A&A 473, 791-803 (2007)
A. J. Fox1 - C. Ledoux2 - P. Petitjean1, 3 - R. Srianand4
1 - Institut d'Astrophysique de Paris, UMR7095 CNRS, Université Pierre et Marie Curie, 98bis Bd. Arago, 75014 Paris, France
2 - European Southern Observatory, Alonso de Córdova 3107, Casilla 19001, Vitacura, Santiago 19, Chile
3 - LERMA, Observatoire de Paris, 61 avenue de l'Observatoire, 75014 Paris, France
4 - IUCAA, Post Bag 4, Ganesh Khind, Pune 411 007, India
Received 13 April 2007 / Accepted 27 July 2007
We present a study of C IV absorption in a sample of 63 damped Lyman- (DLA) systems and 11 sub-DLAs in the redshift range , using a dataset of high-resolution (6.6 km s-1 FWHM), high signal-to-noise VLT/UVES spectra. The complex absorption line profiles show both narrow and broad C IV components, indicating the presence of both warm, photoionized and hot, collisionally ionized gas. We report new correlations between the metallicity (measured in the neutral-phase) and each of the C IV column density, the C IV total line width, and the maximum C IV velocity. We explore the effect on these correlations of the sub-DLAs, the proximate DLAs (defined as those within 5000 km s-1 of the quasar), the saturated absorbers, and the metal line used to measure the metallicity, and we find the correlations to be robust. There is no evidence for any difference between the measured properties of DLA C IV and sub-DLA C IV. In 25 DLAs and 4 sub-DLAs, covering 2.5 dex in [Z/H], we directly observe C IV moving above the escape speed, where is derived from the total line width of the neutral gas profiles. These high-velocity C IV clouds, unbound from the central potential well, can be interpreted as highly ionized outflowing winds, which are predicted by numerical simulations of galaxy feedback. The distribution of C IV column density in DLAs and sub-DLAs is similar to the distribution in Lyman Break galaxies, where winds are directly observed, supporting the idea that supernova feedback creates the ionized gas in DLAs. The unbound C IV absorbers show a median mass flow rate of 22 (r/40 kpc) yr-1, where r is the characteristic C IV radius. Their kinetic energy fluxes are large enough that a star formation rate (SFR) of 2 yr-1 is required to power them.
Key words: galaxies: quasars: absorption lines - galaxies: high-redshift - galaxies: halos - galaxies: ISM - galaxies: kinematics and dynamics
Studying this protogalactic plasma allows one to address two key themes of extragalactic astronomy: galactic winds and the metal budget. Galactic winds are common at high redshift (Veilleux et al. 2005), and must be present in order to enrich the IGM up to its observed metallicity (Aracil et al. 2004; Aguirre et al. 2005,2001). Simulations predict that the level of ionization in winds is high (Kawata & Rauch 2007; Fangano et al. 2007; Oppenheimer & Davé 2006), and direct observations of absorption in high-ionization lines have been made in and around Lyman Break galaxies (LBGs; Adelberger et al. 2005; Shapley et al. 2003; Pettini et al. 2002,2000). Low-redshift studies of galactic outflows have also found a high-ionization component (Strickland et al. 2004; Heckman et al. 2001). Since DLAs represent the largest reservoirs of neutral gas for high-redshift star formation (Wolfe et al. 2005), they are natural sites to look for supernova-driven winds and plasma halos in general.
On the second front, ionized halos are important because of their potential ability to close the metal budget at : there is currently a discrepancy between the global density of metals predicted by integrating the star formation history, and the density of metals actually observed (Pettini 1999; Bouché et al. 2006; Ferrara et al. 2005; Bouché et al. 2007,2005; Sommer-Larsen & Fynbo 2007). The contribution of plasma halos to the metal budget will be particularly significant if the plasma is hot and collisionally ionized, because the cooling times in low-metallicity, low-density, hot halos are extremely long, and so gas injected into these environments can become locked up until the current epoch. For 106 K gas at 10-3 cm-3 and solar metallicity, is yr (Houck & Bregman 1990), so assuming that to first order , we find that the cooling time in gas at one-hundredth solar metallicity (as seen in DLAs) would be approximately equal to the Hubble time. Finding the gas in hot halos is therefore important for tracing the history of the cosmic metals.
In a recent paper (Fox et al. 2007, hereafter Paper I), we discussed the first observations of O VI absorption in DLAs, finding evidence for a hot ionized medium that, modulo certain assumptions on metallicity and ionization, typically contains 40% as many baryons and metals as there are in the neutral phase. Though 12 DLAs with O VI detections were found, detailed kinematic measurements of the O VI absorption are difficult due to the high density of blends with the Lyman- forest. However, if one instead traces the ionized gas with C IV, whose lines lie redward of the Lyman- forest, the blending problems are avoided. Thus, although C IV may trace a lower temperature phase of plasma than O VI, it is a better ion to study for building a sample of statistical size. Partly for this reason, the properties of C IV absorption in the IGM at z>2 have been studied at length (Ellison et al. 2000; Songaila 2006,2005; Aguirre et al. 2005; Schaye et al. 2003; Aracil et al. 2004; Boksenberg et al. 2003; Schaye et al. 2007; Cowie & Songaila 1998; Scannapieco et al. 2006a).
Previous observations of C IV absorption in DLAs (Wolfe & Prochaska 2000a; Lu et al. 1996; Ledoux et al. 1998) and sub-DLAs (Péroux et al. 2007; Dessauges-Zavadsky et al. 2003; Richter et al. 2005; Péroux et al. 2003) have found the C IV profiles generally occupy a more extended (though overlapping) velocity range than the neutral gas profiles. In an attempt to explain these observations, Wolfe & Prochaska (2000b) tested a model of gas falling radially onto centrifugally-supported exponential disks, and found it was unable to reproduce the observed C IV kinematics. On the other hand, Maller et al. (2003) found that a model in which hot gas in halos and sub-halos gives rise to the C IV absorption in DLAs was generally successful in explaining the kinematics.
We continue the study of C IV in DLAs in this paper. We are partly motivated by the recent work of Ledoux et al. (2006), who reported a correlation between low-ion line width and metallicity [Z/H] in a sample of 70 DLAs and sub-DLAs, covering over two orders of magnitude in metallicity (see also Wolfe & Prochaska 1998; Prochaska et al. 2007a; Murphy et al. 2007). Since the line widths of the neutral species are thought to be gravitationally-dominated, the broader lines may be tracing the more massive halos, and so the -[Z/H] correlation has been interpreted as evidence for an underlying mass-metallicity relation. A natural follow-on question is whether a similar correlation exists between the high-ion line width and DLA metallicity. In this paper we investigate whether such a correlation exists, as well as exploring other relations between the properties of C IV and those of the neutral gas. To maximize our sample size, we include observations of C IV absorption in both DLAs and sub-DLAs (and we compare the properties of the C IV absorption in the two samples). There is some evidence that sub-DLAs display larger metallicities than DLAs (Dessauges-Zavadsky et al. 2003; Kulkarni et al. 2007; Péroux et al. 2005), and they have been suggested to be more massive (Khare et al. 2007).
The structure of this paper is as follows. Section 2 covers the observations, sample selection, and measurements. In Sect. 3 we present observed correlations in the data set. In Sect. 4 we discuss the interpretation of these correlations, and we identify a population of absorbers that may trace galactic winds. A summary is presented in Sect. 5.
We also define an alternative measure of total line width as v+-v-, the total velocity range over which C IV absorption is present, regardless of saturation. v+-v- is sensitive to weak but nonetheless interesting satellite components that are not contained within . These weak components are particularly relevant in the search for winds. The drawback of using v+-v- is that it has a larger error than , since v- and v+ are selected by eye (we estimate km s-1), and also that it is sensitive to the signal-to-noise ratio (low optical depth absorption is harder to detect in low S/N data). Our data is of high enough quality to ensure that the second effect should not be a major concern: the noisiest spectrum in our sample has S/N = 25, and the mean S/N is 51 (where the S/N is measured per resolution element at the observed wavelength of C IV).
The measurements of column density, total line width, and mean velocity were conducted independently on 1548 and 1550. To select which of the two C IV transitions to use for our final measurement, we followed the following rules that assess the influence of saturation. If the condition (corresponding to ) is true for 1548, where v0 denotes the velocity where the absorption in strongest, we use 1548 to measure the C IV, otherwise we use 1550. If both lines are saturated (defined here as when or at any point within the line profile), we proceed with a lower limit to the column density and an upper limit to the line width using the results from 1550. However, if one of the two C IV lines is blended, we use the other line for measurement, regardless of the level of saturation. There are four cases where both C IV lines are partly blended, but we still have useful information at other velocities within the line profiles. This can occur when the two C IV lines, separated by 500 km s-1, blend with each other. In these cases, which are flagged in Table 1, we derived our best estimate of log using the sum of the column densities measured over two separate unblended velocity ranges, and assuming and , where for these cases v- refers to the lower bound of absorption of the lower velocity range, and v+ refers to the upper bound of absorption of the higher velocity range.
Histograms of the total C IV line width (using both and v+-v-) and C IV velocity offset are given in Fig. 2. A significant difference between the distributions of and v+-v- can be seen, with the peak in the distribution at 80 km s-1, and the peak in the v+-v-distribution occurring at 200 km s-1. This difference is due to the presence of weak, outlying components which contribute to v+-v- but not to . Both distributions show an extended tail reaching over 1000 km s-1. Figure 2 also shows the distribution of mean C IV velocities.
|Figure 2: Normalized histograms of total C IV line width and absolute velocity offset among our DLA/sub-DLA sample. The distribution of both (dotted) and v+-v- (solid) is shown in the top panel. The peak of the v+-v- distribution is 100 km s-1 broader than the peak of the distribution, reflecting the presence of low optical depth satellite components. There is an extended tail of line widths reaching >1000 km s-1. We have treated the upper limits to as data points when forming the distribution. The distribution of the absolute C IV velocity offset (i.e., the mean C IV velocity relative to the neutral gas) is shown in the lower panel.|
We can assess the level of IGM contamination statistically by comparing the properties of the C IV in our sample with those of C IV in the IGM. This is shown in Fig. 3, where we compare our DLA/sub-DLA C IV column density distribution with the IGM distribution at -3 measured by Boksenberg et al. (2003, data taken from their Fig. 10). A similar IGM C IV distribution is presented by Songaila (2005). We also include in Fig. 3 the distribution of C IV near LBGs, taken from Table 3 in Adelberger et al. (2005); these measurements were made by finding LBGs lying at impact parameters of <1 h70-1co-moving Mpc from QSO sight lines, and then measuring the C IV column densities in the QSO spectra at velocities within 200 km s-1 of the LBG redshift. Strong C IV absorption is also directly observed in LBG spectra by Shapley et al. (2003).
The DLA/sub-DLA C IV absorbers are clearly a different population from the IGM C IV absorbers: the DLA/sub-DLA population shows a mean column density that is higher by almost 1 dex. The distribution of C IV in DLAs and sub-DLAs resembles the distribution of galactic C IV as seen in LBGs, both with mean column densities near 1014 cm-2. In consequence, the highest systems in our sample are least likely to be of IGM origin. Since we report below that the highest systems tend to show the broadest C IV, we come to the conclusion that even the broadest C IV absorbers (that were potentially the most suspect in terms of an association with an individual galaxy), are likely to be galactic.
|Figure 3: Comparison of the normalized C IV column density distributions: (i) in DLAs and sub-DLAs (solid line, this work); (ii) in the IGM at (dashed line; Boksenberg et al. 2003); (iii) around LBGs at (dotted line; Adelberger et al. 2005). In each case, is integrated over all components. Note how a typical DLA/sub-DLA shows (a) considerably stronger C IV than a typical IGM C IV absorber, but (b) a C IV column similar to the mean seen in the LBG distribution. These two findings support our interpretation that the C IV in DLAs and sub-DLAs is galactic rather than intergalactic. We have included the saturated C IV absorbers as data points in the DLA/sub-DLA distribution, using the measured lower limits; this has the effect of artificially truncating the high tail of the solid line.|
|Quan. 1||Quan. 2||Sample||Size|
|C II* detected||21||0.30||1.9|
|a In the intervening DLA sample, sub-DLAs and DLAs at <5000km s-1 from the QSO were excluded. b In the unsaturated sample (considered for correlations involving or ), all saturated C IV lines were removed. c In the Zn II sample, cases where [Z/H] was derived from Si or S were excluded. This only applies to correlations involving [Z/H].|
We now discuss correlations (or lack thereof) between the various measured quantities in our dataset. For reference, a summary of all correlations found and their statistical significance is given in Table 2.
|Figure 4: Comparison of high-ion and low-ion total line width for DLAs (filled circles) and sub-DLAs (open circles). Absorbers at <5000 km s-1 from the QSO redshift are highlighted in square symbols. The dashed line shows where . In 69 of 74 cases the C IV absorption is broader than the neutral absorption. There is a large scatter in at low , but the scatter decreases with increasing . Saturated C IV absorbers are shown with upper limits to .|
|Figure 5: Correlations between the measured C IV properties for both DLAs (filled circles) and sub-DLAs (open circles). Proximate absorbers are highlighted in square symbols. We use v+-v- rather than to measure the line width, since it is defined even in the saturated cases. We annotate the Kendall rank correlation coefficient and its significance on the panel, for various sub-samples. Solid lines show linear least-square bisector fits for the case where all data points are treated equally (including limits). Top panel: C IV column density vs. C IV line width. A correlation is found even when excluding the saturated C IV absorbers (the lower limits on ), although in this case the slope of the fit (shown as a dashed line) is shallower. Bottom panel: comparison between total C IV line width and C IV absolute velocity offset, also showing a significant correlation. These trends show that the stronger C IV absorbers tend to be broader and more offset from the neutral gas than the weaker absorbers.|
|Figure 6: Dependence of C IV properties in DLAs (filled circles) and sub-DLAs (open circles) with neutral-phase metallicity. Proximate absorbers are highlighted in square symbols. In each panel, we annotate the Kendall rank correlation coefficient and its significance, and we show a linear least-square bisector fit (solid line), for the case where all data points are treated equally (including the limits). In panel a), we find a significant correlation between and [Z/H]. This remains true (at 2.5) even in the case where the saturated points are excluded, though in this case the slope is slightly shallower (dashed line). In panel b), we show a 3.4 correlation between and [Z/H], but this correlation is not found when the saturated points are excluded (hence there is no dashed line in this panel). However, if we instead use to measure the line width ( bottom panel), since this statistic is not affected by saturation, a significant (4.1) metallicity-line width correlation does exist. The detected correlations show that high-metallicity systems tend to exhibit strong and broad C IV absorption.|
A linear least-squares bisector fit to the data gives the result:
If we remove the upper limits on (i.e., the saturated absorbers) from the sample, and redo the correlation analysis, we find no significant detection of a correlation between [Z/H] and remains. However, in a sense this is not surprising, since the saturated absorbers tend to show broad C IV lines (Sect. 3.4), so by removing them, we are biased against finding a trend with line width. To further investigate whether saturation was playing a role in setting up this metallicity-line width relation, we looked for a correlation between [Z/H] and As discussed in Sect. 2.3.2, is defined even in the saturated cases. The result was that we detected a positive correlation at the 4.1 level. Since we believe this metallicity-C IV line width correlation to be one of the most important results of this paper, we investigated whether it was seen independently in the lower- and higher-redshift halves of the sample, and found that it was at significance (Fig. 7), even though the mean metallicity of the low-redshift sample is higher than the mean metallicity of the high-redshift sample. Together, these results imply that the C IV line width and metallicity are correlated in DLAs and sub-DLAs.
|Figure 7: Illustration that the correlation between C IV total line width and metallicity exists independently in the lower and upper redshift halves of the sample, even though there is a difference between the mean metallicity of the two sub-samples (the lower-z sample shows systematically higher [Z/H]). The symbols have their same meanings as in Fig. 6. All DLAs and sub-DLAs in each redshift range were included in the correlation analysis and in the linear bisector fits, shown with solid lines.|
Using the sample of 74 DLAs and sub-DLAs,
the best-fit linear least-squares bisector model is:
There are four DLAs (at toward Q0112+306, toward Q2206-199, toward Q2222-396, and toward Q2344+125), and one sub-DLA (at toward Q1451+123) that stand out in Figs. 6b and c due to their unusual properties. These absorbers show narrow C IV lines ( km s-1 in all cases), low velocity offsets ( km s-1 in four of five cases), and low metallicities ([Z/H] between -2.42 and -1.81). These cases play a significant role in generating the correlations discussed here.
|(N in cm-2)||(kms-1)||(kms-1)||(N in cm-2)|
|a Each entry in this table shows the mean and standard deviation of the given property in the given category. b Proximate absorbers are those within 5000km s-1 of the QSO redshift. All others are intervening.|
The resulting values of are shown in Fig. 8. Based on the 74 systems in this sample, we find that the mean and standard deviation of the warm ionized-to-neutral ratio is , i.e. the C IV phase contains >10% of the baryons of the neutral phase. The sub-DLAs show a mean log of 19.33, whereas the DLAs show a mean log of 19.77, a factor of 2.2 higher. This is because sub-DLAs show (on average) similar C IV columns as DLAs, but higher metallicities. We note in the lower panel of Fig. 8 that there is no trend for , which is /(Z/H), to depend on metallicity (although such a trend could be partly hidden by the saturation of C IV in high metallicity systems). Thus, the correlation reported in Sect. 3.4 between and [Z/H] appears to be a simple consequence of the metallicity alone, but does not imply that there is more ionized gas (i.e. more H II) in the high-metallicity systems. We also note that the scatter in is substantial, of order 2 dex in at values of [Z/H] between -2.0 and -1.0.
|Figure 8: Comparison of H II column density in the C IV-bearing gas integrated over all velocities with ( top) and ( bottom) [Z/H] for each system in our sample, assuming an ionization fraction . The average value of the ratio N(H II)/N(H I) over all 74 systems is 0.1, implying that the C IV phase of DLAs and sub-DLAs typically contains >10% of the baryons and metals in the neutral phase. The mean value of is 2.5 times lower in sub-DLAs than in DLAs.|
We searched the literature for C II* measurements in our sample of DLAs/sub-DLAs with C IV. We took 20 data points (11 measurements and 9 upper limits) from Srianand et al. (2005), 9 measurements from Wolfe et al. (2003a), and one from Heinmüller et al. (2006). In Fig. 9, we directly compare the cooling rate with the C IV column density. Below log , there is no trend evident in the data. However, we find that the seven points with the highest are among the systems with the highest cooling rate. Even though a formal correlation between the cooling rate and is detected only at the 1.8 level (the C II* upper limits were excluded in this analysis), we note that the the median logarithmic C IV column density among the systems with log is 13.86, whereas the median among the systems with log is 14.83. This finding is consistent with the results of Wolfe (2007, in preparation), who finds evidence for bimodality in DLAs based on the cooling rate, in the form of significant differences between the metallicities and velocity widths of those DLAs with cooling rates below and above a critical value erg s-1 H-1.
|Figure 9: Dependence of the C IV column density on the cooling rate derived from the ratio, for each DLA (filled circles) and sub-DLA (open circles) where data on C IV and C II* exist. Color-coding is used to denote the metallicity of the gas, as indicated in the legend. The seven data points with the highest all show above-average cooling rates. Among these seven, six show high metallicities.|
We propose that the broad C IV components arise in the hotter phase of DLA plasma that is detected in O VI absorption (Paper I), i.e. the hot ionized medium. This phase will arise following either heat input from supernova in the DLA host galaxy, or by the shock-heating of infalling gas at the virial radius. In the first case, the hot ionized medium may exist in the form of a wind (Kawata & Rauch 2007; Fangano et al. 2007; Oppenheimer & Davé 2006), though Galactic fountain scenarios are also possible (Bregman 1980; Houck & Bregman 1990; Shapiro & Field 1976). The observation that up to 80% of the C IV components are broad is consistent with the origin of the C IV in a wind, since in the models of Oppenheimer & Davé (2006), much of the C IV in galactic winds is collisionally ionized.
We note that type II supernovae will heat interstellar gas to temperatures >106 K, too high for the formation of O VI and C IV lines, and left to itself, gas at a density of 10-3 cm-3 and one-hundredth of the solar metallicity will not cool in a Hubble time. However, if the hot plasma interacts with cool or warm entrained clouds, conductive interfaces (Borkowski et al. 1990), turbulent mixing layers (Esquivel et al. 2006; Slavin et al. 1993), or shock fronts (Dopita & Sutherland 1996) can form between the hot and cool phases, in which the temperatures are favorable for the formation of O VI and C IV lines. These mechanisms, which have been invoked to explain high-ion observations in the extended halo of the Milky Way (Savage et al. 2003; Zsargó et al. 2003; Indebetouw & Shull 2004), in high-velocity clouds (Fox et al. 2004; Collins et al. 2005; Fox et al. 2005; Sembach et al. 2003), and in the Large Magellanic Cloud (Lehner & Howk 2007), can explain the broader C IV and O VI components seen in the DLAs and sub-DLAs. Note that the interpretation of broad high-ion components in DLAs and sub-DLAs as hot and collisionally ionized is different from the interpretation of the O VI components in the IGM at , which (generally) appear to be photoionized (Lopez et al. 2007; Carswell et al. 2002; Bergeron & Herbert-Fort 2005; Bergeron et al. 2002; Levshakov et al. 2003; Reimers et al. 2006), though see Reimers et al. (2001) and Simcoe et al. (2006,2002).
In order to evaluate whether any of the observed C IV components represent winds, we need to determine the escape velocity in each system. We calculate this using , an empirical relation found from analysis of artificial spectra in the simulations of Haehnelt et al. (1998) and Maller et al. (2001). There is a factor of two dispersion around this relation, due to variations in the viewing angle. We then take (appropriate for a spherical halo), so that . The escape speed we have assumed can be taken as an upper limit, because is calculated in the disk and will decrease with radius, and we may be observing C IV at high radii. We have not accounted for drag forces arising due to entrainment between the winds and the galaxy's ISM (Finlator & Davé 2007).
When we search for C IV absorption at
we find it in 25 of 63 DLAs and 4 of 11 sub-DLAs, i.e.
C IV absorption unbound from the central potential well
exists in 40% of cases.
These absorbers, which we refer to as wind candidates, are colored with dark
shading in Fig. 1, for easy identification.
A key property of the wind candidate absorbers is the low column
density of the accompanying neutral gas absorption.
In this respect, the wind candidates are analogous to the highly
ionized high-velocity clouds seen in the vicinity of the Milky Way
(Sembach et al. 1999; Fox et al. 2006; Collins et al. 2004; Ganguly et al. 2005; Fox et al. 2005; Collins et al. 2005; Sembach et al. 1995,2003).
They also resemble the
C IV absorbers reported by
Ryan-Weber et al. (2006).
In a handful of cases, wind candidates are seen at both redshifted and
blueshifted velocities in the same system.
In Fig. 10 (top panel) we
plot the absolute wind C IV columns as a function of metallicity.
We find similar C IV wind column densities in systems that span the
2.5 dex range of metallicity in our sample, even in the
highest metallicity systems.
|Figure 10: Top panel: C IV column density moving above the escape speed (i.e. wind candidate ) as a function of metallicity, for the 25 DLAs and 4 sub-DLAs with C IV absorption at , where . Bottom panel: wind mass outflow rate divided by characteristic C IV radius r40 (in units of 40 kpc) vs metallicity. The median mass outflow rate among these cases is 22r40 yr-1. The large diamond shows the mass outflow rate determined in the z=2.7 Lyman Break Galaxy cB58 by Pettini et al. (2000), assuming r=1 kpc. The filled triangles are the mass flow rates determined for the cool winds in the Martin (2006) sample of low-redshift ultraluminous starburst galaxies and plotted at zero metallicity for convenience. Only the DLA and sub-DLA points have been divided by r40.|
Our results are surprising when viewed in the light of simulations of galactic outflows, which show that dwarf galaxies are more important than massive galaxies for the metal pollution of the intergalactic medium (Nagamine et al. 2004b; Scannapieco et al. 2006b; Ferrara et al. 2000; Tissera et al. 2006; Mac-Low & Ferrara 1999; Kobayashi et al. 2007), since they are incapable of gravitationally confining the metals released by supernovae. The increase in wind escape fraction with decreasing galactic mass may contribute to the origin of the mass-metallicity relationship observed in DLAs and other high-redshift galaxies (Tremonti et al. 2004; Erb et al. 2006; Møller et al. 2004; Nagamine et al. 2004a; Savaglio et al. 2005). The surprising result here is that, given the mass-metallicity relation, the higher metallicity (higher mass) galaxies should decelerate their supernova-driven outflows, so that C IV outflows should not be seen in the high-metallicity systems, but yet we observe the high-velocity components even in systems with . Furthermore, we find that the maximum outflow velocity is correlated to the metallicity.
For any individual C IV component, it is difficult to determine conclusively whether one is seeing a galactic outflow from the DLA galaxy (see Fangano et al. 2007). Two other plausible origins are inflow toward the DLA galaxy (Wolfe & Prochaska 2000b), and the ISM of a separate nearby galaxy. A contribution of C IV from these processes could help to explain both the scatter seen in each panel in Fig. 6, and the cases with large values for . One weakness of the inflow model is that accretion would accelerate the gas up to but not beyond the escape velocity, but yet we see gas moving above the escape velocity. Thus inflow cannot explain the highest-velocity C IV components. The nearby galaxy model has the problem of not readily explaining why the ionized-to-neutral gas ratio is so high in the high-velocity C IV components. In other words, if nearby galaxies are responsible for the high-velocity C IV components, where is their neutral ISM? The outflow explanation, on the other hand, naturally explains the velocities and the ionization properties of the high-velocity C IV absorbers. The outflow model also explains the presence of metals in the ionized gas (they came from the DLA host galaxy), so it does not need to resort to pre-enrichment. A more serious problem is whether we can associate a single DLA with a single galaxy. This assumption may be false since the clustering of galaxies near DLAs has been observed both at low (Chen & Lanzetta 2003) and high (Cooke et al. 2006a; Ellison et al. 2007; Bouché & Lowenthal 2004,2003; Cooke et al. 2006b) redshift. Our wind interpretation implicitly assumes that each high-velocity C IV component arose from a galaxy located in velocity at the point where the neutral line absorption is strongest.
We assume that the outflow exists in an expanding hemispherical shell
moving at velocity v relative to the neutral gas in the DLA galaxy
(we do not assume a full spherical outflow, since we do not generally
see two components corresponding to the near and far side of the galaxy).
We then calculate M, ,
using the following equations:
The only quantity which is not directly measured is the characteristic radius r. If we take a reference value of r=40 kpc, as determined for C IV around LBGs by Adelberger et al. (2005), we find that the 29 C IV wind candidates typically contain (median values) a total mass of , a kinetic energy of erg, a mass flow rate of 22 yr-1, and a kinetic energy injection rate of erg s-1. To put these wind mass flow rates in perspective, Pettini et al. (2002,2000) report a mass flow rate of 60 yr-1 for the outflow from the z=2.7 LBG MS 1512-cB58, and Martin (2006) calculate the mass loss rates in the cool winds of a sample of low-redshift ultraluminous starburst galaxies as between 1 and >14 yr-1.
We calculate that the thermal energy in the C IV wind absorbers is only a few percent of their kinetic energy (assuming a temperature of 105 K). Assuming the clouds fill a region whose depth is similar to its width, a 40 kpc depth and cm-2 correspond to a density of cm-3, assuming a filling factor of unity. In turn, such a medium would exhibit a thermal pressure cm-3 K. If the C IV arose in interface layers between entrained cool/warm clouds and a surrounding hot gas, then the line-of-sight filling factor would be much lower and the C IV could exist in higher-density, higher-pressure, localized regions.
Among the 25 DLAs and 4 sub-DLAs showing wind candidate components, the median rate at which each galaxy delivers metals to the IGM is yr-1. This estimate is robust to changes in the metallicity of the ionized gas with respect to the neutral gas. Assuming that the outflows are driven by the supernovae that follow star formation, we can calculate the star formation rate necessary to power the observed C IV outflows. We use the relation from Efstathiou (2000) between kinetic energy injection rate and star formation rate, erg s-1, with in yr-1. Equating to the observed mean flux of kinetic energy, we find a median required SFR per DLA of 2 yr-1, corresponding (with our assumed r=40 kpc) to a required SFR per unit area ( ) of yr-1 kpc-2. This is consistent with (though close to) the limit on the SFR per unit area derived from low surface brightness features (i.e. DLA analogs) in the Hubble Ultra-Deep Field by Wolfe & Chen (2006), who report yr-1 kpc-2. We thus conclude that, at least to an order-of-magnitude, there is sufficient energy released following star formation to drive the observed high-velocity C IV components in DLAs, supporting the notion that they trace galactic winds.
Infalling clouds can also contribute to the C IV seen in DLAs (e.g. Wolfe & Prochaska 2000b). However, as we have pointed out, infall cannot explain the highest velocity C IV components, which are detected at well over the escape speed. Although metallicity measurements could in theory discriminate between the infall and outflow hypotheses, it is very difficult to directly measure the metallicity of the ionized gas in DLAs. One practical test of the idea that star formation leads to the production of ionized gas in DLAs (and the associated idea that the high-velocity C IV components in DLAs trace winds) would be a detailed comparison between the properties of C IV absorption in DLAs and in LBGs, where outflows are directly observed (Shapley et al. 2003; Pettini et al. 2000,2002). Though we have not yet conducted a full comparison, we do note that the similar C IV column density distributions observed in DLAs and LBGs suggest that the ionized gas in these two classes of object shares a common origin in supernova-driven outflows.
A.J.F. gratefully acknowledges the support of a Marie Curie Intra-European Fellowship awarded by the European Union Sixth Framework Programme. P.P. and R.S. gratefully acknowledge support from the Indo-French Centre for the Promotion of Advanced Research (Centre Franco-Indien pour la Promotion de la Recherche Avancée) under contract No. 3004-3. We thank Alain Smette for kindly providing the spectra of many quasars in the Hamburg-ESO survey prior to publication. We acknowledge valuable discussions with Blair Savage, Jason Prochaska, and Art Wolfe. Finally, we thank the referee for perceptive comments that improved the paper.
|Figure 1: VLT/UVES absorption line spectra of C IV and an optically thin line (typically Si II or Fe II) chosen to trace the neutral phase component structure, for all DLAs and sub-DLAs in our sample. The flux is in arbitrary units, with the bottom of each panel at zero. The light (dark) shaded regions show C IV absorption below (above) , where (see text). The thick vertical line in each C IV panel denotes the optical-depth weighted mean velocity of the profile. The two narrow vertical lines show the velocities corresponding to 5% and 95% of the integrated optical depth; the interval between these velocities defines the total line width . A small letter "B'' within the shaded area indicates a blend; in these cases the other C IV line was used for measurement. The label "Proximate'' implies the absorber is at <5000 km s-1 from the QSO. Note how the wind candidate absorbers (the dark regions) show no absorption in the weak neutral line shown here, implying these absorbers have low column densities of neutral gas and are highly ionized.|