The systematic search for $z\gtrsim5$ active galactic nuclei in the $Chandra$ Deep Field South

4.9 MB·English

by  Anna K. Weigel

Mon.Not.R.Astron.Soc.000,1–??(2013) Printed28January2015 (MNLATEXstylefilev2.2)
(cid:38)
The systematic search for z 5 active galactic nuclei
in the Chandra Deep Field South
Anna K. Weigel1?, Kevin Schawinski1, Ezequiel Treister2,
C. Megan Urry3, Michael Koss1, Benny Trakhtenbrot1
1Institute for Astronomy, Department of Physics, ETH Zurich, Wolfgang-Pauli-Strasse 27, CH-8093 Zurich, Switzerland
2Universidad de Concepción, Departamento de Astronomía, Casilla 160-C, Concepción, Chile
3Physics Department and Yale Center for Astronomy and Astrophysics, Yale University, New Haven, CT 06511, USA
5
1
0 ABSTRACT
2 Weinvestigateearlyblackhole(BH)growththroughthemethodicalsearch
n forz (cid:38)5AGNintheChandraDeepFieldSouth.Webaseoursearchonthe
a Chandra 4-Ms data with flux limits of 9.1 × 10−18 (soft, 0.5−2 keV) and
J 5.5 × 10−17 erg s−1 cm−2 (hard, 2−8 keV). At z ∼5 this corresponds to
6 luminosities as low as ∼1042 (∼1043) erg s−1 in the soft (hard) band and
2 shouldallowustodetectCompton-thinAGNwithM >107M andEd-
BH (cid:12)
dington ratios > 0.1. Our field (0.03 deg2) contains over 600 z ∼5 Lyman
]
Break Galaxies. Based on lower redshift relations we would expect ∼ 20
A
of them to host AGN. After combining the Chandra data with GOOD-
G
S/ACS,CANDELS/WFC3andSpitzer/IRACdata,thesampleconsistsof
. 58 high-redshift candidates. We run a photometric redshift code, stack the
h
GOODS/ACSdata,applycolourcriteriaandtheLymanBreakTechnique
p
- and use the X-ray Hardness Ratio. We combine our tests and using addi-
o tionaldatafindthatallsourcesaremostlikelyatlowredshift.Wealsofind
r five X-ray sources without a counterpart in the optical or infrared which
t
s might be spurious detections. We conclude that our field does not contain
a any convincing z (cid:38)5 AGN. Explanations for this result include a low BH
[
occupation fraction, a low AGN fraction, short, super-Eddington growth

1 modes, BH growth through BH-BH mergers or in optically faint galaxies.
v Bysearchingforz (cid:38)5AGNwearesettingthefoundationforconstraining
0 early BH growth and seed formation scenarios.
8
5 Key words: galaxies: active; X-rays: galaxies; galaxies: high-redshift;
6 (galaxies:) quasars: supermassive black holes;
0
.
1
0
5
1 INTRODUCTION sumedseedformationmodel,almostconstantEdding-
1
tonaccretionorevensuper-Eddingtonepisodesarere-
v: Massive black holes (BHs) with masses >106M(cid:12) re- quiredtomatchtheseobservations(Volonteri&Rees
i sideinmostgalaxies,includingourown(Genzeletal.
X 2005; Volonteri & Silk 2014; Alexander & Natarajan
1996;Magorrianetal.1998;Schödeletal.2003;Ghez
2014).Inourcurrentunderstanding,BHsgrewoutof
r etal.1998,2000,2008).LuminousquasarswithM
a ∼ 109M (Barth et al. 2003: z = 6.4, Willott etBaHl. ∼100−105M(cid:12) seedsbyaccretinginfallingmatteror
(cid:12) merging with a second BH (Rees & Volonteri 2007).
2003: z = 6.41, Trakhtenbrot et al. 2011: z ∼ 4.8)
Twoseedformationmodelsarecurrentlyfavored.One
havebeendetectedatz∼5−7(Fanetal.2000,2001:
scenario predicts that the remnants of massive Pop-
z∼6,Mortlocketal.2011:z=7.085).TheBHspow-
ulation III (Pop III) stars constitute BH progenitors
eringthesequasarsmustthereforebuilduptheirmass
(Madau & Rees 2001; Haiman & Loeb 2001; Bromm
in less than one billion years. Depending on the as-
et al. 2002; Alvarez et al. 2009; Johnson et al. 2012).
The second model is based upon the direct gravita-
tional collapse of massive gas clouds (Loeb & Rasio
? E-mail:[email protected]
(cid:13)c 2013RAS
2 Anna K. Weigel et al.
Treisteretal.(2013)usedtheChandra4-Msdata
in their search for high-redshift (z > 6) AGN. Us-
ing a sample of preselected z = 6−8 Lyman Break
0.5
dropoutandphotometricallyselectedsourcesfromthe
c
se HUDF and CANDELS, Treister et al. (2013) showed
c
ar that none of these sources are detected individually
in in the X-rays. Stacking the X-ray observations does
CS not produce a significant detection either. Treister
A
C et al. (2013) suggested different processes that could
E 0.0
D account for the lack of X-ray counterparts to these
¡
high-redshift sources. The sample could be contami-
a
ndr natedbyalargenumberoflow-redshiftinterlopers.A
a
Ch low BH occupation fraction could explain the lack of
C
E X-raycounterparts.ItispossiblethatBHgrowthonly
D
occurs in dusty and/or small galaxies which were not
−0.5
includedinthisanalysisbecausetheyliebelowthede-
tection threshold. The Treister et al. results can also
beexplainediflargeamountsofgasanddustobscure
−0.5 0.0 0.5 theX-rayemissionofactivelyaccretingBHs,asprevi-
RAChandra¡RAACS in arcsec ouslyproposedbyTreisteretal.(2009)andFioreetal.
(2009).Additionally,itispossiblethataccretionisnot
thedominantBHgrowthmodeintheearlyuniverse.If
Figure1.OffsetbetweenGOODS/ACSandChandrapo- BHsareprimarilygainingmassbymergingwithother
sitionsafterthecorrectionhasbeenapplied.Outofthe740 BHs, X-ray radiation might not probe BH activity.
ChandraX-raysources408possessanopticalcounterpart.
(SeeTreisteretal.2013foramoredetaileddiscussion
Wedetermineameanoffsetof(0.12800,-0.23700)forthese
ofthesepossiblescenarios.)Inadditiontothescenar-
408objectsandcorrectforitbyshiftingtheChandra po-
ios described above, short, super-Eddington growth
sitions.Thegreytriangleindicatesthemeandisplacement
episodes, as proposed by Madau et al. (2014) and
before the correction. The green star illustrates the mean
offsetafterourcorrection.Theblackpointsshowthecor- Volonteri & Silk (2014), also present a possible solu-
rectedobjectpositions.Thegreycircleillustratesanoffset tion. In comparison to constant Eddington accretion,
of0.500. the amount of matter that is accreted during these
super-Eddingtongrowthphasesisthesame.However,
Madau et al. (2014) showed that, if we allow super-
1994; Bromm et al. 2009; Volonteri 2010; Latif et al.
Eddington accretion, a duty cycle of 20% is enough
2013). Both models include uncertainties and predict
to grow a non-rotating 100 M seed BH into a 109
markedly different BH growth histories. More exotic (cid:12)
M objectbyz∼7.Onecouldimaginethattheseed
scenarios,includingBHseedformationviastellardy- (cid:12)
BH grows via five 20 Myr long m˙/m˙ = 4 growth
namical, rather than gas dynamical processes, have Edd
modes, each followed by a 100 Myr phase of quies-
also been suggested ( see Volonteri 2010; Bromm &
cence.Forshort,super-Eddingtongrowthepisodes,we
Yoshida 2011 and references therein). These scenar-
wouldthusexpecttofindfewerBHsthatareactively
ios primarily focus on reproducing the high-redshift
accretingat thesame time.TheTreisteret al.(2013)
quasar population. We must however also be able to
sample could therefore not contain any BHs that are
explaintheexistenceoflessmassiveandluminous,but
actively accreting at the time of observation.
more abundant BHs that we find in galaxies such as
InthisworkwecarefullyexaminetheChandra 4-
theMilkyWay(Treisteretal.2011;Volonteri2010).A
Ms catalog for possible z (cid:38) 5 AGN. We combine the
firststeptowardsconstrainingseedformationmodels
deep Chandra observations with optical and infrared
and determining if they are also valid for such ’nor-
data from GOODS, CANDELS and Spitzer. We use
mal’BHs,ismeasuringtheBHluminosityfunctionat
the Lyman Break Technique and a photometric red-
high redshift.
shift code to estimate the redshift of our targets. We
TheChandra 4-Mscatalog(Xueetal.2011)pro-
also use colour criteria, stacking and the X-ray Hard-
vides X-ray counts for the soft (0.5 keV - 2 keV),
nessRatio.IncontrasttoTreisteretal.2013,webase
hard (2 keV - 8keV) and full (0.5 keV - 8 keV) band
thisanalysisondetectedX-raysourcesinsteadoftry-
for the Chandra Deep Field South (CDF-S). The on-
ing to determine if a high-redshift object possesses a
axis flux limits lie at 9.1×10−18, 5.5×10−17 and
X-ray counterpart. We therefore also analyse X-ray
3.2×10−17erg s−1 cm−2 for the soft, hard and full
sources that would not be classified as Lyman Break
band,respectively.TheCDF-Scoversa0.11deg2area.
Galaxiesbecausetheyareheavilyobscuredintheop-
For our analysis we not only use the Chandra 4-Ms
tical and the infrared.
data,butalsorequirecoveragebytheCANDELSwide
and deep surveys. The effective area of our field is
hence 0.03 deg2 1.
http://archive.stsci.edu/pub/hlsp/candels/goods-s/
gs-tot/v1.0/hlsp_candels_hst_acs-wfc3_gs-tot_
1 README CANDELS GOODS-S Data Release v1.0 readme_v1.0.pdf
(cid:13)c 2013RAS,MNRAS000,1–??
z (cid:38) 5 AGN in the CDF-S 3
740objectsintheChandra
4Mssourcecatalog
Visualpreselection:Issourcecoveredby
No(366) Discardobject
B,V,i,z,J,H?
Yes(374)
Imagesintact? No(3) Discardobject
Yes(371)
Objectclearlyvisiblein
Yes(305) Discardobject
allbands(z < 4)?
No(66)
ClassificationaccordingtoLyman ’lowsignifi-
No(8)
BreakTechniquepossible? canceobjects’
Yes(58)
58objectsinmainsample
Photometric Colour-Colour-
StackingMethod HardnessRatio
redshiftcode Diagram
46HR>0
49sourcescannot
35z(cid:46)4, beclassified, (z<4.3)sources,
6z>5sources, 10HR(cid:54)0
52z(cid:54)5sources 513zz∼(cid:38)57,3sozur∼ces6, 22zz∼∼45,and (z(cid:62)4.3)sources,
2sourcescannot
5z∼6object
beclassified
zphot(6/58) zstacking(21/56) zcolo(u7r/−9c)olour HR(10/56)
5zcaonldouHr−Rco(cid:54)lo0uran>d5zsatnadckzinpgho>t>5? No(55) Discardobject
Yes(3)
additionaldata:VLT/Hawk-IHUGS
KS-band+HST/WFC3HUDFY-band
all3candidates:mostlikelylow-redshift
Figure 2. Flowchart illustrating the analysis steps that we take to determine possible z (cid:38) 5 candidates. The number of
objects that pass each step is given in brackets. After executing this analysis for all 740 Chandra sources, three z (cid:38) 5
candidatesremaininoursample.AdditionalKS-bandanddeepY-banddatadoeshowevershowthattheyaremostlikely
low-redshiftsources.
(cid:13)c 2013RAS,MNRAS000,1–??
4 Anna K. Weigel et al.
We know that the CDF-S contains hundreds of
well constrained z (cid:38) 5 Lyman Break Galaxies (see GOODS/ACS V, i, z CANDELS J, H
GOODS/ACS B CANDELS Y
e.g. Stark et al. 2009; Vanzella et al. 2009; Wilkins 27.6
etal.2010;Bouwensetal.2014;Duncanetal.2014).
Theseshouldallpassourmanualinspection,stacking,
colourcriteriaandphotometricredshiftmeasurement.
To be considered as a high-redshift AGN candidate, g
theymusthoweveralsobedetectedintheX-raysand e
d
pass our X-ray Hardness Ratio test. n
Volonteri&Begelman(2010)showedthattheex- i
C 27.8
pectednumberdensityofhigh-redshiftAGNdepends E
D
on the assumed seed formation model. Both, a detec-
tion and a non-detection, of high-redshift AGN gives
us a lower limit on this number density. Our search
thus constrains possible seed formation scenarios and
sheds light on BH growth modes.
Vito et al. (2013) searched for z >3 AGN in the
28.0
4-Ms CDF-S. They mainly analysed the evolution of
obscuration and AGN space density with redshift. In 53.0 53.2
contrast to this work, Vito et al. (2013) based their RA in deg
analysisonalreadyexistingphotometricandspectro-
scopic information on the Chandra sources. We com-
pare our results to Vito et al. (2013) in Section 5. Figure 3.OverviewoftheareacoveredbytheHST/ACS
This paper is organized as follows. Section 2 de- filters B (red), V, i, z (blue), the HST/WFC3 bands Y
(green),J,H (purple)andour740Chandra4-Mssources.
scribes the data that is used in this work. Sections
Grey points mark sources that are not of interest for this
3 and 4 introduce our redshift tests and illustrate
workbecausetheyarenotcoveredbyenoughbands(B,V,
the results of their combination. We conclude with
i,z,J,H)tousetheLymanBreakTechnique.Blackpoints
a discussion in Section 5 and a summary in section
indicateobjectswithenoughfiltercoveragethatwereelim-
6. Throughout this paper we assume a ΛCDM cos- inatedbecausetheobjectsareclearlyvisibleinallbands.
mology with h0 = 0.7, Ωm = 0.3 and ΩΛ = 0.7. All According to the Lyman Break Technique this indicates
magnitudesaregivenintheABsystem(Oke&Gunn z<4.Yellowstarsmarkthepositionsofthe58potential
1983). high-redshiftAGNthatweanalysemoreclosely.
dra and the GOODS/ACS catalog. We determine a
2 DATA mean offset of RA − RA = 0.12800 and
Chandra ACS
TheChandra4-MssourcecatalogbyXueetal.(2011) DECChandra −DECACS = −0.23700. We adjust the
is the starting point of this work. It contains 740 GOODS/ACS and CANDELS positions of each ob-
sourcesandprovidescountsandobservedframefluxes ject by subtracting the mean displacement from the
inthesoft(0.5keV-2keV),hard(2keV-8keV)and originally given catalog position.
full(0.5keV-8keV)band.AllobjectIDsusedinthis
work refer to the source numbers listed in the Xue
et al. (2011) Chandra 4-Ms catalog. We make use of
3 ANALYSIS
HST/ACSdatafromtheGOODS-southsurveyinthe
opticalwavelengthrange.Weusecatalogsandimages In the following section we describe in detail the set
for filters F435W (B), F606W (V), F775W (i) and of criteria we employed to the data in order to iden-
850LP(z)fromthesecondGOODS/ACSdatarelease tify z (cid:38) 5 candidates. We first exclude objects with
(v2.0) (Giavalisco et al. 2004). We use CANDELS insufficient filter coverage, perform our own aperture
WFC3/IR data from the first data release (v1.0) for photometry and determine the dropout band of each
passbands F105W (Y), F125W (J) and F160W (H) source by manual inspection. We run a photometric
(Groginetal.2011;Koekemoeretal.2011).Todeter- redshiftcode,stacktheGOODS/ACSdataandapply
mine which objects are red, dusty, low-redshift inter- colour criteria. In addition, we use the X-ray data as
lopers, we also include the 3.6 micron and 4.5 micron a photometric redshift indicator. We combine all red-
Spitzer IRAC channels. We use SIMPLE image data shift tests in section 4. Figure 2 illustrates and sum-
from the first data release (DR1, van Dokkum et al. marizes the complete analysis that is detailed in the
2005) and the first version of the extended SIMPLE following subsections.
catalog by Damen et al. (2011). Dahlen et al. (2010), McLure et al. (2011) and
When comparing Chandra, GOODS/ACS and Duncanetal.(2014)showedthatwhenselectinghigh-
CANDELSobjectpositions,aclearoffsetintheChan- redshift objects, a selection based only on colour cri-
dra coordinates is apparent. We illustrate this incon- teriaisnotasreliableascalculatingphotometricred-
sistencyinFigure1.Tocorrectforthisdiscrepancy,we shifts.Especiallyforfaint,lowsignal-to-noiseobjects,
calculate the mean displacement between the Chan- errors and upper limits can lead to scattering out of
(cid:13)c 2013RAS,MNRAS000,1–??
z (cid:38) 5 AGN in the CDF-S 5
byallotherGOODS/ACSandCANDELSfilters.The
Aperture radius: 0.6" opticalandinfraredcounterpartdetectionisprimarily
Aperture radius: 0.5"
based upon the H-band image since this is the deep-
27 Aperture radius: 0.75"
r) Aperture radius: 1" estband.TheH-bandimagesforobjects105and521
to show significant artifacts, we therefore discard them.
c
ra 26 Source 366 is eliminated due to the object’s position
t
x beingattheedgeoftheGOODS/ACSimages.Hence,
E
S 371 possible candidates remain after this first visual
(
B25 preselection.
A
m
e
d
u 24
t 3.2 Aperture Photometry
ni
g
a We perform our own aperture photometry on the
M
23 GOODS/ACS and CANDELS images to gain flux
values and to estimate parameters such as detection
threshold and aperture size. We compare our results
to the GOODS/ACS catalog (Figure 4).
23 24 25 26 27
To perform aperture photometry we use Source-
Magnitude m (GOODS)
AB Extractor(SExtractor,Bertin&Arnouts1996;Bertin
et al. 2002; Holwerda 2005). We determine the coun-
Figure 4. Comparison between our own photometry and terpart position in the H band in a first SExtractor
thefluxvaluesreportedintheGOODScatalog(Giavalisco
run.WethenrunSExtractorontheremainingoptical
etal.2004).WeshowtheABmagnitudevaluesforthez-
and infrared images to establish if the counterpart is
band. Note that the GOODS/ACS catalog only contains
present at the same location. The flux measurements
flux values for an aperture with a 0.70700 and not a 0.7500
arecarriedoutwithincircularapertureswithradiibe-
radius.Nonetheless,wecomparethesevaluestothebright-
nesswemeasuredwithina0.7500radiusaperture.Weusea tween0.300 and100.Wealtertheaperturesizeforfaint
small0.300radiusaperturefortwoobjectsonly.Sincethese sourcesandtopreventcontaminationthroughnearby
twosourcesarenotdetectedintheGOODS/ACScatalog, objects.Forsourceswithasignal-to-noiseratio<1we
theyarenotpartofthiscomparison. usethe1σ sensitivitylimitofthecorrespondingfilter
as an upper limit. The SExtractor parameter values
andaperturesizesaresummarizedinTable1andTa-
the colour selection region. Similar to colour criteria, ble 2. For the 3.6 micron and 4.5 micron bands we
photometricredshiftsstronglydependontheposition rely on the flux values reported in the SIMPLE cat-
oftheLymanBreak.Aphotometricredshiftcodedoes alog (Damen et al. 2011). We make use of the flux
however consider all filter information, including up- values reported for a 1.500 radius aperture. We show
perlimitsanderrors.Furthermore,low-redshiftinter- flux and error values for our main sample in Tables 5
lopers can be identified by including the filters red- and 6.
ward of the Lyman Break. Dahlen et al. (2010) il-
lustrated the discrepancy between colour criteria and
photometric redshifts for the GOODS-S field. Only 3.3 Lyman Break Technique and visual
50% of their photometrically selected z ∼ 4 sources classification
werealsoclassifiedasB-dropoutsaccordingtocolour
The Lyman Break Technique (Steidel et al. 1999; Gi-
criteria.Wethereforeprimarilyuseaphotometricred-
avalisco 2002; Dunlop 2013) employs the pronounced
shift code. The colour criteria, our visual classifica-
feature of the Lyman continuum discontinuity in the
tion, the stacking of the GOODS/ACS data and the
spectral energy distribution (SED) of young, star
X-ray Hardness Ratio provide additional redshift in-
forming galaxies. We use the common terminology of
dications.
referring to Lyman Break Galaxies as ’dropouts’. If a
sourceisnotdetectedintheB-bandoranybluerpass-
bands,butisvisibleinallredderfilters,thisindicates
3.1 Initial Sample Selection
z∼4andwerefertoitasa’B-dropout’.V-dropouts,
Ourinitialsampleconsistsofthe740objectsgivenin i-dropouts and z-dropouts correspond to redshifts of
theChandra4-Mssourcecatalog.Figure3showsthat ∼5, ∼6 and ∼7 respectively.
not all Chandra targets are covered by the GOOD- We classify the 371 possible candidates by eye
S/ACSandCANDELSimages.However,adjacentfil- accordingtotheirdropoutband.Ifanobjectisclearly
tercoverageisnecessaryfortheapplicationoftheLy- visible in all bands, this indicates z < 4. We exclude
manBreakTechnique.Wethereforenarrowthenum- suchsourcesfromoursample.Figure5illustratesthe
ber of possible candidates down to 374 by removing conditionssourceshavetofulfilltobeincludedinthe
sources that are not covered by B, V, i, z, J and further analysis.
H. The Y-band area is small compared to the other
filters. We therefore also include sources that are not Eightsourcesarenotclassifiedaccordingtotheir
coveredbytheY-bandprovidedthattheyarecovered dropout band. These objects are shown in Figures 6.
(cid:13)c 2013RAS,MNRAS000,1–??
6 Anna K. Weigel et al.
208 hard 208 soft 208 B 208 V 208 i 208 z 208 3.6 208 4.5
→eliminated: lack of CANDELS coverage
141 hard 141 soft 141 B 141 V 141 i 141 z 141 Y 141 J 141 H 141 3.6 141 4.5
→eliminated: source is clearly visible by eye in the optical and the infrared, this indicates z<4 according to
the Lyman Break Technique
184 hard 184 soft 184 B 184 V 184 i 184 z 184 Y 184 J 184 H 184 3.6 184 4.5
→kept in sample: by eye classified as a B-dropout, this indicates z∼4 according to the Lyman Break Technique
Figure 5. Classification examples. We only kept source 184 in our sample. The images are 1000 x 1000 in size and were
colourinverted.
Figure 7 shows the hard and soft band counts for all outthefactthatwemayhavefoundoneormorefalse
eightsourcesincomparisontotheentiresample.The detections.Ontheotherhand,wefindmanysourcesof
hard, soft and full band counts are given in Table 1. comparable X-ray brightness that do possess an opti-
caland/orinfraredcounterpartandthataredetected
in both bands (Figure 7). 333, 384 and 643 are also
Five (190, 280, 333, 384, 643) of these eight ob-
closetoabrightgalaxywhichmightbewhythecoun-
jects are especially interesting since they do not have
terpart remains undetected. This could indicate that
a counterpart in the optical or the infrared. We refer
at least some of the sources are real.
tothesesourcesas’low significance objects’.Weshow
these five objects in the upper panel of Figure 6. For
three(104,156,276)oftheseeightobjectsitisunclear TheLymanBreakTechniqueisnotapplicableto
whichsourcerepresentsthecounterpartintheoptical thelowsignificanceobjectsandweareunabletomea-
and infrared since multiple objects are visible in the sureaphotometricredshiftwithoutadetectioninthe
GOODS/ACS and CANDELS bands. These objects optical and the infrared. Since 333 is detected in the
are shown in the lower panel of Figure 6. hardandthesoftband,itistheonlyobjectforwhich
Onlythreeoftheeightobjects(333,156,276)are we can apply our Hardness Ratio test (see Section
detected in the full, the hard and the soft band. 104 3.7).WithHR=−0.01333couldbeapotentialhigh-
and 280 are detected in the soft band only. 384 and redshift AGN candidate. A negative Hardness Ratio
643 are found in the soft and in the full band. 190 is alone does however not convince us of 333 indeed be-
detected in the full band only. Only one of the low ingathigh-redshift.Atthispointwecanthusnotde-
significanceobjects(333)andoneofthethreesources termineifoursourcesarerealhigh-redshiftAGNcan-
withmultiplecounterpartsintheopticalandinfrared didates, false detections or low-redshift objects that
(156) show a signal-to-noise ratio (cid:62) 5 in at least one are optically faint. Hopefully, the forthcoming 7-Ms
ofthebands.Outofthe371objectswithenoughfilter observationsoftheCDF-S(PI:WilliamBrandt,Pro-
coverageandintactimages140(37.7%)haveasignal- posal ID: 15900132) will shed more light on our five
to-noise ratio (cid:62) 5 in the soft, hard or full band. See lowsignificanceobjects.Weeliminatealleightsources
Table3forthenumberofsourceswithsignal-to-noise from our sample. We stress that these targets could
ratios (cid:62)1 and (cid:62)5 in the individual bands. still be high-redshift AGN.
The objects for which we do not detect a coun-
terpart in the optical and the infrared could be spu- Afterdiscardingtargetsthatareclearlyvisiblein
rious detections. On the one hand, Xue et al. (2011) allbandsandeliminatingtheeightobjectsthatcould
report that, for the entire catalog, the probability of notbeclassifiedaccordingtotheLymanBreakTech-
a source not being real is < 0.004. The entire catalog nique, 58 B, V, i, z and Y-dropouts remain in our
doesthereforecontainuptothreespurioussourcesper main sample. For sources that we visually classify as
band. We do however only consider sources that are B-dropoutsthesignal-to-noiseintheV-bandmightbe
also covered by GOODS and CANDELS. The CAN- toolowforadetectionbySExtractor.Byeliminating
DELSwideanddeepsurveyfieldsare 0.03deg2insize visuallyclassifiedB-dropoutswecouldhencebemiss-
and hence only make up ∼ 27% of the CDF-S (0.11 ing objects that might be classified as z ∼ 5 sources
deg2)area.Accordingtothis,wewouldexpecttofind byotherredshifttests.WethereforekeepB-dropouts
∼0.8spuriousdetectionsperband.So,wecannotrule in our sample.
(cid:13)c 2013RAS,MNRAS000,1–??
z (cid:38) 5 AGN in the CDF-S 7
190 full 190 hard 190 soft 190 B 190 V 190 i 190 z 190 Y 190 J 190 H 190 3.6 190 4.5
280 full 280 hard 280 soft 280 B 280 V 280 i 280 z 280 Y 280 J 280 H 280 3.6 280 4.5
333 full 333 hard 333 soft 333 B 333 V 333 i 333 z 333 Y 333 J 333 H 333 3.6 333 4.5
384 full 384 hard 384 soft 384 B 384 V 384 i 384 z 384 Y 384 J 384 H 384 3.6 384 4.5
643 full 643 hard 643 soft 643 B 643 V 643 i 643 z 643 Y 643 J 643 H 643 3.6 643 4.5
104 full 104 hard 104 soft 104 B 104 V 104 i 104 z 104 J 104 H 104 3.6 104 4.5
156 full 156 hard 156 soft 156 B 156 V 156 i 156 z 156 J 156 H 156 3.6 156 4.5
276 full 276 hard 276 soft 276 B 276 V 276 i 276 z 276 Y 276 J 276 H 276 3.6 276 4.5
Figure 6.Sourcesthatcannotbeclassifiedaccordingtotheirdropoutband.Top:SourcesthatareintheChandra 4-Ms
catalog but do not show a clear counterpart in the optical and infrared, ’low significance objects’. 333, 384 and 643 are
close to a bright galaxy which is why we might not be able to detect the counterpart in the optical and infrared. Deeper
observations would be needed to detect possible counterparts. Bottom: Sources with multiple possible counterparts. Out
of these eight objects only 156, 276 and 333 are simultaneously detected in the hard and in the soft band. We are unable
togainredshiftestimatesfortheseeightobjectsandarethusunabletodetermineiftheyareloworhigh-redshiftsources
or if they might be spurious detections. The black circles are centered on the original Chandra position and illustrate the
positionaluncertaintygivenintheXueetal.(2011)catalog.Allimageswerecolourinvertedandare1000 x1000 insize.
3.4 Photometric Redshift measurements lowphotometricredshiftaremostlikelyindeednearby
objects, high photometric redshift results are less re-
Even though the Lyman Break Technique provides a
liable. This is mainly due to template incompleteness
fastandeasywayofidentifyingpossiblecandidates,it
and large photometric errors for faint high-redshift
isnotwithoutcaveats.Dustinred,low-redshiftgalax-
sources.
ies can produce a sharp break in the SED that might
bemistakenfortheLymanBreak(Dunlopetal.2007; The photometric redshifts we determine for our
McLureetal.2010;Finkelsteinetal.2012).Applying objects prove to be highly dependent on the filters
the Lyman Break Technique therefore only produces used as input. We use the photometric redshift code
a sample that may still contain low-redshift interlop- EAZY (Brammer et al. 2008). We apply the de-
ers. Yet, even the results of a photometric redshift fault template set by Brammer et al. (2008) which
codehavetobetreatedcarefully.Whilesourceswitha is part of the EAZY distribution. Five of these six
(cid:13)c 2013RAS,MNRAS000,1–??
8 Anna K. Weigel et al.
ID Hardcounts σHard SNRHard Softcounts σSoft SNRSoft Fullcounts σFull SNRFull
190 27.55 −1.00 − 17.29 −1.00 − 21.99 10.12 2.17
280 15.13 −1.00 − 9.36 4.94 1.89 18.36 −1.00 −
333 51.15 16.91 3.02 52.23 11.59 4.51 103.21 19.66 5.25
384 11.37 −1.00 − 7.25 4.44 1.63 14.98 −1.00 −
643 34.34 −1.00 − 27.16 8.72 3.11 33.54 13.21 2.54
104 49.95 −1.00 − 38.57 12.16 3.17 57.78 −1.00 −
156 65.01 13.43 4.84 61.59 10.59 5.82 126.33 16.40 7.70
276 22.17 9.54 2.32 29.55 7.95 3.72 51.61 11.70 4.41
Table 1.Hard,softandfullbandcountsforsourcesthatcannotbeclassifiedaccordingtotheirdropoutband.Werefer
to the top 5 sources as ’low significance objects’. For objects that are not detected we give an upper limit on the counts.
Wesetσto-1.00andmarkthesignal-to-noiseratiowithadash.AllvaluesweredirectlyextractedfromtheChandra4-Ms
catalog(Xueetal.2011).
1σ fluxlimit
3 104 inµJy/arcsec2
156 5
129706 H R = 0. R = 0.0 BV 44..613660××1100−−22
d band) 233683840343 H H R = - 0.5 iz 81..245857××1100−−22
r
a Table 2. GOODS/ACS flux limits. We determine these
h 2
s meansensitivitylimitsbymeasuringthemeanbackground
t
un fluxforsixdifferentobjects.Foreachobjectwedetermine
Co thebackgroundcountswithinfiveaperturesofvaryingsize.
g(
o
L
1
agesfor30ofour58objects.WeshowtheH bandand
Spitzer stamps for these sources in the appendix. We
do not perform aperture photometry on the Spitzer
1 2 3
images, but simply extract the 1.500 aperture radius
Log(Counts soft band)
flux values from the Spitzer catalog by Damen et al.
(2011).
Figure 7. Counts in the hard and in the soft band for
theChandra 4-Mssources.Wehighlighttheeightobjects If a source is not detected in a passband, we use
thatcouldnotbeclassifiedaccordingtotheLymanBreak the 1σ sensitivity limit of this filter as an upper limit
Technique. These eight sources might be real sources or intheEAZYrun.FortheGOODS/ACSbandsB,V,
spuriousdetections.Theblackpointsillustratethe58ob- i and z we determine the sensitivity limits by mea-
jectsthatareleftinoursample.Thegreypointsshowthe suring the mean background flux. We measure the
positionsoftheadditional502Chandra 4-Msobjectsthat number of background counts for six of our objects.
arealsoclassifiedasAGN.Theywereexcludedbecauseof
For each source, we determine the flux within aper-
insufficient filter coverage or quality or because they are
turesofvaryingsizeatfivedifferentpositions.Forthe
clearlow-redshiftdropouts.Xueetal.(2011)categorizeall
CANDELS bands Y, J and H we rely on the sensi-
oftheseeightobjectsandall58samplesourcesasAGN.
tivity limits reported in Grogin et al. (2011). Table 2
shows the flux limits that we used in our analysis.
EAZY templates were created by using the Blanton IncludingtheSpitzer dataturnsouttobecrucial
&Roweis(2007)algorithmtoreducethetemplateset for our purposes. As an example we show the pho-
byGrazianetal.(2006).Thesixthtemplatedescribes tometric redshift code results we get for source 244
a young starburst with a dust screen following the in Figure 8. By running EAZY without the Spitzer
Calzetti law with A = 2.75 (Calzetti et al. 2000). flux values we gain z = 6.54. Assuming that this
v phot
Due to large uncertainties in the luminosity function photometricredshiftvalueiscorrect,wedeterminean
of high-redshift AGN, we do not include a luminosity absolutemagnitudeofM =−24.63fortheH-band.
H
prior when running EAZY. To gain reliable redshift Bouwens et al. (2014) however give M∗ =−21.2 for
UV
valueswetaketheSpitzer/IRACimagesintoaccount. z ∼6 Lyman Break Galaxies. 244 would therefore be
WecarefullycomparetheH bandCANDELSimages verybrightifitindeedwasatz∼6.Includingthe3.6
totheSpitzer 3.6micronimagestodeterminewhichof and4.5micronSpitzer fluxvaluesinthephotometric
ourcandidatespossessesaSpitzer counterpartandfor redshiftcodeanalysisresultsinz =1.9.Including
phot
which sources the Spitzer flux values can not be used theSpitzer infrareddatahenceprovestobecrucialin
due to source confusion. We include the Spitzer im- revealing low-redshift interlopers.
(cid:13)c 2013RAS,MNRAS000,1–??
z (cid:38) 5 AGN in the CDF-S 9
14 20
H
J
Y
2m− 15 BV z 244, excluding Spitzer bands 10
c z =6.54
phot
i
1
−
s 16
g
r
e 14 2χ z
n 3.6 4.5 phot
i H
)ν Y J
F
ν 15 B 244, including Spitzer bands
V z
x 10
u z =1.9
Fl i phot
(
g 16
o
l
1.0 2.0 3.0 4.0 5.0 0 2 4 6 8 10
Wavelength λ in µm z
phot
Figure 8. Best fit SED for 244 determined by running EAZY without the Spitzer data (top) and with the Spitzer data
(bottom). If we only use the GOODS/ACS and CANDELS filters as input for EAZY, the photometric redshift code will
classify402asaz∼6.54source.EventhoughthefluxvaluesarefitwellbythisSED,theshapeoftheSEDseemsunphysical
foraz∼6.54source.Wewouldexpectthecontinuumfluxtobelowerandbluer.Acomparisontothez∼7UVluminosity
function by Bouwens et al. (2014) shows that 244 would indeed be bright if it was at z ∼ 6.54 (MH = −24.6). The χ2
distributionontherightshowsmultiplesecondarylowredshiftsolutions.SEDshapeandχ2 distributionthussuggestthat
244mightbealow-redshiftobject.InthebottompanelweillustratethebestfitSEDdeterminedbyincludingtheSpitzer
3.6 and 4.5 micron IRAC channels. 244 is now exposed as a low-redshift source. Based upon z ∼ 1.9 we reject 244 as a
possiblehigh-redshiftcandidate.Theshownlimitscorrespondtothe1σ sensitivitylimits.
3.5 Stacking GOODS/ACS data ton1992).Thesecolourcriteriadonotonlydependon
thepositionoftheLymanBreak,butalsousetheover-
We combine the GOODS/ACS images of each object
all SED shape. The criteria we apply here are based
intostacks.Wegeneratedthreestackspersource:(1)
on Vanzella et al. (2009) who used 114 Lyman Break
combinesB andV,(2)B,V andiand(3)B,V,i,z.
GalaxiesfromtheGOODSfield.Thesampleconsisted
Weexaminethesedeeperstacksfordetectionsbyrun-
of 51 z ∼ 4, 31 z ∼ 5 and 32 z ∼ 6 Lyman Break
ning SExtractor on them. We use a detection thresh-
Galaxies. These objects were first chosen by apply-
old of 1.5 σ (DETECT_TRESH = 1.5) and a mini-
ingtheLymanBreakTechniqueandthenfollowedup
mum detection area of 15 pixels above the threshold
spectroscopically.
(DETECT_MINAREA = 15). The remaining SEx-
Objects that fulfill the following condition are
tractor parameter values are left at their default val-
classified as z∼4 sources:
ues. A detection in the first stack indicates a source
that drops out before or in the B-band and therefore
impliesz(cid:46)4.Sourcesthataredetectedin(1),(2)and (B−V)(cid:62)(V−z)∧(B−V)(cid:62)1.1∧(V−z)(cid:54)1.6 (1)
(3) are hence of no interest to us. We assume z ∼ 5
Here ∧ and ∨ represent the logical ’and’ and ’or’ op-
if an object is detected in (2) and (3), but not in (1).
erators respectively. z ∼5 objects need to satisfy the
Only being detected in (3) indicates z ∼ 6. Finally,
constraints given here:
no detection in any stack signals z (cid:38) 7. Using this
approach we find 35 z (cid:46) 4, 5 z ∼ 5, 3 z ∼ 6 and
13 z (cid:38) 7 sources. The stacking analysis proves to be [(V −i)>1.5+0.9×(i−z)]∨[(V −i)>2.0]∧
(2)
inconclusive for two objects (303, 651). Source 303 is (V −i)(cid:62)1.2∧(i−z)(cid:54)1.3∧(S/N) <2
B
detected in (1) and (3), but not in (2). 651 is only
For z∼6 galaxies these conditions apply:
found in stack (2).
(i−z)>1.3∧[(S/N) <2∨(S/N) <2] (3)
B V
3.6 Colour criteria
To classify the 58 possible candidates according to
Weobtainanadditionalredshiftindicationbyapply- colour criteria, we use the magnitude values from
ing colour criteria based upon population synthesis our aperture photometry (see 3.2). If a source is not
models (Guhathakurta et al. 1990; Steidel & Hamil- detected, we use the 1σ sensitivity limit as an upper
(cid:13)c 2013RAS,MNRAS000,1–??
10 Anna K. Weigel et al.
limit. 49 of the 58 sources do not fulfill any colour
criteria. Note that a source that simultaneously
9
possesses upper limits in the B and V or V and z
band is not included in the z ∼ 4 diagram since its 8
position cannot be determined. The same applies
7
for the z ∼ 5 and z ∼ 6 diagrams. Furthermore, we
only use colour criteria to determine z ∼ 4, z ∼ 5 cts6
ac1lnadsshsziofiwe∼sd6athrdeerhoceponolcoueutsnr.-ocTtolhaoleulrn49edcoeiasbgsjarearcimtlsystahttahzta<tcai4nl.lunFsoitgtruabrteee er of obje45 28, 45566, 546, 591
thesecriteria.Basedoncolourcriteriawefind2z∼4 mb 93, 44, 44
(373, 444), 2 z ∼5 (303, 321) and 5 z ∼6 (226, 244, u3 58404462
296, 522, 589) objects. N 62 89, 73, 30, 21, 30
4 33436
2 2, 16, 1, 2, 1, 4,
9703007
3333435
1 83 22 200258, 48, 42, 28, 44, 26, 18, 518539
5 5 632323223645
4, 3, 6 7590, 3, 1, 1, 4, 1, 0, 9, 3, 6, 5, 0, 6 5
3.7 X-rays as a photometric redshift 0 35 17 45 21531946272122181215183055324129 62
indicator 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8
Hardness Ratio HR
The Hardness Ratio (HR), sometimes denoted as X-
raycolour,representsanadditionalindicatorforhigh-
Figure 10. Number of objects per HR bin. This figure
redshift AGN. The Hardness Ratio is defined as:
illustratesthedistributionofour54objectsintermsofHR.
H−S HR < 0 indicates that this source might be at z > 4.34.
HR= (4)
H+S A positive HR suggests that this source might be at z (cid:54)
4.34. Thus ten of our 54 objects might be at z > 4.34.
Here H and S represent the observed frame hard (2 -
Notshownareobjects496and578.Thesetwosourcesare
10keV)andsoft(0.5-2keV)bandcountsrespectively.
only detected in the full band and can therefore not be
Tozeroth-order,anAGNX-rayspectrumfollows constrained by the Hardness Ratio. The numbers in each
apowerlawwithanobscurationdependentturnoverat barcoincidewiththeobjectIDsinthecorrespondingHR
lower energies. In the Compton-thin regime a higher bin.
column density N means that, relative to the hard
H
band,wewilldetectfewercountsinthesoftband.Ina
galaxy’s restframe the X-ray spectrum hence appears zphabs*zpow model and assume a power law slope
harder for higher obscuration. With increasing red- of 1.8 (Turner et al. 1997; Tozzi et al. 2006). Fig-
shift the spectrum gets shifted to lower energies and ure 9 summarizes our results. The left panel illus-
thenumberofcountsthatweobserveinthehardand trates our model for N = 1022cm−2, 1022.5cm−2
H
in the soft band changes accordingly. We can there- and1023.5cm−2 (blue,yellowandgreen,respectively)
fore use the Hardness Ratio as an additional redshift at z = 0.1, 3 and 6 (dot dashed, solid and dashed
indicator. line, respectively). It is evident that the number of
A soft X-ray spectrum is then expected for counts in the soft and hard band changes with red-
Compton-thin (N <1024cm−2) objects. For sources shift.AfterincludingtheChandraRedistributionMa-
H
in which 1024 (cid:46) N H(cid:46)1025cm−2, i.e. transmission- trix (RMF) and Auxiliary Response Files (ARF) for
H
dominatedCompton-thickAGN,thedirectemissionis on-axissources,wemeasurethespectralcountsinthe
stillvisible,althoughtheE<10keVradiationiscom- hard and in the soft band and determine the HR.
pletely obscured by photoelectric absorption, while The right panel of Figure 9 shows our results. Ac-
thedetectedemissionathigherenergiesisreducedby cording to our simulations HR > 0 signifies z < 4.34
Comptonscattering(Comastrietal.2010;Murphy& for sources with N up to 1023.5cm−2. Allowing for
H
Yaqoob 2009). Hence, in these cases we expect to ob- a small amount of transmission, e.g. 1% by using a
serveahardX-rayspectrumevenforsourcesatz>5. zpcfabs*zpow model does not change our results sig-
When N > 1025cm−2, i.e. for reflection-dominated nificantly (HR = 0 for z = 4.3). In our analysis we
H
Compton-thick AGN, we only observe the small frac- hence discard objects with HR > 0. Figure 10, which
tion of the initial emission which is reflected off the illustrates our results, shows that based solely on the
accretion disk or the obscuring material (Ajello 2009; HR, ten of our 58 candidates might be at z > 4.34.
Brightman & Nandra 2012). We can hence not make For496and583theHRcannotbeusedtoconstrain
the same assumptions as for Compton-thin sources. z since they are only detected in the full band. Ta-
Nevertheless,itissafetoassumethatallz(cid:38)5sources ble4summarizestheX-raycounts,signal-to-noisera-
that can be detected individually in the X-rays are tiosandHardnessRatiosforeachofourmainsample
Compton-thin objects. Compton-thick z (cid:38) 5 AGN sources.
would simply be too faint to be detected individually Our results are in good agreement with a similar
and are therefore not part of our sample. analysisbyWangetal.(2004).TheyalsousedXspec
To quantify the HR(z) relation we use the X-ray (Arnaud 1996) to simulate the HR at different red-
spectral fitting tool Xspec to simulate X-ray spec- shifts.Yet,Wangetal.(2004)choseapowerlawslope
tra at different redshifts (Arnaud 1996). We use a of1.9andusedawabsmodel.Ouranalysisshowsthat
(cid:13)c 2013RAS,MNRAS000,1–??