The Lyman Continuum in Starburst Galaxies Observed with the Hopkins Ultraviolet Telescope

The starburst galaxies IRAS 08339+6517, Mrk 1267, Mrk 66, and Mrk 496 (=NGC 6090) were observed with the Hopkins Ultraviolet Telescope (HUT) during the Astro-2 mission. All four galaxies have radial velocities larger than 5000 km s-1, permitting the measurement of their intrinsic Lyman-continuum fluxes redward of the H I absorption edge in our Galaxy. The sample was selected on the basis of having the most favorable conditions for the escape of Lyman-continuum photons. Two σ upper limits of Fλ < 7 × 10-16 ergs s-1 cm-2 Å-1 were obtained for the flux around 900 Å within the rest frame of each galaxy. A set of theoretical spectral energy distributions has been calculated. We show that there exists a tight correlation between the continuum luminosity at 900 Å and the total number of photons emitted in the Lyman continuum, which is valid for high- and low-metallicity galaxies and essentially all relevant initial mass functions and star formation histories. Comparison with the observed values suggests that, on average, less than 3% of the intrinsic Lyman-continuum photons escape from the program galaxies. Models for the ionization of the intergalactic medium at high z by young starbursts require a significant fraction of the ionizing radiation to escape from the galaxy. If the four galaxies observed by us have properties similar to young galaxies at redshift z ≈ 3, such galaxies are not likely to provide Lyman-continuum photons for the ionization of the early universe.


INTRODUCTION
Starburst galaxies are a relatively heterogeneous class of objects with star formation rates high enough that they would exhaust their reservoir of molecular gas on a timescale much less than a Hubble time (Weedman 1987). Starburst galaxies include objects such as nuclear starbursts (Balzano 1983), H II galaxies (Terlevich et al. 1991), blue compact dwarf galaxies (Thuan 1991), and infrared-luminous galaxies (Soifer et al. 1987).
Of particular cosmological interest is the effect such early starbursts could have had on the intergalactic medium. Relatively model-independent arguments from metal production suggests that galaxies forming at z Ն 3 could be a significant, or even dominant, contribution to the metagalactic ionizing radiation field (Miralda-Escudé & Ostriker 1990;Songaila, Cowie, & Lilly 1990;Shapiro, Giroux, & Babul 1994;Madau & Shull 1996). However, models for the contribution of starforming galaxies to the UV background depend sensitively on the assumption of how much radiation leaks out of these gas-rich, dusty environments. Direct measurement of the escaping Lyman continuum flux from starburst galaxies with Hubble Space Telescope (HST) and IUE requires redshifts z Ͼ 0.3, while relatively modest redshifts, z Ͼ 0.015, are required to separate the starburst Lyman edge from attenuation by the Galactic Lyman edge and converging Lyman series. Such low-redshift galaxies are ideal candidates for observations with the Hopkins Ultraviolet Telescope (HUT), whose free wavelength range extends to 912 Å and below.
In this Letter we report measurements of the Lyman continuum of four starburst galaxies with HUT. In the absence of absorption and scattering by interstellar gas and dust, theoretical models for the spectral energy distribution of young starbursts predict significant emission above and below the Lyman edge. Therefore, a detection of, or a significant upper limit to, the Lyman continuum can constrain the fraction of photons escaping absorption within the galaxy and ionizing the surrounding intergalactic medium.

OBSERVATIONS
The observations were performed with HUT on board the Space Shuttle Endeavor during the Astro-2 mission in 1995 March. HUT is a moderate-resolution (R ϭ 400 at 1200 Å) spectrophotometer operating in the wavelength range 820 -1840 Å. A description of the instrument is given by Davidsen et al. (1992), and its in-flight performance and calibration for Astro-2 are described by Kruk et al. (1995).
Our sample is consciousl y biased toward having the most favorable conditions f or a high escape f raction of UV photons. The strong H␣ emission indicates that many ionizing photons are present. From the full sample of UV-bright starburst galaxies presented by Calzetti et al. (1995), the four we have observed are among the highest for their luminosities in the ratio of UV to infrared flux; thus, we have tried to minimize the amount of dust attenuation. Furthermore, the galaxies are rather symmetric and observed essentially face-on, so that the path length of a photon through the disk is minimized.
The combined spectra presented here are in each case the sum of several different observations taken through 12Љ and 20Љ diameter circular apertures. Because of the uncertainties in centering, the integrated fluxes from the starburst region are probably uncertain at a level of about 30%. Comparisons to optical and IUE data may suffer an additional (probably less than 10%) uncertainty because of the extended nature of the emission.
The spectra of the four galaxies obtained with HUT were reduced in the standard way and are shown in Figure 1. The strong emission features seen in the spectra are geocoronal Ly␣ and Ly␤. IRAS 08339ϩ6517, Mrk 66, and Mrk 496 also have strong intrinsic Ly␣ emission that is redshifted with respect to geocoronal Ly␣. The most prominent absorption lines are due to stellar C IV 1550 and Si IV 1400, in particular in IRAS 08339ϩ6517 (see Leitherer, Robert, & Heckman 1995). Further evidence for hot stars comes from strong O VI 1034, which is a unique indicator of the most massive stars (Walborn et al. 1995).
The continuum of the four galaxies is flat or slowly rising from longer wavelengths toward Ly␣. Around or somewhat below Ly␣, the spectra turn over and reach the noise level at 912 Å in the rest frame of each galaxy. An enlarged version of the spectral region around the Lyman edge is shown in Figure  2. Our upper limits on the Lyman continuum flux are computed directly from the count rate data. We calculate the background from the total number of counts in the wavelength range 804 -905 Å and assume that this remains flat across the wavelength range of the redshifted Lyman continuum (912-940 Å). In actuality, the background may be up to 5% higher in the 912-940 Å range because of the scattering wings of the geocoronal Ly␣ line. We compute the Lyman continuum flux from the total number of counts between 915 Å and the redshifted Lyman edge for each galaxy. Our upper limits to the Lyman continuum flux correspond to the maximum source count rates such that the formal Poisson probability of detecting N or fewer counts from a given source ϩ background rate ϫ time is less than 5%. These ''2 '' limits, converted to flux, are shown as thick solid lines in Figure 2. For all galaxies, the measurements of the Lyman continuum are consistent with no flux above the noise level.

COMPARISON WITH MODEL PREDICTIONS
The amount of UV radiation escaping from a starburst galaxy is determined by three parameters: the intrinsic UV spectrum of individual hot stars, the age and mass spectrum of   the stellar population, and the opacity of the starburst galaxy's interstellar medium.
We used the models of  to study the relation between the luminosity of the burst population at 900 Å (L ) and the total number of Lyman continuum photons (N Ly ) for different star formation histories and initial mass functions (IMFs). The purpose of these model calculations is to demonstrate that a measurement of L provides a reliable estimate of N Ly even if the starburst parameters are uncertain. The results are in Figure 3. These models were calculated for a constant star formation rate. For all models considered, there is a very narrow relation between L and N Ly . We find log N Ly /L ϭ 13.28 H 0.16 (photons Å erg Ϫ1 ). A second model series was run for an instantaneous burst. The results are essentially the same: log N Ly /L ϭ 13.07 H 0.50.
The model predictions can be used to derive upper limits on the fraction of Lyman photons escaping from the galaxies. Using the observed upper limits of F in column (2) of Table  2, the distances of Table 1, and correcting for Galactic foreground absorption (col. [3] of Table 2), we obtain lower limits on log N rec /L . They are tabulated in column (4) of Table 2.
Comparison with the model predictions of Figure 3 implies that the observed L is much lower than expected if every ionizing photon escaped from the galaxy. If the emission is isotropic, the fraction of photons below 912 Å escaping ranges from less than 15% in Mrk 66 to less than 0.95% in Mrk 496 (see col. [5] of Table 2).
Absorption by gas and/or dust in the galaxies and in their halos may be responsible for the low fraction of photons escaping. Dust scattering is not likely to be important since the HUT aperture is large enough to encompass all the dust scattering regions. The data do not allow us to quantify the relative importance of gas and dust absorption. The decreasing continuum level below 1100 Å in all galaxies suggests that interstellar dust absorption is significant.
Most of the Lyman continuum photons from a representative hot-star population are emitted at wavelengths within 100 -200 Å of the Lyman break. This coincides with the wavelength of the largest neutral hydrogen opacity. Additionally, dust absorption becomes important toward shorter wavelengths. Both effects together make it unlikely that a significant fraction of photons escapes from the galaxies at wavelengths shorter than those observed by HUT.
Absorption of photons may occur either in the starburst itself or, in addition, in a surrounding halo. Thus, our measurements do not exclude the possibility of the ionization of a halo by the diffuse radiation field escaping from the starburst.

IMPLICATIONS FOR THE ULTRAVIOLET BACKGROUND RADIATION
These are the first direct measurements of the emergent UV flux from actively star-forming galaxies, and the implications for both galactic structure and cosmology are significant. The results represent an important first step toward understanding whether or not young, star-forming galaxies might be responsible for the ionization of the early universe. At present, the known population of QSOs at high redshift appears to fall short by about a factor of a few from being able to ionize the intergalactic medium, which must have happened by at least z ϭ 4 judging by the intergalactic medium's transparency (Miralda-Escudé & Ostriker 1990;Meiksin & Madau 1993). Fall & Pei (1995) showed that this apparent shortfall may be a consequence of dust obscuration, which leads to an underestimate of the observed comoving density of quasars. If dust obscuration is taken into account, the contribution of quasars to the mean ionizing background is high enough to be consistent with limits imposed by the proximity effect.
Lyman continuum photons that accompany the production of metals observed in Ly␣ forest clouds could, in principle, make a significant contribution to the ionizing background at high redshift as well. Such photons could be produced by hot stars formed in young galaxies (Bechtold et al. 1987;Shapiro et al. 1994). The observed metallicity in spirals and elliptical galaxies and the recent detection of C IV absorption lines in Ly␣ forest clouds by Cowie et al. (1995) and Tytler et al. (1995) can be used to constrain the production of ionizing ultraviolet photons (Miralda-Escudé & Ostriker 1990;Madau & Shull 1996). This estimate is uncertain because of the unknown fraction of Lyman continuum photons escaping from the gas-(and possibly dust-) rich star-forming regions in galaxies. Madau & Shull (1996) found that young star-forming galaxies could account for the background intensity at the Lyman edge, in agreement with the limits imposed by the proximity effect if the photon escape fraction is on the order of tens of percents. This fraction is significantly higher than observed in the four galaxies we studied. If the four observed galaxies have properties similar to those of young galaxies at high redshift, they are unlikely to provide a significant contribution to the UV background radiation. Support for the Astro-2 Guest Investigator program was provided by NASA through grant NAG8-1075 from NASA/ Marshall. Helpful suggestions from Daniela Calzetti, Mike Fall, Piero Madau, and Mike Shull are gratefully acknowledged. Support for J. D. L. was provided by NASA through grant HF-1048.01.93A from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. (2), F is the 2 upper limit on the flux at about 900 Å observed with HUT. Col. (3), A 910 is the total Galactic foreground absorption at 910 Å, derived from the observed H I columns (Stark et al. 1992), N H /E(B Ϫ V ) ϭ 4.93 ϫ 10 21 cm Ϫ2 mag Ϫ1 (Diplas & Savage 1994), and the Mathis 1990 extinction curve. Col. (4) is the derived lower limit on log N rec /L . L has been reddening-corrected with A 910 of col. (3). The escape fraction of photons f esc in col. (5) was obtained by dividing 13.3 (see Fig. 3) by the observed values in col. (4).