Strong Spatial Clustering of UV-selected Galaxies with Magnitude Ks<20.5 and Redshift Z~2

0.17 MB·English

by  Kurt L. Adelberger

Received 2004September30;Accepted 2005January7
PreprinttypesetusingLATEXstyleemulateapjv.11/26/03
STRONG SPATIAL CLUSTERING OF UV-SELECTED GALAXIES WITH MAGNITUDE K <20.5 AND
S
REDSHIFT Z ∼21
Kurt L. Adelberger2
CarnegieObservatories,813SantaBarbaraSt.,Pasadena, CA,91101
Dawn K. Erb, Charles C. Steidel, and Naveen A. Reddy
PalomarObservatory,Caltech105–24, Pasadena,CA91125
Max Pettini
5
InstituteofAstronomy,MadingleyRoad,CambridgeCB30HA,UK
0
0
2 Alice E. Shapley3

n UniversityofCalifornia,DepartmentofAstronomy,601CampbellHall,Berkeley,CA94720
a Received 2004 September 30; Accepted 2005 January 7
J
ABSTRACT
7
1 We obtained deep 8.5′×8.5′ near-infrared images within four high-redshift survey fields, measured
the Ks magnitudes of 300 optically selected galaxies with spectroscopic redshift 1.8∼< z ∼< 2.6, and
1 compared the spatial clustering strength of galaxies with K < 20.5 and K > 20.5. We found at
s s
v greater than 95% confidence that the brighter galaxies cluster more strongly. The best-fit correlation
4 lengths for the bright and faint samples are 10±3 and 4±0.8h−1 comoving Mpc, respectively (1σ),
5 although the unusual density of bright QSOs in one of our survey fields may imply that the result is
3
not representative of the universe as a whole. Neglecting this possibility, the correlation length for
1
the optically selected sample with K < 20.5 agrees well with that reported for comparably bright
0 s
near-IR-selectedsamples. Thedifferencesincorrelationlengthbetweenopticallyselectedandnear-IR-
5
selectedsampleshavebeenpresentedasevidencethatthe twotechniquesfindorthogonalpopulations
0
of high-redshift galaxies. Our results favor a more nuanced view.
/
h Subject headings: galaxies: high-redshift — cosmology: large-scale structure of the universe
p
-
o
r 1. INTRODUCTION thedifferencesinmeanstellarmassesandstar-formation
t rates might simply reflect the fact that near-IR surveys
s Near-infrared (IR) surveys of the high-redshift uni-
tendto be shallowerthanopticalsurveys,detectingonly
a verse (Cimatti et al. 2002, Franx et al. 2003, Abra-
v: ham et al. 2004) were designed to find passively evolv- thebrightesttipoftheluminositydistribution. Sincethe
near-IR-selectedstar-forming galaxies are less numerous
i inggalaxiesthathadbeenmissedbyopticalsurveys,but
X than optically selected galaxies, a significant fraction of
they also uncovered an intriguing population of galaxies
ar whose star-formation rates M˙∗ ∼> 200M⊙ yr−1 and stel- tinhgemthceoumldeabnepprroepseernttieisnoofpotipctailcasulrsvoeuyrscewsitihnoaunt cahpapnrge--
lar masses M∗ ∼> 2×1011M⊙ dwarfed those of optically- ciable way. The unusual measured properties of bright
selected galaxies. Since these star-forming galaxies have
near-IRgalaxies do not imply that they are absent from
the properties expected for the progenitors of giant el-
optical surveys any more than the high mean income of
lipticalgalaxies,manyresearchershavebeguntoassume
Luxembourgersimpliesthattheyareapreviouslyundis-
that near-IR surveys find the progenitors of ellipticals,
covered population in the E. U. Second, sub-millimeter
that optical surveys find the progenitors of less massive
surveysshow thatthe majorityof the mostrapidly star-
local galaxies, and (by extension) that the high-redshift
forming galaxies at high redshift are bright enough and
universe is divided into two distinct star-forming popu-
blue enough to be included in existing optical surveys
lations with divergent evolutionary paths.
(e.g., Chapman et al. 2005; see their Figure 6). Finally,
Althoughsome ofthe near-IR-selectedgalaxiesclearly
there are physical reasons to believe that at least some
donotsatisfycommonopticalselectioncriteria(see,e.g.,
of the elliptical galaxies in the local universe descended
Daddi et al. 2004; Forster-Schreiber et al. 2004; van
from the objects detected in optical surveys. Ellipticals
Dokkum et al. 2004), the evidence for two distinct star-
are the only local galaxies that have the spatial distri-
forming populations still leaves room for doubt. First,
butionexpectedfor the descendants ofopticallyselected
star-forming galaxies at z > 1.5, for example (see Adel-
1 Based, inpart, ondata obtained atthe W.M.Keck Observa-
berger et al. 2005).
tory,whichisoperatedasascientificpartnershipbetweentheCal-
ifornia Institute of Technology, the University of California, and Partly in order to help clarify the connection between
NASA, and was made possible by the generous financial support opticalandnear-IRpopulations,ourgrouphasobtained
oftheW.M.KeckFoundation.
K images throughout the course of its optical surveys
2 CarnegieFellow s
3 MillerFellow at 1.4 < z < 3.5. Recent analyses of the Ks images
2 Clustering of UV-selected Galaxies at z ∼2
(Erb et al. 2005, in preparation) reveal that many
optically-selected high-redshift galaxies have the mag-
nitudes K < 20 characteristic of the supposedly dis-
s
tinct bright near-IR-selected populations. Shapley et al.
(2004)showedthattheseoptically-selectedgalaxieswith
K < 20 have star-formation rates, stellar masses, and
s
metallicities similar to those of star-forming galaxies in
near-IR-selected surveys to K = 20. This suggested
s
that there might be substantial overlap between near-
IRandopticalpopulationsathighredshift, aconclusion
reinforced by their estimate that the number density of
high-redshift galaxies with K <20 in optically-selected
s
surveysisatleasthalfaslargeasthe numberdensityre-
ported by near-IR surveys. Daddi et al. (2004, 2005)
reached a similar conclusion from a different starting
point.
The present paper extends the work of Shapley et
al. (2004) in a small way by showing that optically-
selected galaxies at z ∼ 2 with bright K magnitudes
s
clusterasstronglyastheirnear-IR-selectedcounterparts.
Fig. 1.— Redshiftdistributionsforthegalaxiesinoursamples.
This suggests that the K -bright star-forming galaxies
s The top panel shows the raw number of galaxies with Ks ≤ 20.5
detected in optical surveys may not be enormously dif- andKs>20.5(Vega system). Thebottom panel shows thesame
ferent from the K -bright star-forming galaxies that are data normalized to constant area for z > 1. The overall redshift
s
missed. We will discuss the implications further in § 5, distributionfor the “BX”survey of Steidel et al. (2004) is shown
as a smooth curve in the background. The low redshift tail with
afterfirstdescribingourdatain§2,ourmethodformea- z<1isexcludedfromtheclusteringanalysisofthispaper.
suring r in § 3, and our raw results in § 4.
0
2. DATA
Table1. Observedfields
The galaxies we studied lie within the fields GOODS-
N,Q1623,Q1700,andQ2343ofthespectroscopicsurvey Name ∆Ωopta ∆ΩIRb NiImRgc NdIRetd N<IR20.5e
of Steidel et al. (2004). Steidel et al. (2004) obtained
GOODS-N 14.0×13.4 8.7×8.0 84 60 10
U GR images of these fields, constructed a catalog of
n Q1623 16.1×11.6 8.7×8.8 121 98 30
objects with AB-magnitude R ≤ 25.5 whose colors sat- Q1700 11.5×11.0 8.5×7.8 62 57 13
isfied the z ∼ 2 “BX” selection criteria of Adelberger et Q2343 22.5×8.5 8.7×5.4 101 85 22
al. (2004), measured spectroscopic redshifts for several total 695 259 368 300 75
hundredofthe objects inthis catalog,and, afterexclud-
ingobjectswithz <1,foundthatthemeanredshiftand aArea imaged in UnGR and observed spectroscopically, square
arcmin
r.m.s. dispersionz¯±σz =2.21±0.34agreedwellwiththe bArea imaged in Ks and observed spectroscopically, square ar-
range targeted by the photometric selection criteria. To cmin
measure the near-IR magnitudes of some of the galaxies cNumber of sources in the BX catalog that fall within the Ks
imageandhavespectroscopic redshiftsz>1.
in the catalog, we (Erb et al. 2005, in preparation) sub-
dNumber of sources in the BX catalog that are detected in the
sequently used the Wide-Field InfraredCamera (WIRC; Ksimageat3σconfidenceandhavespectroscopicredshiftsz>1.
Wilsonetal. 2003)onthe Hale5mtelescopeto obtaina eNumber of sources in the BX catalog with Ks ≤ 20.5 (Vega
deepK image(∼11hrintegrationtime,5σpoint-source system)andspectroscopicredshiftsz>1.
s
detection threshold of K ≃22.3 in the Vega system) of
s
a roughly 8.5′ ×8.5′ region within the larger optically-
observed area in each of the four fields. A total of 1598 servedBX candidates that have 1.8<z <2.8 is roughly
objects in the photometric BX catalogs of Steidel et al. 0.8, 0.65, 0.5 for K > 20.5, K ≤ 20.5, K ≤ 20, we
s s s
(2004)fellwithinthe areasobservedwithWIRC.We re- estimatefromtheapparentmagnitudedistributionofall
strict our analysis to the 368 of these sources that have BX candidates that the total number of BX candidates
measured redshifts. Most (300) of the sources with red- in our fields with 1.8 < z < 2.8 is 1000, 200, 80 for
shifts were detected in K at 3σ significance or better, the same three magnitude ranges. Although our spec-
s
and 75 were brighter than the K =20.5 threshold that troscopy is far from complete, our spectroscopic catalog
s
we use below to divide our sample into bright and faint of galaxies with Ks < 20.5 is already about an order of
subsamples. Seetable1;allnear-IRmagnitudeshereand magnitude larger than any near-IR selected catalog in
elsewhere are in the Vega system, and all optical magni- the same redshift range.
tudes are AB. K = 20.5 was chosen because it was the
brightest threshosld that left us with enough objects in 3. METHODS
the bright subsample; only 20 objects satisfy the thresh- We estimated the correlation lengths of the samples
old K = 20. The redshift distributions for the bright with two approaches. Both were based on counting
s
and faint subsamples are shown in figure 1. the observed number of galaxy pairs with comoving ra-
The total number of galaxies in our subsamples would dial separation |∆Z| < ℓ and comparing to the num-
be significantly higher if our spectroscopy were 100% ber expected for a reduced correlation function of the
complete. Since the fraction of spectroscopically ob- form ξ(r) = (r/r )−γ. Letting Z denote the comov-
0 i
Adelberger et al. 3
ing distance to redshift z and adopting the shorthand The estimate of r from this approach is barely affected
i 0
Z ≡Z −Z , we counted the observednumber of pairs by even major changes in data quality from one field to
ij i j
intheeachfieldwithradialseparation0≤|∆Z|<ℓand the next.
summedoverfields togetN (0,ℓ),the totalnumberof Although this insensitivity is a significant advantage,
obs
pairs with 0≤|∆Z|<ℓ. the disadvantage of the second approach is that it can
If the selection function P(z) is accurately known, the be noisy. We adopt both approaches because one is less
expected number of pairs with 0 ≤ |∆Z| < ℓ can be susceptibletosystematicerrorsandtheotheris lesssus-
estimatedforanyassumedvalueofr . Supposeweknow ceptible to random. See Adelberger (2005) for further
0
thatthejthgalaxyinafieldhasredshiftz andthatthe discussion.
j
ith galaxy lies at an angular separation of θ . Then In both approaches we adopt ℓ = 20h−1 Mpc, consid-
ij
the probability that the unknown redshift z will satisfy erably largerthan the biggestplausible peculiar velocity
i
0≤|Z |<ℓ is (Adelberger 2005) or redshift measurement error,and assume that the cor-
ij
relationfunctionhas the slopeofγ =1.6measuredfrom
P(z )[2ℓg−1(z )+aI]
P(0≤|Z |<ℓ|z θ )= j j (1) a larger sample of similar galaxies by Adelberger et al.
ij j ij
1+aP(zj) (2005). Adopting γ = 1.8 instead would lower our de-
where a ≡ rγ[f(z )θ ]1−γg−1(z )β(γ), g(z)≡ c/H(z) is rived correlation lengths by less than 10%. Eliminating
the change i0n comj ovijing distancje with redshift, f(z) ≡ pairs with θij > 300′′ (e.g., Adelberger 2005) does not
(1+ z)D (z) is the change in comoving distance with significantly alter the results.
A
angle, D (z) is the angular diameter distance, β(γ) ≡
A 4. RESULTS
B[1/2,(γ−1)/2],Bisthebetafunctionintheconvention
of Presset al. (1992),and I is relatedto the incomplete Equations 2 and 3 lead to the estimates r0 = 3.8,
beta function Ix of Press et al. (1992) through I ≡ 4.4h−1 comovingMpc forthe subsample with Ks >20.5
Ix[1/2,(γ −1)/2] with x ≡ ℓ2/{ℓ2 +[f(zj)θij]2}. The and and r0 = 9.9, 10.8h−1 comoving Mpc for the sub-
expected number of pairs with 0 ≤ |∆Z|< ℓ is equal to sample with Ks ≤ 20.5. To calculate the significance of
the sum of the probabilities P(0 ≤ |Z | < ℓ|z θ ) for theobserveddifferencebetweenthetwosamples’correla-
ij j j
each unique pair, tionlengths,we splitourfull 368-objectsample into two
partsatrandom,ratherthanaccordingtoK magnitude,
pairs s
1 measuredthedifferenceinr betweenthetwoparts,then
N = P(0≤|Z |<ℓ|z θ ), (2) 0
exp 2 ij j ij repeated the process over and over,with a different ran-
Xi6=j dom splitting each time, to estimate the distribution of
wherethesumincludesonlypairsinwhichbothgalaxies ∆r under the nullhypothesisthatr is unrelatedto K
0 0 s
lie in the same field. magnitude. Each random splitting placed 75 of our 368
Ourfirstapproachtowardsestimating r isto find the objects in one subsample and 293 in another,mimicking
0
valuethatmakestheexpectednumberofpairsN (0,ℓ) the 75/293 division of the actual splitting at K =20.5.
exp s
equalthe observednumber N (0,ℓ). By summing over We observed a difference in r as large as the observed
obs 0
the pair counts in all fields when calculating N and difference in fewer than 5% of the randomized catalogs,
exp
N , we areimplicitly assumingthat the selectionfunc- implyingthattheobserveddifferenceissignificantatbet-
obs
tionP(z)doesnotvaryappreciablyfromonefieldtothe ter than 95% confidence.
next. The estimated values of r will be incorrect if this We used two approaches to estimate the size of the
0
is not the case, but the analysis is otherwise insensitive uncertainties in the correlation lengths for the two sam-
to variations in data quality from field to field. ples. First,wemeasuredthedispersioninthethebest-fit
Even if the selection function is reasonably constant valueofr fromeachindividualfield. Althoughinprinci-
0
among the fields, as it probably is, the inferred value ple this empiricalapproachshouldworkwell,in practice
of r will still depend sensitively on its assumed shape r ispoorlydeterminedinindividualfields andthe field-
0 0
(Adelberger 2005). This is the major weakness of equa- to-field dispersion in r is poorly determined since there
0
tion 2. An alternative is to remove the sensitivity to the are only four fields.
selection function by dividing the number of pairs with Second, we broke our observed bright and faint cata-
0≤∆Z <ℓbythenumberwith0≤∆Z <2ℓ. Asshown logs into many smaller sub-catalogs by rejecting a ran-
byAdelbergeretal. (2005),theexpectationvalueofthis domfractionp=0.5–0.9ofthesources,observedhowthe
quotient is dispersioninthebest-fitr valueamongthesub-catalogs
0
depended on the number of objects in the catalog, and
ℓ
Nobs(0,ℓ) ≃ i6=j 0 dZ[1+ξ(Rij,Z)] , (3) extrapolated to the full catalog sizes.
*Nobs(0,2ℓ)+ Pi6=j R02ℓdZ[1+ξ(Rij,Z)] tioWneinaoduorptfincaolmepstriommaitseesvfaolrutehsefo1rσtchoenfistdaenndcaeridntdeervvaials-
where ξ(R,Z) ≡ ξ[(PR2 +RZ2)1/2] and Rij ≡ (1 + for the two analyses (eq. 2 and 3) of the two catalogs,
z )D (z )θ , regardless of the selection function, as r =3.8±0.8,4.4±0.8h−1 comovingMpc forK >20.5
i A i ij 0 s
long as the angular diameter distance changes slowly andr =10±3,11±8h−1 comovingMpcforK ≤20.5.
0 s
with redshift, ℓ is significantly larger than the comov- The consistency of the answers for the two approaches
ing uncertainty in a galaxy’s radial position, 2ℓ is sig- suggests that systematic problems are minimal. We will
nificantly smaller than the selection function’s width, adoptthe the results from eq. 2 for the remainder of the
and N (0,2ℓ) is large enough that h1/N (0,2ℓ)i ≃ paper, since that equation is subject to smaller random
obs obs
1/hN (0,2ℓ)i. Oursecondapproachto estimatingr is uncertainties.
obs 0
to findthe value thatmakesthe right-handside ofequa- Toverifythatthehighvalueofr forthebrightsample
0
tion3equaltheobservedquantityN (0,ℓ)/N (0,2ℓ). reflectedgenuinelarge-scaleclustering,ratherthan(say)
obs obs
4 Clustering of UV-selected Galaxies at z ∼2
seem to be a fundamentally different population. The
left panel of figure 2 shows,for example, that their large
correlationlengthsreflecttheculminationofatrendthat
isapparentamongthefaintergalaxiesthatdominateop-
tical surveys. Although existing optical surveys do not
detect every star-forming galaxy with K < 20.5, the
s
fraction they find appears to be reasonably representa-
tive of the whole.
Since K and R − K are correlated in our sample,
s s
the relationship between r and K implies relationship
0 s
between r and R − K . We find that r increases
0 s 0
from 4.0±0.8h−1 Mpc at R(AB)−K (Vega) < 3.5 to
s
r =9±3h−1 Mpc at R−K >3.5. A correlationof r
0 s 0
with both K and R−K could result from an under-
s s
lying correlation between r and stellar mass M , since
0 ∗
R−K andK arebothcorrelatedwithM (Shapley et
s s ∗
al. 2005,inpreparation). Anunderlyingcorrelationwith
M couldalsoexplainwhyr variessostronglywithK .
∗ 0 s
Spitzer Space Telescope observations analyzed by Shap-
ley et al. (2005, in preparation) show that mass-to-light
Fig. 2.— Left panel: correlation lengths for sub-samples with
ratios in our sample increase significantly with K lumi-
Vega-magnitude Ks ≤ 20.5, 20.5 < Ks ≤ 21.5, and Ks > 21.5. s
Thenumberofgalaxiesinthethreesubsamplesis75,145,and148, nosity, which implies that M∗ is a strong function of Ks
respectively. Thevalues shown arefromequation 2. Largecircles (see the right panel of figure 2). The very high stellar
are for the full sample; their uncertainties in r0 were estimated masses associated with K <20.5 can only be produced
with the approach of § 4. Small squares and crosses show the s
in unusually massive potential wells that formed unusu-
resultsforQ1623andfortheotherthreefields,respectively. Right
panel: Estimatedstellarmassvs. Ksmagnitudeforgalaxiesinthe ally early, and these are exactly the potential wells that
sample of Steidel et al. (2005). Small circles show the results for ought to be most strongly clustered. The increase of
individualgalaxies. Largecircleswitherrorbarsshowthemean± r with K -luminosity may simply reflect the fact that
r.m.s. forgalaxiesindifferentmagnitudebins. 0 s
galaxiesinproto-clusterenvironmentsformedearlierand
were consequently able to convert more of their baryons
into stars by z ∼2. Steidel et al. (2005, in preparation)
the presence of merging sub-units within individual ha- discuss the point in more detail.
los,werepeatedtheanalysisafterexcludingallpairswith The strong dependence of galaxy properties on K
θij <60′′ (i.e., projected separation ∼< 1.1h−1 comoving magnitudecanbeviewedintwoways. Ontheonehands,
Mpc) from the calculation in equations 2 and 3. The sincegalaxieswithKs ∼< 20arenotrepresentativeofthe
best-fit values of r0 were not significantly altered. high-redshift population as a whole, conclusions from
bright near-IR-selected surveys probably apply to only
5. DISCUSSION
alimited subsetofgalaxies. These galaxiesarenotirrel-
Our analysis ofthe largestexisting spectroscopic sam- evant, since they are plausible progenitors for the rarest
ple ofgalaxieswith Ks <20.5atredshift z ∼2 confirms and most massive ellipticals in the local universe, but
previous reports (Daddi et al. 2004) that these galax- their properties should not be ascribed to the dominant
ies are strongly clustered. Their estimated correlation fainterpopulation. Ontheotherhand,theextremeprop-
lengthofr0 =10±3h−1Mpcsignificantlyexceedsthatof erties of galaxies with Ks ∼< 20 may help reveal trends
normal galaxies at high or low redshift (e.g., Adelberger that are difficult to discern in fainter samples, and these
etal. 2005,Budavarietal. 2003). Theexactvalueofthe trends could provide useful insight into the physics of
correlation length at Ks < 20.5 may be more uncertain galaxy formation. Exploring the second possibility will
than this number implies, since the Daddi et al. (2004) require extensive catalogs of high redshift galaxies with
sample is so small(see Adelberger 2005)andsince much Ks ∼< 20. The ability of optical surveys to find them in
of our signal comes from a field (Q1623; see figure 2) large numbers is surely good news.
whichcontainsanunusualconcentrationofbrightQSOs
at2.2∼< z ∼< 2.64, but the increase of clustering strength
with K luminosity has been established with high sig-
s
nificance (>95%; see § 4). We thank the referee, A. Cimatti, for two detailed
The results of this paper, coupled with those of Shap- reports. KLA and AES were supported by fellowships
ley et al. (2004), suggest that star-forming galaxies in fromtheCarnegieInstituteofWashingtonandtheMiller
optically-selectedandnear-IR-selectedhigh-redshiftsur- Foundation. DKE, NAR, and CCS were supported by
veys have similar properties when the surveys are re- grant AST 03-07263 from the National Science Founda-
strictedtoacommonmagnitudelimitKs ∼< 20.5. Galax- tion and by a grant from the Packard Foundation. Sec-
ieswithK <20.5undoubtedlymakeupalargerfraction tion 3 was written in Port of Spain’s Normandie Hotel,
s
of the sources in the near-IR surveys, but they do not which KLA thanks for remarkable hospitality.
Adelberger et al. 5
4 These QSOsarenot inthe BXsample. The sampleanalyzed
may not be a fair sample of the universe, since one of our fields
here contains only 3 QSOs, and their clustering does not signifi-
containsoneoftherichestknownconcentrations ofbrightQSOs.
cantly affect the results. The point is that our surveyed volume
REFERENCES
Adelberger, K.L., Steidel, C.C., Shapley, A.E., Hunt, M.P., Erb, Forster-Schreiber,N.M.etal.2004,ApJ,616,40
D.K.,Reddy,N.A.,&Pettini,M.2004,ApJ,607,226 Franx,M.etal.2003,ApJ,587,L79
Adelberger,K.L.,Steidel,C.C.,Pettini,M.,Shapley,A.E.,Reddy, Press, W. H., Flannery, B. P., Teukolsky, S. A., & Vetterling, W.
N.A.,&Erb,D.K.2005,ApJ,inpress T. 1992, “Numerical Recipes in C”, (Cambridge: Cambridge
Adelberger,K.L.2005,ApJ,inpress UniversityPress)
Budava´ri,T.etal.2003,ApJ,595,59
Shapley,A.E.,Erb,D.K.,Pettini,M.,Steidel,C.C.,&Adelberger,
Chapman, S.C.,Blain, A.W.,Smail,I., & Ivison, R.J.2005, ApJ,
K.L.2004,ApJ,612,108
inpress
Steidel, C.C., Adelberger, K.L., Shapley, A.E., Pettini, M.,
Cimatti,G.etal.2002,A&A,392,395
Daddi, E.,Ro¨ttgering, H.J.A.,Labb´e, I., Rudnick, G.,Franx, M., Dickinson,M.,&Giavalisco,M.2003,ApJ,592,728
Moorwood, A.F.M., Rix, H.W., van den Werf, P.P., & van Steidel, C.C., Shapley, A.E., Pettini, M., Adelberger, K.L., Erb,
Dokkum,P.G.2003,ApJ,588,50 D.K.,Reddy,N.A.,&Hunt,M.P.2004, ApJ,604,534
Daddi, E., Cimatti, A., Renzini, A., Fontana, A., Mignoli, M., vanDokkum,P.G.etal.2004, ApJ,611,703
Pozzetti, L.,Tozzi,P.,&Zamorani,G.2004,ApJ,617,746 Wilson,J.C.etal.2003,SPIE,4841,451