The Horizon-AGN simulation

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?

Over the past few years, our team has developed a suite of novel algorithms, based on recent progress in topology and computational geometry, to trace filaments right to the core of star forming galaxies (the Skeleton, Sousbie et al. 2008, 2011) and provide a formal definition of filaments. Our particular focus will be on creating the framework to quantify systematically the dynamics of the cosmic web jointly with the process of galaxy formation and black hole growth. We will numerically investigate the respective roles of merger/interaction history, secular evolution, stellar feedback, wind and nuclear activity, and AGN feedback on establishing the Hubble sequence and growing black holes (eg, Springel et al. 2005; Ocvirk et al. 2008, Dekel et al. 2009, Haas et al. 2013; Hirschmann et al. 2013), and then produce and analyse mocks and co-analyse them with existing and upcoming surveys to produce diagnostics and tests. A new generation of surveys such as COSMOS, ULTRAVISTA, VVDS and DEEP2 are now reaching critical size at z˖͂1 and even larger surveys such as the ongoing ESO large program VIPERS and the future eBOSS will follow.

One of the main challenges we need to address is the wide dynamical range involved in our endeavour: our purpose is to connect the large-scale structures (on scales >10 Mpc) to the inner regions of galaxies (e.g. bulges and discs, which can only be modelled convincingly when the disc scale height, ˖͂100 pc, is well resolved). For this purpose we run two types of hydrodynamical simulations:

Our flagship simulation (Horizon-AGN) allows us to produce and release catalogues of galaxies, AGN, and the gas distribution on the sky (via two sets of light cones produced on-the-fly): galaxy spectra, colours and magnitudes in different bandwidths (UV, IR, optical), X-ray emission of the hot plasma around galaxies, AGN luminosities, Lyman-alpha forest and metal distribution in the intergalactic medium will all be post-produced.

With high-resolution, high-detail zooms of around ten high-z halos, we will assess the role of inflows (feeding) and outflows (feedback) in growing MBHs observable as z˖͂5-7 quasars. These quasars are primary targets for ALMA, which will measure the host's gas reservoir, star formation rate, kinematics. We will calculate molecular gas masses, star formation rates and gas kinematics for comparison with the observations. Our other interest is to make predictions for the population of high-z quasars expected for JWST and Euclid in near-IR, and eRosita and NuSTAR in X-ray.

With the combination of a large volume simulation and highly detailed zooms, we will be able to analyse the link between galaxy evolution, AGN activity (and black hole growth) and the cosmic environment, together with the cosmological gas accretion (cold filamentary or hot diffuse) versus galaxy mergers. Following Dubois et al. 2012 and Bellovary et al. 2013, we will study the funneling of low-angular momentum gas to the centre of very massive halos at high redshift, thought to accounts for the rapid pace at which the most massive black holes reach observed masses around 109 Msun at an epoch when the Universe is barely one Gyr old. Such selective feeding only of the black holes in the most massive galaxies would naturally explain high-redshift observations (Volonteri & Stark 2011; Di Matteo et al. 2012).

One important aspect that will be explored is the consequence of feedback (AGN and stellar/SNe) on the re-distribution of angular momentum. In Dubois et al. (2013), we showed that AGN feedback in massive halos at high redshift is able to strongly disrupt the cold filamentary structure of the gas accretion, while, in contrast, Powell et al. (2011) argued that supernovae explosions are not able to perturb cold streams even in moderate mass galaxies. Thus, we will test the importance of feedback on altering the structure of the Skeleton of the cosmic web using the unique algorithms developed by our team.

This is the first multi-scale project of this size and scope that uses an adaptive mesh refinement code. The simulations will in fact be performed with the Ramses code (Teyssier, 2002). The code allows one to follow the evolution of three-dimensional systems including the effect of gravity, hydrodynamics, gas cooling and heating processes, and a number of sub-grid prescriptions for the physics of galaxy evolution (star formation, supernovae, AGN, etc.). Ramses is based on the adaptive mesh refinement technique which allows to zoom in with more resolution elements in regions of interests (structures and substructures such as cosmic filaments, dark matter halos, galaxies, giant molecular clouds, etc), and Riemann-based hydrodynamical solver which is very efficient at capturing sharp discontinuities in the gas flow, such as shocks at the interface of halos and cosmic gas or galactic winds and the circum-galactic gas, and gas instabilities that spontaneously arise in the cosmic turbulent medium. The simulation volume is sub-divided in different regions with a Peano-Hilbert curve decomposition that splits the computational effort between the several cores employed for the calculation. Each thread (or core) communicates with its neighbors with MPI calls in order to share variables (hydrodynamical variables, gravitational potential, particles crossing boundaries, etc.). The load balancing between CPUs is regularly recalculated to get optimized performance of the code. During the I/O step, each process outputs its own data, as it is the same for a restart of the simulation: each process reads its own corresponding data.

The main motivation for using this code is that i) it scales very well on highly parallel architectures; ii) it conserves well angular momentum to trace cold flows accurately; iii) it captures shocks very well; iv) it includes a large set of physical processes that are required to get realistic galaxies: gravity, hydrodynamics, cooling, heating, star formation, supernovae feedback and AGN feedback: much observational progress is under way to study the galaxies and quasars in the cosmic web, it is crucial that theoretical work keeps pace, in order to make predictions first, and then guide interpretation of data.

The Ramses code already led to the realization of some of the largest cosmological simulations. The Horizon-MareNostrum simulation performed on 2048 cores of the MareNostrum supercomputer in Barcelona during 4 weeks in 2007 (totaling 1 million CPU hours) was the largest hydrodynamical cosmological simulation of its time. The Horizon-4pi done in 2008 on 6144 cores on the BULL supercomputers in CCRT used 9 Mhours CPU. This simulation is comparable in size and resolution to the MareNostrum simulation, but it includes more physical processes (black holes and AGN; feedback from young stars and SNe Ia, plus metal enrichment through six chemical elements, O, Fe, C, N, Mg, Si). It will also reach lower redshift and produce light cones (dark matter, stars, gas, AGN) on-the-fly. From this numerical experiment we aim at understanding which physical processes lead galaxy formation, specifically focusing on environmental effects, accretion of gas from filaments, galaxy mergers, star formation, stellar and AGN feedback.