The application of the Gunn-Peterson constraint on the amount of smoothly distributed neutral material along the line of sight to distant objects requires the hydrogen component of the diffuse IGM to have been highly ionized by z 5 , and the helium component by z 2.5 . From QSO absorption studies we also know that neutral hydrogen at early epochs accounts for only a small fraction, ~ 10%, of the nucleosynthetic baryons . It thus appears that substantial sources of ultraviolet photons were present at z 5, perhaps low-luminosity quasars  or a first generation of stars in virialized dark matter halos with Tvir ~ 104-105 K , , . Early star formation provides a possible explanation for the widespread existence of heavy elements in the IGM , while reionization by QSOs may produce a detectable signal in the radio extragalactic background at meter wavelengths . Establishing the character of cosmological ionizing sources is an efficient way to constrain competing models for structure formation in the universe, and to study the collapse and cooling of small mass objects at early epochs.
Figure 4. Left: comoving space density of bright QSOs as a function of redshift. The data points with error bars are taken from  (filled dots),  (filled squares),  (crosses), and  (filled pentagon). The empty triangles show the space density of the Parkes flat-spectrum radio-loud quasars with P > 7.2 x 1026 W Hz-1 sr-1 . Right: comoving emission rate of hydrogen Lyman-continuum photons (solid line) from QSOs, compared with the minimum rate (dashed line) which is needed to fully ionize a fast recombining (with ionized gas clumping factor C = 30) EdS universe with bh2 = 0.02. Models based on photoionization by quasar sources appear to fall short at z ~ 5. The data point shows the estimated contribution of star-forming galaxies at z 3, assuming that the fraction of Lyman continuum photons which escapes the galaxy HI layers into the intergalactic medium is fesc = 0.5 (see  for details).
What keeps the universe ionized at z = 5? The problem can be simplified by noting that the breakthrough epoch (when all radiation sources can see each other in the Lyman continuum) occurs much later in the universe than the overlap epoch (when individual ionized zones become simply connected and every point in space is exposed to ionizing radiation). This implies that at high redshifts the ionization equilibrium is actually determined by the instantaneous UV production rate . The fact that the IGM is rather clumpy and still optically thick at overlapping, coupled to recent observations of a rapid decline in the space density of radio-loud quasars and of a large population of star-forming galaxies at z 3, has some interesting implications for rival ionization scenarios and for the star formation activity at < 3 < z < 5.
The existence of a decline in the space density of bright quasars at redshifts beyond ~ 3 was first suggested by , and has been since then the subject of a long-standing debate. In recent years, several optical surveys have consistently provided new evidence for a turnover in the QSO counts , , , . The interpretation of the drop-off observed in optically selected samples is equivocal, however, because of the possible bias introduced by dust obscuration arising from intervening systems. Radio emission, on the other hand, is unaffected by dust, and it has recently been shown  that the space density of radio-loud quasars also decreases strongly for z > 3. This argues that the turnover is indeed real and that dust along the line of sight has a minimal effect on optically-selected QSOs (Figure 4, left). The QSO emission rate (corrected for incompleteness) of hydrogen ionizing photons per unit comoving volume is shown in Figure 4 (right) .
Galaxies with ongoing star-formation are another obvious source of Lyman
continuum photons. Since the
rest-frame UV continuum at 1500 Å (redshifted into the visible band for a
source at z 3) is
dominated by the same short-lived, massive stars
which are responsible for the emission of photons shortward of the Lyman edge,
the needed conversion factor, about one ionizing photon every 10 photons at
1500 Å, is fairly insensitive to the assumed IMF and is independent of the
galaxy history for t >> 107 yr.
Figure 4 (right) shows the estimated
Lyman-continuum luminosity density of galaxies at z
3. (3) The data
point assumes a value of
fesc = 0.5 for the unknown fraction of ionizing
photons which escapes
the galaxy HI layers into the intergalactic medium.
A substantial population of dwarf galaxies below the detection threshold,
i.e. having star-formation rates < 0.3 M yr-1, and with a
space density in excess of
that predicted by extrapolating to faint magnitudes the best-fit
Schechter function, may be expected to form at early times in hierarchical
clustering models, and has been recently proposed by
as a possible candidate for photoionizing the IGM at these epochs. One
should note that, while highly reddened galaxies at high
redshifts would be missed by the dropout color technique (which isolates
sources that have blue colors in the optical and a sharp drop in the
rest-frame UV), it seems unlikely that very dusty objects (with
<< 1) would contribute in any significant manner to the ionizing
3 At all ages 0.1 Gyr one has L(1500) / L(912) 6 for a Salpeter mass function and constant SFR . This number neglects any correction for intrinsic HI absorption. Back.