreasing redshift, but is mostly determined by the halo mass, increasing gradually from less than 10 per cent for ∼1011 M⊙ to greater than 90 per cent at ∼1013 M⊙. In contrast, for accretion on to galaxies, the cold mode is always significant and the relative contributions of the two accretion modes are more sensitive to feedback and metal‐line cooling. On average, the majority of stars present in any mass halo at any redshift were formed from the gas accreted in the cold mode, although the hot mode contributes typically over 10 per cent for Mhalo≳ 1011 M⊙. Thus, while gas accretion on to haloes can be robustly predicted, the rate of accretion on to galaxies is sensitive to uncertain feedback processes. Nevertheless, it is clear that galaxies, but not necessarily their gaseous haloes, are predominantly fed by the gas that did not experience an accretion shock when it entered the host halo.

We quantify the stellar abundances of neutron-rich r-process nuclei in cosmological zoom-in simulations of a Milky Way-mass galaxy from the Feedback In Realistic Environments project. The galaxy is enriched with r-process elements by binary neutron star (NS) mergers and with iron and other metals by supernovae. These calculations include key hydrodynamic mixing processes not present in standard semi-analytic chemical evolution models, such as galactic winds and hydrodynamic flows associated with structure formation. We explore a range of models for the rate and delay time of NS mergers, intended to roughly bracket the wide range of models consistent with current observational constraints. We show that NS mergers can produce [r-process/Fe] abundance ratios and scatter that appear reasonably consistent with observational constraints. At low metallicity, [Fe/H] −2, we predict there is a wide range of stellar r-process abundance ratios, with both supersolar and subsolar abundances. Low-metallicity stars or stars that are outliers in their r-process abundance ratios are, on average, formed at high redshift and located at large galactocentric radius. Because NS mergers are rare, our results are not fully converged with respect to resolution, particularly at low metallicity. However, the uncertain rate and delay time distribution of NS mergers introduces an uncertainty in the r-process abundances comparable to that due to finite numerical resolution. Overall, our results are consistent with NS mergers being the source of most of the r-process nuclei in the Universe.

Unlike spiral galaxies such as the Milky Way, the majority of the stars in massive elliptical galaxies were formed in a short period early in the history of the Universe. The duration of this formation period can be measured using the ratio of magnesium to iron abundance ([Mg/Fe]) 1-4 , which reflects the relative enrichment by core-collapse and type Ia supernovae. For local galaxies, [Mg/Fe] probes the combined formation history of all stars currently in the galaxy, including younger and metal-poor stars that were added during late-time mergers 5 . Therefore, to directly constrain the initial star-formation period, we must study galaxies at earlier epochs. The most distant galaxy for which [Mg/Fe] had previously been measured 6 is at a redshift of z ≈ 1.4, with [Mg/Fe] = 0.45 +0.05 −0.19 . A slightly earlier epoch (z ≈ 1.6) was probed by stacking the spectra of 24 massive quiescent galaxies, yielding an average [Mg/Fe] of 0.31 ± 0.12 7 . However, the relatively low signal-to-noise ratio of the data and the use of index analysis techniques for both studies resulted in measurement errors that are too large to allow us to form strong conclusions. Deeper spectra at even earlier epochs in combination with analysis techniques based on full spectral fitting are required to precisely measure the abundance pattern shortly after the major star-forming phase (z > 2). Here we report a measurement of [Mg/Fe] for a massive quiescent galaxy at a redshift of z = 2.1, when the Universe was 3 billion years old. With [Mg/Fe] = 0.59 ± 0.11, this galaxy is the most Mg-enhanced massive galaxy found so far, having twice the Mg enhancement of similar-mass galaxies today. The abundance pattern of the galaxy is consistent with enrichment exclusively by core-collapse supernovae and with a star-formation timescale of 0.1 to 0.5 billion years -characteristics that are similar to population II stars in the Milky Way. With an average past star-formation rate of 600 to 3,000 solar masses per year, this galaxy was among the most vigorous starforming galaxies in the Universe.We observed the galaxy COSMOS-11494 with the near-infrared multi-object spectrograph MOSFIRE on the Keck I Telescope 8 . It was also observed by two other programmes 9, 10 , and so we incorporated these publicly available archival data. COSMOS-11494 was selected from the 3D-HST survey 11,12 . With a stellar mass M given by log10 M/M⊙ = 11.5 ± 0.1, COSMOS-11494 is among the most massive galaxies at its redshift, and it has a very low star-formation rate of less than 0.6M⊙/ yr (see Methods). Similarly to the typical massive, quiescent galaxy at this redshift, it is smaller than its local counterparts of the same mass, with an effective radius of 2.1 kpc 13 . The MOS-