Star formation in galaxy mergers with realistic models of stellar feedback and the interstellar medium

We use hydrodynamic simulations with detailed, explicit models for stellar feedback to study galaxy mergers. These high-resolution (similar to 1 pc) simulations follow the formation and destruction of individual giant molecular clouds (GMC) and star clusters. We find that the final starburst is dominated by in situ star formation, fuelled by gas which flows inwards due to global torques. The resulting high gas density results in rapid star formation. The gas is self-gravitating, and forms massive (less than or similar to 10(10) M-circle dot) GMC and subsequently super star clusters (with masses up to 10(8) M-circle dot). However, in contrast to some recent simulations, the bulk of new stars which eventually form the central bulge are not born in super-clusters which then sink to the centre of the galaxy. This is because feedback efficiently disperses GMC after they turn several per cent of their mass into stars. In other words, most of the mass that reaches the nucleus does so in the form of gas. The Kennicutt-Schmidt law emerges naturally as a consequence of feedback balancing gravitational collapse, independent of the small-scale star formation microphysics. The same mechanisms that drive this relation in isolated galaxies, in particular radiation pressure from infrared photons, extend, with no fine-tuning, over seven decades in star formation rate (SFR) to regulate star formation in the most extreme starburst systems with densities greater than or similar to 10(4) M-circle dot pc(-2). This feedback also drives super-winds with large mass-loss rates; however, a significant fraction of the wind material falls back on to the discs at later times, leading to higher post-starburst SFRs in the presence of stellar feedback. This suggests that strong active galactic nucleus feedback may be required to explain the sharp cut-offs in SFR that are observed in post-merger galaxies. We compare the results to those from simulations with no explicit resolution of GMC or feedback ['effective equation-of-state' (EOS) models]. We find that global galaxy properties are similar between EOS and resolved-feedback models. The relic structure and mass profile, and the total mass of stars formed in the nuclear starburst are quite similar, as is the morphological structure during and after mergers (tails, bridges, etc.). Disc survival in sufficiently gas rich mergers is similar in the two cases, and the new models follow the same scalings as derived for the efficiency of disc re-formation after a merger as derived from previous work with the simplified EOS models. While the global galaxy properties are similar between EOS and feedback models, subgalaxy-scale properties and the SFRs can be quite different: the more detailed models exhibit significantly higher star formation in tails and bridges (especially in shocks), and allow us to resolve the formation of super star clusters. In the new models, the star formation is more strongly time-variable and drops more sharply between close passages. The instantaneous burst enhancement can be higher or lower, depending on the details of the orbit and initial structural properties of the galaxies; first-passage bursts are more sensitive to these details than those at the final coalescence.