A z=2.5 protocluster associated with the radio galaxy MRC 2104-242: star formation and differing mass functions in dense environments
E. A. Cooke, N. A. Hatch, S. I. Muldrew, E. E. Rigby, J. D. Kurk
MMon. Not. R. Astron. Soc. , 1–14 (2014) Printed 15 September 2018 (MN L A TEX style file v2.2)
A z=2.5 protocluster associated with the radio galaxyMRC 2104-242: star formation and differing massfunctions in dense environments
E. A. Cooke (cid:63) , N. A. Hatch , S. I. Muldrew , E. E. Rigby and J. D. Kurk School of Physics and Astronomy, University of Nottingham, University Park, Nottingham NG7 2RD Leiden Observatory, P.O. Box 9513, 2300 RA, Leiden, The Netherlands Max-Planck-Institut fuer Extraterrestrische Physik, Giessenbachstrasse, D-85748 Garching, Germany
Accepted 2014 March 13. Received 2014 March 10; in original form 2013 December 13
ABSTRACT
We present results from a narrow-band survey of the field around the high redshiftradio galaxy MRC 2104 − α emitters in a 7 sq. arcmin field andcompared the measured number density with that of a field sample at similar redshift.We find that MRC 2104 −
242 lies in an overdensity of galaxies that is 8 . ± . z ∼ . M < M (cid:12) within our small field of view. Based onthe level of overdensity we expect to find ∼
22 star forming galaxies below 10 M (cid:12) in the protocluster and do not detect any. This lack of low mass galaxies affects thelevel of overdensity which we detect. If we only consider high mass ( M > . M (cid:12) )galaxies, the density of the protocluster field increases to ∼
55 times the control fielddensity.
Key words: galaxies: clusters: individual ; galaxies: high-redshift
Locally, the star formation rate (SFR)-mass relation doesnot change as a function of galaxy environment; the frac-tion of galaxies which are star forming differs but the specificstar formation rate (sSFR) is constant irrespective of envi-ronment (Peng et al. 2010). This SFR-mass relation evolveswith redshift, however cluster and field galaxies continue tolie on the same relation up to z = 1 (Muzzin et al. 2012).At higher redshifts, studies have found that this trend of aconstant sSFR between galaxies in the process of forminga cluster (protocluster galaxies) and field galaxies appearsto continue, implying a sSFR independent of environment(Koyama et al. 2013a,b). The existence of a “main sequence”for galaxies suggests that star formation in galaxies proceedsin the same way in (proto)clusters as it does in the field, even (cid:63) e-mail: [email protected] at redshifts z >
2. Protocluster galaxy properties, however,differ from those in the field: the progenitors of low red-shift clusters have previously been found to contain membergalaxies that are older, more star-forming, more metal-richand twice as massive as field galaxies at the same redshift(Steidel et al. 2005; Hatch et al. 2011b; Koyama et al. 2013a;Kulas et al. 2013). This implies that cluster galaxies haveexperienced an accelerated growth in their early years, yettheir sSFRs show no difference from the field up to redshift z = 2.Previously, the SFR-mass relation at z > c (cid:13) a r X i v : . [ a s t r o - ph . GA ] M a r E.A.Cooke et al. cators such as 24 µ m and 250 µ m fluxes, may help to breakthis degeneracy between normal star-forming galaxies andheavily dust-obscured star-bursting objects. Combining thiswith SED-derived masses should provide a better measureof the SFR-mass relation for protocluster and field galaxiesat z > − − − σ overdensity of redgalaxies and the angular correlation function showed thatthe galaxies in this field were more clustered than average(Hatch et al. 2011a). MRC 2104 −
242 lies at z = 2 .
49, whichmeans the H α emission line falls directly within the ISAACnarrow-band filter at 2.29 µ m. This allows us to select star-forming galaxies within a narrow redshift range (∆ z = 0 . α selected galaxies around MRC 2104 − = 70 km s − Mpc − , Ω M = 0 . Λ = 0 . MRC 2104 −
242 lies at a redshift of 2.49 (McCarthy et al.1990) and has been found to lie in an overdensity of redgalaxies ( J − H > H − K + 0 . ∩ J − K > .
5, seeHatch et al. 2011a). We have obtained photometry of thistarget in g (cid:48) , z (cid:48) , J , H , K s , 3.6 µ m, 4.5 µ m, and 24 µ m bandsas well as narrow-band photometry at 2.29 µ m, covering anarea of 2.65 arcmin × α emission line at z = 2 .
49, the redshiftof the radio galaxy. The width of the filter (324 ˚A) allowsus to select H α emitters between 2 . < z < .
51. Thiscorresponds to ∆ v ∼ − , so we expect to detect allprotocluster members. MRC 2104 −
242 was observed in service mode using the HighAcuity Wide-field K-band Imager (HAWK-I) (Kissler-Patig et al. 2008) to obtain the H, J and K s images, and ISAACto obtain the narrow-band (hereafter NB) 2.29 µ m image.Details on the observations and reduction of the H, J andK s data are provided in Hatch et al. (2011a). The NB datawere obtained in 2011 October 8-10th for a total integrationtime of 5.6 h. The ISAAC field of view is smaller than theHAWK-I field of view (2.5 arcmin × × s HAWK-Idata. The pixel scale of the H, J and K s HAWK-I images(0.106 arcsec pixel − ) was degraded to the ISAAC pixel scaleof 0.148 arcsec pixel − . The NB image was convolved to theseeing of the K s of 0.7 arcsec.The total overlapping area of the NB, H, J and K s images is 11.8 sq. arcmin, resulting from the large dither-ing pattern used during the NB observations. To ensurethe image depth was approximately consistent across thewhole image, regions which had less than 30 percent of themaximum exposure time were masked out. The remainingarea is 7 .
09 sq. arcmin. The 3 σ image depths given in Table1 were measured by placing 2 arcsec apertures at multiple( ∼ s image (which was flux-calibrated using 2MASS stars inthe field of view; see Hatch et al. 2011a) and further ad-justments were made to this calibration by comparing the NB − K s colour of stars in the images to the predictedcolours of stars in the Pickles stellar library. Uncertaintiesin the flux calibration are < IRAC (Fazio et al. 2004) observations at 3.6 µ m and 4.5 µ mwere obtained in 2009 during the warm Spitzer mission (PID60112) for a total integration time of 1600 s in both bands.Details of the observations and data reduction can be foundin Galametz et al. (2012). The limiting magnitudes for theIRAC bands were estimated from their completeness curves.
Spitzer
MIPS (Rieke et al. 2004) 24 µ m data was ob-tained as part of the Spitzer
High-redshift Radio Galaxysample survey. Full details of the observations and data re-duction can be found in Seymour et al. (2007).
Herschel
SPIRE (Griffin et al. 2010) 250 µ m imagingwas obtained during the Search for Protoclusters with Her-schel (SPHer) survey. The depth of the SPIRE data of theMRC 2104 −
242 field is identical to that of the three controlfields. A description of the data can be found in Rigby et al.(2014).
Observations in the optical regime ( g (cid:48) and z (cid:48) bands) weretaken in service mode using the Gemini Multi-Object Spec-trograph South (GMOS-S; Hook et al. 2004) instrument on c (cid:13)000
Observations in the optical regime ( g (cid:48) and z (cid:48) bands) weretaken in service mode using the Gemini Multi-Object Spec-trograph South (GMOS-S; Hook et al. 2004) instrument on c (cid:13)000 , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242 Filter Integration Time 3 σ Limit (AB) Instrument g (cid:48) z (cid:48) J H K s µ m 0.44 h 23.0 IRAC4.5 µ m 0.44 h 22.7 IRAC Table 1.
Details of the images used. Limiting magnitudes for theoptical and NIR images were measured using randomly placed2 arcsec apertures. The IRAC image limits were determined fromtheir completeness curves.
16 18 20 22 24NB-2-101234 K s - NB EW = 25A
Figure 1.
Colour-magnitude diagram for the MRC 2104 − K s − NB >
2Σ and K s − NB >
Cerro Pachon, Chile, during the period August–November2010. The z (cid:48) band total integration time was 40 min, and thetotal g (cid:48) band integration time was 3.8 h. The g (cid:48) and z (cid:48) datawere reduced using the Gemini gemtools IRAF package.The usual reduction steps were taken: bias subtraction, flatfielding, and trimming of the image. The z (cid:48) band fringingwas removed using IDL to subtract the fringe frame, whichhad been created using the IRAF package gifringe . Theimages were mosaiced and combined using imcombine .The g (cid:48) image was flux-calibrated by comparing the g (cid:48) − J colour of stars in the image to those predicted using thePickles stellar library (the J image was flux-calibrated using2MASS stars in the field of view; see Hatch et al. 2011a).The z (cid:48) image was then flux-calibrated similarly, using the g (cid:48) − z (cid:48) colour of stars. 3 σ image depths were measured byplacing ∼ We compare our radio galaxy field to three control fieldstaken from the Ultra Deep Survey (UDS), the Cosmic Evo- lution Survey (COSMOS) and the Great Observatories Ori-gins Deep Survey-South (GOODS-S). We have photometryin approximately the same bands as our radio galaxy field( B , z (cid:48) , J , H , K s , 3.6 µ m, 4.5 µ m, 24 µ m, 250 µ m). NB im-ages were taken using the HAWK-I H µ m filter forthe UDS and COSMOS fields and using the NB2090 filterfor the GOODS-S field. These filters detect H α emission at2 . (cid:54) z (cid:54) .
26 and 2 . (cid:54) z (cid:54) .
21 respectively. When cal-culating densities we scale our control field results accord-ing to the different volumes given by each filter. Each of ourcontrol fields is limited by the size of the NB field-of-viewand are all approximately 57 sq. arcmin. We refer to Hatchet al. (2011b) for details on the reduction of the K s and NBimages. The remaining photometry was obtained from pub-lic archives and is described in Capak et al. (2007, 2011);Furusawa (2008); Retzlaff et al. (2010); McCracken et al.(2012); Hartley et al. (2013). The Spitzer data was obtainedfrom the NASA/IPAC Infrared Science Archive. The
Her-schel µ m data was obtained from the H-ATLAS survey(Eales et al. 2010) and re-reduced to have the same depth asthe MRC 2104 −
242 data, see Rigby et al. (2014) for details.
The
SExtractor software package (Bertin & Arnouts1996) was used to create a photometric catalogue of ourdata. We used
SExtractor in dual-image mode, using aweighted NB image as the detection image, to obtain fluxesin all bands. The NB image was weighted with the squareroot of the effective exposure map, which takes backgroundnoise into account. We select as sources those with 25 ad-joining pixels that are 1 σ above the rms background anduse apertures of 2 arcsec in diameter for measuring colours.These apertures are significantly larger than the ∼ . auto apertures. Limiting magnitudes for the optical andNIR bands were estimated by measuring the standard de-viation of the flux densities in 2 arcsec diameter aperturesplaced randomly on the images (Table 1). For the IRAC3.6 µ m and 4.5 µ m bands, SExtractor was optimised with minarea = 4 pixels and detect thresh = 2 . σ abovethe rms background. The NB photometric catalogues werematched with the IRAC catalogues within 1 arcsec using Topcat (Taylor 2005) to produce the full photometric cat-alogue. In order to determine what effect the choice of
SExtractor parameters had on our results, we checkedour methods using three different parameter combinations:2 arcsec fixed apertures (25 adjoining pixels), auto aper-tures for 25 adjoining pixels and auto apertures with 24adjoining pixels. We found that the choice of selection pa-rameters does not significantly affect our results and doesnot alter our conclusions.
To obtain a sample of NB-excess sources we followed themethod of Bunker et al. (1995), selecting sources with excessNB signal relative to the K s band. Sources with a value of K s − NB (cid:62)
2Σ were selected as NB excess sources, with Σdefined as: c (cid:13) , 1–14 E.A.Cooke et al.
Σ = 1 − − . K − NB ) − . zp − NB ) (cid:112) πr ap ( σ NB + σ K ) (1) K and NB are the AB magnitudes in each band, σ val-ues are the SExtractor errors for each band, πr ap is thearea of the aperture used and zp is the zero-point of theimages; here zp = 26 . K s − NB colours against the NBmagnitudes for all sources. Σ quantifies the significance ofthe NB excess and our 2Σ selection corresponds to a com-pleteness cut in star formation rate (SFR) of ∼ (cid:12) yr − .We also exclude sources with NB magnitude fainter than22 .
9. At this limit we are >
80% complete in both the radiogalaxy field and all the control fields. Completeness was cal-culated by comparing the detection catalogues for the NBand deeper K s images. In Figure 2 we plot the complete-ness curves for each field in the NB and K s band. Verticallines indicate where the NB becomes 80% complete. Our NB > . K s − NB > µ m. α emitters Excess NB flux could also be produced from low-redshift( z <
1) emission line contaminants or [O iii ] lines fromsources at z = 3 . α emitters as sources with BzK colours(( z − K s ) − ( B − z ) > − .
2, or equivalently gzK colours:( z − K s ) − (cid:16) ( g (cid:48) − z (cid:48) ) − . . (cid:17) > − . . < z < . (cid:54)
13% from galaxies at z < B band photometry inthe radio galaxy field so we used the g (cid:48) band photometryin its place. We converted the selection criteria using modelgalaxy spectra, redshifted to the lower limit of BzK-selectedgalaxies ( z = 1 .
4) and convolved with B , g (cid:48) and z (cid:48) filters.A line was fit to the g (cid:48) − z (cid:48) versus B − z (cid:48) points to obtainthe selection conversion. Secondly, for sources with IRACdetections, a colour cut of [3 . − [4 . > − . z > . z = 3 .
57 and cannot remove themfrom our sample. After applying our selection to our NB ex-cess sources, we have 18 H α emitters in our sample (from 31 NB excess sources), including the radio galaxy and three“companion” galaxies, which lie within 3 arcsec of the radiogalaxy. 9 of these H α emitters were selected via the IRACcolour selection, and 11 via the BzK criterion (2 were se-lected by both criteria). We select 17 /
25, 9 /
16, 8 /
12 (H α emitters / NB excess sources) from the COSMOS, UDS andGOODS-S control fields respectively.
We estimate the contamination rate of AGN in our controlfields using the
Spitzer
IRAC criterion from Donley et al.(2012). From this selection we estimate that there are twopossible AGN in the COSMOS H α emitter sample and nonein the UDS or GOODS-S samples. We do not have 5.8 µ mand 8 µ m data for the MRC 2104 −
242 field that is deepenough to determine the number of AGN around the radiogalaxy. Assuming the AGN fraction in the MRC 2104 − / H α emitters= 0 . α EMITTERS3.1 Stellar mass
We determined stellar masses by using the SED fitting pro-gramme “Fitting and Assessment of Synthetic Templates”(FAST, Kriek et al. 2009) to fit the photometry of our sam-ple of H α candidates to obtain mass estimates. We assumefrom now on that the NB excess flux in the H α candidates isdue to H α +[N ii ] emission at the redshift of the radio galaxyand we fixed the redshift of the fit to z = 2 .
49. The controlfield galaxy redshifts were set to z = 2 . , . , and 2 .
19 forCOSMOS, UDS and GOODS-S respectively, assuming H α emission from the centre of the NB filters.We used FAST to fit Bruzual & Charlot (2003) stellarpopulation synthesis models with a Chabrier (2003) IMFto our photometry ( B / g (cid:48) , z (cid:48) , J , H , K ,[3 . . /
18 H α emitters in the MRC 2104 −
242 field had detections in theIRAC bands. We fit delayed exponentially declining (SFR ∼ t exp[ − t/τ ]) star formation histories with dust extinction0 < A V < . < log ( τ / yr) < . . < log (age / yr) < . A V ) and the assumedstar formation histories, we do not use these outputs fromthe FAST output as they are likely to be highly unreliable.However, the mass output is robust independent of the ex-act star formation history template that is assumed (Shapleyet al. 2005). Errors in the stellar masses are determined from100 Monte Carlo simulations performed by FAST, with thephotometry being varied within the flux uncertainties. Wealso added a rest-frame template error function to take intoaccount the uncertainties in the model templates. c (cid:13)000
242 field had detections in theIRAC bands. We fit delayed exponentially declining (SFR ∼ t exp[ − t/τ ]) star formation histories with dust extinction0 < A V < . < log ( τ / yr) < . . < log (age / yr) < . A V ) and the assumedstar formation histories, we do not use these outputs fromthe FAST output as they are likely to be highly unreliable.However, the mass output is robust independent of the ex-act star formation history template that is assumed (Shapleyet al. 2005). Errors in the stellar masses are determined from100 Monte Carlo simulations performed by FAST, with thephotometry being varied within the flux uncertainties. Wealso added a rest-frame template error function to take intoaccount the uncertainties in the model templates. c (cid:13)000 , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242
16 18 20 22 24 26 28mag01234 l og ( N ) KNB MRC2104-242 16 18 20 22 24 26 28mag COSMOS 16 18 20 22 24 26 28mag UDS 16 18 20 22 24 26 28mag GOODS-S
Figure 2.
Completeness histograms for the K s (purple lines) and NB (green dashed lines) images in the MRC 2104 −
242 field and controlfields. Vertical lines mark the 80% completeness limit for each NB image.
Figure 3.
Median stacks of MIPS 24 µ m images for H α emitters.Clockwise from top left: MRC 2104 −
242 (14 stamps), COSMOS(17 stamps), GOODS-S (8 stamps), UDS (9 stamps). All imageshave the same scale. Three of the four fields have clear detec-tions, with MRC 2104 −
242 showing a stronger signal. The radiogalaxy and companions are not included in the stack, however theCOSMOS AGN candidates are included.
Some of the photometry for the control fields is deeperthan for the protocluster field. In our analysis only detec-tions to the depth of the MRC 2104 −
242 field were consid-ered in the control fields. We have checked our results usingfull-depth magnitudes for the control field and find that ouroverall conclusions are unaffected by the different depths ofthe images between fields. α -derived SFRs We calculate the K s continuum and convert our NB signalto an H α flux using: f ( K cont ) = w K s f ( K s ) − w NB f (NB) w K s − w NB (2) f (H α ) = w NB [ f (NB) − f ( K cont )] (3)where f ( K cont ) is the continuum flux density in the K s band, f (NB) and f ( K s ) are the flux densities in the NB and K s bands respectively, f (H α ) is the H α flux, and w K s and w NB are the widths of the corresponding filters.These values are corrected for dust extinction calculatedfrom the B − z (cid:48) colour , which corresponds to the rest-frameUV slope, following the method of Daddi et al. (2004): E ( B − V ) = 0 . B − z (cid:48) + 0 . AB (4)Note that here we assume that the extinction for H α isthe same as for the broadband SED. Where sources had g (cid:48) , B or z (cid:48) magnitudes fainter than the 3 σ limiting magnitude(see Table 1) we convolved the best fitting SED templatefor that source with the appropriate filter curve in order toget a magnitude estimate. For the radio galaxy field anysources with g (cid:48) magnitudes fainter than 3 times the limitingmagnitude were convolved with a B filter curve to avoidhaving to convert the colours. For each of the control fieldsand for the radio galaxy field z (cid:48) band, we used the B or z (cid:48) filter curve of the instrument used to obtain the data.Dust-corrected H α luminosities were then calculated,scaling for luminosity distance, and H α SFRs determinedusing the Kennicutt (1998) relation, converted to a Chabrier(2003) IMF:
SF R (M (cid:12) yr − ) = 4 . × − L Hα (erg s − ) (5) µ m SFRs The
Spitzer µ m filter transmits between 20 . . µ m,which corresponds to rest-frame wavelengths of 6 . . µ mfor z = 2 .
49 galaxies. This rest-frame wavelength range isdominated by polycyclic aromatic hydrocarbon (PAH) fea-tures, which have been shown to provide a good measure ofhidden star formation (Siana et al. 2009).The 24 µ m data have a 3 σ detection limit of ∼ .
11 mJy.We have a > σ detection in 24 µ m for the radio galaxy and For the MRC 2104 −
242 field the B − z (cid:48) colour was calculatedusing ( B − z (cid:48) ) = (cid:16) ( g (cid:48) − z (cid:48) )+0 . . (cid:17) at z = 2 . (cid:13) , 1–14 E.A.Cooke et al. a r c m i n s COSMOS
UDS
GOODS-S
Figure 4.
The control fields used in this study. From left: COSMOS, UDS, GOODS-S. The figures show the NB images, with detectedH α sources overlaid as green circles. The AGN candidates in the COSMOS field are highlighted in blue. Each window is 7.5 arcmin × its companions (these sources are blended in the 24 µ m im-age), however the majority of our H α emitters were not in-dividually detected. We therefore stacked the sources to ob-tain a median flux density for each field. The radio galaxyand its companions were not included in the stack, how-ever we include the AGN candidates in the COSMOS fieldas these sources were not individually detected at > σ .Postage stamps of 22 ×
22 pixels (4 . Spitzer µ m FWHM) were created around each H α source, andsources in each field were median stacked (Figure 3). Fluxdensities were then measured from the stacks in 8 pixel(5 arcsec) diameter apertures (Table 2). These rest-frameIR flux densities were converted to SFRs using both themethods outlined in Rujopakarn et al. (2013) (their sec-tion 5) and using equation 14 of Rieke et al. (2009) . Themethod from Rujopakarn et al. (2013) assumes these galax-ies lie on the galaxy main sequence (MS), whereas Riekeet al. (2009) calculate the SFR for (ultra) luminous infraredgalaxies ([U]LIRGs). Without additional information, suchas a measure of the IR bump, we cannot distinguish betweenthe two scenarios for the galaxies in our sample (see Elbazet al. 2011) and so use both methods in our analysis.The detection limit of 0 .
11 mJy corresponds to ∼
145 M (cid:12) yr − or ∼ (cid:12) yr − (MS or ULIRG) at z = 2 . Herschel µ m SFRs The
Herschel
SPIRE 250 µ m filter probes the far-IR bumpfor galaxies at z >
2, allowing the total IR luminosity of dis-tant galaxies to be measured. These data have a 3 σ detectionlimit of ∼
375 M (cid:12) yr − at z = 2 .
5. The radio galaxy and itscompanions are detected in the
Herschel µ m data, and afew other H α sources had > σ detections within 10 arcsec,however due to the large beam size of Herschel we are un-able to robustly identify counterparts. To obtain an estimateof the SFR of the H α emitters we therefore median stacked log ( SF R IR ) = 0 .
108 + 1 . (cid:0) πL d f (cid:1) − f is the flux density in an 8 pixel diameter aperture, L d is the luminosity distance in cm. -90 -60 -30 0 30arcsecs9060300-30 a r csecs Figure 5. K s image of the field around MRC 2104 − α sources are shown with greencircles. The radio galaxy and three companions (see Figure 6) lieat the origin, within the larger green circle of radius 3 arcsec. Thewindow size is 2.65 arcmin × − α emitters in a ∼ all H α sources (not including the radio galaxy and its com-panions). A SFR was derived from the median 250 µ m fluxby modelling the IR bump as an isothermal body of temper-ature 35 Kelvin and β = 1 .
5. This template was normalisedto the detected 250 µ m flux and integrated over 8-1000 µ mto obtain L IR . The L IR was converted to a SFR using theKennicutt (1998) relation adjusted to a Chabier IMF by di-viding the SFRs by 1 .
6. Median stacks of the H α emittersin the UDS, COSMOS and GOODS-S fields were produced c (cid:13) , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242 Field n Flux density ( µ Jy) SFR (MS; M (cid:12) yr − ) SFR (ULIRG; M (cid:12) yr − )MRC 2104 −
242 14 35.7 ± . ± . . ± . ± a . ± . . ± . ± . ± . . ± . b ± Table 2.
Flux densities measured from the 24 µ m stacks in an aperture of radius 5 arcsec. The uncertainties are the standard deviationof 1000 sets of n stacked random regions (where n is the number of H α sources in each field). The SFRs given are calculated from the24 µ m fluxes using relations based on local ULIRGs and main sequence (MS) estimates. a Numbers in brackets for COSMOS are fluxdensity and error values when the AGN candidates are removed from the stack. b There was no detectable signal in the GOODS-S stack,we use the 3 σ value in all SFR calculations. -4 -2 0 2 4arcsecs420-2-4 a r csecs Figure 6.
NB image of the radio galaxy MRC 2104 −
242 and itsthree companion sources, all circled in green. in the same manner, but none of these stacks resulted in asignal above 3 σ significance. The field around MRC 2104 −
242 has a large overdensity ofH α emitters (Figures 4 & 5). Excluding the radio galaxyand three nearby companions there are 14 objects in a7.09 sq. arcmin field, which is 8 . ± . . ± .
8. The field of view around the HzRG is relativelysmall (4 . × . z > ∼
10 Mpc (Venemans et al. 2007; Hatchet al. 2011a). As Chiang, Overzier & Gebhardt (2013) showthis means we cannot say anything for certain about themass of this structure, however, this level of overdensity isof the same order that has been found in other protoclustersat similar redshift (e.g. Kurk et al. 2004; Hatch et al. 2011b;Hayashi et al. 2012). MRC 2104 −
242 is therefore likely toalso lie within a protocluster.We tested to see if there was any preferential cluster-ing of H α sources around the radio galaxy. We did this by comparing the average distance from the radio galaxy toaverage distances calculated from random distributions ofsources. The average distance of the H α sources from theradio galaxy differs from that expected from a random dis-tribution at a 2.6 sigma level. However, this includes thethree companion galaxies within 3 arcsec of the radio galaxy.When these three sources are excluded from the analysis thesignificance is only 1.2 sigma. Therefore there is no strongclustering around the radio galaxy. Hatch et al. (2011a) found a 3 σ overdensity of JHK galaxies( J − H > H − K + 0 . ∩ J − K > . − α emitters. We find 10 JHK galaxies within the ISAACfield-of-view (Figures 5 and 7), one of which is the radiogalaxy. The spatial distribution of the JHK galaxies is pre-sented in Figure 5.Whilst all of our H α emitters are likely to lie withinthe protocluster, the JHK galaxies lie within a much largerredshift range and so it is unclear whether they are associ-ated with the protocluster. Two JHK galaxies, in additionto the radio galaxy, are H α emitters, meaning these galax-ies are highly dust obscured, star forming galaxies which liein the protocluster. One of these is the H α source with a3 σ signal at 24 µ m and 2 σ signal at 250 µ m. Stacking theNB images for the remaining 7 JHK galaxies does not pro-duce a signal, giving an upper limit of SF R ∼ . (cid:12) yr − ,and there is no significant detection ( < σ ) in the stackedMIPS 24 µ m and Herschel µ m images. Hence if the re-maining 7 JHK galaxies are in the protocluster the lack ofNB emission indicates that they are passive, with a sSFR oflog ( sSF R/ yr − ) (cid:54) − . α emitters in theprotocluster and control fields In this section we perform a detailed comparison of the pro-tocluster and control galaxies, including their stellar masses,SFRs, dust extinction ( A V ), and sSFRs. In all followinganalysis the radio galaxy and three companions (see Fig-ure 6) have been removed from the protocluster sample asthese objects are likely to be affected by the radio jets. c (cid:13) , 1–14 E.A.Cooke et al. (M/M O • )0.00.20.40.60.81.01.2 N (a) MRC2104-242Control fieldControl field M > 10 M O • H α / M O • yr -1 (b) 1A V (c) -10 -9 -8 -7 -6log (sSFR H α / yr -1 ) (d) Figure 8.
A comparison of the properties of the protocluster galaxies (black) and control galaxies (red dashed lines) including: ( a ) Mass,( b ) H α SFR (dust corrected), ( c ) A V , ( d ) sSFR. Shaded in red are the mass-selected control field histograms for SFR, A V and sSFR(log( M /M (cid:12) ) > J - H Figure 7.
Near IR colours of galaxies in the MRC 2104 −
242 field.Lines mark the JHK criterion used to select red galaxies at highredshift; galaxies selected this way are shown by red squares. H α emitters are highlighted with green circles. The protocluster galaxies are on average more massive thanthe control field galaxies as shown by Figure 8a. A two-sided Kolmogorov-Smirnov (K-S) test shows a significantdifference between the two samples: K-S p = 2 . × − .The SED fits at masses M < M (cid:12) have large errors as-sociated with them, but even if we exclude these galaxiesfrom our analysis there is still a significant difference (K-S p = 1 . × − ). A similar difference between the massesof protocluster and control galaxies has been found in other z > M > . M (cid:12) objects and no objects with M < M (cid:12) withinour 7 sq. arcmin field-of-view. Our detection method selectson H α equivalent width and galaxies below our complete-ness limit in SFR ( < (cid:12) yr − ) may not be selected. TheH α sample is therefore incomplete at all masses and partic-ularly at low masses due to the mass-SFR relation. Howeverwe emphasise that both the protocluster and the controlfields are incomplete to the same level as we have ensuredthat the selection method is identical in all fields. Hence the difference in mass functions in different environments ispuzzling and is discussed in detail in Section 5.2. The protocluster galaxies typically have higher dust extinc-tion, as calculated from their UV slopes, than the field galax-ies, with a median A V that is twice as large (see Figure 8c).A K-S test shows a significant difference in the dust contentbetween the two environments: K-S p = 3 . × − .Dust extinction correlates strongly with galaxy mass(e.g. Garn & Best 2010) so we tested whether the ob-served trend was a symptom of the mass difference foundin Section 4.3.1 by limiting our analysis to galaxies with M (cid:62) M (cid:12) . In Figure 9 we show the values of A V in boththe protocluster and control fields as a function of mass,with filled red squares highlighting the control field galax-ies with M (cid:62) M (cid:12) . The range of A V reduces for thismass-limited sample and the control field galaxies are moreconsistent with those in the protocluster. There remains asignificant difference in the dust extinction measured in theprotocluster and control galaxies for this sample, howeveronly at a 2 σ level (K-S p = 0 . The H α SFRs corrected for dust extinction using the UVslope are plotted in Figures 8b and 10; there is little differ-ence between the protocluster and control galaxies. A K-Stest results in a probability of 0 . α SFRsagainst the SED-derived stellar masses for both the proto-cluster and control field galaxies. The Daddi et al. (2007)and Santini et al. (2009) correlations showing the “main-sequence” for z ∼ α emitters with M < . M (cid:12) is consistentwith the main sequence, but at higher masses both the pro-tocluster and control field galaxies appear to lie below thisrelation. This suggests that the applied dust-correction forthe high-mass H α emitters is not sufficient and there may beadditional star formation that is heavily optically obscured.It is extremely difficult to correct for dust extinction usingthe UV slope alone (Elbaz et al. 2011) and a far more accu-rate measurement of the total SFR is obtained through theIR luminosity. c (cid:13) , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242 (M/M O • )012345 A V MRC2104-242Control fieldControl field M > 10 M O • Figure 9. A V as a function of galaxy mass for the protocluster(black circles) and control field (red squares) galaxies. There is atrend for increasing dust extinction with galaxy mass, with thetrend becoming steeper at higher masses. In Figure 11 we show the total SFR derived by com-bining the raw H α SFRs with SFRs derived through the IR24 µ m and 250 µ m luminosities. SFRs derived using 24 µ mhave two values depending on whether we assume they haveULIRG SEDs or whether they have main-sequence SEDs.We note that as H α emission is less sensitive to dust at-tenuation than rest-frame UV light, these total SFRs mayslightly overestimate the true SFR. However, the derivedtotal SFRs are almost entirely dominated by the IR, so thecontribution from unobscured H α is likely to be negligible.The Herschel µ m protocluster SFR estimate is in betteragreement with the 24 µ m IR SFR estimate based on localULIRGs (Rieke et al. 2009), although all of these IR esti-mates are in agreement with the main sequence relationship.Whilst the 24 µ m signal could be due to AGN-heated warmdust, the detection of 250 µ m flux (rest-frame 70 µ m) in theprotocluster galaxies indicates that we must be detectingcooler dust heated by UV emission from young, hot stars.The IR+H α SFRs are comparable to the dust-correctedH α SFRs in the control fields, but in the protocluster we finda large discrepancy. The IR+H α SFRs are at least twice asfast (and up to ten times as fast) as the dust-corrected H α SFRs which implies the protocluster galaxies contain moreoptically-obscured star formation than in the control galax-ies. These results imply that the total SFR of the massivegalaxies which reside in dense regions cannot be derived fromH α estimates alone; the protocluster galaxies have highermasses with large dust extinctions, therefore far-IR or sub-millimetre data are required to probe the optically-obscuredstar formation. We note that the large amount of dust ex-tinction may have implications for studies which aim to de-tect protoclusters and study them through Lyman α emis-sion from their member galaxies.The IR SFRs reveal a different picture to the H α SFRs:the protocluster galaxies are forming stars more rapidly thanthe control galaxies but much of this star formation is hiddenfrom optical view. Figure 11 reveals that once this obscuredstar formation is taken into account the protocluster galaxies (M/M O • )1101001000 S F R / M O • y r - MRC2104-242Control field D07 S09
Figure 10.
Dust corrected H α SFRs against stellar mass. Theprotocluster galaxies are plotted as filled, black circles and liemostly at the high mass end of the plot. The control fields areplotted as open, red squares. The radio galaxy and three com-panion galaxies are highlighted with green diamonds. The COS-MOS AGN candidates are plotted as filled red squares. Medianerror bars at the top show the typical uncertainty on the SEDmass estimates in four mass bins (6.5 (cid:54) log ( M/ M (cid:12) ) < . (cid:54) log ( M/ M (cid:12) ) < .
5, 9.5 (cid:54) log ( M/ M (cid:12) ) < .
5, 10.5 (cid:54) log ( M/ M (cid:12) ) < M ∼ . M (cid:12) . lie on the same main sequence of the mass-SFR relation asthe control galaxies.Our IR SFR estimates for both control and protoclus-ter galaxies are consistent with the main-sequence of theSFR-mass relation, suggesting that the majority of theseH α emitters are not undergoing a “bursty” mode of starformation but rather forming stars at the expected rate fortheir mass. This is in agreement with previous protoclusterstudies (Koyama et al. 2013a,b). However, we note that theSFR IR are derived from median stacks, thus our methodwould not be able to find starbursting galaxies if the major-ity of the H α emitters were main sequence galaxies. A fewof the protocluster galaxies have 2 σ detections at 250 µ m,and one has a 3 σ detection at 24 µ m. If we remove from the250 µ m stack those H α emitters with nearby ( (cid:54)
10 arcsec)2 σ detections, the signal of the stack decreases and we donot find a signal above 3 σ (where 3 σ corresponds to an up-per limit of 98 M (cid:12) yr − ). We discuss these galaxies furtherin Section 4.4. Figure 8d compares the specific star formation rates (sSFR)of the protocluster and control galaxies. When the entiremass range of galaxies is taken into account there is a sig-nificant difference in the sSFRs between the two populations(K-S p = 6 . × − ). However this difference is driven bythe disparate mass distributions of galaxies in the two envi-ronments. The shaded red histogram shows the distributionof sSFRs of galaxies with masses M (cid:62) M (cid:12) . For thispopulation there is no significant difference in the sSFRs:K-S p = 0 . c (cid:13) , 1–14 E.A.Cooke et al.
No H α emitters in the protocluster or control fields are de-tected above 3 σ significance at 250 µ m, however there are afew detections with signals > σ . In the MRC 2104 − σ sources, one of which has a 3 σ µ m detection of 0 . µ Jy= 145 ±
60 M (cid:12) yr − (MS) or= 1200 ±
775 M (cid:12) yr − (ULIRG). Their 250 µ m SFRs areplotted in Figure 11 as small black diamonds.In the control fields we only find one source with a > σ detection. The 250 µ m-derived SFR is plotted as a smallred diamond in Figure 11. This source is one of the AGNcandidates in the COSMOS field.We expect 5% of our H α emitters (i.e. < α emitters) to be detected at the 2 σ level due to noise in the250 µ m data. In the protocluster we find three, suggestingthat at least two of them are real sources and not noise. Allthree sources have 250 µ m SFRs which are consistent withstarbursting galaxies, defined such that they lie four timesabove the main sequence (Rodighiero et al. 2011). This sug-gests that the fraction of starbursts is several times higherin the protocluster, with 21% of the H α emitters being star-burst galaxies, compared to just ∼
3% in the control field.
Removing the two AGN detected in the COSMOS field fromour control sample does not significantly change our results.There is still a significant difference in dust content esti-mated from the UV slope (K-S p = 2 × − ) which remainsat a 2 σ level when considering the mass-limited galaxy sam-ples. Furthermore the trends for the sSFRs remain the same:K-S p = 2 . × − and K-S p = 0 . α SFRs decrease for the COSMOS field and the H α SFR dis-tributions become significantly different at a 2 σ level (K-S p = 0 . > M (cid:12) )there is still no significant difference in the SFRs betweenthe two distributions: K-S p = 0 .
43. Excluding the COS-MOS AGN, the starburst galaxy fraction is still higher inthe protocluster than the control field.
The NB filters used for our control fields have different cen-tral wavelengths from the NB229 filter used to select theprotocluster galaxies. Since we select galaxies at slightly dif-ferent redshifts, the luminosity distance to the control fieldgalaxies is slightly less than to the protocluster galaxies. Asthe control field galaxies are at lower redshifts than the pro-tocluster, we probe further down the luminosity function ofthe control field for the same cuts in apparent magnitude.We have tested how this may affect the results by takingthis difference in magnitude into account and applying acut to the control fields at brighter magnitudes. These cuts The median dust-corrected H α SFR for the COSMOS field de-creases by < (cid:12) yr − and the 24 µ m + H α SFR decreases by ∼ (cid:12) yr − . (M/M O • )10100 S F R I R + H α / M O • y r - D07 S0924 µ m ULIRG 24 µ m MS H α µ m 250 µ mMedian stacks Individual Figure 11.
The IR+H α SFRs plotted against median mass val-ues for the protocluster (black filled symbols) and control fields(red open symbols). From left to right: GOODS-S, COSMOS,UDS. 3 σ upper limits are plotted if no signal is observed. Cir-cles and squares (offset in the x -axis for clarity) are SFRs cal-culated from 24 µ m using Rujopakarn et al. (2013) and Riekeet al. (2009) respectively. Triangles are the dust-corrected H α SFRs, with error bars enclosing 68% of the data. The crosses arethe SFRs derived from
Herschel µ m stacks of H α emitters.Overplotted are Daddi et al. (2007) and Santini et al. (2009) rela-tions, labelled D07 and S09. Also plotted are the Herschel
SFRs(IR+H α ) and 1 σ error bars for sources with 250 µ m signal > σ above the background noise: black/red diamonds are protoclus-ter/COSMOS galaxies. These sources are consistent with beingstarbursts ( ∼ × the main sequence); the grey dotted line showsthe Santini et al. (2009) relation multiplied by 4. remove five control field galaxies from our sample, increas-ing the level of overdensity measured in the protoclusterfield to 9 ± . We have shown that the star forming protocluster galaxies at z = 2 . µ m and Herschel µ m data, we find that on average, protocluster and con-trol galaxies lie on the same main-sequence of the SFR-massrelation. This means that at z ∼ .
5, galaxy growth in termsof star formation is regulated predominantly by galaxy massand is not greatly affected by the environment of the hostgalaxy. c (cid:13) , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242 (M/M O • )10 -5 -4 -3 -2 N / M p c MRC2104-242Control fieldControl field x 25
Figure 12.
Galaxy number densities per mass bin for the controlfield (red diamonds) and the protocluster (blue triangles). In greywe also show the control field distribution, scaled by a factor of25, to illustrate the expected number densities in the protocluster.This figure shows a clear excess of galaxies in the protocluster atthe high mass end, however there appears to be a lack of lowmass objects in the protocluster, whereas we detect many lowmass objects in the field.
Figures 8a and 10 show that the protocluster galaxiestypically have higher masses than the control field galaxies,and there are more than twice as many protocluster galax-ies than field galaxies with
M > . M (cid:12) (10 protoclus-ter galaxies compared to only 4 control field galaxies eventhough the protocluster area surveyed is only 4% of the con-trol area). This poses a conundrum: if the SFR is governedby galaxy mass alone at z ∼ .
5, then how did the proto-cluster galaxies gain so much mass so rapidly? The earlyformation of these galaxies must be dependent on their en-vironments at higher redshift, even though at z ∼ . σ detections at 250 µ m in the protoclus-ter, suggesting the presence of starbursting galaxies. If thefraction of galaxies undergoing a starburst is much greaterin denser environments, this may explain the higher masses.Deeper submm observations of protocluster galaxies are es-sential to understanding this issue. We now examine why, on average, galaxy masses differ be-tween the two environments. We find no difference at thehigh mass end of the distributions; taking a mass selectedsample of all H α emitters with M (cid:62) M (cid:12) there is nosignificant difference in the mass distributions. However, wefind no low mass ( M < M (cid:12) ) galaxies in our protoclus-ter sample. This skew in the mass distribution means thatthe strength of the overdensity that we detect depends onthe mass range we examine, e.g. the protocluster numberdensity is ∼
25 times the control field if we only considerobjects with
M > M (cid:12) and ∼
55 times the control fieldat
M > . M (cid:12) (see Figure 12). This large excess of high-mass galaxies suggests the presence of a galaxy protocluster, (M/ (M O • ))0.0010.0100.1001.000 Φ / ( M p c - d l og M - ) ProtoclustersAll galaxies
Figure 13.
Semi-analytic derived mass distributions for all galax-ies (red diamonds) and protocluster galaxies (blue triangles, seetext for details). Dashed lines show the fitted Schechter function.The values of M ∗ and Φ differ by 0 . .
14 dex respectively.The difference in α is 0 .
08 dex. as discussed in Section 4.1. If the MRC 2104 −
242 field doescontain a protocluster then we also expect to find an over-density of low mass galaxies within the field. Although weare incomplete in mass, particularly at low masses, we areincomplete to the same level in the protocluster and the con-trol fields. Since we detect 22 H α emitters at M < M (cid:12) in the control fields, we expect to detect ∼ α emittersin the protocluster, assuming an overdensity of 24, whereaswe do not detect any (Figure 8a). We note that the Koyamaet al. (2013a) study shows that the protocluster aroundMRC 1138 −
262 (the Spiderweb galaxy) also lacks low-massobjects. The difference we find in the average masses be-tween the MRC 2104 −
242 field and the control field is dueto this lack of low mass galaxies in the protocluster, ratherthan a population of extremely massive galaxies.In the following subsections, we consider three possi-ble reasons for this difference in the protocluster and con-trol field mass distributions: an intrinsic difference in massfunctions between the protocluster and the field galaxies;observational effects, such as the higher value of dust ex-tinction in protocluster galaxies or low mass galaxies whichmay have already shut down their star formation; and masssegregation, with high mass galaxies preferentially clusteredaround the radio galaxy.
In order to determine an expected mass function for proto-clusters at z ∼ .
5, we use semi-analytic models to producethe mass distributions of a protocluster and the surroundingfield. We have taken the z = 2 .
42 output of the Guo et al.(2011) semi-analytic model built upon the Millennium DarkMatter Simulation (Springel et al. 2005). The MillenniumSimulation follows the evolution of 2160 dark matter par-ticles from z = 127 to the present day in a box of comovingside length 500 h − Mpc. The simulation adopts a flat ΛCDMcosmology with { Ω , Ω Λ , σ , n, h } = { . , . , . , , . } . c (cid:13) , 1–14 E.A.Cooke et al.
This is consistent with the Two-Degree Field Galaxy Red-shift Survey (2dFGRS; Colless et al. 2001) and the Wilkin-son Microwave Anisotropy Probe first year results (WMAP;Spergel et al. 2003), but is marginally discrepant with thelatest measurements of cosmological parameters (PlanckCollaboration 2013). Haloes were identified using a Friends-of-Friends algorithm (FoF; Davis et al. 1985) with linkinglength 0.2, which were then analysed for bound substruc-tures using subfind (Springel et al. 2001). Only haloes con-taining 20 particles were considered and we note that sim-ilar results can be found with other halo finders (Muldrew,Pearce & Power 2011; Knebe et al. 2011).Galaxies were added to the halo merger tree using theGuo et al. (2011) semi-analytic model, which is an updatedversion of the Croton et al. (2006) and De Lucia & Blaizot(2007) models. A full description of the model, includingmodifications, can be found in those papers. Traditionallysemi-analytic models have been poor at reproducing theredshift evolution of the galaxy stellar mass function. Asshown in figure 23 of Guo et al. (2011), the high mass endof the galaxy stellar mass function is reproduced well inthis redshift range, but there is an over-abundance of lowermass galaxies. In order to minimise the effect of this over-abundance of low mass galaxies, we limit our sample togalaxies with stellar masses greater than 10 . h − M (cid:12) .We identify 1938 clusters in the z = 0 catalogue by con-sidering haloes with masses greater 10 h − M (cid:12) . For each z = 0 identified cluster, we locate the highest mass progen-itor galaxy in the z = 2 .
42 catalogue. We then subsamplecubes of side length 3 . h − Mpc comoving centred on theseprogenitors and compare the mass function with that of thewhole volume, a cube of side length 500 h − Mpc comoving.Fitting a Schechter curve to the semi-analytic derivedmass distributions (Figure 13) we find that the expectedmass function of the protocluster shows no turnover at thefaint end. Indeed, we find the faint end slope for protoclustergalaxies tends to be slightly steeper (by 0.1 dex) than thatfor the whole volume. The distributions differ significantlyin the value of M ∗ and normalisation (differences of 0 . .
14 dex respectively). This means that the difference innumber densities that we observe is not due to a fundamen-tal difference in the shape of the mass functions at z ∼ . Our NB survey selects star forming galaxies with H α emis-sion, down to a dust-uncorrected star formation rate of ∼ (cid:12) yr − . If the low mass protocluster galaxies were pas-sive or heavily obscured by dust, our NB survey would notdetect them. To test if these galaxies are missing in our NBsurvey, we compare the galaxy luminosity functions in theprotocluster field to the control field (Figure 14).We compare the luminosity functions in the K s band,using a J − H > H − K s − .
15 cut to remove galaxies at red-shifts below ∼
1, and at 4.5 µ m, taking a ([3 . − [4 . AB > − . z > .
14 16 18 20 22 24 26K s N / s q a r c m i n MRC2104-242Control fieldMRC2104 - cf14 16 18 20 22 24 264.5 µ m012 N / s q a r c m i n Figure 14.
Number density histograms in the K s band and at4.5 µ m for galaxies with colours J − H > H − K s − .
15 (Vega)and [3 . − [4 . > − . −
242 field, red is the control field and green is thedifference between the two, indicating protocluster candidates.Completeness is shown by the vertical dashed lines. The lack ofprotocluster galaxies at magnitudes brighter than the complete-ness limits, shown by the drop in the green histograms, suggestsa lack of faint galaxies in this protocluster. bands select passive galaxies, as well as the star-forming NBemitters in our sample, albeit with a large contaminationrate. We find an overdensity of bright galaxies ( K s < . µ m < .
5) and a lack of faint galaxies in both K s and 4.5 µ m at magnitudes fainter than 21.9 (AB) and 20.5(AB) respectively. The lack of faint galaxies, at magnitudesbrighter than the completeness limits (shown by the verti-cal dashed lines in Figure 14), suggests that this protoclusterlacks both star forming and passive low mass galaxies.Recently Kulas et al. (2013) found that the metallic-ity of protocluster galaxies did not vary with galaxy mass,whereas field galaxy metallicity decreases with decreasingmass. They found no difference between the two environ-ments at high masses, but at low masses found a significantdifference in metallicity. This suggests that low mass galax-ies are more metal rich in protocluster environments thanin the field. This may also mean that the low mass galaxiesin protoclusters are dustier than those in the field. However,with our current data we find no evidence to suggest thisand it is difficult to test as we do not detect any low massgalaxies in the protocluster. c (cid:13) , 1–14 =2.5 protocluster associated with HzRG MRC 2104-242 Protoclusters at high redshift are not dynamically evolvedand so it is unlikely that large-scale mass segregation hashad enough time to occur: our 2.65 arcmin × × .
28 Mpc in physical coordi-nates. Assuming an average galaxy velocity of 500 km s − ,this gives a crossing time of 2 . z = 2 .
49, the ageof the Universe is 2 .
58 Gyr. This means that there has notbeen enough time for virialization to occur and any dynami-cal friction effects will not be strong enough to produce masssegregation in the protocluster at this redshift.Substructure has, however, been found around radiogalaxies at high redshift. Hayashi et al. (2012) reported thediscovery of a protocluster where there were three distinct“clumps” of galaxies on scales of ∼ −
10 Mpc. They foundthat the highest mass objects resided in the densest clumpat z = 2 .
53, suggesting that higher mass objects may pref-erentially form in denser environments. Kuiper et al. (2010)also found that the most massive and highly star forminggalaxies were located near the radio galaxy of a z ∼ In the previous subsections we have established that theMRC 2104 −
242 protocluster galaxy mass function differsfrom that of the control field for both star forming and pas-sive galaxies. A higher level of dust extinction in only the lowmass protocluster galaxies could produce this effect observa-tionally; with our current data we only find a 2 σ differencein the dust extinction between the protocluster and controlfield galaxies at high masses, and cannot test this at lowermasses. Alternatively, protocluster enivronments may formmore high mass galaxies through monolithic collapse or pro-tocluster galaxies may undergo many more mergers in theearly stages of their growth compared to the field. We findtentative evidence that the fraction of starburst galaxies ishigher in the protocluster, indicating a more rapid growthof galaxies in denser environments. We note that data fromKoyama et al. (2013a) also shows a similar lack of galaxieswith low masses in the MRC 1138 −
262 protocluster, how-ever with only two protoclusters it is difficult to come toany firm conclusions as to why we find this result. In futurestudies it is important that we now progress towards largersamples of protoclusters, in order to obtain a meaningfulstatistical understanding of the formation and evolution ofthese structures.
We have undertaken a NB survey of the field around theHzRG MRC 2104 − −
242 is overdensecompared to blank control fields, with a level of overdensityof 8 . ± . −
242 are moremassive and have more hidden star formation than controlfield galaxies at the same redshift. When we take a massselected field sample we find no difference in the SFR andsSFR between the two environments, and only a minor dif-ference in the dust content.(iii) Star formation at z ∼ . µ m and Herschel data wefind that the average SFR-mass relations are the same irre-spective of environment and both the protocluster and con-trol field galaxies lie close to the main sequence.(iv) We find a large difference in the mass distributions betweenenvironments: we expect to find ∼ M < M (cid:12) and detect none. Thiscould indicate a higher level of dust extinction in low massgalaxies in the protocluster. It may alternatively be due togalaxies in the protocluster forming more high mass galax-ies through monolithic collapse or undergoing many moremergers in the early stages of their growth.(v) We find tentative evidence of a larger fraction of starburstgalaxies in the protocluster than in the control field. Furtherdata is required to confirm the 250 µ m detections, howevera more rapid mode of star formation in denser environmentsmay explain how protocluster galaxies build up their massquicker than in the field.(vi) The overdensity we detect in this small area is highly de-pendent on the mass range we consider. It can range from anoverdensity of 0 (at M < M (cid:12) ) to 55 ( M > . M (cid:12) ).It is important when quantifying protoclusters to comparetheir mass functions, rather than simply number overdensi-ties. ACKNOWLEDGMENTS
We thank Bruce Sibthorpe for help with the Herschel datareduction and Dan Smith for useful discussions. We thankthe referee for their helpful suggestions which improvedthe paper. EAC acknowledges support from the STFC.NAH is supported by an STFC Rutherford Fellowship. SIMacknowledges the support of the STFC Studentship En-hancement Programme (STEP). EER acknowledges finan-cial support from NWO (grant number: NWO-TOP LOFAR614.001.006).The research in this paper is based largely on obser-vations made with ESO Telescopes at the La Silla ParanalObservatory under programme IDs 081.A-0673 and 088.A-0954. This work also made use of observations made withthe Spitzer Space Telescope, which is operated by the JetPropulsion Laboratory, California Institute of Technologyunder a contract with NASA, as well as observations ob-tained at the Gemini Observatory, programme GS-2010B-Q-65. The Gemini Observatory is operated by the Associa-tion of Universities for Research in Astronomy, Inc., under acooperative agreement with the NSF on behalf of the Gem-ini partnership: the National Science Foundation (United c (cid:13) , 1–14 E.A.Cooke et al.
REFERENCES
Bertin E., Arnouts S., 1996, A&AS , 117, 393Bruzual G., Charlot S., 2003, MNRAS , 344, 1000Bunker A. J., Warren S. J., Hewett P. C., Clements D. L.,1995, MNRAS , 273, 513Calzetti D., Armus L., Bohlin R. C., Kinney A. L., Koorn-neef J., Storchi-Bergmann T., 2000, ApJ , 533, 682Capak P. et al., 2007, ApJS , 172, 99Capak P. et al., 2011, ApJ , 730, 68Chabrier G., 2003, PASP , 115, 763Chiang Y.-K., Overzier R., Gebhardt K., 2013, ApJ , 779,127Colless M. et al., 2001, MNRAS , 328, 1039Croton D. J. et al., 2006, MNRAS , 365, 11Daddi E., Cimatti A., Renzini A., Fontana A., Mignoli M.,Pozzetti L., Tozzi P., Zamorani G., 2004, ApJ , 617, 746Daddi E. et al., 2007, ApJ , 670, 156Davis M., Efstathiou G., Frenk C. S., White S. D. M., 1985,ApJ , 292, 371De Lucia G., Blaizot J., 2007, MNRAS , 375, 2Donley J. L. et al., 2012, ApJ , 748, 142Eales S. et al., 2010, PASP , 122, 499Elbaz D. et al., 2011, A&A , 533, A119Fazio G. G. et al., 2004, ApJS , 154, 10Furusawa H. e. a., 2008, ApJS , 176, 1Galametz A. et al., 2012, ApJ , 749, 169Galametz A. et al., 2013, A&A , 559, A2Galametz A. et al., 2010, A&A , 522, A58Garn T., Best P. N., 2010, MNRAS , 409, 421Griffin M. J. et al., 2010, A&A , 518, L3Guo Q. et al., 2011, MNRAS , 413, 101Hartley W. G. et al., 2013, MNRAS , 431, 3045Hatch N. A. et al., 2011a, MNRAS , 410, 1537Hatch N. A., Kurk J. D., Pentericci L., Venemans B. P.,Kuiper E., Miley G. K., R¨ottgering H. J. A., 2011b, MN-RAS , 415, 2993Hayashi M., Kodama T., Tadaki K.-i., Koyama Y., TanakaI., 2012, ApJ , 757, 15Hook I. M., Jørgensen I., Allington-Smith J. R., DaviesR. L., Metcalfe N., Murowinski R. G., Crampton D., 2004,PASP , 116, 425Kennicutt, Jr. R. C., 1998, ARAA , 36, 189Kissler-Patig M. et al., 2008, A&A , 491, 941 Knebe A. et al., 2011, MNRAS , 415, 2293Koyama Y., Kodama T., Tadaki K.-i., Hayashi M., TanakaM., Smail I., Tanaka I., Kurk J., 2013a, MNRAS , 428,1551Koyama Y. et al., 2013b, MNRAS , 434, 423Kriek M., van Dokkum P. G., Labb´e I., Franx M., Illing-worth G. D., Marchesini D., Quadri R. F., 2009, ApJ ,700, 221Kuiper E. et al., 2010, MNRAS , 405, 969Kulas K. R. et al., 2013, ApJ , 774, 130Kurk J. D., Pentericci L., R¨ottgering H. J. A., Miley G. K.,2004, A&A , 428, 793McCarthy P. J., Kapahi V. K., van Breugel W., Subrah-manya C. R., 1990, AJ , 100, 1014McCracken H. J. et al., 2012, A&A , 544, A156Muldrew S. I., Pearce F. R., Power C., 2011, MNRAS ,410, 2617Muzzin A. et al., 2012, ApJ , 746, 188Papovich C., 2008, ApJ , 676, 206Peng Y.-j. et al., 2010, ApJ , 721, 193Planck Collaboration, 2013, ArXiv e-prints, astro-ph/1303.5062Retzlaff J., Rosati P., Dickinson M., Vandame B., Rit´e C.,Nonino M., Cesarsky C., GOODS Team, 2010, A&A , 511,A50+Rieke G. H., Alonso-Herrero A., Weiner B. J., P´erez-Gonz´alez P. G., Blaylock M., Donley J. L., Marcillac D.,2009, ApJ , 692, 556Rieke G. H. et al., 2004, ApJS , 154, 25Rigby E. E. et al., 2014, MNRAS , 437, 1882Rodighiero G. et al., 2011, ApJL , 739, L40Rujopakarn W., Rieke G. H., Weiner B. J., P´erez-Gonz´alezP., Rex M., Walth G. L., Kartaltepe J. S., 2013, ApJ , 767,73Santini P. et al., 2009, A&A , 504, 751Seymour N. et al., 2007, ApJS , 171, 353Shapley A. E., Steidel C. C., Erb D. K., Reddy N. A.,Adelberger K. L., Pettini M., Barmby P., Huang J., 2005,ApJ , 626, 698Siana B. et al., 2009, ApJ , 698, 1273Sobral D., Smail I., Best P. N., Geach J. E., Matsuda Y.,Stott J. P., Cirasuolo M., Kurk J., 2013, MNRAS , 428,1128Spergel D. N. et al., 2003, ApJS , 148, 175Springel V. et al., 2005, Nature , 435, 629Springel V., White S. D. M., Tormen G., Kauffmann G.,2001, MNRAS , 328, 726Steidel C. C., Adelberger K. L., Shapley A. E., Erb D. K.,Reddy N. A., Pettini M., 2005, ApJ , 626, 44Taylor M. B., 2005, in Astronomical Society of the PacificConference Series, Vol. 347, Astronomical Data AnalysisSoftware and Systems XIV, Shopbell P., Britton M., EbertR., eds., p. 29Vandame B., 2004, PhD Thesis, Nice University, FranceVenemans B. P. et al., 2007, A&A , 461, 823This paper has been typeset from a TEX/ L A TEX file preparedby the author. c (cid:13)000