A redshift survey of submillimetre galaxies

A significant fraction of the energy emitted when the Universe was young came from very luminous galaxies largely hidden at optical wavelengths by shrouds of interstellar dust grains, and now resides in the cosmic background radiation at wavelengths near 1mm (Fixsen et al. 1998). The first real insight into the origin of the far-IR/submm background came with the first deep fields observed by the SCUBA submm camera on the JCMT (Smail, Ivison & Blain 1997). Close to 100% of the submm background has now been resolved with the deepest SCUBA maps which exploit a sensitivity boost from gravitational lenses (Smail et al. 2002, Cowie et al. 2002). However, the near-IR/optical faintness of the submm galaxy population conspires with the modest positional precision available from the large (15 arcsec) SCUBA beam and the large surface density of unrelated optically-faint galaxies to render positional coincidence alone inadequate to identify counterparts. This problem is compounded by the small field of view of SCUBA: compiling large samples of submm galaxies is slow, with roughly one detected every night. Because of these difficulties, robust spectroscopic redshifts, a crucial element in understanding their nature and evolution (Blain et al. 1999), have been published for only a handful of submm sources, all atypically bright at optical wavelengths (e.g., Ivison et al. 1998, Barger et al. 1999).

Figure 1. Four representative rest-UV spectra for our submm galaxies. spanning the I = 22.2-26.4 range, ordered brightest (a) to faintest (d). While the optically brightest and faintest sources show AGN characteristics in their spectra, the sources between do not reveal any signs of AGN. Source (b) appears as a typical Lyman-break galaxy. Source (c) is the least compelling of the four, with only a single strong emission line feature, which we interpret as Lyman-alpha. However this source has recently been detected in Balmer-alpha (Smail et al. in prep). Spectroscopic exposures times were 1.5-6.0 hr. One-dimensional spectra were extracted and compared with template spectra and emission-line catalogues to identify redshifts. Most identifications are based on multiple lines, most prominently the Ly-alpha line which varies tremendously in both flux (ranging from 1 to 60uJy) and rest-frame equivalent width (3 to >100A). Weaker stellar/interstellar/AGN features and/or continuum breaks were detected in the spectra, strengthening the redshift identifications. Our highest S/N spectrum (SMMJ163650.0+405733 - Smail et al. 2003) includes features indicative of strong starburst activity (P-Cygni wind absorption profiles, stellar/interstellar absorption lines). Narrow-line Type-II AGN with enhanced NV and/or CIV emission are consistent with half of our spectra. Despite this evidence for AGN in these galaxies, they appear to be relatively weak in energetic terms as compared to those identified in X-ray or optical surveys.

We have been able to overcome these problems by taking advantage of 1.4-GHz radio data from the Very Large Array (VLA) radio telescope. These deep images have fine (~2 arcsec) spatial resolution and a large (30 arcmin FWHM) field of view. They are sensitive to the synchrotron radio emission from cosmic-ray electrons accelerated in the supernovae explosions of the same high-mass stars that heat the dust and generate the far-IR/submm emission. Hence a deep radio image should be an efficient route to pinpoint the positions of many submm galaxies. This is confirmed in fields with both submm and radio observations (Ivison et al. 2002, Chapman et al. 2003a): radio counterparts brighter than 30 uJy can be found for 60-70% of submm galaxies brighter than 5 mJy at 850 µm (which have a surface density of 450 per square degree and contribute about 20% of the submm background). Moreover, the accurate radio positions mean that SCUBA's efficient `photometry' mode can be used to search for submm galaxies (Barger, Cowie, & Richards 2000; Chapman et al. 2001, 2002), raising the detection rate to around ten per night.

With a large sample (~100) of radio identified SCUBA galaxies in hand (Chapman et al. 2002, 2003a; Ivison et al. 2002), we initiated a multi-object optical spectroscopy program in March 2002, using the Keck 10-m telescopes and the LRIS-Blue spectrograph. In 8 clear nights we have targetted 62 radio-detected submm galaxies mostly brighter than 5 mJy at 850µm, and with optical magnitudes in the range I-mag = 22-27. We measured robust redshifts for 35 submm galaxies (preliminary results in Chapman et al. 2003b), representative of the blank-field population (Figs. 1 & 2). This represents an order of magnitude increase over previously published redshift samples, and importantly allows the first constraint on the optically faint majority of the SCUBA galaxies. The redshifts span the range z = 0.7-3.7, with a median of 2.4 and an interquartile range of 1.9-2.8 (Fig. 2). These redshifts allow dust temperatures (T_d) and bolometric luminosities to be determined, assuming that the tight correlation between far-IR and radio emission observed at low redshifts (Condon 1992) remains valid. If the radio emission is boosted by an AGN then both temperature and luminosity are overestimated. There are no obvious cases in the present sample in this category. Typical values are of order 35 K and several 10¹² L_o respectively.

To quantify the whole submm population using this survey we must account for several selection effects. In particular, requiring a radio identification prior to spectroscopy limits the maximum redshift. Studies of galaxies at low and moderate redshifts suggest that changes in the dust properties, especially the dust temperature T_d, modify the relative strength of the emission in the submm and radio wavebands (Blain et al. 2002). This results in a range for the maximum redshift detected in the radio: hot sources can be observed out much higher redshifts than cold sources.

The relationship between different classes of high-redshift galaxy is critical for our understanding of galaxy evolution. Our spectroscopy shows that the submm galaxies are coeval with the important populations of star-forming galaxies and quasars detected at z=2-3 using optical and X-ray techniques (Boyle et al. 2000; Barger et al. 2002; Steidel et al. 1999, 2002). We can use their relative space densities to compare and relate these populations. In a 1000 Mpc³ box at z~2.5, there are 10 Lyman-break galaxies with R-mag < 25.5 and 1 submm galaxy from our sample (with 850µm flux > 5mJy), but the single submm galaxy produces a comparable bolometric luminosity to the 10 Lyman-break galaxies.

Figure 2. The redshift histogram of our submm galaxy sample. We describe the selection effects using an evolving model of the local far-IR luminosity function, in which the dusty galaxies are represented by a range of template spectral energy distributions. The model has been tuned to fit the statistical properties of the submm-radio galaxy population. We plot the predicted redshift distributions for submm galaxies with flux densities S(850µm)>5mJy (solid line) and radio sources with S(1.4GHz)=30-500uJy (dashed line). We expect to miss sources lying between the submm and radio model curves due to our requirement of a radio detection to pinpoint the submm source. The apparent deficit between model and data at z~1.5 may be indicative of the difficulty in obtaining spectra in the spectroscopic desert that spans the range 1.2

We are rapidly approaching a statistically useful sample of SCUBA galaxies with redshifts, however we are already close to exhausting the supply of radio identified examples. Two ongoing projects using SCUBA to followup faint radio sources are underway this semester (PI: Smail, UK; PI: Chapman, international). Both programs are seeking to rapidly uncover large samples of submm galaxies with precise radio positions for Keck optical spectroscopy. These programs are expected to increase our sample size of SCUBA sources with redshifts by 50%. The Smail program was awarded grade 1 weather to additionally attempt 450µm detections as a means to further constrain the SED peak for sources with spectroscopic redshifts, an independent check on the dust temperature and therefore the bolometric luminosity extrapolations. This highly efficient observing mode with SCUBA provides a complementary approach to the blank field mapping campaigns underway with the "SHADES" consortium (see article by Dunlop in this issue). Several exciting studies are underway with this new and expanding sample: the redshift clustering, the dust temperature distribution, the UV spectroscopic properties, a comparison with MAMBO millimeter sources having spectroscopic redshifts, and over the longterm, a detailed comparison with optically selected galaxies and absorption systems at similar redshifts to our SCUBA sample.