1: \documentclass[12pt,preprint]{aastex}
2:
3: %\documentclass[manuscript]{aastex}
4: %\documentclass[preprint2]{aastex}
5: %\documentclass[preprint2,longabstract]{aastex}
6:
7: \newcommand{\vdag}{(v)^\dagger}
8: \newcommand{\myemail}{rji@roe.ac.uk}
9: \newcommand{\sfrd}{$\rho_{\star}$}
10: \newcommand{\sfrduv}{$\rho_{\star\rm (UV)}$}
11: \newcommand{\sfrdrad}{$\rho_{\star\rm (rad)}$}
12:
13: \slugcomment{To appear in AEGIS Special Issue of ApJ Letters}
14:
15: \shorttitle{AEGIS20: a radio survey of the Extended Groth Strip}
16: \shortauthors{Ivison et al.}
17:
18: \begin{document}
19:
20: \title{AEGIS20: a radio survey of the Extended Groth Strip}
21:
22: \author{R.\,J.\ Ivison,\altaffilmark{1}
23: S.\,C.\ Chapman,\altaffilmark{2} S.\,M.\ Faber,\altaffilmark{3}
24: Ian Smail,\altaffilmark{4} A.\,D.\ Biggs,\altaffilmark{1}
25: C.\,J.\ Conselice,\altaffilmark{5}\\ G.\ Wilson,\altaffilmark{6}
26: S.\ Salim,\altaffilmark{7} J.-S.\ Huang\altaffilmark{8}
27: \& S.\,P.\ Willner\altaffilmark{8}
28: }
29:
30: \altaffiltext{1}{UK Astronomy Technology Centre, Royal Observatory,
31: Blackford Hill, Edinburgh EH9\,3HJ, UK}
32: \altaffiltext{2}{Astronomy Department, California Institute of
33: Technology, Pasadena, CA\,91125}
34: \altaffiltext{3}{Department of Astronomy and
35: Astrophysics, University of California, Santa Cruz,
36: Santa Cruz, CA\,95064}
37: \altaffiltext{4}{Institute for Computational Cosmology, Durham
38: University, South Road, Durham DH1\,3LE, UK}
39: \altaffiltext{5}{Physics and Astronomy, University of
40: Nottingham, University Park, Nottingham NG7\,2RD, UK}
41: \altaffiltext{6}{Spitzer Science Center, California Institute of Technology,
42: 1200 E.\ California Bvd, Pasadena, CA\,91125}
43: \altaffiltext{7}{Department of Physics and Astronomy, University of California,
44: Los Angeles, CA\,90095}
45: \altaffiltext{8}{Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,
46: Cambridge, MA\,02138}
47:
48: \setcounter{footnote}{8}
49:
50: \begin{abstract}
51: We describe AEGIS20 -- a radio survey of the Extended Groth Strip
52: (EGS) conducted with the Very Large Array (VLA) at 1.4\,GHz. The
53: resulting catalog contains 1,123 emitters and is sensitive to
54: ultraluminous (10$^{12}$\,L$_{\odot}$) starbursts to $z\le\rm 1.3$,
55: well matched to the redshift range of the DEEP2 spectroscopic survey
56: in this region. We use stacking techniques to explore the
57: $\mu$Jy-level emission from a variety of galaxy populations selected
58: via conventional criteria -- Lyman-break galaxies (LBGs), distant red
59: galaxies (DRGs), UV-selected galaxies and extremely red objects (EROs)
60: -- determining their properties as a function of color, magnitude and
61: redshift and their extinction-free contributions to the history of
62: star formation. We confirm the familiar pattern that the
63: star-formation-rate (SFR) density, \sfrd, rises by {\em at least}
64: $\sim$5$\times$ from $z$ = 0 to 1, though we note highly discrepant
65: UV- and radio-based SFR estimates. Our radio-based SFRs become more
66: difficult to interpret at $z>\rm 1$ where correcting for contamination
67: by radio-loud active galactic nuclei (AGN) comes at the price of
68: rejecting luminous starbursts. Whilst stacking radio images is a
69: useful technique, accurate radio-based SFRs for $z\gg\rm 1$ galaxies
70: require precise redshifts and extraordinarily high-fidelity radio data
71: to identify and remove accretion-related emission.
72: \end{abstract}
73:
74: \keywords{cosmology: observations --- galaxies: evolution}
75:
76: \section{Introduction}
77:
78: The tight correlation between radio and far-IR emission for
79: star-forming galaxies (Helou et al.\ 1985; Garrett 2002; Kovacs et
80: al.\ 2006), allows us to push dust-independent surveys down to lower
81: SFRs than is possible in the confusion-limited far-IR/submm
82: wavebands. Moreover, the high mapping speed of facilities such as the
83: Giant Metre-wave Radio Telescope means we can quickly obtain the large
84: samples of faint sources needed for reliable analyses.
85:
86: The bulk of the far-IR background seen by {\it COBE} (Fixsen et al.\
87: 1998) most likely arises from a large population of luminous and
88: ultraluminous IR galaxies (LIRGs and ULIRGs), their energy originating
89: from dust-obscured star formation and accretion. Individually less
90: luminous than submm galaxies, with $L_{\rm bol}\sim 3\times
91: 10^{11}$\,L$_{\odot}$, these galaxies are believed to be sufficiently
92: numerous to dominate \sfrd\ at $z\sim$ 1 (Dole et al.\ 2006).
93:
94: In this letter we present a new panoramic radio survey -- AEGIS20 --
95: undertaken with the National Radio Astronomy
96: Observatory's\footnote{The National Radio Astronomy Observatory is
97: operated by Associated Universities Inc., under a cooperative
98: agreement with the National Science Foundation.} VLA as part of the
99: All-wavelength Extended Groth Strip International Survey (AEGIS; Davis
100: et al.\ 2006). AEGIS20 was tuned to detect ULIRGs robustly at $z\sim$
101: 1, with a noise level of 10\,$\mu$Jy\,beam$^{-1}$ at 1.4\,GHz (cf.\
102: Hopkins et al.\ 2003; Bondi et al.\ 2003). The resulting catalog,
103: available electronically, contains $\sim$10$^3$ faint radio sources --
104: an order of magnitude more than the 5-GHz survey of this region by
105: Fomalont et al.\ (1991); nearly half are expected to have optical
106: spectra provided by the DEEP2 survey, many with redshifts, as well as
107: photometry across a wide range of wavelengths.
108:
109: The future goal of AEGIS20 is to measure the 1.4-GHz luminosity
110: function, track the evolution of SFRs in LIRGs and ULIRGs and, using a
111: measure of the local galaxy density of each radio source, study the
112: history of star formation as a function of environment. Here, we
113: present the AEGIS20 catalog and utilise the radio image to estimate
114: SFRs for a number of independent and overlapping galaxy populations
115: selected via conventional criteria.
116:
117: \section{Observations and data reduction}
118:
119: Data were obtained at 1.4\,GHz during 2003--05 with the VLA in its B
120: configuration, acquiring seven 3.125-MHz channels every 5\,s in each
121: of four IFs. We obtained data in six positions, spaced by 15$'$ (see
122: Davis et al.\ 2006), concentrating in the northern half of the EGS
123: because of the proximity of 3C\,295 ($S_{\rm 1.4GHz}$ =
124: 23\,Jy). Around 18\,hr of data were acquired for each of the field
125: positions, cycling through them between scans of 1400+621 and 1419+543
126: to monitor bandpass, amplitude and phase. Absolute flux calibration
127: was set using 3C\,286.
128:
129: Calibrated visibilities and associated weights were used to generate
130: mosaics of $\rm 37\times 512^2\times 0.8''^2$-pixel images to quilt
131: the VLA's primary beam in each EGS field position. {\sc clean} boxes
132: were placed tightly around all sources and a series of {\sc imagr} and
133: {\sc calib} tasks were run, clipping the $uv$ data after subtracting
134: {\sc clean} components generated by the third iteration of {\sc
135: imagr}. The central images from each of the pointings were then
136: knitted together using {\sc flatn}, ignoring data beyond the primary
137: beam's half-power point, to produce a large mosaic. The synthesized
138: beam is circular, with a {\sc fwhm} of $\sim$3.8$''$.
139:
140: %
141: %
142: %
143: \section{Sample definition}
144:
145: To define a sample of radio sources we searched signal-to-noise (S/N)
146: images using the {\sc sad} detection algorithm, emulating the
147: technique described by Biggs \& Ivison (2006). Sources with
148: $\ge$4$\sigma$ peaks were fitted with 2-D Gaussians using {\sc jmfit},
149: those with $\ge$5$\sigma$ peaks surviving to be fitted in total
150: intensity. Sources with sizes equal to or smaller than the restoring
151: beam were considered unresolved; their size was constrained to that of
152: the beam. We make no correction for bandwidth smearing in the catalog:
153: this is a small effect ($\sim$5\%) given our mosaicing strategy and
154: the use of B configuration. We detect 38, 79, 171, 496 and 1,123
155: sources with $S_{\rm 1.4GHz}\ge\rm 2,000$, $\ge$800, $\ge$320,
156: $\ge$130 and $\ge$50\,$\mu$Jy (consistent with Simpson et al.\ 2006),
157: where the 5$\sigma$ detection limits at 130 and 50\,$\mu$Jy cover 0.73
158: and 0.04\,deg$^2$. Confusion is not an issue: the source density on an
159: arcmin$^2$ scale is $<$0.01\,beam$^{-1}$.
160:
161: $S_{\rm 1.4GHz}=\rm 50\,\mu Jy$ corresponds to rest-frame 1.4-GHz
162: luminosities, $L_{\rm 1.4GHz}$, of 0.44, 2.3, 6.0 and 12 $\times
163: 10^{23}$\,W\,Hz$^{-1}$ and SFRs of 50, 275, 725 and
164: 1,430\,M$_{\odot}$\,yr$^{-1}$ at $z$ = 0.5, 1.0, 1.5 and 2.0 (for
165: $\Omega_m=0.27, \Omega_\Lambda=0.73$,
166: $H_0=71$\,km\,s$^{-1}$\,Mpc$^{-1}$ --- Spergel et al.\ 2003 --- a
167: Salpeter initial mass function [IMF] with d$N$/d$M\propto M^{-2.35}$
168: over 0.1--100\,M$_{\odot}$ and $S_{\nu}\propto \nu^{-0.8}$).
169:
170: AEGIS20 covers 57\% of the existing DEEP2 region (Davis et al.\ 2006),
171: with $\sim$7,900 unique redshifts available in the 0.28\,deg$^2$
172: common to both surveys. DEEP2 thus covers 35\% of AEGIS20, although
173: its $BRI$ imaging covers 90\% of AEGIS20 (93\% of cataloged
174: sources). Of the AEGIS20 sources with optical imaging, $\sim$36\% have
175: $R_{\rm AB}<\rm 24.1$ counterparts within 1$''$. Since the DEEP2
176: targeting rate is $\sim$70\%, the inclusion rate on DEEP2 masks for
177: faint radio emitters is $\sim$25\%. At present, $\sim$100 of the
178: targeted AEGIS20 radio sources have DEEP2 redshifts --- a very high
179: success rate. Radio properties of DEEP2 galaxies and the spectroscopic
180: properties of the AEGIS20 catalog will be discussed in a forthcoming
181: paper.
182:
183: %
184: %
185: %
186: \section{The radio properties of distant galaxy populations}
187:
188: The wealth of multi-frequency data in AEGIS allows us to mimic the
189: selection of galaxy populations such as DRGs ($J-K>\rm 2.3$; expected
190: to lie at $\rm 1.9 < z < 3.5$ --- Franx et al.\ 2003), as well as LBGs
191: (Steidel et al.\ 2003) and EROs. We investigate the radio properties
192: of several such populations in this section, taking them roughly in
193: order of increasing redshift.
194:
195: We expect to detect only a small fraction of distant galaxies at radio
196: frequencies. In such situations it is common to assess the emission
197: from a galaxy population using a stacking analysis, accomplished
198: either by extracting and co-adding postage stamps centered on the
199: galaxies of interest (`image stacking') or by co-adding flux densities
200: measured at the positions of the galaxies (`pixel stacking'). We adopt
201: both approaches here. To determine the signal lost by pixel stacking
202: we employed radio emitters with S/N = 5--20 pixel$^{-1}$, finding a
203: difference of only 3.9\% between the values returned at the positions
204: of the emitters and cataloged AEGIS20 flux densities. Monte-Carlo
205: simulations show that the mean $S_{\rm 1.4GHz}$ determined by pixel
206: stacking are slightly skewed ($+$0.1\,$\mu$Jy, typically) but are
207: otherwise well described by Gaussian statistics; medians are affected
208: at the $<$0.01-$\mu$Jy level. $S_{\rm 1.4GHz}$ values have been
209: corrected for bandwidth smearing (+5.0\%), for pixel-stacking losses
210: (+3.9\%) and we have excluded galaxies in noisy regions ($\sigma_{\rm
211: 1.4GHz}>\rm 30\,\mu Jy\,beam^{-1}$).
212:
213: We must excise emission due to accretion if we are to determine
214: accurate radio-based SFRs. Morphological classification of most radio
215: emitters is not feasible at the resolution of our data, spectral
216: indices are not to hand and the availability and reliability of AGN
217: indicators at shorter wavelengths differs widely across the
218: EGS. Radio-loud AGN were thus identified and rejected via a $L_{\rm
219: 1.4GHz}$ limit. Following Condon (1992), we adopt $L_{\rm 1.4GHz} <\rm
220: 10^{24}\,W\,Hz^{-1}$ for normal galaxies, an order of magnitude below
221: the break in morphology and luminosity noted by Fanaroff \& Riley
222: (1974). We quote the noise-weighted mean $S_{\rm 1.4GHz}$; where AGN
223: contamination is extreme ($>$5\%), we quote the median, noting the
224: number of obvious AGN. One unfortunate consequence of excising
225: radio-loud AGN on the basis of $L_{\rm 1.4GHz}$ is the exclusion of
226: distant hyperluminous starbursts lying on the far-IR/radio correlation
227: (\S5).
228:
229: We begin with a ultraviolet (UV)-selected catalog containing 4,426
230: galaxies detected at 230\,nm by {\em GALEX}, with DEEP2 redshifts,
231: i.e.\ $R_{\rm AB}<\rm 24.1$, excluding objects with AGN flags (Salim
232: et al.\ 2006). Of these, 3,908 lie within the 0.28\,deg$^2$ of common
233: areal coverage with AEGIS20. We compare SFRs determined in two ways
234: -- via their UV and radio properties, SFR$_{\rm UV}$ and SFR$_{\rm
235: rad}$ -- for the same galaxies. We use UV-based, extinction-corrected
236: SFRs, derived by comparing observed spectral energy distributions
237: (SEDs) with those of model galaxies exhibiting a wide range of
238: properties and SF histories (Salim et al.\ 2005). Appropriate volume
239: corrections for the {\em GALEX}-selected sample are difficult to
240: determine due to a complex selection function dependent on UV/optical
241: magnitudes and spectral characteristics. We therefore correct for the
242: rate with which {\em GALEX} detects DEEP2 galaxies, which is known to
243: fall from 90--75--60\% at $z$ = 0.2--1.0--1.4, but not for our
244: steadily decreasing sensitivity to low-luminosity galaxies (Arnouts et
245: al.\ 2005). Because of this, we are limited to discussing the ratio of
246: \sfrduv\ and \sfrdrad\ within the sample.
247:
248: Robust radio detections of the UV sample were possible by pixel
249: stacking over $\Delta z=\rm 0.2$ bins. Only 11 radio-loud AGN were
250: identified via $L_{\rm 1.4GHz}$, $<<$1\% of the total; having rejected
251: these, noise-weighted means provide the most appropriate measure of
252: SFR for this sample (Table~\ref{table}). SFR$_{\rm UV}$ and SFR$_{\rm
253: rad}$ per UV-selected galaxy both increase with redshift,
254: unsurprisingly since we are probing more UV-luminous galaxies at
255: larger distances. For $z$ = 0--1, \sfrduv\ remains fairly constant
256: whilst \sfrdrad\ rises rapidly. It may seem puzzling that \sfrduv\ at
257: $z\sim\rm 0$ is an order of magnitude higher than \sfrdrad\
258: (Fig.~\ref{sfrd}) --- \sfrdrad\ should be sensitive to all recent star
259: formation, obscured and unobscured, for a constant IMF --- however,
260: Bell (2003) showed that $L_{\rm far-IR}/L_{\rm UV}$ varies by
261: $\ga$30$\times$ between 0.01 and 3\,L$_{\star}$ and that radio data
262: underestimate SFRs in low-luminosity galaxies typical of those
263: detected locally by {\em GALEX}. Hopkins \& Beacom (2006) argue that
264: for the full picture we should add SFR$_{\rm UV}$ and SFR$_{\rm rad}$.
265: \sfrdrad\ and \sfrduv\ achieve parity at $z\sim\rm 0.4$, after which
266: \sfrdrad\ continues to rise until $z\sim\rm 0.7$ (cf.\ Cowie et al.\
267: 2004) when incompleteness seriously impacts the sample. Although it
268: is tempting to speculate that the rise in \sfrdrad\ results from the
269: increasing dominance of dust-obscured IR-luminous galaxies, we must
270: recall our sample's origins. We are witnessing an increasing SFR per
271: {\em UV-selected} galaxy, partly because at $z\sim\rm 1$ we are
272: probing the most UV-luminous galaxies; we are also witnessing an
273: increase in \sfrdrad\ {\em despite} the increasing
274: incompleteness. Adding SFR$_{\rm UV}$ and SFR$_{\rm rad}$, \sfrd\ due
275: to UV-selected galaxies increases as {\em at least} (1+$z$)$^{2.2}$
276: between $z$ = 0 and 1 (cf.\ Schiminovich et al.\ 2005).
277:
278: The mismatch between the absolute and relative rates of SF derived
279: using UV- and radio-based indicators is worrying, particularly the
280: difference between local estimates of \sfrd. The local \sfrduv\
281: matches the \sfrd\ compilation presented by Hopkins \& Beacom (2006),
282: which implies the UV-selected sample accounts for most of the SF in
283: the local Universe, yet the UV sample at 0.0\,$<$\,$z$\,$<$\,0.2
284: accounts for $\ll$1\% of the total cataloged $S_{\rm 1.4GHz}$ in the
285: region of common areal coverage. If the fraction of $S_{\rm 1.4GHz}$
286: due to obscured SF at 0.0\,$<$\,$z$\,$<$\,0.2 exceeds 1\%, \sfrd\
287: would then be at the upper envelope of commonly accepted values.
288:
289: Moving to slightly higher redshifts, we take two catalogs of EROs.
290: One uses the conventional color cut $R-K_{\rm s}>\rm 5.3$ with $K_{\rm
291: s}<\rm 20.5$ and DEEP2 redshifts, 1.0\,$<$\,$z$\,$<$\,1.5 (Conselice et
292: al., in prep). It contains 382 objects in low-noise areas of our radio
293: image. The second (Wilson et al.\ 2006) uses $R_{\rm AB}-3.6\mu\rm m$
294: $>$ 4 to select essentially the same class of objects, but the larger
295: area covered at 3.6\,$\mu$m (and no requirement for redshifts) yields
296: 2,363 objects in 0.26\,deg$^2$ of AEGIS20. We assumed $z=\rm 1.1$ to
297: excise radio-loud AGN from this sample.
298:
299: Both ERO samples are well detected at 1.4~GHz, as shown in Fig.~2. In
300: the spectroscopic sample, mean $S_{\rm 1.4GHz}$ does not vary
301: significantly as a function of color, though it is a function of
302: $K_{\rm s}$ consistent with the findings of Smail et al.\ (2002).
303: Median $L_{\rm 1.4GHz}$ is $\rm 8.1\times 10^{22}$\,W\,Hz$^{-1}$ and
304: the median SFR per ERO is 92\,$\pm$\,7\,M$_{\odot}$\,yr$^{-1}$. In
305: the sample volume set by the redshift limits and survey area this
306: equates to \sfrd\ = 0.07\,M$_{\odot}$\,yr$^{-1}$\,Mpc$^{-3}$ (cf.\
307: Georgakakis et al.\ 2006; Simpson et al., in prep).
308:
309: The larger ERO sample reveals a weak trend for $S_{\rm 1.4GHz}$ to
310: increase with redness; this is confirmed by the increasing detection
311: rate for individual objects (Table~1). Mean $S_{\rm 1.4GHz}$ also
312: declines as $S_{\rm 3.6 \mu m}$ decreases. We expect $S_{\rm 3.6 \mu
313: m}$ to trace stellar mass and distance, and the factor of $\sim$5.5
314: decrease in $S_{\rm 1.4GHz}$ for a for a factor of $\sim$15 decrease
315: in $S_{\rm 3.6 \mu m}$ suggests an increasing SFR per unit stellar
316: mass as redshift increases. The overall \sfrd\ for this sample is
317: consistent with that of the spectroscopic sample, as expected given
318: the limited number of spectroscopic redshifts and the significant
319: sample overlap.
320:
321: Huang et al.\ (in prep) present a catalog selected at $S_{\rm 24\mu
322: m}>\rm 150\mu$Jy, with $S_{\rm 4.5\mu m}>S_{\rm 3.6\mu m}$, aiming to
323: select galaxies and AGN at $z>$ 1.5. Almost 10$^3$ objects lie in
324: low-noise regions of our radio mosaic, overlapping AEGIS20 by
325: 0.26\,deg$^2$. The individual radio detection rate is a strong
326: function of $S_{\rm 24\mu m}$, rising from 30 to 70\% between $S_{\rm
327: 24\mu m}=$ 0.15 and $>$1.2\,mJy. Median $S_{\rm 1.4GHz}$
328: (Table~\ref{table}) is fairly insensitive to $R_{\rm AB}$, varying by
329: $<$2$\times$ over $>$3 mag. Over 40\% of the 24-$\mu$m-selected
330: galaxies have $L_{\rm 1.4GHz}>\rm 10^{24}$\,W\,Hz$^{-1}$ when assuming
331: $z=\rm 2.75$. At this redshift {\em all} S/N $\ge$ 3 measurements
332: imply radio-loud AGN and it is difficult to estimate the SFR: \sfrd\
333: is likely to be high, but so is the level of accretion-related
334: contamination. The median $S_{\rm 1.4GHz}$, 28\,$\mu$Jy, translates
335: into \sfrd\ = 0.20\,M$_{\odot}$\,yr$^{-1}$\,Mpc$^{-3}$ for $z$ =
336: 1.5--4. We would be unsurprised if this is in error by $\times$2;
337: regardless, this is an important star-forming population.
338:
339: Moving on to yet more distant populations, this time to a sample of
340: DRGs selected at $K_{\rm s}<\rm 20.5$ (Vega) with $J-K_{\rm s}>\rm
341: 2.3$ (Conselice et al.\ 2006), 108 of which lie within the
342: 0.11\,deg$^2$ of common coverage with AEGIS20. Although expected to
343: lie at 1.9\,$<$\,$z$\,$<$\,3.5 (Franx et al.\ 2003), Conselice et al.\
344: find that 64\% lie at 1\,$<$\,$z$\,$<$\,2. Galaxies with $z<\rm 1$,
345: evident via DEEP2, have been removed from the sample used here. One of
346: the radio emitting DRGs has $L_{\rm 1.4GHz}$ consistent with
347: radio-loud AGN (or, as noted earlier, a hyperluminous starburst). The
348: mean $S_{\rm 1.4GHz}$ for the DRGs was $\rm 10.1\pm 1.3\,\mu$Jy --
349: faint emission can be seen in the stacked S/N image
350: (Fig.~\ref{stack}). At $z=\rm 1.5$ this corresponds to $L_{\rm 1.4GHz}
351: = (1.2\pm 0.2)\times 10^{23}$\,W\,Hz$^{-1}$ and a mean SFR (per DRG)
352: of $\rm 150\pm 19$\,M$_{\odot}$\,yr$^{-1}$. Knudsen et al.\ (2005)
353: found $\rm 190\pm 50$\,M$_{\odot}$\,yr$^{-1}$ using submm data for a
354: sample of 30 DRGs (adapting to the cosmology and IMF used here),
355: having assumed significantly larger distances. The observed radio
356: emission from $K_{\rm s}<\rm 20.5$ DRGs equates to \sfrd\ =
357: 0.02\,M$_{\odot}$\,yr$^{-1}$\,Mpc$^{-3}$ at $z$ = 1--2.
358:
359: The LBGs of Steidel et al.\ (2003) lie in a noisy region of the radio
360: mosaic. Of the 334 cataloged LBGs, after correction for the
361: astrometric offset in that catalog ($\Delta\alpha=\rm +0.8'',
362: \Delta\delta=\rm +2.6''$), 107 lie within low-noise regions of the
363: radio mosaic; their mean $S_{\rm 1.4GHz}$ was $\rm 2.0\pm 2.3\,\mu$Jy
364: (median, $-$0.6\,$\mu$Jy), consistent with an average SFR of
365: $<$500\,M$_{\odot}$\,yr$^{-1}$ (3\,$\sigma$, for $z=\rm
366: 3$). Restricting the catalog to the 53 LBGs detected at 8\,$\mu$m with
367: IRAC did not change the situation significantly (cf.\ Rigopoulou et
368: al.\ 2006).
369:
370: Finally, Huang et al.\ (2005) describe a population of IR-luminous
371: LBGs (ILLBGs) detected at $S_{\rm 24\mu m}>\rm 60\mu$Jy. Only six of
372: Huang et al.'s 13 objects lie within our radio mosaic. Their median
373: $S_{\rm 1.4GHz}$ is 44.2\,$\mu$Jy, including the one significant
374: detection: Westphal MD99 at $\sim$1\,mJy. This provides tentative
375: support for the assertion that ILLBGs share the high SFRs of submm
376: galaxies, though this is a very small sample in a particularly noisy
377: region of the radio mosaic and accretion-related contamination is
378: possible. If, as Huang et al.\ suggest, ILLBGs lie at
379: 2\,$<$\,$z$\,$<$\,3 (like submm galaxies -- Chapman et al.\ 2005) then
380: their \sfrd\ is similar to that of the 24$\mu$m-selected galaxies with
381: which they will overlap significantly (Fig.~\ref{sfrd}).
382:
383: %
384: %
385: %
386: \section{On radio data as a probe of global SF history}
387:
388: Fig.~\ref{sfrd} shows \sfrd\ for the galaxy populations explored in
389: \S4. The upper envelope of points traces the minimum \sfrd\ as a
390: function of redshift and appears to rise by at least 5$\times$ from
391: $z$ = 0 to 1, a now-familiar pattern (Lilly et al.\ 1996), though this
392: work has led us to question the reliability of many SFR and \sfrd\
393: estimates.
394:
395: Radio-based SFR estimates become increasingly prone to contamination
396: by radio-loud AGN at $z\gg\rm 1$. Unfortunately, a consequence of
397: removing this via a limit on $L_{\rm 1.4GHz}$ is the rejection of
398: luminous star-forming galaxies obeying the far-IR/radio correlation;
399: adopting a median $S_{\rm 1.4GHz}$ is unlikely to be better. In
400: addition, some redshift-limited galaxy populations defined by color
401: appear less well defined than first claimed (Conselice et al.\ 2006),
402: limiting our ability to judge the volume probed. These effects lead to
403: large uncertainties so while it is clear that stacking radio data is
404: useful, accurate SFRs for distant galaxies require precise redshifts
405: together with deep, multi-frequency, high-resolution radio data
406: ($\ll$1$''$, $\sigma_{\rm 0.6GHz}\sim \sigma_{\rm 1.4GHz}\la
407: 1\,\mu$Jy). These will facilitate identification and removal of
408: accretion-related emission via radio luminosity, spectral index,
409: brightness temperature and morphology.
410:
411: %% After the acknowledgments section, use the following syntax and the
412: %% \facility{} macro to list the keywords of facilities used in the research
413: %% for the paper. Each keyword will be checked against the master list during
414: %% copy editing. Individual instruments or configurations can be provided
415: %% in parentheses, after the keyword, but they will not be verified.
416:
417: %{\it Facilities:} \facility{Very Large Array}.
418:
419: \begin{thebibliography}{}
420:
421: \bibitem[Arnouts et al.(2005)]{arn05} Arnouts, S., et al.\ 2005, \apj, 619, L43
422:
423: \bibitem[Bell(2003)]{bell03} Bell, E.F.\ 2003, \apj, 586, 794
424:
425: \bibitem[Biggs \& Ivison(2006)]{bi06} Biggs, A.D., \& Ivison, R.J.\ 2006, \mnras,
426: 371, 963
427:
428: \bibitem[Bondi et al.(2003)]{bon03} Bondi, M., et al.\ 2003, A\&A, 403, 857
429:
430: \bibitem[Chapman et al.(2005)]{cha05} Chapman, S.C., Blain, A.W.,
431: Smail, I., \& Ivison, R.J.\ 2005, \apj, 622, 772
432:
433: \bibitem[Condon(1992)]{con92} Condon, J.J.\ 1992, \araa, 315, 575
434:
435: \bibitem[Conselice et al.(2006)]{con06} Conselice, S.C., et al.\ 2006, \apj,
436: in press (this issue)
437:
438: \bibitem[Cowie et al.(2004)]{cow04} Cowie, L.L., Barger, A.J., Fomalont, E.B., \&
439: Capak, P.\ 2004, \apj, 603, L69
440:
441: \bibitem[Davis et al.(2006)]{dav06} Davis, M., et al.\ 2006, \apj, in press (this issue)
442:
443: \bibitem[Dole et al.(2006)]{dol06} Dole, H., et al.\ 2006, A\&A, 451, 417
444:
445: \bibitem[Fanaroff \& Riley(1974)]{fr04} Fanaroff, B.L., \& Riley, J.M.\ 1974,
446: \mnras, 167, 31
447:
448: \bibitem[Fixsen et al.(1998)]{fix98} Fixsen, D.J., Dwek, E., Mather, J.C., Bennett, C.L.,
449: \& Shafer, R.A.\ 1998, \apj, 508, 123
450:
451: \bibitem[Fomalont et al.(1991)]{fom91} Fomalont, E.B., Windhorst, R.A., Kristian, J.A.,
452: \& Kellerman, K.I.\ 1991, \aj, 102, 1258
453:
454: \bibitem[Franx et al.(2003)]{fra03} Franx, M., et al.\ 2003, \apj, 587, L79
455:
456: \bibitem[Garrett(2002)]{gar02} Garrett, M.A.\ 2002, A\&A, 384, L19
457:
458: \bibitem[Georgakakis et al.(2006)]{geo06} Georgakakis, A., Hopkins, A.M., Afonso, J.,
459: Sullivan, M., Mobasher, B., \& Cram, L.E.\ 2006, \mnras, 367, 331
460:
461: \bibitem[Helou et al.(1985)]{hel85} Helou, G., Soifer, B.T., \& Rowan-Robinson, M.\
462: 1985, \apj, 298, L7
463:
464: \bibitem[Hopkins et al.(2003)]{hop03} Hopkins, A.M., Afonso, J., Chan, B., Cram, L.E.,
465: Georgakakis, A., \& Mobasher, B.\ 2003, \aj, 125, 465
466:
467: \bibitem[Hopkins \& Beacom(2006)]{hb06} Hopkins, A.M., \& Beacom, J.F.\ 2006,
468: \apj, in press (astro-ph/0601463)
469:
470: \bibitem[Huang et al.(2005)]{hua05} Huang, J.-S., et al.\ 2005, \apj, 634, 137
471:
472: \bibitem[Knudsen et al.(2005)]{knu06} Knudsen, K.K., et al.\ 2005, \apj, 632, L9
473:
474: \bibitem[Kov\'acs et al.(2006)]{kov06} Kov\'acs, A., Chapman, S.C., Dowell, C.D.,
475: Blain, A.W., Ivison, R.J., Smail, I., \& Phillips, T.G.\ 2006, \apj, in press
476: (astro-ph/0604591)
477:
478: \bibitem[Lilly et al.(1996)]{lil06} Lilly, S.J., Le Fevre, O., Hammer, F., \&
479: Crampton, D., 1996, \apjl, 460, L1
480:
481: \bibitem[Rigopoulou et al.(2006)]{rig06} Rigopoulou, D., et al.\ 2006, \apj, 648, 81
482:
483: \bibitem[Salim et al.(2005)]{2005ApJ...619L..39S} Salim, S., et al.\ 2005,
484: \apjl, 619, L39
485:
486: \bibitem[Salim et al.(2006)]{sal06} Salim, S., et al.\ 2006, \apj, in press
487: (this issue)
488:
489: \bibitem[Schiminovich et al.(2005)]{sch05} Schiminovich, D., et al.\ 2005, \apj, 619, L47
490:
491: \bibitem[Simpson et al.(2006)]{sim06} Simpson, C., Rawlings, S., Mart\'inez Sansigre,
492: A., Ivison, R.J., Sekiguchi, K., \& Takata, T.\ 2006, \mnras, in press
493: (astro-ph/0609529)
494:
495: \bibitem[Smail et al.(2002)]{sma02} Smail, I., Owen, F.N., Morrison, G.E., Keel, W.C.,
496: Ivison, R.J., \& Ledlow, M.J.\ 2002, \apj, 581, 844
497:
498: \bibitem[Spergel et al.(2003)]{spe03} Spergel, D.N., et al.\ 2003, \apjs, 148, 175
499:
500: \bibitem[Steidel et al.(2003)]{ste03} Steidel, C.C., Adelberger, K.L., Shapley,
501: A.E., Pettini, M., Dickinson, M., \& Giavalisco, M.\ 2003, \apj, 592, 728
502:
503: \bibitem[Wilson et al.(2006)]{wil06} Wilson G., et al.\ 2006, \apj, in press
504: (this issue)
505: \end{thebibliography}
506:
507: \clearpage
508:
509: \begin{figure}
510: \epsscale{0.5}
511: \plotone{f1.eps}
512: \caption{Radio-based estimates of SFR density (\sfrd) for a number of
513: independent and overlapping galaxy populations selected via magnitude
514: and color criteria: UV-selected galaxies (open and filled circles for
515: UV- and radio-based SFRs, joined by dotted and solid lines), $R-K_{\rm
516: s}>\rm 5.3$ EROs (triangle), DRGs (star), ILLBGs (diamond) and
517: 24-$\mu$m-selected galaxies (filled square). These are conservative
518: estimates -- no attempt has been made to correct for accessible
519: volume; contamination by radio-loud AGN is possible at $z\gg\rm 1$
520: (\S5). The upper envelope of points thus traces the minimum \sfrd\ as
521: a function of redshift, as demonstrated by the compilation of \sfrd\
522: data from Hopkins \& Beacom (2006), plotted faintly here.
523: \label{sfrd}}
524: \end{figure}
525:
526: \clearpage
527:
528: \begin{figure}
529: \epsscale{0.5}
530: \plotone{f2.eps}
531: \caption{Stacked S/N images ($\rm 33''\times 33''$) at the positions
532: of the galaxies described in \S4: conventional and IR-luminous LBGs
533: (top row); DRGs and $R-K_{\rm s}>\rm 5.3$ EROs (middle row);
534: IRAC-selected EROs and MIPS-selected galaxies with $S_{\rm 4.5\mu
535: m}>S_{\rm 3.6\mu m}$ (bottom row). Areas of high noise and galaxies
536: detected individually ($\ge$5$\sigma$) were excluded. Contours are
537: plotted at S/N levels of $-3,3,4...10,20...100$ and the greyscale is
538: identical in each case. The number of stacked sources from each
539: population is shown in parentheses. The detection of the IRAC-selected
540: ERO population is so significant that we see
541: secondary structure: the image resembles the dirty beam
542: since none of the individually undetected sources in the ensemble that
543: makes up the stacked image have been {\sc clean}ed.
544: \label{stack}}
545: \end{figure}
546:
547: \clearpage
548:
549: \include{tab1}
550:
551: \end{document}
552: