1: %

2: % Draft for the ``Omukai'' paper in an arbitrary LaTeX template

3: %

4: % [RSK: 12.04.07]

5: % 

6: 

7: % latex draft-02 ;  dvips -Ppdf -G0 -ta4 -o draft-02.ps draft-02 ; ps2pdf draft-02.ps

8: 

9: 

10: 

11: 

12: \NeedsTeXFormat{LaTeX2e}

13: 

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

15: \documentclass[a4paper]{emulateapj}

16: %\documentclass[10pt]{article}

17: %\usepackage[pdftex]{graphicx}

18: %\usepackage{epstopdf}

19: \usepackage{graphicx}

20: \usepackage{natbib}

21: %\usepackage{epsf}

22: %\usepackage{psfrag}

23: %\usepackage{color}

24: \usepackage{epsf}

25: 

26: 

27: 

28: 

29: \def\solmas{$\mathrm{M_\odot}$~}

30: \def\solmasp{$\mathrm{M_\odot}$}

31: \def\tcr{$t_{\mathrm{cr}}$}

32: \def\tff{$t_{ff}$}

33: \def\sims{$\sim$}

34: \def\numdensu{cm$^{-3}$}

35: \def\massdensu{g~cm$^{-3}$}

36: \def\solmet{$\mathrm{Z_{\odot}}$}

37: \def\met{$\mathrm{Z}$}

38: 

39: \def\araa{{\em ARAA}}

40: %\def\aas{{\em Astron.\ Astrophys.\ Suppl.\ Ser.}}

41: \def\aj{{\em AJ}}

42: %\def\anap{{\em Ann.\ Astrophys.}}   %{\em Annales d'Astrophysique}

43: %\def\an{{\em Astron. Nach.}}

44: \def\apj{{\em ApJ}}

45: \def\apjl{{\em ApJ}}

46: \def\apjs{{\em ApJS}}

47: \def\aap{{\em A\&A}}

48: \def\apss{{\em Astrophys.\ Space Science}}

49: \def\baas{{\em Bull.\ Amer.\ Astron.\ Soc.}}

50: \def\bain{{\em Bull.\ Astron.\ Inst.\ Netherlands}}

51: \def\fcp{{\em Fund.\ Cosm.\ Phys.}}

52: \def\jcam{{\em J.\ Comput.\ Appl.\ Math.}}

53: \def\jcp{{\em J.\ Comput.\ Phys.}}

54: \def\jfm{{\em J.\ Fluid Mech.}}

55: \def\mnras{{\em MNRAS}}

56: \def\nat{{\em Nature}}

57: \def\pta{{\em Phil.\ Trans.\ A.}}

58: \def\ptp{{\em Prog.\ Theo.\ Phys.}}

59: \def\prd{{\em Phys.\ Rev.\ D}}

60: \def\pre{{\em Phys.\ Rev.\ E}}

61: \def\prl{{\em Phys.\ Rev.\ Lett.}}

62: \def\prsa{{\em Proc.\ R.\ Soc.\ London A}}

63: \def\pasj{{\em Pub.\ Astron.\ Soc.\ Japan}}

64: \def\pasp{{\em PASP}}

65: \def\pfl{{\em Phys.\ Fluids}}

66: \def\ppl{{\em Phys.\ Plasmas}}

67: \def\rpp{{\em Rep.\ Prog.\ Phys.}}

68: \def\rmp{{\em Rev.\ Mod.\ Phys.}}

69: \def\zp{{\em Z.\ Phys.}}

70: \def\za{{\em Z.\ Astrophys.}}

71: 

72: \newcommand{\mHt}{{\rm H_{2}}}

73: \def\simless{\mathbin{\lower 3pt\hbox

74:    {$\rlap{\raise 5pt\hbox{$\char'074$}}\mathchar"7218$}}}

75: \def\simgreat{\mathbin{\lower 3pt\hbox

76:    {$\rlap{\raise 5pt\hbox{$\char'076$}}\mathchar"7218$}}}

77: \newcommand{\zfrag}{${\rm Z} = 10^{-5} \: {\rm Z_{\odot}} $}

78: \newcommand{\zsix}{${\rm Z} = 10^{-6} \: {\rm Z_{\odot}} $}

79: \newcommand{\zleqsix}{${\rm Z} \leq 10^{-6} \: {\rm Z_{\odot}} $}

80: 

81: \begin{document}

82: 

83: \title{The First Stellar Cluster}

84: 

85: \author

86: {Paul C.\ Clark$^1$, Simon C.\ O.\ Glover$^2$, Ralf S.\ Klessen$^1$}

87: 

88: 

89:  \affil{$^{1}$Zentrum f\"ur Astronomie der Universit\"at Heidelberg, Institut

90: f\"ur Theoretische  Astrophysik, Albert-Ueberle-Str.\ 2, 69120 Heidelberg, Germany

91: \break email: pcc@ita.uni-heidelberg.de, rklessen@ita.uni-heidelberg.de \\

92: $^2$Astrophysikalisches Institut Potsdam, An der Sternwarte 16, D-14482 Potsdam,

93: Germany  \break email: sglover@aip.de}

94: 

95: \begin{abstract} 

96: 

97: We report results from numerical simulations of star formation in the early universe that

98: focus on gas at very high densities and very low metallicities. We argue that the  gas in

99: the central regions of protogalactic halos will fragment as long as it carries sufficient

100: angular momentum. Rotation leads to the build-up of massive disk-like structures which

101: fragment to form protostars. At metallicities ${\rm Z} \approx 10^{-5}\,{\rm Z}_{\odot}$,

102: dust cooling  becomes effective and leads to a sudden drop of temperature at densities above

103: $n = 10^{12}\,$cm$^{-3}$. This induces vigorous fragmentation, leading to a very

104: densely-packed cluster of low-mass stars. This is the first stellar cluster. The mass

105: function of stars peaks below $1\,$M$_{\odot}$, similar to what is found in the solar

106: neighborhood, and comparable  to the masses of the very-low metallicity subgiant stars

107: recently discovered in the  halo of our Milky Way. We find that even purely primordial gas

108: can fragment at  densities $10^{14}\,$cm$^{-3} \le n \le 10^{16}\,$cm$^{-3}$, although the

109: resulting mass function contains only a few objects (at least a factor of ten less than the

110: \zfrag~mass function), and is biased towards higher masses. A similar result is found for

111: gas with \zsix. Gas with Z $\le 10^{-6}\,$Z$_{\odot}$ behaves roughly isothermally at these

112: densities  (with polytropic exponent $\gamma \approx 1.06$) and the massive disk-like

113: structures  that form due to angular momentum conservation will be marginally unstable. As 

114: fragmentation is less efficient, we expect stars with Z $\le 10^{-6}\,$Z$_{\odot}$  to be

115: massive, with masses in excess of several tens of solar masses, consistent with the results

116: from previous studies.

117: 

118: \keywords{stars: formation -- stars: mass function -- early universe -- hydrodynamics --

119: equation of state -- methods: numerical }

120: \end{abstract}

121: 

122: \maketitle

123: 

124: \section{Introduction} Detailed numerical simulations of the formation of the first stars,

125: the  so-called population III stars, indicate that they were probably massive, with masses

126: greater than $20 \: {\rm M_{\odot}}$ \citep{abn02,bcl02,yoha06,oshn07}. The fact that no

127: population III star has ever been observed in the Milky Way provides some observational

128: support for this prediction \citep{tum06}. However, a number of low-mass, extremely

129: metal-poor, stars with $[{\rm Fe}/{\rm H}] < -3$ have been discovered in the  Galactic halo

130: \citep{bc05}, suggesting that the distribution of stellar masses is sensitive to even very

131: low levels of metal enrichment.  Explanations of this apparent change in the IMF have

132: concentrated on the fact that metal-enriched gas has more coolants than its primordial

133: counterpart.  These coolants are suggested to provide an opportunity for efficient

134: fragmentation, since they can keep the gas temperature lower during the collapse process

135: than is possible with pure H$_{2}$ cooling in the primordial gas. If this is the case, then

136: the final IMF will likely contain at least some low-mass stars, even if it still differs

137: significantly from the present-day local IMF.

138: 

139: If metal enrichment is the key to the formation of low-mass stars, then logically

140: there  must be some critical metallicity ${\rm Z_{\rm crit}}$ at which the formation

141: of low mass stars first becomes possible. However, the value of  ${\rm Z_{\rm crit}}$

142: is a matter of ongoing debate. Some models suggest that low mass star formation

143: becomes possible only once atomic fine-structure line cooling from carbon and oxygen

144: becomes effective \citep{bfcl01,bl03,san06,fjb07}, setting a value for ${\rm Z_{\rm

145: crit}}$ at around $10^{-3.5} \: {\rm Z_{\odot}}$. Another possibility, and the one

146: that we explore with this paper, is that low mass star formation is a result of

147: dust-induced fragmentation occurring at high densities, and thus at a very late stage

148: in the protostellar collapse \citep{sch02,om05,sch06}. In this model,

149: $10^{-6}~\simless~{\rm Z_{\rm crit}}~\simless~10^{-4}~{\rm Z_{\odot}}$, where much of

150: the uncertainty in the predicted value results from uncertainties in the dust

151: composition and the degree of gas-phase depletion \citep{sch02,sch06}.

152: 

153: The recent simulations performed by \citet{to06}, which model the collapse of very high

154: density protogalactic gas, provide some support for the dust-induced fragmentation model.

155: Using a simple piecewise polytropic equation of state to describe the thermal evolution of

156: extremely metal-poor protogalactic gas, \citet{to06} show that fragmentation can occur at

157: metallicities as low as ${\rm Z} = 10^{-6} \: {\rm Z_{\odot}}$, and that it becomes more

158: effective as the metallicity increases. However, their study considered only the limiting

159: case of gas with zero angular momentum. Owing to the absence of angular  momentum, the

160: fragments formed in their simulation do not survive for longer than a  dynamical time, as

161: they simply fall to the center of the potential well, where they merge with other fragments.

162: It is also unclear whether the fragmentation process would be as effective if the angular

163: momentum of the gas were non-zero, as would be expected in reality.

164: 

165: We present the results of simulations of the high-density, dust-cooling dominated

166: regime that improve on those of \citet{to06} by including the effects of rotation,

167: and by following a much larger dynamical range of the collapse (ten orders of

168: magnitude in density), as well as employing an equation of state (EOS) which follows

169: that of Omukai et al.\ (2005) more closely. We also perform some simulations of purely

170: primordial gas for comparison. A key feature is the use  of sink particles (Bate,

171: Bonnell \& Price 1995) to capture the formation and evolution of multiple collapsing

172: cores, which enables us to follow the evolution of the star-forming gas over several

173: free-fall timescales and thus to  model the build-up  of a stellar cluster. This

174: differs from previous calculations which either follow the collapse of a single core

175: to high densities \citep{abn02,bl04,yoha06}, or use sink particles to capture low 

176: density ($n < 10^{6}$ \numdensu) fragmentation \citep[e.g.][]{bcl02}.

177: 

178: In following section, we give details of the numerical model (\S\ref{code}) and the initial

179: set-up of the simulations (\S\ref{setup}). The main results of our study are presented in

180: \S\ref{results} and we discuss the origin of the fragmentation in \S\ref{discussion}.

181: Potential caveats with the current model are highlighted in \S\ref{caveats}, and there is a

182: summary of our findings in \S\ref{conclusions}.

183: 

184: 

185: 

186: \begin{figure}[t]

187: %\begin{center}

188: \includegraphics[width=3.1in,height=1.75in]{f1}

189: %\end{center}

190: 

191: \caption{\label{fig:EOS} The three equations of state (EOSs) from Omukai et~al. (2005) that are used

192: in our study. The primordial case (solid line), \zsix~(dotted line), and

193: \zfrag~(dashed line), are shown alongside an example of a polytropic EOS with an

194: effective $\gamma = 1.06$. }

195: 

196: \end{figure}

197: 

198: 

199: 

200: \begin{figure*}[t]

201: \centerline{

202: \includegraphics[width=6.4in,height=4.18in]{f2}

203: }

204: 

205: \caption{\label{fig:sequence} Time evolution of the density distribution in the innermost 400 AU of the

206: protogalactic halo shortly before and shortly after the formation of the first protostar at

207: $t_{\rm SF}$. We plot only gas at densities above $10^{10}\,$cm$^{-3}$.  The dynamical

208: timescale at a density $n = 10^{13}\,$cm$^{-3}$ is of the order of only 10 years.  Dark dots

209: indicate the location of protostars as identified by sink particles forming at $n \ge

210: 10^{17}\,$cm$^{-3}$. Note that without usage of sink particles we would not have been able

211: to follow the build-up of the protostellar cluster beyond the formation of the first object.

212: There are 177 protostars when we stop the calculation at $t = t_{\rm SF} + 420\,$yr. They

213: occupy a region roughly a hundredth of the size of the initial cloud. With

214: $18.7\,$M$_{\odot}$  accreted at this stage, the stellar density is $2.25 \times

215: 10^{9}\,$M$_{\odot}\,$pc$^{-3}$.}

216: 

217: \end{figure*}

218: 

219: 

220: \begin{figure*}[t]

221: \begin{center}

222: \unitlength1cm

223: \begin{picture}(18,5.5)

224: \put(1.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f3a}}

225: \put(7.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f3b}}

226: \put(13.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f3c}}

227: \end{picture}

228: \end{center}

229: \caption{ \label{fig:density} We illustrate the onset of the fragmentation process in the high resolution

230: \zfrag~simulation. The graphs show the densities of the particles, plotted as a

231: function of their x-position. Note that for each plot, the particle data has been centered on

232: the region of interest. We show here results at three different output times, ranging from

233: the time that the first star forms ($t_{\rm sf}$) to 221 years afterwards. The densities 

234: lying between the two horizontal dashed lines denote the range over which

235: dust cooling lowers the gas temperature.}

236: 

237: \end{figure*}

238: 

239: 

240: \begin{figure*}[t]

241: \centerline{

242: 	\includegraphics[width=2.1in,height=2.1in]{f4a}

243: 	\includegraphics[width=2.1in,height=2.1in]{f4b}

244: 	\includegraphics[width=2.1in,height=2.1in]{f4c}

245: }

246: 

247: \caption{\label{fig:masses}  Mass functions resulting from simulations with metallicities

248: ${\rm Z} = 10^{-5} \: {\rm Z_{\odot}}$ (left-hand panel), ${\rm Z} = 10^{-6} \: {\rm

249: Z_{\odot}}$ (center panel), and ${\rm Z} = 0$ (right-hand panel). The plots refer to the

250: point in each simulation at which 19 \solmas of material has been accreted (which occurs at

251: a slightly different time in each simulation). The mass resolutions are 0.002 \solmas and

252: 0.025 \solmas for the high and low resolution simulations, respectively.  Note the

253: similarity between the results of the low-resolution and high-resolution simulations. The

254: onset of dust-cooling in the \zfrag~cloud results in a stellar cluster which has a mass

255: function similar to that for present day stars, in that the majority of the mass resides in

256: the lower-mass objects. This contrasts with the \zsix~and primordial clouds, in which the

257: bulk of the cluster mass is in high-mass stars.}

258: 

259: \end{figure*}

260: 

261: \section{Details of the Calculations} 

262: 

263: \subsection{Numerical Method} 

264: \label{code} 

265: 

266: We follow the evolution of the gas in this study using Smoothed Particle

267: Hydrodynamics (SPH). Our version of the code is essentially a parallelized version of

268: that used by Bate et~al.~(1995), which utilizes adaptive smoothing lengths for the gas

269: particles and a binary tree algorithm for computing gravitational forces and

270: constructing SPH neighbor lists. The calculations were performed on the J{\"u}lich

271: Multi-processor (JUMP) supercomputer at the John von Neumann Institute for Computing,

272: Research Center J{\"u}lich, Germany.

273: 

274: The thermal evolution of the gas in our simulations is modelled using a tabulated

275: equation of state (EOS) that is based on the results of the detailed chemical

276: modelling of \citet{om05}. We use their reported results for the temperature and

277: molecular fraction as functions of gas density (their Figures 1 \& 3) with

278: metallicities  ${\rm Z} = 0$, $10^{-6}$ and $10^{-5} \: {\rm Z_{\odot}}$ to compute 

279: the internal energy density and thermal pressure of the gas at a  range of different

280: densities. These values are used to construct look-up tables which are then used by

281: the SPH code to compute the pressure and internal energy at any required density, for

282: a given gas metallicity, via linear interpolation (in log-log space) between the tabulated values.

283: The resulting equations of state are plotted in Figure

284: \ref{fig:EOS}. By using a tabulated equation of state, we avoid the large

285: computational expense involved in solving the full thermal energy equation, while

286: still obtaining qualitatively correct behavior. 

287: 

288: To model the star formation in this study we use sink particles, as described by Bate

289: et~al.\ (1995). This involves replacing the innermost parts of dense, self-gravitating 

290: regions of gas with particles that can both accrete further material from their 

291: surroundings and interact with other particles in the simulation via gravity.  Sink

292: particles are formed once an SPH particle and its neighbors are gravitationally

293: bound, collapsing (negative velocity divergence), and within an accretion radius, 

294: $h_{acc}$, which is taken here to be 0.4 AU. Gravitational interactions between 

295: the sink particles and all other particles in the simulation are also smoothed, 

296: self-consistently, to $h_{acc}$. Our set-up allows us to identify sink particles 

297: as the direct progenitors of individual stars \cite[for a more detailed discussion, 

298: see e.g.][]{WK01}.

299: 

300: \subsection{Setup and Initial conditions}

301: \label{setup}

302: 

303: Our calculations are designed to start from the point where previous fragmentation 

304: calculations ended. It is now well established that the gas which falls into

305: collapsing dark matter mini-halos undergoes a phase of fragmentation, resulting in

306: the formation of large, self-gravitating clumps, with masses of order 100 \solmasp.

307: The origin of this fragmentation is the rapid cooling driven by ${\rm H_{2}}$ that

308: occurs at densities of around 10 -- $10^{4}$ \numdensu, and the masses of the clumps are

309: similar to the Jeans mass at the corresponding density and temperature in this

310: regime. We thus pick up the evolution from conditions similar to those reached in the

311: study of \citet[][]{bfcl01}. Since the process of chemical enrichment involves

312: supernovae events from the first stars we expect the gas to have a certain initial

313: level of turbulence, which is normally assumed to be absent during the formation of

314: the very first stars from purely primordial gas. We also adopt a net angular momentum

315: for the gas consistent with the results of simulations of cosmological structure

316: formation.

317: 

318: The clouds in this study have a mass of 500 \solmasp, and are modelled using either 2

319: million or 25 million SPH particles. In the higher resolution calculations, this gives a

320: particle mass of $2 \times 10^{-5} \: {\rm M_{\odot}}$ and a mass resolution of 0.002

321: \solmas (Bate \& Burkert 1997), roughly 10 times smaller than the opacity limit set by the

322: \citet[][]{om05} equation of state. These simulations therefore have roughly an order of

323: magnitude of surplus resolution. The low-resolution calculations have a mass resolution

324: roughly equal to the mass at which the gas becomes optically thick. To ensure that the Jeans

325: condition is not violated, the sinks are always formed just before the optically thick

326: regime in all the simulations. Our clouds have an initial radius of 0.17pc, at an initial

327: uniform density of $5 \times 10^{5}$~\numdensu. This corresponds to an initial free-fall

328: time of $t_{\rm ff} = (3\pi/32G\pi\rho)^{1/2} = 5.1 \times 10^{4}$ years. At this scale and

329: density regime, the contributions from dark matter to the gravitational potential are small

330: and are thus not taken into account in our computational set up. One can see from Figure

331: \ref{fig:EOS} that the different gas metallicities have slightly different internal energies

332: at the starting density. Thus, the \zleqsix~calculations have an initial ratio of thermal to

333: gravitational energy of $\alpha = E_{\rm therm}/ |E_{\rm grav}| =0.39$, while the cooler

334: \zfrag~calculations have an initial value of $\alpha = 0.32$. All simulations are given a

335: low level of initial turbulence, with the ratio of turbulent to gravitational potential

336: energy $E_{\rm turb} / |E_{\rm grav}| = 0.1$, and thus an RMS Mach number of

337: $\mathcal{M}_{RMS} \approx 1$. The clouds are set in initially uniform rotation, with $\beta

338: = E_{\rm rot}/ |E_{\rm grav}| = 0.02$. The conditions at the start of our simulation are thus

339: similar to those at which \cite{bcl02} form their sink particles (see their Figure 5 for

340: comparison).We perform low-resolution simulations for all three metallicity cases, and

341: high-resolution simulations for the primordial and \zfrag~cases.

342: 

343: \begin{figure*}[t]

344: \centerline{

345: \includegraphics[width=3.1in,height=3.1in]{f5a}

346: \includegraphics[width=3.1in,height=3.1in]{f5b}

347: }

348: 

349: \caption{ \label{fig:massrho} The mass (upper panels) and the number of Jeans masses (lower panels) are both

350: shown here as a function of gas density, with the high-resolution \zfrag~simulation in the

351: left-hand plot and both the high-resolution primordial and low-resolution \zsix~simulations

352: in the right-hand plot. In the upper panels, we plot the mass residing above a density,

353: $n_{\rm gas}$, as a function of that density, as well as the Jeans mass. In the lower

354: panels, we plot the number of Jeans masses as a function of density. In the left-hand panel,

355: we show via the shaded areas, the heating (pink) and cooling (light blue) phases in the

356: \zfrag~EOS. In the right-hand plot, we show the conditions in the primordial

357: (high-resolution) simulation (solid lines), and \zsix~simulation (dashed lines). The gas

358: conditions are shown at the point of star formation and 50 years earlier, as well as at two

359: instants after star formation, when the clouds have converted around 9.5\,\solmas and

360: 19\,\solmas of gas into protostars. Note that we only label the evolutionary stages in the

361: lower panels, but the same progression with time, going from lower left-most lines to those

362: at the top, applies also to the upper panels. }

363: \end{figure*}

364: 

365: \section{Results}

366: \label{results}

367: 

368: We find that enrichment of the gas to a metallicity of only ${\rm Z} = 10^{-5} \:{\rm

369: Z_{\odot}}$ dramatically enhances fragmentation within the dense, collapsing cloud.

370: We first  focus our discussion on the evolution of the ${\rm Z} = 10^{-5} \: {\rm

371: Z_{\odot}}$ gas cloud, before contrasting with the  $10^{-6} \: {\rm Z_{\odot}}$ and

372: primordial clouds.

373: 

374: The evolution of the high-resolution \zfrag~simulation is illustrated in Figure

375: \ref{fig:sequence}, in a series of snapshots showing the density distribution of the

376: gas. We show several stages in the collapse process, spanning a time interval from

377: shortly before the formation of the first protostar (as identified by the formation of

378: a sink particle in the simulation) to 420 years afterwards. Note that we show only the

379: innermost 1\% of the full computational domain. A complementary view is given in Figure

380: \ref{fig:density}, which shows the particle densities as a function of position.

381: 

382: During the initial contraction, the cloud builds up a central core with a density of

383: about $n = 10^{10}\,$cm$^{-3}$. This core is supported by a combination of thermal

384: pressure and rotation. Eventually, the core reaches high enough densities to go into

385: free-fall collapse, and forms a single protostar. As more high angular momentum

386: material falls to the center, the core evolves into a disk-like structure with density

387: inhomogeneities caused by low levels of turbulence.  As it grows in mass, its density

388: increases. When dust-induced cooling sets in, it fragments heavily into a tightly

389: packed protostellar cluster within only a few hundred years. One can see this behavior

390: in particle density-position plots in Figure \ref{fig:density}. We stop the simulation

391: 420 years after the formation of the first stellar object (sink particle). At this

392: point, the core has formed 177 stars. The evolution in the low-resolution simulation is

393: very similar. The time between the formation of the first and second protostars is

394: roughly 23 years, which is two orders of magnitude higher than the free-fall time at

395: the density where the sinks are formed. Without the inclusion of sink particles, we

396: would only have been able to capture the formation of the first collapsing object which

397: forms the first protostar: {\em the formation of the accompanying cluster would have

398: been missed entirely}.

399: 

400: The mass functions of the protostars at the end of the \zfrag~simulations (both high

401: and low resolution cases) are shown in Figure \ref{fig:masses} (left-hand panel). When

402: our simulation is terminated, the sink particles hold $\sim$ 19 \solmas of gas in

403: total. The mass function peaks somewhere below $0.1\ $M$_{\odot}$ and ranges from below

404: 0.01$\,$M$_{\odot}$ to about $5\ $M$_{\odot}$. It is important to stress here that this

405: is not the final protostellar mass function. The continuing accretion of gas by the

406: cluster will alter the mass function, as will mergers between the newly-formed

407: protostars (which cannot be followed using our current sink particle implementation).

408: Protostellar feedback in the form of winds, jets and H{\sc ii} regions may also play a

409: role in determining the shape of the final stellar mass function. However, a key point

410: to note is that the chaotic evolution of a bound system such as this cluster ensures

411: that a wide spread of stellar masses will persist. Some stars will enjoy favourable

412: accretion at the expense of others that will be thrown out of the system (as can be

413: seen in Figure \ref{fig:sequence}), thus having their accretion effectively terminated

414: (see for example, the discussions in Bonnell \& Bate 2006 and Bonnell, Larson \&

415: Zinnecker 2007). The survival of some of the low mass  stars formed in the cluster is

416: therefore inevitable.

417: 

418: Our calculations demonstrate that the dust-cooling model of \citet{om05} can indeed

419: lead to the formation of low-mass objects from gas with very low metallicity. This

420: suggests that the transition from high-mass primordial stars to Population II stars

421: with a more ``normal'' mass spectrum occurs early in the universe, at metallicities at

422: or below Z $\approx10^{-5}\ $Z$_{\odot}$. Our simulations even predict the existence of

423: brown dwarfs in this metallicity range. Their discovery would be a critical test for

424: our model of the formation of the first star cluster. The first hints of its validity

425: come from the very low metallicity sub-giant stars that have recently been discovered

426: in the Galactic halo \cite[][]{christlieb02,bc05}, which have iron abundances less than

427: $10^{-5}$ times the solar value and masses below one solar mass, consistent with the

428: range reported here. 

429: 

430: Turning our attention to the ${\rm Z} = 0$ and \zsix~calculations, we also find that

431: fragmentation of the gas occurs, albeit at a much lower level than in the ${\rm Z} =

432: 10^{-5} \: {\rm Z_{\odot}}$ run. The mass functions from these simulations are shown in

433: Figure \ref{fig:masses} (middle and right-hand panels), and are again taken when $\sim$

434: 19 \solmas of gas has been accreted onto the sink particles, the same amount as is

435: accreted by the end of the \zfrag~calculations. 

436: 

437: The primordial gas clouds form fewer protostars than the \zfrag~clouds, with the high

438: resolution simulation forming 25 sink particles and the low-resolution simulation

439: forming 22 sink particles. The mass functions are considerably flatter than the present

440: day IMF, in agreement with the suggestion that Population III stars are typically very

441: massive. The fragmentation in the \zsix~simulation is slightly more efficient than in

442: the primordial case, with 33 objects forming. Again we stress that there is a delay of

443: several local free-fall times between the formation of first and second protostars in

444: these simulations: {\em without the inclusion of sink particles, we would have missed

445: the formation of the lower mass objects.}

446: 

447: 

448: \section{Conditions for Fragmentation and Cluster Formation}

449: \label{discussion}

450: 

451: We now discuss the physical origin of the fragmentation in the \zfrag~simulation, and

452: investigate the properties of the forming cluster. 

453: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

454: %RSK: this is new to include the angular momentum and density figure...

455: First, we focus on the distribution of gas in the center of the halo right at the onset of star formation, i.e.\ at the time when we identify the first sink particle. Figure \ref{fig:ang-mom} shows $(a)$ the distribution of rotational speed $L/r$ relative to the Keplerian velocity $v_{\rm kep} = (GM/r)^{1/2}$ and $(b)$ the specific angular momentum $L$ in spherical mass shells around  the halo center as a function of the enclosed mass. The blue curves denote the  $10^{-5}\,$Z$_{\odot}$ case, while the red-dashed curves illustrate the behavior of zero metallicity gas, as discussed further below.   It is evident that at the time when the first sink particle forms, rotational support plays  only a minor role.  The rotational velocity of gas in the inner parts of the halo is still sub-Keplerian by a factor or 4 to 5. This will change rapidly in the subsequent evolution as more and more higher angular momentum gas falls to the center.  In $(c)$ we plot the enclosed mass and in $(d)$ the spherically averaged density as a function of distance from the center. The density roughly follows a power-law distribution with slope $-2\pm0.1$.

456: 

457: \begin{figure}[htbp]

458: \begin{center}

459: \includegraphics[width=8.5cm]{f6.eps}

460: \end{center}

461: 

462: \caption{\label{fig:ang-mom} $(a)$ Radially-averaged rotational velocity $L/r$ normalized

463: to the Keplerian velocity $v_{\rm kep} = (GM_{\rm enc}(r)/r)^{1/2}$ and $(b)$ absolute

464: value of the specific angular momentum $L$ in spherical shells centered on the density

465: peak as function of enclosed mass; $(c)$ $M_{\rm enc}(r)$ and $(d)$ average local gas

466: density $n_{\rm gas}(r)$ as function of distance $r$ to the center. Blue (solid line) 

467: curves denote gas

468: with  Z$=10^{-5}$Z$_{\odot}$ and red (dashed line) curves denote gas with Z$=0$. To guide your eye, a

469: power-law slope of $-2$ is indicated in $(d)$.} 

470: 

471: \end{figure}

472: 

473: We point out that the properties of the star forming clump  in our model  are virtually identical to those arising from full cosmological calculations taking into account the combined evolution of baryons and dark matter over time \cite[for comparison, see, e.g.][]{abn02,yoha06}. If anything, one could argue that the specific angular momentum we consider is somewhat on the low side. We therefore expect that fragmentation will  also occur in full cosmological simulations, if these are continued beyond the formation of the very first stellar object \cite[see e.g.\ ][for some preliminary evidence of this]{yokh07}.

474: %RSK: end

475: %%%%%%%%%%%%%%%%%%%%%%%%%%

476: 

477: To illustrate how the conditions

478: for cluster formation arise we show in Figure \ref{fig:density} the density

479: distribution perpendicular to the rotational axis of the system at different times.  In

480: Figure \ref{fig:massrho}, in the top panel, we plot both the mass of gas which resides

481: above a density $n$ as well as the Jeans mass. In the bottom panel, we plot

482: the number of Jeans masses, which is simply the mass at this density (or higher)

483: divided by the corresponding Jean mass.

484: 

485: 

486: \begin{figure*}[t]

487: \begin{center}

488: \unitlength1cm

489: \begin{picture}(20,5.5)

490: \put(1.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f7a}}

491: \put(7.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f7b}}

492: \put(13.0, 0.50){\includegraphics[width=1.8in,height=1.8in]{f7c}}

493: \end{picture}

494: \end{center}

495: 

496: \caption{\label{fig:lowresxrho} Particle densities as a function of position in the

497: low-resolution simulations, for the primordial (left), \zsix~(middle) and

498: \zfrag~simulations (right). The particles are plotted once the protostars in each

499: simulation have accreted 19 \solmas of gas.}

500: 

501: \end{figure*}

502: 

503: The fragmentation in the low-metallicity gas (the \zfrag~case) is the result of two 

504: key features in its thermal evolution. First, the rise in the EOS curve between

505: densities $10^{9}$cm$^{-3}$ and $10^{11}$cm$^{-3}$ causes material to loiter at this

506: point in the gravitational contraction. A similar behavior at densities around $n =

507: 10^3\,$cm$^{-3}$ is discussed by Bromm et al.\ (2001).

508: The rotationally stabilized disk-like structure, as seen in the plateau at $n \approx

509: 10^{10}$cm$^{-3}$ in Figure \ref{fig:density}, is able to accumulate a significant

510: amount of mass in this phase and only slowly increases in density. Second, once the

511: density exceeds $n \approx 10^{12}$cm$^{-3}$, the sudden drop in the EOS curve lowers

512: the critical mass for gravitational collapse by two orders of magnitude. The Jeans mass

513: in the gas at this stage is only $M_{\rm J} = 0.01\ $M$_{\odot}$, as visible in the top

514: panel of Figure \ref{fig:massrho}. The disk-like structure suddenly becomes highly

515: unstable against gravitational collapse. We see this when looking at the behavior of

516: the number of Jeans masses $N_{J}$  above a certain density $n$ in the bottom panel of

517: Figure \ref{fig:massrho}, which shortly after the formation of the first stellar object

518: increases to values as high as $N_{\rm J} \approx 10^3$. Consequently, the disk

519: fragments vigorously on timescales of several hundred years. A very dense cluster of

520: embedded low-mass protostars builds up, and the protostars grow in mass by accretion

521: from the available gas reservoir. The number of protostars formed by the end of our

522: simulation (177) is nearly two orders of magnitude larger than the initial number of

523: Jeans masses in the cloud set-up.

524: 

525: Because the evolutionary timescale of the system is extremely short -- the free-fall

526: time at a density  of $n = 10^{13}\,$cm$^{-3}$ is of the order of 10 years  --   none

527: of the protostars that have formed  by the time that we stop the simulation have yet

528: commenced hydrogen burning, justifying our decision to neglect the effects of

529: protostellar feedback in this study. Heating of the dust due to the significant

530: accretion luminosities of the newly-formed protostars will occur \citep{krum06}, but is

531: unlikely to be important, as the temperature of the dust at the onset of dust-induced

532: cooling is much higher than in a typical Galactic protostellar core ($T_{\rm dust} \sim

533: 100 \: {\rm K}$ or more, compared to $\sim 10 \: {\rm K}$ in the Galactic case). The

534: rapid collapse and fragmentation of the gas also leaves no time for dynamo

535: amplification of magnetic fields \citep{tb04},  which in any case are expected to be

536: weak and dynamically unimportant in primordial and very low metallicity gas

537: \citep{wid02}.

538: 

539: The cluster forming in our  \zfrag~simulation  represents a very extreme analogue of

540: the clustered star formation that we know dominates in the present-day Universe

541: \cite[][]{ll03}. A mere 420 years after the formation of the first object, the

542: cluster has formed 177 stars (see Figure \ref{fig:sequence}). These occupy a region

543: of only around 400~AU, or $2 \times 10^{-3}$~pc, in size, roughly a hundredth of the

544: size of the initial cloud. With $\sim 19 \,$M$_{\odot}$  accreted at this stage, the

545: stellar density is $2.25 \times 10^{9}$ \solmas pc$^{-3}$. This is about five orders

546: of magnitude greater than the stellar density in the Trapezium cluster in Orion

547: \cite[][]{hh98} and about a thousand times greater than that in the core of 30

548: Doradus in the Large Magellanic Cloud \cite[][]{mh98}. This means that dynamical encounters 

549: will be extremely important during the formation of the first star cluster. The

550: violent environment causes stars to be thrown out of the denser regions of the cluster,

551: slowing down their accretion. The stellar mass spectrum thus depends on both the

552: details of the initial fragmentation process \cite[e.g.\ as discussed by][]{jappsen05,

553: clark05} as well as dynamical effects in the growing cluster \cite[][]{bcbp2001,bbv04}. 

554: This is different to present-day star formation, where the

555: situation is less clear-cut and the relative importance of these two processes may

556: vary strongly from region to region \cite[][]{kmk05, bb06, blz07}. 

557: 

558: Dynamical encounters in the extremely dense protocluster will also influence the binary

559: fraction of the stars that form. Wide binaries will be rapidly disrupted, and so any

560: binary systems that survive will be tightly-bound close binaries \cite[][]{kroupa98}.

561: Recent observations suggest that extremely metal-poor low-mass stars  have a higher

562: binary fraction than that found for normal metal-poor stars, and that the period

563: distribution of these binary systems is also skewed towards tight, short-period

564: binaries (Lucatello et~al.\ 2005). Mass transfer from a close binary companion may also

565: be able to explain the extremely high [C/Fe] ratios measured in the most metal-poor

566: stars currently known \citep{ry05,kom07}. Our results suggest that these stars may

567: originate in conditions similar to those that we find in our simulations. Further work

568: aimed at improving our understanding of the binary statistics of low-metallicity stars

569: formed by dust-induced fragmentation is clearly required.

570: 

571: As mentioned in Section \ref{results}, the primordial and the \zsix~cases also

572: exhibit some fragmentation. Careful analysis of the \citet{om05} EOS for

573: zero-metallicity gas shows roughly isothermal behavior in the density range

574: $10^{14}\,$cm$^{-3} \le n \le 10^{16}\,$cm$^{-3}$, i.e.\ just before the gas becomes

575: optically thick and begins to heat up adiabatically. Conservation of angular momentum

576: again leads to the build-up of a rotationally supported massive disk-like structure

577: which then fragments into several objects. This is understandable, as isothermal disks

578: are susceptible to gravitational instability \cite[][]{bodenheimer95} once they have

579: accumulated sufficient mass. Further, Goodwin et~al.~(2004a,b) show how even very low

580: levels of turbulence can induce fragmentation. Since turbulence creates

581: local anisotropies in the angular momentum on all scales, it can always provide some 

582: centrifugal support against gravitational collapse. This support can then provide a window in 

583: which fragmentation can occur.

584: 

585: In addition to the quasi-isothermal behavior of the gas, both of the

586: \zleqsix~equations of state from Omukai et~al. (2005) also contain a brief phase of

587: cooling during the  collapse, which further aids fragmentation. In the primordial case,

588: this occurs very late in the collapse, just above $n = 10^{14}$ \numdensu, and is due

589: to the onset of  efficient cooling from ${\rm H_{2}}$ collision-induced emission

590: \citep{on98,ra04}. In the \zsix~EOS, the cooling occurs earlier, at around $n = 10^{10}$

591: \numdensu, and is due to a combination of effects -- enhanced ${\rm H_{2}}$ cooling

592: resulting from the rapid increase in the molecular fraction at these densities due to

593: efficient three-body  ${\rm H_{2}}$ formation, and rotational and vibrational line

594: cooling from ${\rm H_{2}O}$ -- that are present in the model of Omukai et~al.\ (2005). 

595: One can see  the emergence of structure at these densities

596: in the particle plots shown in Figure~\ref{fig:lowresxrho}. However, the effect is much less

597: pronounced in the primordial case. Indeed a further low-resolution simulation of the

598: primordial EOS in which this dip was removed yielded almost identical results, forming

599: 17 sink particles instead of 22, suggesting that the quasi-isothermal nature of the 

600: gas is more important than this brief cooling phase.

601: 

602: In comparison to the \zfrag~case, the strength of fragmentation in the

603: \zleqsix~calculations is weak, and only a few objects form for the combination of total

604: mass ($M=500\,$M$_{\odot}$), angular momentum ($E_{\rm rot}/ |E_{\rm grav}|= 0.02$),

605: and level of initial turbulence ($E_{\rm turb}/|E_{\rm grav}| = 0.1$) that we adopted

606: in our simulations. Consequently the stars (sink particles) forming in the

607: \zleqsix~simulations are typically of higher mass than those in the \zfrag~simulations,

608: consistent with the predictions made by \citet[][]{abn02} and \citet[][]{bcl02}.

609: Although the recent high-resolution SPH simulation of primordial gas performed by

610: \citet{yoha06}, which predict a similar EOS to that used in our primordial simulations,

611: find no fragmentation, they do not follow the evolution of the gas beyond the formation

612: of the first collapsing core, since they do not include  sink particles in their study,

613: and so they may miss this subsequent phase of  fragmentation. However, more

614: recent work \citet{yokh07}, does show evidence of fragmentation on scales of around

615: 0.1pc.  

616: 

617: \section{Caveats} 

618: \label{caveats} 

619: 

620: Although the equations of state taken from Omukai et~al.\ (2005) provide an opportunity

621: for  vigorous fragmentation at low metallicities, there are a number of questions which

622: still need to be addressed. One immediate uncertainty is the applicability of the

623: results derived from a 1-zone model to a full three-dimensional collapse calculation,

624: which contains  turbulence and thus local anisotropies in velocity and density

625: structure. In particular,  the existence of a strict relationship between temperature

626: and density is unlikely, even when the cooling time of the gas is short compared to the

627: dynamical time (Whitehouse \&  Bate 2006). If the gas were to  collapse more slowly

628: than is assumed in the Omukai et~al.\  (2005) calculations, or if the opacity of the

629: gas were to be lower, then less heating would occur prior to the onset of efficient

630: dust cooling. 

631: 

632: On the other hand, if the  collapse is faster than Omukai et~al.\ (2005) assume or if

633: the gas opacity is greater, then more heating would occur. Since the amount of heating

634: that occurs during the loitering phase helps to accumulate material at high densities,

635: which then acts as a reservoir for fragmentation, the amount of fragmentation will be

636: sensitive to the thermal evolution of the gas during this phase of the collapse, and is

637: therefore somewhat uncertain. However, we note that unless the temperature dip is

638: eliminated entirely (which seems unlikely), fragmentation will still occur along the

639: lines outlined in this paper. We therefore believe our results to be qualitatively, if

640: not quantitatively, correct.

641: 

642: A related issue is one of dust opacity. In their study, Omukai et~al.\ (2005) use a

643: dust model based on Pollack et~al.\ (1994), which was designed for the study of

644: Galactic dust. However, it is not at all clear that this is the appropriate model  to

645: use, given that high-redshift dust is likely to differ significantly from local dust

646: \citep[see e.g.][]{tod01}. Schneider et~al.\ (2006) have performed a similar  study to

647: that of Omukai et~al.\ (2005), but use a dust model based on the results of

648: \citet{tod01} and \citet{sfs04} in place of the Pollack et~al.\ (1994) dust model. 

649: They find that this leads to qualitatively similar behavior to the Omukai et~al.

650: (2005)  model, but that the onset of effective dust cooling happens at lower

651: metallicity,  \zsix~rather than \zfrag. However, the use of this alternative dust model

652: can also be  questioned, since \citet{noz03} have shown that the composition and

653: properties of the  dust produced by population III supernovae are strongly dependent on

654: the (poorly-constrained)  degree of mixing assumed to occur in the ejecta, and since

655: the model does not take into  account the destruction of grains by thermal sputtering

656: in the reverse shock in the supernova remnant \citep{bs07}, or the growth of grains

657: through accretion or coagulation  during the protostellar collapse \citep[see

658: e.g.][]{fpw05}.

659: 

660: Lastly, the total amount of heating produced by three body H$_{2}$ formation is

661: uncertain. This reaction, which takes place primarily at gas number densities  between

662: $10^{8}$ and $10^{13} \: {\rm cm^{-3}}$, is responsible for much of the  rise in gas

663: temperature prior to the dust-cooling phase. However, there is a  difference of two

664: orders of magnitude between the three-body ${\rm H_{2}}$ formation rate used in the

665: simulations of primordial star-formation performed by  Abel et~al.~(2002) and the most

666: recent theoretical determination by Flower \&  Harris (2007), with other suggested

667: rates spanning the full range in between.  This introduces an additional uncertainty

668: into the thermal evolution of the gas during the high-density loitering phase.

669: 

670: \section{Conclusions}

671: \label{conclusions}

672: 

673: We have performed numerical simulations of star formation in very high density gas (in

674: the range $5 \times 10^{5} \leq n \leq  10^{16} \: {\rm cm^{-3}}$) in the early

675: universe. The aim of the study was to investigate whether the dust-induced

676: fragmentation predicted by Omukai et~al.\ (2005) and Schneider et~al.\ (2002, 2006) 

677: does actually occur in realistic systems, and to begin to constrain the resulting IMF. 

678: The major differences of our work from the only previous numerical study of

679: dust-induced  fragmentation (Tsuribe \& Omukai 2006) are  the inclusion of the effects

680: of rotation,  which prove to be of vital importance, and the use of sink particles to

681: capture the  formation of multiple protostellar objects.

682: 

683: Based on the equations of state reported by Omukai et~al.\ (2005), our results show

684: that fragmentation of protogalactic gas at very high densities above $n_{\rm gas} =

685: 10^{12}\,$cm$^{-3}$ is almost unavoidable, as long as the angular momentum is

686: non-negligible. In this case, rotation leads to the build-up of a massive disk-like

687: structure which provides the background for smaller-scale density fluctuations to grow,

688: some of which become gravitationally unstable and collapse to form stars. At

689: metallicities above Z~$\sim 10^{-5}\,$Z$_{\odot}$ dust cooling becomes effective at

690: densities $n_{\rm gas} \sim 10^{12}\,$cm$^{-3}$ and leads to a sudden drop of

691: temperature which in turn induces vigorous fragmentation. A very dense cluster of

692: low-mass protostars builds up, which we refer to as the first stellar cluster. The mass

693: spectrum peaks below $1\,$M$_{\odot}$, which is similar to the value in the  solar

694: neighborhood \cite[][]{kroupa02,chabrier03} and is also comparable to the mass of the

695: very low metallicity subgiant stars recently discovered in the halo of our Milky Way

696: \citep{christlieb02, bc05}. If the dust induced cooling model proposed by Omukai et al

697: (2005) is accurate, then the high-density, low-metallicity, fragmentation we describe

698: here may be the dominant process which shapes the stellar mass function.

699: 

700: 

701: We find that even  purely primordial Z~$ = 0$~gas with sufficient rotation may fragment

702: at densities $10^{14}\,$cm$^{-3} \le n \le 10^{16}\,$cm$^{-3}$. In this density range,

703: zero metallicity gas is roughly isothermal and the disk-like structure that forms due

704: to angular momentum conservation is marginally unstable. It  fragments into several,

705: quite massive objects, thus supporting the hypothesis that metal-free stars should have

706: masses in excess of several tens of M$_{\odot}$. Similar behavior is found for gas with

707: a metallicity of \zsix.

708: 

709: \begin{acknowledgments}

710: 

711: The authors would like to thank Kazuyuki Omukai, Tom Abel, Volker Bromm, Robi

712: Banerjee and Mordecai-Mark Mac Low for  helpful discussions concerning this topic.

713: In particular, we would also like to thank Naoki Yoshida for many interesting

714: and insightful discussions regarding this paper. All computations described here

715: were performed  at the J{\"u}lich Multi-processor (JUMP) supercomputer at the John

716: von Neumann Institute for Computing, Research Centre J{\"u}lich, Germany. PCC

717: acknowledges  support by the Deutsche Forschungsgemeinschaft (DFG) under grant

718: KL~1358/5.   SCOG acknowledges travel support from the European Commission FP6 Marie

719: Curie  RTN CONSTELLATION (MRTN-CT-2006-035890).

720: 

721: \end{acknowledgments}

722: 

723: \begin{thebibliography}{apj}

724: 

725: 

726: \bibitem[Abel, Bryan, \& Norman(2002)]{abn02}

727: Abel, T., Bryan, G.~L., \& Norman, M.~L. 2002, Science, 295, 93

728: 

729: \bibitem[Bate, Bonnell, \& Price(1995)]{bbp95}

730: Bate, M.~R., Bonnell, I.~A., \& Price, N.~P. 1995, MNRAS, 277, 362

731: 

732: \bibitem[Bate \& Burkert(1997)]{bb97}

733: Bate, M.~R., \& Burkert, A. 1997, MNRAS, 288, 1060

734: 

735: \bibitem[Beers \& Christlieb(2005)]{bc05}

736: Beers, T.~C., \& Christlieb, N. 2005, ARA\&A, 43, 531

737: 

738: \bibitem[Bianchi \& Schneider(2007)]{bs07}

739: Bianchi, S., \& Schneider, R. 2007, MNRAS, 378, 973

740: 

741: \bibitem[Bodenheimer(1995)]{bodenheimer95} 

742: Bodenheimer, P.\ 1995, ARA\&A, 33, 199 

743: 

744: \bibitem[Bonnell et al.(2001)]{bcbp2001} Bonnell, I.~A., Clarke, 

745: C.~J., Bate, M.~R., \& Pringle, J.~E.\ 2001, MNRAS, 324, 573 

746: 

747: \bibitem[Bonnell, Bate \& Vine (2004)]{bbv04} Bonnell, I.~A., Vine, 

748: S.~G., \& Bate, M.~R.\ 2004, MNRAS, 349, 735 

749: 

750: \bibitem[Bonnell \& Bate(2006)]{bb06}

751: Bonnell, I.~A., \& Bate, M.~R. 2006, MNRAS, 370, 488

752: 

753: \bibitem[Bonnell, Larson \& Zinnecker (2007)]{blz07} Bonnell, I.~A., Larson, 

754: R.~B., \& Zinnecker, H.\ 2007, Protostars and Planets V,

755: B. Reipurth, D. Jewitt, and K. Keil (eds.), University of Arizona Press, Tucson, p.149 

756: 

757: \bibitem[Bromm et~al.(2001)]{bfcl01}

758: Bromm, V., Ferrara, A., Coppi, P.~S., \& Larson, R.~B. 2001, MNRAS, 328, 969

759: 

760: \bibitem[Bromm, Coppi, \& Larson(2002)]{bcl02}

761: Bromm, V., Coppi, P.~S., \& Larson, R.~B. 2002, ApJ, 564, 23

762: 

763: \bibitem[Bromm \& Loeb(2003)]{bl03}

764: Bromm, V. \& Loeb, A. 2003, Nature, 425, 812

765: 

766: \bibitem[Bromm \& Loeb(2004)]{bl04}

767: Bromm, V., \& Loeb, A. 2004, New Astron., 9, 353

768: 

769: \bibitem[Chabrier(2003)]{chabrier03}

770: Chabrier, G. 2003, PASP, 115, 763 

771: 

772: \bibitem[Christlieb et~al.(2002)]{christlieb02} 

773: Christlieb, N., Bessell, M. S., Beers, T. C., Gustafsson, B., Korn, A., Barklem, P. S., Karlsson, T., 

774: Mizuno-Wiedner, M., \& Rossi, S.  2002, Nature, 419, 904 

775: 

776: \bibitem[Clark \& Bonnell(2005)]{clark05} Clark, P.~C., \& 

777: Bonnell, I.~A.\ 2005, MNRAS, 361, 2 

778: 

779: \bibitem[Flower \& Harris(2007)]{fh07}

780: Flower, D.~R., \& Harris, G.~J. 2007, MNRAS, 377, 705

781: 

782: \bibitem[Flower et~al. (2005)]{fpw05}

783: Flower, D.~R., {Pineau des For\^ets}, G., \& Walmsley, C.~M. 2005, A\&A, 436, 933

784: 

785: \bibitem[Frebel, Johnson, \& Bromm(2007)]{fjb07}

786: Frebel, A., Johnson, J.~L., \& Bromm, V. 2007, MNRAS, 380, L40

787: 

788: \bibitem[Goodwin, Whitworth, \& Ward-Thompson(2004a)]{gw04a}

789: Goodwin, S.~P., Whitworth, A.~P., Ward-Thompson, D. 2004a, A\&A, 414, 633

790: 

791: \bibitem[Goodwin, Whitworth, \& Ward-Thompson(2004b)]{gw04b}

792: Goodwin, S.~P., Whitworth, A.~P., Ward-Thompson, D. 2004b, A\&A, 423, 169

793: 

794: \bibitem[Hillenbrand \& Hartmann(1998)]{hh98} 

795: Hillenbrand, L.~A., \& Hartmann, L.~W.\ 1998, ApJ, 492, 540 

796: 

797: \bibitem[Jappsen et~al.(2005)]{jappsen05} 

798: Jappsen, A.-K., Klessen, R.~S., Larson, R.~B., Li, Y., \& Mac Low, M.-M.\ 2005, A\&A, 435, 

799: 611 

800: 

801: \bibitem[Lucatello et~al.(2005)]{luc05}

802: Lucatello, S., Tsangarides, S., Beers, T.~C., Carretta, E.,

803: Gratton, R.~G., \& Ryan, S.~G. 2005, ApJ, 625, 825

804: 

805: \bibitem[Komiya et~al.(2007)]{kom07}

806: Komiya, Y., Suda, T., Minaguchi, H., Shigeyama, T., 

807: Aoki, W., \& Fujimoto, M.~Y. 2007, ApJ, 658, 367

808: 

809: \bibitem[Kroupa(1998)]{kroupa98} Kroupa, P.\ 1998, MNRAS, 298, 

810: 231

811: 

812: \bibitem[Kroupa(2002)]{kroupa02} Kroupa, P.\ 2002, Science, 295, 

813: 82 

814: 

815: \bibitem[Krumholz, McKee, \& Klein(2005)]{kmk05}

816: Krumholz, M.~R., McKee, C.~F., \& Klein, R.~I. 2005, Nature, 438, 332

817: 

818: \bibitem[Krumholz(2006)]{krum06}

819: Krumholz, M.~R. 2006, ApJ, 641, L45

820: 

821: \bibitem[Lada \& Lada(2003)]{ll03} 

822: Lada, C.~J., \& Lada, E.~A.\ 2003, ARA\&A, 41, 57 

823: 

824: \bibitem[Massey \& Hunter(1998)]{mh98} 

825: Massey, P., \& Hunter, D.~A.\ 1998, ApJ, 493, 180

826: 

827: \bibitem[Nozawa et~al.(2003)]{noz03}

828: Nozawa, T., Kozasa, T., Umeda, H., Maeda, K., \& Nomoto, K. 2003, ApJ, 598, 785

829: 

830: \bibitem[Omukai \& Nishi(1998)]{on98}

831: Omukai, K., \& Nishi, R. 1998, ApJ, 508, 141

832: 

833: \bibitem[Omukai \& Palla(2003)]{op03}

834: Omukai, K., \& Palla, F. 2003, ApJ, 589, 677

835: 

836: \bibitem[Omukai et~al.(2005)]{om05}

837: Omukai, K., Tsuribe, T., Schneider, R., \& Ferrara, A. 2005, ApJ, 626, 627

838: 

839: \bibitem[O'Shea \& Norman(2007)]{oshn07}

840: O'Shea, B.~W., \& Norman, M.~L. 2007, ApJ, 654, 66

841: 

842: \bibitem[Pollack et~al.(1994)]{pol94}

843: Pollack, J.~B., Hollenbach, D., Beckwith, S., Simonelli, D.~P., 

844: Roush, T., \& Fong, W. 1994, ApJ, 421, 615

845: 

846: \bibitem[Ripamonti \& Abel(2004)]{ra04}

847: Ripamonti, E., \& Abel, T. 2004, MNRAS, 348, 1019

848: 

849: \bibitem[Ryan et~al.(2005)]{ry05}

850: Ryan, S.~G., Aoki, W., Norris, J.~E., \& Beers, T.~C. 2005, ApJ, 635, 349

851: 

852: \bibitem[Santoro \& Shull(2006)]{san06}

853: Santoro, F. \& Shull, J.~M. 2006, ApJ, 643, 26

854: 

855: \bibitem[Schneider et~al.(2002)]{sch02}

856: Schneider, R., Ferrara, A., Natarajan, P. \& Omukai, K. 2002, ApJ, 571, 30

857: 

858: \bibitem[Schneider, Ferrara, \& Salvaterra(2004)]{sfs04}

859: Schneider, R., Ferrara, A., \& Salvaterra, R. 2004, MNRAS, 351, 1379

860: 

861: \bibitem[Schneider et~al.(2006)]{sch06}

862: Schneider, R., Omukai, K., Inoue, A.~K., \& Ferrara, A. 2006, MNRAS, 369, 1437

863: 

864: \bibitem[Tan \& Blackman(2004)]{tb04}

865: Tan, J.~C., \& Blackman, E.~G. 2004, ApJ, 603, 401

866: 

867: \bibitem[Todini \& Ferrara(2001)]{tod01}

868: Todini, P., \& Ferrara, A. 2001, MNRAS, 325, 726

869: 

870: \bibitem[Tumlinson(2006)]{tum06}

871: Tumlinson, J. 2006, ApJ, 641, 1

872: 

873: \bibitem[Tsuribe \& Omukai(2006)]{to06}

874: Tsuribe, T., \& Omukai, K. 2006, ApJ, 642, L61

875: 

876: \bibitem[Whitehouse \& Bate(2006)]{wb06}

877: Whitehouse, S.~C., \& Bate, M.~R. 2006, MNRAS, 367, 32

878: 

879: \bibitem[Widrow(2002)]{wid02}

880: Widrow, L.~M. 2002, Rev.\ Mod.\ Phys., 74, 775

881: 

882: \bibitem[Wuchterl \& Klessen(2001)]{WK01} 

883: Wuchterl, G., \& Klessen, R.~S.\ 2001, ApJ, 560, L185 

884: 

885: \bibitem[Yoshida {\em et~al.}(2006)]{yoha06}

886: Yoshida, N., Omukai, K., Hernquist, L., \& Abel, T. 2006, ApJ, 652, 6

887: 

888: \bibitem[Yoshida et al.(2007)]{yokh07} 

889: Yoshida, N., Oh, S.~P., Kitayama, T., \& Hernquist, L.\ 2007, ApJ, 663, 687

890: 

891: \end{thebibliography}

892: 

893: \end{document}

894: