As discussed above, broad features in the SEDs of galaxies allow for the determination of photometric redshifts, and for a very successful pre-selection of candidate galaxies at 
for follow-up spectroscopy. The same principle can also be used to select a special subset of galaxies at much higher redshifts. A star-forming galaxy has a SED roughly flat down to the Lyman limit at
, beyond which there is a prominent break due to the spectra of the stellar population (see the spectra of the O9 and B0 stars in Fig. 1) and to intervening absorption. Physically this reflects the large ionization cross-section of neutral hydrogen. A galaxy revealing a pronounced break at the Lyman limit is called a Lyman-break galaxy (LBG), and is characterized by a relatively high star-formation rate.
For a LBG at 
, the Lyman break falls in between the U and B bands (see Fig. 2.34). Therefore, by selecting those galaxies in a deep multi-color survey that are undetected (or extremely faint) in the U band, but detected in the B and redder bands, one can select candidate star-forming galaxies in the redshift range 
(Steidel et al., 1996). Galaxies selected this way are called UV drop-outs. Follow-up spectroscopy of large samples of UV drop-out candidates has confirmed that this Lyman-break technique is very effective, with the vast majority of the candidates being indeed star-forming galaxies at .
To date more than 1,000 LBGs with 
have been spectroscopically confirmed. The comoving number density of bright LBGs is estimated to be comparable to that of present-day bright galaxies. However, contrary to typical bright galaxies at corresponding to rest-frame UV at z ~ 3) with inferred star-formation rates in the range of a few times 
up to 
depending on the uncertain amount of dust extinction (Adelberger & Steidel, 2000).
The Lyman break (or drop-out) technique has also been applied to deep imaging surveys in reder bands to select galaxies that drop out of the B band, V band and even the I band. If these are indeed LBGs, their redshifts correspond to z ~ 4, z ~ 5, and z ~ 6, respectively. Deep imaging surveys with the HST and ground-based telescopes have already produced large samples of these drop-out galaxies. Unfortunately, most of these galaxies are too faint to follow-up spectroscopically, so that it is unclear to what extent these samples are contaminated by low redshift objects. With this caveat in mind, the data have been used to probe the evolution of the galaxy luminosity function (LF) in the rest-frame UV all the way from z ~ 0 (using data from the GALEX satellite) to z ~ 6. Over the redshift range 4 ≲ z ≲ 6 this LF is found to have an extremely steep faint-end slope, while the characteristic luminosity 
is found to brighten significantly from z = 6 to z = 4 (Bouwens et al., 2007).
Fig.1 Spectra for stars of different spectral types. fλ is the flux per angstrom, and an arbitrary constant is added to each spectrum to avoid confusion. [Based on data kindly provided by S. Charlot]
Fig. 2.. An illustration of how the ‘Lyman-break’ or ‘drop-out’ technique can be used to select starforming galaxies at redshifts z ∼ 3. The spectrum of a typical star-forming galaxy has a break at the Lyman limit (912A˚), which is redshifted to a wavelength λ ∼ 4000A˚ if the galaxy is at z ∼ 3. As a result, the galaxy appears very faint (or may even be undetectable) in the U band, but bright in the redder bands. [Courtesy of M. Dickinson; see Dickinson (1998)]
