The origin of metals in the circum-galactic medium of massive galaxies at z=3
Sijing Shen, Piero Madau, Anthony Aguirre, Javiera Guedes, Lucio Mayer, James Wadsley
aa r X i v : . [ a s t r o - ph . C O ] S e p submitted to the ApJ Preprint typeset using L A TEX style emulateapj v. 5/2/11
THE ORIGIN OF METALS IN THE CIRCUMGALACTIC MEDIUM OF MASSIVE GALAXIES AT Z = 3 Sijing Shen , Piero Madau , , Anthony Aguirre , Javiera Guedes , , Lucio Mayer , and James Wadsley submitted to the ApJ ABSTRACTWe present a detailed study of the metal-enriched circumgalactic medium (CGM) of a massive galaxyat z = 3 using results from “ErisMC”, a new cosmological hydrodynamic “zoom-in” simulation of adisk galaxy with mass comparable to the Milky Way. The reference run adopts a blastwave scheme forsupernova feedback that generates galactic outflows without explicit wind particles, a star formationrecipe based on a high gas density threshold, and high temperature metal cooling. ErisMC mainprogenitor at z = 3 resembles a “Lyman break” galaxy of total mass M vir = 2 . × M ⊙ , virialradius R vir = 48 kpc, and star formation rate 18 M ⊙ yr − , and its metal-enriched CGM extends asfar as 200 (physical) kpc from its center. Approximately 41%, 9%, and 50% of all gas-phase metals at z = 3 are locked in a hot ( T > × K), warm (3 × K > T > × K), and cold (
T < × K)medium, respectively. We identify three sources of heavy elements: 1) the main host, responsible for60% of all the metals found within 3 R vir ; 2) its satellite progenitors, which shed their metals beforeand during infall, and are responsible for 28% of all the metals within 3 R vir , and for only 5% of thosebeyond 3 R vir ; and nearby dwarfs, which give origin to 12% of all the metals within 3 R vir and 95%of those beyond 3 R vir . Late ( z <
5) galactic “superwinds” – the result of recent star formation inErisMC – account for only 9% of all the metals observed beyond 2 R vir , the bulk having been releasedat redshifts 5 ∼ < z ∼ < − . Theoutflow mass-loading factor is of order unity for the main halo, but can exceed a value of 10 for nearbydwarfs. We stress that our “zoom-in” simulation focuses on the CGM of a single massive system andcannot describe the enrichment history of the intergalactic medium as a whole by a population ofgalaxies with different masses and star formation histories. Subject headings: galaxies: evolution – galaxies: high-redshift – intergalactic medium – method: nu-merical INTRODUCTION
Studies of the ionization, thermodynamic, and kine-matic state of heavy elements in circumgalactic gas holdclues to understanding the exchange of mass, metals,and energy between galaxies and their surroundings.The distribution of observed metals in the intergalacticmedium (IGM) is highly inhomogeneous, with a globalcosmic abundance at z = 3 of [C/H]= − . ± .
13 for gaswith overdensities 0 . < δ <
100 (Schaye et al. 2003).The characteristic epoch of this enrichment and its maindonors remain uncertain. Late supernova-driven “super-winds” from massive galaxies (e.g., Aguirre et al. 2001;Adelberger et al. 2003), early outflows from dwarf galax-ies (e.g., Dekel & Silk 1986; MacLow & Ferrara 1999;Madau, Ferrara, & Rees 2001; Mori, Ferrara, & Madau2002; Scannapieco, Ferrara, & Madau 2002;Furlanetto & Loeb 2003), quasar-driven winds (e.g.,Scannapieco & Oh 2004), and the ejection of gas duringthe merging of protogalaxies (Gnedin 1998) are all likely Department of Astronomy and Astrophysics, University ofCalifornia, Santa Cruz, 1156 High Street, Santa Cruz, CA 95064. Institute of Astronomy, Madingley Road, Cambridge CB30HA, United Kingdom. Institute for Astronomy, ETH Zurich, Wolgang-Pauli-Strasse27, 8093 Zurich, Switzerland. Institute of Theoretical Physics, University of Zurich, Win-terthurerstrasse 190, CH-9057 Zurich, Switzerland. Department of Physics and Astronomy, McMaster Univer-sity, Main Street West, Hamilton L8S 4M1, Canada. to have left some chemical imprint on circumgalacticand intergalactic gas, but their relative contributions arenot constrained by present data, and involve complexbaryonic processes that are difficult to model.Observations clearly show that galactic-scale outflowswith velocities of several hundred km s − are ubiq-uitous in massive star-forming galaxies at high red-shift and in starburst galaxies in the local universe(e.g., Heckman, Lee, & Miley 1990; Pettini et al. 2001;Martin 2005; Veilleux, Cecil, & Bland-Hawthorn 2005;Weiner et al. 2009; Steidel et al. 2010). They also in-dicate that the metal content of the z ∼ VI systems are found to be strongly associated with knownLyman-break galaxies (LBGs), most detectable C IV sys-tems ( N CIV ∼ > . cm − ) lie within 1 proper Mpcfrom an LBG, and roughly one-third of all “intergalactic”absorption lines with N CIV ∼ > cm − are producedby gas that lies within ∼
80 proper kpc from an LBGs(Adelberger et al. 2003, 2005a). Somewhat puzzling, thecomoving mass density of C IV ions in the IGM is foundto drop by a factor of 4 at z ∼ > N -body/smooth particle hydro-dynamic (SPH) simulation of extreme dynamic range,a twin of the “Eris” simulation (Guedes et al. 2011).Termed “ErisMC”, the simulations follows the assemblyof a massive galaxy halo with a spline softening length of120 pc and 26 million dark matter and SPH particles inthe high-resolution region. The feedback from an activegalactic nucleus is neglected, and a star formation recipeis adopted based on a high gas density threshold. Heatingby supernovae occurs in a clustered fashion, and the re-sulting pressure-driven outflows at high redshifts removelow angular momentum metal-enriched gas. As shownbelow, our detailed study of ErisMC’s CGM at z = 3 re-veals that late galactic superwinds – the result of recentstar formation – account for only a small fraction of allthe metals found between ∼
100 and 200 (proper) kpcfrom the center of the main host, and resolves two othersources for the heavy elements found in ErisMC’s envi-ronment: its satellite progenitors – which deposit theirmetals before and during infall – and the orbiting nearbydwarfs.The paper is organized as follows. In § § §
4. The properties of the supernova-driven outflowsfrom the main host are analyzed in §
5. Finally, in § Z ⊙ = 0 . THE ERISMC SIMULATION
The Eris suite of simulations was performed in a
Wilkinson Microwave Anisotropy Probe
Gasoline (Wadsley et al. 2004). Details of the main simulation aregiven in Guedes et al. (2011), and are briefly summarizedhere. The high-resolution region, 4 comoving Mpc on aside, is embedded in a low-resolution, dark matter-only,periodic box of 90 comoving Mpc on a side, and contains13 million dark matter particles and an equal number ofgas particles, for a final dark and gas particle mass of m DM = 9 . × M ⊙ and m SPH = 2 × M ⊙ , respec-tively. The gravitational softening length, ǫ G , was fixedto 120 physical pc for all particle species from z = 9to the present, and evolved as 1 / (1 + z ) from z = 9 Fig. 1.—
The optical/UV stellar properties of ErisMC at z = 3.The images were created with the radiative transfer code Sunrise (Jonsson 2006), and show a rest-frame B , U , and NUV stellarcomposite of the simulated galaxy seen face-on ( top panel ) andand edge-on ( bottom panel ). The 10 kpc bar in the bottom panelis in proper units. to the starting redshift of z = 90. Compton cooling,atomic cooling, and metallicity-dependent radiative cool-ing at low temperatures (Mashchenko et al. 2006) areincluded. A uniform UV background modifies the ioniza-tion and excitation state of the gas and is implementedusing a modified version of the Haardt & Madau (1996)spectrum.Star formation occurs stocastically when cold ( T < hen et al. 2011 3
Fig. 2.—
The mass assembly history of ErisMC. The growthof the stellar component ( red curve ) is compared to the growthof the dark matter halo ( black solid curve ). For comparison, wealso show the stellar mass in Eris ( blue curve ) as well as the best-fit to the median virial mass for progenitors of halos with mass M vir = 8 × M ⊙ at z = 0 in the Bolshoi, MultiDark, andConsuelo simulations (Behroozi, Wechsler, & Conroy 2012). × K), virialized gas reaches a threshold density n SF = 5 atoms cm − and is part of a converging flow. Itproceeds at a rate dρ ∗ /dt = 0 . ρ gas /t dyn ∝ ρ . (1)(i.e. locally enforcing a Schmidt law), where ρ ∗ and ρ gas are the stellar and gas densities, and t dyn is the localdynamical time. Each star particle has an initial mass m ∗ = 6 × M ⊙ and represents a simple stellar popula-tion with its own age, metallicity, and a Kroupa et al.(1993) initial stellar mass function (IMF). Star parti-cles inject energy, mass, and metals back into the ISMthrough Type Ia and Type II SNe and stellar winds, fol-lowing the recipes of Stinson et al. (2006). Each SN IIdeposits metals and a net energy of 0 . × ergs intothe nearest neighbor gas particles, and the heated gashas its cooling shut off (to model the effect of feedbackat unresolved scales) until the end of the momentum-conserving phase of the SN blastwave, which is set bythe local gas density and temperature and by the to-tal amount of energy injected (McKee & Ostriker 1977).Cooling is turned off only for those particles within theblast radius (for a maximum of 32 neighboring gas parti-cles), and no kinetic energy is explicitly assigned to them.For the typical conditions of star-forming clouds resolvedin this study, this translates into just one gas particleheated up by a SN and having its cooling shut off for atimescale t E ∼ × yr. The energy injected by manySNe adds up to create larger hot bubbles and longer shut-off times. The advantage of this physically-motivatedfeedback model compared to other “sub-grid” schemes(e.g. Springel & Hernquist 2003; Oppenheimer & Dav´e2008) is that it keeps galactic outflows hydrodynam-ically coupled to the energy injection by SNe (albeitwith a delay). In combination with a high gas densitythreshold for star formation (which enables energy de-position by SNe within small volumes), this scheme hasbeen found to be key in producing realistic dwarf galax-
50 kpc −3.0 −2.5 −2.0 −1.5 0.0−1.0 50 1 2 3 6 (ρ/<ρ>) log 4
Fig. 3.—
Projected gas density ( top panel ) and gas particle metal-licity ( bottom panel ) of ErisMC’s CGM at z = 3 in a cube of 500(proper) kpc on a side. The galaxy’s stellar disk is edge-on in thisprojection. ies (Governato et al. 2010) and late-type massive spirals(Guedes et al. 2011).Metal enrichment from SN II and SN Ia follows themodel of Raiteri et al. (1996). For SN II, metals arereleased as the main sequence progenitors die and dis-tributed to gas within the blastwave radius. Iron andoxygen are produced according to the following fits tothe Woosley & Weaver (1995) yields: M Fe = 2 . × − (cid:18) m ∗ M ⊙ (cid:19) . M ⊙ , (2) Metals in circumgalactic mediumand M O = 4 . × − (cid:18) m ∗ M ⊙ (cid:19) . M ⊙ . (3)Each SN Ia produces 0.63 M ⊙ of iron and 0.13 M ⊙ ofoxygen (Thielemann et al. 1986) and the metals are dis-tributed between the nearest gas particles. Radiativecooling is not disabled following a SN Ia. Stellar windfeedback was implemented based on Kennicutt et al.(1994), and the returned mass fraction was determinedfollowing Weidemann (1987). The returned gas has thesame metallicity as the star particle.The reference simulation discussed here, “ErisMC”,includes metallicity-dependent radiative cooling at hightemperatures following Shen et al. (2010). Aside fromhigh-temperature metal cooling, Eris and ErisMC wererun with exactly the same setup. ERISMC’S CIRCUMGALACTIC MEDIUM
Figure 1 shows the optical/UV stellar properties ofErisMC at z = 3. The mock images were created us-ing the radiation transfer code Sunrise (Jonsson 2006),which produces spectral energy distributions using theage and metallicities of each simulated star particle, andtakes into account the three-dimensional effect of dustreprocessing. At this epoch, the simulated galaxy hasa virial mass of M vir = 2 . × M ⊙ , a virial ra-dius of R vir = 48 kpc, a total stellar mass of M ∗ =2 . × M ⊙ , and is forming stars at the rate of SFR=18 M ⊙ yr − .The mass assembly history of ErisMC’s stellar compo-nent and dark halo is shown in Figure 2. While ErisMC isin the right range of halo masses ( M vir = 10 . ± . M ⊙ ,Adelberger et al. 2005a) and stellar masses ( M ∗ =10 . ± . M ⊙ , Shapley et al. 2005) of LBGs at redshifts2-3, by z = 3 it has formed 70% more stars than Eris asa consequence of the increased metallicity-dependent ra-diative cooling at high temperatures. We note here thata recent new simulation of the Eris suite (“Eris2”, seeShen et al. 2012) that includes metal diffusion, a Kroupa(2001) IMF that boosts the number of Type II super-novae per unit stellar mass by about a factor of 2 com-pared to Kroupa et al. (1993), and metallicity-dependentradiative cooling at all temperatures, produces by red-shift 3 the same stellar mass of Eris. This shows howuncertanties in the IMF, metal diffusion and gas coolingproperties can affect the star formation history of thesimulated galaxy. A comprehensive analysis of the aver-age star formation rates and histories of galaxies andtheir connection to the underlying growth and merg-ing of dark matter halos has been recently presented byBehroozi, Wechsler, & Conroy (2012), and it is interest-ing to discuss ErisMC in this context. With a specificstar formation rate (sSFR=SFR/ M ∗ ) of 8 . × − yr − at z = 3, ErisMC is consistent with the sSFR expected atthese redshifts for such massive stellar systems (see Fig.4 of Behroozi, Wechsler, & Conroy 2012). Accordingto Behroozi, Wechsler, & Conroy (2012), halos of mass ∼ M ⊙ are the most efficient at forming stars at ev-ery epoch, with baryon conversion efficiencies of 20-40%that are constant to within a factor of 2 over a remarkablylarge redshift range. ErisMC’s efficiency, about 50%, ap-pears then too high for a “typical” M vir = 2 . × M ⊙ halo at z = 3. Figure 2 shows, however, that the dark matter accretion history of Eris (the same as ErisMC) isfar from “typical”: when compared to the median virialmass for progenitors of halos with mass 8 × M ⊙ at z = 0 in the Bolshoi, MultiDark, and Consuelo simu-lations (see fit in Behroozi, Wechsler, & Conroy 2012),Eris’s fractional growth appears to be more skewed to-wards high redshift. While it is conceivable that this maycause Eris (and ErisMC) to form their stars ”too early”,it is fair to keep in mind that zoom-in hydrodynamicalsimulations may generically suffer from having star for-mation efficiencies that are too high at early epochs, asrecently argued by Moster, Naab, & White (2012).A 2D photometric decomposition performed on thedust-reddened rest-frame i -band light distribution withthe Galfit program (Peng et al. 2002) shows the pres-ence in ErisMC of an extended stellar disk with radialscale length R d = 0.6 kpc. The total gas and stellarmetallicities are Z g = 0 . Z ⊙ and Z ∗ = 0 . Z ⊙ , re-spectively. The galaxy’s gaseous disk (defined by all thecold, T < × K gas within 15 comoving kpc from thecenter) is characterized by 12 + log (O / H) = 8 .
4. Thisis below the mass metallicity relation at z = 0 from the Sloan Digital Sky Survey (Tremonti et al. 2004) but inagreement with the z ∼ R vir ) from the center, within the same re-gion we also identify 47 self-bound dwarfs galaxies withmasses M s > M ⊙ (424 with M s > M ⊙ ), many ofthem also forming stars and polluting their surroundings.The total mass of heavy elements in the gas phase within R vir , 2 R vir , and 3 R vir is 1 . × M ⊙ , 2 . × M ⊙ ,and 3 . × M ⊙ , respectively. A region ∼
100 kpc insize is enriched to metallicities above 0.03 solar. Theextent of the metal enriched region is consistent with re-cent observations of circumgalactic metals around LBGs(Steidel et al. 2010).Figure 4 shows a projected gas metallicity and velocitymap in a 500 × ×
10 kpc slice. The arrows indi-cate the direction and magnitude of the mass-weightedpeculiar velocity field relative to the center of ErisMC.Galactic winds can be clearly seen propagating perpen-dicularly to the disk (seen edge-on in this projection)well beyond the virial radius, with average velocities(over the slice) that exceed 250 km s − (we shall see be-low that individual gas particle speeds can reach 800km s − ). The outflows have a bipolar distribution witha smaller opening angle near the base, similar to theobserved morphologies of galactic winds in star-forminggalaxies (Veilleux, Cecil, & Bland-Hawthorn 2005). In-flows along large-scale filaments bring in both pristinegas (gray arrows in the figure), as well as material pre-enriched by nearby dwarfs (colored arrows on the left sideof the figure and in the inset). Inflowing cold streamsthat penetrate deep inside the virial radius are com-monly seen in cosmological hydrodynamical simulations(e.g. Kereˇs et al. 2005; Ocvirk et al. 2008; Dekel et al.2009), and have been shown to give origin to C II absorp-tion with a significant covering factor (Shen et al. 2012).hen et al. 2011 5 Fig. 4.—
Same as Fig. 3, showing the mass-weighted gas metallicity and velocity field in a 500 × ×
10 kpc slice. The galaxy centeris indicated by the plus sign at coordinates ( x, y ) = (0 , − (and 164km s − in the inset). The inset in the bottom left of the panel shows a zoom into the velocity field around one of ErisMC nearby dwarfs( M s = 5 × M ⊙ ) relative to its own center (indicated by a plus sign in the inset). While being accreted, the dwarf is also venting heavyelements into the surroundings. Note that, owing to the averaging process, the metallicities and velocities plotted in this figure can besignificantly lower than the corresponding quantities for individual gas particles. It is instructive at this stage to look at the mass-weighted distribution of all (inside and outside the mainhost) enriched gas at z = 3 in the temperature-densityplane. Figure 5 indicates that metals are spread over alarge range of phases, from cold star-forming material at T < × K and n > n SF = 5 atoms cm − (corre-sponding to δ ≡ ρ/ρ mean > × at z = 3) to hot T > K low density δ = 3 intergalactic gas that can-not cool radiatively over a Hubble time. The black stripin the lower left corner of the figure marks the pristine,adiabatically cooling IGM, while the colored swath inthe lower right corner shows dense, metal-rich gas in thegalaxy disk cooling down below 10 K. Hot enriched gasvented out in the halo by the cumulative effect of SNexplosions can be seen cooling and raining back onto thedisk in a galactic fountain. Intergalactic gas in the range1 ∼ < δ ∼ <
10 shows a strong positive gradient of metallicitywith increasing temperature, and has the largest rangeof metallicities, extending from solar all the way down tozero. We note that this is the temperature-density plane for gas in a limited zoom-in region surrounding the maingalaxy, and that these results may not be representativeof the z = 3 IGM as a whole and of the CGM of othergalaxies.A census of all the gas-phase metals in the cold ( T < × K), warm (3 × K < T < × K), and hot(
T > × K) interstellar and circumgalactic mediumwithin 1 comoving Mpc (250 physical kpc at redshift 3)from the center of ErisMC is depicted in Figure 6 (toppanel) as a function of redshift. The fraction of metals inthe cold gas drops from about 70% at redshift 8 to 50% at z = 3. Hot gas is the second most important reservoir ofgas-phase heavy elements, with a fraction of metals thatincreases from 15% at z = 8 to 40% at z = 3. This phasewould remain undetected in UV spectroscopic studies ofhigh redshift galaxies, and would therefore contribute tothe “missing metals” (Pettini 2006; Bouch´e et al. 2007).The metal mass fraction in warm gas remains around10% at all epochs. Note that the sudden increase in theamount of hot metals at z < Fig. 5.—
Distribution of all enriched gas in the temperature-density plane at z = 3 within 250 physical kpc from the center ofthe main host. The color scale indicates the mass-weighted metal-licity. flow activity at these epochs rather than by the transitionfrom “cold” to “hot” accretion mode in ErisMC.More than 40% of all gas-phase metals at z = 3 lieoutside the virial radius: while cold metal-rich materialtraces large overdensities within the main host, about50% of all warm and 70% of all hot metals are foundin low density δ <
30 regions beyond the virial radius,a point illustrated in the bottom panel of Figure 6. In-tergalactic metals are characterized by a strong temper-ature gradient with overdensity, as the metal-weightedtemperature climbs from 10 K at δ = 1 to 2 × K at δ = 10.The metallicities of the individual cold, warm, and hotgas components, as well as that of the total gas phase,are shown in Figure 7 as a function of gas overdensityin four redshift bins. The gas-phase metallicity distri-bution at z = 3 shows a positive density gradient abovean overdensity of log δ = 2, with little redshift evolu-tion as heavy elements removed from high density regionsare steadily replenished. The mean total metallicity ex-hibits a plateau around log δ = 1 −
2, where the metalstransported by SN-driven winds accumulate, and dropsquickly below log
Z/Z ⊙ = − δ ∼ < z = 3, the mean metal-licity of cold gas drops quickly below log Z/Z ⊙ = − δ ∼ < Fig. 6.—
Top panel:
Evolution with redshift of the mass fractionof heavy elements in the gas phase within 1 comoving Mpc (250physical kpc at z = 3) from the center of the main host. Thethree curves shows the fraction of metals in the cold (
T < × K), warm (3 × K < T < × K), and hot (
T > × )interstellar and circumgalactic gas. Bottom panel:
Cumulativemetal mass fraction in each gas phase at z = 3 as a function ofoverdensity. cosmological volume simulations of IGM pollution thatuse different sub-grid prescriptions for generating galac-tic outflows (Wiersma et al. 2009; Cen & Chisari 2011;Oppenheimer et al. 2011). Cen & Chisari (2011) injectthermal energy and metals from SN feedback on tensof kiloparsecs scales, do not turn off hydrodynamic cou-pling between the ejected metals and the surroundinggas, and, like in ErisMC, find that a large fraction of allthe heavy elements in the gas phase at redshift 3 havetemperatures in excess of 3 × K. Their metallicity-density relation for cold gas also shows a low-densitypeak, albeit shifted towards underdense regions com-pared to ErisMC. For comparison, in the momentum-conserved wind implementation of kinetic feedback byOppenheimer et al. (2011) (where hydrodynamic forcesare temporarily turned off) there is no low-density peakin the metallicity-density relation and metals largely re-side in cool gas. IGM metals reside primarily in a warm-hot component in the simulations of Wiersma et al.(2009), who also use kinetic feedback but with non-decoupled wind models. Measurements of the distribu-tion of carbon in the IGM using pixel statistics yield,hen et al. 2011 7
Fig. 7.—
Mean metallicity of cold, warm, and hot gas within 1 comoving Mpc (250 physical kpc at z = 3) from ErisMC’s center as afunction of gas overdensity in four redshift bins. The thin blue line shows the total gas metallicity.
Fig. 8.—
Fractional amount of metals within 250 physical kpcfrom ErisMC’s center at redshift 3 at overdensity δ ( z = 3) thatwas added to gas particles before redshift z = 4 , , . , . , . at z = 3 and log δ = 0 .
5, [C/H] ≈ − . − . THE ORIGIN OF CIRCUMGALACTIC METALS
In the absence of metal diffusion (Shen et al. 2010),the SPH technique allows us to trace back in time theenrichment history of every gas particle. In this andthe following section we use simulation outputs fromthe ErisMC run and the Amiga’s Halo Finder (AHF,Knollmann & Knebe 2009) to identify the age of the met-als observed at redshift 3 and the sources of pollution –whether nearby dwarfs, satellite progenitors, or the mainhost.Figure 8 shows the fraction of metals at redshift 3 thatwas released to gas particles at overdensity δ ( z = 3) be-fore redshifts z = 4 , , . , . , .
9. The epoch at which agas element is enriched is clearly a sensitive function of itsoverdensity at some later time. The diffuse IGM is typ-ically enriched earlier than high density regions, a trendthat is in agreement with the results of Wiersma et al.(2010) and Oppenheimer et al. (2011). More than 50%(35%) of all z = 3 metals at the average density weresynthesized before z = 5 ( z = 6), while newly producedmetals are mostly confined to high overdensities. Metals in circumgalactic medium Fig. 9.—
Projected gas metallicity map of ErisMC’s CGM at z = 3 in a 500 × ×
300 kpc (proper) box. The galaxy center andits virial radius are indicated by the plus sign and the black circle, respectively. The color coding indicates the mean gas metallicity in acolumn of area dxdy = 7 . × . . The different panels highlight different enrichment redshift ranges. Gas enriched at earlier epochsgenerally lies at larger distance to the main host’s center, as galactic outflows transport metals from the dense regions of star formationinto the CGM on cosmological timescales. At galactocentric distances >
130 kpc the IGM is enriched mostly by nearby dwarfs.
Fig. 10.—
Gaseous metal mass in 2.5 kpc thick radial shells atvarying distances from the center at z = 3. These metals werereleased by z = 3 from ErisMC’s main halo ( solid lines ) and itssatellites and nearby dwarfs ( dashed lines ). The colors indicatemetals produced at different enrichment epochs. There are two possible causes for this “outside-in” (aterm we borrow from Oppenheimer et al. 2011) enrich-ment of ErisMC’s CGM: 1) metals released at earliertimes into the high-density regions of ErisMC main hostand transported into the IGM via galactic winds ontimescales that are comparable to the age of the uni-verse at z = 3; and 2) metals released at earlier times indwarf galaxies and shed into the surrounding intergalac-tic and circumgalactic medium before and/or during in-fall. Dwarf galaxies are referred to as “nearby dwarfs” ifthey are still orbiting outside R vir at z = 3, and as “satel- lite progenitors” if they have been accreted by the mainhost before redshift 3. To identify the time, location,and source of enrichment of a given gas particle, we de-fine a metal mass-weighted redshift as in Wiersma et al.(2010): h z en i = P i ∆ m Z,i z i P i ∆ m Z,i , (4)where ∆ m Z,i is the metal mass gained by the gas particlein an enrichment event at redshift z i ≥
3, and P i ∆ m Z,i is the total metal mass of the particle at z = 3. Gasparticles in ErisMC typically receive metals more thanonce, with about half of them enriched in more thanthree events. For every enrichment episode we also derivethe mass of the satellite or nearby dwarf in which thegas particle resides, M h,i and the distance between thegas particle and the center of ErisMC, d i . Analogouslyto equation (4), we then define a metal mass-weightedsource halo mass and distance from ErisMC center as h M en i = P i ∆ m Z,i M h,i P i ∆ m Z,i (5)and h D en i = P i ∆ m Z,i d i P i ∆ m Z,i . (6)respectively. We have separated gas particles into dif-ferent groups according to their h z en i and plotted inFigure 9 the projected metallicity of each group at red-shift 3. Metals within ErisMC’s virial radius are clearly“younger”, i.e. they are characterized by an enrichmentredshift h z en i between 3 and 3.5. Because of the longhen et al. 2011 9 Fig. 11.—
Metal mass in ErisMC’s CGM as a function of the mean source halo mass h M en i . Each panel indicates the physical distanceof the gas from the main halo’s center at z = 3. Solid lines: main halo plus satellites and nearby dwarfs.
Dashed lines: main halo only.The mass bin in all panels is ∆(log h M en i ) = 0 . wind propagation time, “older” metals with h z en i be-tween 4 and 5 are spread over 100 (physical) kpc per-pendicularly to ErisMC’s disk. Low-metallicity gas inthis enrichment redshift range can be seen as far as 200kpc from the main host center as it is ejected from nearbydwarfs (see also Figs. 3 and 4). There is little materialcontaminated by metals at h z en i > h z en i <
5) galactic “superwinds” –the result of recent star formation in ErisMC’s main host– are found to account for less than 9% of all the metalsobserved beyond 2 R vir ,Figure 10 sheds light on the role played by satellitesand nearby dwarfs in contaminating ErisMC’s circum-galactic medium. It shows the total mass of heavy ele-ments released by the main host and its satellites as afunction of distance from ErisMC’s center at redshift 3.About 60% of all the metals within 100 kpc of the centeroriginate from the main host, and the rest from its satel-lites. Beyond 100 kpc, nearby dwarfs start dominatingthe metal budget. Both the host and the satellites con-tribute to the recent ( z <
4) pollution of gas within thevirial radius. Older metals within R vir typically form insatellite progenitors, collect along the filaments into themain host, and are not blown away by galactic outflows.Note how, within 150 kpc or so, the distribution of met-als from satellites is rather smooth and follows that fromthe main host, an indication that gas polluted by starformation in satellite progenitors is stirred up and wellmixed with ErisMC’s galactic outflows. Spikes due toindividual nearby dwarfs can be seen beyond 85 kpc.It is interesting at this stage to look at the masses of the satellites that contribute to the enrichment of the CGM.Figure 11 shows the metal-weighted mean halo mass, h M en i (defined in eq. 5), for gas at different physical dis-tances from ErisMC’s center. Most of the satellites’ met-als come from systems more massive than 10 M ⊙ , withthe peak of the distribution typically around 10 . M ⊙ .Satellites smaller than 10 . M ⊙ do not cause significantpollution as they are unable to form many stars bothbecause of SN feedback and the suppression of baryonicinfall by the UV background. Also, halos smaller than10 M ⊙ are resolved by less than 1,000 particles in oursimulation, and the inability to properly resolve high den-sity star-forming regions in these objects may result in anumerical suppression of star formation (as our recipe isbased on a high gas density threshold). The right bot-tom panel shows how gas beyond 150 kpc is enrichedonly by nearby dwarfs. Figure 12 provides some insighton the overall transport of enriched gas by showing themetal mass distribution as a function of the enrichmentdistance (as defined in eq. 6). We use comoving dis-tances to highlight departures from pure Hubble flow.Most host metals are released within 50-100 comovingkpc from the center, with those found beyond the virialradius at z = 3 originating earlier from strong galacticwinds launched closer to the center. Metals from satel-lites can have very different kinematics. Some are ejecteddirectly into the IGM by nearby dwarfs, other, initiallydeposited in the halo of the main host by infalling dwarfprogenitors, become part of ErisMC’s galactic outflows.We find that about 40-50% of all the satellite-producedmetals found in a given distance intervals at z = 3 were0 Metals in circumgalactic medium Fig. 12.—
Metal mass distribution at z = 3 as a function of the comoving mean enrichment distance h D en i . The dotted vertical linesand the legend in each panel mark the comoving distance interval from ErisMC’s center of the enriched gas under consideration at z = 3. Solid lines: main halo plus satellites and nearby dwarfs.
Dashed lines: main host only. produced at larger distances and then transported in-wards to their current location. For example, 42% ofall satellite metals found at z = 3 in the range 400-600(comoving) kpc from ErisMC’s center have a mean en-richment distance larger than 600 kpc. All metals foundbeyond 600 comoving (150 physical) kpc were ejected bynearby dwarfs that have not been yet accreted by thehost.To better grasp the kinematics of the material enrichedby satellite progenitors (i.e. dwarf galaxies that have allbeen accreted by the main halo before z = 3), we plotin Figure 13 a metal column density map of ErisMC’scircumgalactic environment at three different observerredshifts, z obs = 3 , , All the heavy elements shownin the figure were produced by satellite progenitors andnearby dwarfs at epochs ≤ z ≤ . All distances arecomoving and the metals are separated according to theirradial peculiar velocities relative to the center of the mainhost, v r , in inflowing ( v r <
0, left panels) and outflowing( v r >
0, right panels). The projected mass distributionis shown in the gray scale. Inflowing metals at z obs = 5have contaminated the vicinity of their satellite hosts,and are now being accreted onto the main host along thefilamentary structure. They are subsequently dispersedduring the infall and tidal disruption of their satellitehosts, become entrained in the galactic wind of the mainhost, and are ultimately ejected in a bipolar outflow. PROPERTIES OF GALACTIC OUTFLOWS
In this section, we study the properties of ErisMC’sgalactic ouflows and compare them with the observations as well as with other galactic wind models adopted incosmological simulations.
Outflow velocity
An example of the evolution with time of the ra-dial velocity of polluted material is shown in Figure14, where we have selected gas particles that wereenriched for the last time in the main host at red-shift 5, and that are unbound at z = 3. Gas ac-cretes with negative peculiar velocity onto the host,is enriched and accelerated to outflow velocities ofa few hundred km s − by blastwave feedback, andslows down as it sweeps the ISM and halo material.The peak velocity is comparable to those observedin high redshift LBGs (e,g., Adelberger et al. 2003;Veilleux, Cecil, & Bland-Hawthorn 2005; Steidel et al.2010). It is also comparable to the veloci-ties adopted in various kinetic feedback models(e.g., Springel & Hernquist 2003; Oppenheimer & Dav´e2006; Dalla Vecchia & Schaye 2008; Wiersma et al. 2009;Choi & Nagamine 2011). After reaching a peak value,the gas’ proper velocity declines rapidly, and particlesare swept by the Hubble flow as they move farther fromthe center into the IGM. This behaviour is typical of allenriched gas particles. The rapid decline of the propervelocity is also seen in the kinetic feedback model ofDalla Vecchia & Schaye (2008), where outflowing parti-cles are allowed to interact hydrodyamically with theISM, and pressure forces within the disk significantly de-crease the wind speed.Observations of local starburst galaxies (e.g.,hen et al. 2011 11 Fig. 13.—
Metal column density map of ErisMC’s circumgalactic environment at three different observer redshifts, z obs = 3 , ,
5. Allthe heavy elements shown in the figure are produced by satellites and nearby dwarfs at epochs 5 ≤ z ≤
7. The projected region has acomoving size of 2 × × and is centered on the main host. The center of the host galaxy and its virial radius are indicated bythe plus sign and black circle, respectively. The color scale indicates the logarithmic of total metal mass a column dxdy = 28 . × . comoving, in units of M ⊙ . Left panels : inflowing material.
Right panels : outflowing material.
Schwartz & Martin 2004; Rupke et al. 2005) haveshown that outflow speeds are correlated with halomasses and star formation rates, and that the correla-tion flattens out for SFR >
10 M ⊙ yr − (Rupke et al.2005). To investigate the existence of such relationshipin the ErisMC simulation, we plot in the upper panel ofFigure 15 the peak velocity of gas particles that becomeunbound after they are enriched in the main host, asa function of their enrichment redshift. We choose thegas peak velocity as this is close to the flow speed whenthe wind is launched, and is relatively unaffected byinteractions between the outflow and the ISM/gaseoushalo. Note that, because of the limited time resolution,we may actually underestimate the peak velocity ofgas particles that reach a maximum speed and thenslow down significantly within the time interval oftwo simulation outputs (∆ t ∼ >
140 Myr). The scatter plot shows that enriched gas particles can often reachvelocities in excess of 600 km s − , with a few of themmoving as fast as ∼ − − , a value that isconsistent with the highest velocity material observed inLBGs at 2 ∼ < z ∼ < − . From redshift 9.3to 3.0, the total mass of the main host halo increasesfrom 5 . × M ⊙ to 2 . × M ⊙ . The mean outflowspeed increases for 5 ∼ < z ∼ < . z ∼ <
5, so there is no obvious correlation between halomass and the peak outflow velocity in ErisMC. Similarly,we find only a weak correlation between the maximumwind velocity and star formation rate. However, ifwe define the mass averaged mean outflow velocity atdistance r as:2 Metals in circumgalactic medium Fig. 14.—
Evolution of the radial velocity of gas particles lastenriched at redshift 5 in the main host. Peculiar and Hubble flowvelocities are indicated with the black and red curves, respectively.The solid, dotted, and dashed lines show the median, 5 percentileand 95 percentile values, respectively.
Fig. 15.—
Peak radial velocity of gas particles that become un-bound after they are enriched in the main host, as a functionof their enrichment redshift. Each dot represents a gas particle.The black and blue solid lines show the mass-weighted and metal-weighted average values, respectively. The colored dashed linesshow the standard deviation from the mean. h v out i ( r ) = P Ni m i v r,i P Ni m i (7)where N is the total number of outflow gas particles in aradial shell of thickness dr =0.02 R vir at distance r , m i isthe mass of particle i and v r,i its outflow radial velocity(relative to the host’s center). There is a correlationbetween the mean outflow velocity and the peak circularvelocity of the host, as shown in Figure 16. Metallicity
Figure 17 shows the metallicity of inflowing and out-flowing material at ErisMC’s virial radius as a functionof redshift. Galactic outflows are enriched to a typi-cal metallicity in the range 0 . − . Z ⊙ since redshift ∼
9, with little dependence on cosmic time. The meanmetallicity of inflowing gas shows a more marked evo-lution, as it increases by about one dex in the interval
Fig. 16.—
Evolution of the mean outflow velocity of all unboundgas particles at different radii for the main host, as a function ofits peak circular velocity.
Fig. 17.—
Mean metallicity of inflowing ( dashed lines ) and out-flowing ( solid lines ) material as a function of redshift. The averageis taken over all gas particles within a thin shell of radius R vir andthickness 0 . R vir . ∼ > z ∼ > .
5. This is a consequence of more processedmaterial falling back onto ErisMC in a “halo fountain”(Oppenheimer et al. 2010) as well as being accreted viainfalling satellites. Only half of the inflowing material at R vir is unprocessed primordial gas. The gas metallicityof inflowing gas at z . Z ∼ . Z ⊙ ) is typical of themetallicity observed in Damped Ly α and Lyman-Limitsystems (e.g., Wolfe et al. 2005). Mass loading
The mass loading factor characterizes the amountof material involved in a galactic outflow, and isdefined as η = ˙ M w / SFR, where ˙ M w is the rateat which mass is ejected. Observations of galac-tic outflows powered by starbursts suggest a widerange of mass loading factors, η = 0 . − σ , η = σ /σ . In our SN-driven blastwavefeedback scheme there is no specific parameter for massloading. In this section we explore the mass loading fac-tor in ErisMC and its variation with star formation andhalo mass.We compute the wind mass loading at a distance r from the center as˙ M w ( r ) = 1∆ r N X i m i v r,i (8)where ∆ r is the thickness of the radial shell, N is thetotal number of outflow gas particles in the shell, m i isthe mass of particle i and v r,i its outflow radial velocity(relative to the host’s center). We have measured themass flux at 0.2, half, and one virial radius, and dividedit by the instantaneous SFR ignoring any time delay be-tween star formation and large-scale outflow, for ErisMCand the 9 most massive dwarf halos within 250 kpc at z = 3. We find (see Fig. 18) a strong correlation of themass-loading factor ( η ) and the mean outflow velocity( h v out i ) with halo mass. While η is of order unity forthe main host, it can exceed 10 (and reaches 80 in onecase) for nearby dwarfs and satellites. Similarly, h v out i increases from ∼
50 km/s for nearby dwarfs to ∼ α , C II ,C IV , Si II , and Si IV as a function of galactocentric im-pact parameter that are in good agreement with thoseobserved at high-redshift by Steidel et al. (2010). SUMMARY
We have presented a detailed study of the metal-enriched CGM of a massive galaxy at z = 3 using re-sults from a zoom-in hydrodynamic simulation of a diskgalaxy with mass comparable to the Milky Way. Ourapproach to understanding the role of inflows, star for-mation feedback, and outflows in governing the gaseousand metal content of galaxies and their environment, isdifferent to that of many recent theoretical efforts: Eris’extreme mass and spatial resolution allows us to followself-consistently the venting of metals by small progeni-tor dwarf satellites and the transport of heavy elementsfrom their production sites into the environment. Thereference run adopts a blastwave scheme for supernovafeedback that generates galactic outflows without explicitwind particles, and a star formation recipe based on ahigh gas density threshold.We have found that ErisMC’s metal-enriched CGM ex-tends as far as 4 virial radii (about 200 physical kpc) fromits center. Approximately 41%, 9%, and 50% of all gas-phase heavy elements within 250 kpc from the center arehot ( T > × K), warm (3 × K > T > × K), and cold (
T < × K), respectively. More than40% of all gas-phase metals lie outside the virial radius:
Fig. 18.—
Top panel:
Mass loading factor at z = 3 for the mainhost and the 9 most massive dwarfs within 250 kpc. The mass fluxwas measured at the virial radius ( black dots , half the virial radius( cyan triangles ) and one fifths of the virial radius ( red diamonds ).The empty square symbols indicate satellite systems of the mainhost. Bottom panel:
Same as the top panel, but for mass-weightedaverage outflow velocity h v out i . while cold metal-rich material traces large overdensitieswithin the main host, about 50% of all warm and 70%of all hot metals are found in low density δ <
30 re-gions beyond ErisMC’s virial radius. Intergalactic metalsare characterized by a strong temperature gradient withoverdensity, as the metal-weighted temperature climbsfrom 10 K at δ = 1 to above 2 × K at δ = 10.SN-driven winds are able to transport metals to regionsat δ ∼ <
10 where they accumulate, creating a peak inthe cold gas metallicity-density relation at
Z/Z ⊙ ≈ . Z/Z ⊙ = − δ ∼ < R vir , and fornone of those found beyond 3 R vir ; 2) its satellite progen-itors – systems accreted by the main halo before redshift3, which shed their metals before and during infall andare responsible for 28% of all the metals within 3 R vir ;and its orbiting nearby dwarfs, which give origin to 12%of all the metals within 3 R vir and 95% of those beyond3 R vir . Late ( z <
5) galactic “superwinds” – the result ofrecent star formation in ErisMC – account for only 9%of all the metals observed beyond 2 R vir , the rest having4 Metals in circumgalactic mediumbeen released at redshifts 5 ∼ < z ∼ <
8. These findings con-firm the ideas put forward by Porciani & Madau (2005),that enrichment from nearby dwarfs may contribute sig-nificantly to (and dominate at large distances) the pol-lution of the CGM around LBGs. Porciani & Madau(2005) argued that massive galaxies at high redshift arelikely to be surrounded by the metal bubbles produced bynearby dwarfs and satellites at earlier epochs, and thatgravity will tend to increase the spatial association be-tween such metal bubbles and LBGs. The strong associ-ation observed between stronger C IV systems and LBGsled Adelberger et al. (2003) to argue that metal-rich “su-perwinds” from LBGs may be responsible for distribut-ing the product of stellar nucleosynthesis on (comoving)Mpc scales. The analytical results of Porciani & Madau(2005) and the simulations presented here show that thisis not the case, as it is nearby dwarfs that dominate themetal enrichment beyond 2-3 R vir of LBGs. This is notto imply that most of the metals observed in the IGM athigh redshift have been ejected by dwarf galaxies: ratherthan, as a function of distance from a massive system,the contribution of the main host becomes sub-dominantcompared to that of its smaller companions. Simula-tions of larger cosmological volumes are needed to assesswhether our results are consistent with observations ofthe diffuse metal-enriched IGM.Substantial amounts of heavy elements are generatedat a larger distance from the main host’s center than their current location, and subsequently are accretedby the host along filaments via low-metallicity cold in-flows. Galactic outflows have velocities of a few hundredkm s − . The outflows decelerate rapidly, and the result-ing long metal transport timescales produce an age gra-dient in metals as a function of distance from the mainhost. The outflow mass-loading factor is of order unityin the main halo, but can exceed a value of 10 for nearbydwarfs.As a Lagrangian particle method, SPH does not in-clude any implicit diffusion of scalar quantities suchas metals. In the absence of some implementationof diffusion, metals are locked into specific particlesand their distribution may be artificially inhomogeneous(Aguirre et al. 2005; Wiersma et al. 2009). A numberof simulations of even higher resolution than ErisMCand including a scheme for turbulent mixing that re-distributes heavy elements and thermal energy betweenthe outflowing material and the ambient gas (Shen et al.2012) are in the making.Support for this work was provided by the NSFthrough grant AST-0908910 and OIA-1124453 (P.M.).Simulations were carried out on NASA’s Pleiades super-computer and the UCSC Pleiades cluster. One of us(P.M.) acknowledges support from a Raymond and Bev-erly Sackler Visiting Fellowship at the Institute of As-tronomy, Cambridge. REFERENCESAdelberger, K. L., Steidel, C. C., Shapley, A. E., & Pettini, M.2003, ApJ, 584, 45Adelberger, K. L., Shapley, A. E., Steidel, C. C., Pettini, M., Erb,D. K., & Reddy, N. A. 2005a, ApJ, 629, 636Adelberger, K. L., Steidel, C. C., Pettini, M., Shapley, A. E.,Reddy, N. A., & Erb, D. K. 2005, ApJ, 619, 697Aguirre, A., Hernquist, L., Schaye, J., Weinberg, D. H., Katz, N.,& Gardner, J. 2001, ApJ, 560, 599Aguirre, A., Schaye, J., Hernquist, L., Kay, S., Springel, V., &Theuns, T. 2005, ApJ, 620, L13Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009,ARA&A, 47, 481Behroozi, P. S., Conroy, C., & Wechsler, R. H. 2010, ApJ, 717,379Behroozi, P. S., Wechsler, R. H., & Conroy, C. 20120, ApJ,submitted (arXiv:1207.6105)Bouch´e, N., Lehnert, M. D., Aguirre, A., P´eroux, C., & Bergeron,J. 2007, MNRAS, 378, 525Cen, R., & Chisari, N. E. 2011, ApJ, 731, 11Cen, R., Nagamine, K., & Ostriker, J. P. 2005, ApJ, 635, 86Choi, J.-H., & Nagamine, K. 2011, MNRAS, 410, 2579Dalla Vecchia, C., & Schaye, J. 2008, MNRAS, 387, 1431Dekel, A., Birnboim, Y., Engel, G.;, Freundlich, J., Goerdt, T.,Mumcuoglu, M., Neistein, E., Pichon, C., Teyssier, R., &Zinger, E. 2009, Nature, 457, 451Dekel, A., & Silk, J. 1996, ApJ, 303, 39Gnedin, N. Y. 1998, MNRAS, 294, 407Governato, F., et al. 2010, Nature, 463, 203Guedes, J., Callegari, S., Madau, P., & Mayer, L. 2011, ApJ, 742,76Furlanetto, S. R., & Loeb, A. 2003, ApJ, 588, 18Haardt, F., & Madau, P. 1996, ApJ, 461, 20Heckman, T. M., Armus, L., & Miley, G. K. 1990, ApJS, 74, 833Jonsson, P. 2006, MNRAS, 372, 2Kennicutt, R. C., Jr., Tamblyn, P., & Congdon, C. E. 1994, ApJ,435, 22Kereˇs, D., Katz, N., Weinberg, D. H., & Dav´e, R. 2005, MNRAS,263, 2Knollmann, S. R., & Knebe, A. 2009, ApJS, 182, 608 Kroupa, P. 2001, MNRAS, 322, 231Kroupa, P., Tout, C. A., & Gilmore, G. 1993, MNRAS, 262, 545Mac Low, M.-M., & Ferrara, A. 1999, ApJ, 513, 142Madau, P., Ferrara, A., & Rees, M. J. 2001, ApJ, 555, 92Mannucci, F., et al. 2009, MNRAS, 398, 1915Martin, C. L. 2005, ApJ, 621, 227Mashchenko, S., Couchman, H. M. P., & Wadsley, J. 2006,Nature, 442, 539McKee, C. F., & Ostriker, J. P. 1977, ApJ, 218, 148Mori, M., Ferrara, A., & Madau, P. 2002, ApJ, 571, 40Moster, B. P., Naab, T., & White, S. D. M. 2012, MNRAS,submitted (arXiv:1205.5807)Ocvirk, P., Pichon, C., & Teyssier, R. 2008, MNRAS, 390, 1326Oppenheimer, B. D., & Dav´e, R. 2006, MNRAS, 373, 1265Oppenheimer, B. D., & Dav´e, R. 2008, MNRAS, 387, 577Oppenheimer, B. D., & Dav´e, R. 2009, MNRAS, 395, 1875Oppenheimer, B. D., Dav´e, R., Katz, N., Kollmeier, J., &Weinberg, D. H. 2011, MNRAS, submitted (arXiv:1106.1444)Oppenheimer, B. D., Dav´e, R., Kereˇs, D., Katz, N., Kollmeier, J.A., & Weinberg, D. H. 2010, MNRAS, 406, 2325Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ,124, 266Pettini, M. 2006, in The Fabulous Destiny of Galaxies: BridgingPast and Present (Paris: Frontier Group), eds. V. Le Brun, A.Mazure, S. Arnouts, & D. Burgarella, p. 319Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J.-G., Dickinson,M., Morwood, A. F. M., Adelberger, K. L., & Giavalisco, M.2001, ApJ, 569, 742Porciani, C., & Madau, P. 2005, ApJL, 625, L43Raiteri, C. M., Villata, M., & Navarro, J. F. 1996, A&A, 315, 105Rupke, D. S., Veilleux, S., & Sanders, D. B. 2005, ApJS, 160, 87Ryan-Weber, E. V., Pettini, M., Madau, P., & Zych, B. J. 2009,MNRAS, 395, 1476Scannapieco, E., & Oh, S. P. 2004, ApJ, 608, 62Scannapieco, E., Ferrara, A., & Madau, P. 2002, ApJ, 574, 590Schaye, J., Aguirre, A., Kim, T.-K., Theuns, T., Rauch, M., &Sargent, W. L. W. 2003, ApJ, 596, 768Schwartz, C. M., & Martin, C. L. 2004, ApJ, 610, 201 hen et al. 2011 15hen et al. 2011 15