astro-ph0309058/ms.tex
1: %%%%%%%%%%%%%%%
2: % Define a switch for submission mode
3: %%%%%%%%%%%%%%%
4: 
5: \newif\ifsubmode
6: \submodefalse
7: %\submodetrue
8: 
9: %%%%%%%%%%%%%%%
10: % Preamble
11: %%%%%%%%%%%%%%%
12: 
13: \ifsubmode
14:   \documentclass[12pt,preprint]{aastex}
15:   \received{}
16:   \revised{}
17:   \accepted{}
18:   \citestyle{aa}
19: 
20: \else
21:   \documentclass[12pt,preprint]{emulateapj}
22:   \font\sevenrm=cmr7 
23:   \def\kmm{{\sevenrm KMM}~}
24:   \citestyle{aa} 
25: \fi
26: 
27: %%%%%%%%%%%%%%%                                                           
28: % User defined macros                                                     
29: %%%%%%%%%%%%%%%                                                           
30:                                                                           
31: \newcommand{\etal}{{et al.~}}                                             
32: \newcommand{\lta}{\lesssim}                                               
33: \newcommand{\gta}{\gtrsim}                                                
34: \newcommand{\kms}{\,\rm km\,s^{-1}}                                       
35: 
36: \newenvironment{inlinefigure}{
37: \def\@captype{figure}
38: \noindent\begin{minipage}{0.999\linewidth}\begin{center}}
39: {\end{center}\end{minipage}\smallskip}
40: 
41: %%%%%%%%%%%%%%%                                                           
42: % Figure Captions
43: %%%%%%%%%%%%%%%                                                           
44:                                                                           
45: \newcommand{\figonecap}{\label{fig1}
46: Size vs. redshift relation. Mean SExtractor half-light radii are
47: plotted with errorbars indicating the standard error of the mean
48: (i.e. the sample standard devation divided by the square-root of
49: the sample size).
50: The solid blue curve shows the expected trend
51: in the WMAP cosmology
52: if physical (proper) sizes do not evolve. The dashed red curve shows
53: the trend if sizes evolve as $H^{-1}(z)$ and the dotted green curve
54: shows $H^{-2/3}(z)$. The curves are all normalized to the mean
55: size at $z \approx 4$ (approximately $r_h = 1.7$ kpc).
56: }
57: 
58: \newcommand{\figtwocap}{
59: \label{fig2}                                                           
60: Observed size distributions. Data for the different samples are shown
61: with Poisson error bars. The curves show a ``no-evolution'' model
62: of the size distribution, tuned to match approximately the observed
63: distribution at $z \approx 4$. The model has a log-normal distribution
64: of radii with a peak at 2.1 kpc (for $L=2.9L^*$) and a width $\sigma =
65: 0.3$. We adopt the size--luminosity relation ($R \propto L^{1/3}$),
66: observed for local disks \citep{2000ApJ...545..781D}.  The models
67: include a 50/50 mix of oblate ellipticals and flat exponential disks.
68: Model galaxies have been inserted into the images in a Monte-Carlo
69: fashion and detected and measured using SExtractor in the same way
70: as the real galaxies. Therefore the model curves faithfully represent
71: the effects of both sample incompleteness and measurement biases. The
72: input distribution (prior to selection and measurement biases) for the
73: $z \approx 4$ bin is shown as the dashed curve.  Objects classified by
74: SExtractor neural-network as stellar at greater than 90\% probability
75: are excluded from the samples (both the data and the simulations). In
76: practice only a few objects with $r_h < 0.1$\arcsec\ are removed.
77: }
78: 
79: \newcommand{\figthreecap}{
80: \label{fig3}
81: (a) Ellipticity distribution for $z \approx 4$ galaxies compared to
82: simulations for pure disk and pure spheroid populations (the blue-dashed
83: and red-solid curves, respectively).  For the simulations, the disk sample
84: (viewed from arbitrary directions) is drawn from a population of oblate
85: optically-thin spheroids with a Gaussian distribution of intrinsic axial
86: ratios with mean $b/a = 0.05$ and $\sigma = 0.01$.  The spheroids are
87: drawn from a population of oblate spheroids with intrinsic axial ratios
88: uniformly populating the range $0.3 < b/a < 0.9$.  The samples for both
89: observations and simulations are limited to galaxies with $23 < I_{775}
90: < 25$ and SExtractor half-light radii $r_{h} > 0.2$\arcsec.
91: 
92: (b) Concentration-index distribution for
93: for $z \approx 4$ galaxies compared to simulations for
94: pure disk and pure spheroid populations (the blue-dashed and
95: red-solid curves, respectively). Samples are the same
96: as in (a).
97: }
98: 
99: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
100: \shorttitle{High-Redshift Galaxy Sizes}
101: \shortauthors{Ferguson et al.}
102: 
103: \begin{document}
104: 
105: \title{The Size Evolution of High Redshift Galaxies\altaffilmark{1}}
106: 
107: \author{
108: Henry C. Ferguson\altaffilmark{2,3}, 
109: Mark Dickinson\altaffilmark{2,3}, 
110: Mauro Giavalisco\altaffilmark{2,11}, 
111: Claudia Kretchmer\altaffilmark{3}, 
112: Swara Ravindranath\altaffilmark{2}, 
113: Rafal Idzi\altaffilmark{3}, 
114: Edward Taylor\altaffilmark{4}, 
115: Christopher J. Conselice\altaffilmark{5}, 
116: S. Michael Fall\altaffilmark{2}, 
117: Jonathan P. Gardner\altaffilmark{6}, 
118: Mario Livio\altaffilmark{2}, 
119: %Jennifer Lotz\altaffilmark{7}, 
120: Piero Madau\altaffilmark{7}, 
121: Leonidas A. Moustakas\altaffilmark{2}, 
122: Casey M. Papovich\altaffilmark{8}, 
123: Rachel S. Somerville\altaffilmark{2}, 
124: Hyron Spinrad\altaffilmark{9}, 
125: Daniel Stern\altaffilmark{10,11}
126: }
127: 
128: \altaffiltext{1}{Based on observations obtained with the NASA/ESA
129: {\it Hubble Space Telescope} (HST), the European Southern Observatory,  
130: and the Kitt Peak National Observatory (KPNO). HST is operated by the
131: Association of Universities for Research in Astronomy, Inc. (AURA)
132: under NASA contract NAS5-26555. KPNO is part of the National Optical
133: Astronomy Observatories, which is operated also by AURA under cooperative
134: agreement with the National Science Foundation.}
135: 
136: \altaffiltext{2}{Space Telescope Science Institute, 3700 San Martin Drive, 
137: Baltimore, MD  21218, USA}
138: 
139: \altaffiltext{3}{Department of Physics and Astronomy, The Johns Hopkins
140: University, 3400 N. Charles St., Baltimore, MD 21218}
141: 
142: \altaffiltext{4}{University of Melbourne, Australia}
143: 
144: \altaffiltext{5}{Caltech, MS 105-24, Pasadena CA}
145: 
146: \altaffiltext{6}{Laboratory for Astronomy and Solar Physics, Code 681, 
147: Goddard Space Flight Center, Greenbelt MD 20771}
148: 
149: \altaffiltext{7}{Department of Astronomy and Astrophysics, University
150: of California, Santa Cruz, CA 95064.}
151: 
152: \altaffiltext{8}{University of Arizona}
153: 
154: \altaffiltext{9}{
155: Dept. of Astronomy, University of California, Berkeley Berkeley CA. 94720-3411}
156: 
157: \altaffiltext{10}{Jet Propulsion Laboratory, California Institute of Technology,
158: Mail Stop 169-506, Pasadena, CA  91109, USA }
159: 
160: \altaffiltext{11}{Visiting astronomer Kitt Peak National Observatory.} 
161: 
162: 
163: \begin{abstract}
164: Hubble Space Telescope images of high-redshift galaxies selected via color
165: and photometric redshifts are used to examine the size and axial-ratio
166: distribution of galaxies as a function of redshift at lookback times $t
167: > 8$ Gyr. These parameters are measured at rest-frame UV wavelengths
168: ($1200< \lambda <2000$ \AA) on images with a rest-frame resolution
169: of less than 0.8 kpc.  Galaxy radii are found to scale with redshift
170: approximately as Hubble parameter $H^{-1}(z)$. This is in accord with
171: the theoretical expectation that the typical sizes of the luminous
172: parts of galaxies should track the expected evolution in the virial
173: radius of dark-matter halos.  The mean ratio of semi-major to
174: semi-minor axis for a bright well-resolved sample of galaxies
175: at $z \sim 4$ is $b/a =0.65$, suggesting that these Lyman 
176: break galaxies are not drawn from a spheroidal population. 
177: However the median concentration index of this sample is $C = 3.5$,
178: which is closer to the typical concentration indices, $C \sim 4$, of nearby
179: elliptical galaxies than to the values, $C<2 $ for local disk galaxies
180: of type Sb and later.
181: 
182: \end{abstract}
183: 
184: \keywords{Galaxies: evolution --- 
185: Galaxies: high-redshifts --- 
186: Galaxies: structure ---
187: Cosmology: observations 
188: }
189: 
190: \section {Introduction}
191: 
192: An important goal of cosmology is to understand how galaxies evolve
193: toward their current sizes and shapes.  The basic framework of galaxy
194: formation within the hierarchical Cold-Dark-Matter (CDM) cosmology was
195: set out by \citet{WR78}, and has been refined by numerous N-body and
196: semi-analytical studies \citep{WF91, KWG93, CAFNZ94, 1999MNRAS.310.1087S}.
197: The formation of galactic disks within dark-matter halos was studied by
198: \citet{1980MNRAS.193..189F}. In this model, dark halos acquire their
199: angular momenta via tidal torques, the angular momentum per unit mass
200: of the baryons and the dark matter are initially the same, and angular
201: momentum is conserved as the baryons collapse and cool to form a disk.
202: With these simplifying assumptions the baryons typically collapse by
203: factors of $\sim 10$ and the resulting disks have rotation curves,
204: surface-density profiles, and scale radii similar to those observed.
205: Further analytical studies have calculated the distribution function of
206: disk-galaxy sizes and the disk-galaxy size-redshift relation \citep{DSS97,
207: MMW98}. Meanwhile cosmological N-body + hydrodynamical simulations have
208: formed disks that appear similar to spiral galaxies, but that tend to
209: be too small to match present-day galaxies (\citealt{NavSteinmetz97};
210: but see \citealt{2000MNRAS.315L..18E}).  It is not yet known whether
211: the source of this discrepancy lies in baryonic, dark-matter physics,
212: or numerical issues in the simulations.
213: 
214: If the general Fall \& Efstathiou view of disk formation is correct,
215: there are several rather robust expectations that are worth exploring
216: through the observations of high-redshift galaxies.
217: 
218: {\em The size-redshift relation.}
219: The sizes of galactic disks forming at a redshift $z$ should be
220: a fixed fraction of the size of the dark-matter halo. The virial-radius
221: of a dark-matter halo scales with redshift and virial velocity 
222: $V_{\rm vir}$ or virial mass $M_{\rm vir}$ as 
223: \begin{equation}
224: R_{\rm vir} = \left[ {{GM_{\rm vir}} \over {100 H^2(z)}} \right]^{1/3} =
225: {{V_{\rm vir}} \over {10 H(z)}},
226: \end{equation}
227: where
228: \begin{equation}
229: H(z) = H_0 [ \Omega_m (1+z)^3 + \Omega_k (1+z)^2 + \Omega_\Lambda ] ^{1/2}
230: \end{equation}
231: is the Hubble parameter at redshift $z$ \citep{CPT92}.
232: Assuming that the exponential scale-length $R_s$ of the baryonic disk
233: scales with the virial radius, the sizes of disks are expected to
234: scale with redshift as $R_s \propto H^{-1}(z)$ at fixed circular
235: velocity, or $R_s \propto H^{-2/3}(z)$ at fixed mass. Observations
236: are more suitable to tracking evolution with fixed luminosity $L$, which
237: may be expected to fall somewhere in between these two functional forms.
238: Previous observations have demonstrated that high-redshift galaxies are
239: small (e.g. \citealt{LickHDFhighz}; \citealt{BBI2003}). The
240: larger sample of galaxies provided by GOODS allows a closer look
241: at the trend with redshift.
242: 
243: {\em The size distribution.}
244: If governed primarily by the angular momenta of their halos, the sizes
245: of disk galaxies should 
246: should show a log-normal distribution proportional
247: to the dimensionless spin parameter:
248: \begin{equation}
249: p(\lambda) d \lambda = {{1} \over {\sqrt{2 \pi \sigma_\lambda}}}
250: \exp \left[ - {{\ln^2 (\lambda/\bar{\lambda})} \over {2 \sigma_\lambda^2}} \right]
251: {{d \lambda} \over {\lambda}}, 
252: \end{equation}
253: where $\lambda$ is related to the total angular momentum $J$, energy $E$,
254: and mass $M$ of the dark matter halo through the definition
255: $\lambda \equiv J |E|^{1/2} G^{-1} M^{-5/2}$. 
256: The predicted functional form for the distribution of disk-galaxy sizes 
257: appears to be a reasonable match to observations of present-day
258: disk galaxies, albeit with
259: a width $\sigma$ that is significantly smaller than
260: predicted by tidal-torque theory \citep{2000ApJ...545..781D}. 
261: It is sensible to imagine that a simple proportionality between galaxy
262: radius and halo radius might hold even for elliptical galaxies, although
263: models are less well developed in this case. It is less 
264: obvious that the size distribution should show the same functional form.
265: 
266: {\em The ellipticity distribution.}
267: The theory described above is specific to disk galaxies. It is thus
268: interesting to test whether or not high-redshift galaxies are predominantly
269: disks. A relatively straightforward test is to compare the observed distribution
270: of axial ratios to the distribution expected for disks or spheroids 
271: viewed at a random set of orientations, applying the classic test
272: outlined by \citet{1970ApJ...160..831S}.  
273: 
274: %The outline of this paper is as follows. The observations and measurements
275: %are discussed in \S\ref{observations}. In \S\ref{results} we show measured the
276: %size-redshift relation, and the size and ellipticity distributions.
277: %Various biases and selection effects are discussed in \S\ref{biases}.
278: %Section \ref{conclusions} presents the conclusions.
279: Throughout this paper we use a cosmology with parameters 
280: $h,\Omega_{\rm tot},\Omega_{m},\Omega_{\Lambda} = 0.7,1.0,0.3,0.7$,
281: in the notation of \citet{CPT92}. The sizes of galaxies quoted
282: in this paper are in physical (proper), not co-moving, units.
283: All magnitudes are on the AB system \citep{Oke74}.
284: 
285: \section{Observations and Measurements}\label{observations}
286: 
287: The Chandra Deep Field South (CDF-S) and Hubble Deep Field North (HDF-N) 
288: were observed with the Hubble Space Telescope (HST) Advanced Camera
289: for Surveys (ACS) as part of the Great Observatories Origins Deep Survey
290: (GOODS).  For the HDF-N, U-band observations were carried
291: out using the Mosaic camera on the 4-m Mayall telescope.
292: Details of the observations, data reduction, and catalog generation
293: are described in \citet{Giavalisco2003a}. Samples of galaxies in
294: broad bins of redshift were constructed using
295: the \citet{Giavalisco2003b} Lyman-break selection for $z>3$
296: and the \citet{Mobasher2003b} CDF-S photometric redshifts 
297: (for two bins spanning the range $z=1 - 1.9$ and $z=1.9-2.8$).
298: The $z \approx 3$ and $4$ samples use data from both the HDF-N and the CDF-S. The 
299: $z \approx 3$ sample is drawn only from the HDF-N, where deep U-band images exist.
300: Because the redshift selection is based entirely on photometry, there
301: is undoubtedly some scatter of galaxies from one redshift bin to
302: another, and a few interlopers from lower redshifts in the high redshift bins.
303: Based on simulations of template galaxies of various spectral types
304: observed at the GOODS S/N, we expect the contamination 
305: to affect only the tails of the observed
306: size, ellipticity, and concentration-index distributions. Spectroscopic
307: confirmations are of course highly desirable.
308: 
309: Two different techniques were used to measure the radii of the galaxies.
310: Half-light radii (denoted by $r_h$) were measured using SExtractor \citep{BA96}. 
311: For this analysis SExtractor performs circular-aperture photometry using
312: an analysis aperture radius that is ten times larger than 
313: the first radial-moment of the light distribution defined by connected
314: pixels more than 0.6 $\sigma$ brighter than the background in a smoothed
315: detection image.
316: \citet{1976ApJ...209L...1P} radii were measured using the prescription adopted for the
317: SDSS project \citep{2002AJ....123..485S}. 
318: The ratio of the semi-minor to semi-major axis $b/a$ is measured
319: by SExtractor using second moments of the light distribution within
320: the analysis aperture.
321: 
322: % The above measurements are subject to selection biases and measurement
323: % biases. These are quantified in \S\ref{biases}. 
324: 
325: To compare the sizes of galaxies in different redshift intervals, we
326: confine our samples to non-evolving rest-frame luminosities between
327: $0.7L^*$ and $5 L*$, where $L^*$ is the characteristic luminosity
328: of a $z=3$ Lyman break galaxy from \citet{SAGDP99}.  Five samples
329: in different bins of redshift ($z \approx 1.4, 2.3, 3.0, 4.0,$ and
330: $5.0$) are constructed.  For each sample we select galaxies within the
331: desired luminosity range based on their apparent $z$-band magnitudes,
332: after accounting for cosmological effects and $k$-corrections assuming
333: a typical Lyman-break galaxy spectral-energy distribution (SED). The
334: adopted SED is from the 1999 version of the \citet{BC93} solar-metallicity models,
335: with a Salpeter initial-mass function, constant
336: star-formation rate, and age of 144 Myr.  This is attenuated assuming $E(B-V) =
337: 0.14$ with the \citet{Calzetti2000} extinction curve to match the mean
338: UV spectral slope seen in the $z \sim 3$ samples \citep{2000ApJ...544..218A}.
339: The luminosity range translates to a $z$-band magnitude
340: range of 23.1 to 25.3 for the $z \sim 4$ sample, with that magnitude range
341: shifting by -2.41, -1.12, -0.64, and +0.46 magnitudes for the samples
342: in the redshift bins at $z \sim 1.4, 2.3, 3.0,$ and $5.0$, respectively.
343: After rejecting sources 
344: classified as stellar (SExtractor \verb!CLASS_STAR!$ > 0.9$) 
345: or with
346: unreliable photometry (SExtractor \verb!FLAG!$ > 3$), the sample sizes (for the
347: lowest to highest redshift bins, respectively) are 21, 73, 140, 386,
348: and 153 galaxies.
349: 
350: \section{Results}\label{results} 
351: 
352: The galaxy sizes are measured in the band closest to rest-frame
353: $1500${\AA}.  At $z = 3,4,$ and $5$, the central
354: wavelengths of the F606W, F775W, and F850LP band filters fall at
355: rest-frame $\lambda = 1500 \pm 40$ {\AA}. For the lower-redshift
356: samples the size measurement was made in the F435W images corresponding
357: to $\lambda \sim 1800$\AA\  at $z = 1.4$ and $\lambda \sim 1300$\AA\ at $z=2.3$.
358: The sizes for any of these samples do not change appreciably
359: when measured through the next redder bandpasses.
360: 
361: 
362: The measured size-redshift relation for our samples is shown in
363: Fig. \ref{fig1}. The observed points are uncorrected for biases
364: or incompleteness (see \S\ref{biases}), and the errorbars represent
365: the standard error of the mean for each sample, ignoring possible
366: effects of clustering \citep{Somerville2003a}
367: or crowding. The measurements are roughly
368: consistent with constant angular sizes above $z = 2.5$.
369: For comparison, in Fig. \ref{fig1} we show three size-redshift
370: relations. The blue line is the angular-size
371: redshift relation for a standard measuring-rod in the WMAP cosmology
372: \citep{Spergel2003}. In this cosmology, angular diameters increase
373: by 23\% from $z=3$ to $z=5$ (for objects with fixed proper diameters).
374: The two lines that decrease with redshift
375: are the expected scalings if size at fixed luminosity tracks 
376: size at fixed virial velocity $r \propto H^{-1}(z)$ or fixed mass
377: $r \propto H^{-2/3}(z)$. The mean half-light radius of 
378: 0.24\arcsec\ at $z \approx 4$ corresponds to 1.7 kpc. 
379: 
380: \ifsubmode
381: \else
382: \begin{inlinefigure}
383: \begin{center}
384: \resizebox{\textwidth}{!}{\includegraphics{f1_color.ps}}
385: \figcaption{ \figonecap }
386: \end{center}
387: \end{inlinefigure}
388: \fi
389: 
390: The size distributions of galaxies in the different redshift bins
391: are shown in Fig. \ref{fig2}. The curves show a ``no-evolution''
392: model tuned to match the observed size distribution at $z=4$.
393: This model accounts for selection and measurement biases 
394: (including cosmological surface-brightness dimming) as
395: described in \S\ref{biases}. The galaxies in
396: the two low-redshift bins are clearly systematically larger than 
397: the no-evolution model would predict. However, given the sample
398: sizes and measurement biases, the trend between redshifts $z=3$ and $z=5$
399: is not easily distinguished from the no-evolution model.
400: 
401: Figure \ref{fig3}a compares the observed distribution of ellipticities
402: for the B-dropout $z \approx 4$ sample to simulations. 
403: The observed population has 
404: a larger proportion of galaxies with axial ratios $b/a \lesssim 0.4$
405: than expected for a purely spheroidal population. A Kolmogorov-Smirnov
406: test indicates a likelihood of 57\% that observed distribution could
407: have been drawn from the model disk distribution, while the probability
408: is $5\times10^{-7}$ that the observed distribution could have been
409: drawn from the model spheroidal distribution. Similar results
410: (at lower significance) are obtained for the $z \sim 3$ and $z \sim 5$ 
411: samples.  However, these statistical probabilities are sensitive to the 
412: size and magnitude distribution of galaxies fed into the simulations.
413: While we have attempted to match the observed distribution of
414: magnitudes and sizes (and colors) at $z=4$ (see Fig. 2), more work
415: is needed to verify that the preference for flattened systems is 
416: indeed robust.  Visual inspection of the images reveals
417: very few galaxies that have morphologies reminiscent of nearby 
418: disk galaxies. The observed flattening distribution is probably 
419: a reflection of the tendency for many of the  Lyman-break galaxies to have a few
420: concentrations of light of nearly equal brightness rather than a 
421: central dominant concentration.
422: 
423: \ifsubmode
424: \else
425: \begin{inlinefigure}
426: \begin{center}
427: \resizebox{\textwidth}{!}{\includegraphics{f2_color.ps}}
428: \figcaption{ \figtwocap }
429: \end{center}
430: \end{inlinefigure}
431: \fi
432: 
433: 
434: The concentration-index distribution for the same sample
435: is shown in Fig. \ref{fig3}b.  The concentration index here is defined
436: as $5 \log(r_{80}/r_{20})$ where $r_{80}$ and $r_{20}$ are radii
437: containing 80\% and 20\% of the galaxy flux, respectively
438: \citep{Conselice2003a}.
439: The observed galaxies span a broader range
440: than the ideal disks and spheroids that populate the simulations.
441: The $z \approx 4$ population tends to be more centrally concentrated than 
442: pure exponentials.
443: Nearby spiral galaxies tend to have central concentrations of light
444: as well, but these also tend to be red, and virtually invisible
445: in the rest-frame ultraviolet where the current measurements are made.
446: 
447: \ifsubmode
448: \else
449: \begin{inlinefigure}
450: \begin{center}
451: \resizebox{\textwidth}{!}{\includegraphics{f3_color.ps}}
452: \figcaption{\figthreecap}
453: \end{center}
454: \end{inlinefigure}
455: \fi
456: 
457: 
458: \section{Selection Effects and Biases}\label{biases} 
459: 
460: In this section we highlight the issues of bias and incompleteness
461: and the tests that have
462: been carried out to verify the results of the previous sections.
463: 
464: {\it Incompleteness.} The completeness limits of the GOODS images
465: have been estimated by inserting artificial galaxies with a
466: wide range of half-light radii and magnitudes into the GOODS $z$-band
467: image, and re-running SExtractor. The simulated galaxies have either $r^{1/4}$-law
468: or exponential surface-brightness profiles, with the axial ratio
469: distributions described in the caption to Fig. 3.
470: The completeness limits as a function of
471: size and magnitude are shown by \citet{Giavalisco2003a}.
472: At $z_{850} = 24.7$, corresponding to the mean magnitude of
473: the $z \approx 4$ sample considered in this letter, the sample appears
474: to be 80\% complete for galaxies with half-light radii $r_h=0.5$\arcsec\
475: and 50\% complete for galaxies with $r_h = 0.9$\arcsec. 
476: The $z>3$ samples are affected by incompleteness for galaxies with sizes 
477: $r_h \gtrsim 3.5$ kpc; for the lower-redshift samples there is no
478: equivalent bias against detecting galaxies with small sizes. Therefore
479: the most robust result is the observation
480: that analogs of the galaxies with small half-light radii in the $z>3$ bins
481: are much less common at the lower redshifts. Without
482: careful modeling (or deeper images) it is not possible to rule out
483: the possibility that galaxies with larger sizes exist at the higher
484: redshifts. It is worth noting that a pure exponential disk galaxy with 
485: a luminosity $L^*$ and a scale length of $>5$ kpc would 
486: be unlikely to be detected at $z>3$ in the GOODS $z$-band images.
487: 
488: {\it Radius-measurement biases.} The results shown in Fig. 1 and 2 
489: are for SExtractor half-light radii. The trends and distributions
490: look virtually identical when Petrosian radii are used. Based
491: on the simulations it is clear that both radius measurements become
492: biased at faint magnitudes and small sizes. For example, for galaxies in the
493: magnitude range of our $z \approx 4$ sample, a galaxy with an
494: intrinsic $r_h = 0.5${\arcsec} would have a measured $r_h = 0.36$\arcsec.
495: Measurement biases will
496: thus pull in the tail of the size distribution; the true size
497: distributions at $z>3$ are almost certainly broader and peaked at slightly
498: larger sizes than the observed distributions shown in Fig. 2. 
499: The model distribution for the $z \sim 4$ sample provides an estimate
500: of the importance of this bias. The true distribution of half-light
501: radii input to the simulation is shown as the dashed line.
502: For this input distribution the effect of selection
503: and the size-measurement bias are minor. A distribution
504: with a more significant tail to large sizes would show more of an effect.
505: We have tried broader distributions and 
506: distributions with a larger mean size and found them to predict more
507: large galaxies than observed in the $z \approx 4$ sample.
508: 
509: \section{Summary and Conclusions}\label{conclusions} 
510: 
511: Samples of galaxies within a fixed and fairly
512: narrow range of rest-frame UV luminosities have been compared from
513: redshifts $z \sim 1 $ to $z \sim 5$. 
514: The sizes of high-redshift galaxies appear to evolve roughly
515: as $H^{-1}(z)$ in agreement with the rather robust expectation from
516: hierarchical theory. This general conclusion is subject to the
517: caveat that we may be seeing different kinds of galaxies at the
518: different redshifts. Our sample selection necessarily targets
519: UV-bright objects; further studies on samples selected
520: in the near-infrared or in subsets of morphology are clearly
521: warranted. For the galaxy sample discussed here, the size 
522: distribution at $z \sim 4$ is 
523: reasonably well represented by a log-normal
524: distribution, but with a smaller mean size than 
525: observed for samples of nearby luminous galaxies.
526: Measurements of the ellipticity and concentration-index
527: distributions suggest that Lyman-break galaxies at $z \sim 4$ represent a 
528: mix of morphologies, with some tendency toward flattened systems.
529: 
530: \acknowledgments
531: 
532: Support for this work was
533: provided by NASA through grant GO09583.01-96A from the Space Telescope
534: Science Institute, which is operated by the Association of
535: Universities for Research in Astronomy, under NASA contract
536: NAS5-26555.
537: Support for this work, part of the {\it Space Infrared Telescope 
538: Facility (SIRTF)} Legacy Science Program, was provided by NASA through 
539: Contract Number 1224666 issued by the Jet Propulsion Laboratory, 
540: California Institute of Technology under NASA contract 1407. 
541: PM acknowledges support by NASA through grant NAG5-11513.
542: 
543: 
544: \bibliographystyle{apj}
545: \bibliography{apjmnemonic,bib}
546: 
547: 
548: \ifsubmode
549: \newpage
550: 
551: \begin{inlinefigure}
552: \begin{center}
553: \resizebox{\textwidth}{!}{\includegraphics{f1.ps}}
554: \figcaption{ \figonecap }
555: \end{center}
556: \end{inlinefigure}
557: 
558: \newpage
559: \begin{inlinefigure}
560: \begin{center}
561: \resizebox{\textwidth}{!}{\includegraphics{f2.ps}}
562: \figcaption{ \figtwocap }
563: \end{center}
564: \end{inlinefigure}
565: 
566: \newpage
567: \begin{inlinefigure}
568: \begin{center}
569: \resizebox{\textwidth}{!}{\includegraphics{f3.ps}}
570: \figcaption{ \figthreecap }
571: \end{center}
572: \end{inlinefigure}
573: 
574: \fi
575: 
576: \end{document}
577: