3. EVOLUTION OF STAR FORMATION IS ALSO WELL TRACED IN THE INFRARED

A region's "current" star formation rate (SFR) is approximately proportional to its population of upper main sequence stars, because of their cosmically brief lifetimes. The presence of massive stars is manifested in several observables which are not strongly produced by old stellar populations: a) direct photospheric radiation, predominantly in the UV. At lower frequencies it extends into the optical, where it produces blue colors in a stellar population. The high-frequency tail extends above 1 Ryd, and includes a substantial fraction of ionizing photons, in the hottest stars. Much of this radiation is absorbed by gas and re-emitted as b) nebular emission lines, mostly in the optical and infrared. The most massive stars hardly live long enough to completely leave the dusty environments in which they were born. Thus, much of their remaining continuum emission may be absorbed by interstellar dust grains, which are warmed to c) re-emit the power thermally in the infrared. Finally, massive stars are the most violent sources of mechanical energy in the ISM, driving winds and supernovae. A fraction of this power accelerates electrons which then emit d) synchrotron radiation.

After a great deal of work on the alternatives, it is H emission which remains the "gold standard" of SFR measurements. It is equivalent to counting the ionizing continuum photons from the hottest (youngest) main-sequence stars at a wavelength that is relatively insensitive to absorption. The observable line luminosity is supposed to equal a constant times the aggregate SFR: L(H) = 1.1 × 10⁴¹ erg/sec per M / year [16].

By contrast, the other leading SFR indicator at high redshift-the ultraviolet continuum-is far more sensitive to dust reddening. The integrated star-forming luminosity of those LBGs detected needs to be corrected upward for extinction by a factor of 2 - 7 ([32], [33], [1], [14], and [38]). Furthermore, these LBGs were selected to be the least dusty ones.

We can obtain a better idea of why the line and continuum SFR estimates disagree (methods a and b above) by looking closely (with HST) at nearby star-forming galaxies. The contours in Figure 5 show our HST imaging of the (continuum-subtracted) H in the small star-forming galaxy NGC 2328. The greyscale shows the optical continuum, which has a considerably different spatial distribution. Note the numerous contour peaks with little or no associated optical light. These HII region complexes with significant reddening account for the majority of the current star formation in this galaxy. Another even more dramatic example of a dust-enshrouded starburst is seen in the nearby galaxy NGC 5253 [11].

Figure 5. Our WFPC-2 continuum-subtracted H image of "Early-Type" Galaxy NGC 2328 from Glassman and Malkan [10], shown as a greyscale plot. The optical continuum is shown with overplotted contours. The field of view is roughly 6 × 10".

3.1. Narrow-Band Infrared Imaging of Emission-Line Galaxies

Large-format multislit spectrographs are not yet available in the near-IR. This will change when instruments such as the NIRMOS (Near-infrared Multi-Object Spectrograph) begin working on the VLT, and Flamingos (in multi-slit mode) on Gemini. For now, ground-based searches are restricted to either narrow spatial or spectral windows. The latter case is obtained with narrow-band interference filter imaging. To speed up the detections, there is often a pre-selection of special fields. These are supposed to have excess galaxies at the targeted redshift, because some object in the field is already known to have that z. Several dozen galaxies at redshifts above 2 have been discovered by detecting their H line emission ([27], [28], [4], [15], [29]). More discoveries will be coming soon with the availability of 1K × 1K near-infrared array detectors.

3.2. Grism Spectroscopy

Slitless spectroscopy receives the full sky background from the entire spectral window at each point on the detector. It is therefore impractical with ground-based infrared telescopes. However, it has been demonstrated to work very well in space, where the backgrounds are much lower. In particular, the NICMOS Camera 3 on HST has an objective grism that disperses the spectra of all objects within its 50" field-of-view. Since it was able to operate simultaneously with other HST instruments, it obtained a substantial number of deep parallel pointings on random fields. The largest set of these were reduced and analyzed by McCarthy et al. [30]. The exposures of a few orbits had a 1 in 3 chance of detecting a line-emitting object in the field; in the longer parallel observations, these odds increase to nearly 100%. In addition, a single very deep NICMOS grism field was observed during a 3-day pointing in the Continuous Viewing Zone ([23]).

The standard assumption in deep near-IR grism surveys of random fields is that any single emission line detected is redshifted H. In only one of the dozen cases tested so far has this assumption been shown to be incorrect; it is likely true better than 90% of the time. This allowed Yan et al. 42] to estimate the global luminosity function of recent star formation in the 0.8 z 1.7 universe.

In some cases, the difficult confirmation with optical spectroscopy is not even necessary because a second emission line is also detectable simultaneously in the NICMOS grism image. Figures 6 and 7 show deep grism spectra (observed wavelengths) in which emission lines other than H detected. In Figure 6 the additional lines on the blue side are H , and the [OIII]5007 / 4959 doublet, at z_e = 1.49. In Figure 7 the only emission-line spectrum which does not show H is shown. This foreground (z_e = 0.35) galaxy shows the strong emission lines of He I 1.083 / P 1.093 µm, [FeII]1.257 and P 1.281 µm [23].

Figure 6. Extracted spectrum of a star-forming galaxy discovered in a deep NICMOS-grism parallel observation (Malkan et al. [23]). At a redshift of 1.49, this low-resolution spectrum shows both H and H, as well as a broad blend of the two [OIII] doublet lines, the redshifted 5007 and 4959Å emission.

Figure 7. Another star-forming galaxy in the same NICMOS parallel field [23]. This is the only grism spectrum in which none of the detected emission lines is H. The lower redshift, ze = 0.35, allows the detection of the He I 1.083 / P 1.093 µm blend and the [FeII]1.257 and P 1.281 µm emission lines.

In its first incarnation, this camera was defocused by the distortion of the NICMOS dewar. This degraded the spatial resolution to about 0.4". The resulting spectral resolution was only R ~ 50. Hopkins et al. [14] used NICMOS Camera 3 for pointed observations of suspected high-redshift galaxies during a campaign when it was in focus. Both their observations and the parallel ones indicated that SFR rates estimated from H emission in galaxies at 0.8 z 1.7 are several times higher than those estimated from their UV continuum.

These large numbers of strong H-emitting galaxies already indicate that search methods using shorter wavlengths are incomplete. Similar incompleteness in surveys of star formation rates has also been found in the local universe by Sakai et al. [35]. If short-wavelength surveys really do miss a substantial amount of obscured star formation activity, we should examine galaxies that are found in the near-IR, to see if they are indeed dustier than those discovered by optical surveys. In Figure 8 we compare the two emission-line indicators of current star formation rates-H and [OII]3727 luminosities. The nine data points represent the NICMOS-selected galaxies, with followup Keck optical spectra from Hicks et al. [12]. The solid line shows the observed average ratio in local spirals, which does not fit the NICMOS galaxies. In these IR-selected galaxies the [OII] line systematically underestimates the SFR. This discrepancy between the two indicators cannot be predicted from the broadband colors, which are not often extremely red. Many of these galaxies must contain additional internal reddening around their star-forming regions. The dashed line shows an indicative amount, of A_V = 3 mag. If the ten galaxies we measured are representative of H selection in general, it produces samples of galaxies which a) would mostly be missed by optical search methods; and b) still need a significant upward correction for dust extinction. We need H and short-wavelength SFR measurements for more of the same galaxies to see how much overlap there is between those selected with optical and IR searches.

Figure 8. Comparison of star formation rates estimated from [OII]3727 and H emission, in z 1 galaxies discovered by the NICMOS grism. Typical uncertainties on the points are 20-30%. The open square indicates the line luminosities predicted by the local relations for an SFR of 10M / year. All of these 9 galaxies observed at Keck by Hicks et al. [12] have relatively weaker [OII] emission than would be predicted by their H , based on local spiral galaxies (which are described by the solid line). Instead, their [OII]/H ratios are suggestive of additional internal dust absorption, which might be as large as AV = 3 mag (shown by the dashed line).

3.3. Nebular Diagnostic Line Emission at Higher Redshift

H is redshifted beyond the prime sensitivity of infrared detectors at z 2.6. By unlucky coincidence, the rising thermal background makes it impossible to compare star formation rates from H with those estimated from the ultraviolet continuum in the LBGs. Thus the only way to make a direct comparison of SFR indicators (i.e., on the same galaxies) is by replacing H with another emission line that is bluer in the optical rest-frame, and less reliable. The advantage of using H as this "silver standard" SFR estimator is that its intrinsic strength relative to H is known accurately from recombination theory [33]. The [OIII] 5007 and 4959Å doublet, however, is usually two or three times stronger, and is not so affected by underlying stellar absorption. In Figure 9 we compare the star formation rates predicted by [OIII] and UV continuum emission in eleven LBG's at z ~ 3. A few galaxies lie near the bottom of the graph, consistent with 1-to-1 agreement. However in most of them, the [OIII] line emission predicts a higher SFR by a factor of 3 to 7 times. This is again consistent with the amounts of internal dust reddening that have been inferred from other observations.

Figure 9. The ratio of star formation rates estimated by [OIII]5007 and UV continuum emission from 10 Lyman break galaxies [38]. [OIII] for the open circles and asterisks was measured from narrow-band imaging; for the remaining points it comes from slit spectroscopy. The [OIII] line reveals several (up to seven) times more current star formation than does the UV continuum, especially in the redder galaxies. This is probably largely attributable to internal dust reddening.