Far-Ultraviolet Spectra of Starburst Galaxies: Stellar Population and the Kinematics of the Interstellar Medium

The far-ultraviolet spectra of the four starburst galaxies NGC 6090, Mrk 66, Mrk 1267, and IRAS 0833+6517 were observed with the Hopkins Ultraviolet Telescope during the Astro-2 mission. Additional data were obtained for IRAS 0833+6517 with the Goddard High-Resolution Spectrograph (GHRS) on the Hubble Space Telescope. We analyze the observations in terms of the stellar content and the kinematics of the interstellar medium, and we discuss the implications of these results for the interpretation of the ultraviolet spectra of high-redshift galaxies. Evolutionary synthesis models are used to constrain the star formation history from the absolute ultraviolet flux and from the stellar components of the absorption lines of Si IV and C IV, and from the far-ultraviolet lines O VI + Lyβ + C II. The spectral energy distributions from these models are used as inputs for the photoionization code CLOUDY to predict [O III]/Hβ and column densities of the cool component of the ionized gas (Si II and C II) associated with the H II region. The results indicate continuous star formation during about 9 Myr, or, alternatively, bursts with ages between 3 and 6 Myr, and a mass consistent with that estimated from the Hα flux. Evidence for significant dilution by a field star population is found in all galaxies except NGC 6090. Most of the interstellar absorption lines are saturated. Their equivalent widths indicate a large velocity dispersion in the gas. Other evidence for large-scale motions of the interstellar gas comes from blueshifts of several hundred km s-1 with respect to the systemic velocity in the interstellar lines of NGC 6090, Mrk 66, and IRAS 0833+6517. These outflows are most likely driven by the starburst. Lyα was detected in emission in three of the galaxies (NGC 6090, Mrk 66, and IRAS 0833+6517). The dereddened Lyα/Hβ ratio in IRAS 0833+6517 is close to the recombination value, indicating that extinction is more important than multiple resonant scattering effects. However, the GHRS spectrum of IRAS 0833+6517 clearly shows that the emission profile of Lyα is asymmetric, the blue wing being absorbed by neutral gas. This indicates that the velocity structure of the neutral gas and the scattering by H I atoms can also play an important role in the escape of the Lyα photons.


INTRODUCTION
One of the most important present astrophysical issues is understanding the formation and evolution of galaxies. One way to address this problem is through the study of highredshift objects. Any high-redshift object can be considered as a star-forming galaxy if it looks like a galaxy and is identiÐed by spectral features consistent with those arising from young stars (types O, B, and A). Many observational 1 Based on observations made with the Hopkins Ultraviolet Telescope and the NASA/ESA Hubble Space T elescope, obtained at the Space Telescope Science Institute, operated by AURA, Inc., under NASA contract NAS5-26555.
2 Adjunct astronomer at the Space Telescope Science Institute.
programs to detect such objects have been carried out during the past few years (e.g., & Hartwick Pritchet 1990 ;et al. & Ham-Lowenthal 1990 ;Djorgovski 1992 ; Steidel ilton Djor-1993 ;McCarthy 1993 ;Wolfe 1995 ;Thompson, govski, & Trauger A small number of star-forming 1995). galaxy candidates have been found at high redshift near QSOs or damped Lya systems (e.g., et al. Lowenthal 1991 ;& Warren et al. et al. Moller 1993 ;Francis et al. & McMahon 1996 ;Djorgovski 1996 ;Hu 1996 ;Teplitz, & McLean or serendipitously (e.g., Malkan, 1995) et al. et al. et al. Ebbels 1996Trager 1997 ;Franx 1997 ;Yee et al. However, it is only recently that the class of 1996). 10 m telescopes have allowed spectroscopic conÐrmation of large numbers of star-forming galaxies at high redshifts, as selected by their broadband colors et al. (Steidel 1996 ; Many of these high-redshift galaxies Lowenthal 1997). show weak or absent Lya emission, a strong UV continuum, compatible with large star formation rates, and absorption features that are quite similar to those found in local starburst galaxies. Low-ionization absorption lines, such as Si II j1260, O I j1302, Si II j1304, C II j1335, Si II j1527, Fe II j1608, and Al II j1670, which are produced in the interstellar medium, are detected in the spectra ; but so are higher ionization lines, such as Si IV j1400 and C IV j1550. All these characteristics make the high-redshift galaxies quite similar to nearby starbursts. However, the P Cygni proÐles of the high-ionization metallic lines produced in the stellar wind of massive stars look weaker than in most nearby starbursts. This could be an e †ect of low metallicity at high redshift, since the strength of the stellar wind lines are metallicity dependent et al. et al. (Robert 1997 ; Walborn One exception is the galaxy MS 1512 [cB58 et 1995). (Yee al. This is the brightest starburst known at high red-1996). shift et al. C IV and Si IV absorption lines (Ellingson 1996). look stronger here than in other high-redshift galaxies, and appear to be quite similar to those of local starbursts. Since the metallicity is expected to be lower than in nearby starbursts, a younger age, an upper-limit mass cuto † or a (M up ), Ñatter initial mass function (IMF) could be required to explain the C IV and Si IV in this high-redshift galaxy.
The ultraviolet (UV) continuum in high-redshift galaxies is quite red, and a large (1 to 3 mag) extinction correction is required to match the continuum slope with the predicted j~2.6 law derived from evolutionary synthesis models & Heckman These inferences are based on (Leitherer 1995). the uncertain assumption that in these high-redshift galaxies, the nature of the dust and the IMF is similar to that in nearby starbursts. However, the overall similarities of the UV spectra of nearby and distant star-forming galaxies are so compelling that local starburst galaxies make very good laboratories in which to explore spectroscopic techniques in the UV that can be used to study high-redshift galaxies, and ultimately star formation and the evolution of galaxies in the early universe.
Four nearby starburst galaxies were selected for observation with the Hopkins Ultraviolet Telescope (HUT) during the Astro-2 mission et al. et (Davidsen 1992 ; Kruk al.
The main goal was a direct measurement of the 1995). escaping Lyman continuum Ñux. Because of the similarities between local starburst galaxies and high-redshift galaxies, and also because the large aperture of HUT can mimic the high-redshift situation rather well (even a narrow slit encompasses the entire central several starbursts), a direct measurement of the far-UV Ñux should be very useful in assessing the importance of star-forming galaxies for the ionization of the early universe & Ostriker (Miralda-Escude In a previous paper, we derived 2 p upper limits for 1990). the Ñux at 900 for each galaxy et al. Ó (Leitherer 1995a). Comparison with stellar population synthesis models suggests that at most only a few percent of the ionizing photons can escape from these starburst galaxies. The implication is that young protogalaxies may not provide the Lyman continuum photons for the ionization of the early universe (see also et al. et al. Deharveng 1997 ;Hurwitz 1997). Until now, no local starburst galaxy has been observed in the far-UV ; thus, these observations give us the opportunity to explore new spectroscopic techniques in the far-UV that can later be applied to the study of the star formation history in high-redshift star-forming galaxies, where the far-UV emission is shifted to the near-UV or the optical. In fact, an absorption feature between 1010 and 1060 was Ó Ó detected in the spectra of these starburst galaxies. This can be attributed to the lines Lyb, O VI j1032, 1038, and C II j1036, 1037. Evolutionary synthesis models of these lines have been performed Delgado, Leitherer, & (Gonza lez Heckman hereafter These were based on a 1997 ; Paper I). stellar library built with spectra of O and B stars, which were used as inputs to an evolutionary synthesis code to predict the line proÐles of O VI ] Lyb ] C II. O VI is a very sensitive indicator for the presence or absence of O stars, which show a P Cygni proÐle. Lyb and C II are very sensitive indicators for B stars. If a population younger than 1 Gyr is present in a galaxy, a strong stellar feature will be observed, from O VI if the dominant population are O stars or from Lyb ] C II if the feature is dominated by B stars.
In we describe the observations, and deals with the°2°3 determination of the reddening through the UV continuum Ñux distribution.
presents a study of the stellar Section 4 population using C IV, Si IV, and O VI proÐle synthesis techniques, and also the absolute Ñux at 1500 Ó. Section 5 addresses the interstellar absorption lines and the kinematics of the interstellar medium. The Lya emission and the e †ect of the extinction and gas absorption in the Lya/Hb ratio are discussed in The implications are taken up in°6. and a summary and conclusions are given in°7,°8.

OBSERVATIONS
The initial motivation for far-UV observations of starburst galaxies was to study their Lyman continuum. The observations were performed with HUT during the Astro-2 mission in 1995 March, covering 820 to 1840 with a Ó Ó moderate resolution (R \ 400 at 1200 We selected four Ó). starburst galaxies that met the following criteria : (1) a large enough redshift to separate the starburst Lyman edge from the Galactic Lyman edge (thus the galaxies have cz º 5000 km s~1) ; (2) large UV Ñux at 1500 (3) strong Ha emis-Ó; sion, indicating an intense source of ionizing radiation within each galaxy ; and (4) low Galactic foreground extinction. The galaxies selected were NGC 6090, Mrk 66, Mrk 1267, and IRAS 0833]6517. presents a summary of Table 1 the properties of these galaxies. The Galactic foreground extinction was derived from the observed H I columns et al. and assuming (Stark 1992) The values are given in Table 2.
The spectra were taken through circular apertures with diameters of 12A and 20A, except for Mrk 66, which was observed through the large aperture only. 12A corresponds to 3.8, 3.8, 4.4, and 5.8 Leitherer (1995a). Additional data for IRAS 0833]6517 were obtained with the HST . The galaxy was observed with the Goddard High-Resolution Spectrograph (GHRS) and the grating G140L, which has a nominal dispersion of 0.57 per diode, Ó through the Large Science Aperture (LSA, 1.7 ] 1.7 arcsec2). One of the exposures, with an integration time of 4216 s, covers 1213 to 1499 and three exposures were Ó Ó , taken covering 1465 to 1752 with a total integration Ó Ó , time of 13,872 s. After standard pipeline processing, all the spectra were combined into a single spectrum covering 1213 to 1752 Ó Ó . The instrumental spectral resolution for a point source observed through the LSA is 0.80 but for an extended Ó, source this depends on the size of the UV source. Using the Galactic C II j1335 interstellar absorption line, we determine that the instrumental resolution is better than 2.3 Ó FWHM for IRAS 0833]6517. The center of the Gaussian Ðt to this line is 1334.4^0.1 giving an uncertainty in the Ó, zero-point wavelength calibration of 0.1 (23 km s~1). The Ó continuum Ñux in the GHRS spectrum is a factor of 3 lower than in the HUT spectrum. This di †erence reÑects the aperture loss in the GHRS spectrum, and tells us that the UV source is more extended than 2A. WFPC2 images of the galaxy at 2200 show that most of the UV light is emitted Ó by bright spots aligned in an internal bar extending 3 ] 0.9 arcsec2 (1.1 ] 0.3 kpc2). Di †use emission is extended about et al. 3A .6 (Robert 1997). The spectra were corrected for redshift with the values given in shows the HUT spectra after a Table 1. Figure 1 wavelet Ðltering was applied Starch, & Bijaoui (Murtagh, except for the galaxies IRAS 0833]6517 and NGC 1994), 6090, for which the spectrum shown is the original, rebinned every 4 pixels. The 1 p error spectrum of the original data, assuming Poisson noise, is shown at the bottom. The S/N ratio at 1500 is 9, 4, 3, and 5 for IRAS 0833]6517, Mrk Ó 1267, Mrk 66, and NGC 6090, respectively. The spectra are in the rest-frame wavelength for each galaxy. also Figure 1 shows the GHRS spectrum of IRAS 0833]6517 corrected for redshift and rebinned to 0.57 pixel~1. The original Ó spectra were oversampled by 0.25 pixel steps.
The most prominent absorption features seen in the spectra are Lyb ] O VI j1032, 1038 ] C II j1036, Si II j1260, O I ] Si II j1303, C II j1335, Si IV j1400, Si II j1526, and C IV j1550. The strongest emission features are the geocoronal Lya and Lyb, and also the intrinsic Lya in emission. An exception is Mrk 1267, which has Lya in absorption.  Kinney (1994),  Ðnd a relationship between the color excess and the change of the UV continuum slope produced by reddening,  Vol. 495 where *b is the di †erence between the observed and the predicted slope. This method of determining the reddening is based on the assumption that the starburst is dominated by only a young burst (less than 10 Myr old). If older bursts contribute signiÐcantly to the UV continuum, the slope would be Ñatter than [2.2, and the extinction determined in this way should be an upper limit to the real value.
We have derived the continuum slope by Ðtting a Ðrstorder polynomial in the (log log j) plane using di †erent F j , windows for the continuum.
gives the average value Table 2 of the slope for each galaxy and the color excess derived. However, the color excess derived depends on the extinction law used. We have also checked that the reddening derived in this way gives similar results, within the errors, if we use the LMC extinction law instead of the law given by et al. However, the E(B[V ) values derived Kinney (1994). using the Milky Way (MW) extinction law instead of the Kinney or LMC law are larger by a factor of 2 (see Table 2).
Because Galactic extinction toward these galaxies is smaller than the reddening derived (see the color Calzetti,

Storchi-Bergmann
No signiÐcant di †erence is found (1994). between the two methods if the MW extinction law is assumed instead of the LMC or that of et al. Kinney (1994). In the same way, we have derived the slope and the color excess of IRAS 0833]6517 using the GHRS spectrum. The slope is somewhat steeper (b \ [1.7) and the color excess lower [E(B[V ) \ 0.1] than the value derived from the HUT spectrum, probably indicating a nonhomogeneous distribution of the dust and/or a change in the mean evolutionary state of the stellar population with the aperture size.

STELLAR POPULATION : LINE PROFILE SYNTHESIS AND ABSOLUTE FLUX SYNTHESIS
The most important stellar features seen in the near-UV region in the spectra of starburst galaxies are the resonance lines C IV j1550 and Si IV j1400. All O stars show a P Cygni proÐle in the C IV lines. In addition, all O supergiants show a P Cygni proÐle in Si IV Bohlin, & Panek (Walborn, 1985). In the far-UV, the most prominent stellar wind feature is the line O VI j1032, 1038. All O and B0 stars develop a P Cygni proÐle or a blueshifted absorption feature in O VI (Morton et al. As with C IV and Si IV, the 1979 ; Walborn 1995). proÐle of O VI is a better indicator than its equivalent width of the properties of massive stars in a stellar cluster ; this is because of the possible strong interstellar contribution in C IV and Si IV, and a possible blend of O VI with Lyb and C II j1036, 1037. Other ultraviolet lines that provide information on the stellar content in a starburst are the resonance line N V j1240 and the recombination line He II j1640. N V develops a strong P Cygni proÐle in the winds of all O and B0 stars. He II shows a broad emission proÐle, and it is formed in the very dense stellar winds of Wolf-Rayet (W-R) stars and O5ÈO3 supergiants Robert, & (Leitherer, Heckman et al. 1995b ;Robert 1997). Evolutionary synthesis models of a stellar population have been computed for C IV, Si IV, N V, and He II by Leitherer, & Heckman and et al. Robert, (1993 (1995), Leitherer (1995b), Models at subsolar metallicity predict weaker C IV Paper I. and Si IV P Cygni proÐles than those generated using libraries of stars in the solar neighborhood et al. (Robert 1997). These models must be applied cautiously for metallicities lower than 1/3 solar.
In the following subsections, we use evolutionary synthesis models to study the evolutionary stage of the stellar population of the four starburst galaxies observed. First, we will apply the C IV, Si IV, and O VI ] Lyb ] C II line proÐle synthesis technique. For C IV and Si IV, this technique has been successfully applied to starbursts et al.  and Seyfert 2 galaxies Conti, 1996) et al. Delgado et al. Our (Heckman 1997 ;Gonza lez 1998). strategy is to Ðnd the best synthetic model that Ðts C IV and Si IV, and then to compare the synthetic O VI proÐle with the observations in order to infer the possible contribution of interstellar Lyb and C II. In most cases, the line proÐle synthesis technique does not provide a unique solution. To distinguish which is the best solution and to further constrain the range of possible models, we use the UV continuum luminosity at 1500 With this luminosity, we can Ó. predict the number of ionizing photons, Q, for each model. The predicted values are compared with the value derived from the Balmer recombination lines. We also use the equivalent width of the Balmer emission lines and the excitation ratio [O III]/Hb. This ratio will be obtained by computing photoionization models, taking as ionizing spectrum the spectral energy distribution of the models. The reason for combining the UV continuum with the Balmer recombination lines and the excitation ratio is to ascertain whether the UV light is mainly the product of O or B stars. If hot stars produce the UV emission, models that Ðt the UV will correctly predict Q and the excitation ratio. To Ðnd the best models that Ðt C IV and Si IV, we Ðrst use the GHRS data and then perform a consistency test with the HUT data.
A comparison of the proÐles with those of NGC 1741, another starburst galaxy, indicates that the stellar absorption lines are not as strong in IRAS 0833]6517 as they are in NGC 1741. This indicates either that IRAS 0833]6517 is probably in a more advanced evolutionary stage than NGC 1741, which has an estimated age of 4È5 Myr et al. (Conti or that the IMF is steeper in IRAS 0833]6517 than 1996), in NGC 1741, or that there is a dilution of the stellar lines in IRAS 0833]6517 due to an underlying population. Note that NGC 1741 was also observed with the GHRS at the same aperture as IRAS 0833]6517, and that it is at a similar distance as IRAS 0833]6517.
Because of a contribution from the interstellar C IV and Si IV (which are stronger in the starburst spectra than in the synthetic models), we do not Ðt the entire line proÐle. Instead, we select two wavelength windows, corresponding to the blue side of the wind proÐle, 1535È1542 (for C IV) Ó and 1382È1390 (for Si IV), and to the red side, 1550È1560 Ó (for C IV) and 1404È1409 (for Si IV), where the proÐles Ó Ó are dominated by the stellar contribution. We compute the s2 parameter between the observations and every model, and the results are plotted for burst models as a function of age, upper mass limit cuto †, and slope of the IMF for C IV in Figures  In the case of an instantaneous burst, we can exclude models younger than 5 Myr based on the values of s2 if the upper mass limit cuto † is higher than 40 or the IMF M _ slope Ñatter than a \ 3 (IMF P M~a). Models with an age of 6 Myr are consistent with the data, but we cannot constrain the upper mass limit cuto † or the slope of the IMF, since for this age the most massive stars have already reached the supernova phase. Any age is compatible with the proÐles if or a \ 3.75. For CSF models, M up \ 30 M _ we can exclude models of any age with an upper mass limit cuto † higher than 40 and a slope of the IMF Ñatter M _ than a \ 3. Figures and show the proÐles of C IV and 4a 4b Si IV for burst models, and Figs. and show these 4c 4d proÐles for CSF models, for an age of 6 Myr.
compares the HUT spectrum with the GHRS Figure 5 spectrum rebinned at the same resolution as the HUT spectrum. An inspection of the C IV and Si IV proÐles indicates no di †erence between the two spectra. Thus, the same conclusions that were found with the GHRS spectrum hold for the HUT spectrum. The di †erences between these spectra are the slope of the continuum (as we have noted previously, the HUT spectrum is Ñatter than the GHRS spectrum) and the strength of some interstellar lines (e.g., Si II j1527 is stronger in the HUT spectrum than in the GHRS spectrum).
The most straightforward conclusion is that O stars must be present in the cluster of IRAS 0833]6517, in order to explain the blue wing of the C IV and Si IV lines, but these stars cannot be more massive than about 40 because M _ , is less than 40 or the age is older than 5 Myr.
shows the proÐle of the absorption feature Figure 6 between 1000 and 1060 The equivalent width of this Ó Ó . feature is 14.4 and the minimum is at 1027 The main Ó, Ó.  contributors to this feature are O VI j1032, 1038, Lyb, and C IIj 1036, 1037. The contribution of both stellar and interstellar C II must be small, since we cannot identify any line at the rest wavelength of C II (for the stellar component) or blueshifted about 2 because of the interstellar component Ó (see for an explanation). We estimate that the equivalent°5 width of the interstellar C II component is 1.3 This is Ó. derived from the equivalent width of interstellar C II j1335, assuming that these lines are on the Ñat part of the curve of growth, so that the ratio of the equivalent width over the wavelength is constant. Although this feature is predicted to be very broad and strong in the poststarburst phase, we can exclude all such poststarburst models because stellar C II is not clearly present (see Figs. 7eÈ7h). This also Paper I, means that we can exclude stellar Lyb as the sole contributor, because the equivalent width of this line in stars hotter than B0 is less than 2 Moreover, this feature Ó (Paper I). cannot be mainly the interstellar Lyb, since an equivalent width of interstellar Lyb of 14.4 implies an H I column Ó density larger than 1022 cm~2, which is much larger than the value derived using a normal dust-to-gas ratio and the reddening (less than 1021 cm~2). We can also exclude the interstellar O VI as the sole contribution for the large value of the equivalent width, the broadening of the feature, and the shift with respect to the rest wavelength. Typical values of the equivalent width of the interstellar O VI detected in the coronal gas in the Galaxy are less than 0.1 Ó (York 1974).
Thus, the feature seems to be mainly the product of stellar O VI ] interstellar Lyb. A direct comparison of the proÐle with a O VI ] Lyb ] C II burst synthetic model at 6 Myr, which is one of the models compatible with the C IV and Si IV proÐles, shows that the observed feature is stronger than the model, probably indicating a stronger contribution of O VI and/or interstellar Lyb. The O VI proÐle of an individual O star is generally similar to the one (Fig. 6) observed. However, there are two important di †erences. First, the emission part of the P Cygni proÐle seems to be destroyed by some absorption in the galaxy, and second, the blue wing of the observed proÐle is broader and stronger than that in the star. We note that the blue wing of Lya (which is detected in emission) is a †ected by absorption in neutral gas that is blueshifted with respect to the rest wavelength (see shows the velocity proÐle of°6). Figure 7 the O VI ] Lyb feature and the Lya wavelength region, with the rest wavelength of Lyb and Lya taken as zero velocity. Thus, we suggest that the excess absorption in the blue wing of the O VI proÐle could be produced by blueshifted interstellar Lyb absorption with a velocity larger than 3000 km s~1. If this is conÐrmed, an extraordinary outÑow of neutral gas in this starburst has been detected.

Ultraviolet Continuum L uminosity and Ha Equivalent W idth
From the line proÐle analysis, we can exclude all the burst models that require stars more massive than 40È50 to be currently present. Burst models younger than 5 we use the UV continuum luminosity to further constrain the range of possible models. shows the 6 Myr burst model with Figure 8 M up \ 30M _ overplotted on the HUT spectrum dereddened by E(B[V ) \ 0.17 using the LMC law. After dereddening, and assuming a distance of 78 Mpc, the luminosity at 1500 is Ó  after correcting by the reddening derived from the Balmer decrement, and assuming the radiation-bounded, large optical depth case B. The best model would be one that Ðts the proÐles of the stellar components of C IV and Si IV and also predicts a Q value equal to that derived from the recombination lines.
gives the predicted values of Q  Margon (1988). cates that the actual star formation rate is higher than the average star formation rate during the last 108 yr. The same conclusion is obtained if we compute the evolutionary indices proposed by Hunter, & Tutukov Gallagher, (1984), which indicate that the present rate is higher than (a c \ 5.2) the average rate during the last 108 yr (a L \ 0.5). We can use the equivalent width of a recombination line such as Ha as an additional constraint. However, this physical parameter should be used with caution, since the equivalent width of any recombination line could be diluted by the existence of an underlying population that could contribute to the continuum at the optical wavelength but not to the ionization of the gas. Nevertheless, models that predict values for the equivalent width of Ha lower than that  Note that these observations were done with an Ó. aperture of 22A, which is similar to the HUT aperture. However, this is a lower-limit value of EW(Ha), since in addition to the e †ect of an underlying population, EW(Ha) may also be a †ected by reddening, since the stars and gas may not be a †ected by the same amount of dust. If we assume that the nebular light is a †ected by reddening derived from the Balmer decrement and that the stellar light is a †ected by reddening derived from the UV continuum slope, then the equivalent width of Ha after dereddening is 264 (assuming the E(B[V ) derived using the UV contin-Ó uum slope and the extinction law of et al. Kinney [1994]). With this value of the equivalent width, we can only exclude burst models older than 8 Myr.

T he Excitation Ratio [O III]/Hb
Finally, we can also use the excitation ratio [O III]/Hb to further constrain the models. This ratio depends mainly on the hardness of the ionizing spectrum, the ionization rate, and the metallicity of the gas. This means there is a dependence on the total mass and age of the starburst. We have computed photoionization models using the code CLOUDY (C90.03 ; taking as ionizing spec-Ferland 1996), trum the spectral energy distribution of the models that Ðt the stellar component of C IV and Si IV, the Q value predicted for each model, a metallicity of 0.6 times solar, and a constant electron density of 100 cm~3. We compute spherical models with di †erent values of the Ðlling factor, /. A change in / is equivalent to a change in the ionization parameter, U, because U is proportional to the Ðlling factor, gives [O III]/Hb for / \ 10~2, U P (QN e /2)1@3. Table 3 10~3, and 10~4. The models that give values of this ratio closest to the observed 1.8 et al. are the 6 (Margon 1988) Myr burst models with / \ 10~3 and 10~4 or a 9 Myr CSF model with / \ 10~2. Note that the variation of the excitation ratio with age reÑects the changes in the spectral energy distribution (SED) with age produced by the variations of the W-R/O number. For the W-R/O ratio M up \ 30 M _ , increases rapidly at D7 Myr, the age at which the most massive stars evolve to the W-R phase ; at this stage, the relative number of electrons able to excite the O`2 increases with respect to the number of hydrogen ionizing photons as a result of the harder SED produced by the W-R stars. We can exclude any CSF model younger than 6È7 Myr, because the predicted [O III]/Hb is much lower than that observed (see We can also exclude any burst model older Table 3). than 6È7 Myr, because for these, Q is a factor of 4 lower than observed, and also because after this age the spectral energy distribution of the burst models is not hard enough to produce the observed high-excitation line ratio of . 0833]6517 is indeed the product of a combination of a Ðeld population and a superposed starburst, the starburst population by itself could be weighted more heavily toward massive stars than suggested by the formal IMF parameters given in Table 3.

C IV and Si IV Synthesis ProÐles
The spectrum of this galaxy is noisier than that of IRAS 0833]6517, which makes the s2 results for C IV and Si IV much less reliable. C IV and Si IV are much weaker here than in the starburst NGC 1741, more similar to those in IRAS 0833]6517. This means that the burst is in an advanced stage of evolution, older than 5 Myr, and/or that M up ¹ 40 To illustrate this, shows the HUT spectrum M _ . The equivalent width is 7.6 Two absorption Ó.
Ó. minima are detected, one associated with C II (probably mainly interstellar), the other with Lyb ] O VI. There is good agreement between the observed proÐle and the 6 MyrÈold burst model, except in the C II line, which is stronger in the starburst spectrum than in the model, indicating a larger contribution from the interstellar component. We also note that this feature is blueshifted about 1 with respect to the models. We do not Ðnd a physical Ó interpretation for this, and attribute it to a worse wavelength calibration here than in the Si II j1260 and C II j1335 wavelength region, where the uncertainty in the zero-point calibration is 0.5 (see Thus, the absorption feature Ó°5). seems to indicate the presence of late O stars (O7ÈO9), for which the O VI stellar feature is still important.   Table 3). have mentioned earlier, the EW(Ha) also could be a †ected by an older underlying population that contributes to the continuum but not to the ionization, making the EW(Ha) lower than the values predicted by the models. Burst models with ages between 6 and 8 Myr are able to reproduce Q and EW (Ha) This means that the starburst in Mrk (Table 4). 1267 is at an advanced stage of evolution. Again, we note the possibility of an underlying Ðeld population that may bias the derived IMF parameters toward less massive stars. The star formation rate derived using the B luminosity of the galaxy after correction for Galactic and intrinsic absorption is 2.3 yr~1. This is similar to the star forma-M _ tion rate derived from the Ha Ñux, 4.5 yr~1. The evolu-M _ tionary indices (0.9) and (0.2) are very similar, a c a L indicating that the present star formation rate is not signiÐcantly higher than the average in the last 108 yr.  (Robert 1997). Vacca (1995) shown that the library at solar metallicity can still be applied to the 30 Doradus region, which has a metallicity of 0.3 The spectrum of Mrk 66 does not show any clear Z _ . stellar contribution in the Si IV and C IV lines, indicating a strong e †ect of metallicity in this starburst, or/and an advanced stage of evolution, and/or dilution by an underlying Ðeld population.  (Calzetti 1994). reddening. This small equivalent width is incompatible with all the CSF models younger than 50 Myr, assuming a Salpeter IMF slope (see Fig. 44 in & Heckman Leitherer 1995) and that the optical continuum light is not a †ected by an underlying population. Both the EW(Ha) and Q are compatible with burst models in the range of 7È9 Myr. Accounting for dilution would lead to starburst populations with correspondingly more massive stars. As in Mrk 1267, the star formation rate derived from the Ha Ñux and the B luminosity are very similar, 1.2 and 2 yr~1, respectively, M _ indicating that the current star formation rate is not higher than average.

C IV and Si IV Synthesis ProÐles
NGC 6090 is the most distant galaxy of the starbursts studied here, and it also has the largest metallicity (close to solar). The C IV and Si IV lines indicate a larger stellar contribution than in the previous objects, probably due to the earlier evolutionary stage of this starburst compared to the other objects, or a more powerful starburst compared to the Ðeld population. To constrain the star formation history from the C IV and Si IV proÐles, we have applied the same technique used for IRAS 0833]6517 ; we Ðtted the two wavelength windows corresponding to the red and blue sides of C IV and Si IV proÐles and then computed the s2 parameter. Figures and show the results for burst and 11 12 CSF models, respectively. The Si IV proÐle is a good discriminator between burst and CSF models. This line shows a conspicuous P Cygni proÐle if very massive stars are present in the starburst and if the star formation process occurs in a very short period of time. This is because the P Cygni proÐle is produced only by blue supergiants, which live only for a short period of time, and whose total number in a cluster is not very large compared with the total number of O stars. We can exclude the CSF models based on the Ðt to Si IV For burst models, the C IV Ðt (Fig. 12). gives a minimum age between 5 and 6 Myr and an IMF with a Salpeter exponent (a \ 2.35) or Ñatter (a \ 1.5), while the Si IV Ðt indicates an age between 3 and 5 Myr and Thus, we can conclude that very massive M up º 60 M _ . stars are present in the starburst and that star formation occurred 3È6 Myr ago in a very short period of time.

O VI ] Lyb ] C II Synthesis ProÐle
The spectrum between 1010 and 1060 shows a P Cygni Ó proÐle. O VI, and therefore O stars, are clearly detected. The equivalent width of the absorption proÐle is 10.6 and the Ó, emission region has an equivalent width of 2.2 Ó. Figure 13 shows the observed feature compared with the burst model and SK [66¡172, an O4 star in the LMC. The stellar and galactic features agree rather well, except in the absorption feature at 1034 This line is the interstellar C II j1036, Ó. 1037, which is blueshifted with respect to the rest wavelength, along with the other interstellar lines (see A°5). comparison with the 5 Myr burst model indicates that the emission is stronger in the observed proÐle than in the models, and also that the absorption region is broader than in the models. This is likely the result of some contribution by interstellar Lyb blueshifted by several and also a Ó, stronger contribution of stellar O VI in blue supergiants than in main-sequence stars, thus making the P Cygni proÐle more conspicuous than predicted by our models. Note that in our modeling we do not make a distinction by luminosity class, since in general the O VI proÐle does not depend much on luminosity (see Fig. 5 of There-Paper I). fore, our stellar library only distinguishes by spectral class.

Ultraviolet Continuum L uminosity and Ha Equivalent W idth
We can exclude the CSF models from the line proÐle synthesis, since such models predict weaker stellar Si IV absorption than that observed. The C IV proÐle lets us exclude burst models with an IMF slope steeper than Salpeter. We use the predicted Q values derived from the luminosity at 1500 to further constrain the possible models. Ó After dereddening by E(B[V ) \ 0.22 using the LMC law, the luminosity at 1500 is 1041.07 ergs s~1 Note that Ó Ó1. we would obtain the same luminosity if we used the MW extinction law and E(B[V ) \ 0.5. The dereddened Ha luminosity [using E(B[V ) from the Balmer decrement] implies that the number of ionizing photons is Q \ 1054.55 s~1. This value is compatible with burst models younger than 3 Myr ; however, from the line proÐle modeling, the best models are bursts with ages between 4 and 6 Myr. However, these models underestimate the Ha Ñux by a factor larger than 2 (see Note that the HUT Ñux is Table 5). about 30% lower than the IUE Ñux ; thus, the missing radiation in the HUT aperture cannot account for the di †erence between the predicted and observed Balmer Ñux. Figure 14     respectively. Burst models of ages 5 and 6 Myr Ó, are compatible with the observed value.
The current star formation rate in NGC 6090 derived from the Ha Ñux is 43 yr~1, which is higher than the M _ average star formation rate derived from the B luminosity of the galaxy, 5.6 yr~1. The a-indices and M _ (a c \ 9 a L \ 0.5) indicate that recent star formation is more conspicuous than in the past.

T he Excitation Ratio [O III]/Hb
Finally, we can use the predicted [O III]/Hb ratio to constrain the models. As in the case of IRAS 0833]6517, we compute photoionization models assuming a spherical geometry, constant electron density of 100 cm~3, metallicity 0.76 times solar, and Q equal to the value predicted for each model. We take as the ionizing spectra the spectral energy distributions from the models that Ðt the stellar component of C IV and Si IV. We have computed models with / \ 10~3 and 10~5. Burst models with ages between 3 and 6 Myr and / \ 10~5 predict a ratio [O III]/Hb close to that observed, 0.75È1.5 & Kinney Huchra, & (Calzetti 1992 ;Hartmann, Geller To summarize, 3È6 Myr burst models are 1984). compatible with the observations of NGC 6090.

THE INTERSTELLAR ABSORPTION LINES
The UV spectra of starburst galaxies are also rich in strong absorption features formed in the interstellar medium. The most important lines are Al II j1670, Fe II j1608, C IV j1550, Si II j1526, Si IV j1400, C II j1335, O I ] Si II j1303, Si II j1260, Lya, C II j1036, O VI j1038, 1032, and Lyb. Some of these lines are superposed on or blended with resonance wind or photospheric absorption stellar lines, for example Si IV j1400, C IV j1550, C II j1036, Lyb, and O VI j1032, 1038. It is also well known that the interstellar absorption lines in starbursts are stronger than the interstellar lines observed in the spectra of nearby stars. This is primarily because of the larger interstellar velocity dispersion in the starbursts. That is, in many starbursts, the stronger absorption lines are optically thick and the equivalent width of the lines is mainly related to the velocity dispersion of the gas instead of to the column density (e.g., et al. & Leitherer Conti 1996 ;. We Ðrst address the question of where these interstellar absorption lines are formed. To do so, we measure the radial velocity of the lines Si II j1260, C II j1335, and Si II j1526 with respect to the adopted systemic velocity for each galaxy. We exclude other resonance lines, such as C IV and Si IV, because these are not pure interstellar lines and because their radial velocity in many cases can reÑect a blueshift due to the stellar winds. A cursory inspection of Figures and clearly indicates that the interstellar 8, 9, 14 lines are blueshifted with respect to the systemic velocity, except in Mrk 1267. In the case of IRAS 0833]6517, Si II j1260 and C II j1335 lines from the Milky Way are also detected. If we assume these lines are at a heliocentric velocity of zero, the corresponding lines in IRAS 0833]6517 are blueshifted with respect to the galaxy systemic velocity by 450 km s~1 and 520 km s~1, respectively. Another signature of the out-Ñowing gas in this galaxy is provided by the Lya proÐle in the GHRS spectrum (see Relative to the systemic Fig. 7). velocity, the Lya emission is redshifted by about 500 km s~1. As in the well-studied case of Haro 2 et al. (Lequeux this is the result of the absorption of the blue half of 1995), the Lya emission line by the outÑowing gas. Thus, in this galaxy the interstellar absorption lines are blueshifted by about 1000 km s~1 with respect to the Lya emission line, and the galaxy systemic velocity is about halfway in between. We will discuss this point further in below.°6 We do not detect absorption lines from the Milky Way in the other three starbursts. This is owing to a combination of a worse S/N and a lower resolution in the HUT spectra than in the GHRS spectrum, and the equivalent width of the lines. Note that the MW C II j1335 is resolved from O I ] Si II j1303 in the GHRS spectrum but not in the HUT spectrum. Since we do not know accurately the location of the centroid of the UV light within the HUT aperture, the true zero point in the HUT wavelength scale (in the starburst rest frame) is ill determined. However, both Mrk 66 and NGC 6090 have Lya in emission. In these two galaxies, the Si II j1260, C II j1335, and Si II j1526 lines are blueshifted with respect to the Lya emission line by about 760 km s~1 and 480 km s~1, respectively. In view of the above results for IRAS 0833]6517, these blueshifts imply high-velocity outÑows of gas, of several hundred km s~1. Mrk 1267 has no Lya emission, and thus we cannot determine if there is an outÑow. We cannot estimate the spatial scale associated with this gas, but it cannot be produced in isolated clouds, since ISM lines cover about 50% of the UV light Thus, this high-velocity gas probably rep- (Fig. 15). resents a Galactic-scale outÑow.
Another indicator of the large-scale motions of the interstellar gas in these starbursts is the broadening of the interstellar lines. In the GHRS spectrum of IRAS 0833]6517, the Si II j1260 and C II j1334 lines appear to be asymmetric, and they can be Ðtted by two components. One component is unresolved (instrumental resolution about 2.3 FWHM), Ó centered on 1261.0^0.9 and has a FWHM of 2.9^0.9.
Ó, The other, which is the dominant contributor to the total equivalent width, is centered on 1258.4^0.7 and has a FWHM of 3.8^1.0 (corresponding to an intrinsic Ó FWHM of about 700^250 km s~1 ; see C II Fig. 15a). j1335 can also be Ðtted by two components. One is unresolved at the rest wavelength, and the other is blueshifted by 2.6^0.2 with a FWHM of 5.0^0.6 (corresponding to Ó, Ó an intrinsic FWHM of about 1000^150 km s~1 ; see Fig. The HUT spectrum does not have sufficient spectral 15b). resolution, and only one Gaussian is sufficient to Ðt Si II j1260 with a FWHM of 7.0^2.2 Taking the FWHM of Ó. the MW C II j1335 as the instrumental resolution, the intrinsic FWHM of Si II j1260 derived from the HUT spectrum is 1060^500 km s~1, which is in agreement with the value obtained with the GHRS spectrum. In the spectra of the other three starbursts, C II j1335 has a FWHM of 6.1^2.5 5.2^1.2 and 5.8^2.0 for NGC 6090, Ó, Ó, Ó Mrk 1267, and Mrk 66, respectively. As we do not know the HUT point spread function of this line, we cannot make a reliable estimate of the intrinsic FWHM of the absorption lines in these three starbursts.
More evidence for large-scale motions of the interstellar gas comes from the equivalent width of the metallic interstellar lines. These are optically thick, and they are in the Ñat part of the curve of growth. This means that the equivalent width of these lines is not proportional to the column density of the gas, as would be the case if they were in the linear part of the curve of growth. Instead, the equivalent width is determined by the velocity dispersion of the gas. We can show that in fact the lines are saturated, taking for example Si II j1260 and Si II j1527. If these were optically thin, the equivalent width of Si II j1527 would be about three times lower than that of Si II j1260. However, the observed ratio is larger than (see Therefore, an 1 3 Table 6). equivalent width of about 2È3 implies a velocity dis-Ó persion larger than 200È300 km s~1 (FWHM \ 470È700 km s~1). This may indicate that several unresolved velocity components are observed. However, we can estimate the minimum Si II and C II column densities by using the equivalent width of the lines Si II j1527 and C II j1334, assuming that they are in the linear part of the curve of growth. This gives minimum values that are several times 1014 cm~2 for each ion.
Where can the Si II and C II lines be formed ? The ionization potentials of these ions (8.1 eV for Si II and 11.3 eV for C II) are smaller than that of hydrogen ; thus, they can be formed in an outÑowing neutral or ionized medium. In°6 we will show clear evidence for the existence of a neutral medium that is Ñowing away from the Lya absorption line, at least in IRAS 0833]6517. However, we will see that the derived column densities of C II and Si II are compatible with their formation in the H II region around the starburst. We have computed photoionization models using the code CLOUDY to determine the column den-(Ferland 1996) sities of C II and Si II, taking as inputs the SED from the starburst models listed in Tables and and the number of 3 5 ionizing photons derived from and assuming a L 1500 , spherical geometry, a constant electron density of 100 cm~3, a Ðlling factor of 0.01, 0.001, and 0.0001, and the appropriate metallicities for IRAS 0833]6517 and NGC 6090. We also assume that Si is depleted with respect to the solar value. Therefore, we use a gas-phase Si abundance of 4 ] 10~4, which is the typical value of the interstellar medium in the Milky Way. The column densities computed have values of several times 1016 cm~2 for C II and several times 1014 cm~2 for Si II (if we assume that the Si is not depleted, the column density of Si II would be 1 order of magnitude larger than this). Models with lower values of the electron density (1 cm~3) predict similar values of the column densities. Within a factor of 2È3, these values do not depend on the starburst model assumed. These values are larger than the lower limits derived from the equivalent width of the lines, indicating that these ions can be formed in an ionized medium.

THE Lya EMISSION
Lya emission can in principle be produced by recombination of hydrogen photoionized by O and B stars. The stellar clusters in starburst galaxies provide sufficient ionizing radiation to produce a large Lya Ñux. However, galaxies observed with IUE show Lya emission weaker than predicted by recombination theory & Terlevich (Meier Hartmann et al. et al. and 1981 ;Terlevich 1993), an anticorrelation between the Lya/Hb ratio and the metallicity is found. One possible interpretation is that Lya photons are attenuated by dust as a result of multiple resonant scattering by hydrogen atoms that increase the path length of the Lya photons and thus the probability that Lya photons will be absorbed by dust. However, other authors & Kinney have shown (Calzetti 1992 ; Valls-Gabaud 1993) that some galaxies have a Lya/Hb ratio consistent with simple recombination theory if the ratio is corrected for reddening using the appropriate extinction law for the metallicity of the galaxy, and if the Hb Ñux is obtained with an aperture similar to that used for the IUE data. This result contrasts with that of et al.
for IZw 18, Kunth (1994) where no Lya emission was detected, even though the metallicity and dust in this galaxy are very low. A few more galaxies observed with HST show the same behavior et al. Furthermore, several galaxies have been (Kunth 1997). detected whose Lya emission line shows an asymmetric proÐle, with the peak emission redshifted with respect to the systemic velocity et al. et al. (Lequeux 1995 ;Kunth 1997). The explanation is that the neutral gas responsible for the absorption is blueshifted with respect to the systemic veloc-ity, as a result of which only Lya photons in the red wing can escape. This suggests that the amount of neutral gas and dust are not the only determining factors for Lya emission, but that the velocity structure of the gas also matters. A similar conclusion was reached by Koratkar, Giavalisco, & Calzetti for a sample of galaxies observed with (1996) IUE. They found no good correlation between the equivalent width of Lya and the reddening or metallicity, and interpreted this as evidence that the interstellar medium in these galaxies is highly inhomogeneous and that the escape of Lya photons is determined by the geometry of the interstellar medium.
In three of the galaxies observed by us, Lya is detected in emission. The Lya/Hb dereddened using the Balmer decrement is shown in Also shown are the equivalent Table 7. width of Lya and the dereddened values, assuming that the gas is a †ected by the reddening derived from the Balmer decrement and the continuum by the reddening derived from the UV slope. The dereddened values in IRAS 0833]6517 and NGC 6090 are very close to the values predicted for bursts younger than 6 Myr (Valls-Gabaud & Fall, These ages are in agreement 1993 ;Charlot 1993). with values derived from the synthesized absorption line proÐles. However, in Mrk 66 the Lya emission line equivalent width is quite low, and the line is detected only in absorption. One possible explanation, according to those models, is that these galaxies are in a poststarburst phase. However, this disagrees with our absorption proÐle analysis and also with the observed strength of the Balmer emission lines.
After correcting for extinction, the Lya/Hb ratio in IRAS 0833]6517 and NGC 6090 is close to the recombination value (D33 ; & Osterbrock In Mrk 66 this Ferland 1985). ratio is much lower. One possible explanation is that the starburst is surrounded by an H I envelope, and the Lyman photons are destroyed by absorption by dust after multiple scattering with the H I atoms, because the dust optical depth is relatively small in Mrk 66 (this is our lowest metallicity starburst) and the number of ionizing photons derived from the Hb Ñux is large enough. However, due to the proximity of the geocoronal Lya and the poor spectral resolution in the HUT spectrum, we did not Ðnd any clear evidence of absorption in this object (although there is a hint of asymmetry in the Lya proÐle) that could be interpreted as being due to absorption by neutral outÑowing gas (see et al. However, we do Ðnd evidence for Lequeux 1995). an outÑow in Mrk 66 from the Si II j1260 and C II j1335 lines as compared to Lya (see°5).
The GHRS spectrum of IRAS 0833]6517 (Fig. 16a) sheds more light on this issue. The peak of the Lya line is redshifted with respect to the systemic velocity, the proÐle and we assume that the blue part of the line is not a †ected by Lya emission. We also assume that the blue wing of Si III is a †ected by another absorption feature seen close to the geocoronal Lya. The results of the Ðtting give b (Fig. 16a) parameters (corrected for the instrumental resolution) of 530 km s~1 and 630 km s~1 for Si III and Lya, respectively. The b parameters indicate an interstellar velocity dispersion (FWHM \ 875 km s~1 and 1000 km s~1 for Si III and Lya, respectively) similar to the values derived from the FWHM of Si II j1260 and C II j1335. Again, we found evidence for large-scale motions of the interstellar gas in this galaxy. Clearly, these large values of b may be the result of the presence of multiple velocity components.
Next, we consider how much of the Lya emission is a †ected by the absorption component, and thus whether in this starburst the velocity structure of the neutral H I gas or the abundance of dust is the determining factor for the escape of Lya photons. The normalized observed proÐle was divided by the Ðtted Voigt proÐle, and the resulting proÐle was Ðtted by a Gaussian emission line. The (Fig. 16b) emission-line Ñux of the Gaussian after correction for the reddening derived from the Balmer decrement is 9.8 ] 10~12 ergs s~1 cm~2, which is 50% larger than before correction for the absorption. The GHRS aperture does not include all the Lyman photons, and the Ñux measured in the HUT aperture is a factor of 4 larger than that in the GHRS aperture. Therefore, we estimate that the true Lyman Ñux is about 4 ] 10~11 ergs s~1 cm~2, implying a Lya/Hb ratio of 26.6, very close to the case B recombination ratio.
The conclusion is that in IRAS 0833]6517, the velocity structure of the neutral gas and the scattering by H I atoms are not the sole determining factors for the escape of the Lyman photons, since the correction factor to the Lya/Hb ratio from extinction is larger than from the absorption of the neutral gas. However, these could be determining factors in low-metallicity objects where the dust-to-gas ratio is low. 7. IMPLICATIONS 7.1. T he Lya L ine One of the techniques used to look for primeval galaxies is to search for Lya emitters ; however, few strong Lya sources have been found at high redshift, with the exception of QSOs. It has been suggested that Lya photons are obscured by multiple resonant scattering with H I atoms, thereby increasing the optical depth across the ISM and thus the probability for absorption by dust (Neufeld 1990). This means that the Lya emission search technique is not always a good tool for detecting candidate star-forming galaxies at high redshift et al. However, (Hartmann 1984). other studies suggest that the weakness of the Lya emission in nearby starburst galaxies is due more to extinction than to resonant scattering, making the technique appropriate, since the extinction in primeval galaxies should be smaller & Kinney Our study (Calzetti 1992 ;Valls-Gabaud 1993). does not unambiguously support either of these suggestions. For two galaxies with larger metallicity and larger Balmer decrement reddening, IRAS 0833]6517 and NGC 6090, the determining factor for Lya emission is obscuration by normal dust extinction. The velocity structure of the H I gas and the scattering it produces seems to be only a secondary e †ect. However, for more metal-deÐcient starbursts, the normal extinction is not so important, and the H I column density and the geometrical distribution of the neutral gas and dust seem to be more important factors. In addition, the mass of the starburst and its age also seem to play some role (at least in Mrk 1267, where Lya is purely in absorption). Bursts that are not very massive and that are in a very evolved state, in combination with large amounts of H I, will preferentially show Lya in absorption. All these considerations indicate that the Lya emission technique is not a straightforward tool for searching for primeval galaxies, owing to the difficulty of interpreting the Lya emission line in terms of the rate of star formation & (Charlot Fall 1993).
A second point is related to the proÐle of the Lya emission line detected in IRAS 0833]6517. This line is asymmetric, with the peak emission redshifted with respect to the systemic velocity and a deep absorption in the blue wing of the line. Evidence of the asymmetry of this line is also present in the other starbursts, but better spectral resolution, comparable to that of the GHRS spectrum, is needed to conÐrm this suggestion. In the three galaxies where Lya is detected in emission, the interstellar lines are blueshifted by several hundred of km s~1 with respect to the systemic velocities. This means that the asymmetric Lya line proÐle is a consequence of the interstellar gas outÑow that absorbs the Lya blue wing via neutral H I gas. Such out-Ñows are undoubtedly a consequence of the star formation activity in these starbursts.
It is interesting to note that a Lya proÐle similar to that seen in IRAS 0833]6517 has been detected in a candidate protogalaxy at a redshift of 3.150 et al. In (Djorgovski 1996). this galaxy, the asymmetry of the line is interpreted as consequence of the line photons passing through lines of sight that intersect the di †erential rotating disk of the galaxy. The proÐle of the line is then used to derive the dynamical mass of the galaxy. Our results suggest an alternative explanation, namely, that the proÐle may be the consequence of an outÑow of the interstellar gas. Thus, Lya should be used with care for mass estimates of high-redshift galaxies.

T he Dynamics of the ISM in Starbursts
Most of the interstellar lines detected in the spectra of these starbursts are saturated. This means that the equivalent width of the lines is more representative of the velocity dispersion than of the column density. The equivalent widths indicate a large velocity dispersion. More direct evidence comes from the broadening of the Si II j1260 and C II j1335 lines in IRAS 0833]6517. These lines are resolved into two components, one unresolved at the rest wavelength, and the other blueshifted by several hundred km s~1, having intrinsic FWHM of 700^250 km s~1 and 1000^150 km s~1, respectively. This broadening is a clear indication of the large-scale motions of the interstellar gas in this starburst. This means that the kinematics of the warm ionized interstellar medium is driven more by the dynamical consequences of the violent star formation process ongoing in these galaxies than by their gravitational potential. The implication is that similar processes can occur in high-redshift galaxies. In this case, any determination of the dynamical mass of the galaxy through the velocity dispersion derived from the interstellar absorption lines would be incorrect. A similar conclusion was reached by & Leitherer from the GHRS spectrum Heckman (1997) of the nearby dwarf starburst galaxy NGC 1705. 7.3. T he Escape of UV and Ionizing Radiation In a previous paper et al. we found that (Leitherer 1995a), the luminosity of the burst at 900 is proportional to the Ó number of Lyman continuum photons for very di †erent IMF and star formation histories : log (Q Lyc /L 900 ) \ 13.28 0.16 (photons erg~1) for CSF and 13.07^0.50 Ó (photons erg~1) for an instantaneous burst. Using Ó published Ha Ñuxes, corrected by the reddening derived from the Balmer decrement, et al. Leitherer (1995a) obtained the ratio of the recombinations to the luminosity at 900 For the four starbursts, the observed values of this Ó. ratio are larger than the predicted theoretical value by almost 2 orders of magnitude, implying that only a few percent of the ionizing photons escape from these galaxies (but see et al. for some caveats). If we now Hurwitz [1997] correct the upper limit of the Ñux at 900 by the extinction Ó derived from the UV slope, we obtain logarithmic ratios of 14.34, 13.22, 13.55, and 13.98 (photons erg~1) for IRAS Ó 0833]6517, Mrk 1267, Mrk 66, and NGC 6090, respectively. With the exception of Mrk 1267, these values are still lower than the theoretical ratio derived from the models. This implies that dust alone is not responsible for the low value of and absorption by gas in the galaxies and L 900 , their halos may be responsible for the low fraction of photons escaping. This indicates that the fraction of ionizing photons that can escape from these galaxies is very small, and that young primeval galaxies are not likely to provide the Lyman continuum photons for the ionization of the early universe.

SUMMARY AND CONCLUSIONS
The present work is a continuation of the study presented by et al. Four starburst galaxies were Leitherer (1995a). observed by HUT from 820 to 1840 with a moderate Ó resolution (R \ 400 at 1200 The galaxies selected are Ó). a †ected by very little Galactic extinction, and they show strong Ha emission, indicating a large supply of ionizing photons. The measurement of the Lyman continuum Ñux suggests that only a small fraction of the ionizing photons escape from these galaxies. This implies that young galaxies may not provide enough photons for the ionization of the early universe. Similar conclusions have been reported by et al. Deharveng (1997) Hurwitz (1997). In this paper, we analyze the HUT spectra of the starbursts galaxies IRAS 0833]6517, Mrk 1267, Mrk 66, and NGC 6090 in order to constrain the stellar population and the kinematics of the interstellar medium. GHRS data for IRAS 0833]6517 are also presented. We use an analysis of the stellar component of the UV absorption lines Si IV, C IV, and O VI ] Lyb ] C II to study the star formation history. The proÐles of these lines are predicted by evolutionary synthesis models. These models are based on stellar libraries built with O and B stars. From the line proÐle analysis Vol. 495 and the absolute Ñux at 1500 we conclude that O stars Ó, are present in these starbursts. The stellar components of Si IV, C IV, and O VI are weaker in IRAS 0833]6517 and Mrk 1267 than in NGC 6090. This indicates that (1) the bursts are more evolved in IRAS 0833]6517 and Mrk 1267 than in NGC 6090, or (2) NGC 6090 has more massive stars, or (3) the lines are diluted by an old population. The latter suggestion becomes even more plausible if we compare NGC 6090 with the Wolf-Rayet galaxy NGC 1741 et al. NGC 1741 is dominated by a starburst, (Conti 1996). just like NGC 6090. In both cases, the UV spectra show strong stellar P Cygni proÐles.
The spectra of these starbursts are rich in interstellar absorption lines. In IRAS 0833]6517, Mrk 66, and NGC 6090 the lines are blueshifted by several hundred km s~1 with respect to the systemic velocity or with respect to Lya detected in emission. These outÑows might be triggered by the starburst activity. This situation di †ers from other local starbursts et al. et al. in which (Conti 1996 ;, no outÑows have been detected in the UV absorption lines. Other evidence for large-scale motions of the interstellar gas comes from the broadening of the interstellar lines in IRAS 0833]6517. For the other three galaxies, it comes from the large value of the equivalent width of the interstellar absorption lines.
Lya in emission was detected in IRAS 0833]6517, Mrk 66, and NGC 6090. The dereddened Lya/Hb ratio is close to the recombination value in IRAS 0833]6517 and NGC 6090, but much smaller in Mrk 66. An analysis of the absorption component of Lya in IRAS 0833]6517 indicates that even though the velocity structure of the neutral H I gas plays an important role, it is not the sole determining factor for the escape of Lya photons. This result suggests that in galaxies with relatively high metallicity and thus dust abundance, such as NGC 6090 and IRAS 0833]6517, the velocity structure of the neutral gas and the scattering by H I atoms are not solely responsible for the escape of the Lyman photons. However, in low-metallicity objects where the extinction is low, it could be the primary factor. W. D., Robert, C., Leitherer, C., & Conti, P. S. 1995, ApJ, 444, Vacca, 647 D. 1993, ApJ, 419, Valls-Gabaud, 7 N. R., Bohlin, J. N., & Panek, R. J. 1985 Yee, 1996, AJ, 111, 1783D. G. 1974