It's not easy being green: The evolution of galaxy colour in the EAGLE simulation

3.7 MB·

by  James. W. Trayford

MNRAS000,1–??(2015) Preprint24June2016 CompiledusingMNRASLATEXstylefilev3.0
It’s not easy being green: The evolution of galaxy colour in
the EAGLE simulation
James W. Trayford1(cid:63), Tom Theuns1, Richard G. Bower1, Robert A. Crain2,
Claudia del P. Lagos3,4, Matthieu Schaller1, Joop Schaye5
1Institute for Computational Cosmology, Durham University, South Road, Durham, DH1 3LE.
2Astrophysics Research Institute, Liverpool John Moores University, 146 Brownlow Hill, Liverpool, L3 5RF.
3International Centre for Radio Astronomy Research, 7 Fairway, Crawley, 6009, Perth, WA, Australia.
6 4Australian Research Council Centre of Excellence for All-sky Astrophysics (CAASTRO), 44 Rosehill Street Redfern,
1
NSW 2016, Australia.
0 5Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, the Netherlands.
2

n
u Accepted.Received;inoriginalform
J

3
2 ABSTRACT
Weexaminetheevolutionofintrinsicu-rcoloursofgalaxiesintheeaglecosmological

] hydrodynamicalsimulations,whichhasbeenshowntoreproducetheobservedredshift
A
z =0.1 colour-magnitude distribution well, with a focus on z <2. The median u-r of
G
star-forming (‘blue cloud’) galaxies reddens by 1 magnitude from z = 2 to 0 at fixed
. stellar mass, as their specific star formation rates decrease with time. A red sequence
h
startstobuild-uparoundz =1,duetothequenchingoflow-masssatellitegalaxiesat
p
thefaintend,andduetothequenchingofmoremassivecentralgalaxiesbytheiractive
-
o galacticnuclei(AGN)atthebrightend.Thisleavesadearthofintermediate-massred
r sequence galaxies at z = 1, which is mostly filled in by z = 0. We quantify the time-
t
s scalesofcolourtransitionfindingthatmostgalaxiesspendlessthan2Gyrinthe‘green
a
valley’.Wefindthetimescaleoftransitiontobeindependentofquenchingmechanism,
[
i.e. whether a galaxy is a satellite or hosting an AGN. On examining the trajectories

2 of galaxies in a colour-stellar mass diagram, we identify three characteristic tracks
v that galaxies follow (quiescently star-forming, quenching and rejuvenating galaxies)
7 and quantify the fraction of galaxies that follow each track.
0
9 Key words: galaxies: colours, galaxies: evolution, galaxies: formation.
7
0
.
1
0
1 INTRODUCTION a firm statistical footing (Willett et al. 2013; Ha¨ußler et al.
6
2013). Blue colours are typically due to light from massive
1 A scatter plot of observed galaxies in optical colour versus
hot, young stars. Broadly speaking, the more massive blue
: broad-band magnitude (or stellar mass) reveals two rela-
v galaxiesarestar-formingdiscsandthefainteronesareirreg-
tivelywell-defineddistinctpopulations,a‘redsequence’and
i ulars.Theredgalaxies,incontrast,areellipticalorlenticular
X a ‘blue cloud’, in a volume-limited sample. While a narrow
types, with the red colour reflecting an old stellar popula-
r red sequence was evident in early datasets (e.g. Sandage &
tion. These galaxies are often referred to as ‘red-and-dead’
a Visvanathan 1978; Larson et al. 1980; Bower et al. 1991),
to stress that their star formation has mostly ceased (e.g.
this striking colour ‘bi-modality’ was perhaps first revealed
Brammer et al. 2009).
most clearly by Strateva et al. (2001), who exploited the
At higher redshifts, z ≥ 1, the blue cloud is clearly
step-change in sample size offered by the Sloan Digital Sky
Survey (sdss, York et al. 2000), the statistics and general dominant and its colour becomes increasingly blue with in-
creasingz.Selectioneffectsbiasagainstthedetectionofred
properties of these two sequences were subsequently char-
galaxies at higher z, but it is clear that the bright end of
acterised quantitatively by e.g. Baldry et al. (2004). The
the red sequence is already in place by z ∼1, albeit with a
u-r colour of a galaxy correlates strongly with its morphol-
ogy(Hubbletype);thegalaxyzoo1 citizenscienceproject small shift in colour (e.g. Wolf et al. 2003; Bell et al. 2004)
andtheMegaMORPHsurveyhaveputthiscorrelationon It is difficult to establish how these sequences arise, or
how individual galaxies evolve in colour space, using ob-
servations alone. This is because both star formation and
(cid:63) E-mail:[email protected](JWT) galaxy destruction by mergers change the number density
1 http://www.galaxyzoo.org/ of galaxies of given mass across time. Faber et al. (2007)
(cid:13)c 2015TheAuthors
2 J.W. Trayford, et al.
noted that the number density of blue galaxies remains ap- inclusters(e.g.Chungetal.2007;Fumagallietal.2014)and
proximately constant below redshift z ∼ 1 whereas that of of the ram-pressure stripped gas behind simulated galaxies
red galaxies increases markedly. This led them to propose that fall onto a cluster (e.g. Roediger & Bru¨ggen 2008).
a model in which oneor more mechanisms operate that de- Even though observations suggest two empirical mod-
crease the star formation rate of blue galaxies, with such els of quenching (i.e. AGN and environmental), models of
‘quenching’ making galaxies redder until they join the red galaxy formation have struggled to reproduce simultane-
sequence. Bell et al. (2012) showed that there is significant ously the detailed distribution of galaxies in the colour-
scatter in the properties of quenched galaxies. One correla- magnitudediagramandthedifferentclusteringpropertiesof
tionthatstoodoutintheirsample,isthatquenchedgalaxies red and blue galaxies. This is true of semi-analytical mod-
usuallyexhibitaprominentbulge,byassociationsuggesting els, which use phenomenological prescriptions to describe
a supermassive black hole. thephysicalprocessesthatleadtoquenching(e.g.Fontetal.
Amodelinwhichanaccretingsupermassiveblackhole 2008;Laceyetal.2015);forexample,Henriquesetal.(2015)
quenches star formation in its host galaxy appears very at- compare the Munich semi-analytical L-galaxies model to
tractive. This is because black hole mass increases rapidly sdssdata.Althoughinmanyaspectsthismodelreproduces
as a function of bulge mass (e.g. Ha¨ring & Rix 2004; Mc- theobservationsbetterthanitspredecessors,limitationsre-
Connell&Ma2013),hencesuchamodelmightexplainwhy main.Forexample,L-galaxies’u−r coloursareconsider-
most massive galaxies are red - as observed. Unfortunately ably more bimodal than observed.
the evidence that star formation in galaxies hosting X-ray Hydrodynamical simulations can, in principle, model
brightAGNisindeedsuppressedappearsinconclusive.Sev- manyphysicalprocessesself-consistently,butlackofnumer-
eral studies have found no correlation between star forma- ical resolution and other limitations of the hydrodynami-
tion rate and X-ray luminosity for an X-ray selected sam- cal integration may limit their realism. Fortunately, rela-
ple of AGN (e.g. Rosario et al. 2012; Harrison et al. 2012; tivelysmallchangestothebasichydrodynamicsscheme(e.g.
Stanley et al. 2015). However a close to linear correlation Price 2008; Hopkins 2013) seem to resolve most numerical
has been observed for galaxies selected in the infrared (e.g. issues,suchthatthedominantuncertaintiesinhydrodynam-
Delvecchio et al. 2015). The fact that the luminosity of an icalsimulationsbecomeassociatedtotheimplementationof
AGN likely varies on a range of time-scales (from hours to unresolved‘subgrid’processesratherthanthedetailsofthe
Myrs) might explain the apparent disparity (Hickox et al. hydrodynamics scheme (Scannapieco et al. 2012; Schaller
2014; Volonteri et al. 2015). Powerful radio galaxies associ- et al. 2015b).
ated with the centres of groups and clusters do appear to The huge dynamic range required to simulate a cos-
disrupt the inflow of cold gas (McNamara & Nulsen 2012). mologically representative volume with the required resolu-
Another well-documented process that quenches star tion to follow the hierarchical build-up of galaxies, presents
formationinagalaxyisrestrictionofitssupplyofgasbyei- a major challenge to numerical simulations. Until recently,
therram-pressurestrippingofdiscgas(Gunn&Gott1972, such simulations did not reproduce the galaxy stellar mass
e.g.)orremovalofhalogas(e.g.strangulationLarsonetal. function well, let alone the detailed colours/clustering of
1980),asthegalaxytraversesaregionofhighergaspressure galaxies. A red/blue bimodality appears in the zoomed-
associated with a group or cluster. The quenching of star- simulationsofCen(2014)eventhoughthesedonotinclude
formationturnsthesesatellitesred(e.g.Knobeletal.2013). AGN. However, the r−band luminosity function of these
Originally suggested by Gunn & Gott (1972), the efficiency simulation contains many more massive galaxies than ob-
of these mechanisms have been investigated using simula- served. Gabor & Dav´e (2012) include the effects of AGN
tions by many groups (e.g. Quilis et al. 2000; Roediger & using a heuristic prescription of heating gas, where cooling
Bru¨ggen2007),withmorerecentlyBah´eetal.(2013)point- is simply switched off in halos deemed massive enough to
ing out that galaxies may be stripped before they become host AGN. They illustrate how this process builds-up a red
satellites, by the gas in the outskirts of massive systems. sequencebelowredshiftz∼2;initiallylower-masssatellites
McCarthy et al. (2008) presented a theoretical framework andmoremassivequenchedcentralsappearinheatedhalos,
that improves upon the simple analysis by Gunn & Gott withacharacteristicdipintheabundanceofredgalaxiesof
(1972), and describes their simulation results well. stellarmassM ∼1010M thatismoreprominentathigher
(cid:63) (cid:12)
Observationalconfirmationthatenvironmentalquench- z. While this simulation may provide valuable insight into
ingindeedoperatesisevidencedbythefactthatredgalaxies thebuildupoftheredsequence,theheuristicnatureofthe
preferentially reside in regions of high galaxy number den- halo heating limits their practical applicability. For lower
sity (Dressler 1980), or equivalently that red galaxies are mass galaxies, Sales et al. (2015) show that the illustris
more strongly clustered than blue galaxies, even at fixed simulation(Vogelsbergeretal.2014)broadlyreproducesthe
mass (e.g. Zehavi et al. 2005), and that the clustering am- colours of satellites, which they attribute to the relatively
plitudeofredgalaxiesdependslittleonmass(incontrastto large gas fractions of satellites at infall.
thatofbluegalaxies,e.g.Coiletal.2008).Thisiscompati- The eagle reference model was calibrated to the z =
ble with a model where red galaxies reside close to, or even 0.1 stellar mass function, black hole masses and sizes of
inside,moremassiveandhencestronglyclusteredhalosthat galaxies and is currently the only hydrodynamical simula-
cause the quenching. Trends between the environment and tion that reproduces these observations. Eagle also repro-
gascontentofgalaxiesprovidefurtherevidence,withgalax- duces many independent galaxy observations, such as the
ies residing in clusters seen to be deficient in both HI and content and ionisation state of gas (Bah´e et al. 2016; Lagos
H gas relative to the field (e.g. Cortese et al. 2011; Boselli et al. 2015b), mass profiles (Schaller et al. 2015a) and evo-
2
et al. 2014). Particularly convincing is the similarity of the lutioninstellarmass,starformationrateandsize(Furlong
trailsofHIgasseentobeemanatingfromgasrichgalaxies et al. 2015b,a). The clustering of galaxies as a function of
MNRAS000,1–??(2015)
Colour evolution in EAGLE 3
colour is investigated in a companion paper (Artale et al., ISM. The free parameters that enter the modules for feed-
in prep.). Trayford et al. (2015) showed that eagle repro- back were calibrated using the redshift z =0.1 galaxy stel-
duces the g-r − M colour magnitude (and the g-r − M ) lar mass function, the z = 0.1 stellar mass-size relation,
r (cid:63)
relation from the gama spectroscopic survey (Driver et al. and the z = 0 stellar mass - black hole mass relation, see
2011) very well. Including a model for dust-reddening com- Crain et al. (2015) for motivation and details. We briefly
putedusingtheskirtradiativetransferscheme(Baesetal. summarise these subgrid modules here, paying particular
2005;Camps&Baes2015)improvesthequantitativeagree- attention to those aspects most crucial for this paper.
ment further (Trayford et al. in prep.). With low-redshift
(z∼0.1) galaxy colours in eagle appearing to be realistic, • Radiative cooling and photo-heating of gas by the
evolving optically-thin UV/X-ray background of Haardt &
studying how they have arisen given the physical feedback
Madau(2001)isimplementedelement-by-elementfollowing
model of the simulation may provide new insight. The evo-
lutionofeaglegalaxycoloursisalsoaffordedcredibilityby Wiersma et al. (2009a).
the reasonable evolution of the eagle galaxy population in • Star formation is implemented as a pressure-law
(Schaye & Dalla Vecchia 2008) so that simulated galaxies
terms of the stellar mass function (Furlong et al. 2015b).
In section 2 we describe the eagle simulations used in reproducetheobservedz=0relationbetweengasandstar
formationsurfacedensityofKennicutt(1998).Eachgaspar-
this study, particularly the aspects of star formation, metal
ticleisassignedastarformationrate,m˙ ,andgasparticles
enrichment and feedback that are most relevant for setting (cid:63)
areconvertedtostarparticlesstochastically.Thestarforma-
intrinsicgalaxycolours.Weinvestigateinsection3theevo-
tionrateiszeroforparticlesbelowthemetallicity-dependent
lution of the galaxy population across the colour-mass dia-
threshold of Schaye (2004). We resample young stars using
gram and correlate colour changes with galaxies becoming
a probability proportional to the estimates of m˙ to reduce
satellitesorhostinganAGN.Insection4weexpoundthese (cid:63)
sampling noise when estimating the galaxy luminosities, as
processes by analysing the behaviour of individual galax-
described in Trayford et al. (2015).
ies,usinggalaxymergertrees.Typicaltime-scalesassociated
• Feedback from star formation is implemented by heat-
withcolourtransitionarepresentedinsection4.2.Weshow
inggasparticlesneighbouringnewly-formedstarparticlesas
that the colour evolution of most galaxies can be described
described by Dalla Vecchia & Schaye (2012). In this purely
well in terms of three generic tracks and quantify the frac-
thermal implementation, a temperature boost ∆T is de-
tion of galaxies that follow each path. Finally, our findings SF
fined and the heating of neighbouring gas particles is sam-
aresummarisedinsection5.Throughoutthisworkwerefer
pledstochasticallygiventheenergyavailableforfeedback.A
todust-free,rest-framecoloursas‘intrinsic’colours,andwe
valueof∆T =107.5Kischosentobehighenoughtomit-
take Z = 0.0127 for the metallicity of the Sun (Allende SF
(cid:12)
igaterapidcoolingduetonumericaleffects,butlowenough
Prieto et al. 2001). Note that while the Z value affects
(cid:12)
to avoid poor sampling of heating events around individual
the normalisation of metallicities in solar units, colours are
star particles (Dalla Vecchia & Schaye 2012).
unaffected by the assumed Z (see Trayford et al. 2015).
(cid:12)
• Seeding, merging, accretion and feedback from super-
massive black holes is implemented as described in Schaye
et al. (2015). Briefly, dark matter halos with virial mass
> 1010h−1M are seeded with a black hole of mass
2 THE EAGLE SIMULATIONS (cid:12)
105h−1M . These can grow through Eddington-limited ac-
(cid:12)
The eagle suite (Schaye et al. 2015; Crain et al. 2015) in- cretion of gas while accounting for the gas angular momen-
cludessimulationsperformedinarangeofperiodicvolumes tum as by Rosas-Guevara et al. (2015), and through merg-
and at various numerical resolutions to enable convergence ers with other black holes, following Springel et al. (2005)
testing. The simulations were performed with the gadget- and Booth & Schaye (2009). Feedback from accreting black
3 tree-SPH code (Springel 2005), but with changes to the holes is also modelled by heating surrounding gas using
SPHandtime-steppingalgorithmcollectivelyreferredtoas an implementation similar to that of stellar feedback. The
anarchy (see appendix of Schaye et al. (2015) for details temperature boost for AGN heating events is chosen to be
and Schaller et al. (2015b) for the relatively minor impact T =108.5K for all the simulations considered here.
AGN
of these changes on the properties of simulated galaxies).
We use the ΛCDM cosmological parameters advocated by Each star particle represents a ‘simple stellar popula-
Planck Collaboration et al. (2014), and initial conditions tion’(SSP),characterisedbyanassumedstellarinitialmass
generated at z = 127 (Jenkins 2013) using second order function(IMF,eagleadaptstheChabrier(2003)IMFover
Lagrangian perturbation theory. We concentrate here on themassrange[0.1,100]M(cid:12)),andassumingthatstarshave
analysingthelargestreferencemodel(Ref-100).Thisisacu- ametallicityinheritedfromtheconvertedgasparticle,with
bic cosmological volume of 100 comoving Mpc (cMpc) on a a single age corresponding to the time that the gas parti-
side,withaninitialgasparticlemassofm =1.81×106M . cle was converted to a star particle. We then use published
g (cid:12)
ThesimulationhasaPlummerequivalentgravitationalsoft- stellarlife-times,evolutionarytracks,andyieldstocompute
ening of (cid:15) =0.7 proper kpc (pkpc) at redshift z=0. the rate at which these stars evolve and lose mass, as well
prop
astherateofcorecollapseandTypeIasupernovaeventsas
described in Wiersma et al. (2009b). The simulation tracks
11 elements (H, He, C, Ni, O, Ne, Mg, Si, S, Ca, Fe) as
2.1 Subgrid model and galaxy identification
well as a ‘total metallicity’ (metal mass fraction) variable
Theeaglereferencemodelimplementssubgridmodulesfor for each gas and star particle. The mass, age, and metal-
physicalprocessesthatoccurbelowtheresolutionlimit,cor- licity of the SSP are input parameters for the population
responding approximately to the Jeans length of the warm synthesis model described below.
MNRAS000,1–??(2015)
4 J.W. Trayford, et al.
• Dark matter halos are identified using the ‘friends- tion, leading to more clearly separated blue/red colour se-
of-friends’ algorithm (fof), linking dark matter particles quences.Indeed,theu(cid:63)-r(cid:63) indextraversesthe4000˚Abreak,
within 0.2 times the mean inter-particle separation into a oftenusedasaproxyforstarformationactivity(e.g.Kauff-
single fof halo. Other particles are assigned to the same mann et al. 2003). The photometry is presented here with-
halo(ifany)asthenearestdarkmatterparticle.Wecharac- out dust effects, comparison is possible with various obser-
terisethemassofthehalobyitsM value.Thisisthe vational data where dust corrections have been estimated
200,crit
mass enclosed within a sphere of radius R centred on (e.g. Schawinski et al. 2014).
200,crit
thelocationoftheparticlewithminimumgravitationalpo-
tentialinthehalo.Thisradiusischosensuchthatthemean
density within this sphere is 200 times the critical density,
3 COLOUR EVOLUTION OF THE ENSEMBLE
given the assumed cosmology.
GALAXY POPULATION
• Galaxies are identified with the subfind algorithm
(Springel et al. 2001; Dolag et al. 2009). subfind identi- Figure 1a shows that a scatter plot of eagle galaxies in a
fies self-bound substructures within halos which we asso- colour-stellar mass diagram, (u(cid:63)-r(cid:63)) vs M , exhibits strong
(cid:63)
ciatewithgalaxies.The‘central’galaxyisthegalaxyclosest bimodality in colour at redshift z≈0. The well defined red
to the centre of the parent fof halo; this is nearly always sequenceresidesatu(cid:63)-r(cid:63)(cid:38)2.2withcoloursbecomingredder
alsothemostmassivegalaxyinthathalo.Theothergalax- with increasing M . The blue cloud is at u(cid:63) −r(cid:63) ≈ 1.3,
(cid:63)
ies in the same halo are its satellites. Particles in a halo with a slope similar to that of the red sequence. These two
not associated with a bound substructure (i.e. satellites) sequences are indicated by red and blue lines to guide the
are assigned to the central galaxy. Central massive galaxies eye, respectively, obtained by a spline fit to the maxima in
(M ≥ 1011 M , say) then have an extended halo of stars theprobabilitydistributionofu(cid:63)-r(cid:63) inbinsofM .Wekeep
(cid:63) (cid:12) (cid:63)
around them, usually referred to as intra-group or intra- the location of these lines fixed in Fig. 1b-d to facilitate
cluster light. Determining the mass or indeed luminosity of comparison at higher z. We clearly see that:
suchalargegalaxyisambiguous,bothinsimulationsandin
(i) The red sequence becomes bluer and less populated
observations. For this reason we impose an aperture on the
with increasing z. It is in place at z ≈ 1 but has mostly
definitionofagalaxy:wefollowSchayeetal.(2015)andcal-
disappeared by z ≈ 2. A gap in the red sequence is notice-
culate masses and luminosities for every subhalo, excluding
able at z≈1 for M ∼109.7 M .
materialthatisoutsidea30pkpcsphericalaperturecentred (cid:63) (cid:12)
(ii) Thebluesequencebecomesbluerwithincreasingz,and
on the subhalo potential minima as well as material that is
exhibits decreasing scatter.
not bound to that subhalo. The 30 pkpc aperture has been
showntomimicanobservationalPetrosianaperture,andre- The main features of the galaxy population that
ducesintra-clusterlightinmassivecentralswhilelowermass drive these trends are illustrated in the bottom panels
galaxies are unaffected (Schaye et al. 2015). of the figure. Figs. 1e & g show that for a narrow stel-
lar mass range around M = 1010.25 M that u(cid:63)-r(cid:63) is
(cid:63) (cid:12)
strongly anti-correlated with the specific star formation
2.2 Galaxy colours
rate, sSFR ≡ M˙ /M , provided the galaxy is star-forming
(cid:63) (cid:63)
The stellar population properties (age, metallicity & as- (log (sSFR/Gyr) (cid:38) −2). This is not surprising since the
10
sumed IMF) of an eagle galaxy are combined with the light in the u(cid:63) filter is dominated by emission from mas-
Bruzual & Charlot (2003) population synthesis model to sive and hence young stars, while r(cid:63) is dominated by the
construct an SED for each star particle. Summing spectra older population. Galaxies in this plot follow a very tight
overallstarswithintheaperturedescribedinsection2.1and relation at a given z, sliding along a narrow locus in colour
convolvingwithafilterresponsefunctionyieldsbroad-band that becomes bluer at higher z. At z = 2, the galaxies fol-
colours,whichwecomputeusingtheugrizYJHKphotomet- low almost the same relation in u(cid:63)-r(cid:63) versus sSFR as at
ric system for optical and near infrared photometry (taken z = 0, with just a small but noticeable offset towards red-
from Doi et al. 2010; Hewett et al. 2006). We express these dercolours(≈0.1mag)forlog (sSFR/Gyr)(cid:38)−1.25.This
10
absolute magnitudes in the AB-system, see Trayford et al. is a result of the redder population being younger on av-
(2015) for more details. erage and hence brighter for a z = 2 star-forming galaxy,
It is well known that dust can alter the optical colour compared to a star-forming galaxy at z =0. For lower star
of a galaxy significantly, particularly for gas-rich discs seen formation rates (log (sSFR/Gyr) (cid:46) −1.25) the old popu-
10
edge-on. We describe a simple model for dust reddening in lationhasmoreinfluenceonu(cid:63)-r(cid:63),andtheyoungeraverage
a previous study (Trayford et al. 2015), as well as a model stellarageofz=2galaxiescausesanoffsettobluecolours.
that uses ray-tracing to account for the patchy nature of Withu(cid:63)-r(cid:63)coloursostronglycorrelatedwithsSFR,the
dustcloudsenshroudingstar-formingregionsdescribedina colour vs M diagram of Fig. 1 is one view of the ‘fun-
(cid:63)
forthcomingstudy(Trayfordetal.,in prep.).However,here damental plane of star-forming galaxies’, discussed recently
we use the ‘intrinsic’ (i.e. rest-frame and dust-free) colours by Lagos et al. (2015a). These authors showed that eagle
of galaxies to examine the changes arising purely from the galaxiesfromdifferentredshiftsfallontoasingle2Dsurface
evolutionoftheirstellarcontent.Tosimplifytheinterpreta- whenplottedinthe3DspaceofM˙ −M andgasfraction(or
(cid:63) (cid:63)
tionwealwaysquoterest-framecolours:thereisthereforeno metallicity),whichtheyattributedtoself-regulationofstar
‘k’-correction needed to compare galaxies in the same band formation. Lagos et al. (2015a) also showed that observed
at different redshifts. We concentrate here on u(cid:63)-r(cid:63) colours galaxies follow very similar trends. The increasingly bluer
(with the (cid:63) referring to intrinsic colours) rather than g(cid:63)-r(cid:63), colours of the blue cloud towards higher z is a consequence
because the u band is more sensitive to recent star forma- of the increased star formation activity at fixed M .
(cid:63)
MNRAS000,1–??(2015)
Colour evolution in EAGLE 5
Figure 1.Colourevolutionofeaglegalaxies.Top row:u(cid:63)-r(cid:63) vsM(cid:63) colour-massdiagramatfourredshifts(z=0.1,0.5,1and2,left
toright).Individualgalaxiesareplottedaspoints,colouredbymedianstellarmetallicity,usingthecolourbarinthebottomrow.The
locations of the red sequence and blue cloud at z = 0.1 (red and blue lines, respectively) are repeated in panels b-d to guide the eye.
Filledredsquaresshowu(cid:63)-r(cid:63)versusM(cid:63) fora10GyroldstellarpopulationwithmetallicityZ(cid:63) equaltothemedianmetallicityatthat
M(cid:63); filled circles are the same, but assuming an exponential distribution of stellar metallicities with the same median. Bottom row,
panelseandg:dependenceofu(cid:63)-r(cid:63)colouronspecificstarformationrate(sSFR,M˙(cid:63)/M(cid:63))forgalaxieswith10<log10(M(cid:63)/M(cid:12))<10.5
(thegreybandinpanela,andgalaxieswithM(cid:63) betweenthetwogreylinesinpaneld)atredshiftz=0.1andz=2,respectively.The
olive line indicates the median u(cid:63)-r(cid:63) as a function of sSFR at z = 0.1 for comparison at z = 2. Panel f: u(cid:63)-r(cid:63) versus median stellar
metallicity for the galaxies of panel e; galaxies with sSFR<10−3.5 Gyr−1, appearing in the green box in panel e, are plotted as green
dots.
The scatter in colour at fixed M on the red sequence quencehasthesameslopeastheredsequenceineagle,itis
(cid:63)
ismostlyduetometallicity,Z,asisclearfromexamination systematically redder by ≈0.25 magnitudes. This is not an
of the u(cid:63)-r(cid:63) distribution of galaxies at a given M with low age effect, but a consequence of stellar populations exhibit-
(cid:63)
sSFR<10−3 Gyr−2,plottedasgreenpointsinFig.1f.The ing a spread in metallicity within an eagle galaxy. In fact,
colour of star-forming galaxies with sSFR > 10−3 Gyr−1 the metallicity distribution function of stars in an eagle
(black points) also depends on Z, but from comparison of galaxy is fairly well described by an exponential distribu-
these panels it is clear that this effect is much smaller than tion.Wethereforegeneratedanothercomparisontoymodel
thedependenceofcolouronsSFRitself-itinducesthesmall for the red sequence colour, in which we impose an expo-
scatter in u(cid:63)-r(cid:63) in panel e, on top of the main trend with nential metallicity distribution and again assume a coeval
sSFR. 10 Gyr old population. The exponential metallicity distri-
bution is defined by a mean value at fixed mass, given by
As discussed by many others, the slope of the red se- theZ (M )dependenceofeaglegalaxies.Thismodelis
quence demonstrates the dependence of colour on Z for plottemdedasfil(cid:63)ledredcirclesanditreproducestheeaglered
galaxies: more massive galaxies are more metal rich and
sequence very well. This simple exercise shows that the as-
hence redder (see Trayford et al. 2015). Because the mass-
sumption that all stars have the same metallicity results in
and light-weighted metallicities are not equivalent, the in-
systematicerrorsinthemetallicityfrombroad-bandcolours.
ternal metallicity distribution for stellar populations in a
galaxy may also affect the normalisation of the red se- The consistent red sequence slope between the toy
quence. To illustrate this, we calculated the median metal- model and eagle suggests that any changes in the inter-
licity,Z (M ),inbinsofstellarmass.Wethencalculated nal stellar Z distribution of eagle galaxies with mass are
med (cid:63)
u(cid:63)-r(cid:63) colours for a 10 Gyr old population with that depen- not strong enough to bias the median colours of red galax-
dence of Z on M , and plot the resulting u(cid:63)-r(cid:63) colour as a ies.Wenotethattheslopeofu(cid:63)−r(cid:63) relationasafunction
(cid:63)
function of M in Fig. 1a as red squares. Although this se- of M in the blue cloud is set by the sSFR-M relation and
(cid:63) (cid:63) (cid:63)
MNRAS000,1–??(2015)
6 J.W. Trayford, et al.
Figure 2. The impact of satellite fraction on the evolution of the u(cid:63)-r(cid:63)vs M(cid:63) colour-stellar mass relation. Top panels: Each square
correspondstoabinincolourandM(cid:63) andiscolouredaccordingtothemediannormalisedsatellitefractioninthatbin,suchthathigher
satellitefractionscorrespondtoreddercolours(seethecolourbar).Thesatellitefractionisnormalisedtotheaveragesatellitefraction
at that stellar mass (bottom panel), removing trends of satellite fraction with stellar mass and redshift. At z = 2, most red galaxies
withM(cid:63)≤1010M(cid:12) aresatellites(redcolourinsatellitefraction).Thistrendpersiststoz=0,althoughitbecomesweakerasgalaxies
classified as centrals also get quenched. Bottom panels: Fraction of galaxies classified as satellites as a function of M(cid:63). At redshift
z=0,thesatellitefractionisnearlyconstantatjustbelow50percentbelowM(cid:63)=1010M(cid:12),anddecreasesabovethatmass.Athigher
z thesatellitefractiondecreasesforallM(cid:63).
not by metallicity effects. Therefore, the similarity between deed, they may not be part of the fof halo (yet). Another
the slopes of the blue and red lines is coincidental. possibility is that some of these galaxies were stripped as
satelliteswhentheyfellinsideamassivehalobuthavetrav-
elledoutagain,theso-calledbacksplashpopulation(Balogh
et al. 2000).
3.1 Satellite colours
At redshift z ≈ 0, the fraction of satellites is ≈ 50 per
The extent to which satellite galaxies are preferentially red cent at M ∼ 109M , decreasing slowly to 30 per cent by
(cid:63) (cid:12)
relativetothegeneralpopulationisillustratedinFig.2.We M ∼1010.5M , and then dropping rapidly towards higher
(cid:63) (cid:12)
divide colour vs mass diagrams at different redshifts into M . Satellite fractions decrease slowly at all M with in-
(cid:63) (cid:63)
equalbinsofu(cid:63)-r(cid:63)andlog10(M(cid:63)/M(cid:12)).Thesatellitefraction creasing z to z ≈ 1, and then drop much faster to below
ineachbiniscomputedandnormalisedbythetotalsatellite 30 per cent at all masses by z = 2. This rapid drop in the
fractionforallgalaxiesinthesamestellarmassrange.Bins satellitefractionwithincreasingz isthereasonthatthered
containing>10galaxiesareshadedbythelog normalised sequence disappears at low M (cid:47)1010M for z(cid:38)2.
10 (cid:63) (cid:12)
satellitefraction,suchthatpositivevaluesindicateahigher
than average satellite fraction for that mass, while negative
valuesindicatealowerthanaveragevalue.Thesatellitefrac-
3.2 AGN host colours
tionasafunctionofstellarmassineagleisplottedforeach
redshift in the bottom panels. The effect of feedback from accreting black holes on galaxy
Galaxies with M ≤1010M that are red are predom- coloursisillustratedinFig.3.Weonlyplotcentralgalaxies
(cid:63) (cid:12)
inately satellites, seen most strikingly at z ≈ 2. At lower z to disentangle satellite quenching from effects induced by
thereisstillatrendforlow-massredgalaxiestobesatellites, AGN. The figure is analogous to Fig. 2, with median black
butthetrendislesspronouncedbecausesomegalaxiesclas- holemassreplacingsatellitefraction.Themedianblackhole
sified as centrals are also red. To some extent this may be mass (M ) as a function of M is plotted for each redshift
• (cid:63)
a consequence of galaxies being quenched by ram-pressure as the bottom panels in Fig. 3.
strippingintheoutskirtsofmoremassivehalos,beforethey At redshift z = 2, there is a very strong trend for red
are classified as being a satellite (e.g. Bah´e et al. 2013). In- galaxieswithM ≈1010M toexhibitunusuallyhighblack
(cid:63) (cid:12)
MNRAS000,1–??(2015)
Colour evolution in EAGLE 7
Figure 3. The impact of black hole mass on the evolution of the u(cid:63)-r(cid:63) vs M(cid:63) colour-stellar mass relation for central galaxies. Top
panels:Eachsquarecorrespondstoabininu(cid:63)-r(cid:63) andM(cid:63),andiscolouredaccordingtothemedianblackholemass,M•,inthatbin,
suchthatlargervaluesofM•correspondtoreddercolours(seethecolourbar).Themedianblackholemassineachsquareisnormalised
tothemedianblackholeatthatstellarmass(bottompanel),removingtrendsofM• withM(cid:63) andredshift.Atz=2thereisatrendfor
redder galaxies to have more massive black holes. This trend is particularly striking for galaxies with M(cid:63) ∼1010M(cid:12) and becomes less
pronouncedathighermasses.Thereisnoobviouscorrelationatlowerstellarmass.Thesetrendspersisttolowerz butbecomeweaker.
Lower panels:Themedianblackholemass,M•,asafunctionofstellarmassisplottedinsolidblack.Dashedblacklinesrepresentthe
16thand84thpercentiles.M• isnearlyindependentofM(cid:63) belowM(cid:63)∼1010M(cid:12),andincreaseswithM(cid:63) abovethischaracteristicmass.
Thistrendisalmostindependentofredshift.
holemasses,morethan4timesthemedianblackholemass 3.3 Colour transformation mechanisms
atthatstellarmass.Thistrendpersistsbutbecomesweaker
athigherM ,andiscompletelyabsentatlowermasses.This CombiningtheresultsofFigs.2and 3enablesustounder-
(cid:63)
correlation between the residuals of the M•−M(cid:63) relation standtheoriginoftheevolutionintheu(cid:63)-r(cid:63) vsM(cid:63) diagram
andthecolourofthegalaxyisstillmostlypresentatz=1, of Fig. 1: galaxies with M(cid:63) <1010 M(cid:12) tend to become red
but begins to be washed out at later times. when they become satellites, whereas galaxies above this
characteristic mass are quenched by their AGN. This rea-
soningalsoexplainswhytheredsequencestartstobuild-up
fromboththelow-massandthehigh-massends,leavingini-
tiallyanoticeablescarcityofredgalaxiesatM ≈109.7M
(cid:63) (cid:12)
at z ≈ 1. Such galaxies are too low mass to host a vigor-
The trend for galaxies with high black hole masses to ously accreting black hole, yet too massive to be satellites
be predominantly red when M (cid:39) 1010 M is likely re- inthetypicallylower-massgroupsatthathigherz.Itisnot
(cid:63) (cid:12)
lated to the largely redshift independent characteristic halo until redshifts z <1 that the more massive halos that host
mass, Mh ∼ 1012 M(cid:12), above which black holes start to M(cid:63) ≈1010 M(cid:12) satellites appear.
growrapidlyineagle,fedbythegrowinghothalosaround The extent to which eagle predicts the characteris-
them (Bower et al. in prep.). The accreting black hole then tic stellar mass above which AGN quenching occurs, and
quenches star formation in its host galaxy, turning it red. the evolution of the under-abundance of intermediate-mass
Acorollaryoftheexistenceofthischaracteristichalomass, red galaxies, not only depends on the details of the sub-
isthatblackholesonlystarttogrowsignificantlywhenthe grid physics but also on the volume that is simulated. This
galaxy’s stellar mass is ∼1010 M (Fig. 3, bottom panels). is because massive clusters are under-represented or simply
(cid:12)
The scatter in the M −M relation results in the transi- absentduetomissinglarge-scalepowerinthedensityfield,
(cid:63) h
tionbetweendormantandrapidlygrowingblackholesbeing and poor sampling of rare objects in the relatively small,
less well-defined in the M −M relation in comparison to periodic eagle volume of 1003 cMpc3. However we believe
• (cid:63)
a M −M plot. that the relevant physics described here is robust, and cor-
• h
MNRAS000,1–??(2015)
8 J.W. Trayford, et al.
roborates the similar conclusions of Gabor & Dav´e (2012) factthatahigherproportionofsatellitesarefoundatthese
who used an ad-hoc model for quenching in massive galax- coloursthanatthecentreofthebluepeak(seeFig.2).Note
ies, as opposed to our physically motivated subgrid scheme that this effect is only observed for the high redshift panel,
for the eagle simulations that are implemented on smaller partiallyduetothebluestbinsfailingtomeettheminimum
(sub-kpc) scales. galaxy count criterion of 10. These extreme starbursts are
A corollary of satellitequenching for lower-mass galax- clearly rarer at low redshift.
ies, and AGN quenching for more massive galaxies, is that For each redshift range, we see that galaxies in more
these low- and high-mass red galaxies tend to inhabit the massivehalos(orangevectors)haveastrongermedianshift
same dark matter halos. The more massive red galaxy is in u(cid:63)-r(cid:63) than their low-mass halo counterparts (purple
thecentralgalaxyofthishaloandisquenchedbyitsAGN. vectors), suggesting that they have a higher likelihood of
Conversely, the lower-mass red galaxies are the satellites of quenching.Galaxies in massive halos also generally exhibit
a massive central red galaxy. As a consequence, the low- more mass loss than the overall population, showing the
andhigh-massredgalaxieshavesimilarclusteringstrengths, role of environment in how galaxies evolve in the (u(cid:63)-r(cid:63),
with both clustering more strongly than blue galaxies. ea- M ) plane.
(cid:63)
gle reproduces the observed clustering as a function of
colour and luminosity well, as will be discussed by Artale
etal.(inprep.).Wenextinvestigatehow,andatwhatrate, 4.2 Evolution of colour populations in eagle
individual galaxies move through the u(cid:63)-r(cid:63) vs M diagram.
(cid:63)
To track the evolution of galaxies selected to be red, blue
or green, we must first define these populations. To do this
we apply cuts that evolve with redshift z for red and blue
galaxies.
4 COLOUR EVOLUTION OF INDIVIDUAL
GALAXIES (u(cid:63)−r(cid:63)) >0.2log (M /M )−0.25z0.6+0.24
red 10 (cid:63) (cid:12)
4.1 The flow of galaxies in the colour-M(cid:63) plane (u(cid:63)−r(cid:63))blue <0.2log10(M(cid:63)/M(cid:12))−0.25z0.6−0.3(1)
Fig.4illustrateshowgalaxiesmovethroughthe(u(cid:63)-r(cid:63),M ) The green galaxies are taken to be those that are not in-
(cid:63)
plane.Selectinggalaxiesinequalbinsofu(cid:63)-r(cid:63)andlog (M ) cluded in either set. These cuts are defined in an ad-hoc
10 (cid:63)
at one redshift, we measure the median difference in u(cid:63)- way to divide the galaxies into three populations at each
r(cid:63) and log (M ) for their descendant galaxies at a second redshift. This is a similar procedure to that used by many
10 (cid:63)
redshift.Weplotthesedifferencesforgalaxiesoveranequal observational studies, with authors adopting differing func-
time period at high (z ≈ 1) and low (z ≈ 0) redshift. This tional forms and normalisations (see e.g. the discussion in
is achieved by using two consecutive snapshots (z = 1.3 Taylor et al. 2015). The exact form of the colour cuts is
and z = 1) for the right panel, corresponding to a time unimportant for our qualitative analysis, but is considered
interval of ≈ 0.9 Gyr, and interpolating the galaxy vectors when we discuss our quantitative results.
between the lowest redshift snapshots (z = 0.1 and z = 0) The evolution of the u(cid:63)-r(cid:63) colours of galaxies, selected
to match the same time period. Descendants that grow in by colour either at high redshift (z = 0.5) or low redshift
M by a factor > 4 through merging into a more massive (z=0.1),isillustratedinFig.5.Webingalaxiesinthethree
(cid:63)
hostgalaxyareeliminatedfromthemeasurement,toprevent colourbinsdescribedaboveandplotthecolourdistribution
them contributing extreme vectors to their bin. ofthedescendantsandmainprogenitors(oddandevenrows,
For both redshift ranges, it is clear that the colours respectively)ofgalaxiesselectedtobered,greenorblue(top
of galaxies generally become redder, with vectors pointing two,middletwo,andbottomtworows,respectively).Weuse
in the positive u(cid:63)-r(cid:63) direction. Exceptions can be seen on the galaxy merger trees to identify descendants and main
the red sequence, in which some red galaxies become star- progenitors.Foreachpanel,theu(cid:63)-r(cid:63) colourdistributionof
formingfollowingagas-richmerger-wediscussthefraction allgalaxiesattheindicatedredshift,withM >1010M ,is
(cid:63) (cid:12)
of such ‘rejuvenated’ galaxies below. Red sequence galax- plotted in grey.
ies show little change in u(cid:63)-r(cid:63), but in general those with Fromthetoptworowsitbecomesclearthatmostgalax-
M (cid:46)1010.75M lose mass, with only the most massive red iesthatareredatz=0.5stayredtoz=0,whereasasub-
(cid:63) (cid:12)
sequencegalaxiesshowingmassgrowth.Consideringthatwe stantialfractionofgalaxiesthatareredatz=0weregreen
donotcountmergersintohostsoffactor>4higherM ,this atz=0.1orevenblueatz=0.5.Galaxiesthataregreenat
(cid:63)
suggeststhatredsequencegalaxiesarebeingstrippedprior z =0.5 predominantly become red at z =0, but a fraction
toadrymergerwithamassivecentral.Wealsoseeevidence ofgreengalaxiesbecomesblue(thirdrow).Galaxiesthatare
of‘massquenching’,withthevectorsforblue-cloudselected green at z = 0 had a range of colours at z = 0.5, although
galaxies becoming steeper with increasing stellar mass. For they were bluer than average (fourth row). The similar and
M > 1010M this is attributable to the presence of AGN dominant blue (red) fractions in the two green progenitor
(cid:63) (cid:12)
(Fig. 3). (descendant) panels suggests that the typical time to tran-
Another notable behaviour seen in Fig. 4 is that the sition through the green valley is shorter than the redshift
bluest galaxies (u(cid:63)-r(cid:63) ≈ 0.5, right panel) tend to change intervalsusedhere.Finally,galaxiesthatareblueatz=0.5
their u(cid:63)-r(cid:63)colour more than the average blue galaxy, to the have a large range of colours at z = 0 with a distribution
extentthattheyend-upontheredsideofthebluesequence that is similar to that of the population as a whole (fifth
atthelaterredshift.Thestrongreddeningandmassincrease row), whereas galaxies that are blue at z = 0 were mostly
ofthesegalaxiessuggeststhattheyarestarburststriggered blue at z=0.5 as well (bottom row).
justpriortoamerger.Suchascenarioisconsistentwiththe The rate at which galaxies with stellar mass M >
(cid:63)
MNRAS000,1–??(2015)
Colour evolution in EAGLE 9
3.0
z=0.0 z=1.0
2.5
2.0
(cid:63)
r
−1.5
(cid:63)
u
1.0
All
0.5 M <1013 M
200,crit
(cid:12)
M >1013 M
200,crit
(cid:12)
0.0
9.5 10.0 10.5 11.0 9.5 10.0 10.5 11.0
log (M /M ) log (M /M )
10 (cid:63) 10 (cid:63)
(cid:12) (cid:12)
Figure4.Theflowofgalaxiesin(u(cid:63)-r(cid:63),M(cid:63))spacebetweentworedshifts,z1 toz2.Rightpanel showsthegalaxyflowbetweenz1=1.3
andz2=1snapshots,aperiodof≈0.9Gyr.Theleftpanel showsthegalaxyflowinterpolatedbetweenthez=0.1andz=0snapshots
toyieldthesametimeperiod,withz1=0.07andz2=0.Blackcirclesrepresentthemeanlocationofgalaxiesatz1,selectedinabinof
u(cid:63)-r(cid:63)-M(cid:63);thesizeofthecircleisproportionaltothelogarithmofthetotalstellarmassingalaxiesinthatbin.Blackvectorsrepresent
the mean motion of the galaxies in that bin between z1 and z2. Orange vectors (purple vectors) are for those galaxies that at redshift
z2 belongtohaloswithvirialmassM200,crit>1013M(cid:12) (M200,crit<1013M(cid:12)).Centrecoordinatesandvectorssamplingfewerthan10
galaxies are not plotted. The overall distribution at the later redshift is plotted as grey contours for comparison. We do not take into
accountgalaxiesmergingintohoststhataremorethanfourtimestheirmass,illustratingsuchmergersinmoredetailinFig.8.
1010M at z = 0.5 change u(cid:63)-r(cid:63) colour over the redshift
(cid:12) Table 1. Properties of the galaxies plotted as main tracks in
range z =0.5 to z =0 (elapsed time ∆t≈5 Gyr) is quan- Figs.7-8.TheSymbol/Figureisgiventoidentifythegalaxieson
tified in Fig. 6. We identify the z = 0 descendant for all the figures. For each galaxy we quote the unique galaxy identi-
galaxieswithM >1010M atz=0.5,computethechange fier(GalaxyID)takenfromtheeaglepublicdatabase(McAlpine
(cid:63) (cid:12)
in colour, ∆(u(cid:63)-r(cid:63)), and plot a histogram of rates, ∆(u(cid:63)- etal.2015),thez=0blackholemass(M•),andindicatewhether
r(cid:63))/∆t.Wealsoidentifyifagalaxyisasatelliteatz=0.5, agalaxywaseverclassifiedasasatellite(y)ornot(n).
or if the mass of its central black hole increases by a factor
≥ 1.5. This threshold is chosen to represent an above av- Sym./Fig. GalaxyID M•/M(cid:12) Satellite
erage black hole growth, while still providing a significant
Circle/7 18169630 7.09×106 n
sample of galaxies. Square/7 15829793 1.03×108 n
The rate of change of the median colour of galaxies Triangle/7 14096270 6.00×107 n
is small, ∆(u(cid:63)-r(cid:63))/∆t ≈ 0.08 mag Gyr−1 to the red, but Circle/8 15197399 1.84×108 y
is larger for galaxies whose black hole grows more than
average (∆(u(cid:63)-r(cid:63))/∆t ≈ 0.09 mag Gyr−1) or those that
are satellites (∆(u(cid:63)-r(cid:63))/∆t ≈ 0.12 mag Gyr−1). Galaxies
that are red at z = 0.5 typically change little in colour to generictracksofcentralgalaxiesthatweillustrateinFig.7.
z = 0, (∆(u(cid:63)-r(cid:63))/∆t ≈ 0.03 mag Gyr−1), except for the In Fig. 8 we also show the track of a central galaxy that is
occasional outlier that becomes blue. The rate of change of very massive at z = 0 (M ∼ 1011M ), to illustrate indi-
(cid:63) (cid:12)
the median colour is larger for galaxies that are green or vidual tracks of satellites that merge with it. More details
blue at z = 0.5, with individual galaxies changing colour of the four galaxies tracked in these panels are given in Ta-
more rapidly, both to the red and to the blue. Galaxies ble 1. The time-scale over which galaxies transition to the
that are satellites can undergo rapid changes to the red, red sequence is compared to that of a passively evolving
∆(u(cid:63)-r(cid:63))/∆t (cid:38) 0.2 mag Gyr−1, whether blue or green at population, by plotting (single) star-burst tracks initiated
z =0.5. Note that this rate is averaged over a considerable at different times (grey curves in middle panels).
period (≈ 5 Gyr), and instantaneous rates of colour change The blue track in Fig. 7 is for a galaxy that remains
for galaxies can be much higher, as explored below. in the blue cloud down to z = 0, forming stars in a blue
disc that grows in time. As its sSFR decreases with time,
andthecontributionofanolderpopulationofstarsbecomes
4.3 Colour-mass tracks of individual galaxies
moreimportant,itslowlyreddens,with∆(u(cid:63)-r(cid:63))≈0.2from
We have examined a large number of tracks of individ- z=1toz=0(elapsedtime≈8Gyr).WeseefromTable1
ual galaxies in (u(cid:63)-r(cid:63), M ) space and have identified three and Fig. 3 that the black hole mass is ≈0.6 dex lower than
(cid:63)
MNRAS000,1–??(2015)
10 J.W. Trayford, et al.
0.20 z=0.5 RedSelection z=0.3 DescendantsofRedSelection z=0.0 DescendantsofRedSelection
q.
e
r0.15
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.20 z=0.5 ProgenitorsofRedSelection z=0.3 ProgenitorsofRedSelection z=0.0 RedSelection
q.
e
r0.15
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.20 z0.=50.5 1.5 G2r.e5enSelection z0.=50.3 De1sc.e5ndantsofG2r.e5enSelection z0.=50.0 De1sc.e5ndantsofG2r.e5enSelection
req.0.15 u∗−r∗ u∗−r∗ u∗−r∗
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.20 z=0.5 ProgenitorsofGreenSelection z=0.3 ProgenitorsofGreenSelection z=0.0 GreenSelection
q.
e
r0.15
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.20 z0.=50.5 1.5 2B.l5ueSelection z0.=50.3 D1e.s5cendantsof2B.l5ueSelection z0.=50.0 D1e.s5cendantsof2B.l5ueSelection
req.0.15 u∗−r∗ u∗−r∗ u∗−r∗
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.20 z=0.5 ProgenitorsofBlueSelection z=0.3 ProgenitorsofBlueSelection z=0.0 BlueSelection
q.
e
r0.15
f
d
e
alis0.10
m
or0.05
N
= =
⇒ ⇒
0.00
0.5 1.5 2.5 0.5 1.5 2.5 0.5 1.5 2.5
u r u r u r
∗ ∗ ∗ ∗ ∗ ∗
− − −
Figure5.Greyhistograms:u(cid:63)-r(cid:63)colourevolutionofallgalaxiesfromredshiftz=0.5toz=0,withredshiftdecreasingfromlefttoright
in each row; each panel is labelled with the corresponding redshift. Only galaxies with M(cid:63) > 1010M(cid:12) are included. Colour selections
are made using Eq. 1 and are as follows: top row: we select red galaxies at z =0.5 (red histogram) and plot the colour distribution of
theirdescendantsatlowz asaredhistogram.Secondrowfromtop:weselectredgalaxiesatz=0(redhistogram),andplotthecolour
distribution of their main progenitors as a red histogram at higher z. Rows 3 and four from the top: as above, but for green galaxies.
Bottom two rows: as above, but for blue galaxies. The background colour of the panel in which galaxies were selected is coloured grey
foreaseofreference.
the median value for its stellar mass (at approximately the compact elliptical galaxy since at least z = 1, maintaining
3rdpercentileofgalaxiesforthatM ),suggestinglowlevels a stellar half-mass radius of < 2.3 pkpc. The similar rate
(cid:63)
of black hole feedback in the galaxy’s history. of colour transition for this galaxy to that of an instanta-
neous starburst suggests rapid quenching of star formation.
The red track in Fig. 7 corresponds to a galaxy that
ConsideringtheblackholemassinTable1andthebottom-
becomes red more rapidly, reddening by ∆(u(cid:63)-r(cid:63)) ≈ 1 in
left panel Fig. 3 shows that the galaxy has a central black
≈ 2 Gyr, joining the red sequence at z = 2. From then on,
holemass≈0.6dexhigherthanthemedianblack-holestel-
itscolourorstellarmasshardlychanges;ithasbeenavery
MNRAS000,1–??(2015)