9. The Universe at high redshift

The method. The breakthrough was obtained with a method that became known as the Lyman-break method. Since hydrogen is so abundant and its ionization cross section so large, one can expect that photons with λ < 912 Å are very heavily absorbed by neutral hydrogen in its ground state. Therefore, photons with λ < 912 Å have a low probability of escaping from a galaxy without being absorbed.

Intergalactic absorption also contributes. In Sect. 5.​7 we saw that each QSO spectrum features a Lyα forest and often also Lyman-limit absorption. The intergalactic gas absorbs a large fraction of photons emitted by a high-redshift source at λ < 1216 Å, and virtually all photons with a rest-frame wavelength λ ≲ 912 Å. As also discussed in Sect. 8.​5.​2, the strength of this absorption increases with increasing redshift. Combining these facts, we conclude that spectra of high-redshift galaxies should display a distinct feature—a ‘break’—at λ = 912 Å for redshifts z ≲ 4, and for higher redshifts, the break shifts more towards λ = 1216 Å. Furthermore, radiation with λ ≲ 912 Å should be strongly suppressed by intergalactic absorption, as well as by absorption in the interstellar medium of the galaxies themselves, so that only a very small fraction of these ionizing photons will reach us.

From this, a strategy for the detection of galaxies at z ≳ 3 emerges. We consider three broad-band filters with central wavelengths λ ₁ < λ ₂ < λ ₃, where their spectral ranges are chosen to not (or only marginally) overlap. If λ ₁ ≲ (1 + z)912 Å ≲ λ ₂, a galaxy containing young stars should appear relatively blue as measured with the filters λ ₂ and λ ₃, and be virtually invisible in the λ ₁-filter: because of the absorption, it will drop out of the λ ₁-filter (see Fig. 9.2). For this reason, galaxies that have been detected in this way are called Lyman-break galaxies (LBGs) or drop-outs. An example of this is displayed in Fig. 9.3.

Large samples of LBGs. The method was first applied systematically in 1996, using the filters specified in Fig. 9.2. As can be read from Fig. 9.4, the expected location of a galaxy at z ∼ 3 in a color-color diagram with this set of filters is nearly independent of the type and star formation history of the galaxy. Hence, sources in the relevant region of the color-color diagram are very good candidates for being galaxies at z ∼ 3. The redshift needs to be verified spectroscopically, but the crucial point is that the color selection of candidates yields a very high success rate per observed spectrum, and thus spectroscopic observing time at the telescope is spent very efficiently in confirming the redshift of distant galaxies. With the commissioning of the Keck telescope (and later also of other telescopes of the 10-m class), spectroscopy of galaxies with B ≲ 25 became possible (see Fig. 9.5). Employing this method, thousands of galaxies with 2. 5 ≲ z ≲ 3. 5 have been detected and spectroscopically verified to date.

From the spectra shown in Fig. 9.5, it also becomes apparent that not all galaxies that fulfill the selection criteria also show a Lyα emission line, which provides one of the explanations for the lack of success in earlier searches for high-redshift galaxies using narrow-band filters. The spectra of the high-redshift galaxies which were found by this method are very similar to those of starburst galaxies at low redshift. It should come as no surprise that the galaxies selected by the drop-out technique feature active star formation, since it was required that the spectrum on the red side of the break—i.e., at (rest-frame) wavelengths above 1216 Å—shows a blue spectrum. Such a blue spectrum in the rest-frame UV is produced only by a stellar population which features active star formation. Furthermore, the luminosity of galaxies in the rest-frame UV and blue range strongly depends on the star-formation rate, so that preferentially galaxies with the highest (unobscured—see below) star-formation rate are selected.

This is a prominent example of the effect that the physical properties of objects selected depend on the selection criteria. One must always bear in mind that, when comparing galaxy populations detected by different methods, the properties can differ substantially. One of the challenges of studies of (high-redshift) galaxies is to get a coherent picture of the galaxy population from samples with a vast variety of selection methods.

The correlation function and halo masses of LBGs. For a large variety of objects, and over a broad range of separations, the spatial correlation function of objects can be described by the power law (7.​28), with a slope of typically γ ∼ 1. 7. However, the amplitude of this correlation function varies between different classes of objects. For example, we saw in Sect. 8.​2.​4 that the amplitude of the power spectrum of galaxy clusters is larger by about a factor 7 than that of galaxies (see Fig. 8.​23); the same ratio holds of course for the corresponding correlation functions. As we argued there, the strength of the correlation depends on the mass of objects; in the simple picture of biasing shown in Fig. 7.​22, the correlation of objects is larger the rarer they are. High-mass peaks exceeding the density threshold needed for gravitational collapse have a lower mean number density than low-mass peaks, so they are therefore expected to be more biased (see Sect. 8.​1.​3) and thus more strongly correlated.

If we now assume that each galaxy lives in a dark matter halo, we can estimate the dark halo mass from the observed correlation function of these galaxies. As we discussed in Sect. 7.​6.​3, dark matter halos have clustering properties which differ from the clustering of the underlying matter density field, and we described that in terms of the halo biasing b _h(M, z), which is a function of halo mass and redshift. The dark matter correlation function can be determined quite accurately from numerical simulations. The ratio of the observed correlation function to the dark matter correlation function then yields the square of the halo bias parameter (7.​68), and comparing that to the numerically-determined function b _h(M, z), the corresponding halo mass can be obtained.

Considering the spatial distribution of LBGs, we find a large correlation amplitude. The (comoving) correlation length of LBGs at redshifts 1. 5 ≲ z ≲ 3. 5 is r ₀ ∼ 4. 2h ⁻¹ Mpc, i.e., not very different from the correlation length of L _∗-galaxies in the present Universe. Since the bias factor of present-day galaxies is about unity, implying that they are clustered in a similar way as the dark matter distribution, this result then implies that the bias of LBGs at high redshift must be considerably larger than unity. This conclusion is based on the fact that the dark matter correlation at high redshifts (on large scales, i.e., in the linear regime) was smaller than today by the factor D ₊ ²(z), where D ₊ is the growth factor of linear perturbations introduced in Sect. 7.​2.​2. Thus we conclude that LBGs are rare objects and thus correspond to high-mass dark matter halos. Comparing the observed correlation length r ₀ with numerical simulations, the characteristic halo mass of LBGs can be determined, yielding ∼ 3 × 10¹¹ M _⊙ at redshifts z ∼ 3, and ∼ 10¹² M _⊙ at z ∼ 2. Furthermore, the correlation length is observed to increase with the luminosity of the LBG, indicating that more luminous galaxies are hosted by more massive halos, which are more strongly biased than less massive ones. If these results are combined with the observed correlation functions of galaxies in the local Universe and at z ∼ 1, and with the help of numerical simulations, then this indicates that a typical high-redshift LBG will evolve into a massive elliptical galaxy by today.

Proto-clusters. Furthermore, the clustering of LBGs shows that the large-scale galaxy distribution was already in place at high redshifts. In some fields the observed overdensity in angular position and galaxy redshift is so large that one presumably observes galaxies which will later assemble into a galaxy cluster—hence, we observe some kind of proto-cluster. We have already shown such a proto-cluster in Fig. 6.​71. Galaxies in such a proto-cluster environment seem to have about twice the stellar mass of those LBGs outside such structures, and the age of their stellar population appears older by a factor of two. This result indicates that the stellar evolution of galaxies in dense environments proceeds faster than in low-density regions, in accordance with expectations from structure formation. It also reveals a dependence of galaxy properties on the environment, which we have seen before manifested in the morphology-density relation (see Sect. 6.​7.​2). Proto-clusters of galaxies have also been detected at higher redshifts up to z ∼ 6, using narrow-band imaging searches for Lyman-alpha emission galaxies (see below).

Satellite galaxies at high redshifts. Whereas the clustering of LBGs is well described by the power law (7.​28) over a large range of scales, the correlation function exhibits a significant deviation from this power law at very small scales: the angular correlation function exceeds the extrapolation of the power law from larger angles at Δ θ ≲ 7″, corresponding to comoving length-scales of ∼ 200 kpc. It thus seems that this scale marks a transition in the distribution of galaxies. To get an idea of the physical nature of this transition, we note that this length-scale is about the virial radius of a dark matter halo with M ∼ 3 × 10¹¹ M _⊙, i.e., the mass of halos which host the LBGs. On scales below this virial radius, the correlation function thus no longer describes the correlation between two distinct dark matter halos. An interpretation of this fact is provided in terms of merging: when two galaxies and their dark matter halos merge, the resulting dark matter halo hosts both galaxies, with the more massive one close to the center and the other one as ‘satellite galaxy’. The correlation function on scales below the virial radius thus indicates the clustering of galaxies within the same halo, whereas on larger scales, where it follows the power-law behavior, it indicates the correlation between different halos. Note that this effect is also well described in the halo model which we discussed in Sect. 7.​7.​3. On large scales, the correlation function is dominated by the two-halo term, whereas on smaller scales, the one-halo term takes over. The transition between these two regimes, which at low redshifts occurs on scales of several hundred kiloparsecs (see Fig. 7.​27), is at smaller scales for high-redshift galaxies, since the high-mass population of galaxy clusters has not formed yet at these early epochs.

Winds of star-forming galaxies. The inferred high star-formation rates of LBGs implies an accordingly high rate of supernova explosions. These release part of their energy in the form of kinetic energy to the interstellar medium in these galaxies. This process will have two consequences. First, the ISM in these galaxies will be heated locally, which slows down (or prevents) further star formation in these regions. This thus provides a feedback effect for star formation which prevents all the gas in a galaxy from turning into stars on a very short time-scale, and is essential for understanding the formation and evolution of galaxies, as we shall see in Sect. 10.​4.​4. Second, if the amount of energy transferred from the SNe to the ISM is large enough, a galactic wind may be launched which drives part of the ISM out of the galaxy into its halo. Evidence for such galactic winds has been found in nearby galaxies, for example from neutral hydrogen observations of edge-on spirals which show an extended gas distribution outside the disk. Furthermore, the X-ray corona of spirals (see Fig. 3.​26) is most likely linked to a galactic wind in these systems.

Indeed, there is now clear evidence for the presence of massive winds from LBGs. The spectra of LBGs often show strong absorption lines, e.g., of Civ, which are blueshifted relative to the velocity of the emission lines in the galaxy. An example of this effect can be seen in the spectra of Fig. 9.5, where in the upper panel the emission line of Civ is accompanied by an absorption to the short-wavelength side of the emission line. Such absorption can be produced by a wind moving out from the star-forming regions of the galaxy, so that its redshift is smaller than that of the emission regions. Characteristic velocities are ∼ 200 km∕s. In one case where the spectral investigation has been performed in most detail (the LBG cB58; see Fig. 9.17), the outflow velocity is ∼ 255 km∕s, and the outflowing mass rate exceeds the star-formation rate. Whereas these observations clearly show the presence of outflowing gas, it remains undetermined whether this is a fairly local phenomenon, restricted to the star-formation sites, or whether it affects the ISM of the whole galaxy.

Connection to QSO absorption lines. A slightly more indirect argument for the presence of strong winds from LBGs comes from correlating the absorption lines in background QSO spectra with the position of LBGs. These studies have shown that whenever the sight-line of a QSO passes within ∼ 40 kpc of an LBG, very strong Civ absorption lines (with column density exceeding 10¹⁴ cm⁻²) are produced, and that the corresponding absorbing material spans a velocity range of Δ v ≳ 250 km∕s; for about half of the cases with impact parameters within 80 kpc, strong Civ absorption is produced. This frequency of occurrence implies that about 1/3 of all Civ metal absorption lines with N ≳ 10¹⁴ cm⁻² in QSO spectra are due to gas within ∼ 80 kpc from those LBGs which are bright enough to be included in current surveys. It is plausible that many of the remaining 2/3 are due to fainter LBGs.

The association of Civ absorption line systems with LBGs by itself does not prove the existence of winds in such galaxies; in fact, the absorbing material may be gas orbiting in the halo in which the corresponding LBG is embedded. In this case, no outflow phenomenon would be implied. However, in that case one might wonder where the large amount of metals implied by the QSO absorption lines is coming from. They could have been produced by an earlier epoch of star formation, but in that case the enriched material must have been expelled from its production site in order to be located in the outer part of z ∼ 3 halos. It appears more likely that the production of metals in QSO absorption systems is directly related to the ongoing star formation in the LBGs. We shall see in Sect. 9.3.5 that clear evidence for superwinds has been discovered in one massive star-forming galaxy at z ∼ 3.

Finally, we mention another piece of evidence for the presence of superwinds in star-forming galaxies. There are indications that the density of absorption lines in the Lyα forest is reduced when the sight-line to the QSO passes near a foreground LBG. This may well be explained by a wind driven out from the LBG, pushing neutral gas away and thus leaving a gap in the Lyα forest. The characteristic size of the corresponding ‘bubbles’ is estimated to be ∼ 0. 5 Mpc for luminous LBGs.

Lyman-break galaxies at low redshifts. One might ask whether galaxies similar to the LBGs at z ∼ 3 exist in the current Universe. Until recently this question was difficult to investigate, since it requires imaging of lower redshift galaxies at ultraviolet wavelengths. With the launch of GALEX an appropriate observatory became available with which to observe galaxies with restframe UV luminosities similar to those of LBGs. UV-selected galaxies show a strong inverse correlation between the stellar mass and the surface brightness in the UV. Lower-mass galaxies are more compact than those of higher stellar mass. On the basis of this correlation we can consider the population of large and compact UV-selected galaxies separately. The larger ones show a star-formation rate of a few M _⊙∕yr; at this rate, their stellar mass content can be built up on a time-scale comparable to the Hubble time, i.e., the age of the Universe. These galaxies are typically late-type spiral galaxies, and they show a metallicity similar to our Galaxy.

In contrast, the compact galaxies have a lower stellar mass and about the same star-formation rate, which allows them to generate their stellar population much faster, in about 1 Gyr. Compared to normal low-redshift galaxies, their metallicity is smaller by about a factor of 2 for a given stellar mass. In addition, they show similar extinction and outflow properties as the LBG at z ∼ 3. Hence, the properties of the compact UV-selected galaxies, which are sometimes called Lyman-break analogs , are quite similar to those of the LBGs seen at higher redshifts, and they may indeed be closely related to the LBG population.

Lyman-break galaxies at high redshift. By variation of the filter set, drop-outs can also be discovered at larger wavelengths, thus at accordingly higher redshifts. The object selection at higher z implies an increasingly dominant role of the Lyα forest whose density is a strongly increasing function of redshift (see Sect. 8.​5.​2). This method has been routinely applied with ground-based observations up to z ∼ 4. 5, yielding so-called B-drop-outs. Galaxies at considerably higher redshifts are difficult to access from the ground with this method. One reason for this is that galaxies become increasingly faint with redshift, rendering observations substantially more difficult. Furthermore, one needs to use increasingly redder filter sets. At such large wavelengths the night sky gets significantly brighter, which further hampers the detection of very faint objects. For detecting a galaxy at redshift, say, z = 5. 5 with this method, the Lyα line, now at λ ≈ 7900 Å, is located right in the I-band, so that for an efficient application of the drop-out technique only the I- and z-band filters or NIR-filters are viable, and with those filters the brightness of the night sky is very problematic (see Fig. 9.6 for an example of a drop-out galaxy at very high redshift). Furthermore, candidate very high-redshift galaxies detected as drop-outs are very difficult to verify spectroscopically due to their very faint flux and the fact that most of the diagnostic spectral features are shifted to the near-IR. In spite of this, we will see later that the drop-out method has achieved spectacular results even at redshifts considerably higher than z ∼ 4, where the HST played a central role.

Spectral breaks. The Lyman-break technique is a special case of a method for estimating the redshift of galaxies (and QSOs) by multi-color photometry. This technique can be employed due to the spectral breaks at λ = 912 Å and λ = 1216 Å, respectively. Spectra of galaxies also show other characteristic features. As was discussed in detail in Sect. 3.​5, the broad-band energy distribution is basically a superposition of stellar radiation (if we ignore for a moment the presence of dust, which can yield a substantial infrared emission from galaxies). A stellar population of age ≳ 10⁸ yr features a 4000 Å-break because, due to a sudden change in the opacity at this wavelength, the spectra of most stars show such a break at about 4000 Å (see Fig. 3.​33). Hence, the radiation from a stellar population at λ < 4000 Å is less intense than at λ > 4000 Å; this is the case particularly for early-type galaxies, as can be seen in Fig. 3.​36, due to their old stellar population.

Principle of the method. If we assume that the star-formation histories of galaxies are not too diversified, the spectral energy distributions of these galaxies are expected to follow certain patterns. For example, if all galaxies had a single episode of star formation, starting at redshift z _f and lasting for a time τ, then the spectra of these galaxies, for a given initial mass function, would be characterized by these two parameters, as well as the total stellar mass formed (see Sect. 3.​5); this latter quantity then yields the amplitude of the spectrum, but does not affect the spectral shape. When measuring the magnitude of these galaxies in n broad-band filters, we can form n − 1 independent colors. Next suppose we form a multi-dimensional color-color diagram, in which every galaxy is represented by a point in this (n − 1)-dimensional color space. Considering only galaxies at the present epoch, all these points will lie on a two-dimensional surface in this multi-dimensional space, instead of being more or less randomly distributed.

Next, instead of plotting z = 0 galaxies, we consider the distribution of galaxies at some higher redshift z < z _f. This distribution of points will be different, mainly due to two different effects. First, a given photometric filter corresponds to a different rest-frame spectral range of the galaxy, due to redshift. Second, the ages of the stellar populations are younger at an earlier cosmic epoch, and thus the spectral energy distributions are different. Both of these effects will cause these redshift z galaxies to occupy a different two-dimensional surface in multi-color space. Generalizing this argument further, we see that in this idealized consideration, galaxies will occupy a three-dimensional subspace in (n − 1)-dimensional color space, parametrized by formation redshift z _f, time-scale τ and the galaxy’s redshift z. Hence, from the measurement of the broad-band energy distribution of a galaxy, we might expect to be able to determine, or at least estimate, its redshift, as well as other properties such as the age of its stellar population; this is the principle of the method of photometric redshifts.

Of course, the situation is considerably more complicated in reality. Galaxies most likely have a more complicated star-formation history than assumed here, and hence they will not be confined to a two-dimensional surface at fixed redshift, but instead will be spread around this surface. The spectrum of a stellar population also depends on its metallicity, as well as absorption, either by gas and dust in the interstellar medium or hydrogen in intergalactic space (of which the Lyman-break method makes proper use). On the other hand, we have seen in Sect. 3.​6 that the colors of current-day galaxies are remarkably similar, best indicated by the red sequence. Therefore, the method of photometric redshifts may be expected to work, even if more complications are accounted for than in the idealized example considered above.

The method is strongly aided if the galaxies have characteristic spectral features, which shift in wavelength as the redshift is changed. If, for example, the spectrum of a galaxy was a power law in wavelength, then the redshifted spectrum would as well be a power law, with the same spectral slope—if we ignore the different age of the stellar population. Therefore, for such a spectral energy distribution is would be impossible to estimate a redshift. However, if the spectrum shows a clear spectral break, then the location of this break in wavelength depends directly on the redshift, thus yielding a particularly clean diagnostic. In this context the 4000 Å-break and the Lyα-break play a central role, as is illustrated in Fig. 9.7.

Calibration. In order to apply this method, one needs to find the characteristic domains in color space where (most of) the galaxies are situated. This can be done either empirically, using observed energy distributions of galaxies, or by employing population synthesis model. More precisely, a number of standard spectra of galaxies (so-called templates) are used, which are either selected from observed galaxies or computed by population synthesis models. Each of these template spectra can then be redshifted in wavelength. For each template spectrum and any redshift, the expected galaxy colors are determined by integrating the spectral energy distribution, multiplied by the transmission functions of the applied filters, over wavelength [see (A.25)]. This set of colors can then be compared with the observed colors of galaxies, and the set best resembling the observation is taken as an estimate for not only the redshift but also the galaxy type.

Practical considerations. The advantage of this method is that multi-color photometry is much less time-consuming than spectroscopy of galaxies. Whereas some modern spectrographs allow one to take spectra of ∼ 1000 objects simultaneously, images taken with wide-field cameras of ∼ 1 deg² on 4-m class telescopes record the fluxes of ∼ 10⁵ galaxies in a one hour exposure. In addition, this method can be extended to much fainter magnitudes than are achievable for spectroscopic redshifts. The disadvantage of the method becomes obvious when an insufficient number of photometric bands are available, since then the photometric redshift estimates can yield a completely wrong z; these are often called catastrophic outliers. One example for the occurrence of extremely wrong redshift estimates is provided by a break in the spectral energy distribution. Depending of whether this break is identified as the Lyman-break or the 4000 Å-break, the resulting redshift estimates will be very different. To break the corresponding degeneracy, a sufficiently large number of filters spread over a broad spectral range must be available to probe the spectral energy distribution over a wide range in wavelengths. As a general rule, the more photometric bands are available and the smaller the uncertainties in the measured magnitudes, the more accurate the estimated redshift. Normally, data from four or five photometric bands are required to obtain useful redshift estimates. In particular, the reliability of the photometric redshift benefits from data over a large wavelength range, so that a combination of several optical and NIR filters is desirable.

The successful application of this method also depends on the type of the galaxies. As we have seen in Sect. 6.​8, early-type galaxies form a relatively well-defined color-magnitude sequence at any redshift, due to their old stellar populations (manifested in clusters of galaxies in form of the red cluster sequence), so that the redshift of this type of galaxy can be estimated very accurately from multi-color information. However, this is only the case if the 4000 Å-break is located in between two of the applied filters. For z ≳ 1 this is no longer the case if only optical filters are used. Other types of galaxies show larger variations in their spectral energy distribution, depending, e.g., on the star formation history, as mentioned before.

Photometric redshifts are particularly useful for statistical purposes, for instance in situations in which the exact redshift of each individual galaxy in a sample is of little relevance. However, by using a sufficient number of optical and NIR filters, quite accurate redshift estimates for individual galaxies are achievable. For example, with eight optical and NIR bands and accurate photometry, a redshift accuracy of Δ z ∼ 0. 03(1 + z) was obtained, as demonstrated in Fig. 9.8 by a comparison of photometric redshifts with redshifts determined spectroscopically for galaxies in the field of the HDF-North. If data in additional photometric bands are available, e.g., using filters of smaller transmission curves (‘intermediate width filters’), the redshift accuracy can be increased even more, e.g., Δ z ∼ 0. 01(1 + z) was obtained using a total of 30 photometric bands.

The Lyman-break technique is a special case of the photometric redshift method; it relies on only three photometric bands to select galaxies in a given redshift range, whereas in general, more bands are needed to obtain reliable redshift estimates. There are other cases where a few bands are sufficient for a fairly reliable selection of particular kinds of galaxies, or particular redshift regimes, some of which should be mentioned here.

Selection of 1. 5 ≲ z ≲ 2. 5 galaxies. The success of the Lyman-break method is rooted in the fact that the observed colors of star-forming galaxies in a carefully selected triplet of filters is essentially independent on details of the star-formation history, metallicity and other effects, due to a very strong spectral break. This is illustrated in Fig. 9.4. The same figure also shows that the colors of galaxies with somewhat lower redshift are also very similar; for example, one sees that galaxies with z ∼ 1. 8 all have U _n − G ∼ 0 and . At that redshift, the Lyα-line is shortward of the U _n-band filter, and thus a star-forming galaxy has a rather flat spectrum across all three filters. As the redshift increases above z ∼ 2, the Lyα-line moves into the U_n-band filter and thus increases the flux there; however, as we have seen, a large fraction of LBGs have rather low Lyα-flux, thus affecting the color only marginally. For redshifts higher than ∼ 2. 5, the break moves into the U_n-band, and the objects redden in U _n − G and move onto the same sequence where LBGs are selected. Thus, with a single set of three filters (and thus the same optical images), one can select galaxies over the broad range of 1. 5 ≲ z ≲ 3. 5.

BzK selection. While the filter combination used for the Lyman-break galaxies selects star-forming galaxies at high redshift, it misses galaxies with a passive stellar population. One has therefore investigated whether another combination of filters, and thus different colors, may be able to identify high-redshift passive galaxies. Indeed, such a filter set was found; the combination of the B-, z- and K-band filters provides a successful tool to search for galaxies with 1. 4 ≲ z ≲ 2. 5, as illustrated in Fig. 9.9. K-band selected galaxies with 1. 4 ≲ z ≲ 2. 5 occupy specific regions in a B − z versus z − K color-color diagram.² In this redshift range, the 4000 Å-break is located redward of the z-band, thus such galaxies display a fairly red z − K color if they are not forming stars at a high rate. The lack of active star formation also causes the B − z color to be rather red, since the B-band probes to the rest-frame UV-region of the spectrum. Such galaxies are located in the upper right corner of the diagram in Fig. 9.9. In case the galaxies in this redshift range are actively forming stars, the 4000 Å-break is weaker, but instead the B − z color is rather blue, so that these galaxies are located in the upper left corner of the diagram. As Fig. 9.9 shows, this selection of high-redshift galaxies is very efficient.

The BzK-selected galaxies with active star formation have redder colors in the rest-frame UV than the Lyman-break galaxies which are selected based on their UV flux, although there is a significant overlap between the two populations in the sense that a substantial fraction of galaxies are found by both methods. However, the most actively star-forming galaxies are missed with the BzK-method since those show little-to-no 4000 Å-break, thus no longer have a sufficiently red z − K color, and would lie below the solid line in Fig. 9.9.

Distant red galaxies. Another method to select high-redshift passive galaxies is based on their rest-frame optical colors. From local galaxies we know that the 4000 Å-break is the most prominent feature in the spectral energy distribution of stellar populations with no or little star formation. At redshifts 2 ≲ z ≲ 4, this break is located between the observed J- and K-band filters; hence we expect that passive galaxies are red in their J − K color. As Fig. 9.10 shows, J − K ≳ 2. 3 as soon as the redshift increases beyond z ≳ 2. Perhaps surprisingly, this is true even if the stellar population is as young as 0. 25 Gyr, for which the redshift of the transition to J − K > 2. 3 occurs at only slighter larger redshift. Furthermore, this color selection is able to find also galaxies with ongoing star formation, provided they also have an old stellar population; this is due to the fact that much of the star formation is accompanied by substantial dust obscuration. At redshift z = 2, the J-band corresponds to the rest-frame B-band, which is substantially affected by extinction, leading to a red J − K color. High-redshift galaxies selected according to J − K > 2. 3 are called distant red galaxies (DRGs). The fact that there is very little overlap in the galaxy population selected according to their UV-properties and the DRG population immediately shows the necessity to apply several very different selection criteria for high-redshift galaxies to obtain a complete census of their population.

Narrow-band selection. We mentioned the method of narrow-band selection before. If a source has a strong emission line, and if the observed wavelength of the emission line matches the spectral response of a narrow-band filter, then the ratio of fluxes obtained in this narrow-band image compared to a broad-band image would be much larger than for other sources without a strong emission line at the corresponding wavelength.

After a substantial population of high-redshift galaxies were found with the Lyman Break technique, it became known that about 60 % of these galaxies show very strong Lyα emission lines. It was then possible to design narrow-band filters that were particularly tuned to detect objects with strong Lyα emission lines at a particular redshift. Several thousand Lyα emitters (LAEs) were detected with this method, extending up to redshift z ∼ 7. These galaxies are on average considerably fainter than LBGs, and therefore allow one to probe the fainter end of the luminosity function of star-forming galaxies. Their faintness, on the other hand, make more detailed spectroscopic studies very challenging, and thus the relation of these Lyα emitters to the other galaxy populations at similar redshifts is not easy to determine.

Furthermore, candidate objects detected in narrow-band images require spectroscopic follow-up, since there are many possible contaminants that may enter the selection. Galaxies, and in particular AGNs, at lower redshifts can display strong emission lines of other atomic transitions and need to be ruled out with a spectrum. Due to the cumulative effect of the Lyα forest, a high-redshift (z ≳ 4) Lyα emitter should show essentially no flux at shorter wavelengths, and so some of the Lyα emission-line candidates can be rejected if continuum flux bluewards of the narrow band is detected.

The Hubble Deep Field North. In 1995, an unprecedented observing program was conducted with the HST. A deep image in four filters (U₃₀₀, B₄₅₀, V₆₀₆, and I₈₁₄) was observed with the Wide Field/Planetary Camera 2 (WFPC2) on-board HST, covering a field of ∼ 5. 3 arcmin², with a total exposure time of about 10 days. This resulted in the deepest sky image of that time, displayed in Fig. 1.​37. The observed field was carefully selected such that it did not contain any bright sources. Furthermore, the position of the field was chosen such that the HST was able to continually point into this direction, a criterion excluding all but two relatively small regions on the sky, due to the low HST orbit around the Earth. Another special feature of this program was that the data became public right after reduction, less than a month after the final exposures had been taken. Astronomers worldwide immediately had the opportunity to scientifically exploit these data and to compare them with data at other frequency ranges or to perform their own follow-up observations. Such a rapid and wide release was uncommon at that time, but is now seen more frequently. Rarely has a single data set inspired and motivated a large community of astronomers as much as the Hubble Deep Field (HDF) did (after another HDF was observed in the Southern sky—see below—the original HDF was called HDF North, or HDFN).

Follow-up observations of the HDF were made in nearly all accessible wavelength ranges, so that it became the best-observed region of the extragalactic sky. Within a few years, more than ∼ 3000 galaxies, 6 X-ray sources, 16 radio sources, and fewer than 20 Galactic stars were detected in the HDF, and redshifts were determined spectroscopically for more than 150 galaxies in this field, with about 30 at z > 2. Never before could galaxy counts be conducted to magnitudes as faint as it became possible in the HDF (see Fig. 9.11); several hundred galaxies per square arcminute could be photometrically analyzed in this field.

Detailed spectroscopic follow-up observations were conducted by several groups, through which the HDF became, among other things, a calibration field for photometric redshifts (see, for instance, Fig. 9.8). Most galaxies in the HDF are far too faint to be analyzed spectroscopically, so that one often has to rely on photometric redshifts.

HDFS and the Hubble Ultra Deep Field. Later, in 1998, a second HDF was observed, this time in the southern sky. In contrast to the HDFN, which had been chosen to be as empty as possible, the HDFS contains a QSO. Its absorption line spectrum can be compared with the galaxies found in the HDFS, by which one hopes to obtain information on the relation between QSO absorption lines and galaxies. In addition to the WFPC2 camera, the HDFS was simultaneously observed with the cameras STIS (51″ × 51″ field-of-view, where the CLEAR ‘filter’ was used, which has a very broad spectral sensitivity; in total, STIS is considerably more sensitive than WFPC2) and NICMOS (a NIR camera with a maximum field-of-view of 51″ × 51″) which had both been installed in the meantime. Nevertheless, the overall impact of the HDFS was smaller than that of the HDFN; one reason for this may be that the requirement of the presence of a QSO, combined with the need for a field in the continuous viewing zone of HST, led to a field close to several very bright Galactic stars. This circumstance makes photometric observations from the ground very difficult, e.g., due to stray-light.

One of the immediate results from the HDF was the finding that the morphology of faint galaxies is quite different from those in the nearby Universe. Locally, most luminous galaxies fit into the morphological Hubble sequence of galaxies. This ceases to be the case for high-redshift galaxies. In fact, galaxies at z ∼ 2 are much more compact than local luminous galaxies, they show irregular light distributions and do not resemble any of the Hubble sequence morphologies. By redshifts z ∼ 1, the Hubble sequence seems to have been partly established.

In 2002, an additional camera was installed on-board HST. The Advanced Camera for Surveys (ACS) has, with its side length of 3.​​^′4, a field-of-view about twice as large as WFPC2, and with half the pixel size (0.​​^″05) it better matches the diffraction-limited angular resolution of HST. Therefore, ACS is a substantially more powerful camera than WFPC2 and is, in particular, best suited for surveys. With the Hubble Ultra Deep Field (HUDF), the deepest image of the sky was observed and published in 2004 (see Fig. 9.12). The HUDF is, in all four filters, deeper by about one magnitude than the HDF, reaching a magnitude limit of m _AB ≈ 29. The depth of the ACS images in combination with the relatively red filters that are available provides us with an opportunity to identify drop-out candidates out to redshifts z ∼ 6; quite a number of such candidates have already been verified spectroscopically.

Lyman-break galaxies at z ∼ 6 seem to have stellar populations with masses and lifetimes comparable to those at z ∼ 3. This implies that at a time when the Universe was 1 Gyr old, a stellar population with mass ∼ 3 × 10¹⁰ M _⊙ and age of a few hundred million years (as indicated by the observed 4000 Å break) was already in place. This, together with the apparently high metallicity of these sources, is thus an indication of how quickly the early Universe evolved. The z ∼ 6 galaxies are very compact, with half-light radii of ∼ 1 kpc, and thus differ substantially from the galaxy population known in the lower-redshift Universe.

Half of the HUDF was also imaged by the near-IR NICMOS camera onboard HST, but only after the installment

of the Wide-Field Camera 3 (WFC3) on HST, with a near-IR channel and a much larger field-of-view and much better sensitivity than NICMOS, could the HUDF be imaged to comparable sensitivity levels in the NIR as in the optical. WFC3 mapped the HUDF in three NIR bands, Y, J and H, down to a limiting magnitude of m _AB ≈ 28. 5. These long wavelengths allowed the systematic search for Lyman-break galaxies at redshifts beyond 6, as will be discussed below.

In September 2012, the deepest view of the Universe ever taken was released: the eXtreme Deep Field (XDF), shown in Fig. 9.13. It covers about 4. 7 arcmin² in nine optical and NIR filters, reaching a limiting magnitude of m _AB ≈ 30.

Further deep-field projects with HST: GOODS, GEMS, COSMOS. The great scientific harvest from the deep HST images, particularly in combination with data from other telescopes and the readiness to make such data available to the scientific community for multi-frequency analysis, provided the motivation for additional HST surveys. The GOODS (Great Observatories Origins Deep Surveys) project is a joint observational campaign of several observatories, centering on two fields of ∼ 16′ × 10′ size each that have been observed by the ACS camera at several epochs between 2003 and 2005. One of these two regions (GOODS-North) contains the HDFN, the other a field that became known as the Chandra Deep Field South (CDFS), also containing the HUDF. The Chandra satellite observed both GOODS fields with a total exposure time of ∼ 2 × 10⁶ s and ∼ 4 × 10⁶ s, corresponding to about 24 and 48 days, respectively. Also, the Spitzer observatory took long exposures of these two fields, and several ground-based observatories are involved in this survey, for instance by contributing an ultra-deep wide-field image ( ∼ 30′ × 30′) centered on the Chandra Deep Field South and NIR images in the K-band. The data themselves and the data products (like object catalogs, color information, etc.) are all publicly available and have led to a large number of scientific results.

The multi-wavelength approach by GOODS yields an unprecedented view of the high-redshift Universe. Although these studies and scientific analysis are ongoing (at the time of writing), quite a large number of very high-redshift (z ≳ 5–6) galaxies were discovered and studied: a sample of more than 500 I-band drop-out candidates was obtained from deep ACS/HST images.

Even larger surveys were conducted with the HST, including the Galaxy Evolution from Morphology and SEDs (GEMS) survey, covering a field of 30′ × 30′ centered on the CDFS mapped in two filters. For this field, full coverage with a 17-band (5 broad bands, and 12 intermediate width bands) optical imaging survey (COMBO17) is available. The largest contiguous field imaged with the angular resolution of HST is the ∼ 2 deg² COSMOS survey. This sky area was also imaged with other space-based (Spitzer, GALEX, XMM-Newton, Chandra) and ground-based (Subaru, VLA, VLT, UKIRT and others) observatories. Its large field enables the study of the large-scale galaxy and AGN distribution at high redshifts. The broad wavelength coverage from the radio to the X-ray regime, renders the COSMOS field a treasure for observational cosmology for years to come. As discussed in Sect. 8.​4, the COSMOS field was used for a detailed cosmic shear analysis, where the broad wavelength coverage helped enormously to determine accurate photometric redshifts of the source galaxies.

Deep fields have been observed in many other frequency ranges, and as mentioned before, they are often taken on the same sky areas to allow for multi-frequency studies of the sources. We mention here the Chandra Deep Fields, one in the North (CDFN) in the direction of the HDFN, the other in the South (CDFS, see Fig. 9.14) in the direction of the HUDF, with a total exposure time of two and four million seconds, respectively.

With the extremely faint limiting flux of the CDFS, the source density on the sky reaches more than 10⁴ deg⁻², which is comparable to the source density seen in rather shallow optical imaging, like the SDSS images. However, the mean redshift of the X-ray sources is considerably higher than that of the optical sources, due to the large fraction of AGNs. As we can see from Fig. 9.15, the X-ray population of compact sources in ultra-deep X-images is composed of three components. By far dominating is the population of AGNs, whose number counts exhibit the shape of a broken power law—steep at the bright end, flatter at the low-flux end. For high fluxes, they follow approximately the Euclidean slope, , whereas at fluxes below the breakpoint at , one finds for the soft band, and even slightly flatter in the hard band. Going to fainter X-ray fluxes, the fraction of obscured AGNs at high redshifts increases.

Approaching the fainter flux levels, the population of normal galaxies becomes increasingly important. At the limiting flux of the CDFS, they constitute almost 50 % of the source population. They are mainly late-type star-forming galaxies, with the X-ray emission mostly due to X-ray binaries. Early-type galaxies contribute just a small fraction of ∼ 10 % to the galaxy counts, where their X-ray emission is a due to a combination of X-ray binaries and hot gas.

Galaxies at high redshift are faint and therefore difficult to observe spectroscopically. For this reason, the brightest galaxies are preferentially selected (for detailed examination), i.e., basically those which are the most luminous ones at a particular z—resulting in undesired, but hardly avoidable selection effects. For example, those Lyman-break galaxies at z ∼ 3 for which the redshift is verified spectroscopically are typically located at the bright end of the LBG luminosity function. The sensitivity of our telescopes is insufficient in most cases to spectroscopically analyze a rather more typical galaxy at z ∼ 3.

The magnification by gravitational lenses can substantially boost the apparent magnitude of sources; gravitational lenses can thus act as natural (and inexpensive!) telescopes. The most prominent examples are the arcs in clusters of galaxies: many of them have a very high redshift, are magnified by a factor ≳ 5, and hence are brighter by about ≳ 1. 5 mag than they would be without the lens effect (see Fig. 9.16).³ In addition to boosting the observable fluxes, the gravitational lens effect yields a spatial magnification of the sources: the sources appear larger than they really are, thus increasing the effective angular resolution with which they can be observed.

Lyman-break galaxies at z ∼ 3. An extreme first example of this effect is represented by the galaxy cB58 at z = 2. 72, which is displayed in Fig. 9.17. It was discovered in the background of a galaxy cluster and is magnified by a factor ∼ 30. Hence, it appears brighter than a typical Lyman-break galaxy by more than three magnitudes. For this reason, one of the most detailed spectra of all galaxies at z ∼ 3 was taken of this particular source. Several more such examples were found subsequently, including the so-called Cosmic Eye, a Lyman-break galaxy at z = 3. 07, shown in Fig. 9.18. The typical magnification of these lensed LBGs is μ ∼ 30.

Further examples. Furthermore, at least two highly magnified Lyα emitters (LAEs) at z ≈ 5 were found. Whereas the study of LAEs is usually hampered by the faint continuum, the strong magnification (μ > 10 in these two systems) enables an investigation of the underlying stellar population from broad-band photometry. One of these two systems has an estimated stellar mass of ≲ 10⁸ M _⊙, one of the lowest mass systems found so far at high redshifts. Thanks to the increased effective angular resolution provided by gravitational lensing magnification, a spatially resolved view of the star-forming regions in a z = 2. 33 sub-millimeter galaxy was obtained. The observations are compatible with the picture that star formation in this object occurs in the cores of giant molecular clouds, as in the local Universe, but these regions are ∼ 100 times larger than local star-forming sites.

Apparently extreme sources. One can argue that there is a high probability that the flux of the apparently most luminous sources from a particular source population is magnified by lensing. The apparently most luminous IRAS galaxy, F10214+47, is magnified by a factor ∼ 50 by the lens effect of a foreground galaxy, where the exact value of the magnification depends on the wavelength, since the intrinsic structure and size of the source is wavelength-dependent—hence the magnification is differential. Other examples are the QSOs B1422+231 and APM 08279+5255, which are among the brightest QSOs despite their high redshifts; hence, they belong to the apparently most luminous sources in the Universe. In both cases, multiple images of the QSOs were discovered, verifying the action of the lens effect. Their magnification, and therefore their brightness, renders these sources preferred objects for QSO absorption line spectroscopy (see Fig. 5.​55). The Lyman-break galaxy cB58 mentioned previously is another example, and we will see below that the most luminous sub-millimeter sources are gravitationally lensed.

Employing natural telescopes. Most of the examples of highly magnified sources mentioned before were found serendipitously. However, the magnification effect can also be utilized deliberately, by searching for high-redshift sources in regions which are known to produce strong magnification effects, i.e., fields around clusters of galaxies: for a massive cluster, one knows that distant sources located behind the cluster center are substantially magnified. It is therefore not surprising that some of the most distant galaxies known have been detected in systematic searches for drop-out galaxies near the centers of massive clusters. One example of this is shown in Fig. 9.19, where a galaxy at z ∼ 7 is doubly imaged by the cluster Abell 2218 (see Fig. 6.​51), and by means of this it is magnified by a factor ∼ 25.

The CLASH and HLS surveys. In order to fully exploit the power of natural telescopes, a large treasury program was carried out with HST, imaging 25 massive clusters with 16 filters of the ACS and WFC3/IR instruments. This CLASH (Cluster Lensing And Supernova survey with Hubble) survey is designed to obtain accurate spectral energy distributions of very faint galaxies in the cluster fields. Furthermore, the data will allow high-quality strong and weak lensing analysis of these clusters, and many results are already available at the time of writing. The observing strategy with multiple visits per cluster field also enables the detection of supernovae, either of cluster galaxies, or gravitationally lensed background supernovae.

In a similar spirit, the Herschel Lensing Survey (HLS) targeted 44 massive clusters, in order to find magnified far-IR sources; some of the results from this survey will be discussed in Sect. 9.3.3 below.

We have seen that in order to apply the Lyman-break technique, one needs images in three filters, one on the blue side of the Lyman break, and two on the red side. These two redder filters are required to demonstrate that these galaxies have a rather flat spectrum for wavelengths longer than the break, i.e., that they are indeed actively star forming, to minimize the possibility that the source is simply a very red one and drops out of the shorter wavelength filter just for this reason. Therefore, the application of the Lyman-break method on deep optical images is restricted to redshifts of z ≲ 5. In order to move towards higher redshifts, images in the near-IR are required.

Combining the images of the HUDF with deep NIR imaging from NICMOS and with ground-based telescopes, the redshift z ∼ 6 barrier could be overcome, by searching for i-band dropouts. Many of these were found, and several of the brighter ones were spectroscopically confirmed. Furthermore, ultra-deep optical and NIR imaging from the ground, covering larger sky areas than possible with the HST, have brought large harvest in selecting z ∼ 6 galaxies using the Lyman-break technique. Currently, more than 30 galaxies are known with spectroscopic redshifts between 6 and 6.5.

As we saw before, QSOs at these redshifts have essentially no flux shortward of the Lyα emission line, i.e., the intergalactic medium at z ∼ 6 is sufficiently neutral to absorb almost all UV-photons. This could mean that we are approaching the redshift of reionization of the Universe, and it is not immediately obvious that one may expect a substantial population of higher-redshift objects. On the other hand, the photons needed for reionizing the Universe must come from the first galaxies formed, and these must occur at even higher redshifts. In addition, since the first results from WMAP, we have known that the redshift of reionization is substantially higher than z = 6, with the Planck CMB results estimating this redshift to be at z ∼ 10. Thus, searching for higher-redshift sources is a highly valuable scientific aim, in order to understand the early stages of the evolution of galaxies.

Once at z ∼ 7, the Lyα line is at λ ∼ 1 μm, and hence one needs at least two NIR images, plus a very deep optical image in the z-band, in order to find LBG candidates at these redshifts. As discussed above, the latest camera onboard HST, the WFC3, has a sensitive near-IR channel, and its mapping of the HUDF provided an excellent data set for this purpose. Together with less deep, but larger-area imaging by WFC3 over the GOODS field, some 100 candidates at z ∼ 7, and more than 50 at z ∼ 8 have been found (see Fig. 9.20 for several z ∼ 8 candidates). Spectroscopic verification of these sources, however, is very challenging, since now very sensitive near-IR spectroscopy is needed.

For one of the high-z candidates in Fig. 9.20, the detection of a spectral line in a deep NIR spectrum was claimed—see Fig. 9.21—yielding a redshift of z = 8. 55. Whereas the redshift is based on a single emission line, the identification of this line as Lyα is considered to be secure, since the HST photometry shown in Fig. 9.20 rules out all other redshift, except a low-probability alternative at z ∼ 2. In this case, the emission line would be [Oii], which is a doublet. This doublet would have been clearly resolved in the spectrum and can safely be ruled out. Thus, if the emission line in Fig. 9.21 is real, this galaxy is the highest-redshift spectroscopically confirmed galaxy yet found. However, reobservation of the same source with two other NIR spectrographs failed to reproduce the emission line. Whether or not this galaxy is at z = 8. 55 thus remains an open issue, but this example impressively illustrates the difficulty of securing redshifts for z ≳ 7 galaxies. We also need to keep in mind that many LBGs at lower redshift do not show a Lyα emission line; no spectroscopic redshift of analogs of such sources at high redshifts can be obtained with current telescopes.

Employing natural telescopes, the hunt for even higher-redshift sources becomes promising. The cluster MACS J0647+7015 shown in Fig. 9.22 is a target of the CLASH survey, and thus has been imaged with HST in many different filters. One of the multiply-imaged sources in this cluster is a J-band drop-out; its spectral energy distribution puts it at a redshift z ∼ 11. Such a high redshift is also supported by the lensing geometry and the location of the three images. Whereas no spectroscopic confirmation of this high redshift is available up to now, the next generation space telescope JWST (see Chap. 11) will probably be able to confirm (or not) this redshift estimate.

On the other hand, the narrow-band selection of high-z sources is more promising in this respect. Although there are many contaminating sources—emission line objects at lower redshifts—a clear detection of a source through the narrow-band technique at least yields a good indication that the source has an emission line at that wavelength, and thus spectroscopy is promising.

The wavelengths of the narrow-band filters are best chosen as to minimize the sky brightness. This then defines preferred spectral windows, which in turn define the redshifts of the Lyα emitting galaxies that can be detected with this technique. By far the most productive telescope in this respect is Subaru, due to its unique combination of aperture and field-of-view of its Suprime-Cam camera. Specifically designed narrow-band filters for the highest redshifts target LAEs at z ∼ 5. 7, 6.6, 7.0 and 7.3. Many hundreds of LAEs were found with this techniques, including several dozens spectroscopically confirmed at redshift z ≳ 6. 6. The redshift record holder are galaxies at z = 6. 96 and z = 7. 213.

In many cases photometric redshifts may be the only method for obtaining redshift information, until more powerful telescopes come into operation. As mentioned before, photometric redshifts are the more reliable the more bands are available, and the better the photometric accuracy is. Deep fields like the GOODS field are therefore best suited for photo-z studies of high-redshift galaxies, owing to the broad wavelength coverage and their depth. In Fig. 9.23, two galaxies selected by the Lyman-break technique are shown, with optical, NIR and MIR observations available from HST and Spitzer. In both cases, the best-fitting high-redshift and low-redshift spectral energy distributions are shown, and in both cases, the low-redshift fit is very much worse than the high-z one, so that a low redshift can be ruled out with very high confidence in both cases. This photo-z technique has yielded substantial samples of z ≲ 8 galaxies, which form a rather robust base for statistical studies of the high-z galaxy population.

The highest redshift QSO. Whereas over most of the past ∼ 50 years QSOs were the redshift record holders, this is currently no longer the case: independent of whether the galaxy shown in Fig. 9.21 is indeed at z = 8. 55, the photo-z of several galaxies are sufficiently robust to place them at z > 7. Concerning QSOs, the SDSS has discovered several z ∼ 6 objects, with the highest redshift one at z = 6. 42 (see Fig. 9.1). This was for many years the record holder; only recently, a QSO with z > 7 was found, whose spectrum is shown in Fig. 9.24. It was found with a color selection, based on optical and near-IR photometry. The spectrum clearly confirms its high redshift of z = 7. 085, based on several emission lines. Its high luminosity of ∼ 6 × 10¹³ L _⊙ implies a very massive black hole with mass M ∼ 2 × 10⁹ M _⊙, as estimated from the line width of the Mgii emission line in combination with the luminosity. This large SMBH mass had to be assembled within the first 800 million years of the Universe. This strengthens the constraints on rapid black hole formation, relative to the previously discovered QSOs with redshifts z ≲ 6. 4.⁴

One class of galaxies, the so-called starburst galaxies, is characterized by a strongly enhanced star-formation rate, compared to normal galaxies. Whereas our Milky Way is forming stars with a rate of ∼ 3M _⊙∕yr, the star-formation rate in starburst galaxies can be larger by a factor of more than a hundred. Dust heated by hot stars radiates in the FIR, rendering starbursts very strong FIR emitters. Many of them were discovered by the IRAS satellite (‘IRAS galaxies’) ; they are also called ULIRGs (Ultra Luminous InfraRed Galaxies).

The reason for this strongly enhanced star formation is presumably the interaction with other galaxies or the result of merger processes, an impressive example of which is the merging galaxy pair known as the ‘Antennae’ (see Fig. 9.25). In this system, stars and star clusters are currently being produced in very large numbers. The images show a large number of star clusters with a characteristic mass of 10⁵ M _⊙, some of which are spatially resolved by the HST. Furthermore, particularly luminous individual stars (supergiants) are also identified. The ages of the stars and star clusters span a wide range and depend on the position within the galaxies. For instance, the age of the predominant population is about 5–10 Myr, with a tendency for the youngest stars to be located in the vicinity of strong dust absorption. However, stellar populations with an age of 100 and 500 Myr, respectively, have also been discovered; the latter presumably originates from the time of the first encounter of these two galaxies, which then led to the ejection of the tidal tails. This seems to be a common phenomenon; for example, in the starburst galaxy Arp 220 (see Fig. 1.​15) one also finds star clusters of a young population with age ≲ 10⁷ yr, as well as older ones with age ∼ 3 × 10⁸ yr. It thus seems that during the merging process several massive bursts of star-cluster formation are triggered.

It was shown by the ISO satellite that the most active regions of star formation are not visible on optical images, since they are completely enshrouded by dust (left panel in Fig. 9.26). A map at 8 μm obtained by the Spitzer Space Observatory (Fig. 9.26, right panel) shows the hot dust heated by young stars, where this IR emission is clearly anti-correlated with the optical radiation. Indeed, the mid-infrared emission is strongest in the region between these two colliding galaxies; apparently, this is the location where the current star formation is most intense. Maps of the CO emission also show that these regions contain a large reservoir of molecular gas. Furthermore, the large-scale X-ray radiation in this galaxy collision shows the efficiency with which gas is heated to high temperatures by the supernova events that accompany massive star formation, with a corresponding chemical enrichment of the gas with α-elements. Many of the X-ray point sources are due to high-mass X-ray binaries. Obviously, a complete picture of star formation in such galaxies can only be obtained from a combination of optical and IR images.

Combining deep optical and NIR photometry with MIR imaging from the Spitzer telescope, star-forming galaxies at high redshifts can be detected even if they contain an appreciable amount of dust (and thus may fail to satisfy the LBG selection criteria). These studies find that the comoving number density of ULIRGs with L _IR ≳ 10¹² L _⊙ at z ∼ 2 is about three orders of magnitude larger than the local ULIRG density. These results seem to imply that the high-mass tail of the local galaxy population with M ≳ 10¹¹ M _⊙ was largely in place at redshift z ∼ 1. 5 and evolves passively from there on. We shall come back to this aspect below.

Ultra-luminous Compact X-ray Sources. Observations with the Chandra satellite have shown that starburst galaxies contain a rich population of very luminous compact X-ray sources (Ultra-luminous Compact X-ray Sources, or ULXs; see Fig. 9.27). They are formally defined to have an X-ray luminosity > 10³⁹ erg∕s. Similar sources, though with lower luminosity, are also detected in the Milky Way, where these are binary systems with one component being a compact star (white dwarf, neutron star, or black hole). The X-ray emission is caused by accretion of matter (which we discussed in Sect. 5.​3.​2) from the companion star onto the compact component, and are called X-ray binaries.

Some of the ULXs in starbursts are so luminous, however, that the required mass of the compact star by far exceeds 1 M _⊙ if the Eddington luminosity is assumed as an upper limit for the luminosity [see (5.​25)]. The detection of these ULXs in the 1980s by the Einstein observatory thus came unexpectedly, since one does not expect to form black holes with a mass larger than ∼ 10M _⊙ in supernova explosions. Thus, one concludes that either the emission of these sources is highly anisotropic, hence beamed towards us, or that the sources are black holes with masses of up to ∼ 200M _⊙. In the latter case, we may just be witnessing the formation of intermediate mass black holes in these starbursts. In fact, recently a ULX was found with a peak luminosity of 10⁴² erg∕s, which, assuming that the Eddington limit is not exceeded, implies a black hole mass of > 500M _⊙ as the origin of this source, located ∼ 4 kpc away from the center of an edge-on spiral galaxy.

This latter interpretation is also supported by the fact that the ULXs are concentrated towards the center of the galaxies—hence, these black holes may spiral into the galaxy’s center by dynamical friction, and there merge to a SMBH. This is one of the possible scenarios for the formation of SMBHs in the cores of galaxies, a subject to which we will return in Sect. 10.​4.​5. Furthermore, the similarity of the properties of ULXs with those of X-ray binaries may indicate that ULXs are just scaled-up version of these more common sources.

As mentioned several times previously, the population of galaxies detected in a survey depends on the selection criteria. Thus, employing the Lyman-break method, one mainly discovers those galaxies at high redshift which feature active star formation and therefore have a blue spectral distribution at wavelengths longwards of Lyα. The development of NIR detectors enabled the search for galaxies at longer wavelengths. Of particular interest here are surveys of galaxies in the K-band, the longest wavelength window that is reasonably accessible from the ground (with the exception of the sub-millimeter to radio domain).

The NIR waveband is of particular interest because the luminosity of galaxies at these wavelengths is not dominated by young stars. As we have seen in Fig. 3.​34, the luminosity in the K-band depends rather weakly on the age of the stellar population, so that it provides a reliable measure of the total stellar mass of a galaxy.

Characteristics of EROs. Examining galaxies with a low K-band flux, one finds either galaxies with low stellar mass at low redshifts, or galaxies at high redshift with high optical (due to redshift) luminosity. But since the luminosity function of galaxies is relatively flat for L ≲ L _∗, one expects the latter to dominate the surveys, due to the larger volume at higher z. In fact, K-band surveys detect galaxies with a broad redshift distribution. In Fig. 9.28 the z-distribution of galaxies in the K20-survey is shown. In this survey, objects with K _s < 20 were selected in two fields with a combined area of 52 arcmin², where K_s is a filter at a wavelength slightly shorter than the classic K-band filter. After excluding stars and Type 1 AGNs, 489 galaxies were found, 480 of which have had their redshifts determined. The median redshift in this survey is z ≈ 0. 8.

Considering galaxies in a (R − K) vs. K color-magnitude diagram (Fig. 9.29), one can identify a population of particularly red galaxies, thus those with a large R − K. These objects were named Extremely Red Objects (EROs); about 10 % of the galaxies in K-selected surveys at faint magnitudes are EROs, typically defined by R − K > 5. Spectroscopic analysis of these galaxies poses a big challenge because an object with K = 20 and R − K > 5 necessarily has R > 25, i.e., it is extremely faint in the optical domain of the spectrum. However, with the advent of 10-m class telescopes, spectroscopy of these objects has become possible.

The nature of EROs: passive ellipticals versus dusty starbursts. From these spectroscopic results, it was found that the class of EROs contains rather different kinds of sources. To understand this point we will first consider the possible explanations for a galaxy with such a red spectral distribution. As a first option, the object may be an old elliptical galaxy with the 4000 Å-break redshifted to the red side of the R-band filter, i.e., typically an elliptical galaxy at z ≳ 0. 8. For these galaxies to be sufficiently red to satisfy the selection criterion for EROs, they need to already contain an old stellar population by this redshift, which implies a very high redshift for the star formation episode in these objects; it is estimated from population synthesis models that their formation redshift must be z _form ≳ 2. 5. A second possible explanation for large R − K is reddening by dust. Such EROs may be galaxies with active star formation where the optical light is strongly attenuated by dust extinction. If these galaxies are located at a redshift of z ∼ 1, the measured R-band flux corresponds to a rest-frame emission in the UV region of the spectrum where extinction is very efficient.

Spectroscopic analysis reveals that both types of EROs are roughly equally abundant. Hence, about half of the EROs are elliptical galaxies that already have, at z ∼ 1, a luminosity similar to that of today’s ellipticals, and are at that epoch already dominated by an old stellar population. The other half are galaxies with active star formation which do not show a (prominent) 4000 Å-break but which feature the emission line of [Oii] at λ = 3727 Å, a clear sign of recent star formation. Further analysis of EROs by means of very deep radio observations confirms the large fraction of galaxies with high star-formation rates. Utilizing the close relation of radio emissivity and FIR luminosity, we find a considerable fraction of EROs to be ULIRGs at z ∼ 1.

Spatial correlations. EROs are very strongly correlated in space. The interpretation of this strong correlation may be different for the passive ellipticals and for those with active star formation. In the former case the correlation is compatible with a picture in which these EROs are contained in clusters of galaxies or in overdense regions that will collapse to a cluster in the future. The correlation of the EROs featuring active star formation can probably not be explained by cluster membership, but the origin of the correlation may be the same as for the correlation of the LBGs.

The number density of passive EROs, thus of old ellipticals, is surprisingly large compared with expectations from the model of hierarchical structure formation that we will discuss in Chap. 10.

FIR emission from hot dust is one of the best indicators of star formation. However, observations in this waveband are only possible from space, such as was done with the IRAS and ISO satellites, and more recently with Spitzer and Herschel. Depending on the dust temperature, dust emission has its maximum at about 100 μm, which is not observable from the ground. At longer wavelengths there are spectral windows longwards of λ ∼ 250 μm where observations through the Earth’s atmosphere are possible, for instance at 450 and 850 μm in the sub-millimeter waveband. However, the observing conditions at these wavelengths are extremely dependent on the amount of water vapor in the atmosphere, so that the observing sites must by dry and at high elevations. In the sub-millimeter (sub-mm) range, the long wavelength domain of thermal dust radiation can be observed, which is illustrated in Fig. 9.30.

Developments. Since about 1998 sub-mm astronomy has experienced an enormous boom, with two instruments having been put into operation: the Sub-millimeter Common User Bolometer Array (SCUBA), operating at 450 and 850 μm, with a field-of-view of 5 arcmin², and the Max-Planck Millimeter Bolometer (MAMBO), operating at 1200 μm. Both are bolometer arrays which initially had 37 bolometers each, but which later were upgraded to a considerably larger number of bolometers. With the opening of the 12-m APEX (Atacama Pathfinder Experiment) telescope (see Fig. 1.​28) in Chile, equipped with powerful instrumentation, a further big step in sub-mm astronomy was achieved. In addition, the Herschel satellite (Fig. 1.​33), operating at wavelength between 55 and 670 μm, allowed imaging of large sky areas at these wavelengths, due to a much lower noise level than can be achieved from the ground, though with considerably worse spatial resolution due to the smaller aperture compared to ground-based sub-mm telescopes. Figure 9.31 shows a Herschel image of the COSMOS field.

Apart from single-dish observatories, interferometers operating at these wavelengths yield substantially higher angular resolution, for example the very successful IRAM Plateau de Bure interferometer in the French Alps consisting of six 15-m antennas. The Atacama Large Millimeter Array (ALMA; see Fig. 1.​29) with its 54 12-m and 12 7-m antennas, inaugurated in 2013, marks a huge leap in terms of resolution and sensitivity in this waveband regime.

The negative K-correction of sub-mm sources. The emission of dust at these wavelengths is described by a Rayleigh–Jeans spectrum, modified by an emissivity function that depends on the dust properties (chemical composition, distribution of dust grain sizes); typically, one finds This steep spectrum for frequencies below the peak of the thermal dust emission at λ ∼ 100 μm implies a very strong negative K-correction (see Sect. 5.​6.​1) for wavelengths in the sub-mm domain: at a fixed observed wavelength, the rest-frame wavelength becomes increasingly smaller for sources at higher redshift, and there the emissivity is larger. As Fig. 9.32 demonstrates, this spectral behavior causes the effect that the flux in the sub-mm range does not necessarily decrease with redshift. For z ≲ 1, the 1∕D ²-dependence of the flux dominates, so that up to z ∼ 1 sources at fixed luminosity get fainter with increasing z. However, between z ∼ 1 and z ∼ z _flat the sub-mm flux as a function of redshift remains nearly constant or even increases with z, where z _flat depends on the dust temperature T _d and the observed wavelength; for T _d ∼ 40 K and λ ∼ 850 μm one finds z _flat ∼ 8. We therefore have the quite amazing situation that sources appear brighter when they are at larger distances. This is caused by the very negative K-correction which more than compensates for the 1∕D ²-decrease of the flux. Only for z > z _flat does the flux begin to rapidly decrease with redshift, since then, due to redshift, the corresponding restframe frequency is shifted to the far side of the maximum of the dust spectrum (see Fig. 9.30). Hence, a sample of galaxies that is flux-limited in the sub-mm domain should have a very broad z-distribution. The dust temperature is about T _d ∼ 20 K for low-redshift spirals, and T _d ∼ 40 K is a typical value for galaxies at higher redshift featuring active star formation. The higher T _d, the smaller the sub-mm flux at fixed bolometric luminosity.

Counts of sub-mm sources at high Galactic latitudes have yielded a far higher number density than was predicted by early galaxy evolution models. For the density of sources as a function of limiting flux S, at wavelength λ = 850 μm, one obtains

The identification of sub-mm sources. At first, the optical identification of these sources turned out to be extremely difficult: due to the relatively low angular resolution of single-dish sub-millimeter telescopes (for example, MAMBO has a beam with FWHM of ∼ 11″ at λ = 1. 2 mm), the positions of sources can only be determined with an accuracy of several acrseconds.⁵ Typically, several faint galaxies can be identified on deep optical images within an error circle of this radius. Furthermore, Fig. 9.32 suggests that these sources have a relatively high redshift, thus they should be very faint in the optical. An additional problem is reddening and extinction by the same dust that is the source of the sub-mm emission.

The identification of sub-mm sources was finally accomplished by means of their radio emission, since a majority of the sources selected at sub-mm wavelengths can be identified in very deep radio observations at 1. 4 GHz. Since the radio sky is far less crowded than the optical one, and since the VLA achieves an angular resolution of ∼ 1″ at λ = 20 cm, the optical identification of the corresponding radio source becomes relatively easy. One example of this identification process is shown in Fig. 9.33. With the accurate radio position of a sub-mm source, the optical identification can then be performed. In most cases, they are very faint optical sources indeed, so that spectroscopic analysis is difficult and very time-consuming. Another method for estimating the redshift results from the spectral energy distribution shown in Fig. 9.30. Since this spectrum seems to be nearly universal, i.e., not varying much among different sources, some kind of photometric redshift can be estimated from the ratio of the fluxes at 1.4 GHz and 850 μm, yielding reasonable estimates in many cases.

Redshift distribution of SMGs. The median redshift of sources with 850 μm flux larger than 5 mJy and 1. 4 GHz flux above 30 μJy is about 2.2. However, at these sensitivities about half the sub-mm sources are unidentified in the radio, and hence their redshift distribution could be different from those with radio counterparts. In some of the sources with radio identification, an AGN component, which contributes to the dust heating, was identified, but in general newly born stars seem to be the prime source of the energetic photons which heat the dust. The optical morphology and the number density of the sub-mm sources suggest that we are witnessing the formation of massive elliptical galaxies in these sub-mm sources.

The great capabilities of interferometric observations enables a different method for identification and redshift determination of SMGs. The high angular resolution of the Plateau de Bure and, in particular, ALMA interferometers can pinpoint a sub-mm source very accurately, allowing the identification with optical or infrared sources. In addition, the large bandwidth of the ALMA receivers can take spectra of these sources over an appreciable range of wavelengths, and thus identify emission lines of molecules, predominantly those of the CO and water molecules, or fine-structure lines of atoms (specifically, ionized and neutral carbon), therefore removing the need for obtaining optical spectra of these optically faint sources. The potential of this method was recently established: even by observing with an incomplete array of telescopes, ALMA detected emission lines in 23 out of 26 sources selected by the South Pole Telescope (SPT; see Fig. 1.​31) at 1. 4 mm wavelength.

Once the precise locations of the sub-mm sources are determined with interferometric observations, they can be identified on deep optical images, and their redshifts be determined from optical spectroscopy. Hence, much of the potential bias in the redshift distribution of these sources, which are caused by the fact that only about half of the SMGs have a radio identification, are removed.⁶ Indeed, if one compares the redshift distributions of both samples, shown in Fig. 9.34, one sees that they are significantly different. Using radio-identification as an intermediate step, and neglecting those SMGs for which no radio source could be identified, biases the redshift distribution of dusty star-forming galaxies low. The population extends over a significantly larger redshift range than concluded previously—in accord with expectations from the large negative K-correction. In addition, the redshifts of SMGs can be determined with the PdB and ALMA interferometers directly, using molecular line spectroscopy, without the need for optical spectroscopy. In particular the large bandwidth of the ALMA receiver allows one to cover a broad range of wavelengths, and thus of redshifts for which molecular lines can be detected and identified.

Halo masses of sub-mm galaxies. As we discussed in Sect. 8.​1.​2, one can estimate the mass of the dark matter halo in which objects reside, by comparing their clustering properties with those of dark matter halos, as obtained from cosmological simulations. Sub-mm galaxies identified in a survey conducted by APEX (see the left panel of Fig. 9.35) allowed an estimate of the clustering length r ₀ (defined such that the two-point correlation function is unity at separation r ₀) of these sources, yielding r ₀ ≈ (7. 5 ± 2)h ⁻¹ Mpc. In the right panel of Fig. 9.35, this measurement of r ₀ is related to that of other source populations. The clustering length of sub-mm sources is very similar to that of QSOs at the same redshift, and considerably larger than that for Lyman-break galaxies. Comparing this to the clustering length of dark matter halos with different masses, shown as dotted curves in Fig. 9.35, we conclude that SMGs live in relatively massive halos of several times 10¹² M _⊙ at z ∼ 2. This results was recently confirmed by detecting a weak lensing signal around a sample of ∼ 600 relatively bright SMGs, which yields a characteristic halo mass of these galaxies of ∼ 10¹³ M _⊙. This mass is comparable to that of current epoch massive elliptical galaxies, and suggests the interpretation that high-redshift SMGs evolve into present-day ellipticals. A large fraction of their stellar population is formed in the epoch at which the galaxy is seen as a SMG; by the end of this period, most of the gas in the galaxy is used up (or some fraction of it may be expelled), and in the remaining evolution little or no star formation occurs.

Additional support for this idea is provided by the fact that the sub-mm galaxies are typically brighter and redder than (restframe) UV-selected galaxies at redshifts z ∼ 2. 5. This indicates that the stellar masses in sub-mm galaxies are higher than those of LBGs.

A joint investigation of z ∼ 2 sub-mm galaxies at X-ray, optical and MIR wavelengths yields that these sources are not only forming stars at a high rate, but that they already contain a substantial stellar population with M ∼ 10¹¹ M _⊙, roughly an order of magnitude more massive than LBGs at similar redshifts. The large AGN fraction among sub-mm galaxies indicates that the growth of the stellar population is accompanied by accretion and thus the growth of supermassive black holes in these objects. Nevertheless, the relatively faint X-ray emission from these galaxies suggests that either their SMBHs have a mass well below the local relation between M _• and stellar properties of (spheroidal) galaxies, or that they accrete at well below the Eddington rate. Furthermore, the typical ratio of X-ray to sub-mm luminosity of these sources is about one order of magnitude smaller than in typical AGNs, which seems to imply that the total luminosity of these sources is dominated by the star-formation activity, rather than by accretion power. This conclusion is supported by the fact that the optical counterparts of SMGs show strong signs of merging and interactions, together with their larger size compared to optically-selected galaxies at the same redshifts. This latter point shows that the emission comes from an extended region, as expected from star formation in mergers, rather than AGN activity.

Merging SMGs. The Herschel satellite found a strong sub-mm source, whose subsequent reobservations with different telescopes revealed this to be a merger of two SMGs. This source, called HXMM01, is shown in Fig. 9.36. In the near-IR, the image shows four main sources, of which the two brightest ones are galaxies at intermediate redshifts. The two fainter ones are galaxies at z = 2. 31. Those coincide with strong sub-mm sources, as seen by the interferometric images at 880 μm, as well as through their strong molecular line emission. This pair of sources has a separation of 19 kpc, and is most likely undergoing a merger. The two foreground galaxies will cause a moderate lensing magnification of the SMGs, with an estimated μ ∼ 1. 6. Thus, even after magnification correction, this source belongs to the brightest SMGs, with a corrected flux of ∼ 20 mJy at λ = 850 μm. This source is extreme in several properties: Its dust temperature is estimated to be T ≈ 55 K, which is larger than that of most starburst galaxies. The estimated bolometric infrared luminosity is L ∼ 2 × 10¹³ L _⊙, with a corresponding star-formation rate of . At this rate, the molecular gas of ∼ 2. 3 × 10¹¹ M _⊙ will be turned into stars on a timescale of only 70 Myr. The brevity of this time interval implies that objects similar to HXMM01 should be rare. The molecular gas mass in this object is comparable with the stellar mass already present, i.e., the gas-mass fraction in this object is about 50 %. At the end of its merging process and star-formation episode, the resulting galaxy will have a stellar mass of ∼ 10¹¹ M _⊙, i.e., the stellar mass of a massive elliptical galaxy.

Although HXMM01 is not the only pair of merging SMGs yet discovered, it is brighter than the other ones by a factor of ∼ 2 (magnification corrected); correspondingly, its star-formation rate is also about twice that of the other merging systems. There are indications that the far-IR luminosity, and thus the star-formation rate, at fixed molecular mass (as measured by the luminosity in molecular CO lines), is higher by about a factor of three compared to galaxies that are not observed to be merging. This then is a direct hint at the possibility that merging triggers star-formation events, or bursts of star-formation.

In fact, there are several observations which suggest that the very large star-forming rates of many of the most luminous SMGs are due to merging events. Arguably the most direct one comes from interferometric observations of the molecular gas in SMGs, which shows that most of them have morphological and kinematical properties that identify them as merging systems. Hence, in accordance with the high merger rate of local ULIRGs, the extreme SMGs may also be triggered by merger events.

If dusty galaxies are at very high redshifts z ≳ 5, the peak of their spectral energy distribution shown in Fig. 9.30 is observed at wavelengths longward of 500 μm. Hence, for such sources the flux density is expected to increase with wavelength for λ ≲ 500 μm. The SPIRE instrument on the Herschel Space Observatory observed at 250, 350 and 500 μm, and blank survey fields at these frequencies can thus be used to search for very high-redshift sub-millimeter galaxies.

One such object found by this selection method is shown in Fig. 9.37. The galaxy HFLS3 is an extreme starburst at z = 6. 34, with an estimated star-formation rate of ∼ 3000 M _⊙∕yr, some factor of 20 larger than the local starburst Arp 220 (see Fig. 1.​15). With an estimated gas mass of ∼ 10¹¹ M _⊙, this object would transform all its gas mass into stars on a time-scale of only ∼ 30 Myr, assuming a constant star-formation rate. Spatially resolved molecular spectroscopy shows a velocity profile of this galaxy, which can be used to estimate a dynamical mass of ∼ 2. 7 × 10¹¹ M _⊙, yielding a gas-mass fraction of ∼ 40 %, similar to what is found in typical z ∼ 2 sub-millimeter galaxies.

Magnification bias of sub-millimeter sources. We mentioned before that some of the apparently most luminous sources of any kind have a large probability of being gravitationally lensed. The reason for this can be seen as follows: If we consider a population of sources, then their distribution in luminosity is most frequently described by a Schechter-like function. That means that for low luminosities, the luminosity function behaves as a power law in L, whereas for L > L ^∗, where L ^∗ characterizes the break in the luminosity function, the density of sources decreases exponentially with L. In particular that means that there are essentially no sources with luminosity L ≳ 5L ^∗.

The probability that any given high-redshift source is gravitationally lensed and significantly magnified is small, of order 10⁻⁴. Thus, if one picks random sources, the fraction of lensed ones among them will be similarly small. However, we can not pick random sources, but only sources above the flux limit of our observations. The situation in flux-limited samples can be quite different, since the magnification by lensing affects the mix of sources which are above the flux threshold. The probability that a source undergoes a magnification larger than μ can be shown to behave like μ ⁻², provided the source is sufficiently small. Hence, large magnifications are correspondingly rare. But the probability for large magnifications decreases as a power law in μ—compared to the exponential decrease in the luminosity of sources for L > L ^∗. There will be a point where the (low-amplitude) power law overtakes the exponential, or in other words, where a source above a given flux threshold is more likely to be highly magnified, than having a very high intrinsic luminosity. An example of this effect are the bright z ∼ 3 Lyman-break galaxies shown in Figs. 9.17 and 9.18, whose inferred luminosity is much larger than the L ^∗ of this population of galaxies.

This is illustrated in Fig. 9.38, where the upper end of the 500 μm source counts are shown, together with a model decomposition of the counts into lensed and unlensed SMGs, nearby spiral galaxies and radio AGNs. As we just argued, for relatively low fluxes, the fraction of lensed sources is increasingly small. However, due to the steep decline of the unlensed counts, beyond a certain flux level they start to dominate the counts of SMGs. From that figure, one therefore expects that a SMG with S ≳ 60 mJy at 500 μm has a fairly high probability of being lensed, whereas a SMG with S ≳ 100 mJy almost certainly is a lensed source.

Hence, selecting sources with S ≳ 100 mJy, and cleaning the source catalog for nearby spirals (they are bright and big in the optical, and can thus easily be identified), one expects to obtain a clean lensed sample. Indeed, all five candidates selected from the survey data on which Fig. 9.38 is based, are most likely lensed, with the putative lens being identified on optical and NIR images. Furthermore, interferometric observations have confirmed the lensing nature of several of the candidates, by finding multiple images.

Using ALMA imaging of sources from the South Pole Telescope survey, selected at 1.4 and 2 mm to have the spectral index at these wavelengths corresponding to dust, and cleaning the sample for nearby objects (e.g., excluding sources detected by IRAS or in low-frequency radio catalogs), a large fraction of these sources turn out to be gravitationally lensed. In Fig. 9.39, ten of these are shown, where not only the multiple images of the sources (which are all at high redshift) are shown, but also a near-IR image of the field which clearly shows the lensing galaxy. Hence, selecting bright sources in the (sub)-millimeter regime yields a very high success rate of finding gravitational lens systems. All these lens systems would be missed in optical surveys, due to the faintness of the SMGs at optical and NIR wavelengths.

In our discussion of QSO absorption lines in Sect. 5.​7, we mentioned that the Lyα lines are broadly classed into three categories: the Lyα forest, Lyman-limit systems, and damped Lyα systems, which are separated by a column density of N _HI ∼ 10¹⁷ cm⁻² and N _HI ∼ 2 × 10²⁰ cm⁻², respectively. The origin of the Lyα forest, as discussed in some detail in Sect. 8.​5, is diffuse, highly ionized gas with small density contrast. In comparison, the large column density of damped Lyα systems (DLAs) strongly suggests that hydrogen is mostly neutral in these systems. The reason for this is self-shielding : for column densities of N _HI ≳ 2 × 10²⁰ cm⁻² the background of ionizing photons is unable to penetrate deeply into the corresponding hydrogen ‘cloud’, so that only its surface is highly ionized. Interestingly enough, this column density is about the same as that observed in 21 cm hydrogen emission at the optical radius of nearby spiral galaxies.

DLAs can be observed at all redshifts z ≲ 5. For z > 5 the Lyα forest becomes so dense that these damped absorption lines are very difficult to identify. For z ≲ 1. 6 the Lyα transition cannot be observed from the ground; since the apertures of optical/UV telescopes in space are considerably smaller than those on the ground, low-redshift DLAs are substantially more complicated to observe than those at higher z.

The neutral hydrogen mass contained in DLAs. The column density distribution of Lyα forest lines is a power law, given by (8.​30). The relatively flat slope of β ∼ 1. 6 indicates that most of the neutral hydrogen is contained in systems of high column density. This can be seen as follows: the total column density of neutral hydrogen above some minimum column density N _min is and is, for β < 2, dominated by the highest column density systems. In fact, unless the distribution of column densities steepens for very high N _HI, the integral diverges. From the extended statistics now available for DLAs, it is known that dN∕dN _HI attains a break at column densities above N _HI ≳ 10²¹ cm⁻², rendering the above integral finite. Nevertheless, this consideration implies that most of the neutral hydrogen in the Universe visible in QSO absorption lines is contained in DLAs. From the observed distribution of DLAs as a function of column density and redshift, the density parameter Ω _HI in neutral hydrogen as a function of redshift can be inferred. Apparently, Ω _HI ∼ 10⁻³ over the whole redshift interval 0 < z < 5, with perhaps a small redshift dependence. Compared to the current density of stars, this neutral hydrogen density is smaller only by a factor ∼ 3. Therefore, the hydrogen contained in DLAs is an important reservoir for star formation, and DLAs may represent condensations of gas that turn into ‘normal’ galaxies once star-formation sets in. Since DLAs have low metallicities, typically 1/10 of the Solar abundance, it is quite plausible that they have not yet experienced much star formation.

The nature of DLAs. This interpretation is supported by the kinematical properties of DLAs. Whereas the fact that the Lyα line is damped implies that its observed shape is essentially independent of the Doppler velocity of the gas, velocity information can nevertheless be obtained from metal lines. Every DLA is associated with metal absorption line systems, covering low- and high-ionization species (such as Siii and Civ, respectively) which can be observed by choosing the appropriate wavelength coverage of the spectrum. The profiles of these metal lines are usually split up into several components. Interpreted as ionized ‘clouds’, the velocity range Δ v thus obtained can be used as an indicator of the characteristic velocities of the DLA. The values of Δ v cover a wide range, with a median of ∼ 90 km∕s for the low-ionization lines and ∼ 190 km∕s for the high-ionization transitions. The observed distribution is largely compatible with the interpretation that DLAs are rotating disks with a characteristic rotational velocity of v _c ∼ 200 km∕s, once random orientations and impact parameters of the line-of-sight to the QSO are taken into account.

Search for emission from DLAs. If this interpretation is correct, then we might expect that the DLAs can also be observed as galaxies in emission. This, however, is exceedingly difficult for the high-redshift DLAs. Noting that they are discovered as absorption lines in the spectrum of QSOs, we face the difficulty of imaging a high-redshift galaxy very close to the line-of-sight to a bright QSO (to quote characteristic numbers, the typical QSO used for absorption-line spectroscopy has B ∼ 18, whereas an L _∗-galaxy at z ∼ 3 has B ∼ 24. 5). Due to the size of the point-spread function this is nearly hopeless from the ground. But even with the resolution of HST, it is a difficult undertaking. Another possibility is to look for the Lyα emission line at the absorption redshift, located right in the wavelength range where the DLA fully blocks the QSO light. However, as we discussed for LBGs above, not all galaxies show Lyα in emission, and it is not too surprising that these searches have largely failed. Only very few DLA have been detected in emission, with some of them seen only through the Lyα emission line at the trough of the damped absorption line, but with no observable continuum radiation. This latter fact indicates that the blue light from DLAs is considerably fainter than that from a typical LBG at z ∼ 3, consistent with the interpretation that DLAs are not strong star-forming objects. But at least one DLA is observed to be considerably brighter and seems to share some characteristics of LBGs, including a high star-formation rate. In addition, a couple of DLAs have been detected by [Oiii] emission lines. Overall, then, the nature of high-redshift DLAs is still unclear, due to the small number of direct identifications.

For DLAs at low redshifts the observational situation is different, in that a fair fraction of them have counterparts seen in emission. Whereas the interpretation of the data is still not unambiguous, it seems that the low-redshift population of DLAs may be composed of normal galaxies.

The spatial abundance of DLAs is largely unknown. The observed frequency of DLAs in QSO spectra is the product of the spatial abundance and the absorption cross section of the absorbers. This product can be compared with the corresponding quantity of local galaxies: the detailed mapping of nearby galaxies in the 21 cm line shows that their abundance and gaseous cross section are compatible with the frequency of DLAs for z ≲ 1. 5, and falls short by a factor ∼ 2 for the higher-redshifts DLAs.

The search for high-redshift galaxies with narrow-band imaging, where the filter is centered on the redshifted Lyα emission line, has revealed a class of objects which are termed ‘Lyman-α blobs’. These are luminous and very extended sources of Lyα emission; their characteristic flux in the Lyα line is ∼ 10⁴⁴ erg∕s, and their typical size is ∼ 30 to ∼ 100 kpc. Some of these sources show no detectable continuum emission in any broad-band optical filter. Hence, these sources seem to form a distinct class from the Lyman-alpha emitters discussed previously.

The nature of these high-redshifts objects remained unknown for a long time. Suggested explanations were wide-ranging, including a hidden QSO, strong star formation and associated superwinds, as well as ‘cold accretion’, where gas is accreted onto a dark matter halo and hydrogen is collisionally excited in the gas of temperature ∼ 10⁴ K, yielding the observed Lyα emission. It even seems plausible that the Lyman-α blobs encompass a range of different phenomena, and that all three modes of powering the line emission indeed occur. As a common feature, most of the Lyman-α blobs are associated with luminous galaxies, and are associated with strong infrared emission.

Chandra observations of 29 Lyα blobs detected five on them in X-rays; one of them is shown in Fig. 9.40. The X-ray sources are AGNs with L _X ∼ 10⁴⁴ erg∕s and rather large obscuration. Furthermore, these sources emit infrared light from warm dust. The energy output of the AGN is sufficiently large to power the Lyα emission through photoionization. Hence, the AGN hypothesis has been verified for at least some of the sources.

Two of these Lyα blobs were discovered by narrow-band imaging of the aforementioned proto-cluster of LBGs at z = 3. 09. Both of them are sub-mm sources and therefore star-forming objects; the more powerful one has a sub-mm flux suggesting a star-formation rate of ∼ 1000M _⊙∕yr. Spatially resolved spectroscopy extending over the full ∼ 100 kpc size of one of the Lyα blobs shows that across the whole region there is an absorption line centered on the Lyα emission line. The optical depth of the absorption line suggests an Hi column density of ∼ 10¹⁹ cm⁻², and its centroid is blueshifted relative to the underlying emission line by ∼ 250 km∕s. The spatial extent of the blueshifted absorption shows that the outflowing material is a global phenomenon in this object—a true superwind, most likely driven by energetic star formation and subsequent supernova explosions in these objects. Therefore, it appears likely that Lyα blobs are intimately connected to massive star-formation activity.

Being able to detect galaxies at high redshifts, we may first consider their abundance and investigate how their luminosity distribution compares with that galaxies in the local Universe. We are interested in a possible evolution of the luminosity function of galaxies with redshift, as this would clearly indicate that the galaxy population as a whole changes with redshift. The source counts from the Hubble Deep Field (Fig. 9.11) and its strong deviations from the non-evolution models provide a clear indication that the galaxy population evolves in redshift. This point will be considered here in somewhat more detail.

The most convenient way of summarizing the results is a representation of the estimated luminosity function in terms of a fit with a Schechter function (3.​52). In that, the luminosity function is characterized by an overall normalization Φ ^∗, a characteristic luminosity L ^∗ (or, equivalently, an absolute magnitude M ^∗) above which the abundance decreases exponentially, and a power-law slope α of the luminosity function at L ≪ L ^∗. An evolution of Φ ^∗ with redshift indicates that the abundance of luminous galaxies evolves. If L ^∗ depends on z, one may conclude that the luminosity of a typical galaxy is different at higher redshifts. Finally, the faint-end slope α determines what fraction of the total luminosity of the galaxy population is emitted by the fainter galaxies—see (3.​58).

The UV-luminosity function. Since high-redshift galaxies are selected using quite a variety of methods, as discussed in Sect. 9.1, and since the nature of the detected galaxies depends on their selection method, one has to consider different types of luminosity functions. For example, the Lyman-break method selects galaxies by their rest-frame UV radiation, so that from these surveys, the UV-luminosity function of galaxies can be obtained. Since the Lyman-break technique is applicable over a very wide redshift range, the UV-luminosity function has been obtained for redshifts between 2 and 7.

As already indicated by the galaxy counts shown in Fig. 9.11, the rest-frame UV-luminosity function of galaxies evolves strongly with redshift. In the redshift interval 2 ≲ z ≲ 4, the characteristic luminosity L ^∗ is about three magnitudes brighter than that of the local UV-luminosity function as determined with GALEX. This immediately shows that a typical galaxy at these redshifts is far more actively forming stars than local galaxies. Furthermore, the faint-end slope α is steeper for the high-redshift galaxies than for local ones. In fact, estimates yield α ∼ −1. 6, indicating that much of the UV-luminosity density at high redshifts is emitted from rather faint galaxies—galaxies which are currently not observed due to the limited sensitivity of our instruments. Therefore, the overall abundance of UV-luminous galaxies is considerably larger in the redshift interval 2 ≲ z ≲ 4 than it is today. Since the UV-radiation is produced by massive (and thus young) stars, this implies that the star-formation activity at those redshifts was much more intense than at the current epoch.

Going to even higher redshifts, the abundance of UV-selected galaxies decreases again, as can be seen in Fig. 9.41. The evolution is such that the characteristic luminosity decreases with higher z, and at the same time the faint-end slope steepens towards an estimated value of α ∼ −1. 8. Hence, for these very high redshifts, most of the luminosity in the UV is emitted from faint sources. Recent attempts to find credible z ∼ 10 galaxies using the Lyman-break technique yielded upper limits to the abundance of these objects, which yields upper limits to the luminosity function at this redshift. It appears that the decrease with z, visible in Fig. 9.41, accelerates towards even higher redshift.

In detail, these results are still burdened with quite some uncertainties, given the difficulties to identify very high redshift sources. Most of the conclusions are based on photometric redshifts only, since the spectroscopic verification of a z ∼ 7 galaxy is extremely difficult, given that all spectral features blueward of the Lyα emission line are invisible due to intergalactic absorption, and that the radiation redward of the Lyα-line is redshifted into the near-IR. Hence, some of the sources may be misidentified and are in fact lower-redshift objects. Furthermore, since the identification of these very high redshift sources requires very deep observations, carried out only in a small number of fields, one must be aware of sampling variance—the fact that the distribution in a single small field may not be representative of the overall distribution. However, the general trends just discussed are established by now, providing a clear view of the evolution of the galaxy population with cosmic time.

Optical/NIR luminosity function. The rest-frame optical light is a somewhat better indicator of the total stellar mass of galaxies than is the UV-radiation. However, only when going to the NIR is the luminosity of star-forming galaxies not dominated by the radiation from newly-born stars; in addition, the K-band light is rather unaffected by dust obscuration. To assess the rest-frame K-band emission of high-redshift galaxies, one needs mid-IR observations which became possible with the Spitzer Space Telescope. The results of a combined analysis of optical, near-IR and mid-IR data show again a dramatic change of the luminosity function with redshift. The characteristic density of galaxies Φ ^∗ decreases with redshift, as one might expect—there should be fewer galaxies around at higher redshifts. For example, at z ∼ 2, Φ ^∗ is about a factor 3.5 smaller than in the local Universe. In parallel to this, however, the characteristic luminosity L ^∗ increases with z, by about one magnitude up to redshift 2. Hence it seems that a typical galaxy was brighter in the past. This phenomenon is quite counter-intuitive, given that the theory of structure formation predicts that more massive objects form in large abundance only at later redshifts—as follows from hierarchical structure formation. Another way to see this phenomenon is displayed in Fig. 9.42, which shows the comoving density of galaxies with fixed rest-frame K-band luminosity as a function of redshift, normalized to the corresponding local density. For rather low luminosities, the density decreases, but for high K-band luminosities, it increases by a factor ∼ 5 over a broad range in redshift, reaching a maximum at z ∼ 1. 5, and thereafter slowly decreases, but even at z ∼ 4 the density is still higher than in the local Universe.

It thus seems that the typical galaxy at high redshift has a larger stellar mass than currently, or that the ratio of high-mass to low-mass galaxies was substantially larger at high z. This implies that with increasing cosmic time, the galaxy population becomes increasingly dominated by those with lower mass. This phenomenon has received the name downsizing. Models of galaxy evolution in a hierarchical universe need to be able to describe this effect; we will return to this in Chap. 10.

Mid-IR luminosity function. Whereas the rest-frame UV-radiation indicates the level of star formation which is unobscured by dust, it misses those star-forming galaxies which are heavily obscured by dust. Their activity can best be seen in the rest-frame mid- and far-IR. At a fixed wavelength, the emitted flux depends on the amount of heat absorbed by the dust—and reradiated as thermal dust emission—and on the dust temperature. Therefore, the most reliable indicator of the obscured star formation rate is the bolometric infrared luminosity. Whereas this is not directly observable—the combination of sensitivity and field-of-view of far-IR detectors allows one to study only relatively bright objects—the combination of observations at optical, near-IR and mid-IR can be used to estimate the dust temperature and thus to derive the bolometric IR luminosity from the Spitzer 24 μm data and the derived dust temperature.

The corresponding evolution of the IR luminosity function is shown in Fig. 9.43 for several redshift bins. Although for the higher-redshift bins only the highest luminosity sources can be observed, the figure shows a dramatic evolution towards higher redshift: The number density of luminous sources increases by a large factor compared to the local one. The trend is similar to that shown in Fig. 9.42 for the K-band luminosity function, but even stronger. The increase of very strongly star-forming galaxies with redshift is stronger than that with large stellar masses. Combined with the evolution of largely unobscured star-formation, shown in Fig. 9.41, we therefore conclude that the star-forming activities had a much higher level in earlier epochs of cosmic evolution than it has today. We will come back to this point in more detail in Sect. 9.6

Integrating the Schechter function fits shown in Fig. 9.43 over luminosity, one obtains the total infrared luminosity emitted per unit comoving volume. The redshift evolution of this IR luminosity density is shown in Fig. 9.44. It must be pointed out that these estimates carry quite some uncertainty, since they require the extrapolation of the luminosity function to much fainter levels than those where data are available. In particular, the faint-end slope of the Schechter function is not at all well determined at high redshifts. Therefore, the detailed behavior of the luminosity density beyond z ∼ 1 may be slightly different from what is shown in the figure. However, the contribution of the LIRGs (defined as L ≥ 10¹¹ L _⊙) and ULIRGs (L ≥ 10¹² L _⊙) to the luminosity density, also shown in Fig. 9.44, is much better determined. While the IR luminosity density increases by a factor ∼ 20 between today and z ∼ 1, and stays roughly constant up to z ∼ 2. 5, the contribution from ULIRGs increases by at least a factor 100 over the same redshift range.

These results were confirmed with Herschel blank-field surveys, centered on fields for which multi-band observations were previously available (such as the GOODS fields or COSMOS). Observing in the far-IR, Herschel samples the peak of the spectral energy distribution directly, and fewer extrapolations are necessary to derive the bolometric infrared luminosity than required for using Spitzer data only. The analysis of the Herschel data showed that the evolution of the IR luminosity function with redshift is indeed dramatic. If one parametrizes the luminosity function as a Schechter function, the characteristic luminosity L ^∗ in the infrared increases like (1 + z)^3. 5 for 0 ≲ z ≲ 2, and ∝ (1 + z)^1. 6 for 2 ≲ z ≲ 4. The normalization Φ ^∗ of the Schechter function decreases with redshift, with for 0 ≲ z ≲ 1. 1, and for 1. 1 ≲ z ≲ 4. In agreement with the results shown in Fig. 9.43, the comoving space density of very IR-luminous sources increases by huge factors from today to higher redshifts, before it starts to decline beyond redshift z ∼ 3.

The color bimodality, seen prominently in the local population of galaxies (see Sect. 3.​1.​3), has been in place at least since z ∼ 2. As shown in Fig. 9.45, using spectroscopy of a 4. 5 μm-flux limited sample of galaxies in the GOODS South field, for which deep photometry is available over a wide range of optical and infrared bands (including HST and Spitzer), the color bimodality can be clearly seen in all redshift intervals.

At even higher redshift, the sample of galaxies on the red sequence gets increasingly contaminated by dusty star-forming galaxies. However, one can account for this reddening and obtain dust-corrected colors for those galaxies. After this correction, the color bimodality can be detected out to redshifts z ∼ 3, implying that already at young cosmic epochs, galaxies with an old stellar population coexisted with those which actively formed stars. Accordingly, the red sequence was formed early on, as can also be seen in Fig. 9.45.

This observational results implies that even at high redshifts, a large fraction of galaxies exists with a passively evolving stellar population. Whereas the star-forming galaxies—LBGs and SMGs—are the more spectacular objects at these high redshifts, many galaxies had formed their stars at even earlier epochs. From what we mentioned above—see, e.g., Fig. 9.42—the more massive galaxies seem to conclude most of the built-up of their stellar population at the highest redshifts. In parallel with the color-magnitude relation, also the local color-density relation was in place at least since z ∼ 1.

The fact that the population of galaxies evolves strongly with redshift raises the expectation that the galaxies at high redshift are different from those in the local Universe. Here we will point out some of these differences.

The Hubble sequence and galaxy morphology. The majority of present day massive galaxies fall onto the morphological Hubble sequence (Fig. 3.​2). In addition, we have seen that local galaxies can be classified according to their color, with most of them being either member of the red sequence or the blue cloud, as seen in Fig. 3.​38, with a tight correspondence between the Hubble classification and the galaxy color and morphological parameters, such as the Sérsic index.

The situation at high redshifts is quite different. At z ≳ 3, most of the galaxies are strongly star forming, with a rather small population of quiescent galaxies. The star-forming objects do not appear at all to have a regular morphology, rather, they are irregular, or clumpy. In Fig. 9.46, five z ∼ 2 galaxies are shown as a IJH-color composite image. The irregular, knotty structure is easily seen, and many of the bright knots are clearly well separated from the center of the galaxy—these galaxies do not appear to fall on the Hubble sequence. These clumps have a characteristic size of ∼ 1 kpc, and they seem to be projected onto a kind of disk galaxy. In fact, using high angular resolution integral field spectroscopy, the rotation of several of such z ∼ 2 galaxies could clearly be shown. But these disk are not rotating quietly, in contrast to local disk galaxies, they have a very large velocity dispersion. This fact renders the interpretation of the observed velocity field in terms of Kepler rotation more complicted than for thin, kinematically cold disk galaxies.

However, we should keep in mind that the rest-frame UV light distribution is dominated by star-forming regions. The second and third column in Fig. 9.46 show the surface brightness of these galaxies in two different filters separately, and an estimate of the rest-frame U − V color is shown in the fourth column. We can see that the clumps are typically significantly bluer than the rest of the galaxy, and hence their stellar population has a younger age than the underlying disk. The stellar mass contained in the clumps make a 7 % contribution to the total stellar mass of the galaxy, but they contribute about 20 % to the star-formation rate. Finally, the right column shows the reconstructed stellar surface mass density. Now the picture is a quite different one: the stellar mass is seen to be centrally concentrated, and no prominent off-center clumps are present.

Quiescent galaxies become more abundant towards lower redshift; it is estimated that the stellar mass contained in quiescent objects increases by a factor ∼ 15 between redshifts 3 and 1, and by another factor of ∼ 3 from there until today. In other word, the number of passive red galaxies has at least doubled since z = 1 until today, so that many of the early-type galaxies in the current Universe arrived on the red sequence at rather low redshift. For z > 2, peculiar galaxies dominate the galaxy population, with some quiescent, spheroidal galaxies already present then, but a negligible disk population. At a redshift around z ∼ 2, the abundance of spheroidal and disk galaxies together start to overtake the peculiar population, where this redshift depends on mass: at higher mass, the fraction of Hubble sequence-like galaxies is higher than at lower masses, indicating that they finish their morphological evolution earlier. Thus, starting from z ∼ 2, the Hubble sequence is gradually built up.

Size evolution. Red, quiescent galaxies at z ∼ 2 not only have a regular morphology compared to the clumpy star-forming galaxies, but they also are more massive and more compact. The latter aspects can be seen in Fig. 9.47, where in the left panel the effective radius is plotted as a function of stellar mass, for galaxies with Sérsic index n > 2. 5 and different redshift bins. Compared to the local population of early-type galaxies, higher-redshift spheroidal galaxies are significantly smaller at fixed stellar mass. The effective radius as a function of redshift, for a fixed stellar mass, is shown in the upper panel on the right. The size evolution is fitted with a power law of the form . The decrease in radius at fixed stellar mass by a factor of ∼ 3 at z ∼ 1. 5 implies that these galaxies have a stellar density larger by a factor ∼ 30 than present day early-type galaxies—these high-redshift galaxies are very different from the current population. The higher density also implies a larger velocity dispersion, which is indeed observed, as seen in the bottom panel on the right in Fig. 9.47.

Indeed, such compact galaxies are very rare in the local Universe—that means that the typical z ∼ 2 quiescent galaxy must have evolved significantly to fit into the local zoo of galaxies. Two principal possibility for this evolution exist: either the galaxies grow in size, at fixed stellar mass, or they accumulate more mass in their outer parts, thereby growing in mass and in size, such that they become less compact in this evolution. The latter possibility seems to be closer to the truth, as shown by simulations. Minor merging processes can yield an evolution that is compatible with the observational finding. Thus, early-type galaxies seem to grow inside-out: Their inner region was in place at earlier epochs, their outer parts were added lateron by merging processes.

The interstellar medium in high-redshift galaxies differs from that of local galaxies in a number of properties, of which we mention just a few here.

Metallicity. For local galaxies, there is a clear trend of increasing metallicity with increasing mass, as shown in Fig. 3.​40. A similar trend is observed for Lyman-break galaxies at z ∼ 2, except that the normalization of the mass-metallicity relation is smaller by a factor ∼ 2, as can be seen in Fig. 9.48. Of course, this result does not come unexpectedly, since at earlier redshifts, there was less time to enrich the ISM.

Whereas the metallicity of star-forming galaxies is lower than at the current epoch, at least some galaxies managed to enrich their ISM to about Solar values, as can be inferred from the metallicity of the broad-line emitting gas in high-redshifts QSOs. Not only did these objects form supermassive black holes with M _• ≳ 10⁹ M _⊙, but the chemical evolution in these objects was already mature at a small cosmic age. Of course, luminous high-z QSOs are rare and most likely populate the most massive halos available at these epochs.

Gas content. Second, the high star-formation rate implies the presence of a large reservoir of gas. As we have seen, in the current Universe the gas-mass fraction of even late-type spiral galaxies is below ∼ 30 %; in contrast to that, high-z star-forming galaxies typically have a gas-mass fraction of ∼ 50 %.

Instead of considering the absolute star-formation rate , it is often meaningful to study the specific star-formation rate, , i.e., the star-formation rate per unit stellar mass. This quantity has the units of an inverse time: the inverse of the specific star-formation rate is the time it would take to built up the stellar mass present if the star-formation rate would be a constant. At a fixed gas-mass fraction, this time is very similar to the time-scale on which all the gas in a galaxy is transformed into stars. The specific star-formation rate of Lyman-break galaxies increases by a factor ∼ 10 between the current epoch and z ∼ 1. 5, but then appears to stay remarkably constant out to z ∼ 7 (where, of course, the results at the highest redshifts carry an appreciable uncertainty).

Dust. The fact that the far-IR emission is the best indicator for star formation already indicates that these galaxies must contain a significant dust abundance. The dust temperature, which determines the spectral shape in the far-IR, can vary in these dusty galaxies over quite a substantial range, 25 K ≲ T _d ≲ 65 K, as determined recently from Herschel observations. The impact of the dust temperature on the spectral shape means that single-band selection, e.g., the flux at 850 μm, can bias against objects with hotter dust temperature.

Whereas most of the properties of z ∼ 6 QSOs are indistinguishable from those of low-redshift QSOs, there are signs that they differ in their near-IR properties. In local QSOs, the UV/optical continuum emission is believed to be mainly due to the accretion disk, whereas the near-IR radiation is due to hot dust, heated by the AGN. The ratio of NIR-to-optical luminosity of z ≲ 5 QSOs is confined to a rather small range around ∼ 1. In a sample of 21 z ∼ 6 QSOs, there are two sources without detected NIR emission, yielding an upper limit to the NIR-to-optical flux ratio that is one order of magnitude smaller than the value typically observed. Using a control sample of more than 200 lower-z QSOs, not a single one has this flux ratio as low as the sources at z ∼ 6. The lack of detectable NIR emission can be attributed to the lack of dust in these systems.

A clue for the origin of this lack of dust is obtained from a second finding: the NIR-to-optical flux ratio for lower-z QSOs shows no correlation with the SMBH mass as estimated from the width of broad emission lines. In contrast to that, there seems to be a strong dependence of this flux ratio on the SMBH mass in the sample of z ∼ 6 QSO, in that the ratio increases with increasing M _•. A simple interpretation of this result could be that these high-redshift QSOs were able to build up their SMBH and the corresponding accretion disk, but that they were unable yet to form large masses of dust. The larger M _•, the more evolved is the AGN, and the more dust was created. It remains to be seen whether this interpretation survives further observational tests.

The first point to note from Fig. 9.49 is the relatively flat energy distribution between the UV- and the mm-regime. Since both, the UV-radiation and the far-IR radiation originate almost entirely from star-formation, the flat energy distribution implies that essentially half of the energetic photons emitted from newly-formed stars are absorbed by dust and reradiated in the FIR. Hence, estimates of the star-formation activity from UV-flux alone will on average be biased low by ∼ 50 %.

Absolute measurements of the intensity of the background radiation are difficult to obtain, since it requires an absolute calibration of the instruments. The filled circles and stars with error bars in Fig. 9.50 show estimates of the background radiation level in the optical and near-IR. The black arrows show the integrated light of all sources which are detectable in the deepest observations with HST; within the error bars of the estimated level of the background radiation, these results are compatible with the background being solely due to the superposition of individual optical and near-IR sources, i.e., galaxies and (to a lesser degree) AGNs.

Observations of background radiation in the infrared are very difficult to accomplish, in particular due to the thermal radiation from the instruments and the zodiacal light. However, the DIRBE and FIRAS instruments onboard COBE provided a measurement (in fact, the detection) of the CIB. The question now is whether the CIB can be understood as well as being due solely to individual sources.

Confusion limit. Since mid- and far-IR observations are only possible from space, finding the answer to that question is challenging. Infrared observatories in space have a rather small aperture which, together with the long wavelength, yields a rather large point-spread function (PSF). This implies that when one observes to low flux limits, where the mean angular separation of sources on the sky becomes comparable to the size of the PSF, these sources can not be separated. This yields a lower flux limit for the detection of individual sources, called the confusion limit. The smaller the telescope, the shallower is the confusion limit reached. For example, the flux limit down to which individual sources could be identified with the Spitzer satellite at 160 μm corresponds to only 7 % of the CIB at this wavelength. The much larger mirror on the Herschel satellite lowered the confusion limit such that individual sources can be identified which account for about 52 % of the CIB. Going to larger wavelength, the confusion limit is even more severe.

Stacking. However, one can dig deeper into the source counts with a technique called stacking. Taking the position of sources detected at some smaller wavelength (where the confusion limit is fainter), and adding up the flux in the longer wavelength band around all these positions, one obtains the mean long-wavelength flux of these sources. With this method, one will miss all fainter sources which do not have a detected counterpart in the short-wavelength input catalog, so that wavelength should be selected carefully. Given the characteristic spectrum of FIR-bright sources shown in Fig. 9.30, one expects that most of the sources radiating in the FIR will have an appreciable flux at 24 μm. Since Spitzer was particularly sensitive at this wavelength, the corresponding source catalog is best for a stacking analysis. Furthermore, if the redshifts of the sources selected at 24 μm is known, the stacking analysis can be used to determine the redshift distribution of the contributions to the CIB in the FIR. With stacking, the source counts can be followed to about three times lower flux than the confusion limit of individual sources permits.

The state of the art is defined by deep fields observed with the Herschel observatory, owing to its large aperture and sensitive instrumentation. From observing the well-studied GOODS, COSMOS, and ECDFS fields, where detailed multi-waveband data from other observatories are available, Herschel was able to attribute between 65 and 89 %, depending on wavelength, of the estimated CIB level between 100 and 500 μm to resolved sources or sources seen at 24 μm. A moderate extrapolation of the source counts to fainter flux limits then shows that the bulk, if not all, of the CIB comes from galaxies or AGNs.

In addition, the redshift distribution of the CIB could be determined. At wavelengths below ∼ 160 μm, more than half of the CIB radiation comes from sources at z < 1, whereas at longer wavelength, the source distribution shifts to increasingly higher redshifts. The major fraction of the CIB is due to galaxies with infrared luminosities in the range 10¹¹ to 10¹² L _⊙, i.e., due to LIRGs.

The added flux of sources, either individually detected or obtained from a stacking analysis, yields a lower limit to the extragalactic background light, which in the UV, optical and near-IR regime is smaller than estimates of the total intensity of the background, as seen in Fig. 9.50, although these latter measurements have fairly larger error bars. Hence, the question arises whether there are other sources of the background light not identified as individual sources—for example, very low surface brightness galaxies that could escape detection. In fact, this question can be answered from observations of blazars (see Sect. 5.​2.​6) at energies in the GeV and TeV regime, as will be described next.

Attenuation of γ -rays: Condition for pair production. High-energy photons from distant sources propagate through the extragalactic background radiation field. If the photon energy is high enough, then by colliding with one of the background photons, it may produce an -pair, in which case it does not reach the Earth. Thus, this pair production attenuates the flux from the source.

In order for pair production to occur, the product of the energies of the background-light photon (ε) and the photon from the source (E _γ ) must be sufficiently high. If the two photons propagate in opposite direction (head-on collision), then the threshold condition is ε E _γ  > (m _e c ²)². In general, if the photon directions enclose an angle θ, this gets modified to We see that the threshold energy is smallest for head-on collisions, where θ = π. The cross-section for this process is small for photon energies very close to the threshold, reaches its maximum at about twice the threshold energy (9.3), and decreases again for larger energies. At the maximum of the cross-section, the relation between the two photon energies can be written in practical units,

from which we see that for the attenuation of TeV photons, extragalactic background photons in the near-IR are most efficient, whereas radiation in the tens-of-GeV-regime can be attenuated by UV-photons.

Optical depth. In order to derive the efficiency of the attenuation, one needs to calculate the optical depth τ _{γ γ} (E _γ , z), which depends on the energy of the γ-ray and the source redshift. To obtain τ _{γ γ} (E _γ , z), the pair-production cross section needs to be integrated along the line-of-sight to the source, multiplied by the spectral energy density of the background radiation; for this, the redshift evolution of the background radiation needs to be accounted for. Since the extragalactic background light can be observed only at the current redshift, one needs to model its redshift evolution, based on what is known about the source population. A particular model is shown in the top left panel of Fig. 9.51, which also includes the CMB, and the corresponding photon number density per logarithmic photon energy interval is shown in the top right panel. Based on the background light model, the optical depth for pair production can be calculated, which is shown in the bottom left panel of the same figure. τ _{γ γ} (E _γ , z) is a strong function of both, the γ-ray energy and the redshift. The plateau in τ _{γ γ} at energies ∼ 2 TeV is due to the minimum of the background radiation spectrum at ∼ 10 μm.

The observed flux S _obs(E _γ ) is related to the intrinsic (i.e., non-attenuated) flux S _int(E _γ ) by where the attenuation factor is plotted in the bottom right panel of Fig. 9.51. We can see that the attenuation factor has a very steep decline with photon energy; for example, based on this model we would not expect to see 20 TeV photons from any source at z ≳ 0. 1. This steep, exponential decline implies that the attenuation factor is very sensitive to the model of the extragalactic background light; conversely, if observations yield constraints on the attenuation factor, then strong constraints on the background light can be obtained, at wavelengths depending on the detection of the attenuation, according to (9.4).

Observational constraints on the attenuation. As mentioned in Sect. 5.​5.​4, blazars can emit at GeV and TeV energies, most likely caused by their jet pointing towards us. The Fermi Gamma-Ray Space Telescope, operating in the energy range between 200 MeV and 300 GeV, and the air Cherenkov observatories H.E.S.S., MAGIC and VERITAS, observing in the range between ∼ 50 GeV and 100 TeV, have detected more than 1000 blazars in the GeV-range and more than 30 at TeV energies. Whereas blazars in the GeV-range are observed out to redshifts z ≥ 1, essentially all the TeV blazars are at low redshift, most of them having z ≲ 0. 2. This is indeed what one expects, based on the results in Fig. 9.51.⁷ In principle, these observations of the spectral energy distribution could be used to determine the attenuation factor; however, in order to employ (9.5), one needs some knowledge about the intrinsic flux distribution S _int(E _γ ).

There are various ways how realistic estimates for τ _{γ γ} can be obtained from the observations. The first of these is to base the intrinsic flux distribution on models of the γ-ray emission, and constrain these models by observations at somewhat lower photon energies. However, the models are sufficiently uncertain to preclude very accurate predictions, and thus the corresponding results on τ _{γ γ} are correspondingly uncertain. Second, since the relativistic electrons responsible for the inverse Compton effect that presumably causes the γ-ray emission, are expected to result from acceleration by shock fronts, as mentioned in Sect. 5.​1.​3, the slope of the electron distribution can not be arbitrarily flat, and thus the resulting inverse Compton radiation is also limited in slope; in the notation of Sect. 5.​1.​3, s ≳ 2 and thus α ≳ 0. 5. Assuming this value of the spectral index as a limit for the intrinsic flux distribution, the observed energy distribution can be translated into upper bounds on the attenuation. An even weaker assumption is used by a third methods, where one requires that the intrinsic flux S _int(E _γ ) = S _obs(E _γ ) exp[τ _{γ γ} (E _γ , z)] does not (exponentially) increase with photon energy, as would be the case if the background light intensity would be overestimated.

Results. Depending on which of these methods are used, the results will differ slightly. A particular result is shown in Fig. 9.50, where the magenta curve indicates the upper bound on the extragalactic background light obtained from the high-energy observations of blazars. This upper bound is almost coincident with the lower bound obtained from the resolved source counts in the UV, optical and near-IR regime, strongly arguing that there are no other significant contributions of the background light than the observed galaxies and AGN. It thus seems that the spectral intensity of the background light in this spectral regime is now rather well determined. This conversely implies that the optical depth τ _{γ γ} is very well constrained, which in turn allows us to derive the intrinsic flux distribution from the observed one. In the future, this method will therefore yield more detailed constraints on the emission mechanism for high-energy radiation from blazars and other AGNs.

The first X-ray experiment in astronomy, a balloon flight in 1962, discovered a diffuse X-ray emission across the sky, confirmed by the first X-ray satellites which discovered not only a number of extragalactic X-ray sources (such as AGNs and clusters of galaxies), but also an apparently isotropic radiation component. The spectrum of the cosmic X-ray background (CXB) is a very hard (i.e., flat) power law, cut off at an energy above ∼ 40 keV, which can roughly be described by with E ₀ ∼ 40 keV. A recent estimate of the spectrum of the CXB is shown in Fig. 9.52. The estimates from different instruments agree in general, though differences in the level are clearly visible. These differences can have a number of origins, including cosmic variance (the spectral shape of the CXB is usually determined from rather small fields, so there could be variations from field to field), stray light entering the telescope, and remaining calibration uncertainties of the instruments. Together, the CXB is known with an uncertainty of ∼ 20 %.

Initially, the origin of this radiation was unknown, since its spectral shape was different from the spectra of sources that were known at that time. For example, it was not possible to obtain this spectrum by a superposition of the spectra of known AGNs.

ROSAT, with its substantially improved angular resolution compared to earlier satellites (such as the Einstein observatory), conducted source counts at much lower fluxes, based on some very deep images. From this, it was shown that at least 80 % of the CXB in the energy range between 0.5 and 2 keV is emitted by discrete sources, of which the majority are AGNs. Hence it is natural to assume that the total CXB at these low X-ray energies originates from discrete sources, and observations by XMM-Newton and Chandra have confirmed this.

However, the X-ray spectrum of normal AGNs is different from (9.6), namely it is considerably steeper (about S _ν  ∝ ν ^−0. 7). Therefore, if these AGNs contribute the major part of the CXB at low energies, the CXB at higher energies cannot possibly be produced by the same AGNs. Subtracting the spectral energy of the AGNs found by ROSAT from the CXB spectrum (9.6), one obtains an even harder spectrum, resembling very closely that of thermal bremsstrahlung. Therefore, it was supposed for a long time that the CXB is, at higher energies, produced by a hot intergalactic gas at temperatures of k _B T ∼ 30 keV.

This model was excluded, however, by the precise measurement of the thermal spectrum of the CMB by COBE, showing that the CMB has a perfect blackbody spectrum. If a postulated hot intergalactic gas were able to produce the CXB, it would cause significant deviations of the CMB from the Planck spectrum, namely by the inverse Compton effect (the same effect that causes the SZ effect in clusters of galaxies—see Sect. 6.​4.​4). Thus, the COBE results clearly ruled out this possibility.

Deep observations with Chandra and XMM (e.g., in the CDFS shown in Fig. 9.14) have finally resolved most of the CXB also at higher energies, as seen in Fig. 9.53. From source counts performed in such fields, more than 75 % of the CXB in the energy range of 2 keV ≤ E ≤ 10 keV could be resolved into discrete sources. Again, most of these sources are AGNs, but typically with a significantly harder (i.e., flatter) spectrum than the AGNs that are producing the low-energy CXB. Such a flat X-ray spectrum can be produced by photoelectric absorption of an intrinsically steep power-law spectrum, where photons closer to the ionization energy are more efficiently absorbed than those at higher energy. According to the classification scheme of AGNs discussed in Sect. 5.​5, these are Type 2 AGNs, thus Seyfert 2 galaxies and QSOs with strong intrinsic self-absorption. We should recall that Type 2 QSOs have only been detected by Chandra—hence, it is no coincidence that the same satellite has also been able to resolve the high-energy CXB.

However, at even higher energies most of the CXB was still unaccounted for—even the observed Type-2 AGNs could not account for it. It thus seems that there is a population of sources in the Universe which dominate the X-ray emission at high energies, still escape the observations at low X-ray frequencies. These could be heavily obscured AGNs, where only the hard X-rays manage to escape the emitting region. With the X-ray telescope onboard the Swift satellite, a significant number of such heavily obscured AGNs were found. Their estimated number density, together with their spectral energy distribution, make it plausible that they are the missing population of ‘hidden black holes’ responsible for the hard CXB.

We will start by discussing the most important indicators of star formation.

Emission in the far infrared (FIR). This is radiation emitted by warm dust which is heated by hot young stars. Observations yield for the approximate relation between the FIR luminosity and the SFR For this relation it is assumed that all the energetic photons from newly born hot stars are absorbed locally and heat the dust; more generally, this expression yields the SFR that is dust enshrouded. The wavelength range over which L _FIR is determined should be large, covering the two decades from 8 μm to 1 mm, so that this luminosity is essentially independent of the dust temperature. However, in most cases the observation do not cover such a broad spectral region, so that interpolation and extrapolation of the spectral behavior is required to determine L _FIR, based on template spectra constructed from very well observed sources. This procedure therefore carries an intrinsic uncertainty in the determination of the SFR. For large samples of sources, one often uses the 24 μm flux to obtain an estimate of L _FIR; this wavelength is seen as a good compromise between the need to go to large wavelengths and the decreasing sensitivity and field-of-view of infrared instrumentation. The Spitzer Space Telescope played a very important role in the studies of star formation at these wavelengths.

Radio emission by galaxies. A very tight correlation exists between the radio luminosity of galaxies and their luminosity in the FIR, over many orders of magnitude of the corresponding luminosities. Since L _FIR is a good indicator of the star-formation rate, this should apply for radiation in the radio as well (where we need to disregard the radio emission from a potential AGN component). The synchrotron radio emission of normal galaxies originates mainly from relativistic electrons accelerated in supernova remnants (SNRs). Since SNRs appear shortly after the beginning of star formation, caused by core-collapse supernovae at the end of the life of massive stars in a stellar population, radiation from SNRs is a nearly instantaneous indicator of the SFR. Once again from observations, one obtains

H α emission. This line emission comes mainly from the Hii-regions that form around young hot stars with M ≳ 10M _⊙. As an estimate of the SFR, one uses For redshifts z ≳ 2, the observed Hα lines moves into the near-IR part of the spectrum and is thus much more difficult to observe. One can also employ other emission lines, such as Hβ or Lyα. However, whereas Hβ can be observed in the optical to higher redshifts, due to its shorter wavelength, it is a weaker line. For Lyα, the uncertainty to convert line flux into a SFR is much larger, since it is a transition to the ground state of hydrogen (a so-called resonance line). Because of that, a Lyα photon is easily absorbed by neutral hydrogen, which is then excited and reemits a Lyα photon in a random direction. Effectively, thus, this process is a scattering of the photon. In order to leave the interstellar medium of a galaxy, a Lyα photon may be subject to many such scatters, which implies that is path inside the ISM is very much enhanced. This longer path then also increases the probability for the photon to become absorbed by dust. Therefore, the conversion of Lyα flux that leaves a galaxy and the SFR is burdened with a large uncertainty. Alternatively, one can also consider recombination lines of hydrogen coming from transition between higher energy states of the atom, such a Bracket γ, a transition that occurs in the NIR of the spectrum, or transition lines that have millimeter or radio wavelengths.

UV radiation. This is mainly emitted by O and B stars, i.e., by hot young stars thus indicating the SFR in the most recent past, with This relation assumes that the UV-flux can leave the galaxy without being attenuated by dust absorption (and it neglects that there may be an AGN contribution to the UV luminosity). However, in most galaxies this is not a reasonable assumption, and the observed L _UV must be corrected for this effect. Due to the wavelength-dependence of dust absorption, extinction is always connected to reddening, thus affecting the spectral slope of the UV-radiation. One therefore expects that the redder the UV-spectrum, the larger are the effects of dust obscuration. To quantify this effect, one has to assume an intrinsic, unobscured spectral shape, and to make assumptions about the dust properties, specifically regarding its wavelength dependence of the extinction coefficient (see Fig. 2.​6). The unobscured spectral shape can be obtained from the shape of the IMF at the high-mass end (i.e., over that mass region of stars from which the UV-radiation is emitted), whereas there is considerable uncertainty about the variation of dust properties between different objects.

As a result of this procedure, one finds that only a small fraction of UV-photons actually leaves an UV-selected galaxy. A typical value for this escape fraction in a Lyman-break galaxy is ∼ 0. 2, which means that the observed UV-flux has to be corrected by a factor ∼ 5 to obtain the corresponding SFR.

X-ray luminosity. We have seen (Fig. 9.15) that non-active galaxies are X-ray emitters. Most of the X-ray emission is due to high-mass X-ray binaries which are members of a young stellar populations. About 25 % of the X-ray emission from a normal galaxy is due to bremsstrahlung from a hot interstellar medium; since its heating is provided by star-formation activity, it should also scale with the star-formation rate. Hence, if a contribution from an AGN can be excluded, the X-ray luminosity should be a good indicator for the star-formation rate. One finds a tight relation, where the X-ray luminosity is integrated from 0. 5 to 8 keV, and where the scatter in this relation is estimated to be less than a factor of 1.5.

Comparison. Applied to individual galaxies, each of these estimates is quite uncertain, which can be seen by comparing the resulting estimates from the various methods (see Fig. 9.54). For instance, Hα and UV photons are readily absorbed by dust in the interstellar medium of the galaxy or in the star-formation regions themselves. Therefore, the relations above should be corrected for this self-absorption, which is possible when the reddening can be obtained from multi-color data. It is also expected that the larger the dust absorption, the stronger the FIR luminosity will be, causing deviations from the linear relation SFR_FIR ∝ SFR_UV. After the appropriate corrections, the values for the SFR derived from the various indicators are quite similar on average, but still have a relatively large scatter.

There are also a number of other indicators of star formation. The fine-structure line of singly ionized carbon at λ = 157. 7 μm is of particular importance as it is one of the brightest emission lines in galaxies, which can account for a fraction of up to 1 % of their total luminosity. The large abundance of carbon, together with the fact that this transition can be collisionally excited even at low temperature, result in this line to have a major role in the cooling of the neutral interstellar medium, and thus signifies the presence of star-forming regions. At its wavelength, this line is difficult to observe and until recently has been detected only in star-forming regions in our Galaxy and in other local galaxies. However, more recently this line was detected from the host galaxies of high-redshift QSOs and SMGs, where it shifted into more accessible spectral windows.

The density of star formation, ρ _SFR, is defined as the mass of newly formed stars per year per unit (comoving) volume, typically measured in . Therefore, ρ _SFR as a function of redshift specifies how many stars have formed at any time. By means of the star-formation density we can examine the question, for instance, of whether the formation of stars began only at relatively low redshifts, or whether the conditions in the early Universe were such that stars formed efficiently even at very early times.

Investigations of the SFR in galaxies, by means of the above indicators, and source counts of such star-forming galaxies, allow us to determine ρ _SFR. The plot of these results (Fig. 9.55) is sometimes called “Madau diagram”. In about 1996, Piero Madau and his colleagues accomplished, for the first time, an estimate of the SFR at high redshifts from Lyman-break galaxies in the Hubble Deep Field North. For these early results, the intrinsic extinction was neglected. In order to correct for this extinction, the progress in FIR and sub-millimeter astronomy was extremely important, as we saw in Sect. 9.3.3.

There is a strong increase in ρ _SFR from the current epoch to z = 1 by about a factor 10, a further slight increase towards z ∼ 2, and a decrease at redshifts beyond z ∼ 3. These results have more recently been confirmed by investigations with the Spitzer and Herschel satellites, observing a large sample of galaxies at FIR wavelengths. Whereas the star-formation rate density at low redshifts is dominated by galaxies which are not very prominent at FIR wavelength, this changes drastically for redshifts z ≳ 0. 7, above which most of the star-formation activity is hidden from the optical view by dust.⁸ The increasing importance of dust-obscured star formation is concluded from the very strong redshift evolution of the infrared luminosity function of galaxies shown in Fig. 9.43, and the corresponding evolution of the infrared luminosity density (Fig. 9.44).

Integrating the star-formation density over cosmic time, one obtains the stellar mass density as a function of redshift, shown in Fig. 9.56. From this we conclude that most stars in the present-day Universe were already formed at high redshift: star formation at earlier epochs was considerably more active than it is today. Although the redshift-integrated star-formation rate and the mass density of stars determined from galaxy surveys slightly deviate from each other, the degree of agreement is quite satisfactory if one recalls the assumptions that are involved in the determination of the two quantities: besides the uncertainties discussed above in the determination of the star-formation rate, we need to mention in particular the shape of the IMF of the newly formed stars for the determination of the stellar mass density. In fact, Fig. 9.56 shows that we have observed the formation of essentially the complete current stellar density.

The difference between the observed stellar mass density and the one predicted from Fig. 9.55 is largest at high redshifts, which may be due to the uncertainties with which ρ _SFR is determined at high redshifts. With the more recent very deep fields observed with HST, the unobscured star-formation rate can be estimated at larger redshifts, based on the luminosity function of Lyman-break galaxies (Fig. 9.41). The result is shown in Fig. 9.57, which shows a steep decline of ρ _SFR towards higher redshift.⁹ Hence, the bottom line is that there was an epoch in the Universe, between redshifts ∼ 1 and ∼ 4, where the star-formation activity was largest. This epoch coincides with the period in our Universe where the QSO activity was highest—indicating that the built-up of the supermassive black hole mass in the Universe happened in parallel to the formation of the stellar population. The close correlation between SMBH mass and properties of the stellar population discussed in Sect. 3.​8.​3 may thus find a first explanation in this parallel evolution.

Different modes of star formation. Whereas most of the star formation in the local Universe occurs in spiral and irregular galaxies at a modest rate (so-called quiescent star formation), the star-formation activity at higher redshifts was dominated by bursts of star formation, as evidenced in the sub-mm galaxies and in LBGs. At a redshift z ∼ 1, the latter has apparently ceased to dominate, yielding the strong decline of the star-formation rate density from then until today. This behavior may be expected if bursts of star formation are associated with the merging of galaxies ; the merger rate declines strongly with time in models of the Universe dominated by a cosmological constant. This transition may also be responsible for the onset of the Hubble sequence of galaxy morphologies around z ∼ 1.

The starburst-AGN connection. The just-mentioned coincidence of the ‘QSO epoch’ with the peak of the star-formation activity can either have a statistical origin, or there can be a connection object-by-object, in the sense that QSO are hosted in galaxies with active star formation. For physical reasons, one would in fact expect such a direct connection, since both processes, star formation and AGN fueling, need a gas supply.

Indeed, in a large fraction of QSOs, clear signs of star-formation activity are found, in some cases at a level where the QSO host galaxy appears as a ULIRG. Conversely, in many of the low-redshift star-forming galaxies, signs of the presence of an AGN are seen. Furthermore, a direct connection between these two processes in a given galaxy does not necessarily have to be observable: it is conceivable that a fresh supply of gas (say, from a major merging of two galaxies) first leads to a strong star-formation activity, and that the accretion onto the central black hole occurs with some delay—or in reverse order.

Statistical studies based on the MIR and FIR emission properties of X-ray selected AGNs suggest that for low-luminosity AGNs, there is no correlation between AGN luminosity and star-formation rate. However, at high AGN luminosity, such a correlation is indeed found, providing a strong hint for the connection between the built-up of the black hole mass and the stellar population in individual sources.

In this chapter, we have considered various aspects of galaxies in the high-redshift Universe. Our discussion of this very quickly evolving field is not complete, but concentrates on some of the central issues. Before we move to another class of high-redshift sources, we want to summarize some of the points mentioned before: