Differential Morphology Between Rest-frame Optical and UV Emission from 1.5 < z < 3 Star-forming Galaxies

0.38 MB·English

by  N. A. Bond

Draftversion January10,2011
PreprinttypesetusingLATEXstyleemulateapjv.11/10/09
DIFFERENTIAL MORPHOLOGY BETWEEN REST-FRAME OPTICAL AND UV EMISSION FROM
1.5<z <3 STAR-FORMING GALAXIES
Nicholas A. Bond, Eric Gawiser
PhysicsandAstronomyDepartment, RutgersUniversity,Piscataway, NJ08854-8019, U.S.A.
and
Anton M. Koekemoer
SpaceTelescopeScienceInstitute, 3700SanMartinDrive,Baltimore,MD21218,U.S.A.
1 Draft version January 10, 2011
1
0 ABSTRACT
2 We present the results of a comparative study of the rest-frame optical and rest-frame ultraviolet
n morphological properties of 117 star-forming galaxies (SFGs), including BX, BzK, and Lyman break
a galaxies with B < 24.5, and 15 passive galaxies in the region covered by the Wide Field Camera 3
J EarlyReleaseScienceprogram. Using the internalcolordispersion(ICD)diagnostic,we findthatthe
6 morphologicaldifferencesbetweentherest-frameopticalandrest-frameUVlightdistributionsin1.4<
z < 2.9 SFGs are typically small (ICD ∼ 0.02). However, the majority are non-zero (56% at > 3σ)

] and largerthan we find in passive galaxiesat 1.4<z <2, for which the weightedmean ICD is 0.013.
O The lack of morphological variation between individual rest-frame ultraviolet bandpasses in z ∼ 3.2
C galaxiesargues againstlargeICDs being causedby non-uniformdust distributions. Furthermore, the
absenceofa correlationbetweenICDandgalaxyUV-opticalcolorsuggeststhatthe non-zeroICDs in
.
h SFGs are produced by spatially distinct stellar populations with different ages. The SFGs with the
p largest ICDs (&0.05) generally have complex morphologies that are both extended and asymmetric,
- suggesting that they are mergers-in-progress or very large galaxies in the act of formation. We also
o
find a correlation between half-light radius and internal color dispersion, a fact that is not reflected
r
t by the difference in half-light radii between bandpasses. In general, we find that it is better to use
s
a diagnostics like the ICD to measure the morphologicalproperties of the difference image than it is to
[ measure the difference in morphologicalproperties between bandpasses.

Subject headings: cosmology: observations — galaxies: formation – galaxies: high-redshift – galaxies:
2
structure
v
5
2 1. INTRODUCTION 2005),Ginicoefficients(Lotz et al.2006),andS´ersicpro-
5 file fitting (e.g. Cassata et al. 2005), among others. At
The Hubble Sequence of galaxies (Hubble 1936)
1 is seen out to z ∼ 1.5 (Glazebrook et al. 1995; 2.z .3.5, star-forming galaxy (SFG) sizes range from
0. van den Bergh et al. 1996; Griffiths et al. 1996; <1kpcto∼5kpc,withthelargestoftenexhibitingmul-
1 Brinchmann et al. 1998; Lilly et al. 1998; Simard et al. tiple photometric components (e.g. Bouwens et al. 2004;
0 Ravindranath et al. 2006; Oesch et al. 2009). Morpho-
1999; van Dokkum et al. 2000; Stanford et al. 2004;
1 logicalanalyseshaverevealedthatmostofthesesystems
Ravindranath et al. 2004) beyond which high-redshift
: are disturbed and disk-like (i.e., with exponential light
v galaxies typically appear compact and irregular (e.g.
profiles), with only ∼ 30% having light profiles consis-
i Giavalisco et al. 1996; Lowenthal et al. 1997; Dickinson
X tent with galactic spheroids (e.g., Ferguson et al. 2004;
2000; van den Bergh 2001). Galaxies at high redshift
Lotz et al. 2006; Ravindranath et al. 2006; Petty et al.
r were initially identified using the Lyman-break tech-
a 2009). Because high-redshift galaxies are typically faint
nique (Steidel et al. 1996), wherein z > 2.5 galaxies
andclumpy/irregular,itcanbedifficulttoquantifytheir
are selected using a flux discontinuity in the continuum
morphology in a meaningful way. Using a sample of 97
(e.g. a U-band dropout) caused by absorption from
Lyman-α emitters, Bond et al. (2009) found that even
intervening neutral hydrogen (e.g. Rafelski et al. 2009;
the half-light radius could not be accurately measured
Yan et al. 2009; Oesch et al. 2010; Bunker et al. 2009;
without S/N &30 within a 0′.′6 aperture.
Hathi et al. 2010). Subsequent techniques, such as BzK
In 2003, Papovichet al. (2003, P03) introduced a dif-
(Daddi et al. 2004) and ‘BX’ (Adelberger et al. 2004)
ferential morphological diagnostic known as the inter-
color selection, as well as narrow-band selection of Lyα
nal color dispersion (ICD) that could be used to quan-
emitters (Hu & McMahon 1996), have allowed for the
tify the morphological differences between bandpasses
photometric identification of star-forming galaxies over
for astronomicalobjects. When applied to galaxies, this
a wider redshift range.
proved to be particularly useful for comparing the rest-
The morphologicalproperties of high-redshift galaxies
frame ultraviolet light distribution, which often appears
arequantifiedusingawiderangeoftechniques,including
clumpy even in low-redshift galaxies, to the rest-frame
theconcentrationandasymettryindices(Conselice et al.
optical light distribution. If the young stars in a galaxy
are distributed differently from the old stars (as is usu-
[email protected]
[email protected] ally the case at low redshift), then that galaxy will
[email protected]
2 Bond et al.
have a large ICD. Similarly, inhomogeneous dust dis-
tributions can lead to spatial variations in the level of TABLE 1
GalaxySampleProperties
obscuration of rest-frame UV light, leading to a large
ICD. Their initial study focused on low-redshift galax-
Sample Number RedshiftRange Reference
ies and foundthat mid-type spirals exhibited the largest
ICDs (ξ ∼ 0.2), irregular galaxies were intermediate
(ξ ∼ 0.1), and early-type galaxies had negligible mor- SFGmain 100 1.4<z<2.9 1
SFGhighz 17 2.9<z<3.5 1
phologicalvariationbetweenbands(ξ ∼0). Theauthors PassGal 15 1.4<z<2.0 2
also concluded that ICDs could be reliably measuredfor
high-redshift galaxies so long as the angular resolution References. — (1) Balestra et al. 2010; (2)
(& 0.5 beam kpc−1≃ 0′.′1 at z = 2) and signal-to-noise Cameronetal. 2010
(S/N &80) were sufficient. 2.2. Wide Field Camera 3 Imaging
Followinguponthisexploratorystudy,Papovich et al.
Inordertoprobetherest-frameopticallightofz ∼2.3
(2005)(P05) computed the ICDs of separate flux-limited
galaxysamplesatz ∼1andz ∼2.3usingacombination galaxies, we use HST/WFC3 F160W images (hereafter
of Hubble Space Telescope (HST) Wide-Field Planetary designatedH160images)takenaspartoftheWFC3ERS.
TheH imagingconsistedof20orbitsovertenvisitsto
Camera 2 and Near Infrared Camera and Multi-Object 160
the GOODS-S field (Program 11563), leading to a total
Spectrometerimaging. TheauthorsfoundthattheICDs
of z ∼ 1 galaxies were generally larger and more var- of 60 exposures. The exposures were then reduced and
calibrated (Koekemoer et al. 2010, in prep.), incorpo-
iedthanthoseoftheirhigher-redshiftcounterparts,with
rating SPARS100 dark frames and correcting for resid-
the former sample including many spiral galaxies and
ual gain differences between the mosaic quadrants. This
the latter being dominated by irregularsand interacting
also included the removal of large-scale scattered light
systems. Theyinterpretthesechangesasbeingduetoin-
residuals and satellite trails. The exposures were then
creasingstellar populationheterogeneityat low redshift,
combinedusingMultiDrizzle(Koekemoer et al.2002)us-
with gaseous disks forming around older spheroids. In
ing a pixfrac of 0.8 and a square kernel, to produce fi-
addition, they speculated that the high-redshift galaxies
nal drizzled images with a pixel scale of 60 mas. For
with largeICDs were actually mergers-in-progressbased
′′
a 1.2 fixed aperture, a typical H -band, 5σ detection
upon their large apparent asymmetries. 160
limit is m = 27.2. In order to ensure accurate align-
With the installation of the Wide Field Camera 3 AB
(WFC3) on HST, we now have a new, improved win- ment of the IR imaging with the GOODS ACS imaging,
the WFC3 exposures were individually aligned to the
dow into the rest-frame optical light of galaxies at
1 . z . 3. The deep NIR images being taken as GOODS-S v2.0 z-band catalog and to each other. Even
for single exposures, the number of sources matched to
part of the WFC3 Hubble Ultra Deep Field 2009 (GO
the reference catalog was ∼ 500 and the final astromet-
11563: PI Illingworth)andWFC3 EarlyReleaseScience
ric solution appears to be robust to ∼10 mas or better.
(ERS, Windhorst et al. 2010) programs allow for a de-
Furtherdetailsonthisreductionarepresentedinaforth-
tailed study of high-redshift galaxy morphologies down
to H ∼26 mag. In this paper, we present the results coming paper (Koekemoer et al. 2010, in prep.)
160
ofacomparativestudyoftherest-frameopticalandrest-
2.3. Galaxy Sample
frame ultraviolet morphological properties of 132 SFGs
in the WFC3 ERS region of GOODS-S, using the ICD Our galaxy sample is broken down into three subsam-
as our primary diagnostic. Throughout we will use AB ples: 1.4 < z < 2.9 SFGs (SFGmain), 2.9 < z < 3.5
magnitudes and assume a concordance cosmology with SFGs (SFGhighz), and 1.4 < z < 2 passive galaxies
H = 71 km s−1 Mpc−1, Ω = 0.27, and Ω = 0.73 (PassGal). The passive galaxy subsample includes 15
0 m Λ
(Spergel et al. 2007). With these values, 1′′ =8.2 physi- objects found in the WFC3 ERS region using a YHVz
cal kpc at z =2.5. color selection technique (Cameron et al. 2010). No fur-
ther cuts were made on this subsample.
2. DATAANDSAMPLE
TheSFGsubsamplesaredrawnfromtheVLT/VIMOS
Our morphological analyses are performed on star- spectroscopic observations described in Popesso et al.
forming and passive galaxies at 1.5 < z < 3.6 using (2009) and Balestra et al. (2010). The VLT/VIMOS
HST/ACS and HST/WFC3 imaging in the GOODS-S pointings targeted high-redshift galaxies in GOODS-S
field. Thedataproductsandgalaxysamplearedescribed with B < 24.5, including U-band dropouts, BzK color-
below. selected objects (Daddi et al. 2004), sub-U-dropouts
(similar to ’BX’ selection, Adelberger et al. 2004),
2.1. Advanced Camera for Surveys Imaging
and X-ray sources in the Giacconi et al. (2002) and
In the Chandra Deep Field-South, the southern half Lehmer et al. (2005) catalogs. For this analysis, we ex-
of the GOODS survey (Giavalisco et al. 2004) covers ∼ clude objects with redshift quality flag C (poor qual-
160arcmin2ofskyandhasHST/ACSobservationsinthe ity) and objects satisfying the BzK color selection for
F435W, F606W, F775W, and F850LP filters (hereafter passively-evolving galaxies. In order to exclude unob-
designated B , V , I , and z ). For this study, scured AGN from our subsample of SFGs, we remove
435 606 775 850
weuse only the I image,whichwasmultidrizzledto a anyobjecttargetedasanX-raysourceby Popesso et al.
775
′′
pixelscaleof60mas. The effectiveexposuretime ofthis (2009), as well as any object that is within 2 of an
′′
survey is variable across the GOODS area, but for 1.2 X-ray source in the expanded catalogs of Virani et al.
fixed aperture, a typical I -band, 5σ detection limit is (2006) and Luo et al. (2008).
775
m =27.4. Of the remaining SFGs in the WFC3 ERS region, 100
AB
Differential Morphologies of 1.5<z <3 SFGs 3
TABLE 2
GalaxySamples
Number Sample α δ z
447 SFGmain 3:32:01.117 −27:44:04.830 2.6355
455 SFGmain 3:32:01.302 −27:42:43.973 2.2998
462 SFGmain 3:32:01.509 −27:44:22.574 2.5119
479 SFGmain 3:32:02.067 −27:45:13.878 2.3131
482 SFGhighz 3:32:02.167 −27:42:30.428 3.2630
* Thistableisonlyastub. Amanuscriptwithcompletetables
isavailableathttp://www.nicholasbond.com/Bond1004.pdf
lie within 1.4 < z < 2.9 and are assigned to SFGmain,
while a further 17 objects are within 2.9 < z < 3.6 and
are assigned to SFGhighz. This separation is motivated
by P03, who found that the internal color dispersion is
most effective at identifying the signatures of heteroge-
neous stellar populations when the two filters cover op-
posite sides of the Balmer and 4000 ˚A breaks. Since we
are using images taken with the H and I filters on
160 775
HST, the light from young stars will dominate both im-
ages for galaxies at z &2.9 (that is, those in SFGhighz)
and we only expect to see a large ICD in one of these
galaxies if there is a large column of inhomogeneously
distributed dust.
The galaxy subsamples are described in Table 1. For Fig.1.—Internalcolordispersionasafunctionofcentroidoffset
more detail on the selection of these subsamples, see the between I775 and H160-band images for a series of Monte Carlo
references givenin the table. The sky positions and red- simulations of an SFG in the WFC3 ERS, where each point is
computedwitharandomrealizationoftheskynoiseintheimages.
shiftsofindividualobjectsinourparentsamplearelisted
The dashed line indicates an iterative quadratic fit to the data
inTable2,orderedbycatalognumber. Inthispaper,we (discardingoutliers).
will refer to SFGs according to their line number in ver-
sion 2.0 of the VIMOS master catalog (Balestra et al.
2010), excepting the passive galaxies, which will be de-
notedP1-P15. AlthoughitislistedinTable2,weexclude
′′
contamination by interloping sources within the 1.2
object 1379 from the majority of our analysis because it
apertures to be ∼ 17% based on the background sky
was not detected in either the H image or the I
160 775
density of sources with H < 26. However, because
image. 160
the SFGs in our sample were selected using ground-
3. METHODOLOGY based photometry, the contamination rate is probably
much lower because an interloper within ∼ 1′′ of a
In the following section,we describle ourmethodology
high-redshift galaxy would have blended with it and
forthemeasurementoffixed-aperturequantities,includ-
changed its apparent color. Since these “contaminated”
ing half-light radius and light centroid, as well as the
high-redshift galaxies will sometimes fall outside of the
internal color dispersion, a differential diagnostic of the
high-redshift galaxy color selection regions, they are
morphological differences between two bandpasses. Our
less likely to be selected than their uncontaminated
analysisisbasedupontechniquespresentedinP03,P05,
counterparts.
and Bond et al. (2009).
3.1. Fixed-aperture magnitudes and half-light radii 3.2. Internal color dispersion
Because star-forming galaxies are often clumpy, con- The internal color dispersion attempts to quantify the
sistofmultiplecomponents,andmaybeinteractingwith differences in the distribution of an object’s light be-
nearbygalaxies,there is some ambiguity in choosing the tween two observed bandpasses without making model-
appropriateapertureformorphologicalanalysis. Wewill dependent assumptions. When comparing between the
usea simpleapproachinthis paper,measuringallquan- rest-UV and rest-optical, large ICDs can arise if the
tities(withtheexceptionoftheinternalcolordispersion, young stars in a galaxy are distributed differently from
′′
see Section 3.2) within a fixed 1.2 aperture. the old stars or the dust is inhomogeneously distributed
Our methodology for measuring object magnitudes, within the galaxy, leading to spatial variations in the
centroids,and fixed-aperture half-lightradii is described level of obscuration of rest-frame UV light. At low red-
indetailinBond et al.(2009). Allofthesemeasurements shift, P03 found that irregular galaxies have ξ ∼ 0.1,
are performed on the original drizzled images, prior to while mid-type spiral galaxies have ξ ∼ 0.2. High-
the point spread function (PSF) convolution described redshift star-forming galaxies are morphologically sim-
below. ilar to irregular galaxies, but have had less time to form
Using Source Extraction software (SExtractor, previous generations of stars, so the majority would be
Bertin & Arnouts 1996), we estimate the maximum expected to have ξ <0.1.
4 Bond et al.
For a noise-free image, the ICD is given by,
N
(I −αI −β)2
P 2,i 1,i
ξ(I ,I )= i=0 , (1)
1 2 N
(I −β)2
P 2,i
i=0
where the sum is performed over N pixels in a chosen
aperture, I and I are the fluxes in pixel i, α is the
1,i 2,i
ratio of the total fluxes between cutout 2 and cutout 1,
andβisthedifferenceinthebackgroundbetweenthetwo
cutouts. Since we perform SExtractor sky subtraction
on all cutouts prior to analysis (see Bond et al. 2009),
we will assume β =0 for all of our measurements.
If we wish to accurately compute a differential mor-
phological diagnostic between two observed bandpasses,
we must ensure that (1) the effective PSF is the same
in the two images and (2) there is no astrometric offset.
If the PSF is well approximated by a Gaussian in both
frames, the former requirement can be satisfied by con-
volving the lower-resolutionimage with a Gaussian that
hasσ =pσ22−σ12. Unfortunately,thePSFoftheWFC3
H -band imaging is not well approximatedby a Gaus-
160
sian(Bond et al. 2007), so we must convolveeachimage
withthePSFoftheother. WeestimatedthePSFofeach
image using a stack of stars from Altmann et al. (2006)
Fig. 2.—Ratiobetweentheuncertaintyininternalcolordisper-
that fell within the WFC3 ERS region and had temper-
sion in our Monte Carlo simulations and the uncertainty as de-
atures consistent with K and M-type stars, a regime in terminedbytheanalyticalapproximationofP03(seeEquation4)
which photometric confusion with galaxies is minimal. for the objects in our z ∼ 2.5 SFG subsample. All points have
We further restrictedourstackto 29 starsthatwereiso- σξm,sims/σξ,P05 > 1 as a result of correlated noise in the drizzled
lated (i.e., the only photometric component within 0′.′6), HST images, but the analytical approximation becomes particu-
unambiguously stellar (SExtractor stellarity >0.9), and larlybadathighsignal-to-noise(i.e. smallσξ).
unsaturated. Experimentswithstarsofavarietyofspec- onanobject-by-objectbasis basedupon a visualinspec-
tral types reveal that the internal color dispersion is in- tion of the cutouts.
sensitivetothespectralshapeofthePSFacrosstheH
160
filter, with ICDs varying by < 10% of the typical ICD 3.3. Monte Carlo Simulations
uncertainty in SFGmain.
For this study, we estimate the uncertainties on the
Noisecontributionsfromtheskybackgroundinimages
internal color dispersion measurements for individual
of high redshift galaxies will affect the terms in Equa-
galaxies in our sample using object-by-object Monte
tion 1. In order to estimate the amplitude of this con-
tribution, we extracted 400 5′′×5′′ cutouts from random Carlo simulations. In order to do this, we first ex-
tractedthegalaxiesinoursamplefromtheirH cutout
positionsontheH andI images. Cutoutsweredis- 160
160 775 using SExtractor (DETECT MINAREA = 5 and DE-
carded if they contained any detected sources. We then
TECT THRESH =1.65). Mock H cutouts were then
computed a noise term, 160
created by adding the extracted light profile to cutouts
N N centeredatrandompositionsontheH image(seeSec-
160
ξn(α,N)=Median(XI22,i)−2αMedian(XI1,iI2,i) tion 3.2). We used the same H160-extracted galaxylight
distributions to generate mock I cutouts, normaliz-
i=0 i=0 775
ingthe lightdistributionsuchthatthe detectedflux was
N
+α2Median(XI12,i), ethqeuaolbtjoectth’satIin t−heHrealcIo7l7o5r.imFaignea.llyT,hwiseaacdcdoeudntnsofioser
775 160
i=0
realizations from the I image.
(2) 775
Theothersignificantsourceofuncertaintyinourmea-
where the medians were computed over the set of blank surements is the unknown astrometric shift between the
cutouts. This was then subtracted from the numera- WFC3 and ACS imaging. In order to simulate this, we
tor and denominator of Equation 1 when computing the appliedrandomtwo-dimensionalshiftstothemockH
160
ICDs of the objects in our sample. cutouts, with a mean amplitude of 10 mas (our astro-
′′ ′′
We calculated the ICDs using 0.6 to 1.5 circular metric uncertainties, see Section 2.2).
apertures on 5′′×5′′ cutouts in the HST/WFC3 and Oursimulationsaredesignedto approximatethe scat-
HST/ACS images, each centered at the flux-weighted ter induced by the sky background on ICD measure-
′′
centroid of all SExtractor detections within 1.2 of the ments of an objectwith an intrinsic ξ =0 and will allow
object positions reported in Balestra et al. (2010) and us to distinguish objects with intrinsic non-zero ICDs.
Cameron et al. (2010). The sizes of the apertures used Systematic sources of error, such as astrometric offsets
tocomputetheICDsforindividualgalaxieswereselected andnon-uniformskybackgrounds,onlyproducepositive
Differential Morphologies of 1.5<z <3 SFGs 5
scatter,sothe errordistribution is non-Gaussianand we
estimatedthe bias andpositive scatterusing the median
and third quartile, where we define,
σm =1.496[Q3(ξ)−Median(ξ)]. (3)
ξ
We performeda total of400 simulations for eachobject.
Figure 1 shows the dependence of the ICD on the rel-
ative astrometric shift between I and H for one of
775 160
our simulations. The dashed line indicates an iterative
quadraticfittothedata(discardingoutliers). Eachmea-
surement includes random realizations of the sky noise,
leading to the scatter seen about the best-fit curve. For
the objects in our sample, the median astrometric shift
at which the resulting systematic error in the ICD ex-
ceeds the random ICD uncertainty is ∼ 0.64 pixels, al-
though the precise behavior depends upon the shape of
the light distribution. Point sources, for example, will
exhibit more sensitivity to astrometric errors than well-
resolved galaxies.
InFigure2,wecomparethesimulateduncertaintiesfor
the SFGs in our sample to those that would be derived
from the analytical approximationof P03,
N
2/N (B2 +α2B2 )
p pix P 2,i 1,i
δ(ξ)= i=0 , (4)
N N
(I −β)2− (B −αB )2
P 2,i P 2,i 1,i
i=0 i=0
where B and B are the fluxes in pixel i in a cutout
1,i 2,i
extracted from a blank region of sky. As the expres-
sion shows, large ICD uncertainties are generally found
when a source is detected at low signal-to-noise in one
ofthe images. Athighredshift, the morphologicalvaria-
tions that we see between wavebands are small, even for
galaxieswith a lotofdynamicalactivity,so sourceswith
individual detections that have S/N & 30 can still have
uncertainties on their ICD comparable to or larger than
themedianICDinoursample(ξ ∼0.02). Astheauthors
of P03 note, their expression will underestimate the un-
certaintiesinthepresenceofcorrelatednoise. SinceHST
images are typically drizzled to correct for oversampling
ofthe PSF,oursimulationsdoinfactyieldlargeruncer-
tainties than predicted by P03. P05 accounted for this
discrepancy in drizzled images by correcting the uncer-
tainty by a single empirically-determined multiplicative
factor, C(ξ). Although C(ξ) = 6 would be a reasonable
approximation at σξ &0.01, it would underestimate the Fig.3.—Internalcolordispersion(top)andICDsignal-to-noise
uncertainty for some of the brighter objects in our sam- (bottom) as a function of redshift. We indicate 1.4 < z < 2.9
ple (σ . 0.01) by factors up to ∼ 3. Hereafter, we SFGs with triangles, 2.9<z <3.5 SFGs with open squares, and
ξ,P05 1.4< z <2 passive galaxies with stars. All error bars are drawn
will only use the uncertainties derived from our Monte fromourMonteCarlosimulationsandgalaxieswithσm>0.05are
ξ
Carlo simulations.
notplotted. Inthetoppanel,thedottedlineindicatesξ=0.01and
inthe bottom panel, the two dotted linesindicate an ICD signal-
4. RESULTS to-noiseof1and3. Objectswithξ(F775W,F160W)/σm >3show
ξ
Belowwepresentthe resultsofamorphologicalanaly- statisticallysignificantmorphological differences between theI775
sis of the high-redshift star-forming and passive galaxies andH160-bandimages. Whilethemajorityofstar-forminggalaxies
at1.4<z<2.9lieabovethisthreshold,veryfewpassivegalaxies
in our sample. All measured quantities, including cen- or z >3 galaxies have ξ >0 at the depth and resolution of these
troids, magnitudes, half-light radii, and ICDs, of these images.
galaxies are compiled in Table 3.
are in SFGmain except for object 863,which was identi-
4.1. Internal color dispersion fiedasaLyαemitter(LAE)atz =3.1inGronwall et al.
(2007). While it is possible that the extended V-band
4.1.1. Star-forming galaxies at 1.4<z <2.9
emission seen in the cutout originates from the same
SomeexampleI andH cutoutsfor15SFGswith high-redshift source as the Lyα emission, the large ICD,
775 160
ξ >0.05areshowninFigure8. Ofthegalaxiesshown,all distinctdisk-likemorphology,andlargesizecomparedto
6 Bond et al.
the majorityof LAEsatthat redshift(Bond et al.2009) at the ∼ 2σ (ξ ≃ 0.3) level. The I light distribution
775
suggest that it may be a low-redshift galaxy superim- of P14 is also faint (σm = 0.07), but the ICD is greater
ξ
posed on a high-redshift LAE. The remaining galaxies than zero at ∼ 5σ. Its rest-frame optical light distribu-
exhibit complex morphologies,many with multiple com- tionappearsdisk-likewithacompactcentralcomponent
ponentsinboththerest-frameopticalandrest-frameul- thatdoesnotappearinthe rest-frameultravioletimage,
traviolet. Someoftheobjects,suchas1204and1598,re- possiblysuggestingthepresenceofalargecolumnofdust
semblethe“chaingalaxies”ofCowie et al.(1995),which in the galaxy center. Object P4 is bright in both band-
are believed to be massive galaxies in the act of forma- passes,buthastwocomponentswithin1′.′2withverydif-
tion, while others (447 and 560) have two-component ferent I −H colors, yielding an anomalously large
775 160
morphologies suggestive of mergers. ICD. The top left component has colors consistent with
The distribution of internal color dispersions is shown a passive galaxy (Cameron et al. 2010) and the bottom
as a function of redshift in Figure 3. The 100 galaxies right component may be a low-redshift interloper or a
in SFGmain exhibit a wide range of ICDs, 56 of which nearby star-forming galaxy.
are non-zero at > 3σ (see bottom panel). Of the galax- Theremaining10objectshavesingle-component,com-
ies with small uncertainties in the ICD (σm < 0.01), pact morphologies that appear similar in the two band-
ξ
only 22% have ξ < 0.01, suggesting that the majority passes. Of the objects with σm < 0.01, 4 of 8 also have
ξ
of color-selectedSFGs in this redshift range have under- ξ <0.01. For comparison, only 16 of 74 objects in SFG-
lying populations of old stars that are spatially distinct main with σm <0.01 have ξ <0.01, suggesting that the
ξ
fromtheyoungstars. ThemedianICDinthissubsample stellar populations in the passively-selected galaxies are
is 0.023. more uniformly distributed than those in SFGs.
There is no apparent correlation between ξ and z in
SFGmain, except for a possible downturn at z & 2.6. 4.1.3. Star-forming galaxies at 2.9<z<3.5
It may be that the ICD becomes less sensitive to stellar
For galaxies with 2.9 < z < 3.5, the I and H
population heterogeneity as the H filter approaches 775 160
160 bands both cover light blueward of the 4000 ˚A break
the rest-frame 4000 ˚A break. Alternatively, it may be
andthe ICD becomes a poor tracer ofstellar population
duetoevolution-galaxiesareexpectedtoevolvetowards
heterogeneity. If the ICDs of the SFGmain subsample
lessstellarpopulationheterogeneityandlowerdustmass
aredominated by this heterogeneity(as opposedto dust
with increasing redshift (Hayes et al. 2010). There is a
inhomogeneity), we expect to see a sharp dropoff in the
smallsubsetof21SFGswithaverylargeICD(ξ >0.05),
ICDs of the SFGhighz subsample. This dropoff does ap-
which appear to make up a much larger fraction of the
pear in the top panel of Figure 3, where 8 of 11 galaxies
SFG subsample at z < 2. This is consistent with the with σm < 0.01 also have ξ < 0.01. Of the entire sub-
findings of P05, who found that galaxies with very large ξ
sample of 16 galaxies, only 4 have ξ > 0 at > 3σ and
ICDs went from 7% of their H-band-selected sample at
z ∼2.3 to 25% at z ∼1. only 1 at >5σ.
Figure 4 plots ξ as a function of I775 − H160 color. 4.2. Half-light radii and centroid offsets
We find no evidence for a correlation in SFGmain. Al-
High-redshift star-forming galaxies sometimes show
though this is in agreement with the bulk properties of
strong morphological variation between bandpasses, as
the P05 sample, they found that the galaxies with very
large ICDs had very red I −H colors. We see no indicatedbyalargeICD,butdifferentialdiagnosticslike
775 160
the ICD are only useful if they provide us with infor-
evidenceforthiseffectinSFGmain,butconsideringthat
mation that single-band diagnostics, such as half-light
P05 only had 7 galaxies with ξ > 0.05, the apparent
radius and centroid offset, cannot provide. In Figure 5,
differences may be due solely to small-number statistics.
The lack ofa correlationbetweenξ andI −H sup- weplotthe rest-frameopticalhalf-lightradiusasa func-
775 160
tion of the rest-frame UV half-light radius for the SFGs
ports the hypothesis that the morphological differences
in our sample. In most cases, the half-light radii differ
between the rest-frame ultraviolet and rest-frame opti-
by . 20%, with the rest-frame UV light typically being
cal are due primarily to stellar population heterogeneity
more extended for the larger objects. The typical SFG
ratherthaninhomogeneousdustdistributions,asgreater
ismarginallylarger(∼5%)inthe H imageevenafter
dust columns would both redden the galaxies and pro- 160
PSF convolution. The points in Figure 5 are also color
duce larger ICDs. A large population of old stars can
also redden a galaxy’s I −H color, but a galaxy coded according to the magnitude of the ICD for that
775 160
object (blue is smallest, red is largest). Although the
with a larger fraction of old stars will only have a larger
objects with the largestdifference in half-light radius do
ICD if the two populations are spatially segregated and
generally have large ICDs, there are many objects that
the oldstarsarenot dominatingthe emissionin the I
775
exhibit a great deal of morphological variation between
and H bands.
160
bandpasses with relatively little difference in half-light
4.1.2. Passive galaxies radius. Figure 6 indicates a clear increase in the me-
dianICDwithincreasingrest-frameopticalhalf-lightra-
TheI775andH160cutoutsforthe15passivegalaxiesin dius. AlthoughconvolutionwiththePSFwillartificially
our PassGal subsample are shown in Figure 9. Objects decrease the ICD for objects near the resolution limit
P1, P3, and P15 are faint in I775 and have σξm > 0.1, (re . 2 kpc), the trend continues to larger radii. The
so there is little we can say about their differential mor- largestobjects(>3kpc)haveamedianICDanddisper-
phologies. Theremaybeadiffusecomponenttotherest- sionthatis∼2.5timesthatofSFGswith2<r <3kpc.
e
frame ultraviolet light of P3 that is not present in the The correlation between ICD and half-light radius may
rest-frameoptical,but the ICD is only greaterthanzero indicatethatthenon-zeroICDsarebeingcausedbystar
Differential Morphologies of 1.5<z <3 SFGs 7
Fig.4.— Internal color dispersion as a function of I775−H160 Fig. 5.—Half-lightradiusinH160 asafunctionofhalf-lightra-
color. PointsanderrorbarsarethesameasinFigure3andgalax- diusinI775forSFGsat1.4<z<2.9,wherepointsarecolor-coded
ies with σm > 0.05 are not plotted. There is no evidence of a accordingtotheirinternalcolordispersion. Thehalf-lightradiiare
correlationξbetween ICD and rest-UV-to-rest-optical color in the similar between the two bandpasses (∼ 20% scatter), with a me-
2.9<z<3.5SFGs(blackpoints). dian ratio of reF160W/reF775W = 1.02. Objects sometimes show
strong morphological variation between bandpasses, as indicated
formation from gas that is dynamically young; that is, byalargeICD,butthisisnotgenericallyaccompaniedbyalarge
gasthatis spatiallysegregatedfromadynamicallyolder differenceinhalf-lightradius.
stellar population and has been recently accreted onto
the galaxy’shostdarkmatterhalo. By contrast,clumpy that they may be mergers-in-progress. This hypothesis
star formation that arises from, for example, disk insta- isdifficultto provewithimagingalone,butthe factthat
bilities or stellar feedback, would not act to increase a many of these galaxies appear clumpy in the rest-frame
galaxy’s size in the rest-frame optical. optical suggests that it is not just the young stars that
Unlikehalf-lightradius,thecentroidoffsetdoesappear have inhomogeneous distributions, but also older gener-
to give some indication of the morphological variation ationsofstars thatwouldhavehadtime to mix ina sin-
between bandpasses (see Figure 7). However, there is a gle,steadily-accretingsystem. Furthermore,P05showed
greatdealofscatter,evenoutto the largestcentroidoff- thatsingle galaxieswithhalf-lightradii >3 kpc arerare
sets. Galaxies with the largestICDs (ξ >0.05)typically at z ∼2, and that ∼50% of the objects in their sample
havecentroidoffsetsof∼0′.′1,correspondingto0.83kpc with ξ > 0.08 are larger than 3 kpc. If the large ICDs
at z =2.3. are simply due to clumpy star formation, it is occurring
ona scalemuchlargerthanis typicalforgalaxiesatthis
5. DISCUSSION redshift.
The morphological differences between the rest-frame Although many of the passive galaxies in our PassGal
optical and rest-frame UV light distributions in 1.4 < subsamplearecentrallycompactandhaveICDsthatare
z < 2.9 SFGs are typically small (ξ ∼ 0.02), but are consistent with zero, in line with the conventional pic-
statistically significant (more than half are non-zero at ture of this class of galaxies (van Dokkum et al. 2004;
> 3σ) and are larger than we find in the majority of Daddi et al.2005;Cimatti et al.2008;Franx et al.2008;
passive galaxies at 1.4 < z < 2. We find that these dif- Damjanov et al. 2009; Szomoru et al.2010), atleastfive
ferences are due to heterogeneous stellar populations, in of the galaxies have non-zero ICDs at > 3σ. This
agreementwithP05,bothbecausetheICDsshownocor- suggests that a significant fraction of largely passive
relation with UV-optical color and because a sample of galaxies at this epoch are still actively forming stars,
higher-redshift galaxies shows much smaller ICDs (typi- possibly in clumpy stellar disks (Stockton et al. 2008;
cally ξ < 0.01) when compared between two rest-frame McGrath et al. 2008; Cassata et al. 2010).
UV bandpasses. We cannot rule out a contribution to It is important that we continue to explore the wave-
the ICDfromanon-uniformdustdistribution,butthese length dependence of high-redshift galaxy morphologies,
two facts suggest that it is sub-dominant. asdifferentregionsofthespectrumprobedifferentphysi-
The SFGs with the largest internal color dispersion calcomponentsofthehostgalaxy. Becausehigh-redshift
(&0.05,seeFigure 8)generallyhavecomplex morpholo- galaxiesareoftenclumpy andirregular,diagnosticssuch
gies that are both extended and asymmetric, suggesting asthehalf-lightradiusandconcentrationindexaregood
8 Bond et al.
for obtaining a quantitative measure of the morphology
in a single band but are usually poor at capturing the
filter-to-filtervariations(e.g.,seeFigure5). Asageneral
ruleforgalaxiesatz &2,itisbettertomeasurethemor-
phologicalpropertiesofthedifferenceimagethanthedif-
ferencein morphologicalpropertiesbetweenbandpasses.
The internal colordispersion, whichgives the rootmean
squared amplitude of the difference image, is one exam-
ple of this, but any morphological diagnostic that can
be applied to a single-band light distribution can also
be applied to a two-band difference image. An obvious
next step would be to measure higher-order moments
of the differential light distribution, but given the rela-
tively small values of ξ we find in high-redshift galaxies,
one would likely require greater imaging depth in both
bands to draw meaningful conclusions.
WethankCaseyPapovichandSwaraRavindranathfor
their helpful comments on the implementation of the in-
ternalcolordispersionandMartinAltmannfortheuseof
Fig.6.—Internalcolordispersionasafunctionoffixed-aperture hissampleofstarsintheECDF-S.Supportforthiswork
half-light radius in H160. Points and error bars are the same as wasprovidedbyNASAthroughgrantnumberHST-AR-
in Figure 3 and galaxies with σm > 0.05 are not plotted. For
ξ 11253.01-A from the Space Telescope Science Institute,
2.9<z<3.5SFGs,thereisaclearincreaseinthemeanICDasre whichis operatedby AURA, Inc.,under NASA contract
increases,even forobjects muchlargerthan theWFC3 resolution
limit(re ∼1kpc), possiblyindicatingthat largeICDs arearising NAS 5-26555 and by the National Science Foundation
fromsystemsthatarenotyetdynamicallysettled. under grant AST-0807570.
REFERENCES
???? Franx,M.,vanDokkum,P.G.,Schreiber,N.M.F.,Wuyts,S.,
08.1 Labb´e,I.,&Toft,S.2008,ApJ,688,770
Adelberger,K.L.,Steidel,C.C.,Shapley,A.E.,Hunt,M.P., Giacconi,R.etal.2002,ApJS,139,369
Erb,D.K.,Reddy,N.A.,&Pettini,M.2004, ApJ,607,226 Giavalisco,M.,Steidel,C.C.,&Macchetto, F.D.1996,ApJ,470,
Altmann,M.,M´endez,R.A.,vanAltena,W.,Korchargin,V.,& 189
Ruiz,M.T.2006, inRevistaMexicanadeAstronomiay Giavalisco,M.etal.2004, ApJ,600,L93
AstrofisicaConferenceSeries,Vol.26,RevistaMexicanade Glazebrook,K.,Ellis,R.,Santiago, B.,&Griffiths,R.1995,
AstronomiayAstrofisicaConferenceSeries,64–65 MNRAS,275,L19
Balestra,I.etal.2010,A&A,512,A12+ Griffiths,R.E.etal.1996,inIAUSymposium,Vol.168,
Bertin,E.&Arnouts,S.1996, A&AS,117,393 ExaminingtheBigBangandDiffuseBackgroundRadiations,
Bond,H.E.,KimQuijano,J.,etal.2007,“WideFieldCamera3 ed.M.C.Kafatos&Y.Kondo,219–+
InstrumentHandbook, Version1.0”(Baltimore:STScI) Gronwall,C.etal.2007,ApJ,667,79
Bond,N.A.,Gawiser,E.,Gronwall,C.,Ciardullo,R.,Altmann, Hathi,N.P.etal.2010,arXiv:astro-ph/1004.5141
M.,&Schawinski,K.2009,AstrophysicalJournal,705,639 Hayes,M.,Schaerer,D.,Ostlin,G.,Mas-Hesse,J.M.,Atek, H.,
Bouwens,R.J.,Illingworth,G.D.,Blakeslee,J.P.,Broadhurst, &Kunth,D.2010,ArXive-prints
T.J.,&Franx,M.2004,ApJ,611,L1 Hu,E.M.&McMahon,R.G.1996, Nature,382,231
Brinchmann,J.etal.1998,ApJ,499,112 Hubble,E.P.1936,RealmoftheNebulae,ed.E.P.Hubble
Bunker,A.etal.2009,arXiv:astro-ph/0909.2255 Koekemoer,A.M.,Fruchter,A.S.,Hook,R.N.,&Hack,W.
Cameron,E.,Carollo,C.M.,Oesch,P.A.,Bouwens,R.J., 2002,inThe2002HSTCalibrationWorkshop:Hubbleafter
Illingworth,G.D.,Trenti,M.,Labbe, I.,&Magee,D.2010, theInstallationoftheACSandtheNICMOSCoolingSystem,
arXiv:astro-ph/1007.2422 ed.S.Arribas,A.Koekemoer,&B.Whitmore,337–+
Cassata,P.etal.2005, MNRAS,357,903 Lehmer,B.D.etal.2005, ApJS,161,21
—.2010,ApJ,714,L79 Lilly,S.etal.1998,ApJ,500,75
Cimatti,A.etal.2008, A&A,482,21 Lotz,J.M.,Madau,P.,Giavalisco,M.,Primack,J.,&Ferguson,
Conselice,C.J.,Blackburne,J.A.,&Papovich,C.2005,ApJ, H.C.2006,ApJ,636,592
620,564 Lowenthal,J.D.etal.1997,ApJ,481,673
Cowie,L.L.,Hu,E.M.,&Songaila,A.1995,AJ,110,1576 Luo,B.etal.2008,ApJS,179,19
Daddi,E.,Cimatti,A.,Renzini,A.,Fontana, A.,Mignoli,M., McGrath,E.J.,Stockton, A.,Canalizo,G.,Iye,M.,&Maihara,
Pozzetti, L.,Tozzi,P.,&Zamorani,G.2004,ApJ,617,746 T.2008, ApJ,682,303
Daddi,E.etal.2005,ApJ,626,680 Oesch,P.A.etal.2009,ApJ,690,1350
Damjanov,I.etal.2009, ApJ,695,101 —.2010, ApJ,709,L16
Dickinson,M.2000,inRoyalSociety ofLondonPhilosophical Papovich,C.,Dickinson,M.,Giavalisco,M.,Conselice,C.J.,&
Transactions SeriesA,Vol.358,Astronomy,physicsand Ferguson,H.C.2005,ApJ,631,101
chemistryofH+,2001–+ Papovich,C.,Giavalisco,M.,Dickinson,M.,Conselice,C.J.,&
3
Ferguson,H.C.etal.2004,ApJ,600,L107 Ferguson,H.C.2003,ApJ,598,827
Differential Morphologies of 1.5<z <3 SFGs 9
TABLE 3
Photometric Properties
Number Sample αa δa z H160 I775 rHb rIb ξc σξm d
(ABmags) (ABmags) (′′) (′′)
447 SFGmain 3:32:01.102 −27:44:04.869 2.635 23.225±0.026 24.354±0.051 0.46 0.43 0.148 0.024
455 SFGmain 3:32:01.311 −27:42:43.869 2.300 23.619±0.029 24.360±0.059 0.36 0.31 0.029 0.005
462 SFGmain 3:32:01.506 −27:44:22.597 2.512 24.234±0.068 25.206±0.120 0.22 0.15 0.024 0.009
479 SFGmain 3:32:02.063 −27:45:13.900 2.313 23.729±0.052 23.653±0.026 0.14 0.07 0.047 0.009
482 SFGhighz 3:32:02.163 −27:42:30.408 3.263 24.108±0.046 24.971±0.103 0.29 0.24 0.022 0.011
a PositionofWFC3H160-bandcentroid
b Half-lightradiuscomputedwithin1′.′2fixedaperturebeforeconvolution
c InternalcolordispersionbetweenF160WandF775Wimages
d UncertaintyoninternalcolordispersionbasedonMonteCarlosimulations(seeSection3.3)
Petty,S.M.,deMello,D.F.,Gallagher,J.S.,Gardner,J.P.,
Lotz,J.M.,MattMountain,C.,&Smith,L.J.2009,AJ,138,
362
Popesso,P.etal.2009,A&A,494,443
Rafelski,M.,Wolfe,A.M.,Cooke, J.,Chen,H.,Armandroff,
T.E.,&Wirth,G.D.2009,ApJ,703,2033
Ravindranath,S.etal.2004,ApJ,604,L9
—.2006, ApJ,652,963
Simard,L.etal.1999,ApJ,519,563
Spergel,D.N.etal.2007,ApJS,170,377
Stanford,S.A.etal.2004,AJ,127,131
Steidel,C.C.,Giavalisco,M.,Dickinson,M.,&Adelberger,K.L.
1996,AJ,112,352
Stockton, A.,McGrath,E.,Canalizo,G.,Iye,M.,&Maihara,T.
2008,ApJ,672,146
Szomoru,D.etal.2010,ApJ,714,L244
vandenBergh,S.2001,AJ,122,621
vandenBergh,S.,Abraham,R.G.,Ellis,R.S.,Tanvir,N.R.,
Santiago, B.X.,&Glazebrook, K.G.1996,AJ,112,359
vanDokkum,P.G.,Franx,M.,Fabricant,D.,Illingworth,G.D.,
&Kelson,D.D.2000,ApJ,541,95
vanDokkum,P.G.etal.2004,ApJ,611,703
Virani,S.N.,Treister,E.,Urry,C.M.,&Gawiser,E.2006,AJ,
131,2373
Windhorst,R.A.etal.2010,arXiv:astro-ph/1005.2776
Yan,H.,Windhorst,R.,Hathi,N.,Cohen, S.,Ryan,R.,
O’Connell,R.,&McCarthy,P.2009,ArXiv:astro-ph/0910.0077
Fig.7.—Internalcolordispersionasafunctionofcentroidoffset
between the H160 and I775 images forasample ofgalaxies inthe
WFC3ERS.PointsanderrorbarsarethesameasinFigure3and
galaxies with σm > 0.05 are not plotted. Although objects are
ξ
morelikelytohavealargepositionoffsetwhentheirICDislarge,
the two quantities are not tightly correlated and there are many
objectswithξ>0.01andanoffset<0′.′05.
10 Bond et al.
Fig.8.—H160 and I775 cutouts of 151.4<z<2.9star-forminggalaxies withξ>0.05. Thecutouts are1′.′8(30pixels) onasideand
havebeenconvolved withthePSFofthecomparisonimage(i.e. H160 withI775 andviceversa).