PROPERTIES AND ORIGIN OF THE HIGH-VELOCITY GAS TOWARD THE LARGE MAGELLANIC CLOUD^*

The interactions of galaxies and the nearby intergalactic medium (IGM) through the accretion of matter onto galaxies or the expulsion of matter and energy in winds from galaxies are crucial for the evolution of both the galaxies and the IGM. The star formation histories, gas content, and metallicity of a galaxy are co-dependent on both internal processes and on the interaction between the galaxy and the local IGM. Today's star-forming galaxies are continuing to form by accreting gas from the nearby IGM, from matter condensing out of a hot corona (e.g., Peek et al. 2008), from the stripped ISM of smaller dwarf systems (e.g., Putman et al. 1998), or even from matter ejected by earlier star formation episodes (Bertone et al. 2007; Bouché et al. 2007); much of the accretion may proceed through analogs to the high-velocity clouds (HVCs) found about the Milky Way (MW, e.g., Wakker et al. 2007; Lockman et al. 2008; Thom et al. 2008). The continued formation of stars in galaxies without depleting the available gas and the observed metallicity distribution of long-lived stars in our galaxies require a continuous infall of low-metallicity matter (e.g., van den Bergh 1962; Tinsley 1981; Matteucci 2003; Bland-Hawthorn 2008). Balancing the accretion of new matter, feedback from strong star formation can drive large-scale circulations of gas away from the disk or outflows that feed matter into the nearby IGM (e.g., Veilleux et al. 2005; Simcoe et al. 2006; Oppenheimer & Davé 2008). These competing processes are at the heart of the evolution of galaxies and the IGM.

Most of the current observational information of galactic interaction and winds comes from the observation of emission lines of hydrogen and metals (e.g., Martin 1998; Veilleux et al. 2005; Strickland & Heckman 2007). These observations have yielded crucial information, such as the multi-phase nature (from very hot ionized gas to cold neutral gas) and large-scale morphology of these features. But because their density-squared dependence, emission line observations are heavily biased toward the highest-density regions, which may only trace a relatively small fraction of the total mass and energy (Strickland & Stevens 2000; Veilleux et al. 2005). Measurements that rely on absorption lines are less biased to the highest densities, but they generally probe only directions toward the brightest stellar clusters, losing detail as the stars are integrated in the spectrograph slit, and are restricted to starburst galaxies (Heckman et al. 2001; Cannon et al. 2005). However, one counterexample is the Magellanic Clouds that are near enough to allow absorption line measurements toward many individual stars, and thus are excellent laboratories for studying recycling processes of energy and matter in low-mass, low-metallicity galaxies.

The Large Magellanic Cloud (LMC) is the brightest dwarf disk galaxy in the sky situated at ∼50 kpc, while the Small Magellanic Cloud (SMC) is an irregular dwarf galaxy, a little farther away at ∼60 kpc. With a lower metallicity and a higher gaseous mass, the SMC is less evolved than the LMC and has a different star formation history. These galaxies are also much smaller than the MW but are quite common in the universe, hence they can be used to better understand their impact on their external environments. It has been known for a long time that these galaxies are associated with large gaseous structures between them (the Magellanic Bridge) and trailing (the Magellanic Stream) or leading (the Leading Arm) them (e.g., Mathewson et al. 1974; Putman et al. 1998; Brüns et al. 2005). In addition, many FUV spectra of LMC, SMC, and Magellanic Bridge stars have revealed neutral, weakly, and highly ionized at blue-shifted high velocities relative to the Clouds and inconsistent with those of the MW or the Clouds (e.g., Savage & de Boer 1981; Welty et al. 1999; Lehner et al. 2001; Danforth et al. 2002; Hoopes et al. 2002; Howk et al. 2002; Lehner 2002). These components are signatures of high-velocity clouds (HVC) at v_LSR∼ 80–160 km s⁻¹, previously unseen in H i 21 cm emission observations. Subsequent deep observations toward the LMC have also revealed some of these HVCs in H i 21 cm emission (Staveley-Smith et al. 2003).

From the mid-1990s to 2006, one of the favored models predicted that the Magellanic Stream and Leading Arm had formed from a close interaction between the Clouds and the MW about 1.5 Gyr ago, where the gas was pulled from the SMC via tidal forces, while the Bridge was believed to have formed through a close encounter between the LMC and SMC some 200 Myr ago, where the gas and stars were pulled from the SMC via tidal interaction (Gardiner & Noguchi 1996). These models had already some shortcoming as they could not explain the low metallicity of the Bridge gas and stars (Rolleston et al. 1999; Lehner et al. 2008) or the absence of stars in the Stream. Even more dramatically, the Hubble Space Telescope (HST) measurements of the proper motions of the LMC and SMC have changed our view of these galaxies and their interaction with the MW (Kallivayalil et al. 2006a, 2006b; Piatek et al. 2008). Instead of being longtime companions of the MW, these galaxies may be just passing through for the first time, and the LMC/SMC may not be bound to the MW anymore, although, according to Shattow & Loeb (2009), there appears to be still enough uncertainties in the proper motions and circular velocities of the LMC and MW to allow these two galaxies to be bound. Nevertheless, these new measurements appear to rule out models that involve mainly tidal forces, as these forces are ineffective without multiple passages (Besla et al. 2007; Ruzicka et al. 2009). Ram pressure from a low-density, hot-ionized, and extended MW halo may provide the mean to remove a large amount of gas from the disk of the LMC (Mastropietro et al. 2005; Mastropietro 2008). Alternatively or additionally to ram pressure, stellar feedback from the LMC could provide another source of gas for the Magellanic Stream (Olano 2004; Nidever et al. 2008; Besla et al. 2007). But is there any evidence for a generalized stellar feedback beyond the thick disk of the LMC?

Some evidence was presented that the HVCs at v_LSR∼ 100–160 km s⁻¹ in front the LMC may be signatures of a generalized galactic outflow from the LMC (Staveley-Smith et al. 2003; Lehner & Howk 2007) rather than originating from the MW (de Boer et al. 1990). Shedding light on the origin of these HVCs is important for our general understanding of galaxy evolution, but appears even more pressing in the context of the new Clouds' proper motions and their implications. We have therefore embarked in an effort to characterize the HVCs³ that are between the MW and the Clouds. An unprecedented number of background targets is available behind the HVCs since over the last eight years the Far Ultraviolet Spectroscopic Explorer (FUSE) has collected over 230 spectra of early-type stars in the SMC and LMC. In contrast, the best studied HVC complex in the UV is complex C with only 11 sightlines (e.g., Collins et al. 2007). Because of the multi-phase nature of these features, the FUV spectrum is ideal to undertake such a study: (1) the FUV spectra are at high enough spectral resolution to estimate the column density and decipher high-velocity absorption relative to the MW or the Clouds, and (2) the FUV bandpass provides access to a wide range of gas phases, from the molecular clouds, to the neutral atoms, to low-, intermediate-, and highly ionized gas.

In this paper, we focus on the HVC toward the LMC, with three main aims: (1) to derive the metallicity of these HVCs, (2) to characterize the physical and ionization conditions, and (3) to describe the distribution of the properties (kinematics, ionization, etc...) of the HVCs. As the distance of these clouds cannot be bracketed to better than d < 50 kpc, the LMC distance, the metallicity is one of the most promising ways to determine the origin(s) of the HVCs (Savage & de Boer 1981; Wakker et al. 2001) knowing the metallicity of the LMC, SMC, and MW and their chemical evolution (e.g., Russell & Dopita 1992; Pagel & Tautvaisiene 1998).

In order to estimate the metallicity, we compare the column density of O i and H i. O i is the best metal proxy for H i since its ionization potential and charge-exchange reactions with hydrogen ensure that the ionization of H i and O i are strongly coupled in galactic environments (Jenkins et al. 2000; Lehner et al. 2003). Furthermore, oxygen is only mildly depleted into dust grains (≲−0.1 dex based on the interstellar O i estimate in the Galactic disk, e.g., Meyer et al. 1998; Jensen et al. 2005), implying that the O i/H i ratio is the best indicator of the metallicity, especially in partially ionized gas. The ionization level of the gas can be probed with the ratio Fe ii/O i. Indeed while Fe can be depleted into dust (hence the Fe/O is a priori expected to be subsolar), we show that this ratio is supersolar, strongly suggesting that the HVC gas is largely ionized (independently confirmed by the systematic presence of N ii and O vi absorption).

Our paper is organized as follows. In Section 2, we describe our sample, the instruments, and methods to derive the velocities and column densities for O i, Fe ii, and H i associated with the HVCs, as well as discuss the issues when comparing emission with absorption data (O i and Fe ii are seen in absorption in the FUV spectra of stars, while H i is obtained from deep H i 21 cm emission observations). The metallicity and relative-abundance estimates of the HVCs are presented in Section 2.4, while in Section 3 we present their properties (abundances, ionization, depletion, kinematics) and distribution. In Sections 4 and 5, we discuss the implication of our findings and summarize our results, respectively.

2.1. The LMC Sample

2.2. UV Data and Measurements

2.2.1. Instruments and Calibrations

The available resonance lines of O i and Fe ii are all present in the FUV bandpass observed with FUSE and HST/STIS. Descriptions of the FUSE instrument design and in-flight performance are found in Moos et al. (2000) and Sahnow et al. (2000), and see also Dixon et al. (2007) for an update. Information about STIS and its in-flight performance is given in the STIS Instrument Handbook (Kim Quijano et al. 2007). The spectra of 10 stars in our sample were obtained with both FUSE and STIS E140M (see Tables 1 and 2). The spectral resolution for the STIS E140M data is 7 km s⁻¹ with a detector pixel size of 3.5 km s⁻¹, while the FUSE resolution is about 20 km s⁻¹. We note that generally the FUSE spectra were obtained through the large aperture (LWRS: 30'' × 30'') to maximize throughput, except in very crowded regions where the medium aperture (MDRS: 4'' × 20'') was used to ensure that only the selected star was in the FUSE aperture.

Table 2. Relative Abundances in the HVC

Name	[Si ii/O i]	[Si ii/S ii]	[Fe ii/Si ii]
4U0532-664	+0.68 ± 0.03	> + 0.23	−0.06 ± 0.04
BRRG140	+0.33 ± 0.07	> − 0.48	−0.08 ± 0.09
Brey22a	+0.37 ± 0.06	−0.22 ± 0.10	−0.03 ± 0.06
LH54-425a	+0.32 ± 0.06	−0.28 ± 0.12	−0.18 ± 0.04
Sk–65°22	+0.70 ± 0.160.12	> − 0.12	−0.14 ± 0.08
Sk–67°101	+0.47 ± 0.06	+0.03 ± 0.12	−0.18 ± 0.06
Sk–67°104	> + 0.33	> − 0.05	> − 0.03
Sk–67°106	+0.23 ± 0.08	−0.15 ± 0.12	−0.06 ± 0.08
Sk–67°107	+0.38 ± 0.06	−0.10 ± 0.14	−0.04 ± 0.04
Sk–67°111a	+0.34 ± 0.06	−0.11 ± 0.09	−0.06 ± 0.04
Sk–67°166a	+0.38 ± 0.06	> − 0.08 ± 0.14	−0.03 ± 0.03
Sk–67°211	+0.50 ± 0.05	> − 0.44	−0.04 ± 0.03
Sk–69°246	+0.60 ± 0.03	> − 0.56	> − 0.30
Sk–71°45	+0.20 ± 0.03	−0.25 ± 0.10	−0.09 ± 0.03

Notes. All stars are from the STIS+FUSE sample except for those marked with a that are from FUSE only and where P ii is taken as a proxy for S ii. The adopted solar references are [Si/O]_☉ = −1.20, [Si/S]_☉ = +0.37, [Si/P]_☉ = +2.08, and [Fe/Si]_☉ = −0.07 from Lodders et al. (2009).

Download table as:  ASCIITypeset image

The STIS E140M data processing is fully described in Lehner & Howk (2007) for four stars. We applied the same data reduction for the remaining six stars and we refer the reader to this paper for further calibration information. The FUSE data were retrieved fully calibrated from the FUSE Magellanic Clouds Legacy Project archive. In order to maximize the S/N, we use the fully co-added spectra produced by CalFUSE version 3.2 (Dixon et al. 2007). We systematically cross-checked in the regions of interest that no spurious features contaminated the lines under investigation and that the resolution of the lines was not degraded. The S/N per resolution element are typically ≳5–40 in the FUSE spectra.

The absolute FUSE wavelength scale remains uncertain and needed to be corrected. The zero point was established by shifting the average FUSE velocity of Ar i and O i MW absorption to the observed average heliocentric velocity of the MW component in the H i 21 cm spectra. As Fe ii may be present in both the ionized and neutral gas, we compare the velocities of Fe ii at λ < 1080 Å and >1080 Å in order to ensure that the various segments of the detectors were aligned properly. The relative wavelength calibration for FUSE is often better than ±5 km s⁻¹, but there are still sometimes some stretches on small wavelength scales that can be as large as ±10 km s⁻¹.

The STIS spectra have an excellent wavelength calibration, with a velocity uncertainty of ∼1–3 km s⁻¹. The heliocentric velocity was then corrected to the dynamical local standard of rest frame (LSR) frame. We note that the MW O i λ1302 and H i 21 cm components align well, further ensuring that the MW H i component can be used to align the FUSE spectra.

2.2.2. Contamination and Continuum Placement

As we are interested in lines that are shifted by about 100–180 km s⁻¹ in the LSR frame from their rest position, contamination from other lines may be problematic. However, a positive aspect of the velocity shift for the O i lines is that the strong terrestrial airglow lines that may affect them, in particular O i λ1039, are not an issue here. For O i λ1302, there is no contamination by any other features. The O i transitions at λ < 1000 Å were hopelessly contaminated in about 55% of our sightlines (complicated stellar continuum, H₂ or H i contamination). For the remaining 45%, we were able to confidently use O i λ948 and/or λ936 in the LSR velocity interval [ + 90, + 180] km s⁻¹. The other weaker transitions were rarely detected or often could not be used owing to a too complex stellar continuum.

The HVC component of the O i λ1039 line can be partially contaminated by the LMC component of the 5–0 R(2) H₂λ1038.689 line. To estimate the contribution to the O i HVC profile from this line, we fitted a Gaussian profile to the 4–0R(2) H₂λ1051.498 line. This transition was picked because no other line contaminates it, the stellar continuum near this line can be easily modeled, and the strengths (fλ) of both transitions are about the same. Such a technique is further discussed and displayed in, e.g., Lehner et al. (2004). When the blended H₂ line was not too strong, O i λ1039 could be decontaminated and the resulting column density was in agreement within 1σ uncertainty with that measured in the (uncontaminated) O i λλ948, 936 lines, giving us further confidence in our method to deblend the O i λ1039 line.

Although Fe ii has many transitions in the FUSE wavelength range, the most useful was often the strongest transition available in the FUSE bandpass at 1144.938 Å, as it is uncontaminated by other interstellar lines and strong enough to be detected. However, in many cases, we were also to be able to use weaker transitions at lower wavelengths. For the stars observed with STIS, we also made use of the Fe ii λ1608 line.

In order to make our measurements, we first needed to model the stellar continuum near the absorption lines. The stellar continuum was often simple enough to be fitted with low-order Legendre polynomials (≲4). However, in more complicated cases, we fitted the stellar continuum with higher-order polynomials. In these cases, several continua were tested to be certain that the continuum error (see below) was robust (see Sembach & Savage 1992). In several cases, we have estimates from multiple O i and Fe ii lines that give similar column densities within 1σ uncertainty, implying that the continua are reliable. For the cases where we do not have information from other lines, we are confident in our modeling because similar types of stars are used, which have similar overall continuum behavior.

2.2.3. Column Densities and Velocities

To estimate the column density and velocities we adopted the apparent optical depth (AOD) method described in details by Savage & Sembach (1991). In this method, the absorption profiles are converted into apparent column densities per unit velocity N_a(v) = 3.768 × 10¹⁴ln [F_c/F_obs(v)]/(fλ) cm⁻² (km s⁻¹)⁻¹, where F_c is the continuum flux, F_obs(v) is the observed flux as a function of velocity, f is the oscillator strength of the absorption, and λ is in Å (atomic parameters were adopted from Morton 2003). The total column density was obtained by integrating over the absorption profile N = ∫N_a(v)dv. The values of the velocity, v, is obtained from v = ∫vN_a(v)dv/N_a (see Sembach & Savage 1992). The errors for the individual transitions include both statistical and continuum errors (see Sembach & Savage 1992). When no detection was observed, we estimated a 3σ upper limit following the method described by Lehner et al. (2008), where the 1σ equivalent width was estimated from integrating the continuum over a velocity interval Δv. If O i or Fe ii was detected, we adopted Δv from one of the detected species (this is valid as the profiles when both species are detected span a similar velocity extent). Otherwise, we set Δv = 50 km s⁻¹, a rough average based on sightlines where high-velocity absorption is detected.

Zoom In Zoom Out Reset image size

Savage & Sembach (1991) showed that the AOD method is adequate for data with b_line ≳ 0.25–0.50b_inst, where b_line is the intrinsic b value of the line and b_instr is the b value of the instrument. Since b≡ FWHM/1.667, for FUSE, b_inst ≈ 10 km s⁻¹. Therefore, we assume that a negligible fraction of the gas has b ≪ 2 km s⁻¹, or T ≪ 3800 K if thermal motions dominate the broadening. Our assumption is supported by the fact that the H i data show no narrow emission. As the peak apparent optical depth is generally less than 1, we expect that saturation is not important. Unresolved saturation can in fact be checked by comparing the apparent column densities estimated from the various transitions that have at least Δ[log(fλ)] ≳ 0.2. Within 1σ, the apparent column densities of O i λλ1302, 1039, 948, 936, on one hand, and Fe ii λ1608, 1144, 1143, 1125, 1121, 1096, on the other hand, generally agreed. The exceptions were for sightlines where the high-velocity absorption of O i λ1302 was such that τ_a ≫ 1 in the line center of the HVC component. Our final results for the HVCs with +90 ⩽ v_LSR ⩽ +175 km s⁻¹ are summarized in Table 1. The adopted column densities are a weighted average of the various estimates when more than one transition was available.

2.3. H i Data and Measurements

2.4. Estimates of [O i/H i] and [Fe ii/O i]

In the last column of Table 1, we list the ratio O i/H i and Fe ii/O i corrected for the solar oxygen abundance, where we use the standard square-bracket notation [Xⁱ/Y^j] ≡ log N(Xⁱ) − log N(Y^j) − log(X/Y)_☉. In this work, we adopted the solar oxygen abundance log(O/H)_☉ = −3.27 and iron abundance log(Fe/H)_☉ = −4.54, recently recommended by Lodders et al. (2009). We emphasize that the various profiles were integrated over a similar velocity range in order to compare the various species. Figure 2 shows that, within the errors, the average velocities of the O i and H i HVC profiles are aligned. For Fe ii and O i, there is also a good agreement in most cases. There is evidence in several O i and Fe ii profiles of at least two components separated by 20–40 km s⁻¹ (see Figure 1 and Figures 5 and 6 in Lehner & Howk 2007). In most cases, the velocity structures of the O i and Fe ii absorption profiles follow each other, but there are cases where they do not (even though the velocity intervals of the profiles are the same). These differences suggest changes in the ionization structure with velocity along a given sightline, which can cause discrepancies between O i and Fe ii average velocities. So in all cases, we compared the total H i and O i column densities, and apply the same rule when comparing Fe ii and O i.

Zoom In Zoom Out Reset image size

The errors on [Fe ii/O i] include both the statistical and continuum placement, while the errors on [O i/H i] include those and in addition the baseline uncertainties from the H i data. The errors do not include the uncertainties on solar abundances (see Section 4 for more details on the solar O abundance). In addition for [O i/H i], there is a "beam" error that arises from the comparison of data obtained from a 14'–16' beam compared to the pencil beam of the FUV observations. The N(H i) represents an "average" column density over a region near the line of sight toward the star. Wakker et al. (2001) studied this effect, and found that, for the HVCs, N(H i) measured with a half-degree radio beam can differ by up to a factor 2–3 (either way) from the value measured with a 10' or 1' beam. The distribution of N(H i; 36')/N(H i; 9') has a dispersion of about a factor 1.5, or 0.17 dex. Comparing N(H i) measured with a 9' beam or with a 1' beam or through Ly α absorption gives 0.06 dex. Using Wakker et al.'s results, Lehner et al. (2004) estimated a beam error of 0.10 dex for a 21' beam and 0.06 dex for a 12' beam. We therefore estimate that Parkes 14'–16' data have a beam error of about ±0.08 dex.

As there are fields of view with several stars within 14'–16', we can test the validity of the estimated beam error using the O i column densities. For example, near the field defined by R.A. =(4.945 ± 0.005)^h and decl. =(−66.415 ± 0.005)° intervals, there are nine estimates of N(O i) but only a single H i pointing. The mean column density of O i and dispersion are 10^14.80±0.06 cm⁻². The dispersion is consistent with a beam error of ±0.08 dex. In Figure 3, we show the spatial distribution of the O i column density projected on an Hα map of the LMC. While on large-scale structure there is variation in N(O i), on small scales (<15'), there is generally no evidence of variation other than noise fluctuation. An exception is toward BI272 and Sk–67°266 that are separated by 8.6' (and the H i pointings are apart by 9.1'), where the O i columns are different by at least 0.6 dex, but in that case the H i column densities are also likely very different (along Sk–67°266 no H i is detected but it is toward BI272). This may be an example where the edge of a cloud is probed. In summary, statistically, a beam error of 0.08 dex appears to be adequate for the present sample.

Zoom In Zoom Out Reset image size

3.1. Sky Covering Factor and Multi-phase Gas

3.2. Distribution of the Average Velocities

In Figure 4, we show the average LSR velocity the Fe ii HVC absorption for each sightlines overplotted on the Hα map of the LMC. In contrast to the column densities, a few trends can be noted in the distribution of the velocities with coordinates. In most regions, there is little variation in the velocities on small scales. The second trend is that the HVC can be separated in roughly two LSR velocity zones: at R.A. <5.2^h, the average LSR velocity is 140 ± 17 km s⁻¹, while at R.A. >5.2^h, it is 114 ± 14 km s⁻¹. Most of the HVCs at R.A. <5.2^h are roughly projected on the "W" arm of the LMC, while at larger R.A., they are projected on the main body of the LMC and regions with many supergiant shells.

Zoom In Zoom Out Reset image size

As the (Galactic) LSR frame may introduce some spurious velocity gradients, we also estimated the velocities in the LMC standard-of-rest (LMCSR) frame using the velocity vector of the LMC, v_LMC = (−86 ± −12, − 268 ± 11, 252 ± 16) km s⁻¹ (Kallivayalil et al. 2006a),

$$\begin{eqnarray*} &&v_{\rm LMCSR} = v_{\rm GSR}+86 \cos l \cos b + 268 \sin l \cos b - 252 \sin b, \end{eqnarray*}$$

where the Galactic standard of rest (GSR) frame is v_GSR = v_LSR + 220sin lcos b, and 220 km s⁻¹ corresponds to the velocity of the solar circular rotation around the MW.⁵ Figure 5 shows an anti-correlation between the LMCSR velocity and the R.A. Although there is some scatter in this figure, there is little doubt of a general increase in the |v_LMCSR| with increasing R.A. (or with the Galactic latitude), which is highlighted by the solid line, a linear fit to the data using a "robust" least absolute deviation method. No pattern is observed with the declination (see Figure 4).

Zoom In Zoom Out Reset image size

This apparent velocity gradient with the R.A. is unlikely to be caused by the underlying motion of gas in the LMC disk. Indeed, on average, for the sightlines included in our sample, the difference in the LMC disk average velocities (estimated using the H i emission data) between the two R.A. zones is small. Hence, this gradient may be more related to phenomena producing these HVCs rather than the disk–gas motions of the LMC (assuming these HVCs have their origin in the LMC, a hypothesis that appears valid as we discuss in Section 4). And indeed, the R.A. >5.2^h zone is where 30 Dor and most of the supergiant shells are (see Figure 2 in Kim et al. 1999), regions likely to produce stronger galactic outflows than in the spiral arm. In Section 4, we further discuss in more details the implications of this difference in the kinematics.

3.3. Absolute and Relative Abundances

In Figure 6, we show the distributions of [O i/H i] and [Fe ii/O i]. The histograms include all the measurements but not the limits. The median, mean, and 1σ dispersion for [O i/H i] are −0.53 and −0.51 ± ^0.12_0.16 (44 data points),⁶ respectively, and for [Fe ii/O i] are +0.30 and +0.33 ± ^0.14_0.21 (77 data points). With the Kaplan–Meier product limit estimator method that can be used to determine the mean of a set of data points containing limits (e.g., Feigelson & Nelson 1985; Isobe et al. 1986), we find [O i/H i] = −0.45, entirely consistent with the result above. For [O i/H i], the dispersion is similar to the statistical plus systematic uncertainties, and hence the scatter is likely mostly due to the measurements. However, for [Fe ii/O i], the errors in the individual sightlines are much smaller than the dispersion, implying that those are real variations due to changing conditions in the ionization and/or dust depletion (see below).

Zoom In Zoom Out Reset image size

In Figure 7, [O i/H i] and [Fe ii/O i] are compared to N(O i). For [O i/H i], we also show the lower limits (i.e., when H i is not detected). There is a lack of data at 14.8 ≲ log N(O i) ≲ 15.3 and [O i/H i] < −0.8 or [O i/H i]> − 0.2, indicating that neither very subsolar nor solar absolute abundances are found in these HVCs. We emphasize that if HVC gas with [O i/H i] < −0.8 were common, it should have been easily detected as N(H i)>10^18.9 cm⁻² whereas gas with [O i/H i]> − 0.1 would imply N(H i) < 10^18.5, typically smaller than detected. As we discuss in Section 4, the abundance of oxygen is consistent with the O abundance in the LMC, suggesting that the origin of these HVCs is gas that has been ejected from the LMC. We did not find any discernable patterns between [O i/H i] and R.A. or decl. or velocities.

Zoom In Zoom Out Reset image size

In order to correctly interpret the [Fe ii/O i] ratio, we need to know if chemical evolution or dust depletion (besides ionization) plays a role when comparing Fe and O (e.g., Savage & Sembach 1996; Jenkins 2004; Sofia 2004; Howk et al. 2006). In the MW gas, refractory elements like Si and Fe are more depleted from the gas phase than S or P (or O) if dust grains are present. In Galactic halo-like clouds, we would expect that [Fe/O] ≃ −0.4 if ionization was unimportant. Chemical evolution of the gas can also affect the relative abundances. For example, while the bulk of O is produced in massive stars (>12 M_☉), only about a 1/3 of Fe is produced in the remnants of these massive stars; the bulk of Fe is thought to be made in Type Ia supernovae (e.g., Russell & Dopita 1992). Hence, if some gas was recently enriched by a supernova, [Fe/O] would be expected to be subsolar. And indeed, supernova SN 1987A produced a ratio [Fe/O] ∼ −0.6 dex solar (Dopita 1990), unfortunately a level quite similar to dust depletion, implying a degeneracy in the possible origins of the [Fe/O] deficiency. The α-elements (e.g., S, Si), on the other hand, are believed to follow the chemical evolution of O. So if Si is depleted relative to (undepleted) S, this should result mainly from a dust depletion effect (assuming the ionization effect is negligible).

Besides the Fe ii/O i ratio, only toward a few sightlines can we do an analysis of the relative abundances of singly ionized species because in most cases the S/N is not high enough to detect species other than Fe ii and O i (and saturated N ii) in the FUSE spectra. However, the 10 STIS sightlines allow us to access stronger transitions, such as Fe ii λ1608 and Si ii λ1526, as well as the undepleted element S, via S ii λλ1253, 1259. In Table 2, we summarize the relative abundances [Si ii/O i], [Si ii/S ii] (or [Si ii/P ii] for the 4 FUSE sightlines), and [Fe ii/Si ii] estimated in the HVCs along the 10 STIS sightlines plus the four FUSE stars with the highest S/N spectra. As Si and O follow the same chemical evolution, only ionization and depletion should affect [Si ii/O i]. If a mild dust depletion exists in the HVC gas, −0.2 ≲ [Si ii/O i] ≲ 0 would be expected. As we observe +0.2 ≲ [Si ii/O i] ≲ +0.7 (with a mean of [Si ii/O i] = +0.48 ± ^0.15_0.25), this implies that ionization is extremely important and its effect dominates over depletion. Si and S also follow the same chemical evolution, so again only depletion and ionization may affect this result. However, the ionization effect should be milder as two singly ionized species are compared. In a few cases, there seems to be a mild depletion of Si at the level of 0.1–0.2 dex based on the comparison of Si and S. However, toward the X-ray binary 4U0532-664, [Si ii/S ii]> + 0.23 (it has also one of the highest [Si ii/O i] ratios) suggests that ionization effects cannot be neglected even when only singly ionized species are compared. Hence, the intrinsic dust depletion of Si may be larger. Finally, [Fe ii/Si ii] varies between a solar value and −0.3 dex solar. This is somewhat consistent with a Galactic halo-like depletion pattern observed in the MW (e.g., Savage & Sembach 1996; Howk et al. 1999). We note that according to models of diffuse ionized gas (Sembach et al. 2000), Fe ii, P ii, and S ii can be more affected by ionization than Si ii. Unfortunately, neither Fe iii nor S iii can be estimated reliably and accurately. It is, however, beyond doubt that ionization is important as illustrated independently by the high-detection rates of N ii and O vi (see Section 3.1). Following the conclusion from [Si ii/S ii], the mean value [Fe ii/O i] = +0.33 ± ^0.14_0.21 implies that the HVC gas is very ionized toward every directions probed in our sample, >50% on average but could be as large as >90% toward many directions. Even if ionization effects dominate the relative abundances, there is evidence of dust depletion in these HVCs in view of the subsolar [Si ii/S ii] ratio toward several sigthlines. In view of the Si depletion, it is extremely likely that Fe is depleted as well, and nucleosynthesis must play only a little role (if any) in the Fe defiency in these HVCs.

We now offer some comments on the implications of the kinematics, ionization, and abundances of the HVCs for their origin. The first implication is that the HVCs across the LMC are extremely likely to be an HVC complex, i.e., they have the same origin. We draw this conclusion from both kinematics, metallicity, and ionization arguments. In the previous section, we saw that there is no strong evidence for a large variation in the metallicities of the HVCs over the face of the LMC (see Figure 6). While the HVCs cover a large interval of LSR velocities between +90 and +175 km s⁻¹, there seems to be a gradient in the velocity with the R.A. across the LMC. This contrasts from earlier studies that use smaller samples and found no pattern (see Savage & de Boer 1981; Danforth et al. 2002). Along all sightlines, the HVCs are largely photoionized (>50–80%) and also collisionally ionized (based on the high incidence of high-velocity O vi absorption). These results strongly suggest that these HVCs are part of a general complex of dominantly ionized gas moving at high velocity relative to both the LMC and the MW. The H i 21 cm contours shown in Figure 8 of S03 further support this conclusion. The neutral part of this HVC complex covers the sky from decl. ∼−74° to −65° and R.A. ∼4^h to 6^h based on the combined O i and H i observations, and hence the complex covers a large portion of the LMC area. In our sample, the H i column densities range from <10^18.4 (<10^16.6 cm⁻² if O i is used as a proxy for H i after correcting for the metallicity) to about 10^19.2 cm⁻², and therefore these HVCs are possible local analogs of the Lyman limit systems seen at higher redshifts in the spectra of quasi-stellar objects (QSOs).

Recent works have presented some indirect evidence that the HVCs seen toward the LMC are likely related to outflows from the LMC. Staveley-Smith et al. (2003) found that the peak H i column densities are seen projected onto H i voids (such as supergiant shells, e.g., LMC 3) within the LMC disk and furthermore are connected to the disk of the LMC with spatial and kinematics bridges in position-velocity plots, suggesting that they are ejection of matter from the LMC disk. Lehner & Howk (2007) found a systematic similarity in the O vi/C iv ratio between the LMC and HVC components (which were quite dissimilar from the IVC and MW components), suggesting again that the HVC has its origins in the LMC. In the localized region of 30 Dor, Redman et al. (2003) presented high-resolution emission of [O iii] showing high-velocity features at similar velocities that the HVCs studied here, which again can be traced back to the 30 Dor regions in position-velocity plots, and hence likely trace the superwind of 30 Dor. Using a Gaussian decomposition of the H i 21 cm, Nidever et al. (2008) argued that supergiant shells in the southeast H i overdensity region of the LMC are blowing out gas and energy. Below we argue that our results combined with these findings leave little doubt that the LMC has indeed strong outflows of matter and energy from stellar feedback occurring within its disk.

The metallicity of the HVC and the distance to the HVC are the two most direct and important quantities for deciphering the origin of a given HVC (e.g., Wakker et al. 2001; Zech et al. 2008). The distance of this HVC complex is likely to remain unknown, leaving us only the metallicity information. In Section 3.3, we show that the mean oxygen abundance for the HVCs in our sample is [O/H] = −0.51 ± ^0.12_0.16. By definition, [O/H] is relative to the solar O abundance, so this abundance rules out the origin of this gas from outflows within the solar circle. However, is it possible the gas could have originated from the outer regions of the Galactic disk, since the abundances of the MW seem to vary as a function of the Galactocentric radius, R_g. Several studies suggest indeed a slight gradient in O/H from −0.01 to about −0.07 dex kpc⁻¹ (e.g., Daflon & Cunha 2004; Esteban et al. 2005; Rudolph et al. 2006, and references therein). As there is still much uncertainty in the gradient value itself (and the value could vary with R_g), we follow Cescutti et al. (2007) who divided the data in various bins of R_g and find that the lowest [O/H] has a mean and a deviation of −0.2 ± 0.2 dex at Rg>9.5 kpc (at smaller R_g, [O/H]> − 0.1), which is still notably larger than the metallicity of the HVC complex (even when depletion is taken into account, see below). Hence, if the HVC had its origin from a Galactic fountain phenomenon, its metallicity would require to have originated in the outer regions of the MW where a large cluster of massive stars with much subsolar metallicities than found elsewhere in the MW would need to be present. This seems unlikely and to the best of our knowledge there is no observational evidence of such a low-metallicity cluster of massive stars in the MW disk. Therefore, the metallicity of HVC does not support a MW origin and combined with the other observational findings discussed above and below it seems safe to reject the MW hypothesis as the origin for the HVC complex.

The next two more likely candidates for the origin of the HVC complex are the LMC and SMC. In order to compare the abundances, we first need to know the present-day O abundances of these galaxies. As there is currently no reliable estimate of the interstellar oxygen abundance in absorption in these galaxies, we use for comparison the stellar and nebular abundances. The recent results from the analysis of 133 and 80 B-type stars in the LMC and SMC yield a present-day mean oxygen abundance of −3.66 ± 0.13 and −4.01 ± 0.21 dex in the LMC and SMC, respectively (Hunter et al. 2009). These values are quite consistent with the O abundances estimated in H ii regions within the SMC and LMC (Kurt & Dufour 1998). The average of the values derived in the H ii regions and B-type stars yields −3.64 ± 0.10 dex for the LMC and −4.00 ± 0.16 dex for the SMC. In the last 20 years, the solar abundance of oxygen has changed by nearly 0.2 dex (e.g., Anders & Grevesse 1989; Grevesse et al. 1996; Asplund et al. 2006). Here, we adopted the most recent recommended O abundance of −3.27 dex (Lodders et al. 2009) (an intermediate value between the low and high estimates), which is an average value from the recent estimates of Caffau et al. (2008), Ludwig & Steffen (2008), and Meléndez & Asplund (2008). As we use it as a systematic reference, the uncertainty in the solar O abundance does not affect our conclusions. The above LMC and SMC abundances relative to solar abundances are −0.40 ± 0.10 and −0.73 ± 0.16 dex solar, respectively. These abundances are consistent with those measured from the LMC and SMC Cepheids (Luck et al. 1998) and there is no evidence for a large-abundance variation in these galaxies.

Although there is some overlap with the SMC abundance, the mean and the dispersion of O/H of the HVC is closer to the canonical LMC value. Furthermore, several lower limits are well above the SMC value and approach the LMC O abundance. This strongly favors an LMC origin for the HVC complex. As we compare O i with H i, ionization does not affect this ratio. However, dust depletion of oxygen could affect HVC O/H abundance, albeit by no more than 0.1 dex. Indeed, in the Galactic disk the evidence for oxygen dust depletion is found to be ≲0.1 dex with the adopted solar oxygen abundance (Meyer et al. 1998; André et al. 2003; Cartledge et al. 2004; Jensen et al. 2005). It is currently unclear if the O depletion is different in the more diffuse gas than in the Galactic disk and it is often assumed to be the same (e.g., Welty et al. 1999) and in any case it is likely less or equal to the Galactic disk value based on the depletion patterns from other elements. In Section 3.3, we argue that there is some evidence of dust depletion in these HVCs. Hence, if oxygen is depleted by about 0.05–0.1 dex, the average HVC O abundances listed above would converge to the canonical LMC O abundance. A scenario where a pre-existing low metallicity gas may have reduced the overall abundance of the HVC seems less compelling because this would require that the HVC had mixed with the (unknown) low-metallicity component and had time to fully mix as there is no evidence of extremely low-metallicity gas in these HVCs.

Therefore, the HVC metallicity is consistent with an origin mostly from the LMC itself in form of outflowing matter. An alternative scenario would be that the HVC is the result of a hydrodynamical interaction between the LMC and the hot halo of the MW, which, via ram pressure, could remove gas from the LMC disk. In this case, the HVC would also be expected to have an abundance similar to the present-day LMC. However, such models rule out any LSR velocities at 100–180 km s⁻¹ for the ram pressure gas near the center of the LMC; only 40° away from the LMC center, ram pressure models produce gas at LSR velocities 100–160 km s⁻¹ (see Figure 12 in Mastropietro et al. 2005).

The multi-phase structure of the HVC complex, the large fraction of ionized gas, and the presence of dust grains are all also consistent with an energetic galactic outflow. Galactic winds are indeed known to be extremely multi-phase and to entrain dust grains far away from the Galactic disk (e.g., Veilleux et al. 2005; Strickland & Heckman 2007; Howk 2009). The presence of a diffuse hot gas and shock heated and ionized gas would be expected in galactic outflows (e.g., Veilleux et al. 2005). The high ions (O vi, C iv, Si iv) are signatures of such collisionally ionized gas and suggest the presence of a diffuse hot (>10⁶ K) phase interacting with the cooler (10⁴ K) neutral and ionized gas (see discussion in Lehner & Howk 2007). This HVC complex also shares similar ionization properties to those of the complex multi-phase structures of the HVCs toward the globular cluster M 5 in our MW, HVCs believed also to probe energetic events that occur in the central region of the MW (Zech et al. 2008).

Finally, the observed velocity gradient of the HVC across the LMC can also be simply explained in the context of outflows from the LMC. Figure 2 in Kim et al. (1999) shows a larger concentration of supergiant shells at R.A. >5.2^h than at lower R.A. The combined effect of these supergiant shells must produce stronger outflows in the central region of the LMC than in one of its arm. The high velocities of the gas observed toward 30 Dor (Redman et al. 2003) are also quite consistent with the higher velocities observed at R.A. >5.2^h, strongly suggesting that 30 Dor (unsurprisingly) participates to the overall outflow from the LMC.

In order to know if those outflows can escape the gravitational potential of the LMC, the rotation velocity of the LMC needs to be known. Unfortunately, the rotation curves using different methods and tracers allow a large interval of rotational velocities, from about 70 to 107 km s⁻¹ (e.g., Alves & Nelson 2000; Olsen & Massey 2007; van der Marel et al. 2008), which would imply escape velocities from the LMC gravitational potential from ∼100 to ∼150 km s⁻¹. Results based on the LMC proper motions yield an even higher v_rot ≃ 120 km s⁻¹ (Piatek et al. 2008), and hence higher escape velocity. In many cases, the absolute LMCSR velocities of the HVC range from 100 to 170 km s⁻¹ (see Figure 5). The tangential velocities and inclination of the outflows are not known, but in view of such large-radial velocities, it seems reasonable to conclude that the LMC may have a galactic wind as long as v_rot ≲ 100 km s⁻¹. Furthermore, if the HVC gas does not escape the LMC potential, there should be signatures of infalling material. We searched for high positive velocity absorption relative to the LMC disk and only found less than a handful sightlines with Fe ii absorption at +50–70 km s⁻¹ relative to the LMC disk, possible signatures of infalling gas. The two most compelling sightlines are shown and discussed in Howk et al. (2002). Therefore, as these authors concluded with their smaller sample, there is no strong evidence of infalling material in these directions, and hence the HVC complex does not seem to be the result of an outflow associated with a galactic fountain and hence is more likely a galactic wind. At 150 km s⁻¹, ignoring possible drag forces and initial higher velocities, it would take about 30 to 65 Myr to reach a distance of 5 to 10 kpc, respectively. Therefore, a small galaxy (1–3 × 10¹⁰ M_☉) such as the LMC is capable of producing outflows that may pollute its surroundings beyond its gravitational potential.

The mass of the neutral gas in the outflow can be roughly estimated via,

$$\begin{eqnarray*} &&M^{\rm neut}_{\rm HVC} = f_{\rm cov} \frac{\overline{N}({\hbox{\rm O\,{\sc i}}})}{({\rm O/H})_{\rm HVC}} \mu _{\rm H} m_{\rm H} A_{\rm HVC}, \end{eqnarray*}$$

where the covering factor is f_cov = 0.7 from our detection rate, the average O i column density is $\overline{N}({\hbox{\rm O\,{\sc i}}}) \simeq 10^{14.9}$ cm⁻², m_H = 1.673 × 10⁻²⁴ g is the atomic mass of hydrogen, and μ_H ≃ 1.3 corrects for the presence of helium. The HVC seen in absorption covers the LMC disk, so we take the area of the HVC, A_HVC, to be similar to the disk of the LMC assuming a diameter of 7.3 kpc, the diameter of the H i disk (Kim et al. 1998). This yields M^neut_HVC ∼ 1.5 × 10⁶ M_☉. Using the H i data with log N(H i)>17.4, we estimate an H i mass of 0.4 × 10⁶ M_☉ for the HVC gas, which gives after correcting for helium 0.5 × 10⁶ M_☉. The main difference between these values is the difference in the covering factors between absorption and emission data. The absorption method also likely overestimates the mass because it assumes the same H i column density while the H i emission estimate misses the lower column density gas, and therefore we just adopt the average value of ∼10⁶ M_☉. These estimates also assume that the HVC complex is at 50 kpc. If the HVC is 10 kpc away from the LMC toward us, and assuming no projection correction, the mass would be 0.6 × 10⁶ M_☉. The neutral mass is, however, only the tip of the iceberg as the HVC is largely photoionized and collisionally ionized. Including the photoionized high-velocity gas probed by Fe ii and N ii, the mass of the HVC increases by at least a factor 2–3 (see above). Using the measurements of the O vi absorption in the HVC component (Lehner & Howk 2007), the average O vi column density (based on four measurements) is 10^13.9 cm⁻², which would yield a H ii column density for the collisionally ionized gas of N_coll(H ii) ≈ 10^19.3 cm⁻² assuming that the metallicity is the same as the neutral HVC and the O vi ion fraction is 0.2. The latter value cannot be much larger than 0.2, but could be much smaller if the temperature of the gas is far (either way) from the peak temperature of O vi in collisionally ionized equilibrium (Gnat & Sternberg 2007), hence N_coll(H ii)>10^19.3 cm⁻². Therefore, the total mass of the outflow is more likely to be >(0.5–1) × 10⁷ M_☉. While this mass is much smaller than the total masses of the Stream and Bridge (Brüns et al. 2005; Lehner et al. 2008), it is similar to or possibly even larger than other large HVCs such as Complex C (Wakker 2001; Wakker et al. 2007; Thom et al. 2008). In view of our results, it is plausible that Complex C might be the remnant of an outflow from a hidden dwarf galaxy that is being accreted by the MW. The same fate could await the LMC HVC complex.

Using the FUSE LMC database, we search for absorption at high-LSR velocity (+90 ⩽ v_LSR ⩽ +175 km s⁻¹) of neutral, ionized, and molecular species in the spectra of early-type stars located in the LMC. We use sensitive Parkes observations of H i 21 cm for comparison, and for 10 stars, we complement the FUSE data with STIS E140M data. Our final sample consists of 139 lines of sight, 44 of them showing both O i absorption and H i emission at high-LSR velocities, an unprecedented number of sightlines where the metallicity can be estimated in a large HVC complex. Our main results for the HVCs toward the LMC are as follows.

• 1.  
  The detection rates of high-velocity O i λ1039, Fe ii λ1144, and N ii λ1083 absorption toward our LMC stars are about 70%. The detection rate of O vi λ1031 could be even higher, ∼90%. While the quality of the data changes from sightline to sightline, the high detection rates imply that the HVCs cover a substantial area of the LMC. In contrast, the H i 21 cm emission spectra along the same sightline only reveal the HVC at a 32% rate.
• 2.  
  Since the neutral, ionized, and highly ionized HVCs are found at similar velocities along a given sightline, the HVC has a complex multi-phase structure. However, we did not find any evidence of cold (T < 10³ K) or molecular gas, which is not surprising in view of the low H i column density of the HVCs (log N(H i) ≲ 19.2).
• 3.  
  We find a mean and dispersion of [Si ii/O i] = +0.48 ± ^0.15_0.25 (14 sightlines). Since Si and O have similar chemical evolution and are only mildly depleted in diffuse gas, this implies that the gas is >50–80% ionized. Using a larger sample, we find [Fe ii/O i] = +0.33 ± ^0.14_0.21 (77 sightlines), implying that the gas is largely ionized toward every direction probed, >50% on average but as large as >90% toward many sightlines, consistent with the high detection rate of N ii. The high occurrence of O vi (and substantial column densities, see Howk et al. 2002; Lehner & Howk 2007) implies that the HVC is also largely (possibly dominantly) highly ionized.
• 4.  
  Using the ratios of Si ii/S ii and Si ii/Fe ii, we conclude that there is some evidence of mild elemental depletion, which implies the presence of dust grains in these HVCs.
• 5.  
  From the sample where limits are excluded, we derive the mean metallicity of the HVC [O/H] = −0.51 ± ^0.12_0.16 (or O/H = 0.31 ± 0.10 solar, 44 sightlines). Many lower limits imply [O/H]> − 0.6, and including those, the Kaplan–Meier estimator gives [O/H] = −0.45. There is no evidence of intrinsic variation in the abundance of oxygen, suggesting that these HVCs are part of a single complex that has the same origin.
• 6.  
  Based on stellar and nebular abundances, the present-day O abundances of the LMC and SMC are −0.4 and −0.7 dex solar. Therefore, the HVC metallicity toward the LMC is more consistent with the LMC O abundance. Dust depletion may affect the O interstellar abundance by a small amount (<0.1 dex). In that case, the (small) correction would increase the metallicity of the HVC, and hence [O/H] of the HVC complex would converge to the present-day canonical LMC O abundance.
• 7.  
  We do not find strong dependence in the relative or absolute abundances with any other quantities (e.g., Galactic coordinates, H i column density, or velocity). However, the average velocities of the HVCs in the LMCSR or LSR frames show a gradient with the right ascension (or Galactic latitude). Interestingly, most of the supergiant shells believed to produce strong outflows in galaxies are situated at R.A. >5.2^h, where the deviation velocities of the HVCs from the average velocities of the LMC disk are the highest.
• 8.  
  We estimate that the mass of the neutral gas of the HVC complex is ∼(0.5–1) × 10⁶ M_☉ if the HVC complex is at 40–50 kpc. Including the photoionized and collisionally ionized gas, the total mass is likely >0.5 × 10⁷ M_☉. This is comparable to or larger than the total mass of Complex C.
• 9.  
  All of our observational results are consistent with a generalized galactic outflow from the LMC as the origin of the HVC complex at LSR velocity 90–175 km s⁻¹ distributed across the LMC, including the metallicity and kinematics of the HVCs as well as earlier findings from radio, optical and UV observations. This origin also naturally explains why the HVC is principally ionized, with both signatures of photoionization and collisional ionization. Dust grains are found in other known major galactic winds, and there is evidence of dust in this HVC complex as well. However, we do not find any strong evidence of infalling material, therefore the HVC gas is likely to escape the LMC potential, suggesting that the LMC has a galactic wind. Ultimately, this HVC complex may be accreted by the MW and serve as additional fuel for future star formation within the MW. Based on these findings, it is possible that other large HVC complexes found in the MW halo could be the result of outflows from nearby low-mass galaxies (galaxies that could be now hidden by the MW disk).

We are grateful to Naomi McClure-Griffiths for making the GASS data available to us. Support for this research was made by NASA through grant NNX08AJ51G. This research has made use of the NASA Astrophysics Data System Abstract Service and the Centre de Données de Strasbourg (CDS).