Annu. Rev. Astron. Astrophys. 2005. 43: 769-826 Copyright © 2005 by Annual Reviews. All rights reserved

4. LOCAL CENSUS OF GALACTIC WINDS IN STAR-FORMING GALAXIES

In the next two sections, we present a census of galactic winds. Each newly discovered outflow source provides different insights into the wind process, and often reveals new and complex behavior. Therefore, it is necessary to summarize what is known of the wind phenomenon in terms of basic observational parameters (e.g. morphology, kinematics, energetics). The subject of galactic winds is largely in its infancy; therefore a discussion of the phenomenology is warranted here. We start by summarising the search techniques used to find outflow sources, the limitations of each technique and the inferred detection rates.

4.1. Search Techniques & Observational Limitations

The wind phenomenon is likely to be sufficiently complex that a multiwavelength approach is necessary in order to probe all gas phases. Line and continuum emission from warm, hot, and relativistic plasma have been used with great success to identify or confirm GWs. Choice emission lines are H, [N II] 6548, 6583, [S II] 6716, 6731, [O III] 5007, and [O II] 3727 in the optical, and Pa, Br, and H₂ 2.122 µm in the near-IR. Ly 1216 should be avoided because resonant scattering and selective dust absorption severely distort its emission profile and intensity (e.g., Tenorio-Tagle et al. 1999; Keel 2005). CXO and the XMM-Newton Observatory have provided powerful new tools for complementary searches in soft X rays, whereas facilities such as the upgraded Very Large Array (VLA) will continue efficient radio studies of winds.

Enhanced emission along the minor axis of the host galaxy does not necessarily imply a GW; other origins include tidal interaction and ram-pressure stripping by the ICM. Kinematic signatures of extraplanar, non-gravitational effects are needed to be certain. Double-peaked emission-line profiles attributable to an expanding, shocked shell are often evident in wind galaxies. Emission-line ratios typical of shock excitation (e.g., [N II] 6583 / H 1; [S II] 6716, 6731 / H 0.5) indicate winds, but these diagnostics favor powerful ones with large outflow velocities and a weak starburst radiation field (see Section 4.8 for more detail).

Selection effects hamper detection of GWs: "Since deceleration is what powers all of the emission process, the brightness distribution due to a particular emission process will be dominated by the sites of optimum local deceleration (corresponding to an optical density and temperature contrast) within the range of decelerations that can give rise to that process" (Braun 1985). Observable winds are likely those that are only slightly more pressured than the galaxy potential. Highly energetic winds (>> 10⁵⁸ erg or >> 10⁴⁴ erg s^-1 from Equation [15]) may escape undetected, although this will depend on such details as how energy is injected at the source, and the amount of gas in and around the host galaxy. Winds << 10⁴¹ erg s^-1 would have little observable impact on the ISM of an external galaxy.

The need to distinguish wind-related emission from the background favors detection of highly inclined outflows in edge-on galaxies. This is true for most of the well-known starburst-driven winds. In a few less inclined galaxies (e.g., NGC 253, NGC 1808), outflows were first suspected on the basis of absorbing dust filaments seen against the galaxy body (e.g., Phillips 1993). In NGC 253, extended H filaments are projected onto disk H II regions, requiring detailed 3D kinematics to separate the different components. Similar blending occurs at X-ray and radio wavelengths because compact sources are commonly associated with the inner disk. In edge-on systems, kinematical deprojection leads to large uncertainties in wind energetics. Other complications include uncertain dereddening at the base of the outflow, relating the post-shock to the pre-shock velocities, and high density bias introduced by the density-squared dependence of the emission measure.

Absorption-line techniques have been used with great success to search for the unambiguous wind signature of blueshifted absorbing material in front of the continuum source. This method favors detection of winds in face-on systems and therefore complements the emission technique. The equivalent width and profile of the absorption line are used to estimate the amount of outflowing material along the line of sight and to determine its projected kinematics. For an unsaturated line, the column density of the absorbing material scales linearly with the equivalent width of the line and therefore is arguably a better probe of the whole range of density in the wind than the line or continuum emission. For saturated lines, the change in equivalent width reflects primarily a change in the velocity dispersion of the absorbing clouds, covering fraction or both.

UV interstellar absorption lines are useful. Low-ionization lines such as Si II 1260, O I + Si II 1303, C II 1334, Si II 1526, Fe II 1608, and Al II 1670 are particularly well suited to avoid confusion with possible stellar photospheric features of higher ionization (e.g., C III 1176, O IV 1343, S V 1501) and stellar wind features with P Cygni-type profiles (e.g., N V 1238, 1242, Si IV 1393, 1402, C IV 1548, 1550, He II 1640). A few key optical lines have also been used in successful wind searches in z 0.5 galaxies: Na I D 5890, 5896 and K I 7665, 7699. Extension to the far-UV domain with the Far Ultraviolet Spectroscopic Explorer (FUSE) spacecraft has recently allowed the use of the O VI 1032, 1038 lines to probe the coronal-phase (T = a few × 10⁵ K) gas in GWs. Such measurements constrain the amount of radiative cooling (see Section 4.5).

4.2. Detection Rate

There are now estimates of the frequency of wind occurrence in nearby star-forming galaxies for most wavebands. Detailed studies of extended optical line emission around various starburst and quiescent galaxies reveal a clear trend between starburst strength and the presence of kinematically confirmed winds or extraplanar diffuse ionized gas (e.g., Heckman, Armus, & Miley 1990; Hunter, Hawley, & Gallagher 1993; Pildis, Bregman, & Schombert 1994; Lehnert & Heckman 1995, 1996; Marlowe et al. 1995; Veilleux et al. 1995; Rand 1996; Hunter & Gallagher 1997; Kim, Veilleux, & Sanders 1998; Martin 1998; Miller & Veilleux 2003a, 2003b; Rossa & Dettmar 2000, 2003a, 2003b; although see Meurer 2004). Complementary results from systematic searches for blueshifted optical absorption lines (e.g., Heckman et al. 2000; Rupke, Veilleux, & Sanders 2002, 2005a, 2005b; Schwartz & Martin 2004; Martin 2005), complex UV emission and absorption lines (e.g., Lequeux et al. 1995; Kunth et al. 1998; Heckman et al. 1998; Gonzalez Delgado et al. 1998), extended emission in soft X rays (e.g., Read, Ponman, & Strickland 1997; Dahlem, Weaver, & Heckman 1998; Pietsch et al. 2000; McDowell et al. 2003; Ehle et al. 2004; Huo et al. 2004; Strickland et al. 2004a, 2004b; Ehle 2005; Grimes et al. 2005) and at radio wavelengths (e.g., Hummel, Beck, & Dettmar 1991; Irwin, English, & Sorathia 1999; Irwin, Saikia, & English 2000; Dahlem et al. 2001) all confirm this dependence on starburst strength.

A key indicator is 60-to-100 µm IRAS color, S₆₀ / S₁₀₀, which tracks dust temperature in star-forming regions. "Warm" 60-to-100 µm IRAS color - S₆₀ / S₁₀₀ 0.4 - is often used to classify galaxies as starbursts. Warm galaxies generally have large IR luminosities ( 10^10.5 L), excesses (L_IR / L_opt 2), and galaxy-averaged far-IR (FIR) surface brightnesses (L_FIR / R₂₅² 2 × 10⁴⁰ erg s^-1 kpc^-2; e.g., Rossa & Dettmar 2003a). This is not surprising because the SFR scales with the IR luminosity according to equation (5), and the dust temperature is expected to scale with the UV energy density. The conditions on the IR luminosity and surface brightness in warm galaxies translate into SFR 5 M yr^-1 and SFR / R₂₅² > 0.001 M yr^-1 kpc^-2. Almost all of these galaxies have extraplanar ionized gas, and most also host a wind. More than 75% of ultraluminous IR galaxies (ULIRGs) with IR luminosities > 10¹² L (S₆₀ / S₁₀₀ 0.5) have winds (e.g., Rupke et al. 2002, 2005b; Martin 2005). The degree of nucleation of star formation activity increases with SFR, so it is not surprising to detect so many winds among ULIRGs (where SFR 100 M yr^-1 and R_* < 1 kpc).

4.3. Morphology

The outflows in most star-forming galaxies have a bipolar distribution perpendicular to the disk. Opening angles are 2 10° - 45° near the base, increasing to ~ 45° - 100° above the disk as expected from simulations (Section 2.4). But the detailed gas distribution is often complex, shows large galaxy-to-galaxy variations, and depends on the specific gas phase. In optical line emission, outflow structures range from the classic egg-shaped nuclear superbubble of NGC 3079 ¹ (Fig. 3; Ford et al. 1986; Veilleux et al. 1994; Cecil et al. 2001) to the bipolar double-loop morphology of Arp 220 (Heckman, Armus, & Miley 1987), the biconical structure of M82 (Fig. 4; Bland & Tully 1988; Shopbell & Bland-Hawthorn 1998; Ohyama et al. 2002) and NGC 1482 (Fig. 5a; Veilleux & Rupke 2002), and the frothy and filamentary morphology of the outflow in the dwarf NGC 1569 (e.g., Martin, Kobulnicky, & Heckman 2002). The line-emitting structures are often limb-brightened, indicating that much of the optically emitting gas resides on the surface of largely hollow structures. In the few objects examined with the Hubble Space Telescope (HST) (~ 0.1"), the line-emitting gas is resolved into complexes of clumps and filaments (e.g., M82) or streams of filamentary strands and droplets (e.g., NGC 3079) with volume filling factors f 10^-3. The morphology and kinematics (discussed in Section 4.4) of this gas suggest that it originated in the cool disk and became entrained in the outflow.

Figure 3. NGC 3079 imaged with HST (red for H+[N II], green for I-band) and CXO (blue). (a) large-scale emission across 15×5 kpc. Numerous H filaments rise above the disk. Note the V-shaped wind pattern extending in X rays from nucleus; for clarity, we have suppressed the diffuse X-ray emission across the superbubble, where it is generally clumped (see [c]). (b) The 1 × 1.2 kpc superbubble in H + [N II] emission, with log-scaled intensities. It is composed of 4 vertical towers of twisted filaments; the towers have strikingly similar morphologies. (c) Close-up of the wind-swept, circumnuclear region. Note how X-ray emission (blue) clumps along the optical filaments of the superbubble at the limit of CXO's resolution. A prominent dust filament at left drops out of the wind.

Figure 4. M82, imaged by the WIYN telescope in H (magenta) and HST in BVI continuum colors (courtesy Smith, Gallagher, & Westmoquette). Several of the largest scale filaments trace all the way back to super-starclusters embedded in the disk.

Figure 5. (a) Starburst galaxy NGC 1482 imaged in H (red), [N II] (green), and CXO (blue). The bar is 3 kpc long. For this and the other three panels, N is up and E is left, and intensities in all bands are log-scaled. (b) Excitation map of NGC 1482 showing the shock-excited ([N II] 6583 / H 1 in black), biconical structure due to the wind above and below the star-forming disk ([N II] / H < 1 in orange). The scale is the same as in (a), and the white crosses indicate the locations of two bright star-forming regions in the disk. (c) Seyfert galaxy NGC 4388 imaged in H (red), [N II] (green), and CXO (blue). The bar is 8 kpc long. The kinematics of the extraplanar gas 8 kpc north of the nucleus are dominated by the AGN-driven outflow. The long H trail extending further to the north-east may be due to ram-pressure stripping by the ICM in the Virgo cluster. (d) Central region of the Circinus galaxy imaged in [O III] (blue) and blueshifted (between -150 and 0 km s-1) H (red). The bar is 1 kpc long. The [O III] emission traces the ionization cone and conical wind produced by the AGN, whereas the H emission east of the nucleus is dominated by the circumnuclear starburst. Spectacular filaments and bow shocks are seen on scales of ~ 500 - 900 pc.

These structures vary in vertical extent from ~ 1 kpc to 20 kpc (see Veilleux et al. 2003 and references therein). Large filaments seemingly unrelated to the nuclear structure sometimes appear in deeper exposures; e.g., faint X-shaped filaments extend >8 kpc from the nucleus of NGC 3079 (Heckman et al. 1990; Veilleux et al. 1994) and rise ~ 4 kpc from the galactic plane but connect to the inner (R 1.5 kpc) galactic disk rather than to the nucleus itself. Cecil et al. (2001) suggest that these filaments have the shape expected (Schiano 1985) for the contact discontinuity/ISM associated with lateral stagnation of the wind in the galaxy thick disk/halo. The lateral extent of the wind is necessarily much smaller in the disk, where one expects to find a "ring shock." This feature (see Section 2.4) has been detected in a few objects, based on line emission and warm molecular gas emission (e.g., NGC 253: Sugai, Davies, & Ward 2003; see below).

A growing set of ~ 1"-resolution CXO data show that the bright, soft X ray and H filaments in these winds have strikingly similar patterns on both small and large scales (~ 0.01 - 10 kpc; e.g., Strickland et al. 2000 , 2002, 2004a; Cecil, Bland-Hawthorn, & Veilleux 2002a; McDowell et al. 2003). This tight optical-line/X-ray match seems to arise from cool disk gas that has been driven by the wind, with X rays being emitted from upstream, standoff bow shocks or by conductive cooling at the cloud/wind interfaces. This is not always the case for the fainter soft X-ray emission. For instance, the X-ray emission near the X-shaped filaments of NGC 3079 (Fig. 3a) is not significantly edge brightened, suggesting a partially filled volume of warm gas within the shocked wind, not a shell of conductively heated gas (Cecil, Bland-Hawthorn, & Veilleux 2002; see also Huo et al. 2004; Ott, Walter, & Brinks 2005). In a few objects, absorption by foreground neutral ISM also affects the distribution of the soft X-ray emission (e.g., NGC 253: Strickland et al. 2000; M82: Stevens, Read, & Bravo-Guerrero 2003). Intrinsically hard, diffuse X-ray emission has been detected in a few wind galaxies. In M82, this gas is more nucleated but less filamentary than the soft X-ray emission; it probably traces the hot, high-pressure wind fluid (Griffiths et al. 2000; Stevens et al. 2003; see Section 4.9 for more detail).

Several wind galaxies have large radio halos (e.g., M82: Z 5 kpc, Seaquist & Odegard 1991; NGC 253: ~ 9 kpc, Carilli et al. 1992; NGC 3079: ~ 11 kpc, Irwin & Saikia 2003; NGC 4631: ~ 9 kpc, Hummel & Dettmar 1990). Galaxies with thick ionized disks (S₆₀ / S₁₀₀ > 0.4) also tend to show extraplanar synchrotron radio emission (e.g., Dahlem et al. 2001). Given the significant polarization of the radio emission and the lack of point-by-point correspondence with the X-ray and optical line emission, non-thermal synchrotron from magnetized (B few × 10 µG), relativistic electrons is the favored explanation for most of the radio emission. The pattern of magnetic field lines in NGC 3079 (Cecil et al. 2001) suggests that the relativistic electrons are produced in the starbursting disk and then advected from the disk by the wind. Some emission may also come from electrons accelerated locally in internal wind shocks. Steepening of the spectral index of the radio continuum emission in M82 (Seaquist & Odegard 1991), NGC 253 (Carilli et al. 1992), NGC 4631 (Ekers & Sancisi 1977; Hummel 1991) and NGC 891 (Allen, Sancisi, & Baldwin 1978; Hummel 1991) indicates energy losses of the electrons on their way from the starburst, either from synchrotron losses or inverse Compton scattering of the relativistic electrons against the IR photons produced in the nuclear region. In some objects, the relativistic component of the wind appears to decouple from the thermal component beyond the H-emitting structures (e.g., NGC 3079, Duric & Seaquist 1988), perhaps participating in a "cosmic ray wind" rather than a thermal wind (Breitschwerdt & Schmutzler 1999).

Unambiguous evidence for entrained neutral gas has been detected from dwarf galaxies (e.g., Puche et al. 1992; Stewart et al. 2000; Schwartz & Martin 2004) to ULIRGs (e.g., Heckman et al. 2000; Rupke et al. 2002, 2005a, 2005b; Martin 2005). This gas often extends up to several kiloparsec, but morphological constraints are sparse. Detailed long-slit spectra show some degree of correlation with the warm ionized gas (Section 4.4). Kinematically disturbed molecular gas has also been detected in a few GWs. The best case for molecular gas entrained in a GW is in M82, where a detailed kinematical decomposition of the CO gas into wind and disk components reveals a wide-angle (opening angle 2 110°) pattern related loosely to the outflow seen at other wavelengths (Stark & Carlson 1984; Seaquist & Clark 2001; Walter, Weiss, & Scoville 2002). A narrow, shock-excited SiO chimney extends ~ 500 pc above the disk (Garcia-Burillo et al. 2001), and is also found in NGC 253 (Garcia-Burillo et al. 2000). Shocked H₂ gas is detected at the base of the outflow of NGC 253 (Sugai et al. 2003), consistent with the cloud-crushing model of Cowie, McKee, & Ostriker (1981) and Ohyama, Yoshida, & Takata (2003) where C-type shocks with V_shock 40 km s^-1 are compressing the star-forming molecular disk.

GW structures sometimes tilt relative to the minor axis of the host, and are asymmetric to the nucleus. Tilts and asymmetries near the starbursts (M82: Shopbell & Bland-Hawthorn 1998; Galaxy: Bland-Hawthorn & Cohen 2003; NGC 253: Sugai et al. 2003) probably reflect asymmetries in the starbursting population and in the density distribution of the cool, star-forming disk. Asymmetries on large scales may be due to density fluctuations in the halo of the host galaxy, and therefore can be a probe. The IGM through which the galaxy is moving may also influence the morphology of the wind structure on large scales (e.g., radio halo in M82, Seaquist & Odegard 1991).

4.4. Kinematics

Spectra of the warm ionized component in GWs shows double-peaked emission-line profiles with a split ranging from a few 10's of km s^-1 in some dwarf galaxies (e.g., Marlowe et al. 1995; Martin 1998) to about 1500 km s^-1 in NGC 3079 (Filippenko & Sargent 1992; Veilleux et al. 1994). Constraints on the phase-space distribution of gas within the outflows are necessary to deproject observed velocities. The dense 2D spatial coverage of Fabry-Perot and IF spectrometers is ideal, but usually only narrow-band imaging and long-slit spectroscopy are available to constrain the outflow geometry. The observed kinematic patterns indicate that most warm gas lies on the surface of expanding bubbles/ellipsoids or flows along the walls of conical structures. The substantial broadening often seen in individual kinematic components indicates that the flow is not purely laminar or that the cone walls are composites of distinct filaments with a range of velocities.

The most detailed studies of outflow kinematics can be compared quantitatively with the hydro simulations described in Section 2.4. The deprojected outflow velocity in open-ended conical winds often increases with radius as expected for entrained gas in a free-flowing wind (e.g., Murray et al. 2005); examples are M82 (Shopbell & Bland-Hawthorn 1998) and NGC 3079. Gas near the top of the partially ruptured bubble of NGC 3079 is entrained in a mushroom vortex (Cecil et al. 2001), as predicted theoretically (e.g., Suchkov et al. 1994). In these open-ended winds, there is a clear correlation between outflow velocity and gas phase temperature. For example, cool molecular gas is ejected out of M82 at a maximum deprojected outflow velocity of ~ 230 km s^-1 (Shen & Lo 1995; Walter et al. 2002), well below the inferred velocities of the warm ionized gas (525 - 655 km s^-1; Shopbell & Bland-Hawthorn 1998). Coronal gas traced by the O VI 1032 absorption line in NGC 1705 is more blueshifted than the neutral gas, whereas the warm photoionized gas appears to have intermediate velocities (Heckman et al. 2001 a). Smaller outflow velocities in the neutral gas relative to the ionized components are also often seen in LIRGs and ULIRGs (Rupke et al. 2005b). These systematic kinematic variations with phase temperature are consistent with entrainment of gas clouds by a hot wind if the warmer phase has smaller column densities than the cool gas, perhaps as a result of cloud erosion (Section 2.4).

Entrainment of disk material is supported by other evidence, including the detection of rotation in the outflow gas (e.g., M82: Shopbell & Bland-Hawthorn 1998, Greve 2004; NGC 3079: Veilleux et al. 1994; NGC 1482: Veilleux & Rupke 2002) and the field reversal across the NGC 3079 radio lobe that suggests the return of entrained, magnetized material to the disk (Cecil et al. 2001).

Wind kinematics are poorly constrained. This fluid, being hot and tenuous (Section 4.9 and Equation [8]), is hard to detect, so current X-ray instruments do not constrain its velocity. A lower limit follows from the expected terminal velocity of an adiabatic wind at the measured X-ray temperature T_X: V_W² c_s² = k T_X / µ m_p, where c_s is the isothermal sound speed of the hot phase, µ m_p, the mean mass per particle, and = 2.5 - 5.0, a scale factor that depends on the fraction of thermal energy that is radiated (e.g., Appendix B in Efstathiou 2000). This is a lower bound because it accounts only for thermal energy and neglects possibly substantial (e.g., Strickland & Stevens 2000) bulk motion. Measured X-ray temperatures in GWs are quite uniform (~ 0.2 - 2 × 10⁷ K; Martin 1999; Heckman et al. 2000; Strickland et al. 2004a), and imply V_W 500 - 900 km s^-1. These are well below V_∞ from equation (16) unless / << 1. Note that the X-ray temperature is weighted by the emission measure, so it probes only dense shocked material between disk and wind gas. The measured X-ray temperature is therefore likely a lower limit to the wind temperature, and wind velocities are also lower limits.

Our knowledge of the kinematics of neutral gas in GWs has improved considerably thanks to Na I D absorption-line surveys (e.g., Heckman et al. 2000; Rupke et al. 2002, 2005a, 2005b; Schwartz & Martin 2004; Martin 2005). The profile of the interstellar Na I absorption feature is fit with multiple components to determine the bulk and turbulent velocities of the neutral gas. The resulting distribution of Na I D velocities is skewed to the blue (relative to systemic); this is interpreted as outflow. Line full widths at half maximum (FWHMs) average ~ 275 km s^-1 in large starbursts, much larger than the thermal velocity dispersion of warm neutral gas. This broadening comes from the superposition of distinct kinematic components with several radial velocities, as seen in the warm gas phase. The projected "maximum" velocities in the outflowing components (equal to the centroid velocity plus one-half the velocity width) average 300 - 400 km s^-1, and attain ~ 600 km s^-1 (although 1100 km s^-1 is seen in F10378+1108). Figure 6a plots the maximum outflow velocities against host-galaxy mass. A linear least-squares fit suggests that outflow and circular velocities correlate positively, but this correlation is mainly due to the dwarf galaxies (Martin 2005, Rupke et al. 2005b).

Figure 6. Maximum Na I D absorption-line outflow velocities as a function of (a) circular velocities, and (b) star formation rates. Red skeletal stars are star-forming dwarfs from Schwartz & Martin (2004) and red open stars are infrared-selected starbursts from Rupke et al. (2005a, 2005b). Filled blue circles and filled black squares are Seyfert 2s and Seyfert 1s from Rupke et al. (2005c). The dashed line in (a) represents the escape velocity for a singular isothermal sphere with rmax / r = 10, whereas the dashed lines in (b) are characteristic velocities of ram-pressure accelerated clouds (Murray et al. 2005) for column densities of 1020 cm-2 (top line) and 1021 cm-2.

Figure 6b compares the outflow velocities with the host galaxy SFR (L_IR). There is some indication of a trend of increasing outflow velocities with increasing SFR, particularly when data from Schwartz & Martin (2004) on dwarf galaxies are included. Figure 6b shows that cloud entrainment in a wind (Equation [A5] in Murray et al. 2005) is able to reproduce these velocities, except perhaps for F10378+1108 where the large outflow velocity may also require radiation pressure. (Martin 2005 has also argued for radiation-pressure driving in other objects). UV absorption-line measurements appear to confirm the positive correlation between outflow velocities and SFR's (Heckman 2004).

4.5. Mass Outflow Rates & Energetics

The multiphase nature of GWs greatly complicates the task of estimating the outflow masses and energies. A multiwavelength approach that considers all phases is essential. Given the density-squared dependence of the emission measure, diagnostics that rely on line or continuum emission favor the densest material, yet this may be only a relatively small fraction of the total mass and energy. This is especially relevant for the wind fluid, which is expected to dominate the energetics of the outflow, but contributes very little to the emission (Strickland & Stevens 2000). Corrections must be applied for the volume filling factor, f. Constraints on f can be derived from estimates on the volume of the emitting material and assumptions about its emissivity (temperature). Measurements that rely on absorbing column densities are less subject to density inhomogeneities, but require a strong source of background light and therefore are limited to the brightest parts of the starburst. Assumptions about filling factor and outflow extent must then be made to estimate the mass of outflowing material and the energetics.

After deprojecting observed velocities, the dynamical timescales of starburst-driven outflows (t_dyn R / V_out) range from 0.1 to 10 Myr. The mass of warm ionized gas inferred to take part in the outflow is ~ 10⁵ - 10⁶ M in dwarf galaxies and ~ 10⁵ - 10⁷ M in powerful starbursts. The dynamical timescales yield mass outflow rates ranging from 0.1 M yr^-1 to 10 M yr^-1, with a trend for the rate to increase with increasing SFR. A similar trend may exist between star formation activity and the amount of extraplanar gas in non-starburst galaxies (e.g., Dettmar 1992; Rand, Kulkarni, & Hester 1992; Miller & Veilleux 2003a; Rossa & Dettmar 2003a).

Bulk plus "turbulent" KEs inferred from optical emission-line spectra span a broad range: e.g., NGC 1482: 2 × 10⁵³ erg (Veilleux & Rupke 2002); NGC 3079: ~ 10⁵⁴ erg (Cecil et al. 2001); M82: ~ 2 × 10⁵⁵ erg (Shopbell & Bland-Hawthorn 1998). These are typically several times lower than thermal energies derived from X-ray data. Luminosities of galactic X-ray halos scale roughly with IR luminosities or disk SFRs (Strickland et al. 2004a, 2004b), but the corresponding X-ray cooling rates amount to 10% of the SN heating rates in these objects. Cooling by the optical line-emitting material is even smaller.

Results from Na I D studies suggest that GWs entrain considerable neutral material, ~ 10⁴ - 10⁷ M (0.001 - 1.0 M yr^-1) in dwarfs and ~ 10⁸ - 10¹⁰ M (~ 10 - 1000 M yr^-1) in ULIRGs. These rates generally exceed the mass injection rate from SNe (Equation [1]). However, they are highly uncertain because: (1) the geometry of the neutral outflow is poorly known (here the simple model of a mass-conserving free wind from Rupke et al. 2005b is used with the assumption that gas lies in a thin shell at a radius ~ 5 kpc for the large starbursts; and at smaller radii for the dwarfs; the estimated masses scale linearly with this radius); (2) depletion of Na onto grains affects the strength of the Na I D absorption line (here we assume a depletion factor ~ 9); (3) the ionization potential of Na is low (5.139 eV) so considerable Na is ionized even when hydrogen is neutral; this must be accounted for when calculating H⁰ masses (the ionization correction is assumed to be close to the Galactic value of ~ 10).

Assuming that these estimates of neutral gas masses are correct, the ratios of to the global SFRs span / SFR 0.01 - 10, consistent with those found by Martin (1999) for the warm ionized medium of ten galaxies. Parameter , the mass entrainment efficiency, shows no obvious dependence on SFR except perhaps for a decreasing trend at high SFR ( 10 - 100 M yr^-1). The inferred KE increases with increasing SFR: ~ 10⁵⁰ - 10⁵⁴ erg among dwarfs but ~ 10⁵⁶ - 10⁵⁹ erg among LIRGs and ULIRGs; corresponding power outputs are ~ 10³⁶ - 10³⁹ and 10⁴¹ - 10⁴⁴ erg s^-1, respectively. Such energies and powers exceed those of the outflowing warm ionized gas, and imply thermalization efficiencies 10%. Contrary to some expectations (e.g., Silk 2003), the trend with SFR flattens among ULIRGs, perhaps due to the complete evacuation of the gas in the wind's path, a common neutral gas terminal velocity for LIRGs and ULIRGs, and/or a decrease in the efficiency of thermalization of the SN energy.

CO studies of GWs are very important because so much mass is required to see molecular gas that its detection dramatically increases the inferred energies. A good illustration is M82: Walter et al. (2002) deduced from CO observations that > 3 × 10⁸ M of H₂ is involved in the outflow, and its KE ~ 10⁵⁵ erg becomes comparable to the KE in the warm filaments. We discussed in Section 3.2 the GW of the Milky Way, where cold material seems to dominate the KE of the outflow. Detailed millimeter studies of a representative set of GWs will be needed to confirm the dynamical importance of the molecular gas component.

The latest addition to the mass and energy budgets of GWs is the coronal (T ~ 10⁵ K) phase traced by O VI. Dynamical information on this component is currently sketchy, outflowing gas being detected in absorption in only two (dwarf) galaxies so far: NGC 1705 (v_out 100 km s^-1; Sahu & Blades 1997; Heckman & Leitherer 1997; Heckman et al. 2001) and NGC 625 (v_out 30 km s^-1; Cannon et al. 2005). The mass of coronal gas derived from these data is uncertain because of possible line saturation, ionization corrections, and assumptions on the gas geometry. In NGC 1569, Heckman et al. (2001a) estimate a mass of ~ 6 × 10⁵ M and KE of ~ 3 × 10⁵² erg in the coronal phase. These are only ~ 1% of the values in the warm ionized phase, so the coronal component is unimportant dynamically. But what about its radiative losses? The O VI 1032, 1038 lines are key because they produce ~ 30% of the coronal cooling. O VI emission was detected in NGC 4631 (Otte et al. 2003), but not in NGC 1705 (Heckman et al. 2001a), M82 (Hoopes et al. 2003), and NGC 891 (Otte et al. 2003). These measurements limit the radiative cooling of coronal gas to 10 - 20% of the SN heating rate in these objects. To within a factor of 2, this is identical to the cooling rate of the X-ray-emitting gas.

4.6. Escape Fraction

The fraction of outflowing material that can escape the gravitational potential of the host galaxy is an important quantity but is difficult to determine accurately. It should be recalled that within the context of CDM theory, the virial radius of the dark halo is ~ 250h₇₀^-1 kpc for an L_* galaxy, with the IGM beyond.

Can winds reach the IGM? The main uncertainty comes from our lack of constraints on halo drag. Silich & Tenorio-Tagle (2001) have argued that drag may severely restrict a wind and limit its escape fraction. Drag by a dense halo or tidal debris may be especially important in high-luminosity starbursts because many are triggered by galaxy interactions (e.g., Veilleux, Kim, & Sanders 2002b). Large H I halos may also prevent dwarf galaxies from venting a wind (e.g., NGC 4449: Summers et al. 2003). Conversely, Strickland et al. (2004b) suggested that high-latitude hot gas above the disk can actually help the outflow to escape.

A popular way to estimate the escape fraction is to compare the outflow velocity with the local escape velocity derived from a gravitational model of the host galaxy. This is often a simple, truncated isothermal sphere. If truncated at r_max, then the escape velocity v_esc at radius r is related to the rotation speed v_c and r_max by v_esc(r) = 2^1/2 v_c [1 + ln(r_max / r)]^1/2. The escape velocity is not sensitive to the exact value of r_max / r [e.g., for r_max / r = 10 - 100, v_esc (2.6 - 3.3) × v_c]. The curve in Figure 6a is for r_max / r = 10. If halo drag is tiny, material that exceeds v_esc may escape into the IGM. With this simple assumption, Rupke et al. (2005b) find that ~ 5 - 10% of the neutral material in starburst-driven winds will escape. This may only be a lower limit: much of the gas above v_esc may have already mixed with the IGM and would be invisible in Na I D absorption.

Given the correlation between outflow velocity and gas-phase temperature mentioned in Section 4.4, the escape fraction is surely larger for warm and hot phases. Indeed, warm gas in several dwarfs (including possibly M82) exceeds escape (e.g., Martin 1998; Devine & Bally 1999; Lehnert, Heckman, & Weaver 1999). Similarly, velocities derived in Section 4.4 from X-ray temperatures, V_w 500 - 900 km s^-1, exceed escape for galaxies with v_c 130 - 300 km s^-1. Recall that these are lower limits to the wind terminal velocities, so galaxies with v_c 130 km s^-1 may not retain hot, metal-enriched material (Martin 1999). As discussed in Section 7.1.2, this galaxy-mass dependence on metal retention makes definite predictions on the effective yield which appears to have been confirmed by observations.

4.7. Energy Injection Zone

HST images of nearby starbursts show that the star-forming regions can be extremely complex and luminous super-star clusters (SSCs), each having thousands of young ( 50 Myr) stars within a half-light radius 10 pc. But the clustered part of star formation accounts for only ~ 20% of the integrated UV light in nearby optically-selected starbursts (Meurer et al. 1995). Most of the star formation is distributed diffusely in these objects. It is unclear which mode of star formation (clustered vs. diffuse) drives starburst winds. Both seem to in the outflow from the dwarf galaxy NGC 1569, whose wind seems to emanate from the entire stellar disk rather than just the central 100 pc near the SSCs (Martin et al. 2002). In M82, Shopbell & Bland-Hawthorn (1998) deduced from the diameter of the inner outflow cone a relatively large energy injection zone, ~ 400 pc. But chimneys at the base of the wind suggest localized venting of hot gas (Wills et al. 1999; see also Fig. 4). Heckman et al. (1990) used the gas pressure profile (Equation [10]) to deduce relatively large (a few hundred parsec) vertical sizes for the energy injection zones in several edge-on galaxies. But one should be wary of the possibility that the [S II] 6716, 6731 emission lines that they used to derive density can be severely contamined by foreground or background disk emission; Veilleux et al. (1994) showed that this occurs in NGC 3079. Additionally, these may measure only vertical pressure profiles in the foreground/background galaxy disk, not the pressure profile in the central starburst.

Given the dependence of thermalization efficiency on density (Section 2.1.1), in high-density environments diffuse star formation in the low-density ISM, not clustered star formation, may drive the wind most efficiently. Hence, Chevalier & Fransson (2001) have warned about using the radio continuum as a proxy for mechanical luminosity in starburst galaxies. Here, radio emission comes predominantly from SN remnants that are expanding in the dense ( 10³ - 10⁴ H atoms cm^-3) interclump medium of molecular clouds. These radiate most of their mechanical energy, so they do not drive GWs. On the other hand, SNe that detonate in a lower-density medium heat gas and drive winds, but are largely invisible in the radio. The large gas densities at the centers of ULIRGs (~ 10^4-5 cm^-3) (Downes & Solomon 1998; Bryant & Scoville 1999; Sakamoto et al. 1999) seem to imply a large volume filling factor of molecular clouds and possibly strong radiative losses, perhaps explaining the small mass entrainment efficiencies in ULIRGs (Section 4.5).

4.8. Excitation Properties & Evolution

Superbubbles and GWs are time-dependent, dynamic systems. How an outflow evolves is tied directly to the history of its energy source, i.e. the star formation history of the host (starburst versus quiescent star formation). As a GW evolves spatially (Section Section 2.2 - 2.3), the gaseous excitation of any entrained material also changes because of two processes: (1) photoionization by hot OB stars in the starburst and by the hot X-ray-emitting wind, and (2) radiative shocks formed at the interface of the fast-moving wind and the slow-moving disk ISM. As expected from radiative shock models (Dopita & Sutherland 1995, 1996), the importance of shock-excited line emission relative to photoionization by the OB stars in the starburst appears to scale with the velocity of the outflowing gas (e.g., Lehnert & Heckman 1996; Veilleux et al. 2003; Rupke et al. 2005b). NGC 3079 is an extreme example of a shock-excited wind nebula, with outflow velocity of order 1500 km s^-1 (e.g., Filippenko & Sargent 1992; Veilleux et al. 1994; Cecil et al. 2001). The excitation contrast between the star-forming host and the shock-excited wind material can in principle be exploited to search efficiently for GWs (Fig. 5b; Veilleux & Rupke 2002).

The dynamical state of an outflow also affects its gaseous excitation: Compact, pre-blowout superbubbles are less porous to ionizing radiation from hot stars than fragmented, post-blowout superbubbles or GWs. Such fine-tuning between outflow velocities and self-shielding of the starburst may explain why GWs dominated by OB-star photoionization are rare. "Inverted" ionization cones, where stellar photoionization dominates over shocks, have been detected in M82. This wind has cleared channels beyond the two prominent starbursting knots, allowing ionizing radiation to escape to large radii (Shopbell & Bland-Hawthorn 1998).

4.9. Wind Fluid

There is little direct evidence for the wind fluid in starbursts, because it is tenuous and hot, and therefore a poor X-ray radiator (e.g., Strickland & Stevens 2000; Strickland et al. 2000). The best evidence for it is in M82 (Griffiths et al. 2000; Stevens et al. 2003), where the hottest gas is 75 pc from the center, and has T ~ 4 × 10⁷ K and pressure P / k 10⁹ K cm^-3 if the X rays are mostly thermal (Griffiths et al. 2000). Then the hot fluid is overpressured relative to the disk ISM and drives the large-scale wind.

Hard (1 - 6 keV) X-ray emission is resolved in the dwarf galaxy NGC 1569 (Martin et al. 2002) and its temperature exceeds escape velocity (~ 80 - 110 km s^-1; Martin 1999). Interestingly, the spectral softening of X rays with radius that is expected from adiabatically cooling winds (e.g., Chevalier & Clegg 1985) is not seen, perhaps because of large mass loading.

The metal content of the X-ray-emitting gas may constrain mass loading, although significant theoretical and observational uncertainties remain. As discussed in Section 2.1.1, SN explosions are expected to dominate when GWs develop. So, the metallicity of the wind fluid is regulated by SNe yields. Unfortunately, the oxygen and iron yields of massive stars are only known to an accuracy of ~ 2 - 3 because of uncertainties in the critical ¹²C(, )¹⁶O reaction rate, and on the mass limit above which stars do not contribute to the yield (considerable reimplosion of heavy elements may affect stars of 30 M; e.g., Woosley & Weaver 1995). Determining the metallicity of X-ray-emitting gas observationally is notoriously difficult because of uncertainties in the atomic physics and because of the degeneracies inherent in fitting multi-component spectral models to data of low spectral resolution. The X-ray-emitting gas is a multi-phase medium with a range of temperatures, densities, and absolute/relative metal abundances, possibly located behind cool absorbing material of unknown metallicity and column density. The problem is therefore under-constrained and one must assume many unknowns. Presently there seems to be evidence that the /Fe ratio is slightly super-solar in the inner wind of M82 (Stevens et al. 2003; Strickland et al. 2004a) and in the wind filaments of NGC 1569 (Martin et al. 2002), as expected if stellar ejecta from SNe II contribute to the wind fluid. If confirmed, these modest /Fe enrichments would further support the idea that mass loading by disk material contributes significantly to the X-ray emission (recall the kinematic evidence for disk mass loading in Section 4.4). Martin et al. (2002) compared these measurements with predictions from SNe models of Woosley & Weaver (1995) to estimate a mass-loading factor of ~ 10 in NGC 1569.

4.10. Dust

Evidence is mounting that dust is often entrained in GWs (dust in the wind of our Galaxy was discussed in Section 3.2). Far-IR maps of a few GWs show extended cold dust emission along the galaxy minor axis, suggesting entrainment (e.g., Hughes, Robson, & Gear 1990; Hughes, Gear, & Robson 1994; Alton, Davies, & Bianchi 1999; Radovich, Kahanpää, & Lemke 2001). Color maps reveal elevated dust filaments in several GWs (e.g., NGC 1808: Phillips 1993; M82: Ichikawa et al. 1994; NGC 253: Sofue, Wakamatsu, & Malin 1994; NGC 3079: Cecil et al. 2001). In a few systems, including M82, extended polarized emission along the outflow axis indicates dust (e.g., Schmidt, Angel, & Cromwell 1976; Scarrott et al. 1991, 1993; Alton et al. 1994; Draper et al. 1995). Extended red emission, a broad emission band commonly seen in Galactic reflection nebulae, exists in the halo of M82 (Perrin, Darbon, & Sivan 1995; Gordon, Witt, & Friedmann 1998). Far-UV maps of M82 made with the Ultraviolet Imaging Telescope reveal a UV-bright southern cone that is consistent with scattering by dust in the wind (e.g., Marcum et al. 2001); recent Galaxy Evolution Explorer (GALEX) data confirm this conclusion (Hoopes et al. 2005). Other support for dusty outflows comes from the strong correlation between nuclear color excesses, E(B - V), and the equivalent widths of blueshifted low-ionization lines in star-forming galaxies at low redshifts (e.g., Armus, Heckman, & Miley 1989; Veilleux et al. 1995b; Heckman et al. 2000).

Optical measurements can estimate dust masses if one knows the geometry of the dust filaments and the amount of foreground starlight and forward scattering. The dust mass in an outflow can best be estimated from far-IR data, despite uncertainties associated with the disk/halo decomposition and assumptions about the dust temperature distribution and emissivity law. Alton et al. (1999) estimate that ~ 2 - 10 × 10⁶ M of dust is outflowing from M82. Radovich et al. (2001) used the same technique to derive 0.5 - 3 × 10⁶ M of outflow into the halo of NGC 253. The dynamical times yield dust outflow rates of ~ 1 M yr^-1.

The fate of this dust is uncertain; there are no direct constraints on its kinematics in outflows. However, the short sputtering timescale for silicate/graphite dust in GWs (e.g., Popescu et al. 2000; see also Aguirre 1999) suggests that grains of diameter 0.3 µm would not survive long if in direct contact with the wind. Moreover, there does not seem to be a tight spatial correlation between the extraplanar warm ionized medium of starbursts and quiescent galaxies and the dust filaments (e.g., Cecil et al. 2001; Rossa et al. 2004). Most likely, the dust is embedded in the neutral or molecular component of the outflow and shares its kinematics. Assuming a Galactic gas-to-dust ratio, the neutral gas outflow rates in LIRGs and ULIRGs (Rupke et al. 2005b) translate into dust outflow rates of ~ 0.1 - 10 M yr^-1. Given the kinematics of the neutral gas and assuming no halo drag, ~ 5-10% of the entrained dust may escape the host galaxy. Wind-driven ejection of dust from galaxies may feed the reservoir of intergalactic dust (e.g., Stickel et al. 1998, 2002), although tidal interactions and ram-pressure stripping are also efficient conveyors of dust into the ICM of rich clusters.

¹ NGC 3079 has both a nuclear starburst and an AGN, but from the morphology and kinematics of the line emitting gas, Cecil et al. (2001) conclude that this outflow is starburst, not AGN, powered. Back.