3. PRIMEVAL GALAXIES AND THE FAR-INFRARED BACKGROUND

Searches at optical wavelengths for primeval galaxies using the Ly line have not been successful, although systematic efforts over large ranges in volume and redshift have been going on for many years. An elusive and still hypothetical creature, a primeval galaxy is usually defined by its hunters as an early type galaxy going though a dramatic initial star formation event, perhaps much like the luminous starburst galaxies we see in the local universe. A good recent review of the subject is given by Djorgovski and Thompson (1992). There are several possible explanations for this lack of success. One is that primeval galaxies must lie beyond a redshift of about 10, which is the approximate limit of large scale searches to date, a result that would be consistent with a baryonic dark matter model with primeval isocurvature fluctuations (Peebles 1987). Alternatively, it may be that the dark matter models in which galaxies grow much more recently by gravitational instability out of a scale-invariant spectrum of primeval adiabatic density fluctuations are correct. However, the most sensitive searches are now coming into conflict even with the predictions of cold dark matter galaxy formation simulations such as those of Baron and White (1987). Finally it may be that if primeval galaxies do exist, they are sufficiently dusty that the Ly photons are extinguished and much of the energy of the object appears at far-infrared wavelengths (e.g., Kaufman and Thuan 1977, van den Bergh 1990).

Two recent far-infrared/submillimeter PG models are those of Djorgovski and Weir (1990) and Franceschini et al. (1991). The model of Djorgovski and Weir was actually designed to fit the 700µm excess emission over the CMB claimed to be detected by Matsumoto et al. (1988). That excess has now been shown to be non-existent by COBE (Mather et al. 1990), however the model is still of interest for far-infrared bright PGs in general. Based on the observed spectral energy distributions of the nearby far infrared bright galaxies M82 and Arp 220, the model had a range of possible initial mass functions and burst timescales of 10 to 200 Myr. The models were constrained not to exceed the formation of a solar metal abundance during the burst phase. The numerical results are not directly applicable to the situation I am discussing here, however Djorgovski and Thompson (1992) used the model to conclude that the then available COBE limits ruled out more than a few percent of the stars in ellipticals and the bulges of spirals having been formed in dusty PGs, unless the redshift corresponding to the epoch of galaxy formation is less than z_f = 3, and/or the dust is unusually warm.

The model of Franceschini et al. (1991) is also based on the spectral energy distributions of local star forming galaxies, and they constrained the energy output to be that required to produce a solar metal abundance in 2 x 10⁸ years. Their resulting models for formation epochs of z_f = 2 and 4.3 are shown in Figure 6, where it may be seen that, as concluded by Djorgovski and Thompson, FIRAS strongly limits models with even moderate formation redshifts and dust temperatures like those of local universe starburst galaxies.

3.1. IRAS F10214+4724 - A Possible Protogalaxy

The conclusion one may draw from the models described above is that a scenario in which most galaxies went through a dusty early phase similar to local starburst galaxies may be in conflict with the FIRAS limits. However, on the other side of the coin there is some recent evidence in favor of the existence of large amounts of dust in galaxies at early epochs, and possibly even one example of a dusty PG: the extremely luminous galaxy IRAS F10214+4724 at z = 2.286 discovered by Rowan-Robinson et al. (1991). This galaxy is arguably the most luminous object in the universe with a luminosity of 10¹⁴ h^-2 L (h = H₀ / 100 km/s/Mpc; q₀ = 0.5), and a dust mass estimated from submillimeter observations of 2.5 x 10⁸ h^-2 M (Rowan-Robinson et al. 1993; Downes et al. 1992). Evidence for large masses of dust has also been found in several high redshift quasars (Andreani et al. 1993).

The controversy over the interpretation of F10214+4724, as for lower redshift ultraluminous infrared galaxies, concerns the dominant source of the extremely high far-infrared luminosity detected by IRAS. There is little doubt that both a luminous starburst and a non-stellar active nucleus are present in the source. There is abundant evidence that F10214+4724 is a primeval galaxy undergoing rigorous star formation, including ~ 10¹¹ h^-2 M of molecular gas (Solomon, Downes and Radford 1992), a UV-to-radio continuum energy distribution which is most simply interpreted as a powerful starburst (Rowan-Robinson et al. 1993; Mazzei and De Zotti 1993), and a radio source which is extended on a scale of about 2.5 h^-1 kpc (Lawrence et al. 1993). Likewise, there is substantial evidence for an embedded AGN: high excitation emission lines (Rowan-Robinson et al. 1991), and strong polarization (Lawrence et al. 1993). In addition, new results from near-infrared (rest-frame optical) spectroscopy show [NII] / H and [OIII] / H emission-line ratios to be typical of those found in type 2 Seyfert galaxies (Eisenhardt et al. 1993).

The near-infrared (rest-frame optical) continuum morphology observed using a 256² InSb array on the Keck telescope shows at least 3 continuum components that appear to be physically associated over a physical scale of 25 h^-1 kpc (Matthews et al. 1993), suggesting a small cluster since each object is more luminous in the rest frame r band than a local L* galaxy, and the main, southern object has almost 100 L* in rest frame r. A number of faint sources (K > 21 mag) are also seen within 20" of the central source that may be galaxies in an associated cluster. This image is reproduced in Figure 7. The Keck results also show that the brightest H source is now resolved on a scale consistent with that of the radio source, (0".5 ~ 2.5 h^-1 kpc), supporting the star formation origin for the H emission.

Figure 7. Image of F10214+4724 at 2.2µm from Matthews et al. (1993). The object is just right of center in this 40" x 40" image.

Can any protogalaxy model plausibly explain the tremendous luminosity of this object? It is clearly enriched in heavy elements already, as evidenced not only by the emission line spectrum which includes lines of C, N, Ne and Mg, but also by the presence of the dust itself, and this enrichment must also be explained by any plausible model. In particular, must the dust have been created in an earlier generation of stars? If the dust was created in the envelopes of evolved stars, as may be the case for much of the dust formed in our galaxy at the present epoch, then we must be seeing F10214+4724 at an age of at least 1 Gyr.

Elbaz et al. (1992) have developed starburst models for F10214+4724. They found that a model with a bimodal initial mass function (IMF) can achieve both the very high observed L / M_gas ratio of 750 L / M and the strong enrichment, reaching Z_Fe,C,O,Si > Z and M_dust < M_metals, in less than 10⁸ years. Hamann and Ferland (1993) have developed detailed chemical evolution models for QSOs, also concluding that high metallicities can be reached rapidly: > 10 Z in < 1 Gyr. The model of Elbaz et al. has a bursting component with a lower mass limit to the IMF of 3 M and a star formation rate of 6200 M / yr. They were unable to achieve a fit with a single IMF. The source of the dust in the Elbaz et al. model is not evolved stars but supernova remnants. While it is not known whether supernova remnants can be responsible for significant amounts of dust formation, there is evidence that SN 1987A has produced 0.1 M of dust (Dwek et al. 1992). At a supernova rate of 1.25 x 10^-12 (L / L) / yr (Solomon, Radford and Downes 1992) for 10⁸ years, remnants like 1987A could easily produce the the estimated dust mass of 2.5 x 10⁸ h^-2 M.

Mazzei and De Zotti (1993) have successfully modeled the spectral energy distribution of F10214+4724 using the population synthesis models of Mazzei et al. (1993) (see Figure 3). They find a good fit at an age of 1 Gyr (for H = 50 km/s/Mpc; q₀ = 0.5) with a star formation rate of 3 < 10⁴ M / yr; a fit with a much younger age is also possible. Mazzei and De Zotti show that F10214+4724 could plausibly fade to a z = 0 elliptical with bolometric luminosity less than 10¹³ L.

3.2. A Protogalaxy Model Based on IRAS F10214+4724

The models described above for the contribution to the far-infrared background by infrared-bright protogalaxies, those of Djorgovski and Weir (1990) and Franceschini et al. (1991), are based on the spectral energy distributions of local universe starburst galaxies. If it is truly powered by star formation, then F10214+4724 provides us with the opportunity of using a high redshift object with known luminosity and spectral energy distribution as a template for protogalaxies, thus eliminating the uncertainties introduced by assuming that local universe objects are good analogs of protogalaxies, or by adopting a model spectral energy distribution with an assumed dust temperature and luminosity. It also allows us to avoid the large K-corrections involved in redshifting local templates to cosmological distances.

I have therefore developed a simple model to determine the contribution to the far-infrared background of a population of galaxies, forming with a protogalactic burst like that observed in F10214+4724. If a QSO contributes significantly to the luminosity of F10214+4724 then the model predictions can be treated as upper limits to the background emission unless the coeval existence of a QSO with the starburst is a common feature in the formative stages of all ellipticals (cf. Hamann and Ferland 1993 and references therein).

I assume all galaxies have a SED of similar shape to F10214+4724. Galaxies are assigned luminosities according to a luminosity function. The variable parameters of the model are the cosmology, the formation redshift z_f, the burst duration t, the luminosity function, and the factor, f, by which F10214+4724 is assumed to be brighter or fainter than the characteristic luminosity at the knee of the luminosity function, L_FQ1214 = f L*, where the luminosity function is given by (L)dL = *(L / L*) e^-L/L* d (L / L*) (Schechter 1976). Here * is the characteristic space density. Four different local luminosity functions were considered. Note that using a local luminosity function to define the distribution of galaxy luminosities at high redshift is equivalent to adopting a mass function, as long as the evolutionary behavior with lookback time of the L / M ratio does not vary greatly with galaxy mass.

The range of parameters considered is given in Table 2. For most models the luminosity function was restricted to elliptical galaxies only, since spirals are not expected to have formed with a dramatic initial burst. The factor f was restricted to 10 or higher because F10214+4724 is undoubtedly a very rare and unusually luminous object. Estimates for the expected surface density of protogalaxies are in the range 10³ to 10⁵ per square degree, depending on the cosmology, the epoch of formation and the duration of the bright phase (e.g., Djorgovski and Thompson 1992). From the detection statistics we can estimate a surface density of objects like F10214+4724 of 1.5 x 10^-3 per square degree; allowing a factor of ± 10 on this estimate since the statistics are very crude (only one object has been detected, and that very close to the detection limit of the IRAS survey) it follows that objects like F10214+4724 are at least 10⁵ times less numerous that "typical" primeval galaxies. For a Schechter LF, this translates roughly to f > 10.

Table 2. Protogalaxy Model Based on IRAS F10214+4724

Parameter	Range Considered	Baseline Model
H0	50, 100	50
	0, 1	1
zf	1.5-10	2 - 5
t	1.0 to 2.0 x 108 yrs	108 yrs
f	10-100	10
Luminosity	Shanks et al. 1991, Es only	Shanks et al. 1991
Function	Efstathiou et al. 1988, all galaxies
	Franceschini et al. 1988a, Es only
	Tammann et al. 1979, Es only

Luminosity Function	*	M*
(H0=100)	(# Mpc-3)	(mag.)
Shanks et al. 1991	0.0096	-19.00	-0.07
Efstathiou et al. 1988	0.0156	-19.68	-1.7
Franceschini et al. 1988a	0.0032	-19.60	-1.0
Tammann et al. 1979	0.0031	-19.45	-0.77

Full details of the model are given in Lonsdale (1994, in preparation). Figure 8 summarizes the results of the models compared to the data of Figure 1. Figure 9 illustrates the blue and K-band number counts for the model population of F10214-like protogalaxies, compared to observational data.

Figure 8. Protogalaxy model predictions compared to the COBE data. Baseline model of Table 2 (heavy solid line); other lines show effects of changing other parameters: redshift range (light solid lines): zf = 5-10, 2 - 10, 1.5 - 2.5; cosmology (short dashed lines): = 0 and H0 = 100 (upper line) and 50 (lower line); LF (long dashed lines): Tammann et al. (lower line), Franceschini et al. (middle line), Efstathiou et al. (upper line - includes disk galaxies); f (dot-dash lines): f = 50 (upper line), f = 100 (lower line).

The low protogalaxy models shown in Figure 8 are in conflict with the FIRAS limits. The = 1 models are mostly consistent with FIRAS except the model using the Franceschini et al. (1988a) LF including all galaxies (not only ellipticals), which is in conflict with the more stringent FIRAS limit, and the high redshift range model which is only marginally consistent with the stringent limit. None of the models are in conflict with the current DIRBE observations. Therefore, basically the entire range of parameter space that has been explored is allowed for a high universe. An acceptable fit for a low universe would require z_f = 5 or lower, and/or f < 10, and/or a burst duration shorter than 2 x 10⁸ yrs.

Figure 9. Blue and K-band galaxy number counts compared to the prediction from the baseline protogalaxy model (see Table 2). The references to the data can be found in Chokshi et al. (1993).

The number count predictions are small compared to the observed counts, therefore it is not surprising that only one object like F10214+4724 has so far been discovered by serendipitous spectroscopic follow-up studies of faint field galaxies. Systematic surveys of 2.2µm-selected objects in the 15 to 18th magnitude range, where such objects could account for 10% of the sample, might be the most fruitful.

To summarize, the main result of the model presented here is that it is quite possible that a large fraction of the light of forming galaxies is hidden in the far-infrared wavelength region. Thompson and Djorgovski (1992) and Franceschini et al. (1991) concluded from their models that to hide galaxy formation in the far-infrared would require quite low values of z_f (z_f < 5) and/or warm dust temperatures. Both of these requirements are the result of the FIRAS limits. The F10214+4724 model is consistent with these results because this object does indeed contain relatively warm dust: Downes et al. (1992) derive a dust temperature of 80K for the far-infrared/submillimeter emission. Thus this model demonstrates the plausibility of a significant background from protogalaxies in the far-infrared most convincingly since it based on the real SED of a dust-rich, star-forming galaxy with known luminosity at a known (cosmological) redshift, rather than on a local universe analog or a theoretical thermal spectrum. In particular, the > 100µm spectral shape, which is a critical constraint compared to the FIRAS observations, has been directly measured for this object.