Comparing galaxy formation in semi-analytic models and hydrodynamical simulations

MNRAS 474, 492–521 (2018) doi:10.1093/mnras/stx2770 Advance Access publication 2017 October 25 Comparing galaxy formation in semi-analytic models and hydrodynamical simulations Peter D. Mitchell,1‹ Cedric G. Lacey,2 Claudia D. P. Lagos,3 Carlos S. Frenk,2 Richard G. Bower,2 Shaun Cole,2 John C. Helly,2 Matthieu Schaller,2 Violeta Gonzalez-Perez4 and Tom Theuns2,5 1 Université Lyon, Université Lyon1, Ens de Lyon, CNRS, Centre de Recherche Astrophysique de Lyon UMR5574, F-69230, Saint-Genis-Laval, France for Computational Cosmology, Department of Physics, University of Durham, South Road, Durham DH1 3LE, UK 3 International Centre for Radio Astronomy Research, 7 Fairway, Crawley, 6009, Perth, WA, Australia 4 Institute of Cosmology and Gravitation, Portsmouth University, Dennis Sciama Building, Burnaby Road, Portsmouth PO1 3FX, UK 5 Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium 2 Institute ABSTRACT It is now possible for hydrodynamical simulations to reproduce a representative galaxy population. Accordingly, it is timely to assess critically some of the assumptions of traditional semi-analytic galaxy formation models. We use the EAGLE simulations to assess assumptions built into the GALFORM semi-analytic model, focusing on those relating to baryon cycling, angular momentum and feedback. We show that the assumption in GALFORM that newly formed stars have the same specific angular momentum as the total disc leads to a significant overestimate of the total stellar specific angular momentum of disc galaxies. In EAGLE, stars form preferentially out of low-specific angular momentum gas in the interstellar medium due to the assumed gas density threshold for stars to form, leading to more realistic galaxy sizes. We find that stellar mass assembly is similar between GALFORM and EAGLE but that the evolution of gas properties is different, with various indications that the rate of baryon cycling in EAGLE is slower than is assumed in GALFORM. Finally, by matching individual galaxies between EAGLE and GALFORM, we find that an artificial dependence of active galactic nucleus feedback and gas infall rates on halo mass-doubling events in GALFORM drives most of the scatter in stellar mass between individual objects. Put together our results suggest that the GALFORM semi-analytic model can be significantly improved in light of recent advances. Key words: galaxies: evolution – galaxies: formation – galaxies: haloes – galaxies: stellar content. 1 I N T RO D U C T I O N Semi-analytic galaxy formation models are established tools for connecting the predicted hierarchical growth of dark matter (DM) haloes to the observed properties of the galaxy population (e.g. Cole et al. 2000; Somerville et al. 2008b; Guo et al. 2011). Unlike empirical abundance matching (e.g. Conroy, Wechsler & Kravtsov 2006; Moster et al. 2010) or halo occupation distribution models (e.g. Berlind & Weinberg 2002), semi-analytic models employ a forward-modelling approach and are constructed such that they contain as much as possible of the baryonic physics that is thought to be relevant to galaxy evolution, albeit at a simplified, macroscopic level. The simplified, macroscopic nature of semi-analytic models means that they are computationally inexpensive to evaluate. Compared to hydrodynamical simulations, this lack of computa-  E-mail: tional expense meant that until recently it was uniquely possible for semi-analytic models to predict realistic galaxy populations (e.g. Bower et al. 2006; Croton et al. 2006; Henriques et al. 2013). Recently, advances in computational resources combined with improvements in the uncertain modelling of feedback have allowed hydrodynamical simulations to predict galaxy populations which reproduce observations at an equivalent level to semi-analytic models for representative volumes (Vogelsberger et al. 2014; Schaye et al. 2015; Dubois et al. 2014; Davé, Thompson & Hopkins 2016). It is timely therefore to review the underlying assumptions which underpin semi-analytic models and assess their validity against state-of-the-art hydrodynamical simulations. As in semi-analytic models, hydrodynamical simulations are forced to implement uncertain subgrid modelling to approximate the effect of massive stars and black holes on galaxy evolution. This means that, for example, the dynamics of outflowing gas in these simulations are not necessarily realistic (irrespective of whether a realistic galaxy population is produced). Importantly however, the  C 2017 The Author(s) Published by Oxford University Press on behalf of the Royal Astronomical Society Accepted 2017 October 23. Received 2017 October 22; in original form 2017 June 27 Comparing SAMs and hydrodynamical simulations of comparison possible between GALFORM and EAGLE. Namely, to directly measure all of the mass, metal and angular momentum exchanges between different discrete baryonic reservoirs in EAGLE and compare with the corresponding quantities in GALFORM. As such, we consider here how to compartmentalize baryons in EAGLE between the corresponding discrete components that are tracked in semi-analytic models. In particular, we carefully consider how to separate the interstellar medium (ISM) from more diffuse halo gas in the circumgalactic medium (CGM) in EAGLE on physical grounds. The layout of this paper is as follows. We introduce the EAGLE simulations, the GALFORM semi-analytic model and describe our analysis methodology in Section 2. We present a first comparison of the models by analysing stellar masses in Section 3. We compare star formation thresholds and efficiencies as well as the angular momentum of star-forming gas in Section 4. We discuss feedback from supernovae (SNe) and active galactic nuclei (AGNs) in Section 6.2 and the resulting baryon cycle in Section 7. We discuss the consequences of qualitative differences between gas infall rates on to galaxies in the two models in Section 8. Finally, we summarize our main results in Section 10. Throughout this paper, we denote the units of distances in proper kiloparsecs as pkpc and comoving kiloparsecs as ckpc. Also throughout, log refers to the base 10 logarithm and ln refers to the natural logarithm. 2 M O D E L L I N G G A L A X Y F O R M AT I O N To facilitate a direct comparison of the EAGLE simulations and the GALFORM model, we follow Guo et al. (2016) by running GALFORM on a DM-only version of the reference EAGLE simulation run with a 1003 Mpc3 box (L100N1504 in the convention introduced by Schaye et al. 2015). As described by Guo et al. (2016), both simulations where performed with the same cosmological parameters taken from Planck Collaboration XVI (2014), and with the same initial conditions, following the method of Jenkins (2010). 2.1 EAGLE The EAGLE simulations are a suite of hydrodynamical simulations of the formation and evolution of galaxies within the context of the cold dark matter cosmological model. (...truncated)