0808.2391/ms.tex
1:  %%
2:  %% This is a sample manuscript marked up using the
3:  %% AASTeX v5.x LaTeX 2e macros.
4: 
5: \documentclass[12pt,preprint]{aastex}
6: \usepackage{psfig}
7:  
8:  %% manuscript produces a one-column, double-spaced document:
9:  
10:  %\documentclass[manuscript]{aastex}
11:  
12:  %% preprint2 produces a double-column, single-spaced document:
13:  
14:  %% \documentclass[preprint2]{aastex}
15:  
16:  %% Sometimes a paper's abstract is too long to fit on the
17:  %% title page in preprint2 mode. When that is the case,
18:  %% use the longabstract style option.
19:  
20:  %% \documentclass[preprint2,longabstract]{aastex}
21: %\usepackage{graphicx}
22: %\usepackage{epstopdf}
23: \usepackage{psfig}
24:  %% If you want to create your own macros, you can do so
25:  %% using \newcommand. Your macros should appear before
26:  %% the \begin{document} command.
27:  %%
28:  %% If you are submitting to a journal that translates manuscripts
29:  %% into SGML, you need to follow certain guidelines when preparing
30:  %% your macros. See the AASTeX v5.x Author Guide
31:  %% for information.
32:  
33:  \newcommand{\vdag}{(v)^\dagger}
34:  \newcommand{\myemail}{ccarilli@nrao.edu}
35: %\newcommand{\etal}{\mbox{et al.\}}
36: 
37: \newcommand{\etal}{\mbox{et~al.}}
38: \newcommand{\Msol}{\mbox{$M_{\odot}$}}
39: \newcommand{\msun}{\mbox{$M_{\odot}$}}
40: \newcommand{\Msun}{\mbox{${\bf M_{\odot} }$}}
41: \newcommand{\Lsol}{\mbox{$L_{\odot}$}}
42: \newcommand{\lsun}{\mbox{$L_{\odot}$}}
43: \newcommand{\ergsec}{\mbox{erg s$^{-1}$}}
44: \newcommand{\beam}{\mbox{beam$^{-1}$}}
45: \newcommand{\Mdyn}{\mbox{$M_{\rm dyn}$}}
46: \newcommand{\Mgas}{\mbox{$M_{\rm gas}$}}
47: \newcommand{\HH}{\mbox{H$_2$}}
48: \newcommand{\Mgasfive}{\mbox{$M_{\rm gas}(r<500{\rm pc})$ }}
49: \newcommand{\Halpha}{\mbox{H$\alpha$} }
50: \newcommand{\tm}[1]{\tablenotemark{#1}}
51: \def\deg      {{\ifmmode^\circ\else$^\circ$\fi}} %%% Overwrites TeX \deg
52: \def\Ho       {{$H_{0}$} }
53: \def\qo       {{$q_{0}$} }
54: \def\kmsMpc   {{\ km\ s$^{-1}$\ Mpc$^{-1}$} }
55:  
56: %% You can insert a short comment on the title page using the command below.
57:  
58: \slugcomment{to appear in the Astrophysical Journal}
59:  
60:  %% If you wish, you may supply running head information, although
61:  %% this information may be modified by the editorial offices.
62:  %% The left head contains a list of authors,
63:  %% usually a maximum of three (otherwise use et al.).  The right
64:  %% head is a modified title of up to roughly 44 characters.
65:  %% Running heads will not print in the manuscript style.
66:  
67: \shorttitle{Star formation rates in Lyman break galaxies.}
68: \shortauthors{Carilli et al.}
69:  
70:  %% This is the end of the preamble.  Indicate the beginning of the
71:  %% paper itself with \begin{document}.
72:  
73:  \begin{document}
74:   
75:  \title{Star formation rates in Lyman break galaxies: radio stacking of 
76: LBGs in the COSMOS field and the sub-$\mu$Jy radio source population.}
77:  
78: \author{ 
79: C.L. Carilli\altaffilmark{1},
80: Nicholas Lee\altaffilmark{1},
81: P. Capak\altaffilmark{2},
82: E. Schinnerer\altaffilmark{3},
83: K.-S. Lee\altaffilmark{4},
84: H. McCraken\altaffilmark{5},
85: M.S. Yun\altaffilmark{4},
86: N. Scoville\altaffilmark{2},
87: V. Smol{\v c}i{\' c}\altaffilmark{2},
88: M. Giavalisco\altaffilmark{4},
89: A. Datta\altaffilmark{1},
90: Y. Taniguchi\altaffilmark{6}
91: C. Megan Urry\altaffilmark{7,8}
92: }
93: 
94: \altaffiltext{$\star$}{Based on observations in the COSMOS Legacy Survey 
95: including those taken on the HST, Keck, NRAO-VLA, Subaru, KPNO 4m, CTIO 4m,
96: and CFHT 3.6m. The Very Large Array of the National Radio Astronomy
97: Observatory, is a facility of the National Science Foundation
98: operated under cooperative agreement by Associated Universities, Inc}
99: 
100: %
101: \altaffiltext{1}{National Radio Astronomy Observatory, P.O. Box 0, Socorro, NM
102: 87801-0387}
103: %
104: \altaffiltext{2}{California Institute of Technology, MC 105-24, 1200 East
105: California Boulevard, Pasadena, CA 91125}
106: %
107: \altaffiltext{3}{Max Planck Institut f\"ur Astronomie, K\"onigstuhl
108:   17, Heidelberg, D-69117, Germany}
109: %
110: \altaffiltext{4}{Dept. of Astronomy, University of Massachusetts, Amherst, MA}
111: %
112: \altaffiltext{5}{IAP, Paris, France}
113: %
114: \altaffiltext{6}{Graduate School of Science and Engineering, Ehime University, 
115: Bonkyo-cho, Matsuyama, 790-8577m Japan}
116: %
117: \altaffiltext{7}{Department of Astronomy, Yale University, New Haven, CT 06511, USA}
118: \altaffiltext{8}{Yale Center for Astronomy and Astrophysics, Yale University, P.O.~Box 208121, New Haven, CT 06520}
119:  
120: \begin{abstract}
121: 
122: We present an analysis of the radio properties of large samples of
123: Lyman Break Galaxies (LBGs) at $z \sim 3$, 4, and 5 from the COSMOS
124: field.  The median stacking analysis yields a statistical detection of
125: the $z \sim 3$ LBGs (U-band drop-outs), with a 1.4 GHz flux density of
126: $0.90 \pm 0.21 \mu$Jy. The stacked emission is unresolved, with a size
127: $< 1"$, or a physical size $< 8$kpc.  The total star formation rate
128: implied by this radio luminosity is $31\pm 7$ $M_\odot$ year$^{-1}$,
129: based on the radio-FIR correlation in low redshift star forming
130: galaxies.  The star formation rate derived from a similar analysis of
131: the UV luminosities is 17 $M_\odot$ year$^{-1}$, without any
132: correction for UV dust attenuation.  The simplest conclusion is that
133: the dust attenuation factor is 1.8 at UV wavelengths. However, this
134: factor is considerably smaller than the standard attenuation factor
135: $\sim 5$, normally assumed for LBGs. We discuss potential reasons for
136: this discrepancy, including the possibility that the dust attenuation
137: factor at $z \ge 3$ is smaller than at lower redshifts.  Conversely,
138: the radio luminosity for a given star formation rate may be
139: systematically lower at very high redshift. Two possible causes for a
140: suppressed radio luminosity are: (i) increased inverse Compton cooling
141: of the relativistic electron population due to scattering off the
142: increasing CMB at high redshift, or (ii) cosmic ray diffusion from
143: systematically smaller galaxies.  The radio detections of individual
144: sources are consistent with a radio-loud AGN fraction of 0.3\%. One
145: source is identified as a very dusty, extreme starburst galaxy (a
146: 'submm galaxy').
147: 
148: \end{abstract}
149:  
150:  \keywords{galaxies: formation --- galaxies: evolution --- galaxies:
151: radio --- surveys}
152:   
153: \section{Introduction}
154: 
155: The power of discovering high redshift galaxies via the broad-band
156: drop-out technique, ie. Lyman Break galaxies (LBGs), is now well
157: established (Steidel et al. 1999; 2000). Using this technique, over a
158: thousand galaxies have now been detected at $z > 2$.  Detailed studies
159: show that these galaxies have stellar masses between 10$^{10}$ and
160: 10$^{11}$ $M_\odot$, and a (comoving) volume density at $z \sim 3$ of
161: $\sim 0.005$ Mpc$^{-3}$ (Giavalisco 2002).
162: 
163: A number of issues are still being investigated for LBGs. One
164: important issue is deriving the total star formation rate. Besides
165: model parameters such as the IMF and star formation history, the dust
166: correction in the UV remains under investigation.  Steidel et
167: al. (1999) originally estimated a typical UV dust attenuation factor
168: of about 5, based on optical spectroscopy.  Adelberger \& Steidel
169: (2000) applied a similar method to a larger sample of LBGs, as well as
170: observations at other wavebands (radio, submm), and conclude: "...the
171: mean extinction at 1600 \AA ~ for LBGs, a factor of 6 in our best
172: estimate, could lie between a factor 5 and a factor of 9."  More
173: recently, Reddy \& Steidel (2004) have derived the average dust
174: correction factors for UV-selected galaxies at $z \sim 2$ based on
175: deep X-ray, radio, and optical spectroscopic studies of galaxies in the
176: GOODS north field. They find that, for galaxies with total star
177: formation rates $> 20$ $M_\odot$ year$^{-1}$, the dust attenuation
178: factor is between 4.4 and 5.1.
179: 
180: A second issue for LBGs is the AGN fraction. Shapley et al. (2003)
181: found that $\sim 3\%$ of LBGs at z$\sim$3 have optical emission line
182: spectra that are consistent with AGN.
183: 
184: The Cosmic Evolution Survey (COSMOS), covering 2 \sq\deg, is a
185: comprehensive study of the evolution of galaxies, AGN, and dark matter
186: as a function of their cosmic environment. The COSMOS field has
187: state-of-the-art multiwavelength observations, ranging from the radio
188: through the X-ray (Scoville et al. 2007; Capak et al. 2007). A major
189: part of this study is to identify the largest samples of LBGs to date.
190: In section 2 we describe the LBG samples from the COSMOS survey based
191: on U, B, and V-band drop-out searches.
192: 
193: In this paper we consider the radio properties of these large samples
194: of LBGs, similar to the study of $z = 5.7$ Lyman-$\alpha$ emitting
195: galaxies by Carilli et al. (2007). Observations of the COSMOS field
196: have been done at 1.5$"$ resolution (FWHM) at 1.4 GHz with the Very
197: Large Array (Schinnerer et al.  2007). We employ the new VLA image
198: that has added integration time in the central $\sim 1^o$ (Schinnerer
199: in prep.), giving an rms at the field center of $\sim 7$ $\mu$Jy
200: beam$^{-1}$.  In section 3.1 we search for radio counterparts to
201: individual LBGs, while in section 3.2 we perform a stacking analysis
202: to derive the statistical properties of the samples.  In section 4 we
203: discuss the implications these observations have on the questions of
204: dust extinction and the AGN fraction in LBGs. We adopt a standard
205: concordance cosmology.
206:  
207: \section{A conservative COSMOS LBG sample}
208: 
209: We have identified large samples of U,B, and V-band drop-out galaxies
210: using the COSMOS photometric data (Capak et al. 2007).  For an
211: initial, robust analysis of the radio properties of high z LBGs, we
212: have adopted a series of conservative criteria to identify LBGs at the
213: different redshifts.  The U, B, and V-band drop selection criteria are
214: give in Table 1, where u' = CFHT u, i' = Subaru i, and r' = Subaru r
215: (Lee et al. in prep.).
216: 
217: It is well documents that LBG samples can include contaminants
218: (see Steidel et al. 2000, Capak et al. 2004, Bouwens et
219: al. 2007).  In the U and B-band drop-out selection, the contamination
220: comes from A and M stars along with $z<0.3$ star forming galaxies,
221: while the V-band drop-out selection is contaminated with M and T stars
222: along with $1.5<z<2.5$ obscured galaxies.  Since these contaminating
223: objects can be sources of radio emission, we take several steps to
224: produce a clean sample of high redshift objects.  We begin by removing
225: all ACS point sources with a star- or AGN-like spectral energy
226: distribution using the star/galaxy/AGN classifier described in Robin
227: et al. (2007).  In addition we remove all stars identified by a BzK
228: diagram (Daddi et al. 2007) at $K_{AB}<22$, this removes very red
229: stars which may not be identified by the Robin et al. (2007)
230: classifier.  We then use the BPZ photometric redshifts from Mobasher
231: \& Mazzei (2000) to remove objects with a best fit photo-z at $z<2.5$,
232: $z<3.5$, and $z<4.5$ for the U, B, and V-band drop-outs, respectively.
233: Finally we remove objects with a $>5$\% probability of being at
234: $z\leq1$.
235: 
236: The resulting sample is largely clear of contaminants.  Redshifts
237: were obtained with DEIMOS on Keck-II using the 830l/mm grating and
238: 3.5-4h of integration time for a sample of color selected objects with
239: $i<26$ and $0.6<z<1.4$ or $z>4$ (Capak et al. in Prep).  The
240: spectroscopic sample matches 2 U-band drop-outs, 18 B-band drop-outs,
241: and 7 V-band drop-outs in our cleaned LBG sample.  Of these objects,
242: only 1 V-band drop-out object, which is a faint M star, is not in the
243: expected redshift range, yielding a contamination fraction of
244: $4\pm4$\%.  However, the U-band drop-outs sample was not specifically
245: targeted, so the constraint on the contamination fraction for these
246: objects is weak.  Of the remaining 26 objects at $z>3$, one B-band
247: drop-out shows an AGN signature, implying a residual AGN fraction of
248: $4\pm4$\%.
249: 
250: Although our sample is clean, it is not complete.  In general, the
251: phot-z cut creates a bias towards galaxies which have good photometry,
252: and closely match the expected UV-Optical spectral templates used for
253: the photometric redshifts.  Objects with unusual spectra, such as
254: AGN/galaxy composites, or objects with reddening that does not conform
255: to the Calzetti et al. (1997) or Milky Way laws, will likely be
256: excluded.  Furthermore, we have explicitly excluded optically bright
257: broad line AGN, and the probability requirements systematically bias
258: us against faint ($i<25$) objects.  The flux bias occurs because faint
259: objects have broad probability distributions due to their poor
260: photometric measurements.
261: 
262: We are also potentially biased against heavily obscured objects which
263: will have fainter UV fluxes, and hence poor phot-z fits.  An effort to
264: use intermediate band photometry (Ilbert et al. in prep) and a
265: systematic deep spectroscopic follow-up (Lilly et al. 2007, Capak et
266: al. in prep) is underway to fully characterize this sample and
267: understand the biases.
268: 
269: Table 2 summarizes the current samples. Column 2 indicates the
270: redshift range implied for a given drop-out band. Column 3 shows the
271: number of sources in the sample. Even with these conservative
272: selection criteria, we identify thousands of high redshift LBGs due to
273: the very wide COSMOS field.  In this initial study, we feel it
274: important to work with such a conservative sample of sources, in order
275: to make robust conclusions concerning the radio properties of LBGs.
276: 
277: For reference, the radio luminosity density at a rest frame frequency
278: of 1.4 GHz, for a source with $S_{\rm 1.4GHz} = 60\mu$Jy, assuming a
279: spectral index of $-0.8$ (Condon 1992), at $z \sim 3$ is: $L_{\rm
280: 1.4GHz} = 3 \times 10^{31}$ ergs s$^{-1}$ Hz$^{-1}$.  For comparison,
281: the nearby radio AGN source M87 has $L_{\rm 1.4GHz} = 9\times 10^{31}$
282: erg s$^{-1}$ Hz$^{-1}$, while the luminous starburst galaxy Arp 220
283: has $L_{\rm 1.4GHz} = 4\times 10^{30}$ erg s$^{-1}$ Hz$^{-1}$.  In
284: M87, the radio emission is powered by a relativistic jet emanating
285: from the supermassive black hole with $M_{BH} \sim 2.5\times 10^9$
286: $M_\odot$ (Gebhardt et al. 2000), while in Arp 220 the radio emission
287: is driven by star formation with a total star formation rate $\sim
288: 500$ $M_\odot$year$^{-1}$ (Carilli \& Yun 2000; Yun \& Carilli
289: 2001). Even at the sensitivity of the COSMOS survey, we are limited to
290: detecting either radio AGN sources, or extreme starburst galaxies,
291: ie. galaxies with star formation rates $>$ a few thousand $M_\odot$
292: year$^{-1}$.
293: 
294: \section{The Radio properties of Lyman Break Galaxies}
295: 
296: \subsection{Individual sources}
297: 
298: We have searched for radio emission from each of the LBGs in the
299: COSMOS field.  At each position we determine the surface brightness,
300: and the rms noise in the region. The median rms in the LBG fields was
301: $\sim 15\mu$Jy beam$^{-1}$ for all three samples, implying a median
302: $4\sigma$ detection limit of 60$\mu$Jy beam$^{-1}$.  Our radio images
303: have a synthesized beam FWHM = 1.5'', hence we have searched for radio
304: sources brighter than 4$\sigma$, within 1$''$ radius of the LBG
305: optical position.  Column 4 in Table 2 lists the number of 4$\sigma$
306: radio counterparts in each sample.  Roughly, for each drop-out band,
307: we detect radio counterparts for $\sim 0.6\%$ of the UV sources.
308: 
309: We need to correct for the number of detections we expect by chance in
310: such a survey, either due to image noise, or faint sources unrelated
311: to the LBG.  We calculate this correction factor by performing
312: searches with identical criteria as for the LBG positions, but at
313: random positions selected to be $20''$ away from the coordinates of
314: the LBG samples.  This should give us a random sampling of the COSMOS
315: field, with noise statistics that match the LBG regions.  We find 19
316: 4$\sigma$ random detections in the U-band drop-out fields, 4 for the
317: B-band drop-outs, and 3 for the V-band drop-outs.  These correspond to
318: 0.29\%, 0.28\%, and 0.49\%, respectively.  We apply these corrections
319: in column 5 of Table 2.
320:  
321: Overall, we detect radio counterparts to $\sim 0.6\%$ of the UV
322: sources, with the expected number of detections at random being $\sim
323: 0.3\%$. These results imply that only a small fraction, $\sim 0.3\%$,
324: of the LBGs, are radio sources brighter than $\sim 4\sigma$,
325: where, on average $4\sigma \sim 60\mu$Jy.
326: 
327: \subsection{A stacking analysis}
328: 
329: We investigate the statistical radio properties of the LBGs through a
330: stacking analysis. We adopt the median stacking method used by White
331: et al. (2007) in their stacking analysis based on the FIRST radio
332: survey.  White et al. perform detailed calculations that show that a
333: median stacking analysis is superior to a mean stacking, since it is
334: robust to small numbers of bright sources, and it does not require any
335: maximum allowed flux density cut-off prior to stacking.  Moreover, they
336: show that: '...in a limit where almost all of the values in the sample
337: are small compared to the noise, it is straight-forward to interpret
338: the median stack measurements as representative of the mean for the
339: population.' The COSMOS sample obeys this criterion.
340: 
341: Figure 1 shows the images derived by a median stacking of all the
342: sources in each drop-out band. These images are derived by extracting
343: sub-images centered around each LBG with a size of $18''\times
344: 18''$. We then 'stack' these sub-images, aligned at the center
345: (on-source) position, and derive the median surface brightness at each
346: pixel position.  The off-source pixels serve as test positions for a
347: random stacking analysis. We find that the mean values for the off
348: positions equal zero, to within the noise, as expected for random
349: positions. We also derive the statistical noise of the analysis from
350: these off-positions.
351: 
352: Column 6 in Table 2 lists the peak surface brightnesses derived at the
353: stacked LBG position, plus the rms off-source noise, for each sample.
354: We obtain a clear detection of radio emission in the U-band drop-out
355: sample, with a value of $0.9\pm0.21\mu$Jy. 
356: The other bands show marginal detections, although at $<
357: 3\sigma$ value, and hence we consider these upper limits.
358: 
359: Our results are consistent with the upper limit found by Ivison et
360: al. (2007), who performed a 1.4 GHz stacking analysis of 107 $z \sim
361: 3$ LBGs in the extended Groth strip (the AEGIS survey), for which they
362: obtained a median value of $-0.6 \pm 2.3\mu$Jy.
363: 
364: The stacked source is unresolved, with a size $< 1"$. The implied
365: physical size is $< 8$kpc, hence the typical source must be galaxy
366: size, or smaller.
367: 
368: \section{Discussion}
369: 
370: \subsection{The AGN fraction and massive starbursts}
371: 
372: At the sensitivity limits of current deep radio surveys, such as
373: COSMOS, detection of individual sources at $z > 2$ is limited to
374: either extreme starbursts (star formation rates $> 1000$ $M_\odot$
375: year$^{-1}$), or radio jet sources with luminosities comparable to
376: M87.
377: 
378: After correcting for random detections, we obtain a detection rate of
379: $\sim 0.3\%$.  Shapley et al. (2003) find an AGN fraction in LBG
380: samples of $\sim 3\%$, based on optical spectroscopy.  If all of
381: our detected sources are radio AGN, then the implied radio loud
382: fraction is $\sim 10\%$. Interestingly, a value of 10\% is the
383: canonical value for radio loud AGN based on studies of nearby
384: galaxies (eg. Ivezic et al. 2002; Petric et al. 2007).  
385: 
386: We emphasize caution in interpreting our results on the radio loud
387: fraction of AGN at high redshift, for a number reasons. First, in a recent
388: comparison of the FIRST and SDSS surveys, Jiang et al. (2007) find
389: that the radio loud fraction likely depends on both optical luminosity
390: and redshift. This study is not directly comparable to our COSMOS
391: study, since they were limited to much more luminous sources at high
392: redshift, by about two orders of magnitude in the radio.
393: 
394: Second, the above analysis does not consider the possibility that
395: some of the 0.3\% of the radio detections are extreme, highly obscured
396: starbursts, comparable to the submm galaxies. Indeed, in a companion
397: paper, we present the discovery of a submm galaxy in our
398: radio-selected LBG COSMOS sample (Capak et al. 2008). The
399: radio-detected LBG has a spectroscopic redshift of $z =4.5$, and a
400: radio flux density of $45\pm 10\mu$Jy. Thermal emission from warm dust
401: is detected from this galaxy at 250GHz using the  MAMBO
402: bolometer camera, with a flux density of $3.4\pm 0.7$mJy.  This
403: source is the first submm galaxy yet identified (spectroscopically) at
404: $z > 4$ (Capak et al. 2008).
405: 
406: And third are the caveats mentioned in section 2 concerning our
407: conservative source selection criteria, eg. excluding obvious
408: optically bright broad line AGN.
409: 
410: \subsection{The sub-$\mu$Jy radio source population}
411: 
412: The unprecedented size of the COSMOS LBG sample has allowed us to
413: reach sub-$\mu$Jy sensitivity levels in the stacking analysis of the
414: LBG samples. We obtain a clear statistical detection of $S_{\rm 1.4GHz} =
415: 0.90\pm 0.21\mu$Jy for the U-band ($z \sim 3$) drop-outs. The implied
416: rest frame 1.4 GHz luminosity density is $L_{\rm 1.4\rm GHz} = 5.1\times
417: 10^{29}$ erg s$^{-1}$ Hz$^{-1}$, assuming a spectral index of $-0.8$
418: (Condon 1992).
419: 
420: Assuming the radio emission is driven by star formation, we can derive
421: a total star formation rate (0.1 to 100 $M_\odot$), from the rest
422: frame 1.4 GHz luminosity density. There are a number of recent
423: calibrations of this relationship for nearby galaxies using, eg.  the
424: IRAS and NVSS surveys.  These studies adopt the star formation rates
425: derived from the far-IR emission based on the relations in Kennicutt
426: (1998), assuming a Salpeter IMF, and then calibrate the radio
427: conversion factor using the tight radio to far-IR correlation for star
428: forming galaxies.  We adopt the conversion factor from the study of
429: Yun, Reddy, and Condon (2001): 
430: 
431: $$\rm SFR = 5.9 \pm 1.8 \times 10^{-29} {\sl L}_{\rm 1.4GHz} ~
432: M_\odot~ year^{-1}, $$
433: 
434: \noindent where $L_{\rm 1.4\rm GHz}$ is in erg s$^{-1}$ Hz$^{-1}$.
435: From this, we calculate a mean star formation rate of $31\pm 7$
436: $M_\odot$ year$^{-1}$. Note that the conversion factor above is within
437: 10\% of that derived by Bell (2003).
438: 
439: The mean observed UV luminosity at a rest frame wavelength of 1600 \AA
440: ~of the U-band drop-out sample is: $L_{2000A} = 1.2\times 10^{29}$ erg
441: s$^{-1}$ Hz$^{-1}$ (Capak et al. 2008 in prep).  Using equation (1) in
442: Kennicutt (1998) for the relationship between UV luminosity and total
443: star formation rate, we derive a total star formation rate of 17
444: $M_\odot$ year$^{-1}$, uncorrected for dust attenuation.
445: 
446: The ratio of radio derived star formation rate to UV derived star
447: formation rate is $1.8\pm 0.4$.  The simplest conclusion is that the
448: UV emission is attenuated by dust by a factor of 1.8. However, this
449: factor is considerably smaller than the standard factor $\sim 5$
450: adopted for LBGs (see Section 1). We consider some possible reasons
451: for this difference.
452: 
453: One possibility is that the dust attenuation factor for the $z \sim 3$
454: COSMOS LBG sample is indeed smaller than for other LBG samples. Most
455: of the studies of dust attenuation of LBG galaxies have been at $z <
456: 3$ (see section 1), and hence it is possible that at higher redshift
457: the dust attenuation factor decreases. An argument against this
458: decrease is the recent X-ray and optical study of a sample of LBGs at
459: $z \sim 3$ by Nandra et al. (2002), who also derive a dust attenuation
460: factor of about 5. We also re-emphasize that the COSMOS LBG sample
461: analyzed herein de-selected sources with poor phot-z fits, which can
462: occur for very heavily obscured objects (Section 2).
463: 
464: On the other hand, Wilkins, Trentham, \& Hopkins (2008) present a
465: detailed comparison of the build up of stellar mass in galaxies versus
466: the cosmic star formation rate density. They conclude that there is a
467: discrepancy between the rate of stellar mass creation, and the star
468: formation rates derived from the UV luminosities, at high redshift, if
469: one assumes the standard factor $\sim 5$ dust attenuation. The
470: discrepancy is in the sense that the UV derived star formation rates
471: are too high. They find that the peak in the discrepancy occurs at $z
472: \sim 3$, and they suggest that the UV-derived star formation rates at
473: this redshift may be over-estimated by a factor $\sim 4$. They also
474: point out that '...this large deviation at high redshift offers an
475: explanation for why the integrated star formation history implies a
476: local stellar mass density in excess of that measured.'
477: 
478: A second possibility is that the conversion factor of radio luminosity
479: to star formation rate is higher in the $z \sim 3$ LBG COSMOS sample
480: than has been derived for low redshift galaxies.  Radio studies of
481: 24$\mu$m selected galaxies by Appleton et al. (2004) imply that the
482: radio conversion factor is constant out to $z \sim 1$, ie. that the
483: radio-FIR correlation is constant out to this redshift.  Reddy \&
484: Steidel (2004) extend this conclusion out to $z \sim 2$ in their
485: extensive study of UV selected galaxies. Most recently, Ibar et
486: al. (2008) conclude that the radio conversion factor remains constant
487: out to $z \sim 3$, using a 24$\mu$m source sample with redshifts from
488: the SXDF. However, given the need for individual source detections in
489: the radio, their $z \sim 3$ radio sources have star formation rates
490: about two orders of magnitude larger than the stacking results
491: presented herein, and hence a direct comparison is problematic.
492: 
493: One physical reason why we might expect the radio conversion factor to
494: diverge at the highest redshifts is increased relativistic electron
495: cooling due to inverse Compton scattering off the cosmic microwave
496: background (CMB). The ratio of relativistic electron energy losses due
497: to synchrotron radiation, to energy losses due to inverse Compton
498: radiation, equals the ratio of the energy density in the magnetic
499: field to that in the photon field.  The energy density in the CMB
500: increases as: $U_{CMB} = 4.2\times 10^{-13} (1+z)^4$ ergs cm$^{-3}$.
501: Figure 2 shows a comparison of $U_{CMB}$ with the typical energy
502: densities in the magnetic fields in different regions in galaxies. We
503: show the range of fields considered typical for spiral arms ($\sim$
504: few $\mu$G), and for starburst galaxy nuclei (of order 100$\mu$G; Beck
505: et al. 1994; see review by Beck 2005).  The important point is that IC
506: losses off the CMB will dominate synchrotron losses in a typical ISM
507: at $z \ge 0.5$, and dominate in starburst nuclei at $z \ge 4$.
508: 
509: We note that inverse Compton losses will not affect the thermal
510: electrons responsible for Free-Free emission from star forming
511: galaxies. Such Free-Free emission may dominate the total radio
512: emission from galaxies at rest frequencies between roughly 40 GHz and
513: 100 GHz (Condon 1992), and hence become an important factor in radio
514: continuum studies of very high redshift galaxies.
515: 
516: A depressed radio luminosity for a given star formation rate could
517: also arise if the $z \sim 3$ LBGs are systematically smaller
518: galaxies. The standard model that produces the radio-FIR correlation
519: (Condon 1992) requires a cosmic ray processing box size $\ge 1$
520: kpc. It has been observed that dwarf galaxies at low redshift depart
521: from the radio-FIR correlation by about a factor of 2, in the sense of
522: being radio under-luminous.  The hypothesis is that the cosmic rays
523: diffuse out of the galaxy on timescales shorter than required to
524: maintain the standard radio-FIR correlation (Yun et
525: al. 2001). Giavalisco (2002) describes high $z$ LBGs as having typical
526: half-light radii of 4 to 7 kpc, significantly larger than typical
527: dwarf galaxies. However, he points out that: "..frequently the
528: galaxies have disturbed or fragmented morphologies, with one bright
529: core, or multiple knots embedded in diffuse nebulosity, reminiscent of
530: merger events."
531: 
532: In this paper, we report the first robust statistical (median)
533: detection of sub-$\mu$Jy radio emission from LBG galaxies at $z \sim
534: 3$. This detection was made possible by the very wide area, and depth,
535: of the Cosmos field. While our physical interpretation of the result
536: remains inconclusive, there are a number of future studies we are
537: pursuing to address the interesting questions raised concerning the UV
538: attenuation factor for LBGs, and the the radio luminosity to star
539: formation rate conversion factor, at $z \ge 3$. The most important
540: study involves obtaining optical spectra of a large sample of LBGs
541: from the COSMOS sample. Spectra will elucidate the nature of the
542: sources, and allow for a study of the dust correction factor as a
543: function of galaxy type (AGN, starburst, elliptical...). Also, the
544: selection criteria for this LBG sample were very conservative (Section
545: 2).  We will further refine (and increase) our LBG samples using the
546: Cosmos multiband photometry, and perform statistical analyses of the
547: radio properties as a function of eg. stellar mass, or $A_V$. We are
548: also exploring the IR properties of these samples with Spitzer. Such
549: IR studies have particular relevance in the light of the results of
550: Ivison et al. (2007), who find that for the seven IR luminous $z \sim
551: 3$ LBGs in their AEGIS sample ($S_{24\mu m} > 60\mu$Jy), the median
552: 1.4 GHz flux density is 44$\mu$Jy, comparable to that expected for
553: sub-mm galaxies. Lastly, unambiguous identification of radio AGN using
554: eg. optical spectra, will provide input into faint radio source
555: population models that are being generated in the context of planning
556: for future, large area radio telescopes (Wilman et al. 2008). The
557: results presented herein indicate that future deep (sub-$\mu$Jy), wide
558: field surveys with the Expanded Very Large Array, will detect
559: routinely the radio emission from individual LBGs out to $z \sim 3$.
560:  
561: \acknowledgments
562:  
563: The HST COSMOS Treasury program was supported through NASA grant
564: HST-GO-09822.  We gratefully acknowledge the contributions of the
565: entire COSMOS collaboration consisting of more than 70 scientists.
566: More information on the COSMOS survey is available \\ at {\bf
567: \url{http://www.astro.caltech.edu/cosmos}}.  CC thanks the
568: Max-Planck-Gesellschaft and the Humboldt-Stiftung for support through
569: the Max-Planck-Forschungspreis. We thank the referee for helpful 
570: comments. 
571:  
572: % {\it Facilities:} \facility{HST (ACS)}, \facility{HST (NICMOS)},
573: %\facility{HST (WFPC2) \facility{VLA}}.
574:   
575: \begin{thebibliography}{}
576: 
577: \bibitem[]{}Appleton, P.N., Fadda, D. T.; Marleau, F. et al. 
578: 2004, ApJS, 154, 147
579: 
580: \bibitem[]{}Adelberger and Steidel 2000, ApJ, 544, 218
581: 
582: \bibitem[]{}Bell, E. 2003,  ApJ, 586, 794
583: 
584: \bibitem[]{}Beck, R. 2005, in 'Magnetic fields in the Universe: From
585: Laboratory and Stars to Primordial Structures,' AIP Conference
586: Proceedings,  784,  343
587: 
588: \bibitem[]{}Beck, R. Carilli, C.L., Holdaway, M.A., Klein, U. 1994, 
589: 292, 409
590: 
591: \bibitem[]{}Bouwens,R.J.,  Illingworth, G.D., Franx, M.,
592: Ford, H. 2007, ApJ, 670, 928
593: 
594: \bibitem[]{}Capak, P. et al. 2008, ApJ letters, 681, L53
595: 
596: \bibitem[]{}Capak, P. et al. 2007, ApJS, 172, 99
597: 
598: \bibitem[]{}Capak, P. et al. 2004, AJ, 127, 180
599: 
600: \bibitem[]{}Carilli, C.L. et al. 2007, ApJS, 172, 518
601: 
602: \bibitem[]{}Carilli, C.L., Yun, M.S. 2000, 530, 618
603: 
604: \bibitem[]{}Calzetti, D. 1997, AJ, 113, 162
605: 
606: \bibitem[]{}Condon, J. 1992, ARAA, 30, 575
607: 
608: \bibitem[]{}Daddi, E., Dickinson, M., Morrison, G. et al.
609: 2007, ApJ, 670, 156
610: 
611: \bibitem[]{}Gebhardt, K. et al. 2000, ApJ, 539, L13
612: 
613: \bibitem[]{}Giavalisco, M. 2002, ARAA, 40, 579
614: 
615: \bibitem[]{}Ibar, E., Cirasoulo, M., Ivison, R., Best, P., Smail, I.
616: et al. 2008, MNRAS, 386, 953
617: 
618: \bibitem[]{}Ivezic, Z., Menou, K., Strauss, M., Knapp, J., et
619: al. 2002, AJ, 124, 2634
620: 
621: \bibitem[]{}Ivison, R.J. Chapman, S.C., Faber, S.M., Smail, I., Biggs,
622: A. et al.  2007, ApJ, 660, L77
623: 
624: \bibitem[]{}Jiang, L., Fan, X., Ivezic, Z. et al. 2007, ApJ, 656, 680
625: 
626: \bibitem[]{}Kennicutt, R. 1998, ARAA, 36, 189
627: 
628: \bibitem[]{}Meurer, G., Calzetti, D., Heckman, T. 1999, ApJ, 512, 64
629: 
630: \bibitem[]{}Mobasher, B. \& Mazzei, P. 2000, A\& A, 363, 517
631: 
632: \bibitem[]{}Nandra, K. et al. 2002, ApJ, 576, 625
633: 
634: \bibitem[]{}Petric, A., Carilli, C.L., Bertoldi, F., Beelen, A., Cox,
635: P., Omont, A.  2006, AJ, 132, 1307
636: 
637: \bibitem[]{}Reddy, N. \& Steidel, C. 2004, ApJ, 603, L13
638: 
639: \bibitem[]{}Robin, A.C., Rich, R.M., Aussel, H. et al.
640: 2007, ApJS, 172, 545
641: 
642: \bibitem[]{}Schinnerer, E. et al. 2007, ApJS, 172, 46S
643: 
644: \bibitem[]{}Scoville, N. et al. 2007, ApJS, 172, 38S
645: 
646: \bibitem[]{}Shapley, A., Steidel, C., Pettini, M., 
647: Adelberger, K. 2003, ApJ, 588, 65
648: 
649: \bibitem[]{}Steidel, C., Adelberger, K., Shapley, A., Pettini, M.,
650: Dickinson, M., Giavalisco, M. 2000, ApJ, 532, 170
651: 
652: \bibitem[]{}Steidel, C.C., Adelberger, C., Giavalisco, M., Dickinson,
653: M., Pettini, 1999, ApJ, 519, 1
654: 
655: \bibitem[]{}Wilkins, S., Trentham, N., Hopkins, A. 2008, MNRAS, 385, 687
656: 
657: \bibitem[]{} White, Richard L., Helfand, David J., Becker, Robert H., 
658: Glikman, Eilat; de Vries, Wim 2007, ApJ, 654, 99
659: 
660: \bibitem[]{}Wilman, R.J, Miller, L., Jarvis, M., Mauch, T., Levrier, F.
661: et al. 2008, MNRAS, 388, 1335
662: 
663: \bibitem[]{}Yun, M.S., Reddy, N., Condon, J. 2001, ApJ, 554, 803
664: 
665: \bibitem[]{}Yun, M.S., Carilli, C.L. 2002, ApJ, 568, 88
666: 
667: \end{thebibliography}
668: 
669: 
670: \clearpage
671: \newpage
672: 
673: \begin{table}\label{LBGtable}
674: \begin{center}
675: \caption{COSMOS drop-out selection}
676: \begin{tabular}[ht]{|c|c|c|c|}
677: \tableline
678: U-band drop-outs & ($\mbox{u'} - B) \geq 0.7  (B - \mbox{i'})
679: + 0.7$ & ($\mbox{u'} - B) \geq 1.4$ &  ($B - \mbox{i'}) \leq 2.5$ \\
680: B-band drop-outs & ($B-V) \geq 0.7~(V - \mbox{i'}) + 0.7$ &  ($B-V)
681: \geq 1.0$ & $(V - \mbox{i'}) \leq 2.5$ \\
682: V-band drop-outs & $(V-\mbox{r'}) \geq 0.3~(\mbox{r'}-\mbox{i'}) +1.0$ &  ($V-\mbox{r'})\geq 1.4$ & $(\mbox{r'}-\mbox{i'}) \leq 3.0 $\\
683: \tableline
684: \end{tabular}
685: \end{center}
686: \end{table}
687: 
688: 
689: \begin{table}\label{LBGtable}
690: \begin{center}
691: \caption{Statistics of Lyman Break Sample and Results of the $4\sigma$ Counterpart Search}
692: \begin{tabular}[ht]{|c|c|c|c|c|c|}
693: \tableline
694: Band & Redshift & \# of LBGs &  \# of $4\sigma$ Sources & 
695: Corrected \# of Sources & Median flux density \\ 
696: ~ & ~ & ~ & ~ & ~ & $\mu$Jy \\
697: \tableline
698: U & 2.5--3.5 & 6457 &  43 (0.67\%)  & 24 (0.38\%) & $0.90\pm 0.21$ \\
699: B & 3.5--4.5 & 1447 &  8 (0.55\%)  & 5 (0.34\%) & $0.83\pm 0.42$ \\
700: V & $>$ 4.5 & 614 &    5 (0.81\%) & 2 (0.33\%) & $1.72\pm 0.68$ \\
701: \tableline
702: \end{tabular}
703: \end{center}
704: \end{table}
705: 
706: 
707: \begin{figure}
708: \psfig{file=f1a.ps,width=3in}
709: \vskip -3.74in
710: \hspace*{3in}
711: \psfig{file=f1b.ps,width=3in}
712: \psfig{file=f1c.ps,width=3in}
713: \caption{Images at 1.4 GHz of the median stacking results 
714: for the different drop-out samples. In each case, the contour
715: levels are -3, -2, -1, 1, 1, 2, 3, 4 $\times$ the rms value in the image, and
716: the cross indicates the source stacking position. The synthesized  
717: beam FWHM is shown in the lower left corner. 
718: Top left: U-band drop-out result, with rms = 0.21 $\mu$Jy beam$^{-1}$.
719: Top Right: The B-band drop-outs result, with rms = 0.42 $\mu$Jy beam$^{-1}$.
720: Bottom: The V-band drop-outs result, with rms = 0.68 $\mu$Jy beam$^{-1}$.
721: }
722: \end{figure}
723: 
724: 
725: \begin{figure}
726: \psfig{file=f2.ps,width=5in}
727: \caption{The energy density in the CMB as a function of
728: redshift, plus shaded regions showing typical ranges for magnetic
729: fields in spiral arms, and starburst galaxy nuclei.}
730: \end{figure}
731: 
732:  
733: \end{document}
734:  
735: 
736: \begin{table}\label{V_4_sigma}
737: \begin{center}
738: \caption{$4\sigma$ Counterparts to V-band drop-outs}
739: \begin{tabular}[ht]{|c|c|c|c|}
740: \tableline
741: COSMOS ID & RA & Dec & Flux Density ($\mu{}$Jy) \\ 
742: \tableline
743:  18641  &  10 00  5.71  &  02 37 58.32  &   4.8457E-05 \\ 
744:   9167  &  09 59 15.92  &  01 52 28.33  &   4.1299E-05 \\ 
745:  17871  &  10 01 27.08  &  02 08 55.80  &   5.4538E-05 \\ 
746:   1205  &  10 02 24.76  &  01 36 59.27  &   1.6492E-04 \\ 
747:   5364  &  10 00 54.52  &  02 34 35.07  &   4.4867E-05 \\ 
748: \tableline
749: \end{tabular}
750: \end{center}
751: \end{table}
752: 
753: 
754: 
755: \begin{table}\label{B_4_sigma}
756: \begin{center}
757: \caption{$4\sigma$ Counterparts to B-band drop-outs}
758: \begin{tabular}[ht]{|c|c|c|c|}
759: \tableline
760: COSMOS ID & RA & Dec & Flux Density ($\mu{}$Jy) \\ 
761: \tableline
762:   3352  &   10 01  9.48 &    01 46  1.47  &   6.8627E-05 \\
763:   4765  &   10 01 29.87 &    02 15 15.56  &   3.8704E-05 \\
764:   7942  &   10 00 19.21 &    02 13 38.91  &   8.7268E-05 \\
765:   8752  &   10 02 36.00 &    02 32 46.16  &   6.1271E-05 \\
766:  10248  &   10 00 57.49 &    02 42 20.59  &   1.3570E-04 \\
767:   3351  &   10 01  9.45 &    01 46  3.16  &   5.9979E-05 \\
768:   6546  &   10 01 32.07 &    02 04 29.06  &   3.8391E-04 \\
769:   5156  &   10 02 56.36 &    02 24  5.92  &   2.7007E-04 \\
770: \tableline
771: \end{tabular}
772: \end{center}
773: \end{table}
774: 
775: 
776: \begin{deluxetable}{|c|c|c|c|}
777: \tablehead{
778: \colhead{COSMOS ID} & \colhead{RA} & \colhead{Dec} & \colhead{Flux Density ($\mu{}$Jy)}} 
779: \tablecaption{$4\sigma$ Counterparts to U-band drop-outs \label{U_4_sigma}}
780: \tablewidth{0pt}
781: \startdata
782: \tableline
783:   2367 &   10 01 19.14  &   02 26 17.50 &    3.9663E-05 \\
784:    788 &   09 58 58.11  &   02 37  7.13 &    5.1523E-05 \\
785:   9401 &   10 01  1.32  &   02 21 59.39 &    2.2910E-04 \\
786:  13697 &   10 01 46.23  &   02 29  6.41 &    1.0377E-04 \\
787:   5309 &   10 02 43.71  &   01 53 53.35 &    7.8327E-05 \\
788:   1415 &   10 00 46.62  &   02 46 44.07 &    8.2603E-04 \\
789:  13186 &   10 00 37.77  &   01 49 34.51 &    5.0641E-05 \\
790:   6955 &   10 00 11.04  &   01 53 14.38 &    9.3061E-05 \\
791:   1288 &   09 59  0.91  &   01 56 45.26 &    4.2922E-05 \\
792:   8217 &   09 59 30.34  &   02 41 59.85 &    5.3797E-05 \\
793:  15072 &   10 00 36.53  &   02 18 28.22 &    4.8128E-05 \\
794:   7672 &   10 00 54.99  &   02 42 53.61 &    5.7537E-05 \\
795:  14091 &   10 01 11.57  &   02 28 41.09 &    3.6775E-05 \\
796:  15303 &   10 00 54.71  &   01 38 12.74 &    6.2560E-05 \\
797:   8185 &   10 01  9.25  &   01 42 41.12 &    8.5213E-05 \\
798:  12159 &   09 59 29.59  &   01 49 26.43 &    4.0179E-05 \\
799:    481 &   10 02 28.77  &   02 17 21.77 &    2.9689E-03 \\
800:   7009 &   09 59  1.38  &   02 52  4.03 &    9.8580E-05 \\
801:   2839 &   10 01  9.48  &   01 46  1.47 &    6.8627E-05 \\
802:   2408 &   10 00 53.77  &   02 16 14.80 &    1.9877E-04 \\
803:  12795 &   10 02  3.36  &   02 48 17.95 &    7.2447E-05 \\
804:   1016 &   09 59 34.08  &   02 17  6.47 &    1.4397E-04 \\
805:   1894 &   10 01 57.02  &   02 16 12.22 &    1.1569E-04 \\
806:  13001 &   09 58  3.24  &   01 58 56.20 &    8.8272E-05 \\
807:   5560 &   10 01 32.07  &   02 04 29.06 &    3.8391E-04 \\
808:  10033 &   10 00  9.56  &   02 01 42.26 &    3.6324E-05 \\
809:    339 &   09 58 25.21  &   02 27 28.08 &    7.2221E-05 \\
810:   4257 &   09 58 25.18  &   02 14 56.19 &    1.0408E-04 \\
811:  10643 &   09 57 48.53  &   01 40  5.12 &    1.0038E-04 \\
812:  2838  &   10 01  9.45  &   01 46  3.15 &    5.9979E-05 \\
813:  13718 &   10 00 42.39  &   02 09 39.73 &    1.7848E-04 \\
814:  16563 &   10 00 52.24  &   01 58 10.80 &    4.7588E-05 \\
815:  14571 &   09 58 45.20  &   02 08 23.68 &    4.2184E-05 \\
816:    861 &   09 58 58.06  &   02 37  7.97 &    5.1523E-05 \\
817:   8606 &   09 59 58.16  &   02 42  4.83 &    6.1345E-05 \\
818:   8539 &   09 59 48.35  &   02 02 38.64 &    3.3381E-05 \\
819:  13545 &   09 58 30.19  &   01 58 52.64 &    9.0971E-05 \\
820:  12412 &   09 59 50.92  &   02 48 36.76 &    7.3366E-05 \\
821:   6655 &   10 00 19.22  &   02 13 38.89 &    8.7268E-05 \\
822:  12942 &   10 02 24.32  &   01 38 32.52 &    6.6815E-05 \\
823:  13569 &   09 58 34.43  &   01 37 55.39 &    6.7613E-05 \\
824:   7558 &   10 02 28.44  &   02 32 32.82 &    2.1540E-04 \\
825:  15498 &   09 59 39.13  &   01 47 35.13 &    4.0474E-05 \\
826: \tableline
827: \enddata
828: \end{deluxetable}
829: 
830: 
831: 
832: 
833: \begin{figure}[ht]
834: \includegraphics[width=6in]{SFRplot.ps}
835: \caption{Radio Flux Densities for Different SFRs and Redshifts}
836: \label{SFRplot}
837: \end{figure}
838: 
839: 
840: 
841: \begin{figure}[ht]
842: \includegraphics[width=6in]{J1000.eps}
843: \caption{An r,z,3.6 color image of J100054.52+023435.1 with the radio contours (10,20,30 $\mu$Jy).  r (cyan) is mostly Lyalpha, z (green to orange) is UV emission and 3.6um (red) is rest frame optical $+$ Halpha.  Blue objects are in the foreground.}
844: \label{J1000}
845: \end{figure}
846: