The NewHorizon simulation

Hydrodynamical cosmological simulations are increasing their level of realism by considering more physical processes, having more resolution or larger statistics. However, one usually has to either sacrifice the statistical power of such simulations or the resolution reach within galaxies. Here, we introduce the NewHorizon project where a zoom-in region of  ~ (16  Mpc)3 embedded in a larger box is simulated at high resolution. A resolution of up to 40  pc, typical of individual zoom-in state-of-the-art resimulated halos is reached within galaxies, allowing such a simulation to capture the multi-phase nature of the interstellar medium.



Observational information on the properties of galaxies and their dependence on environment is becoming available for galaxies up to redshift two and beyond. Modern simulations have established a tight connection between the dynamics of the large-scale structure of matter and the evolution of the physical properties of forming galaxies, and their central black holes. Key questions formulated decades ago are nevertheless not satisfactorily answered. What are the main drivers determining the morphology of galaxies? Is the “downsizing” and early quenching of star formation in massive systems due to Active Galactic Nucleus (AGN) feedback, as commonly assumed, or to reduced gas infall from cosmic flows, or to both?

  In standard cosmological models of galaxy formation, gas trapped in collapsed dark matter halos is funneled through cold filamentary streams at high redshift. Those highways of gas accretion allow for a rapid replenishment of the reservoir of cold gas in high-redshift galaxies that are, hence, dominated by the gas component and favorise the formation of very clumpy discs. As it is observed, high-redshift galaxies are, therefore, very different from our local Universe galaxies that are more likely fed with a hot isotropic gas accretion. It is also a prediction of hydrodynamical cosmological simulations where ga- laxies look more clumpy, turbulent and distorted at high-redshift than they are at low-redshift.

At high-redshift, it is expected that the feedback of supernovae (SNe) and activity of central black holes (active galactic nuclei, AGN) have a stronger impact on galaxies and their nearby environment. Our pro- posed numerical project will be a simulation of a piece of the universe at z ≥ 2 with cosmological gas dynamics and several important galaxy gas physics (that we will detail further). The aim of this wo rk is to provide both a satisfactory description of the clumpy structure of high-redshift galaxies with their detailed cosmological infall and a representative large statistical sample of galaxies. With that simulation at hand, we will be able to understand the importance that anisotropic collimated gas infall, gas-rich mergers, SN feedback and AGN activity have on the evolution of galaxy morphologies at the peak epoch of cosmic star formation activity.

Previous large-scale hydrodynamical cosmological simulations Horizon- AGN [12], Illustris [38], Eagle or MassiveBlack-II [20] have been able to provide large mock datasets of galaxies at z = 0. Despite their success in providing a fresh view on the link between galaxy populations at z = 0 and the cosmic accretion and feedback activity, their limited spatial resolution and mass resolution due to large volumes (∼ 100 Mpc) reduced their capacity to provide insights on detailed galaxy kinematics (gas turbulence, gas fragmentation, etc., with ∆x = 1 kpc) and cosmic gas inflows (limited in halo mass resolution to a few 1010 M⊙ and cold gas cosmic streams are smeared out). These issues are even more problematic at high redshift, when galaxies were more compact and the interstellar gas more fragmented.

Basic properties

This new hydrodynamical cosmological simulation  focuses on the evolution of galaxies before and at the peak of the cosmic star formation (z ≃ 2), with a sufficiently high spatial, ∆x ≃ 35 pc, and mass, Mres ≃ 2 × 10^5 M⊙, resolution in order to capture both the collimated cold filamentary flows (see figure 1) and its an- gular momentum transfers in wide range of galaxies, as well as to resolve the clumpiness of the high-redshift gas-rich galaxies (see figure 2). This choice of dark matter (DM) mass resolution  allows us to resolve halos with masses larger than Mh,min > 2 × 108 M⊙. For comparison, Magellanic Cloud-like halos are resolved with 50 000 DM particles and their galaxy sampled with ≃ 8 cells in radius. This simulation, performed with the ramses code in ΛCDM cosmology, include DM particles, gas hydrodynamics, gas heating and cooling in the presence of SN-released metals, star formation, stellar feedback (including stellar winds, type II and type Ia SNe), black hole mass and spin evolution and AGN feedback, and tracer particles to follow the Lagrangian history of the gas.

The simulation is performed within a box of 140Mpc length with the initial conditions of the Horizon-AGN simulation [12], and a high-resolution zoom sphere of 20 Mpc diameter, where only into that “zoomed” region the highest spatial and mass resolution will be reached. The rest of the simulation volume is performed at much lower mass and spatial resolutions (refinement is not permitted) in order to capture the large-scale modes that would otherwise be missing in a standard uniformly resolved 20 Mpc box size. Note that the zoom region is chosen such that its total matter density is that of the average density of the Universe, and from the Horizon-AGN simulation so that we can use those two simulations to infer the effects of resolution. We choose a region of average density in order to capture the bulk population of AGN and galaxies.

The sub-grid models that we  employ for the NewHorizon simulation are those that several members of this collaboration have recently developed and tested in high-resolution simulations, therefore, they are the state-of-the-art of what can be included, without sacrificing the CPU time (radiative transfer and magnetic field are not included as they respectively cost an extra factor 2 and 4 respectively in CPU time). This includes :

Science case

The NewHorizon simulation is  designed to investigate the evolution of the circum-galactic medium (CGM), and in particular cold flows, the AGN activity with good resolution (that of usually individual halo zoom simulations), large enough statistics (we expect several 1000 galaxies within the stellar mass range 10^6 < M_gal < 10^11 M⊙) and detailed physical processes. In parallel,   MUSE on the VLT, that  provides  high-spatial resolution images of Lyman-alpha emitters and of stellar kinematics, makes NewHorizon absolutely timely. Observational sets of galaxy kinematics at high redshift are still very scarce, and integral field unit instruments such as MUSE (as well as JWST) will surely revolutionize our understanding of galaxy kinematics and evolution as well as of the properties of the CGM.

The  low  surface  luminosity  Universe  (a  super-set  of  the  low-mass Universe) will be an important frontier in studies of galaxy evolution  over  the  next  decade.  Dwarf  galaxies  are  the  fun-damental  building  blocks  of  galaxy  formation  in  hierarchical models. They represent the largest population in number. These dwarf galaxies can only be observed in detail at low redshift. The dynamical origin of these dwarfs remains largely unexplained, but seems to pose strong constraints on the intensity of stellar feedback, which remains one of the poorly constrained parame-ters of cosmological models of galactic formation. A large partof the tension between theory and observation lies in this regime (density of dwarf galaxy numbers, core-cusp problem, etc.). Surveys such as LSST (the optical reference instrument of the nextdecade) will probe this regime, and we will observe for the firsttime thousands of dwarfs at cosmological distances, but they willlargely be at z<0.5 (those that can be resolved by the instrument). Recent  work  quantify  the  details  of  the  formation  of  low  sur-face brightness galaxies at high redshift, but these mocks cannotbe easily compared to the observations, because LSST will notcorrectly resolve galaxies at z=0.7. EUCLID and JWST are tooshallow to provide high S/N images of the dominant dwarfs.

Most stars are born in stellar disks. Major mergers destroy some of these disks recurrently in the history of the Universe, but some have survived until today, including our own Milky Way.  Understanding the  long-term survival of these stellar disks is therefore an essential ingredient of modern cosmology.   The stabilization of the thickness of disks (thin and thick) in our Galaxy is also a dynamical problem that has recently been revived in the light of the APOGEE and Gaia surveys. Star formation generally occurs on the circular (non intersecting) orbits of the gas, so young stars typically form a very thin disk. However, chemo-dynamic observations of old Milky Way stars have all shown that thick disks are very common. The simultaneous formation of thin and thick stellar disks (S0 and Sc galaxies) is therefore an important puzzle for the theory of galactic formation. Various dynamic mechanisms, internal or external, have been proposed to explain the observed thickening or rejuvenation, but their respective impacts and roles remain to be quantified. Some major violent events contribute to the extended vertical distribution of stars in disks: accretion of galactic satellites, major mergers of gas-rich systems, or even gravitational instabilities in massive turbulent disks. Violent mergers certainly have a strong impact on galactic structure, but the thickening of stellar disks could also result from the slow and continuous heating of pre-existing thin disks, through multiple minor mergers. The induced density waves may also increase their dispersion in velocity, which can be converted into vertical motion via deflections on giant molecular clouds. Finally, the corresponding radial migration will also play an important role in their secular evolution.

The  epoch of  cosmic environment settling allows certain resonant stellar processes to take over to define the morphology of galaxies (bar formation, radial migration, disk heating and thickening, etc.). These discs are cold and therefore fragile dynamical systems for which rotation provides an important reservoir of free energy, and where orbital resonances play a key role. The availability of this free energy leads to a strong amplification of certain stimuli, while the width of the relevant resonances tends to localize their dissipation, with the net result that even a small disturbance can lead to disks evolving towards substantially distinct quasi-equilibria. These disks are furthermore immersed in various sources of perturbations, ranging from fluctuations coming from the cosmic environment,   shot noise coming from the finite number and short life span of giant molecular clouds in the interstellar medium, to globular clusters and substructures in orbit around the galaxy.  Spiral arms and central bars provide other sources of coherent stimulation. The cosmic history of galactic disks must therefore include the common responses to all these various stimuli (internal and external).   Bars have long been recognized as the most easily quantifiable morphological tracer, both in observations and in models (e.g. Bournaud, Combes 2002). Observationally the fraction of bars in galaxies is low until the redshift epoch z=0.7, then increases rapidly when the redshift decreases to 60-70% at z=0.3 until today (Sheth et al. 2008, Comeron et al. 2012).

Existing cosmological (zoom) simulations (Niaho, Apostle, Auriga, Fire, Vela, Eris), which now reach the resolution to model all these effects in principle, are beginning to produce thin disks but  are typically struggling to reproduce the observed fraction of barred galaxies: in fact, the presence of a bar requires a stellar disk that is both massive enough to be axi-symmetrically unstable, but also isolated/unperturbed enough to allow internal resonances to amplify the instability. This fraction thus represents a strong constraint on  stellar formation regulation and feedback mechanisms. Too much feedback makes the disc too light, too little feedback leads to a cusp or bulge that deflects the orbits underlying the bar.  The correct modelling of feedback will not be  sufficient by itself to explain the observed bar fractions: the angular momentum distribution of  accreted gas in the last billions of years of the Universe (Bournaud, Combes 2002), and the profile of the underlying dark matter halos (Athanassoula 2000), also seem to play a critical role in the evolution of  bars.  The observed fraction of bars  is an obvious diagnostic in this context, as we expect a significant evolution of this fraction between redshift 0.7 and 0.3.  Our ability to models and explain this is therefore a strong constrain on galaxy formation models.

NewHorizon's main objectives are to understand the importance of secular processes in galactic disks at low redshift: fraction of strong bars, ultra-thin disks, S0 galaxies and reproduce the distribution of low surface brightness galaxies that dominate the budget in number of galaxies at all masses.   The NewHorizon simulation is outstanding because of the volume re-simulated at a resolution of 35 pc up to redshifts z<0.5: most other galactic-scale simulations have opted for zooms centered on L* galaxies and thus probe a very specific environment in a limited volume. Most of these simulations  also do not include  AGN feedback.  The NewHorizon simulation  produces several thousands of dwarves at these redshifts and must in fact reproduce the local Universe as well as possible, in order to fully rely on its predictions at higher redshifts.
NewHorizon also  provides a self-consistent cosmological framework (through the formation of several hundred massive disks) to probe the interaction between  competing mechanisms for the stabilization of thin and thick disks,  the formation of bars (strong and weak).  As a subset of Horizon-AGN, it models the fraction of gas/stars at  redshift 2  (a crucial epoch because it corresponds to the peak of the stellar formation history) of 50%/50% (Welker et al. 2017), in good agreement with  observations (Tacconi et al. 2017), whereas the vast majority of competing simulations model rather 20%/80% or even 15%/85% (FIRE simulations: Oklopcic et al. 2016; ILLUSTRIS simulations: Keres et al. 2015).

Beyond these two flagship scientific cases, NewHorizon produces a statistical sample of high-resolution simulated galaxies sampling the Hubble sequence in their cosmic environment in order


This work was granted access to the HPC resources of CINES under the allocations 2013047012, 2014047012, 2015047012, c2016047637, A0020407637 made by GENCI and KSC-2017-G2-0003 by KISTI, and as  a ``Grand  Challenge’'  project  granted  by  GENCI  on  the  AMD-Rome  extension of the Joliot Curie supercomputer at TGCC, and under the the allocation 2019-A0070402192 made by GENCI. This research is part of the Spin(e) (ANR-13-BS05-0005, and Horizon-UK projects. This work has made use of the Horizon cluster on which the simulation was post-processed, hosted by the Institut d’Astrophysique de Paris. We warmly thank S. Rouberol for running it smoothly.