1: \documentclass{emulateapj}

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

3: \usepackage{epsfig}

4: \usepackage{apjfonts}

5: \usepackage{graphics}

6: 

7: \newcommand{\ssim}{{\cal S}}

8: \newcommand{\tsim}{{\cal T}}

9: \newcommand{\HI}{\ion{H}{1}} 

10: \newcommand{\HeII}{\ion{He}{2}} 

11: \newcommand{\FeII}{\ion{Fe}{2}} 

12: \newcommand{\CIII}{\ion{C}{3}} 

13: \newcommand{\CII}{\ion{C}{2}} 

14: \newcommand{\CIV}{\ion{C}{4}} 

15: \newcommand{\SiIII}{\ion{Si}{3}} 

16: \newcommand{\SiII}{\ion{Si}{2}} 

17: \newcommand{\SiIV}{\ion{Si}{4}} 

18: \newcommand{\NV}{\ion{N}{5}} 

19: \newcommand{\OVI}{\ion{O}{6}} 

20: 

21: \newcommand{\tsiiv}{\tau_{\rm SiIV}}

22: \newcommand{\tciv}{\tau_{\rm CIV}}

23: \newcommand{\thi}{\tau_{\rm HI}}

24: \newcommand{\tsiiii}{\tau_{\rm SiIII}}

25: 

26: 

27: \newcommand{\msun}{{\rm M}_\odot}

28: 

29: \newcommand{\lya}{Ly$\alpha$} 

30: \newcommand{\lyb}{Ly$\beta$}

31: 

32: \lefthead{Aguirre et al.}

33: \righthead{Simulations and observations of the IGM}

34: 

35: \def\gsim{\;\rlap{\lower 2.5pt

36:  \hbox{$\sim$}}\raise 1.5pt\hbox{$>$}\;}

37: \def\lsim{\;\rlap{\lower 2.5pt

38:    \hbox{$\sim$}}\raise 1.5pt\hbox{$<$}\;}

39: \def\msol{{\rm\,M_\odot}}

40: \def\yr{{\rm\,yr}}

41: \def\kpc{{\rm\,kpc}}

42: \def\kms{\rm\,km\,s^{-1}}

43: \def\mpc{{\rm\,Mpc}}

44: \def\au{{\rm\,AU}}

45: \def\del{{\partial}}

46: \def\gm{{\rm\,g}}

47: \def\cm{{\rm\,cm}}

48: \def\sec{{\rm\,s}}

49: \def\erg{{\rm\,erg}}

50: \def\lya{{{\rm Ly}\alpha}}

51: \def\kev{{\rm\,keV}}

52: \def\mic{{\,\mu{\rm m}}}

53: \def\ev{{\rm\,eV}}

54: \def\kms{{\rm\,km\,s^{-1}}}

55: \def\K{{\rm\,K}}

56: \def\spose#1{\hbox to 0pt{#1\hss}}

57: %\def\lta{\mathrel{\spose{\lower 3pt\hbox{$\mathchar''218$}}

58: %     \raise 2.0pt\hbox{$\mathchar''13C$}}}

59: %\def\gta{\mathrel{\spose{\lower 3pt\hbox{$\mathchar''218$}}

60: %     \raise 2.0pt\hbox{$\mathchar''13E$}}}

61: \def\iun{{\rm\,nW\,m^{-2}\,sr^{-1}}}

62: \def\zsol{{\,Z_\odot}}

63: \def\tc{{t_{\rm c}}}

64: \def\tp{{t_{\rm p}}}

65: 

66: \shorttitle{Simulations vs. observed intergalactic metals}

67: \shortauthors{Aguirre et al.}

68: 

69: \begin{document}

70: 	

71: \title{Confronting cosmological simulations with observations of

72:   intergalactic metals}

73: \author{Anthony Aguirre\altaffilmark{1}, Joop~Schaye\altaffilmark{2}, \\

74: Lars Hernquist\altaffilmark{3}, Scott Kay\altaffilmark{4,5}, Volker Springel\altaffilmark{6}, Tom Theuns\altaffilmark{7,8}}

75: \altaffiltext{1}{Department of Physics, University of California at Santa Cruz,1156 High Street, Santa Cruz, CA  95064; aguirre@scipp.ucsc.edu}

76: \altaffiltext{2}{School of Natural Sciences, Institute for Advanced

77: Study, Einstein Drive, Princeton NJ 08540}

78: \altaffiltext{3}{Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge MA 02138}

79: \altaffiltext{4}{Astronomy Centre, University of Sussex, Falmer, Brighton BN1 9QH, UK}

80: 

81: \altaffiltext{5}{Astrophysics, Denys Wilkinson Building, Oxford OX1 3RH, UK}

82: 

83: \altaffiltext{6}{Max Planck Institute for Astrophysics, Karl-Schwartzschild Strasse 1, Garching, Munich, D-85740, Germany}

84: \altaffiltext{7}{Institute for Computational Cosmology, Department of

85: Physics, University of Durham, South Road, Durham, DH1 3LE, UK}

86: \altaffiltext{8}{University of Antwerp, Campus Drie Eiken,

87: Universiteitsplein 1, B-2610 Antwerp, Belgium}

88: 

89: \setcounter{footnote}{0}

90: 

91: \begin{abstract}

92: Using the statistics of pixel optical depths, we compare \HI, \CIV\

93: and \CIII\ absorption in a set of six high quality $z\sim 3-4$ quasar

94: absorption spectra to that in spectra drawn from two different

95: state-of-the-art cosmological simulations that include galactic

96: outflows. We find that the simulations predict far too little \CIV\

97: absorption unless the UVB is extremely soft, and always predict far

98: too small \CIII/\CIV\ ratios.  We note, however, that much of the

99: enriched gas is in a phase ($T\sim 10^5-10^7\,$K,

100: $\rho/\langle\rho\rangle\sim 0.1-10$, $Z\gtrsim 0.1 Z_\odot$) that

101: should cool by metal line emission -- which was not included in our

102: simulations. When the effect of cooling is modeled, the predicted

103: \CIV\ absorption increases substantially, but the \CIII/\CIV\ ratios

104: are still far too small because the density of the enriched gas is too

105: low. Finally, we find that the predicted metal distribution is much

106: too inhomogeneous to reproduce the observed probability distribution

107: of \CIV\ absorption.  These findings suggest that strong $z\lesssim 6$

108: winds cannot fully explain the observed enrichment, and that an

109: additional (perhaps higher-$z$) contribution is required.

110: \end{abstract}

111: \keywords{intergalactic medium --- quasars: absorption lines

112: %cosmology: miscellaneous --- 

113:  --- galaxies: formation }

114: 

115: \section{Introduction}

116: \label{sec-intro}

117: 

118: Analysis of quasar absorption spectra has revealed that the

119: intergalactic medium (IGM) has been polluted with heavy elements such

120: as carbon, silicon and oxygen (see, e.g., Cowie et al.\ 1995; Songaila

121: \& Cowie 1996; Ellison et al.\ 2000; Schaye et al. 2003, hereafter

122: S03; Aguirre et al. 2004, hereafter A04; Simcoe et al 2004; Aracil et

123: al. 2004; Boksenberg et al. 2003) At the same time, observations of

124: starburst galaxies (e.g., Shapley et al.\ 2003) have revealed powerful

125: galactic outflows resulting from feedback processes in galaxies with

126: rapid star formation.

127: 

128: This has led to a picture in which the observed enrichment results

129: from a phase of strong galactic outflows at $z \gtrsim 2$ during the

130: epoch of galaxy formation, and various numerical (e.g., Gnedin 1998;

131: Aguirre et al. 2001; Scannapieco, Ferrara, \& Madau 2002; Cen,

132: Nagamine \& Ostriker 2004) and semi-analytic (e.g., Furlanetto \& Loeb

133: 2003) models have been able to very roughly account for the observed

134: level of metal enrichment.  But these comparisons have left many

135: unanswered questions: can, for example, simulations reproduce the

136: detailed density- and redshift-dependent metal distribution? Can they

137: reproduce the abundance ratios of different elements and ions? How is

138: enrichment tied to feedback in galaxies, which is required to suppress

139: runaway star formation?

140: 

141: Recently, both the numerical simulations and observational analyses

142: have improved to the degree that some of these questions can be

143: meaningfully addressed. On the observational side, statistical

144: analyses of pixel optical depths (S03; A04) have inferred the

145: distribution of carbon and silicon using absorption by \HI, \CIV,

146: \CIII, \SiIV, and \SiIII\ (see also Simcoe et al. 2004, who finds

147: consistent results from line fitting of \CIV\ and \OVI.)  These

148: studies have shown quantitatively that the observed intergalactic

149: carbon enrichment is highly inhomogeneous, density-dependent, nearly

150: redshift-independent, underabundant (relative to silicon), relatively

151: cool ($T < 10^5\,$K), and persistent at some level even in gas near

152: the cosmic mean density.  Meanwhile, hydrodynamic simulations (e.g.,

153: Theuns et al. 2002, hereafter T02; Springel \& Hernquist 2003a,

154: hereafter SH03) have been produced that probably include all of the

155: galaxies relevant for $z \lesssim 6$ enrichment, use prescriptions for

156: feedback that generate galactic winds, and track metals. SH03 and

157: Hernquist \& Springel (2003) have shown in detail that the star

158: formation rate of their simulations has converged, i.e. that the

159: included feedback is sufficient to solve the overcooling problem; T02

160: have shown that their simulation does {\em not} significantly disrupt

161: the Ly$\alpha$ forest, but can roughly account for earlier

162: observations of \CIV\ absorption if the UVB is extremely soft and the

163: yield is $3\times$ solar.

164: 

165: Clearly, it is of interest to carry out detailed comparisons between

166: state-of-the art observations and simulations that include galactic

167: winds.  One way to do this is to simply compare the observationally

168: inferred distribution of metals as a function of density and redshift

169: to that predicted by the simulations.  This could, however, be

170: misleading. For example, hot, collisionally ionized carbon would be

171: undetectable by observational studies focusing on \CIV. A far more

172: direct and robust method is to directly compare simulated and observed

173: absorption spectra. This Letter describes such a comparison.  We will

174: draw several qualitatively new conclusions from a comparison of

175: absorption by \HI, \CIV, and \CIII\ in in a set of 6 $z\sim 3$ QSO

176: spectra to that in simulated spectra drawn from the SH03 and T02

177: simulations.

178: 

179: \section{Observations, simulations, and method}

180: \label{sec-simobs}

181: 

182: We compare our simulations to the observed pixel statistics published

183: in S03 for the redshift range $2.479 \le z \le 4.033$ (the full sample

184: covers $1.654 \le z \le 4.451$; we employ a smaller range so that all

185: of the data can be combined without binning in redshift). The data

186: come from six quasar spectra: Q0420-388, Q1425+604, Q2126-158,

187: Q1422+230, Q0055-269, and Q1055+461, that were taken with either the

188: Keck/HIRES or the VLT/UVES instrument. See Table 1 of S03 for

189: information on the sample.

190: 

191: The simulated spectra are drawn from cosmological SPH simulations,

192: using the method described in A02: noise, detector resolution,

193: wavelength coverage, and pixelization are chosen to match the

194: corresponding observed spectra. The ionization balance is computed

195: using CLOUDY\footnote{See

196: \texttt{http://www.pa.uky.edu/$\sim$gary/cloudy}.}, with the same

197: three models for the spectral shape of the UV background (UVB) as we

198: used in S03 and A04: `QG' is a Haardt \& Madau (2001) model with

199: contributions from quasars and galaxies; `Q' includes quasars only,

200: and `QGS' is a softened `QG': the flux is reduced by 90\% above 4 Ryd.

201: All models are normalized to the \HI\ ionization rate measured by S03,

202: and the simulation metallicities are converted to carbon number

203: densities using the solar abundance $({\rm C/H})_\odot=-3.45$ of

204: Anders \& Grevesse (1989).

205: 

206: Results are shown for three simulations.  The first, `NF', was used

207: and described in A02, S03 and A04: it uses $2\times 256^3$ particles

208: in a $12h^{-1}$\,Mpc box with $(\Omega_m, \Omega_\Lambda, \Omega_bh^2,

209: h, \sigma_8, n, Y) = (0.3, 0.7, 0.019, 0.65, 0.9, 1.0, 0.24)$. This

210: simulation has no galactic outflows, but for each UVB model we add to

211: the particles the carbon distribution inferred in S03 for that UVB

212: (see Tables 2 and 3 of S03). The second simulation, `\tsim', is

213: described in T02: it uses $2\times 128^3$ particles in a

214: $5h^{-1}$\,Mpc box with the same cosmological parameters as NF.  Here,

215: however, all supernova thermal energy is deposited as feedback, and

216: gas is prevented from cooling for $10^7\,$yr (see Kay et. al. 2002) so

217: that strong winds are generated which enrich the IGM. The third

218: simulation, `\ssim', is described in SH03 as their `Q4' model: it uses

219: $2\times 216^3$ particles in a $10h^{-1}\,$Mpc box with $(\Omega_m,

220: \Omega_\Lambda, \Omega_bh^2, h, \sigma_8, n, Y) = (0.3, 0.7, 0.02,

221: 0.7, 0.9, 1.0, 0.24)$.  It employs both a sub-grid feedback

222: prescription and a wind mechanism in which a velocity of 484\,km/s is

223: imparted to certain gas particles in star-forming regions (see

224: Springel \& Hernquist 2003b for details).  The simulations have

225: approximately the same mass resolution (the baryon particle mass is

226: $\approx 1.1\times10^6\msol$ in both), which is adequate to resolve

227: the Ly$\alpha$ forest and to include relatively small galaxies. The

228: use of two simulations allows us to compare the effects of two

229: different feedback prescriptions.

230: 

231: To compare the carbon absorption in simulated and observed spectra, we

232: have employed the pixel optical depth technique described in A02, S03

233: and A04. This technique (see Cowie \& Songaila 1998; Ellison et

234: al. 2000; A02; S03) has several advantages over traditional

235: line-fitting. Perhaps most important here is the ability to measure

236: \CIII\ absorption, which is very difficult to do using line fitting

237: because \CIII\ is not a multiplet and falls in the Ly$\beta$

238: forest. Optical depths for \HI\,(Ly$\alpha$), \CIV\,(1548\AA), and

239: \CIII\, (977\AA)\, absorption are extracted from each spectrum for the

240: \HI\ absorption region between the QSO's Ly$\alpha$ and Ly$\beta$

241: emission wavelengths, excluding also a small region near the QSO to

242: avoid proximity effects (see S03).  The optical depths are corrected

243: for strong contaminating lines, \CIV\ self-contamination by its

244: doublet, and Ly$\alpha$ contamination of \CIII, as described in A02

245: and S03.

246: 

247: The pixel optical depths are binned for each QSO and the QSOs are

248: combined as described in A04. In brief, for each QSO the \CIV\ optical

249: depths $\tau_{\rm CIV}$ are binned in \HI\ optical depth $\tau_{\rm

250: HI}$ and the median is taken in each bin.  Next, the

251: noise/contamination level, i.e.\ the median $\tau_{\rm CIV}$ optical

252: depth at very low $\tau_{\rm HI}$, is subtracted from each bin and the

253: ratio $({\rm median}\, \tau_{\rm CIV})/\tau_{\rm HI}$ is computed.

254: Finally, $\tau_{\rm HI}$ bins for different QSOs are combined. The

255: result is a plot of median corrected $\tau_{\rm CIV}/\tau_{\rm HI}$

256: vs. $\tau_{\rm HI}$, as shown in Fig.~\ref{fig-civnocool} (upper

257: left).  The same procedure is applied to obtain plots of $\tau_{\rm

258: CIII}/\tau_{\rm CIV}$ vs. $\tau_{\rm CIV}$.

259: 

260: \section{Results}

261: \label{sec-res}

262: 

263: \begin{figure*}

264: \epsscale{1.1}

265: \plottwo{f1a.ps}{f1b.ps}

266: \plottwo{f1c.ps}{f1d.ps}

267: \figcaption[] {Binned optical depth ratios $({\rm med}\,\tau_{\rm

268: CIV})/\tau_{\rm HI}$ vs.\ $\tau_{\rm HI}$ (\emph{left}) and $({\rm

269: med}\,\tau_{\rm CIII})/\tau_{\rm CIV}$ vs.\ $\tau_{\rm CIV}$

270: (\emph{right}) for observations (data points with 1- and 2-$\sigma$

271: errors) and various models. \emph{Top:} Models NF-QG (the Q and QGS

272: backgrounds yield similar results), \ssim-QG \ssim-Q, \ssim-QGS, and

273: \tsim-QG, as per the legend. The effect of changing the UVB in the

274: \tsim\ simulations is similar. \emph{Bottom:} as in the top panels,

275: but particles with cooling time $t_c < t_H$ are set to $T=2\times

276: 10^4\,$K. The same models are shown, except that the dotted line

277: (`C1') corresponds to \ssim-QG if only particles with $t_c < 0.1 t_H$

278: cool (the effect of this change is similar for the other models).

279: \label{fig-civnocool}}

280: \end{figure*} 

281: 

282: Figure~\ref{fig-civnocool} shows median optical depths $({\rm

283: med}\,\tau_{\rm CIV})/\tau_{\rm HI}$ vs.\ $\tau_{\rm HI}$

284: (\emph{left}) and $({\rm med}\,\tau_{\rm CIII})/\tau_{\rm CIV}$ vs.\

285: $\tau_{\rm CIV}$ (\emph{right}) for observations (data points) and

286: three simulations.  The dotted line shows the non-feedback (NF)

287: simulation upon which the carbon distribution of S03 has been

288: imposed.\footnote{At high-$\tau_{\rm HI}$ the predicted $\tau_{\rm

289: CIV}/\tau_{\rm HI}$ (coming mostly from Q1422+230) is a bit low

290: because S03 forced a power-law fit to $Z(\delta)$ with a

291: redshift-independent index.} The other lines show the SH03 (\ssim) and

292: T02 (\tsim) simulations with the three choices of UVB (QG, Q, and

293: QGS). Except for the extremely soft UVB QGS, the simulations woefully

294: under-predict the median \CIV\ absorption at all $\tau_{\rm HI}$.  In

295: Table 1 (column 4) we quantify this by providing the best-fit offset

296: to each set of simulated \CIV/\HI\ optical depths (e.g., simulation

297: \ssim-QG is $\approx 1.2\,$dex too low). A super-solar yield could

298: somewhat ameliorate this but seems unlikely for carbon, which is

299: underabundant relative to $\alpha$-elements in the IGM (e.g., A04).

300: Although the QGS models are only 0.26-0.64 dex too low overall, they

301: cannot reproduce the observed shape of $\tau_{\rm CIV}/\tau_{\rm HI}$

302: vs.\ $\tau_{\rm HI}$: there is too much absorption at low density (low

303: $\tau_{\rm HI}$) and too little at high-density.

304: 

305: The prime reason for this failure can be seen in Fig.~\ref{fig-zvt},

306: which shows the metallicity, temperature, and density of a random

307: subsample of the \ssim\ simulation particles (the \tsim\ simulation is

308: similar).  The metal rich intergalactic (overdensity $\log\delta < 2$)

309: gas is almost entirely at $T=10^5-10^7\,$K.  Because the \CIV/C

310: fraction rapidly falls off at $T \gtrsim 10^5\,$K, this gas is

311: essentially invisible in \CIV, except at very low density if the UVB

312: is extremely soft.

313: 

314: Even for such an extreme UVB, however, the feedback simulations

315: predict far too little absorption by \CIII\ relative to \CIV, as can

316: be seen from the upper-right panel of Fig.~\ref{fig-civnocool} which

317: shows $({\rm med}\,\tau_{\rm CIII})/\tau_{\rm CIV}$ vs.\ $\tau_{\rm

318: CIV}$ for the same models\footnote{For the Q and QG UVBs, there is

319: insufficient CIV absorption to get a signal; we have thus run models

320: for higher yields (see Table 1).}. The \CIII/\CIV\ ratio drops rapidly

321: for $T \gtrsim 10^5\,$K and falls roughly linearly with decreasing

322: density for $T\sim 10^4\,$K (See Fig. 7 of S03). Thus, the fact that

323: the \CIV\ visible in the feedback simulations is accompanied by

324: insufficient \CIII\ means that the enriched gas is too hot and/or of

325: too low density. Note, on the other hand, that the NF simulation

326: reproduces the \CIII/\CIV\ values quite well once the carbon

327: distribution is chosen to match the \CIV/\HI\ values.

328: 

329: The problem that the metals in the feedback simulations are too hot

330: may, however, have a solution.  In both simulations the enriched gas

331: is relatively metal-rich by IGM standards ($Z \sim 0.1-1\zsol$) and

332: should, in fact, be able to cool via metal line emission, which was

333: {\em not} included in the simulations. The contours in

334: Fig.~\ref{fig-zvt} show $\log(t_{c}/t_H)$, where $t_c$ is the

335: radiative cooling time for gas that is in collisional ionization

336: equilibrium computed for overdensity $\delta\equiv

337: \rho/\langle\rho\rangle=1$ ($t_c \propto \delta^{-1}$) using

338: Sutherland \& Dopita (1993), and $t_H$ is the $z=3$ Hubble

339: time. Nearly all of the $\delta \sim 10$ gas and much of the

340: $\delta\sim 1$ gas has $t_{c} < t_H$ and should, in the absence of

341: heating, cool to $T\sim 10^4\,$K which would increase its visibility

342: in \CIV.  To test the importance of cooling, we have generated

343: simulated spectra for which all particles with $t_c < t_H$ are set to

344: $T=2\times 10^4\,$K; see bottom panels of Fig.~\ref{fig-civnocool}. In

345: this case, the QG simulations can roughly match the observed

346: $\tau_{\rm CIV}/\tau_{\rm HI}$ values, although the trend with

347: $\tau_{\rm HI}$ is still not reproduced. The QGS models with cooling

348: now predict too much \CIV\ absorption.

349: 

350: The employed cooling prescription is rather {\em ad hoc}. The particle

351: metallicities may be unreliable because the simulation assumes perfect

352: mixing at the particle level and zero mixing between particles once

353: they leave the star forming gas. Furthermore, heating is ignored in

354: the calculation of the cooling times, the gas may not be in

355: collisional ionization equilibrium, conduction is ignored, and the

356: density may not remain constant as the gas cools. To further

357: illustrate the sensitivity of the results to the cooling prescription,

358: the dotted curves in the lower panels of Fig.~\ref{fig-civnocool} show

359: the results for the S-QG simulation if the gas cools only when $t_c <

360: 0.1t_H$. The \CIV\ absorption is lower by about 0.5~dex, nearly

361: independent of $\tau_{\rm HI}$. Varying the final temperature between

362: $1-3\times 10^4\,$K has a relatively small effect.

363: 

364: \begin{figure}

365: \epsscale{1.05} \plotone{f2.eps} \figcaption[]{Metallicities (in solar

366: units) and temperatures of particles in the \ssim\ simulation (only a

367: random 0.5\% of metal enriched particles are shown, and 0.05\% of the

368: metal-free particles are shown in the scattered bar near

369: $Z=10^{-4}$). The particle colors reflect the log of the gas

370: overdensity as shown by the colored bar: intergalactic $\delta \alt

371: 100$ gas is blue-green; galactic gas (with effective temperatures

372: given by the sub-grid multiphase model) is yellow-red.  The

373: contours indicate $\log(t_c/t_H)$ at $\delta=1$ and $z=3$.  Since $t_c

374: \propto \delta^{-1}$, particles inside the `0' contour at $\delta=1$

375: can cool, as can particles at $\delta=10$ inside the `1' contour.

376: \label{fig-zvt}}

377: \end{figure} 

378: 

379: Although metal line cooling may help resolve the discrepancy for

380: \CIV/\HI, it actually makes the problem worse for \CIII/\CIV\ as can

381: be seen from the bottom right panel of Fig.~\ref{fig-civnocool}. This

382: indicates that in the simulations the \CIV\ absorption arises in gas

383: with too low a density. The level of $\log \tau_{\rm CIII}/\tau_{\rm

384: CIV}\sim -0.5$ predicted by the simulations indicates that the \CIV\

385: absorption with $\tau_{\rm CIV} \gtrsim 0.1$ arises in $\delta\sim

386: 1-5$ gas, whereas the corresponding observed absorption appears to

387: occur in gas of $\delta\sim 10-30$ (see Fig.\ 7 of S03).  The increase

388: in density that would very likely accompany the gas cooling may

389: alleviate this problem, but this will have to be determined using

390: future simulations that include metal-line cooling.

391: 

392: Information on the homogeneity of the observed metal distribution can

393: be inferred by using additional percentiles of the $\tau_{\rm

394: CIV}(\tau_{\rm HI})$ distribution (S03).  In columns 3--5 of Table~1

395: we give the best-fit offset to the simulations in the 31st, 50th and

396: 84th percentiles of the \CIV\ distribution.  We find that models that

397: can roughly reproduce the medians cannot simultaneously reproduce the

398: other percentiles. In all cases, the metals are {\em too

399: inhomogeneously distributed} as compared to the observations (e.g., in

400: S-QG with cooling, the 31st percentile is too low by $\sim0.8\,$dex

401: while the 84th is too high by $\sim0.6\,$dex, indicating a wider

402: distribution).  This appears to be true independent of the UVB and the

403: cooling prescription and is hence a rather robust inconsistency. It

404: is also interesting because while small-scale unresolved physical

405: effects such as a multiphase structure of the IGM and outflows (or

406: emission from interfaces between phases) are potentially quite

407: important, they seem likely to make the metals appear {\em less}

408: rather than more uniformly distributed.

409: 

410: %\begin{deluxetable*}{llccccll} 

411: \tabletypesize{\scriptsize}

412: \begin{deluxetable}{llcccc}

413: \tablecolumns{10} 

414: \tablewidth{0pc} 

415: \tablecaption{Comparisons made \label{tbl:comp}}

416: \tablehead{ Sim-

417: & & \multicolumn{3}{l}{Offset\tablenotemark{a} for percentile\tablenotemark{b} in CIV/HI} & in CIII/CIV\\

418: %\colhead{Offs.\tablenotemark{c}} & 

419: \colhead{UVB} & \colhead{Cool\tablenotemark{c}} & \colhead{31} & \colhead{50} &

420:  \colhead{84}

421: & \colhead{50}} 

422: \startdata %

423: NF-QG &  -  &  $  0.26 \pm 0.13 $  &  $  0.15 \pm 0.06 $ & $  0.12 \pm 0.05 $   &  $  0.16 \pm 0.05 $ \\

424: \ssim-QG &  -  &    NA\tablenotemark{d}   &  $  1.22 \pm 0.09 $  &  $  0.73 \pm 0.07 $  &  $  1.00 \pm 0.11$\tablenotemark{e}\\

425: \ssim-Q &  -  &  NA\tablenotemark{d}    &  $  1.41 \pm 0.10 $  &  $  0.73 \pm 0.07 $ & $  1.05 \pm 0.12 $\tablenotemark{e}\\

426: \ssim-QGS &  -  &    $  3.46 \pm  0.24 $& $  0.64 \pm 0.06 $ & $  0.41 \pm 0.05 $  &  $  0.73 \pm 0.08 $ \\

427: \tsim-QG &  -  &   NA\tablenotemark{d}  &  $  0.94 \pm 0.07 $  &  $  0.69 \pm 0.06 $   &  $  0.41 \pm 0.07 $\tablenotemark{e}\\

428: \tsim-Q &  -  &   NA\tablenotemark{d}  &  $  1.43 \pm 0.10 $   & $  0.94 \pm 0.07 $  &  $  0.58 \pm 0.07 $\tablenotemark{e}\\

429: \tsim-QGS &  -  &  $  1.40 \pm 0.22 $ &   $  0.26 \pm 0.06 $ & $  0.02 \pm 0.05 $ & $  0.56 \pm 0.07  $t \\

430: \ssim-QG &  C  & $  0.82 \pm 0.17 $   &  $  0.05 \pm 0.05 $   & $ -0.59 \pm 0.07  $   & $  0.76 \pm 0.07$  \\

431: \ssim-Q &  C  &   $  1.14 \pm 0.24 $   & $  0.28 \pm 0.06 $  & $ -0.16 \pm 0.07 $   &  $  0.87 \pm 0.07$ \\

432: \ssim-QGS &  C  &  $ -0.01 \pm 0.13 $    & $ -0.45 \pm 0.06 $  &  $ -1.42 \pm 0.05 $  & $  0.96 \pm 0.08 $  \\

433: \tsim-QG &  C  &  $  1.18 \pm 0.19 $    & $  0.13 \pm 0.05 $  & $ -0.31 \pm 0.06  $  & $  0.77 \pm 0.08 $  \\

434: \tsim-Q &  C  &  $  1.29 \pm 0.21 $   &  $  0.42 \pm 0.06 $  &   $  0.04 \pm 0.06 $ & $  0.68 \pm 0.08 $   \\

435: \tsim-QGS &  C  &  $  0.10 \pm 0.14 $    &  $ -0.28 \pm 0.06 $ &  $ -1.27 \pm 0.06 $  & $  1.07 \pm 0.12 $  \\

436: \ssim-QG &  C1  &   $  2.92 \pm 0.20 $  & $  0.48 \pm 0.06 $   &  $  0.11 \pm 0.06 $  &  $  0.65 \pm 0.06 $ \\

437: \ssim-Q &  C1  &  $  3.34 \pm 0.27 $  &  $   0.63 \pm 0.06 $   & $  0.34 \pm 0.06 $   &  $  0.78 \pm 0.07 $ \\

438: \ssim-QGS &  C1  & $  1.60 \pm 0.22 $   &  $  0.20 \pm 0.06 $   & $ -0.46 \pm 0.06 $   &  $  0.75 \pm 0.06 $ \\

439: \tsim-QG &  C1  &  $  1.86 \pm 0.23 $  & $  0.44 \pm 0.06 $   &  $  0.18 \pm 0.05 $  &  $  0.57 \pm 0.06 $ \\

440: \tsim-Q &  C1  &   $  4.65 \pm 0.32 $    & $  0.81 \pm 0.07 $  &  $  0.47 \pm 0.05 $   & $  0.44 \pm 0.07 $  \\

441: \tsim-QGS &  C1  &  $  0.68 \pm 0.18 $   & $  0.09 \pm 0.06 $  &  $ -0.42 \pm 0.06 $   & $  0.84 \pm 0.07 $  

442: \enddata 

443: \tablenotetext{a}{The log of the best-fit offset to the simulations.

444: A positive value indicates that the prediction for the percentile is

445: too low. Errors are given by $\Delta\chi^2=1$.}

446: \tablenotetext{b}{These correspond to -0.5, 0.0, and +1.0 $\sigma$ in a lognormal distribution of $\tau$ about the median; see S03. There are 22, 41, and 35 degrees of freedom for percentiles 31, 50, and 84, in CIV/HI, and 24 d.o.f. for CIII/CIV.}

447: \tablenotetext{c}{For `C' models particles with $t_c<t_H$ are set to $T=2\times10^4$\,K. For `C1' models this is done if $t_c < 0.1t_H$.}

448: \tablenotetext{d}{For these models the simulation values were too low to obtain a reliable fit.}

449: \tablenotetext{e}{In the CIII/CIV results for these models, a yield given by the CIV/HI offset was used; these are marked by an `F' in Fig.~\ref{fig-civnocool}.}

450: \end{deluxetable} 

451: %\end{deluxetable*} 

452: 

453: 

454: \section{Conclusions}

455: \label{sec-conc}

456: 

457: A major goal for cosmological simulations is to develop a prescription

458: for feedback with which observations of the IGM and galaxies can be

459: simultaneously reproduced. We have compared in detail the statistics

460: of \CIV\ and \CIII\ absorption in a set of six high-quality $z\sim

461: 3-4$ quasar spectra to that in simulated spectra drawn from two

462: tate-of-the-art cosmological SPH simulations with a wide range of UVB

463: models and two different prescriptions for feedback: the simulation of

464: Springel \& Hernquist (2003a), which predicts a numerically resolved

465: star formation history that is consistent with the observations and

466: the simulation of Theuns et al.\ (2002) which matches observations of

467: the \HI\ Ly$\alpha$ forest. We have come to the following conclusions:

468: 

469: \begin{itemize}

470: 

471: \item {The simulations predict far too little \CIV\ absorption unless

472: the UVB is extremely soft.}

473: 

474: \item {In all cases, the simulations predict far too small \CIII/\CIV\

475: ratios.}

476: 

477: \item The simulations predict that many of the heavy elements reside

478:   in gas that is metal-rich ($Z \ga Z_\odot$) and hot ($10^5\,

479:   \lesssim T < 10^7\,$K). Much of this gas should be able to cool via

480:   metal lines, which was not included in the simulations.  It will be

481:   important to accurately model cooling in future numerical

482:   simulations.

483: 

484: \item If a crude cooling prescription is applied, the \CIV\ absorption

485: increases significantly, but the predicted \CIII/\CIV\ ratio is still

486: far too low because the metals reside in gas that is too low density.

487: Cooling should increase the enriched gas density, and this can be

488: estimated in future simulations with metal-line cooling.

489:  

490: \item Independent of the cooling prescription, the metal distribution

491:   in the simulations is too inhomogeneous to match the observed

492:   distribution of $\tau_{\rm CIV}(\tau_{\rm HI})$.

493: 

494: \end{itemize}

495: 

496: Numerical simulations with strong outflows from $M_{\rm baryon}

497: \gtrsim 10^8\msol$ galaxies deposit most intergalactic metals in hot,

498: metal-rich bubbles that preferentially inhabit voids and comprise a

499: relatively small filling factor, which allows them to avoid overly

500: disrupting the Ly$\alpha$ forest (Theuns et al.\ 2002).  However,

501: these very attributes appear to prevent them from reproducing the

502: observations of absorption by heavy elements.

503: 

504: If metal cooling is efficient or if the gas has an unresolved

505: multiphase structure then these winds may account for some of the

506: observed enrichment; otherwise they would be largely hidden from

507: current observations.  In either case it appears that an additional

508: ingredient -- either in the form of another enrichment mechanism, or

509: higher-$z$ enrichment that is unresolved by the simulations -- is

510: needed.

511: 

512: \acknowledgements This work was supported by the W.M.~Keck foundation,

513: NSF grants PHY-0070928, AST 02-06299, and AST 03-07690, NASA ATP

514: grants NAG5-12140, NAG5-13292, and NAG5-13381, a PPARC Advanced

515: Fellowship, and Silicon Graphics/Cray Research.

516: 

517: \begin{thebibliography}{}

518: 

519: \bibitem[Aguirre et al.(2001)]{2001ApJ...560..599A} Aguirre, A.,

520: Hernquist, L., Schaye, J., Weinberg, D.~H., Katz, N., \& Gardner, J.\

521: 2001, \apj, 560, 599

522: 

523: \bibitem[Paper(I)]{paper1} Aguirre, A., Schaye, J., \& Theuns, T.\

524: 2002, \apj, 576, 1 (A02)

525: 

526: \bibitem[Aguirre et al.(2004)]{2004ApJ...602...38A} Aguirre, A., Schaye, 

527: J., Kim, T., Theuns, T., Rauch, M., \& Sargent, W.~L.~W.\ 2004, \apj, 602, 

528: 38 (A04)

529: 

530: \bibitem[Anders \& Grevesse(1989)]{1989GeCoA..53..197A} Anders, E.~\& 

531: Grevesse, N.\ 1989, \gca, 53, 197 

532: 

533: \bibitem[Aracil, Petitjean, Pichon, \& Bergeron(2004)]{2004A&A...419..811A} 

534: Aracil, B., Petitjean, P., Pichon, C., \& Bergeron, J.\ 2004, \aap, 419, 

535: 811 

536: 

537: \bibitem[Boksenberg et al.(2003)]{boksen} Boksenberg, A., Sargent, W.L.W., \& Rauch, M.\ 2003, \apjs, submitted; astro-ph/0307557

538: 

539: \bibitem[Cen, Nagamine, \& Ostriker(2004)]{2004astro.ph..7143C} Cen, R., 

540: Nagamine, K., \& Ostriker, J.~P.\ 2004, astro-ph/0407143

541: 

542: \bibitem[Cowie \& Songaila(1998)]{1998Natur.394...44C} Cowie, L. L. \& 

543: Songaila, A. 1998, \nat, 394, 44 

544: 

545: \bibitem[Cowie et al.(1995)]{1995AJ....109.1522C} Cowie,

546: L.~L., Songaila, A., Kim, T., \& Hu, E.~M.\ 1995, \aj, 109, 1522 

547: 

548: \bibitem[Ellison et al.(2000)]{2000AJ....120.1175E} Ellison, S.~L., Songaila, A., Schaye, J., \& Pettini, M.\ 2000, \aj, 120, 1175 

549: 

550: \bibitem[Furlanetto \& Loeb(2003)]{2003ApJ...588...18F} Furlanetto, 

551: S.~R.~\& Loeb, A.\ 2003, \apj, 588, 18

552: 

553: \bibitem[Gnedin(1998)]{1998MNRAS.294..407G} Gnedin, N.~Y.\ 1998, \mnras, 

554: 294, 407 

555: 

556: \bibitem[Haardt \& Madau(2001)]{haardt01:cuba}

557: Haardt, F.~\& Madau, P. 2001, astro-ph/0106018

558: 

559: \bibitem[Hernquist \& Springel(2003)]{2003MNRAS.341.1253H} Hernquist, L.~\& 

560: Springel, V.\ 2003, \mnras, 341, 1253 

561: 

562: \bibitem[Kay, Pearce, Frenk, \& Jenkins(2002)]{2002MNRAS.330..113K} Kay, 

563: S.~T., Pearce, F.~R., Frenk, C.~S., \& Jenkins, A.\ 2002, \mnras, 330, 113 

564: 

565: \bibitem[Paper(II)]{paper2} Schaye, J., Aguirre, A., Kim, T., Theuns,

566: T., Rauch, M., \& Sargent, W.L.W.\ 2003, \apj, 596, 768 (S03)

567: 

568: \bibitem[Scannapieco, Ferrara, \& Madau(2002)]{2002ApJ...574..590S} 

569: Scannapieco, E., Ferrara, A., \& Madau, P.\ 2002, \apj, 574, 590 

570: 

571: 

572: \bibitem[Simcoe, Sargent, \& Rauch(2004)]{2004ApJ...606...92S} Simcoe, 

573: R.~A., Sargent, W.~L.~W., \& Rauch, M.\ 2004, \apj, 606, 92

574: 

575: \bibitem[Shapley, Steidel, Pettini, \& 

576: Adelberger(2003)]{2003ApJ...588...65S} Shapley, A.~E., Steidel, C.~C., 

577: Pettini, M., \& Adelberger, K.~L.\ 2003, \apj, 588, 65

578: 

579: \bibitem[Songaila \& Cowie (1996)]{1996AJ....112..335S} Songaila, A.  \&

580: Cowie, L. L. 1996, \aj, 112, 335

581: 

582: \bibitem[Springel \& Hernquist(2003)]{2003MNRAS.339..312S} Springel, V.~\& 

583: Hernquist, L.\ 2003a, \mnras, 339, 312 (SH03)

584: 

585: \bibitem[Springel \& Hernquist(2003)]{2003MNRAS.339..289S} Springel, V.~\& 

586: Hernquist, L.\ 2003b, \mnras, 339, 289 

587: 

588: \bibitem[Sutherland \& Dopita(1993)]{1993ApJS...88..253S} Sutherland, 

589: R.~S.~\& Dopita, M.~A.\ 1993, \apjs, 88, 253

590: 

591: \bibitem[Theuns et al.(2002)]{2002ApJ...578L...5T} Theuns, T., Viel, M., 

592: Kay, S., Schaye, J., Carswell, R.~F., \& Tzanavaris, P.\ 2002, \apjl, 578, 

593: L5 (T02)

594: 

595: 

596: \end{thebibliography}

597: \end{document}

598: