1: %\documentclass[longnamesfirst,apjl]{emulateapj}

2: \documentclass[12pt,preprint]{aastex}

3: %\usepackage{apjfonts,natbib,epsfig}

4: \usepackage{natbib,epsfig}

5: \bibliographystyle{apj3}

6: \tighten

7: 

8: 

9: 

10: \newcommand{\Mpch}{$h\, {\rm Mpc}^{-1}\, $}

11: \newcommand{\mpc}{{\rm\,Mpc}}

12: \newcommand{\impc}{{\rm\,Mpc}^{-1}}

13: 

14: \newcommand{\hmpc}{h^{-1}{\rm\,Mpc}}

15: \newcommand{\ihmpc}{h{\rm\,Mpc}^{-1}}

16: \newcommand{\hmpcC}{h^{-3}{\rm\,Mpc^3}}

17: \newcommand{\ihmpcC}{h^3 {\rm\,Mpc}^{-3}}

18: \newcommand{\kmsmpc}{{\rm\ km\ s^{-1}\ Mpc^{-1}}}

19: \newcommand{\eV}{{\rm\ eV}}

20: \newcommand{\om}{\Omega_m}

21: \newcommand{\omhh}{\om h^2}

22: \newcommand{\ob}{\Omega_B}

23: \newcommand{\obhh}{\ob h^2}

24: \newcommand{\ok}{\Omega_K}

25: \newcommand{\on}{\Omega_\nu}

26: \newcommand{\da}{d_A}

27: \newcommand{\ns}{n_S}

28: \newcommand{\nt}{n_T}

29: 

30: \newcommand{\ts}{T/S}

31: \newcommand{\As}{A^2_S}

32: \newcommand{\tot}{{\rm t}}

33: \newcommand{\hal}{{\rm h}}

34: \newcommand{\Poi}{{\rm P}}

35: \newcommand{\lin}{{\rm lin}}

36: \newcommand{\nlin}{{\rm nl}}

37: \newcommand{\shell}{{\rm s}}

38: \newcommand{\dirac}{{\rm D}}

39: \newcommand{\halo}{{\rm h}}

40: 

41: \newcommand{\bp}{{\cal C}}

42: \newcommand{\bfw}{{\mathbf{w}}}

43: \newcommand{\bfx}{{\mathbf{x}}}

44: \newcommand{\bfl}{{\mathbf{l}}}

45: \newcommand{\bfr}{{\mathbf{r}}}

46: \newcommand{\bfk}{{\mathbf{k}}}

47: \newcommand{\bfP}{{\mathbf{P}}}

48: \newcommand{\bfPnorm}{{\mathbf{P}_{\rm norm}}}

49: \newcommand{\bfF}{{\mathbf{F}}}

50: \newcommand{\bfp}{{\mathbf{p}}}

51: \newcommand{\veck}{{\bf k}}

52: \newcommand{\vecl}{{\bf l}}

53: \newcommand{\vecr}{{\bf r}}

54: \newcommand{\vecx}{{\bf x}}

55: \newcommand{\thetavect}{{\vec{\theta}}}

56: \newcommand{\Pvect}{{\vec{P}}}

57: \newcommand{\Pbias}{P_{\rm bias}}

58: \newcommand{\ra}{r_a}

59: \newcommand{\what}{\hat{C}}

60: \newcommand{\rms}{{\it rms}}

61: 

62: \newcommand{\h}{{\rm nl}}

63: \newcommand{\vel}{{\rm vel}}

64: \newcommand{\Cov}{{\rm Cov}}

65: \newcommand{\Corr}{{\rm Corr}}

66: \newcommand{\kmax}{k_{\rm max}}

67: \newcommand{\kfid}{k_{\rm fid}}

68: \newcommand{\kpivot}{k_{\rm pivot}}

69: \newcommand{\knorm}{k_{\rm norm}}

70: \newcommand{\Pinit}{P_{\Phi}}

71: \newcommand{\Phat}{\hat{\bf P}}

72: \newcommand{\fsky}{f_{\rm sky}}

73: \newcommand{\sn}{{\rm SN}}

74: \newcommand{\dgr}{D_{\rm gr}}

75: \newcommand{\tableskip}{\\[-6pt]}

76: \newlength{\tskip}\setlength{\tskip}{5pt}

77: \newlength{\colwidth}\setlength{\colwidth}{3.5in}

78: \newcommand{\colskip}{@{\hspace{0.3in}}}

79: \newcommand{\etal}{et al.}

80: 

81: \newcommand{\beq}{\begin{equation}}

82: \newcommand{\eeq}{\end{equation}}

83: \newcommand{\beqa}{\begin{eqnarray}}

84: \newcommand{\eeqa}{\end{eqnarray}}

85: \newcommand{\beam}{B_\ell}      % The beam function

86: \newcommand{\nbar}{\bar{n}}     % The selection function

87: \newcommand{\Dp}{\Delta p}

88: \newcommand{\fpt}{\tilde{f}'}

89: \def\simlt{\lesssim}

90: \def\simgt{\gtrsim}

91:   \newcommand{\wj}{\left(

92:                            \begin{array}{ccc}

93:                            l_1  &  l_2  & l_3 \\

94:                              0  &  0    &  0

95:                            \end{array}

96:                            \right)}

97: \newcommand{\wjm}{\left(

98:                            \begin{array}{ccc}

99:                            l_1  &  l_2  & l_3 \\

100:                             m_1  &  m_2   &  m_3

101:                            \end{array}

102:                            \right)}

103: \newcommand{\bi}{B_{l_1 l_2 l_3}}

104: \newcommand{\bip}{B_{l_1' l_2' l_3'}}

105: \newcommand{\deld}{\delta^{\rm D}}

106: \newcommand{\bn}{\hat{\bf n}}

107: \newcommand{\bm}{\hat{\bf m}}

108: \newcommand{\bl}{\hat{\bf l}}

109: \newcommand{\bk}{\hat{\bf k}}

110: \newcommand{\rad}{r}    % comoving radial distance

111: \newcommand{\tableskabsip}{\tablevspace{3pt}}

112: \newcommand{\dop}{{\rm dop}}

113: \newcommand{\sky}{{\rm sky}}

114: \newcommand{\isw}{{\rm ISW}}

115: \newcommand{\sw}{{\rm SW}}

116: \newcommand{\ov}{{\rm OV}}

117: \newcommand{\se}{{\rm S}}

118: \newcommand{\ri}{{\rm ri}}

119: \newcommand{\len}{{\rm len}}

120: \newcommand{\Ylm}[1]{Y_{l_#1}^{m_#1}}

121: \newcommand{\Ylmn}{Y_{l}^{m}}

122: \newcommand{\alm}[1]{a_{l_#1 m_#1}}

123: \newcommand{\almn}{a_{l m}}

124: \newcommand{\Dk}{\frac{d^3{\bf k}}{\left( 2\pi \right) ^3}}

125: 

126: 

127: %\newcommand{\etal}{{\it et al. }}

128: \newcommand{\siml}{\lesssim}

129: \newcommand{\simg}{\gtrsim}

130: \newcommand{\lsim}{\lesssim}

131: \newcommand{\gsim}{\gtrsim}

132: \newcommand{\psim}{\mbox{\raisebox{-1.0ex}{$~\stackrel{\textstyle  

133: \propto}

134: {\textstyle \sim}~$ }}}

135: \newcommand{\vect}[1]{\mbox{\boldmath${#1}$}}

136: \newcommand{\lmk}{\left(}

137: \newcommand{\rmk}{\right)}

138: \newcommand{\lnk}{\left\{ }

139: \newcommand{\nn}{\nonumber}

140: \newcommand{\rnk}{\right\} }

141: \newcommand{\lkk}{\left[}

142: \newcommand{\rkk}{\right]}

143: \newcommand{\lla}{\left\langle}

144: \newcommand{\p}{\partial}

145: \newcommand{\rra}{\right\rangle}

146: \newcommand{\so}{M_\odot}

147: \newcommand{\mch}{{\cal M}}

148: \newcommand{\vex}{{\vect x}}

149: \newcommand{\ver}{{\vect r}}

150: \newcommand{\mb}{M_\bullet}

151: 

152: \slugcomment{{\it Astrophysical Journal} in press}

153: 

154: 

155: 

156: \begin{document}

157: %\title{Spitzer Stacking Evidence for a Population of sub-L$_{*}$ Galaxies at Redshifts 2 to 3 and Unresolved IR Fluctuations}

158: \title{Contribution to Unresolved Infrared Fluctuations from Dwarf Galaxies at Redshifts of $2-3$}

159: \author{Ranga-Ram Chary\altaffilmark{1},

160: Asantha Cooray\altaffilmark{2}, 

161: Ian Sullivan\altaffilmark{1}}

162: \altaffiltext{1}{Division of Physics, Mathematics, and Astronomy,  

163: California Institute of Technology, Pasadena, CA 91125. Email: rchary@caltech.edu}

164: \altaffiltext{2}{Center for Cosmology, Department of Physics and  

165: Astronomy, University of California, Irvine, CA 92697. E-mail:  

166: acooray@uci.edu}

167: \begin{abstract}

168: In order to understand the origin of clustered anisotropies detected in {\it  

169: Spitzer} images between 3.6 and 8\,$\mu$m, we stack the {\it Spitzer} IRAC/Great Observatories

170: Origins Deep Survey (GOODS) images at pixel locations corresponding to

171: faint, $z_{AB}\sim27$ mag, optical sources with no obvious IR counterparts. We obtain a

172: strong detection of the sources

173: with a stacked median flux at 3.6$\mu$m of 130$\pm$5 nJy above the background. 

174: The wealth of multi-wavelength imaging data in GOODS enables a similar stacking analysis to be undertaken

175: at various wavelengths between the ultraviolet and near-infrared bands. We obtain strong stacked detections of these

176: optically faint sources over the entire wavelength range 

177: which places constraints on the average properties of these sources.  We find that

178: the flux spectrum of the median, stacked source

179: is consistent with a $L\lesssim0.03~L_{*,{\rm UV}}$ galaxy with a 90\% confidence interval for the redshift of $1.9-2.7$.

180: These sources produce a 3.6 $\mu$m absolute background intensity between 0.1 and 0.35 nW m$^{-2}$ sr$^{-1}$

181: and the clustered IR light could account for $\sim30-50$\% of fluctuation power in the IR background

182: at 4 arcminute angular scales. Although the exact redshift distribution of these sources is unknown, these galaxies

183: appear to contain $5-20$\% of the co-moving stellar mass density at $z\sim2.5$. 

184: \end{abstract}

185: 

186: \keywords{large scale structure of universe --- diffuse radiation ---  

187: infrared: galaxies}

188: 

189: \section{Introduction}

190: The intensity of the cosmic near-infrared background (IRB) is a  

191: measure of the total light emitted by stars and galaxies in the universe which is not  

192: thermally reprocessed by dust. 

193: The absolute background has been estimated with the

194: Diffuse Infrared Background Experiment (DIRBE; Hauser  

195: \& Dwek 2001) and the Infra-Red Telescope in Space (IRTS; Matsumoto et al. 2005)

196: which results in values that are about a factor of two higher than the

197: intensity obtained by integrating the light from individually detected galaxies.

198: This is most likely attributable to the large uncertainties associated with the removal of

199: foreground zodiacal light from the DIRBE observations (Dwek et al. 2005). 

200: This hypothesis has been strengthened by the TeV spectrum of gamma-ray blazars

201: which indicate that the total IRB intensity is smaller than the DIRBE/IRTS estimates

202: (Aharonian et al. 2005) and that $\sim$ 90\% of the IRB light probably arises in

203: known galaxy populations.

204: Yet, attempts have been made to explain

205: the difference between the measured and resolved IRB intensity 

206: with sources at the epoch of reionization (e.g., Santos, Bromm \& Kamionkowski 2002; Salvaterra \& Ferrara  

207: 2003; Cooray \& Yoshida 2004; Fernandez \& Komatsu 2006). If first-light galaxies are to explain   the missing

208: intensity completely, then an extreme scenario is needed with the conversion of at least 5\% of  

209: all baryons to stars (Madau \& Silk 2004). 

210: 

211: Instead of the absolute intensity, recent works 

212: have concentrated on spatial   fluctuations of the IRB 

213: (Cooray et al. 2004; Kashlinsky et al. 2004).

214: While an interpretation of unresolved fluctuations is subject to

215: extremely uncertain astrophysical modeling of underlying faint populations,

216: fluctuations in deep {\it Spitzer}  images have  been fully attributed to first-light galaxies containing 

217: Pop~III stars during reionization (Kashlinsky et al. 2005). An alternative study

218: shows that a reasonable fraction ($>50$\%) of unresolved fluctuations is arising from

219: faint, unresolved sources at lower redshifts (Cooray et al. 2007). Thus, while 

220: two independent studies find a  similar clustering amplitude for

221: IRB fluctuations (Kashlinsky et al. 2004; Cooray et al. 2007), they  

222: differ in the interpretation related to the contribution from Pop~III stars due to differences in the background 

223: light ascribed to faint foreground galaxies unresolved by {\it Spitzer}.

224: 

225: The suggestion for a low redshift origin in Cooray et al. (2007)  comes from predictions using a halo model (Cooray \& Sheth 2002) 

226: for the 3.6 $\mu$m population matched to 

227: their luminosity functions  (Babbedge et al. 2006) and clustering power spectra (Sullivan et  al. 2007).

228: In Cooray et al. (2007) this low-redshift population was identified

229: as the faint optical galaxies that are resolved in the {\it Hubble} Advanced Camera for Surveys (ACS; Giavalisco et al.  

230: 2004) images of the Great Observatories Origins Deep Survey (GOODS; Dickinson et al. 2003) fields, 

231: but not detected in  deep {\it Spitzer} Infrared Array Camera (IRAC)   images  of the same

232: fields. With models, it was estimated  that this faint galaxy population  has a 3.6 $\mu$m  intensity between  

233: 0.1 and 0.8 nW m$^{-2}$ sr$^{-1}$. The analysis in Cooray et al. (2007) did not exclude a high redshift

234: contribution to the IRB, with an  upper  limit on the 3.6 $\mu$m absolute background intensity from

235:   $z>8$ sources of  0.6   nW m$^{-2}$ sr$^{-1}$. 

236: In comparison, if all unresolved IR fluctuations are generated by faint $z>6.5$ sources which are below the detection

237: threshold of {\it Spitzer}, then the 

238: IRB intensity for such sources is $\gtrsim$1 nW m$^{-2}$ sr$^{-1}$ at 3.6 $\mu$m (Kashlinsky et al. 2007c).

239: 

240: In this {\it Letter}, we further study the physical origin of fluctuations  

241: in the IR background. Since faint ACS optical sources are undetected

242: in deep {\it Spitzer} IRAC images, we stack their pixel locations in IRAC to establish

243: the average IR flux. We also utilize the wealth of multiwavelength imaging 

244: data in the GOODS fields to obtain a stacked broadband

245: spectral energy distribution of these faint optical galaxies between UV and near-IR bands.

246: Leaving the redshift as a free parameter, we fit the average optical to IR flux spectrum 

247: with galaxy population synthesis models and find that the stacked flux spectrum is best fit

248: by a $\sim0.03$~L$_{*, {\rm UV}}$ galaxy at $1.9<z<2.7$,

249: similar to a scaled-down population of Lyman-break galaxies.

250: We also measure the expected IRB fluctuation clustering spectrum  

251: from these sources. 

252: 

253: We  

254: summarize the stacking analysis in the next section

255: and discuss results on the average optical to IR flux spectrum

256: in \S~3. We discuss clustering of these faint optical  

257: sources in \S~4. In \S~5 we discuss the implications of our measurements.

258: The magnitudes quoted throughout this paper are

259: AB magnitudes.

260: 

261: 

262: \begin{figure}[!t]

263: \centerline{\psfig{file=Fig1.ps,width=3.3in,angle=0}    }

264: \caption{

265: The average-stacked 3.6 $\mu$m image from GOODS-N with

266: pixel locations of all 6160 ACS sources unmatched and unmasked in IRAC.

267: The median stacked flux is  $(130 \pm 5)$ nJy (or $=26.1 \pm 0.1$ magnitude) 

268: and is detected with a  

269: statistical signal-to-noise ratio higher than 25. 

270: }

271: \label{fig:stack}

272: \end{figure}

273: 

274: \begin{figure}[!t]

275: \centerline{\psfig{file=Fig2.ps,width=3.9in,angle=0}    }

276: \caption{

277: The stacked ultraviolet to IR flux spectral energy distribution

278: of faint galaxies detected by ACS but not detected individually by 

279: {\it Spitzer}/IRAC in GOODS-S (black line and data points). The SED is consistent 

280: with the sub-L$_{*,{\rm UV}}$ 

281: galaxy population at $z\sim2.5$. The parameters of the stacked galaxy

282: for GOODS-S are as shown in the legend. 

283: The convolution of the fit SED with the different passbands is shown with empty circles.

284: The stacking in GOODS-N results in a best

285: fit redshift of $z\sim1.9$ and the SED shown as the dotted line (See text for details). 

286: }

287: \label{fig:spec}

288: \end{figure}

289: 

290: \begin{deluxetable}{ccc}

291: \tablecaption{Results from Stacking Analysis\label{table:data}}

292: \tablehead{

293: \colhead{} &

294: \colhead{GOODS-N} & 

295: \colhead{GOODS-S} \\

296: }

297: \startdata

298: $\Delta \Omega$/arcmin$^2$ &   169  & 171 \\

299: N$_{\it ACS}$ &  6160 & 5441 \\

300: U & 28.3 $\pm$ 0.1 & 28.4 $\pm$ 0.1 \\

301: B & 28.0 $\pm$ 0.1 & 27.8 $\pm$ 0.1 \\

302: V & 27.5 $\pm$ 0.1 & 27.4 $\pm$ 0.1 \\

303: i & 27.2 $\pm$ 0.1 & 27.1 $\pm$ 0.1 \\

304: z & 27.1 $\pm$ 0.1 & 27.0 $\pm$ 0.1 \\

305: HK$\arcmin$ & 27.3 $\pm$ 0.3 & ... \\

306: J & ... & 26.9 $\pm$ 0.15 \\

307: H & ...	& 26.7 $\pm$ 0.15 \\

308: Ks & ... & 26.6 $\pm$ 0.15 \\

309: 3.6 $\mu$m  & 26.1 $\pm$ 0.1 &  26.0 $\pm$ 0.1 \\

310: 4.5 $\mu$m & 26.2 $\pm$ 0.1 & 26.2 $\pm$ 0.1 \\

311: 5.8 $\mu$m & 26.2 $\pm$ 0.1 & 26.0 $\pm$ 0.1 \\

312: 8.0 $\mu$m & 26.3 $\pm$ 0.15 & 26.0 $\pm$ 0.15 \\

313: %\hline

314: %\end{tabular}\\[12pt]

315: %\begin{minipage}{3.5in}

316: \enddata

317: \tablecomments{N$_{\it ACS}$ is the total number of faint ACS sources over an  

318: effective area

319: $\Delta \Omega$ that were included in the stack. 

320: Uncertainties

321: in the stacked flux (tabulated in AB magnitudes) 

322: are mostly dominated by calibration systematics.}

323: %\end{minipage}

324: %\end{center}

325: \end{deluxetable}

326: 

327: 

328: 

329: \section{Stacking Analysis}

330: 

331: To establish the average IR intensity of faint optical sources that  

332: are unresolved in {\it Spitzer} IRAC images we stack

333: the IRAC images at the spatial coordinates of faint, optical sources which are 

334: detected in the GOODS optical data. The stacking begins by first masking all $>3\sigma$ detected

335: IRAC sources with $m_{\rm AB}$(3.6$\mu$m)$<26.7$ mag in the GOODS/IRAC mosaics\footnote{A description

336: of the GOODS mosaics can be found at http://ssc.spitzer.caltech.edu/legacy/}. This limit

337: is fainter than the 50\% completeness limit of 24.7 mag.

338: The mask has a radius of 13.5$\arcsec$ for a source with 18 mag

339: and scales linearly with magnitude down to 2.4$\arcsec$ for 

340: sources fainter than 22 mag.

341: We then identified sources in the GOODS ACS catalogs (Giavalisco et  

342: al. 2004) which are in unmasked

343: regions of the IRAC mosaics and are therefore IRAC undetected.

344: While there are close to $22,000$ ACS sources which are unmatched  

345: with IRAC sources to within $\sim0.5\arcsec$,

346: only 6,160 (GOODS-N) and 5,441 (GOODS-S) sources are sufficiently

347: separated from brighter IRAC sources so as to be in unmasked regions  

348: of the IRAC image. 

349: The IRAC images for each GOODS field were stacked at the pixel locations corresponding to  

350: these $\sim$6000 ACS sources. The stacking was undertaken by

351: making 12$\arcsec$ image cutouts from the masked IRAC mosaic, centered

352: on each ACS source position. The cutouts were then coadded using

353: a 3$\sigma$ clipped median resulting in a median stacked image

354: of the sources, as shown in Figure 1. 

355: A count of the number of sources which

356: contribute to the stacked flux is also maintained resulting in 

357: an effective exposure map.

358: Pixels which have been masked do not contribute either flux or exposure in the stacking

359: process.

360: Circular aperture photometry

361: was performed on this stacked image and the flux measured in a 3.6$\arcsec$

362: radius beam. The sky background was estimated from the stacked image

363: by measuring the sigma-clipped median of pixels within

364: a surrounding sky annulus spanning $5-6\arcsec$. 

365: The profile of the source had a full-width at half-maximum (FWHM) of 1.6$\arcsec$

366: and was consistent with an unresolved point source at the resolution

367: of {\it Spitzer}.

368: Aperture corrections were estimated by performing identical photometry on a calibration 

369: point source source and correspond to a factor of 1.14.

370: After background subtraction and aperture corrections to the photometry were performed,

371: the resultant stacked 3.6$\mu$m flux density is 130$\pm$1.3 nJy.

372: 

373: In order to assess the reliability of the stacked flux, we stacked

374: the IRAC images at an identical number of positions which are offset

375: from the nominal ACS source positions by random amounts, up to 9$\arcsec$

376: from the original source coordinates. This has the

377: advantage of accounting for any possible contamination of the stacked flux due to

378: the wings of nearby sources which may not be completely masked, while measuring

379: the same local sky background. The random stacks were repeated 100 times and the

380: flux of the stack measured. The standard deviation of these values

381: results in a flux density of 5 nJy which is a factor of 4 larger than the 

382: uncertainty simply due to the background noise term due to source confusion

383: and the low-level flux from the wings of masked sources.

384: Thus, the stacking results in a strong detection of these faint optical galaxies

385: and we adopt a stacked 3.6$\mu$m flux density of 130$\pm$5 nJy.

386: 

387: This procedure was repeated at each IRAC wavelength independently for each

388: GOODS field to assess the reliability of the result. Table~1 summarizes stacking the results for

389: all IRAC wavelengths between 3.6 $\mu$m and 8.0 $\mu$m.

390: Through the stacking analysis, we obtain a strong detection of these faint optical galaxies and

391: measure a flat flux spectrum (F$_{\nu}\propto\nu^{0.08\pm0.3}$). This is consistent with

392: the flat frequency spectrum of IR fluctuations between 3.6 $\mu$m and 8.0 $\mu$m

393: found by Kashlinsky et   al. (2005).

394: 

395: In order to estimate the total contribution to the IRB intensity of these sources,

396: we multiply the median stacked flux by the exposure map.

397: Based on the surface density of about 6000 galaxies in each GOODS field,

398: we establish the absolute 3.6 $\mu$m IR intensity of these

399: faint optical sources  to be 0.12$\pm$0.01 nW m$^{-2}$ sr$^{-1}$. We note that this 

400: estimate is larger than the value obtained by simply multiplying the quoted number of sources

401: ($\sim$6000) with the median flux. This is because the number of sources measurement requires that

402: a source be unmasked out to a radius of 3.6$\arcsec$ while the exposure map takes into

403: account the exact number of sources which contribute to the stacked flux in each pixel.

404: Since some of the source may be partially masked within the 3.6$\arcsec$ radius,

405: the number of sources is smaller than the number of sources as estimated from the exposure map.

406: 

407: Furthermore, this estimate of 0.12 nW m$^{-2}$ sr$^{-1}$ only  

408: accounts for sources that remain outside the IRAC mask and are not confused with bright  

409: IRAC sources. Assuming the average flux from the stack also applies to

410:  sources affected by the mask, we estimate the absolute IRB intensity to be

411: as high as 0.35 nW m$^{-2}$ sr$^{-1}$. This is well within the range of  

412: 0.1 to 0.8 nW m$^{-2}$ sr$^{-1}$ estimated for the contribution of faint sources

413: below the individual point source detection limit in IRAC images

414: through clustering models (Sullivan et al. 2007).

415: 

416: To further study the nature of these sources, beyond {\it Spitzer}/IRAC wavelengths, we adopted

417: a similar procedure and stacked the same population 

418: at other wavelengths at which the GOODS fields have imaging

419: data. Specifically, for GOODS-N, we utilize the KPNO U-band 

420: \citep{Capak04}, {\it Hubble}/ACS $BViz$ (Giavalisco et al. 2004), and

421: near-infrared HK' \citep{Capak04}. For GOODS-S, we use the CTIO U-band

422: and the ESO/ISAAC near-infrared data in the $JHK$ bands in addition to the

423: {\it Hubble} and {\it Spitzer} data. For the {\it Hubble} data, we simply

424: measure the median flux values of the sources since they are individually detected

425: at the ACS wavelengths. For the other wavelengths, we excised subimages for each source

426: and estimated a sigma clipped average of the subimages

427: to obtain the stacked flux, as was done for the IRAC data. Aperture corrections, background estimates

428: and noise calculations were performed as described for the IRAC data.

429: 

430: \begin{figure}[!t]

431: \centerline{\psfig{file=Fig3.ps,width=3.0in,angle=-90}}

432: \caption{

433: The top solid circles show the angular power   spectrum of unresolved

434: IR fluctuations in IRAC channel 1 (3.6 $\mu$m) of GOODS-N field. 

435: The solid squares show the angular power spectrum 

436: that masks the locations of faint ACS

437: sources that are undetected in IRAC.

438: The same power spectrum published in Cooray et al. 2007 is shown with open squares and the difference accounts for a new correction we have introduced here

439: for the window function of the ACS source mask.

440: With ACS sources masked, the fluctuation power spectrum amplitude is reduced 

441: at $\ell \sim 7 \times 10^3$ by a factor of about 2 ($\pm 0.6$). 

442: The circles and dashed lines below show the predicted power spectrum

443: for faint optical sources with their 3.6 $\mu$m number counts 

444: distributed with  two faint-end slopes:

445: the open and filled circles are for clustering in GOODS-N and GOODS-S assuming the steeper slope for counts

446: with $\alpha \sim 0.6$, while the dashed lines bracket the clustering  

447: for a flatter slope with $\alpha \sim 0$ ($dN/dm \propto m^{\alpha}$ when

448: $m > 26$)

449: For reference, the two solid lines are predictions from clustering models 

450: of the faint source population with the

451:  range covering the uncertainty in model parameters (Sullivan et al. 2007).}

452: \label{fig:irb}

453: \end{figure}

454: 

455: \section{Redshift and Stellar Mass Density Estimates}

456: 

457: In order to understand the physical properties of

458: these faint galaxies, we fit the stacked flux in Table 1 with Bruzual \& Charlot (2003; BC03) population synthesis models. Redshift,

459: mass, extinction, age and e-folding time of star-formation were left

460: as free parameters. The redshift range was $0.1-6$ in steps of 0.1, extinction range was A$_{\rm V}=0-5$ in steps of 0.1,

461: age $t$, varied from 10$^5$ yr to the age of the Universe at each redshift, while the e-folding time $\tau$ 

462: spanned the entire range from instantaneous to constant star-formation.

463: For GOODS-N, we find a best fit redshift of 1.9,

464: stellar mass of 1.3$\times10^{8}$~M$_{\sun}$, A$_{\rm V}$=1.4 mag, $t$=10 Myr and $\tau$=200 Myr, 

465: indicating an ongoing reddened starburst (Figure 2).

466: For GOODS-S, which has far better near-infrared data,

467: we find a best fit redshift of 2.1, stellar mass 

468: of 1.4$\times10^{8}$~M$_{\sun}$, A$_{\rm V}$=1.1 mag, $t$=15 Myr, $\tau$=200 Myr. 

469: The 90\% confidence interval for the redshifts as determined from the lowest chi-square fit

470: at each redshift, spans the range $1.9<z<2.9$ for GOODS-S and $1.3<z<2.7$ for GOODS-N.

471: The $U-$band detection implies that the majority of the sources must be at $z<2.5$.

472: The sharp increase in the flux in the IRAC channels compared to the NIR bands implies

473: that the 1.6$\mu$m bump must be in the IRAC passbands. This implies $z>1.3$. Thus, even a qualitative

474: comparison of the stacked flux densities with a typical galaxy SED results in a redshift range similar

475: to what we estimate from the chi-square values. We adopt $1.9<z<2.7$ as the redshift range of these sources.

476: The uncertainty in the derived stellar mass is about a factor of two, ranging

477: from $1.3-3.0\times10^{8}$~M$_{\sun}$.

478: 

479: For our best fit template, the rest-frame 1500\AA\ UV luminosity

480: of the stacked galaxy is 1.8$\times$10$^{9}$~L$_{\sun}$ while

481: its rest-frame V-band luminosity is 1.2$\times$10$^{9}$~L$_{\sun}$.

482: Since L$_{*,{\rm UV,z=3}}$=5.8$\times$10$^{10}$~L$_{\sun}$ and 

483: L$_{*,{\rm V,z=3}}$=8$\times$10$^{10}$~L$_{\sun}$ \citep{Steidel99, Mar07}, it implies that these

484: faint objects are galaxies which are about $\sim30-60$ times fainter than

485: the characteristic luminosity of field galaxies at these redshifts.

486: 

487: Although the exact redshift distribution of these faint galaxies

488: is not known at this time, it is illustrative to estimate

489: the fraction of the stellar mass density hidden in these objects.

490: We assume that the $\sim$6000 galaxies that we have stacked on

491: are distributed between $1.9<z<2.7$ with a stellar mass of 1.4$\times$10$^{8}$~M$_{\sun}$.

492: This corresponds to a co-moving stellar mass density of 2$\times$10$^{6}$~M$_{\sun}$\,Mpc$^{-3}$ which is a strong lower limit. Assuming that our

493: stacked flux is typical for the $\sim$22000 ACS sources which are unmatched

494: to IRAC sources, would imply a total stellar mass density in faint galaxies

495: of 6.7$\times$10$^{6}$~M$_{\sun}$\,Mpc$^{-3}$ at $z\sim2.4$.  For comparison, the stellar mass density in Lyman-break

496: galaxies at these redshifts is 3.6$\times$10$^{7}$~M$_{\sun}$\,Mpc$^{-3}$ \citep{MED03}.

497: Thus, the faint, sub-L$_{*}$ galaxy population could account for upto 20\% of the  total stellar mass density at $z \sim 2.4$.

498: 

499: \section{Expected Clustering}

500: 

501: We also measure the clustering of these sources to establish the amplitude of unresolved fluctuations at

502: IRAC bands. Since we are interested in the anisotropy of IR light,

503: to establish the power spectrum of fluctuations,

504: we need the flux distribution within this population. The stacking analysis only allows us to establish either the total

505: or the average flux of this sample. This forces us to make an estimate

506: of the flux distribution and use that to predict clustering of unresolved fluctuations  

507: produced by these sources. The angular power spectrum

508: of IR fluxes associated with these sources  is determined

509: using the same technique as described in Sullivan et al. (2007).  

510: We account for the flux distribution by randomly assigning IR flux to the $\sim$  

511: 6000 optical source locations   between 70  nJy and 700 nJy such that the total flux assigned 

512: is the same as the total flux measured for these sources from the stack.

513: We assign fluxes such that the number counts trace the  extrapolated slope from known  IRAC

514: counts down to $m_{\rm AB}(3.6\mu m)$ of 24.7 mag. Given the uncertainty in the

515: faint-end slope, we take the slope $\alpha$ ($dN/dm \propto m^\alpha$) to be

516: either 0.6 or 0.0. We tested how our

517: predictions change with variations to these parameters

518: and found consistent results within errors. Since the flux assignment is 

519: random, to get an average of the expected clustering, we randomize the

520: assignments and the fluxes and use a Monte-Carlo approach to obtain  

521: the mean and variance of clustering.

522: 

523: We summarize our results in Figure~3, where we also

524: compare with a direct measurement of the clustering of IR  

525: fluctuations and expectations based on the clustering  models of Sullivan et al. (2007; solid lines). In Figure~3, we also show  the IR fluctuation spectrum when the faint optical

526: sources that we stack on are masked from the IRAC image with open squares

527: (Cooray et al. 2007). The solid squares show the revised measurement after correcting for a bias

528: associated with the window function introduced by the large mask.

529: To understand this correction noted in Kashlinsky et al. (2007a), 

530: we note that the fluctuation spectrum measurable in Fourier (multipole) space in the presence of a mask is

531: $\hat{C}_\ell = \sum _{\ell'} W_{\ell \ell'} C_{\ell'}$, where

532: $W_{\ell \ell'}$ is the window function associated with the mask and $C_{\ell'}$ is the power spectrum of interest. We recover the latter 

533: with measurements of $\hat{C}_{\ell}$  by first generating  $W_{\ell \ell'} $ associated with our mask (Appendix A of Hivon et al. 2002) 

534: and then iteratively inverting for $C_{\ell}$ using an inversion similar to

535: Dodelson \& Gazta\~naga (2000). The inversion agrees to 10\% with a separate

536: estimate of the angular power spectrum using a likelihood method where

537: one maximizes the likelihood of an estimated $C_\ell^{\rm est}$ with  a model spectrum $C_{\ell}^{\rm inp}$ 

538: using $C_\ell^{\rm est}=\sum_{\ell'} W_{\ell'} C_{\ell'}^{\rm inp}$ given the measurements

539: $\hat{C}_{\ell}$ and the error $\sigma_{\ell}$. This procedure can be described

540: as minimizing the $\chi^2$ with $\chi^2=(\hat{C}_{\ell}-C_{\ell}^{\rm est})^2/\sigma_{\ell}^2$.

541: 

542: As shown in Figure~3, the difference between the fluctuation amplitude of unresolved IR fluctuations

543: and the fluctuation amplitude with ACS sources masked is about a factor of 2 ($\pm$ 0.6) at $\ell \sim 7 \times 10^3$.

544: Our measurements then suggest that, from this difference, up to 50\% of IR fluctuations can be accounted by faint ACS sources.

545: It could be that by accounting for further fainter optically sources, this fraction can be increased, but

546: it is quite

547: unlikely that 100\% of IR fluctuations reported in Kashlinsky et al. (2005, 2007b) are generated by $z>6.5$ galaxies hosting mostly Pop~III stars

548: that are optically invisible.

549: 

550: The difference between IR fluctuations with and without ACS sources masked must be reproduced by the 

551: clustering of IR light produced by the same ACS sources. With flux assignments based on the

552: stacking analysis, we find 

553: that the difference is reproduced (though these indirect estimates of

554: the expected clustering IR fluctuations produced by faint ACS sources are uncertain due to uncertainties in the flux distribution).

555: We conclude that irrespective of whether we measure the power spectrum in

556: the IRB fluctuations after masking optical sources or whether we measure the power spectrum of faint optical

557: galaxies with an average flux based on our stacking analysis, we find that the power in faint ($z_{AB}\sim27$ mag), $z\sim2.5$, 

558: galaxies accounts for at least $\sim$30$-$50\% of the power in the IR background fluctuations

559: on angular scales of $\sim$4$\arcmin$, where the measurement is not strongly sensitive to cosmic variance. 

560: 

561: On similar angular scales, Kashlinsky et al. (2007a) find a power of 0.015 nW~m$^{-2}$~sr$^{-1}$ for the optical galaxies

562: with $z_{AB}>$26.6 mag in HDFN-E2, without a quantification of the uncertainty.  

563: They also demonstrate that the power increases as brighter galaxies are included. 40\% of the galaxies

564: that go into our stacking analysis

565: are brighter than this threshold and therefore the power they measure is a lower limit to our

566: power spectrum shown in Figure 3.

567: This can be compared to their measured value of 0.035 nW~m$^{-2}$~sr$^{-1}$

568: for the amplitude of power in the IR background fluctuations in the HDFN on the same angular 

569: scale/multipole moment.

570: Making use of their estimate of the fluctuations associated with faint, optical galaxies at low redshifts, we find that 

571: Kashlinsky et al. (2007a, 2007b) also find a $\sim$ 40\% contribution from such sources to

572: fluctuations at $\sim$ 4 $\arcmin$ angular scales, though their contribution is ignored in 

573: their subsequent interpretation

574: that assumes all fluctuations are from Pop~III stars before reionization.

575: The analysis of the optical sources in GOODS-S is not presented in Kashlinsky et al. (2007a). The 

576: fundamental discrepancy between our two groups

577: arises on larger angular scales, at 400 arcsec or larger, where they claim the power spectrum is rising. We are unable

578: to constrain the power spectrum on such large angular scales ($\ell \sim 1500$) due to the cosmic variance associated with the

579: limited size of the GOODS fields used in our analysis, which becomes important at angular scales above 400 arcseconds.

580: 

581: \section{Discussion}

582: 

583: The Kashlinsky et al. (2007c) interpretation for IR fluctuations involves

584: $z>6.5$ galaxies with an IR background intensity of $\gtrsim$ 1 nW m$^{-2}$  

585: sr$^{-1}$ at 3.6 $\mu$m. This is a large intensity given that all resolved sources, so far,  

586: lead to an IR background intensity between $\sim$ 6 nW m$^{-2}$ sr$^{-1}$ 

587: %when $m_{\rm AB}(3.6\mu m)<27$ 

588: (Sullivan et al. 2007; Fazio et al. 2004) and $\sim$ 10 nW m$^{-2}$ sr$^{-1}$ (Levenson \& Wright 2008).  

589: To avoid individual  detections of a large number of $z>6.5$ galaxies, such a

590: high background intensity must then be hidden in a source population with  

591: a large surface density but with individual faint fluxes of around 10 nJy for  

592: each source. Given the large surface density of the   population, pixel to pixel intensity

593: variations  are a small fraction of the total intensity produced by  

594: those sources.

595: 

596: The alternative explanation is that a reasonable fraction of fluctuations arises from

597: a population of sources that have a low surface density but are just below

598: the {\it Spitzer}/IRAC detection threshold. The results from the stacking analysis presented here 

599: suggest that faint optical galaxies, with an average source flux of  

600: around 130 nJy at 3.6 $\mu$m, contribute between $\sim$30$-$50\% of the power

601: in the fluctuations on the largest angular scales that we are able

602: to measure.   The average IR absolute intensity produced by these sources is in the range

603: 0.12-0.35 nW m$^{-2}$ sr$^{-1}$ at 3.6 $\mu$m, in agreement with the

604: Kashlinsky et al. (2005) estimate that the contribution

605: from faint extragalactic sources is $\sim$0.15 nW m$^{-2}$ sr$^{-1}$. 

606: However, they claim that such faint objects

607: cannot account for the strong clustering signal in unresolved IR light. We have demonstrated in this paper that a significant

608: fraction of the fluctuations do arise from these faint galaxies.

609: These sub-L$_{*}$ galaxies contribute up to 20\% of the stellar mass density at $z \sim 2.5$

610: and can be described as a low luminosity version of well-studied Lyman-break galaxies.

611: 

612: While these sources do not fully explain all IR fluctuations measured

613: with deep IRAC images, in Cooray et al.  

614: (2007), we placed a conservative upper limit that any

615: contribution from $z>6.5$ sources must have an absolute intensity less than  

616: about 0.6 nW m$^{-2}$ sr$^{-1}$.  Beyond {\it Spitzer} IRAC, studies have also been conducted at  

617: lower near-IR wavelengths with NICMOS. Though limited 

618: by the small field-of-view of NICMOS, they show that   IR fluctuations at 1.6 and 1.25 $\mu$m are

619: more consistent with a $z<8$ origin  than a high redshift  

620: interpretation (Thompson et al. 2007a, 2007b).

621: Beyond IR fluctuations, it will be interesting

622: to further study the nature of the faint, dwarf galaxy population at $z=2$ to 3 

623: to understand the contribution it makes towards bridging the systematic offset between the 

624: co-moving stellar mass density and the integrated star-formation rate density (Hopkins \& Beacom 2006).

625: 

626: {\it Acknowledgments:}

627: We thank Mark Dickinson for helpful suggestions. We also acknowledge the 

628: contributions of various members of the GOODS team who are associated with key

629: aspects of data processing and handling.

630: This research was funded by NASA grant NNX07AG43G, NSF CAREER grant AST-0645427,

631: award number 1310310 from {\it Spitzer} for Archival Research, 

632: and program number HST-AR-11241.01-A by NASA through

633: a grant from the Space Telescope Science Institute, which is operated  

634: by the Association of Universities for Research in Astronomy, Incorporated,  

635: under NASA contract NAS5-26555.

636: 

637: %\input{ref}

638: 

639: \begin{thebibliography}{99}

640: \frenchspacing

641: 

642: \bibitem[Aharonian et al]{Aha}

643:    Aharonian, F. et al. 2006, Nature, 440, 1018

644: 

645: \bibitem[Babbedge et al.]{Bab06}

646:    Babbedge, T. S. R., Rowan-Robinson, M., Vaccari, M. et al. 2006,  

647: MNRAS, 370, 1159

648: 

649: %\bibitem[Bertin \& Arnouts 1996]{Ber96}

650: %Bertin, E. \& Arnouts, S. 1996, A\&AS, 117, 393

651: \bibitem[Bruzual \& Charlot 2003]{BC03}

652: Bruzual, G., \& Charlot, S., 2003, MNRAS, 344, 1000

653: 

654: \bibitem[Capak et al. 2004]{Capak04}

655:    Capak, P., et al., 2004, AJ, 127, 180

656: 

657: \bibitem[Cooray \& Sheth]{CooShe02}

658:    Cooray, A. \& Sheth, R. 2002, Phys. Rep., 372, 1

659: %%CITATION = ASTRO-PH 0206508;%%

660: 

661: \bibitem[Cooray et al. 2004]{Coo04a}

662:    Cooray, A. et al. 2004, ApJ, 606, 611; erratum-ibid.\ 2005, ApJ,  

663: 622, 1363.

664:   %%CITATION = ASTRO-PH 0308407;%%

665: 

666: \bibitem[Cooray \& Yoshida 2004]{CooYos04}

667:    Cooray, A. \& Yoshida, N. 2004, MNRAS, 351, L71

668: %%CITATION = ASTRO-PH 0404109;%%

669: 

670: 

671: \bibitem[Cooray et al. 2007]{Coo07}

672:    Cooray, A. et al. 2007, ApJ, 659, L91

673:   %%CITATION = ASTRO-PH 0612609;%%

674: 

675: \bibitem[Dickinson et al. 2003]{Dic03}

676:    Dickinson, M. et al. 2003, in Proceedings of the ESO Workshop, The  

677: Mass of Galaxies at Low and High Redshifts'', eds. R. Bender \&  A.  

678: Renzini, p. 324 (astro-ph/0204213)

679: 

680: \bibitem[Dickinson et al. 2003]{MED03}

681:    Dickinson, M. et al. 2003, ApJ, 587, 25

682: 

683: \bibitem[Dodelson \& Gazta\~naga]{Dod00}

684:   Dodelson, S. \& Gazta\~naga, E. 2000, MNRAS, 312, 774

685: 

686: \bibitem[Dwek et al.]{Dwek}

687: Dwek, E., Arendt, R. \& Krennrich, F. 2005, ApJ, 635, 784

688: 

689: \bibitem[Giavalisco et al.]{Gi04}

690:    Giavalisco, M. et al. 2004, ApJL, 600, 93

691: 

692: \bibitem[Fazio et al]{Faz04}

693:    Fazio, C. G. et al. 2004, ApJS, 154, 39

694: 

695: \bibitem[Fernandez \& Komatsu]{FerKom06}

696:    Fernandez, E. \& Komatsu, E. 2006, ApJ, 646, 703

697: 

698: \bibitem[Hauser \& Dwek]{HauDwe01}

699:    Hauser, M.G. \& Dwek, E. 2001, ARA\&A, 39, 249

700: 

701: \bibitem[Hopkins \& Beacom 2006]{HopBea06}

702:   Hopkins, A. M., \& Beacom, J. F., ApJ, 651, 142

703: 

704: \bibitem[Hivon et al.]{Hiv02}

705:   Hivon, E. et al. 2002, ApJ, 567, 2

706: 

707: %\bibitem[Kashlinsky 2004]{Kas05}

708: %  Kashlinsky, A. 2005, Phys. Rept., 409, 361.

709: 

710: \bibitem[Kashlinsky et al 2004]{Kasetal04}

711:    Kashlinsky, A. et al. 2004, ApJ, 608, 1

712: 

713: \bibitem[Kashlinsky et al 2005]{Kasetal05}

714:    Kashlinsky, A. et al. 2005, Nature, 438, 45

715: 

716: \bibitem[Kashlinsky et al 2007a]{Kasetal07a}

717:    Kashlinsky, A. et al. 2007a, ApJ, 666, L1

718: 

719: \bibitem[Kashlinsky et al 2007b]{Kasetal07b}

720:    Kashlinsky, A. et al. 2007b, ApJ, 654, L5

721: 

722: \bibitem[Kashlinsky et al 2007c]{Kasetal07c}

723:    Kashlinsky, A. et al. 2007c, ApJ, 654, L1

724: 

725: %\bibitem[Kelsall et al 1998]{Keletal98}

726: %Kelsall, T. et al. 1998, ApJ, 508, 44

727: 

728: \bibitem[Levenson \& Wright 2008]{Lev08}

729: Levenson, L. \& Wright, E. L. ApJ submitted, arXiv:0802.1239

730: 

731: \bibitem[Madau \& Silk]{Madau}

732: Madau, P. \& Silk, J. 2005, MNRAS, 359, L37

733: 

734: \bibitem[Marchesini et al. 2007]{Mar07}

735: Marchesini, D., et al. 2007, \apj, 656, 42

736: 

737: \bibitem[Matsumoto et al 2005]{Matetal05}

738:    Matsumoto, T. et al. 2005, ApJ, 626, 31

739: 

740: \bibitem[Salvaterra \& Ferrara]{SalFer03}

741:    Salvaterra, R. \&  Ferrara, A. 2003, MNRAS, 339, 973

742: 

743: %\bibitem[Salvaterra et al.]{Saletal06}

744: %   Salvaterra, R. et al. 2006, MNRAS is press, astro-ph/0512403

745: 

746: \bibitem[Santos et al]{Sanetal02}

747:    Santos, M. R., Bromm, V., \& Kamionkowski, M. 2002, MNRAS, 336, 1082

748: 

749: \bibitem[Steidel et al. 1999]{Steidel99}

750: {Steidel}, C.~C., et al. 1999, \apj, 519, 1

751: 

752: \bibitem[Sullivan et al.]{Sul07}

753:    Sullivan, I. et al. 2007, ApJ, 657, 37

754: %%CITATION = ASTRO-PH 0609451;%%

755: 

756: \bibitem[Thompson et al.]{Tho06}

757:    Thompson, R. et al. 2007a, ApJ, 657, 669

758: 

759: \bibitem[Thompson et al.]{Tho07}

760:    Thompson, R. et al. 2007b, ApJ, 666, 658

761: 

762: \bibitem[Wright \& Reese]{WriRee01}

763:    Wright, E. L. \& Reese, E. D. 2000, ApJ, 545, 43

764: 

765: \end{thebibliography}

766: 

767: \end{document}

768: 

769: 

770: 

771: 

772: 

773: 

774: 

775: 

776: 

777: 

778: 

779: 

780: 

781: